Download as pdf or txt
Download as pdf or txt
You are on page 1of 344

Zhengyang Jiang

Kellogg School of Management, Northwestern University, and NBER

Lecture Notes on
International Finance

March 7, 2024

This is a preliminary draft of my lecture notes on international finance, with an emphasis on the theoretical
foundation. Its preliminary nature is reflected in the fact that some sections are still missing or incomplete.
I will continue to update the notes and welcome comments and suggestions.

I am grateful to my students for their help and feedback. I am also grateful to my colleagues, coauthors, and
friends for their comments and suggestions. All errors are mine.

This note is subject to future revisions.


For the latest version, please visit https://sites.google.com/site/jayzedwye/research.
For comments and questions, please contact me at zhengyang.jiang@kellogg.northwestern.edu.

Electronic copy available at: https://ssrn.com/abstract=4668578


Contents

0 Preface 6

I Introduction 13

1 A Benchmark Two-Country Economy 14


1.A Model Set-up 15
1.B Exchange Rate Accounting 21
1.C Complete-Market Solution 25
1.D Asset Market and Goods Market Views of Exchange Rates 31

2 Puzzles: Challenges to Making Sense of Data 35


2.A Challenges to Making Sense of Exchange Rates 35
2.B Challenges to Making Sense of Quantities and Flows 42

II Understanding the Exchange Rates 50

3 Risk Premia and Factor Structure 51


3.A The No-Arbitrage Approach 52
3.B Currency Risk Premia in the Time Series 64
3.C Currency Risk Premia in the Cross-Section 69
3.D Currency Risk Premia in the Long Run 78

Electronic copy available at: https://ssrn.com/abstract=4668578


jiang – lecture notes on international finance 3

4 Convenience Yields 88
4.A An Illustrative Model 90
4.B Exchange Rate Accounting 94
4.C Measuring the Convenience Yields 100
4.D Connecting the Short Term with the Long Term 104
4.E Discussions (TODO) 107

5 Incomplete Markets 108


5.A An Illustrative Model 109
5.B The No-Arbitrage Approach 120
5.C Multi-Currency Dynamics and International Spill-Over 127
5.D Incomplete Markets vs. Convenience Yields (TODO) 131

6 Monetary and Fiscal Policies 132


6.A Introducing the Nominal Layer 133
6.B Model Set-Up 135
6.C Characterizations under Flexible Prices 142
6.D Characterizations under Sticky Prices 146
6.E Comparing Monetary and Fiscal Policies 154

III Understanding the Quantities and Flows 159

7 Global Imbalances and the Exorbitant Privilege 160


7.A The Insurance Provision View 161
7.B The Reserve Currency Paradox 170
7.C The Safe Asset View 172
7.D The Stability of the International Monetary System 181

8 Government Debt 185


8.A Backward-Looking Accounting 187

Electronic copy available at: https://ssrn.com/abstract=4668578


4 zhengyang jiang

8.B Forward-Looking Valuation 190


8.C The Transversality Condition 200
8.D An Example Economy 204
8.E The Public Debt Valuation Puzzle 207

9 Portfolio Choice and Asset Demand 219


9.A Net Foreign Assets Accounting 220
9.B The Mean-Variance Approach 227
9.C Applications of the Mean-Variance Approach (TODO) 230
9.D The Demand System Approach 230
9.E Application to International Portfolio Allocation 237

10 Market Segmentation and Financial Intermediation 241


10.A A Model of International Financial Intermediation 241
10.B Comparing Segmented Markets with Convenience Yields 248
10.C A Model of Domestic Financial Intermediation 259

A Proof of Selected Results 269


A.1 Proposition 1.3 in Section 1.C 269
A.2 Proposition 1.4 in Section 1.D 270
A.3 Proposition 3.1 in Section 3.A 271
A.4 Proposition 3.2 in Section 3.A 271
A.5 Proposition 3.3 in Section 3.A 272
A.6 Proposition 3.4 in Section 3.B 273
A.7 Proposition 3.5 in Section 3.B 275
A.8 Proposition 3.6 in Section 3.C 276
A.9 Proposition 3.7 in Section 3.C 278
A.10 Proposition 3.8 in Section 3.C 278
A.11 Proposition 3.9 in Section 3.D 279
A.12 Proposition 3.11 in Section 3.D. 280
A.13 Proposition 4.3 in Section 4.B 281

Electronic copy available at: https://ssrn.com/abstract=4668578


jiang – lecture notes on international finance 5

A.14 Lemma 5.1 in Section 5.A 282


A.15 Proposition 5.1 in Section 5.A 284
A.16 Lemma 5.2 in Section 5.A 286
A.17 Proposition 5.2 in Section 5.A 290
A.18 Proposition 5.3 in Section 5.A 293
A.19 Proposition 5.4 in Section 5.B 296
A.20 Proposition 5.5 in Section 5.B 297
A.21 Proposition 5.6 in Section 5.B 297
A.22 Proposition 5.7 in Section 5.C. 298
A.23 Proposition 5.8 in Section 5.C. 298
A.24 Proposition 6.2 and 6.3 in Section 6.C 302
A.25 Proposition 6.4 in Section 6.D 303
A.26 Proposition 6.5 in Section 6.D 304
A.27 Proposition 6.6 in Section 6.D 305
A.28 Proposition 7.1 in Section 7.A 307
A.29 Proposition 7.2 in Section 7.C 308
A.30 Proposition 8.1 in Section 8.A 309
A.31 Proposition 8.2 in Section 8.B 310
A.32 Proposition 8.3 in Section 8.B 311
A.33 Proposition 8.4 in Section 8.B 313
A.34 Proposition 8.5 in Section 8.B 314
A.35 Proposition 8.6 in Section 8.C 314
A.36 Proposition 8.7 in Section 8.C 316
A.37 Proposition 8.8 in Section 8.E 317
A.38 Proposition 8.9 in Section 8.E 318
A.39 Proposition 9.1 in Section 9.A 320
A.40 Proposition 9.2 in Section 9.B 320
A.41 Propsition 9.3 in Section 9.B 322
A.42 Proposition 10.1 in Section 10.A 322
A.43 Proposition 10.2 in Section 10.B 323
A.44 Proposition 10.3 in Section 10.C 323
A.45 Proposition 10.4 in Section 10.C 325

Bibliography 327

Electronic copy available at: https://ssrn.com/abstract=4668578


0
Preface

What is international macroeconomics and finance? Here is an an-


swer I found by asking ChatGPT.
International macroeconomics and finance is a subfield of economics that
deals with the economic interactions and relationships between countries. It
includes the study of international trade, exchange rates, and the balance of
payments, as well as the economic policies of governments and their impacts
on international economic conditions.
In international macroeconomics and finance, economists and policymakers
analyze how countries can achieve stability and prosperity in an increasingly
interconnected global economy. They also study how economic events and
policies in one country can affect other countries, and how countries can work
together to address global economic challenges.
Some key issues in international macroeconomics and finance include trade
liberalization, exchange rate regimes, capital flows, international financial
crises, and economic development. It is an important area of study for poli-
cymakers, economists, and business leaders, as it helps them understand and
navigate the complex economic interactions between countries.

A pretty impressive summary, right? In the narrower scope of this


lecture note, we will focus on the following four types of issues that
fall under the umbrella of international macroeconomics and finance.
First, we would like to understand the determinants of exchange
rates and international asset prices. As with generic asset pricing, we
would like to understand what drives exchange rates and interna-
tional asset prices to fluctuate ex-post, what drives some currencies
and assets to have higher expected returns ex-ante, and how to char-
acterize the risk-return trade-off. The international asset prices in-
clude bond yields across countries, which, as we will see, are closely
related to the currency market, as well as international equity prices,
though they will receive much shallower coverage in this note.
Second, there is a quantity dimension, which includes capital
flows, international portfolio positions and imbalances, as well as
imports and exports of various goods. Another key agenda in inter-
national macroeconomics and finance is to connect these quantities

Electronic copy available at: https://ssrn.com/abstract=4668578


preface 7

to the exchange rates and asset prices. We would like to understand


how these quantities evolve between countries, sectors, and agents,
and whether they share common drivers with exchange rate fluctua-
tions.
Third, we also need to understand the roles played by monetary
and fiscal policies in regulating and shaping the asset price and
quantity dynamics. Besides the conventional interest rate policies
and foreign exchange market interventions, unconventional monetary
policies such as large scale asset purchases and sales have become
very important and change our understanding of the monetary trans-
mission mechanism. Moreover, the government debt and deficits
have risen to unprecedented levels in the post-war history. The fiscal
outlooks also shape how investors price the government liabilities
including debt and currencies.
Finally, from a global perspective, we need to understand what
constitutes the international monetary system, what makes it stable
or unstable, where it comes from, and how it is going to evolve.
Of particular interest is the asymmetric core-periphery structure
suggested by exchange rates, capital flows and imbalances data. We
would like to understand what enables the U.S. to currently occupy
the central position, and how its centrality is shaping the financial
markets.
A good understanding of these four levels of issues in interna-
tional macroeconomics and finance helps us understand the world
and design better policies and institutions. Pioneers in this field have
thought about these issues throughout history. To motivate the issues
that we study, let us consider some of my favorite historical episodes
that bear economic significance for this research field.
First, consider the exchange rate fluctuations during the interbel-
lum years. The left panel of Figure 1 plots the annual government
surplus in billion local currency units. The U.K., France, and Ger-
many all incurred fiscal deficits during and after the World War I.
The U.K. had the most modest fiscal deficits, followed by France, and
Germany had the largest deficits.
These large fiscal deficits were accompanied by dramatic currency
depreciation. The right panel of Figure 1 plots these countries’ nom-
inal exchange rates against the U.S. dollar. All three currencies went
off their pegs to the gold. In the cross-section, consistent with the
U.K. running the smallest fiscal deficits, the U.K. pound depreciated
the least. The French franc depreciated more than the U.K. pound,
and the German mark depreciated the most. In the time series, after
Germany introduced the new currency, the Reichsmark, in 1924, the
exchange rate stabilized as the fiscal deficits were reduced.
These cross-sectional and time-series patterns suggest a strong

Electronic copy available at: https://ssrn.com/abstract=4668578


8 jiang – lecture notes on international finance

link between fiscal deficits and exchange rates. We will discuss this
link in detail in Chapter 6. This period also witnessed the rise of the
U.S. dollar as the dominant international currency and the transfer
of international financial center from the U.K. to the U.S. We will dis-
cuss the financial hegemon and the architecture of the international
monetary system in Chapter 7.

20
100
0

-20

-40 10-1

-60

-80 10-2

-100

-120 10-3

-140

-160 10-4
1910 1915 1920 1925 1930 1935 1940 1910 1915 1920 1925 1930 1935 1940

Figure 1: Fiscal Surpluses and Exchange


Rates in the Interbellum Years. Source:
Second, consider the dollar exchange rate after the break-down Mitchell [2007].
of the Bretton Woods system in 1973. Before 1973, many foreign cur-
rencies were pegged to the U.S. dollar at fixed exchange rates, which
was in turn pegged to gold. As this system broke down, the ex-
change rates became floating. Figure 2 plots the trade-weighted real
dollar index, which measures the strength of the U.S. dollar in real
terms against a basket of foreign currencies weighted by their trade
shares. We observe several periods of large and persistent dollar ap-
preciation and depreciation. For example, the dollar depreciated by
over 25% in the 70s after it was depegged from the gold, followed
by appreciation by almost 50% in the early 80s aided by higher U.S.
interest rates. Then, concerned about the dollar’s strength, the U.S.
and other major countries reached the Plaza Accord in 1985. They
intervened the foreign exchange market to orchestrate dollar depre-
ciation against the French franc, the German mark, the Japanese yen
and, the British pound. The dollar appreciated again in the mid-90s,
followed by dollar depreciation in early 2000s due to increased risk
appetite for investing abroad. This run of dollar depreciation and
capital outflows came to an end when the global financial crisis hit
in 2008, which led to a flight to safety and dollar appreciation. The
dollar depreciated after the panic subsided. Finally, the dollar experi-
enced a decade-long appreciation since early 2010s, which lasted till
this day.

Electronic copy available at: https://ssrn.com/abstract=4668578


preface 9

While this is far from a complete account of the dollar’s exchange


rate movement, it begs two obvious questions. First, is there a sys-
tematic pattern in the dollar’s exchange rate in relation to global
crisis? Second, are there regime shifts in the main drivers of the dol-
lar exchange rate? We will discuss these questions throughout this
note.

Figure 2: The U.S. Real Exchange Rate.


Source: FRED.
20

10

-10

-20

-30

-40
1975 1980 1985 1990 1995 2000 2005 2010 2015 2020

Third, consider the evolution of sovereign bond yields in the Eu-


rozone, which share the same currency denomination. Figure 3 plots
the sovereign yields in different Eurozone countries, which shows
three distinct regimes. First, before the global financial crisis in 2008,
the yields were almost identical. Second, after the crisis, the yields
diverged sharply as the European sovereign debt crisis unfolded. For
example, Germany’s and France’s bond yields remained as low as 2%
in early 2010s, whereas Italy’s and Portugal’s bond yields rose above
10%. Third, the yields converged again after 2015, possibly aided by
the ECB bond purchase programs this time.
It turns out that these yield differentials are driven not only by the
default spreads, which reflect the fact that different sovereign debt
has different default probabilities and hence carries different risk
premia, but also by the convenience yields, which we will discuss in
Chapter 4. Moreover, the monetary and fiscal policies played impor-
tant roles in shaping these yield differentials. We will discuss these
issues in Chapter 6.
Finally, consider the net flows to the U.S. Treasury market. Fig-
ure 4 plots the annualized net flows into U.S. Treasury notes and
bonds, and decomposes the flows into three components: domestic

Electronic copy available at: https://ssrn.com/abstract=4668578


10 jiang – lecture notes on international finance

Figure 3: 5-Year Bond Yields in


country Eurozone. Source: Jiang, Lustig,
20 Van Nieuwerburgh, and Xiaolan
France
[2020c].
Netherlands
15 Austria
Belgium
Finland
10
Italy
Ireland
5 Spain
Portugal

0 Germany

2005 2010 2015 2020


year

investors, the Fed, and the rest of the world (ROW). We observe three
different regimes. First, from the mid 1970s until the mid 1990s, the
U.S. domestic agents, including the financial sector such as banks,
insurance companies, pensions and the household sector, absorbed
a significant fraction of the new debt issuance. They were the major
players in the Treasury market, whereas the flows and the positions
of the Fed and the ROW were relatively small.
There was a distinct shift in the mid-to-late 1990s, when the ROW
became the main Treasury buyers while the domestic investors be-
came net sellers. In this second regime through 2015, the ROW was
by far the most important buyer of U.S. Treasurys. Their inflows
were particularly pronounced during the global financial crisis,
which is consistent with the flight to safety observed in safe dollar
bond prices. The Fed also started playing a more active role since
the global financial crisis, as it undertook quantitative easing and
expanded its balance sheet.
Since 2015, we have entered a third regime characterized by
weaker demand from the ROW. In fact, contrary to its countercyclical
purchases of the U.S. Treasurys in previous decades, it became a net
seller in the Treasury market during the Covid-19 crisis. In compari-
son, the Fed became a much more active buyer during the Covid-19
crisis. Bond purchases from the ROW re-emerged after the Covid-19
crisis, albeit smaller in magnitude than prior to 2015. In Chapters
7 and 9, we will discuss different approaches to understanding the
cross-border asset allocations and the implications for the capital
flows and exchange rates.
The goal of this note is to develop a coherent theoretical frame-
work for understanding all these phenomena in international macroe-

Electronic copy available at: https://ssrn.com/abstract=4668578


preface 11

Figure 4: Net Flows to U.S. Treasury


Notes and Bonds Normalized by U.S.
Domestic
0.10 GDP. Source: Jiang, Krishnamurthy, and
FED
ROW
Lustig [2022a].
Annualized Flows/GDP

0.05

0.00

1970 1980 1990 2000 2010 2020


Year

conomics and finance. That said, I do not mean to give the impres-
sion that all is known. Instead, much remains to be understood, and
a lot is left out by this note. I hope this note offers curious readers a
starting point for further exploration.
This note is organized by three parts: (1) introduction, (2) under-
standing exchange rates, and (3) understanding quantities and flows.
In introduction, I start by building a benchmark two-country model
of international business cycles that is a natural generalization of the
standard closed-economy real business cycle model. This benchmark
model offers a clear starting point, although it falls short in terms
of explaining many stylized facts in the exchange rate and quantity
data. Then, the part on exchange rates develops a series of extensions
and modifications of the benchmark model that can help us under-
stand the exchange rate dynamics. Building on this part, the part on
quantities and flows further develops frameworks that can help us
speak to the quantity dynamics.
In the course of working on this note, I have benefited from many
discussions with my colleagues, coauthors, and friends. I am thank-
ful to Xuning Ding, Jialu Sun, and Yudan Ying for their research
assistance for related research projects. I am grateful for Ravi Jagan-
nathan, Robert Korajczyk, and the William and Mary Breen Fund for
funding support. All errors in this note are mine.
I assume that the readers have taken an entry-level graduate
course in macroeconomics or asset pricing. I also assume that the
readers are familiar with the basic tools in dynamic programming
and (for a few sections) stochastic calculus. I will try to keep the
math simple, and give priority to the development of intuitions.
So, welcome to the field of international macroeconomics and fi-
nance! I very much hope you enjoy the journey. As many of you have
seen closed-economy macroeconomics and finance, let me end with

Electronic copy available at: https://ssrn.com/abstract=4668578


12 jiang – lecture notes on international finance

some quotes from More Is Different: Broken symmetry and the nature
of the hierarchical structure of science [Anderson, 1972]. Despite ad-
dressing a totally different research field in science, this article may
provide us with some useful insights for our uneven path to transi-
tion from one-country, closed-economy economics to international,
open-economy economics.
The reductionist hypothesis may still be a topic for controversy among philoso-
phers, but among the great majority of active scientists I think it is accepted
without question. The workings of our minds and bodies, and of all the an-
imate or inanimate matter of which we have any detailed knowledge, are
assumed to be controlled by the same set of fundamental laws, which except
under certain extreme conditions we feel we know pretty well...
The main fallacy in this kind of thinking is that the reductionist hypothesis
does not by any means imply a "constructionist" one: The ability to reduce
everything to simple fundamental laws does not imply the ability to start
from those laws and reconstruct the universe. In fact, the more the elementary
particle physicists tell us about the nature of the fundamental laws, the less
relevance they seem to have to the very real problems of the rest of science,
much less to those of society. The constructionist hypothesis breaks down
when confronted with the twin difficulties of scale and complexity. The behav-
ior of large and complex aggregates of elementary particles, it turns out, is not
to be understood in terms of a simple extrapolation of the properties of a few
particles. Instead, at each level of complexity entirely new properties appear,
and the understanding of the new behaviors requires research which I think is
as fundamental in its nature as any other. That is, it seems to me that one may
array the sciences roughly linearly in a hierarchy, according to the idea: The
elementary entities of science X obey the laws of science Y.
But this hierarchy does not imply that science X is "just applied Y." At each
stage entirely new laws, concepts, and generalizations are necessary, requiring
inspiration and creativity to just as great a degree as in the previous one.
Psychology is not applied biology, nor is biology applied chemistry...
The arrogance of the particle physicist and his intensive research may be
behind us (the discoverer of the positron said "the rest is chemistry"), but
we have yet to recover from that of some molecular biologists, who seem de-
termined to try to reduce everything about the human organism to "only"
chemistry, from the common cold and all mental disease to the religious in-
stinct. Surely there are more levels of organization between human ethology
and DNA than there are between DNA and quantum electrodynamics, and
each level can require a whole new conceptual structure.
In closing, I offer two examples from economics of what I hope to have said.
Marx said that quantitative differences become qualitative ones, but a dialogue
in Paris in the 1920’s sums it up even more clearly:
FITZGERALD: The rich are different from us.
HEMINGWAY: Yes, they have more money.

Electronic copy available at: https://ssrn.com/abstract=4668578


Part I

Introduction

Electronic copy available at: https://ssrn.com/abstract=4668578


1
A Benchmark Two-Country Economy

Summary

• We develop a benchmark two-country model of international real business cycles. Restrictions


on the exchange rate et are imposed by both goods market conditions:

pt α c F,t 1 − α c F,t
= = ,
p∗t exp(−et ) 1 − α c H,t α c∗H,t

and asset market conditions:

1 = Et [exp(mt+1 − ∆et+1 + rt∗ )] = Et exp(m∗t+1 + ∆et+1 + rt ) .


 

• Trading in the risk-free bond market implies that the currency’s expected return is determined
by its risk premium:

1
Et [rxt+1 ] = −covt (m∗t+1 , ∆et+1 ) − vart (∆et+1 ),
2
and that the exchange rate level is determined by the expectation of future interest rates and
currency risk premia:
∞ ∞
et = ∑ Et [rt+ j − rt∗+ j ] − ∑ Et [rpt+ j ] + ē.
j =0 j =0

• When markets are complete, we can further simplify the exchange rate movement and cur-
rency expected return as

∆et+1 = mt+1 − m∗t+1 ,


1 1
Et [rxt+1 ] = vart (m∗t+1 ) − vart (mt+1 ).
2 2

We start with a benchmark economy with international trade in


goods and bonds, which is the basis for extensions in later chapters.
We characterize the dynamics of the exchange rate and the currency
return in this model, and, in doing so, showcase some standard tech-

Electronic copy available at: https://ssrn.com/abstract=4668578


a benchmark two-country economy 15

niques.

1.A Model Set-up

1.A.1 Households
We consider an endowment economy with two countries, home and
foreign. Each country has a continuum of identical households.
Equivalently, we can think of a representative household in each
country. Each country has a unique type of goods, labeled as home
goods and foreign goods. The home households receive an endow-
ment of yt units of the home goods, and the foreign households
receive an endowment of y∗t units of the foreign goods.
Home and foreign households derive utility from consuming both
countries’ goods. Let c H denote the home households’ consumption
of the home goods, and c F denote the home households’ consump-
tion of the foreign goods. The home households’ aggregate consump-
tion is a Cobb-Douglas aggregation of the home and foreign goods:

ct = (c H,t )α (c F,t )1−α ,

which means that c H,t units of the home goods and c F,t units of the
foreign goods can be combined to form ct units of the home consump-
tion bundle, which is the unit of account for the home households’
aggregate consumption. In our analysis, we carefully distinguish
between the home consumption bundle and the home goods. Simi-
larly, foreign households also combine the home and foreign goods
into the foreign consumption bundle, which we will specify later. Some
papers also refer to the home and foreign goods as the intermediate
goods, and the home and foreign consumption bundles as the final
goods. These terms will carry more meaning when we discuss the
production side of the economy such as in Section 3.C.
The expected lifetime utility for home households is

" #
E0 ∑ δt u(ct ) ,
t =0

where u(ct ) is a generic utility function that takes the home house-
holds’ aggregate consumption ct as the argument, and δ is the sub-
jective discount factor. We usually impose some regularity conditions
on the utility function, so that it is twice continuously differentiable,
increasing in ct , and concave.
The log real exchange rate et is defined as the log exchange ratio
between the home consumption bundle and the foreign consumption
bundle. That is, to afford 1 unit of the home consumption, we need
to spend exp(et ) units of the foreign consumption. Our convention

Electronic copy available at: https://ssrn.com/abstract=4668578


16 jiang – lecture notes on international finance

is such that et increases when the home currency appreciates in real


terms, which means the home consumption bundle becomes more
expensive.
The home and foreign risk-free bonds are denominated in the
units of their respective consumption bundles. For now, we focus
on the one-period bond only: 1 unit of the bond issued in period t
promises to pay out 1 unit of consumption bundle in the next period
t + 1. The home bond’s interest rate is rt , implying that the price for
1 unit of the bond is exp(−rt ) units of home consumption bundles in
period t. Similarly, the foreign bond’s interest rate is rt∗ , implying that
the price for 1 unit of the bond is exp(−rt∗ ) units of foreign consump-
tion bundles in period t. The home and foreign households can freely
trade both types of bonds.
Using the home consumption bundle as the numéraire, we can
express the home households’ budget constraint as

pt yt + b H,t−1 exp(rt−1 ) + bF,t−1 exp(rt∗−1 − et ) = ct + b H,t + bF,t exp(−et ),


(1.1)

where yt denotes the endowment of the home goods, pt denotes the


price of the home goods in the unit of the home consumption bundle,
b H,t denotes the market value of the home risk-free bond held by the
home households, and bF,t denotes the market value of the foreign
risk-free bond held by the home households. In this equation, the
left-hand side is the home households’ income from the endowment
in period t plus their savings in the bond market from period t − 1,
and the right-hand side is the home households’ expenditure for the
consumption in period t and their investments in the bond market
for consumption in future periods.
We can introduce other tradable assets such as equities and long-
term bonds by extending this budget constraint. A particularly useful
case is when the households can trade the complete set of contingent
claims. We will consider this complete-market case in Section 1.C.

1.A.2 Intertemporal Solution


We write the home households’ Lagrangian as
∞ ∞
" #
E0 ∑ δt u(ct ) + ∑ ζ t ( pt yt + bH,t−1 exp(rt−1 ) + bF,t−1 exp(rt∗−1 − et ) − ct − bH,t − bF,t exp(−et )) .
t =1 t =1

The first-order conditions w.r.t. ct , b H,t , and bF,t are

δt u′ (ct ) − ζ t = 0,
Et [−ζ t + ζ t+1 exp(rt )] = 0,
Et [−ζ t exp(−et ) + ζ t+1 exp(rt∗ − et+1 )] = 0.

Electronic copy available at: https://ssrn.com/abstract=4668578


a benchmark two-country economy 17

These equations imply the following Euler equations:


 ′ 
u ( c t +1 )
1 = Et δ ′ exp(rt ) ,
u (ct )
 ′ 
u (c )
1 = Et δ ′ t+1 exp(−∆et+1 + rt∗ ) ,
u (ct )

which describe how the households trade off consumption and sav-
ing intertemporally. For example, we can express the first Euler equa-
tion as

u′ (ct ) = Et δu′ (ct+1 ) exp(rt ) .


 

The left-hand side represents the increase in utility from consuming


a small amount of the home consumption bundle in period t, which
is u′ (ct ) in the limit. The right-hand side represents the expected
increase in the utility from saving this additional consumption bun-
dle in home bonds, earning interests and receiving exp(rt ) in period
t + 1, and deriving u′ (ct+1 ) utils from each additional unit of con-
sumption in period t + 1. The households discount these future utils
by the subjective discount rate δ, and equalize the left- and right-
hand sides so that they are indifferent between these two options.
We denote the marginal utility growth as

def u ′ ( c t +1 )
exp(mt+1 ) = δ ,
u′ (ct )

which is commonly referred to as the stochastic discount factor (SDF)


or the pricing kernel in the asset pricing literature. In this note, we
will focus on the relationship between the SDF and the exchange
rate. That said, if the same households can also trade the claim to the
endowment, then, the same SDF also prices the endowment claim.
def
Let mt,t+k = ∑kj=1 mt+ j denote the cumulative SDF from period t to
t + k. Then, the price of the home country’s endowment claim is


" #
pm
t = Et ∑ exp(mt,t+k ) pt+k yt+k .
k =1

Similarly, if the claim to the households’ consumption stream is trad-


able, we can also price the households’ wealth as

" #
pw
t = Et ∑ exp(mt,t+k )ct+k .
k =1

1.A.3 Within-Period Solution


Let p∗t denote the price of the foreign goods in the unit of the foreign
consumption bundle. Using the home consumption bundle as the

Electronic copy available at: https://ssrn.com/abstract=4668578


18 jiang – lecture notes on international finance

numéraire, the price of the home consumption bundle is 1. Then, the


law of one price implies

1 · (c H,t )α (c F,t )1−α = 1 · ct = pt c H,t + p∗t c F,t exp(−et ). (1.2)

Substitute this equation into the Lagrangian:


"
E0 ∑ δt u((c H,t )α (cF,t )1−α )
t =1

#
+ ∑ ζ t ( pt yt + bH,t−1 exp(rt−1 ) + bF,t−1 exp(rt∗−1 − et ) − pt c H,t − exp(−et ) p∗t cF,t − bH,t − bF,t exp(−et )) .
t =1

The first-order conditions w.r.t. c H,t , c F,t are


c F,t 1−α
δt u′ (ct )α( ) − ζ t pt = 0,
c H,t
c H,t α
δt u′ (ct )(1 − α)( ) − ζ t p∗t exp(−et ) = 0,
c F,t

which implies that the consumption ratio c F,t /c H,t between foreign
and home goods is a function of the relative prices:
pt α c F,t
∗ = . (1.3)
pt exp(−et ) 1 − α c H,t

Moreover, plug this solution into (1.2), we obtain

1−α 1
ct = pt c H,t + pt c H,t = pt c H,t ,
α α
α 1
ct = p∗t c F,t exp(−et ) + p∗t c F,t exp(−et ) = p∗ c F,t exp(−et ),
1−α 1−α t
which implies that the expenditure shares for home and foreign
goods are constant under the Cobb-Douglas aggregator and a general
utility function:

pt c H,t = αct , (1.4)



pt c F,t exp(−et ) = (1 − α ) c t . (1.5)

Moreover, plug in (c H,t )α (c F,t )1−α = ct , we can express the goods’


prices (i.e., pt and p∗t exp(−et )) in the unit of the home consumption
bundle as functions of the relative consumption weights:
  1− α  α
c F,t c H,t
pt = α , p∗t exp(−et ) = (1 − α) .
c H,t c F,t

A commonly used notion is the terms of trade, defined as the


ratio between the price index of exported goods and the price index
of imported goods from the perspective of a given country. In our
setting, the home country exports home goods and imports foreign

Electronic copy available at: https://ssrn.com/abstract=4668578


a benchmark two-country economy 19

goods. As such, the terms of trade are simply the ratio of prices
between home and foreign goods:
def pt
exp(tott ) = ∗ ,
pt exp(−et )

which allows us to express goods’ prices in the unit of the home


consumption bundle as
 1− α −α
1−α 1−α
 
pt = α exp(tott ) , p∗t exp(−et ) = (1 − α ) exp(tott ) .
α α

Likewise, we can derive the same problem from the foreign house-
holds’ perspective, and obtain
1− α −α
1−α 1−α
 
p∗t = α exp(−tott ) , pt exp(et ) = (1 − α) exp(−tott ) ,
α α

which implies that the real exchange rate and the terms of trade are
tightly connected:

et = (2α − 1)tott .

So, we can simplify Eq. (1.3) to


 
et α c F,t
exp = .
2α − 1 1 − α c H,t

That is, within a period, the households choose between consuming


home and foreign goods by comparing the relative prices as captured
by the real exchange rate et and the relative consumption weights α
and 1 − α.

1.A.4 Foreign Country


For foreign households, symmetrically, we define the foreign con-
sumption bundle as

c∗t = (c∗F )α (c∗H )1−α .

Given α > 1/2, the foreign consumption bundle leans towards the
foreign goods, whereas the home consumption bundle leans towards
the home goods.
The foreign households’ budget constraint is

p∗t y∗t + b∗H,t−1 exp(rt−1 + et ) + b∗F,t−1 exp(rt∗−1 ) = c∗t + b∗H,t exp(et ) + b∗F,t

The Lagrangian is
∞ ∞
" #
E0 ∑δ t
u(c∗t ) + ∑ ζ t∗ ( p∗t y∗t + exp(rt−1 + et )b∗H,t−1 + b∗F,t−1 exp(rt∗−1 ) − c∗t − b∗H,t exp(et ) − b∗F,t ) .
t =1 t =1

Electronic copy available at: https://ssrn.com/abstract=4668578


20 jiang – lecture notes on international finance

The intertemporal solution implies


" #
u′ (c∗t+1 )
1 = Et δ ′ ∗ exp(rt + ∆et+1 ) ,
u (ct )
" #
u′ (c∗t+1 ) ∗
1 = Et δ ′ ∗ exp(rt ) ,
u (ct )
and the within-period solution implies

pt exp(et ) 1 − α c F,t
∗ = ∗ ,
pt α c H,t

c∗F,t
pt exp(et ) = (1 − α) ,
c∗H,t
!1− α
c∗H,t
p∗t =α .
c∗F,t

1.A.5 Market Clearing


In the goods market, the endowment is equal to the sum of home
and foreign consumption:
yt = c H,t + c∗H,t ,
y∗t = c F,t + c∗F,t .
In the bonds market, the bonds are in zero net supply:
0 = b H,t + b∗H,t ,
0 = bF,t + b∗F,t .

1.A.6 Macro Synthesis


Throughout this book, we study the competitive equilibrium defined
in the usual fashion: all households maximize their utilities taking
goods prices, asset prices, and exchange rates as given, and the mar-
kets for goods and assets are cleared.
The only exogenous variables are the endowments:
(yt , y∗t )∞
t =0 .

There are 15 endogenous variables in each period t:


(ct , c H,t , c F,t , b H,t , bF,t , pt , c∗t , c∗H,t , c∗F,t , b∗H,t , b∗F,t , p∗t , rt , rt∗ , et )∞
t =0 ,

plus two auxiliary variables exp(mt+1 ) and exp(m∗t+1 ) that denote the
home and foreign SDFs:
u ′ ( c t +1 )
def
exp(mt+1 ) = δ ,
u′ (ct )
′ ∗ )
def u ( c
exp(m∗t+1 ) = δ ′ t+∗ 1 .
u (ct )

Electronic copy available at: https://ssrn.com/abstract=4668578


a benchmark two-country economy 21

The model implies the following 16 equations in each period, one


of which is redundant because the market clearing conditions add
up to the sum of households’ budget constraints. These 16 equations
include 2 consumption aggregation equations,

ct = (c H,t )α (c F,t )1−α ,


c∗t = (c∗F,t )α (c∗H,t )1−α ,

4 household budget constraints,

pt yt + b H,t−1 exp(rt−1 ) + bF,t−1 exp(rt∗−1 − et ) = ct + b H,t + bF,t exp(−et ),


ct = pt c H,t + p∗t c F,t exp(−et ),
p∗t y∗t + b∗H,t−1 exp(rt−1 + et ) + b∗F,t−1 exp(rt∗−1 ) = c∗t + b∗H,t exp(et ) + b∗F,t ,
c∗t = pt c∗H,t exp(et ) + p∗t c∗F,t ,

2 equations describing the households’ within-period consumption


choices,

pt α c F,t 1 − α c F,t
= = ,
p∗t exp(−et ) 1 − α c H,t α c∗H,t

2 goods market clearing conditions,

yt = c H,t + c∗H,t ,
y∗t = c F,t + c∗F,t ,

2 bond market clearing conditions,

0 = b H,t + b∗H,t ,
0 = bF,t + b∗F,t ,

and 4 Euler equations,

1 = Et [exp(mt+1 ) exp(rt )] ,
1 = Et [exp(mt+1 ) exp(−∆et+1 + rt∗ )] ,
1 = Et exp(m∗t+1 ) exp(rt∗ ) ,
 

1 = Et exp(m∗t+1 ) exp(∆et+1 + rt ) .
 

1.B Exchange Rate Accounting

Among the equilibrium conditions we derive in the last section, par-


ticularly relevant for asset pricing are the four Euler equations:

1 = Et [exp(mt+1 + rt )] , (1.6)
1= Et [exp(mt+1 − ∆et+1 + rt∗ )] , (1.7)
Et exp(m∗t+1 + rt∗ ) ,
 
1= (1.8)
Et exp(m∗t+1 + ∆et+1 + rt ) .
 
1= (1.9)

Electronic copy available at: https://ssrn.com/abstract=4668578


22 jiang – lecture notes on international finance

For the discussion in this section, we assume the random variables


are jointly normally distributed. It is possible to extend this analysis
to the non-normal case using co-entropy instead of covariance. Un-
der joint normality, the foreign households’ Euler equations can be
expressed as
1
0 = Et [m∗t+1 ] + vart (m∗t+1 ) + rt∗ ,
2
1 1
0 = Et [m∗t+1 ] + vart (m∗t+1 ) + Et [∆et+1 ] + vart (∆et+1 ) + covt (m∗t+1 , ∆et+1 ) + rt .
2 2

1.B.1 Accounting for the Currency Expected Return


To capture the relative performance of the home and foreign curren-
cies, we define the log excess return of the home currency against the
foreign currency as

rxt+1 = ∆et+1 + rt − rt∗ .

This excess return captures the return of the strategy that takes a
long position on the home bond and a short position on the foreign
bond, which is exposed to the bilateral exchange rate movement.
Then, the Euler equations imply the following result:

Proposition 1.1. The home currency’s expected log excess return is deter-
mined by the covariance between the log foreign SDF and log exchange rate
movement minus a Jensen’s term:
1
Et [rxt+1 ] = −covt (m∗t+1 , ∆et+1 ) − vart (∆et+1 ). (1.10)
2
We interpret the right-hand side of Eq. (1.10) as the currency risk
premium in log, since the covariance term describes how the exchange
rate comoves with the foreign investors’ SDF. If the covariance is
positive, the home currency tends to appreciate when the foreign
households’ marginal utility is high. Then, the home currency is
a good hedge from the perspective of the foreign households and
should earn a low risk premium.
The right-hand side of Eq. (1.10) also has a variance term − 12 vart (∆et+1 )
that we refer to as the Jensen’s term. To understand this term, it is
useful to consider the risk premium expression for the level of the
currency return:

log Et [exp(rxt+1 )] = −covt (m∗t+1 , ∆et+1 ). (1.11)

Compared with the currency risk premium in log defined in Eq.


(1.10), the currency risk premium in level conveniently has no Jensen’s
term. Empirically, for developed countries’ currencies, the exchange
rate volatility is roughly 10% per annum. This magnitude implies a

Electronic copy available at: https://ssrn.com/abstract=4668578


a benchmark two-country economy 23

Jensen’s term of − 12 vart (∆et+1 ) = −0.5%, which is an order of mag-


nitude smaller than the covariance term for many currency pairs we
usually consider.
We can also derive the currency risk premium from the home
households’ perspective:
1
Et [−rxt+1 ] = −covt (mt+1 , −∆et+1 ) − vart (∆et+1 ),
2
log Et [exp(−rxt+1 )] = −covt (mt+1 , −∆et+1 ),

which implies that the foreign currency has to offer a higher risk
premium if it tends to depreciate when the home SDF is high.
If we combine the expected return expressions from the home and
foreign households’ perspectives, we obtain the following expression:

Et [rxt+1 ] = −Et [−rxt+1 ]


1 1
−covt (m∗t+1 , ∆et+1 ) − vart (∆et+1 ) = covt (mt+1 , −∆et+1 ) + vart (∆et+1 ).
2 2
(1.12)

Recall that all these restrictions are derived from the Euler equations
for holding the risk-free bonds. So, allowing the households to freely
trade in the risk-free bond markets imposes restrictions not only
between the currency expected return and the covariance between
the SDF and the exchange rate movement from each country’s per-
spective, but also between the home and the foreign perspectives.
Specifically, the home and foreign investors need to agree on the
equilibrium currency risk premium after the second-order adjustment
by the Jensen’s term.

1.B.2 Accounting for the Exchange Rate Level


Let us denote the home currency’s risk premium (including the
Jensen’s term) as
def 1
rpt = Et [rxt+1 ] = −covt (m∗t+1 , ∆et+1 ) − vart (∆et+1 ),
2
which implies

et − Et [et+1 ] = rt − rt∗ − rpt . (1.13)

If we regard the next period’s expected exchange rate level Et [et+1 ]


as a reference point, then, Eq. (1.13) implies that the deviation of
the current exchange rate from this reference point is determined
by the interest rate differential and the currency risk premium. The
home currency is stronger relative to the reference point if the home
interest rate is higher than the foreign interest rate and if the home
currency’s risk premium is low.

Electronic copy available at: https://ssrn.com/abstract=4668578


24 jiang – lecture notes on international finance

We can generalize this intuition under the assumption that the


real exchange rate is stationary. In this case, there exists a long-run
exchange rate level,
def
ē = lim Et [et+ j ].
j→∞

Using this as a more natural reference point for the exchange rate
level, we can decompose today’s exchange rate level in the following
way [Campbell and Clarida, 1987, Froot and Ramadorai, 2005]:

Proposition 1.2. The exchange rate level is equal to the sum of expected
future interest rate differentials, the sum of expected future currency risk
premia, and the long-run exchange rate level:
∞ ∞
et = ∑ Et [rt+ j − rt∗+ j ] − ∑ Et [rpt+ j ] + ē. (1.14)
j =0 j =0

This formula shows that, if the home currency is currently stronger


than its long-run mean, it is either because the home currency is of-
fering a higher interest rate than the foreign currency, or because it
has a lower risk premium. This decomposition of the exchange rate
level is similar to the Campbell and Shiller [1988] decomposition of
the equity price-dividend ratio into a cash flow component, a dis-
count rate component, and a vanishing transversality component,
i.e.,
∞ ∞
pdt = ∑ κ j−1 Et [∆ log dt+ j ] − ∑ κ j−1 Et [rt+ j ] + jlim
→∞
κ j Et [ pdt+ j ].
j =1 j =1

In the exchange rate formula (1.14), the interest rate term ∑∞ j =0 Et [ r t + j −


rt∗+ j ]
can be interpreted as the cash flow component: if the home
bond earns a higher interest rate, then, the investors are indeed ex-
pecting higher cash flows from holding it, and should therefore im-
pute a higher valuation to the home currency. Since the exchange
rate is the relative price between the home and foreign countries, the
interest rate differential rt+ j − rt∗+ j enters the formula to capture the
relative magnitude of the cash flows. Likewise, the risk premium
term ∑∞ j=0 Et [rpt+ j ] can be interpreted as the discount rate compo-
nent: if the home currency is risky and therefore earns a higher risk
premium, then, the investors should use a higher discount rate for
the home currency and impute a lower valuation. Finally, the long-
run exchange rate level ē can be interpreted as the transversality
component. The key difference between the exchange rate formula
(1.14) and the equity pricing formula above is that there is no dis-
counting by κ in the exchange rate formula, which follows from the
fact that no approximation was made in deriving the exchange rate
formula.

Electronic copy available at: https://ssrn.com/abstract=4668578


a benchmark two-country economy 25

Moreover, it is simple to go one step further to decompose the


exchange rate innovation:

∞ ∞
(Et − Et−1 )[et ] = ∑ (Et − Et−1 )[rt+ j − rt∗+ j ] − ∑ (Et − Et−1 )[rpt+ j ],
j =0 j =0

which offers a decomposition formula for the volatility of the unex-


pected exchange rate movement.
This discussion makes it clear that the exchange rate is a forward-
looking variable, which incorporates information about future in-
terest rates and risk premia. As a result, if some macro or financial
variables capture variations in interest rates or risk premia, the ex-
change rate should be able to predict these variables [Engel and West,
2005].

1.C Complete-Market Solution

When the markets are complete, households in both countries can


trade any Arrow-Debreu securities (a.k.a. contingent claims). Then,
the households’ budget constraints, such as Eq. (1.1), should be re-
vised to reflect these investment opportunities.
There is a more convenient approach to solving this case. By the
First Welfare Theorem, the equilibrium outcome is observationally
equivalent to the equilibrium outcome under a social planner, who
maximizes a weighted sum of the households’ welfare [Negishi,
1960]:


" #
E0 ∑δ t
(πu(ct ) + (1 − π )u(c∗t )) ,
t =0

where the weight π is known as the Pareto weight and is endoge-


nously determined by the households’ initial wealth.1 1
The allocation problem in the compet-
The social planner tells the home and foreign households how itive economy can be formulated as a
fictitious social planner’s optimization
much to consume, subject to the resource constraints problem even when the markets are in-
complete [Cuoco and He, 1994]. In this
case, the Pareto weight π is stochas-
yt = c H,t + c∗H,t , tic and related to different agents’
marginal utilities. We will consider one
y∗t = c F,t + c∗F,t .
such setting in Chapter 5.

The social planner’s Lagrangian is

∞ ∞ ∞
" #
E0 ∑δ t
(πu(ct ) + (1 − π )u(c∗t )) + ∑ ζ H,t (yt − c H,t − c∗H,t ) + ∑ ζ F,t (y∗t − c F,t − c∗F,t ) ,
t =1 t =1 t =1

Electronic copy available at: https://ssrn.com/abstract=4668578


26 jiang – lecture notes on international finance

which implies the following first-order conditions


 1− α
c F,t
w.r.t. c H,t : δt πu′ (ct )α = ζ H,t ,
c H,t

c∗F,t
w.r.t. c∗H,t : t
δ (1 − π ) u ′
(c∗t )(1 − α) = ζ H,t ,
c∗H,t
 α
c H,t
w.r.t. c F,t : δt πu′ (ct )(1 − α) = ζ F,t ,
c F,t
!1− α
c∗H,t
w.r.t. c∗F,t : δt (1 − π )u′ (c∗t )α = ζ F,t .
c∗F,t

Plug in the within-period solution in Section 1.A, and we obtain


the following equilibrium condition:

πu′ (ct ) = (1 − π )u′ (c∗t ) exp(et ), (1.15)

which describes the social planner’s optimal risk-sharing rule: it sets


the allocation so that the home households’ marginal utility, weighed
by its Pareto weight π, is equal to the foreign households’ marginal
utility, weighed by its Pareto weight (1 − π ) and adjusted by the real
exchange rate exp(et ). The real exchange rate converts the marginal
utility from the foreign consumption unit to the home consumption
unit.
This equilibrium condition implies a linear allocation rule.

Proposition 1.3. Equilibrium consumption is a linear function of endow-


ments. Specifically, for some value k t , the consumption of home and foreign
goods is
αk t 1−α
c H,t = yt , c∗H,t = yt ,
(1 − α) + αk t (1 − α) + αk t
(1 − α ) k t α
c F,t = y∗ , c∗F,t = y∗ ,
α + (1 − α ) k t t α + (1 − α ) k t t
and the aggregate consumption is

α α (1 − α )1− α
ct = α
k yα (y∗t )1−α ,
1− α t t
((1 − α) + αk t ) (α + (1 − α)k t )
α α (1 − α )1− α
c∗t = 1− α
y1−α (y∗t )α ,
α t
((1 − α) + αk t ) ( α + (1 − α ) k t )
where k t can be solved via the following implicit equation:


π u′ (ct ) α c F,t 1−α c H,t
 
= 1.
1 − π u′ (c∗t ) 1 − α c H,t c∗F,t

The proof is presented in Appendix A.1. This optimal risk-sharing


rule implies that the equilibrium allocation of home and foreign

Electronic copy available at: https://ssrn.com/abstract=4668578


a benchmark two-country economy 27

goods to home and foreign households is driven by two considera-


tions. First, when the home bias in consumption α is greater, more
home goods are allocated to home households and more foreign
goods are allocated to foreign households. Second, an endogenous,
stochastic variable k t further adjusts the shares of the aggregate en-
dowment of home and foreign goods that are allocated to the home
households as opposed to the foreign households. When k t is higher,
the home households consume more home and foreign goods relative
to the foreign households:

c H,t α c F,t 1−α


∗ = kt , ∗ = kt .
c H,t 1−α c F,t α

Once we solve k t , we can solve the equilibrium allocations without


solving for the goods’ and the assets’ prices in the competitive equi-
librium.
We illustrate the equilibrium allocation in a simple numerical ex-
ample. Suppose the home and foreign households have identical
CRRA preferences: u(c) = c1−γ /(1 − γ), with γ = 2, α = 0.7, and
π = 0.5. The endowment yt is stochastic and y∗t is fixed at 2. Fig-
ure 1.1 traces out the equilibrium consumption allocations ct and c∗t
for the home and foreign households as we vary yt . We can see that,
as the home endowments become higher, both countries’ consump-
tion increases. The fact that the foreign households’ consumption
responds to the home endowment shocks reflects international risk-
sharing. Moreover, the home country’s consumption increases more
than the foreign country’s consumption: the hypothetical social plan-
ner assigns a greater share of the increase in the quantity of home
goods to the home households because they derive more utility from
consuming home goods due to home bias.
The international risk-sharing can be further illustrated by the
international transfer in equilibrium. Figure 1.2 plots the net interna-
tional transfers of home and foreign households as fractions of their
respective endowments. When the home households receive a low
endowment yt , the social planner transfers resources from the foreign
country to the home country to balance their marginal utilities. As
a result, the home households receive a net transfer paid out by the
foreign households.

1.C.1 Implied Exchange Rate Dynamics

Using the households’ optimality conditions that we derived under


the competitive equilibrium, we can back out the prices from quan-
tities. In particular, the optimal risk-sharing rule Eq. (1.15) implies a

Electronic copy available at: https://ssrn.com/abstract=4668578


28 jiang – lecture notes on international finance

Figure 1.1: Equilibrium Consumption


Allocation.
1.8

1.6

1.4

1.2

0.8

0.6

0.4

0.2

0
0 0.5 1 1.5 2 2.5 3 3.5 4

Figure 1.2: Equilibrium International


Transfer.
0.4

0.3

0.2

0.1

-0.1

-0.2

-0.3
0 0.5 1 1.5 2 2.5 3 3.5 4

relationship between the real exchange rate and marginal utilities:


π
et = log u′ (ct ) − log u′ (c∗t ) + log ,
1−π
which can be written as

∆et+1 = mt+1 − m∗t+1 . (1.16)

This condition indicates that the bilateral exchange rate movement


is solely determined by the difference between the two countries’
marginal utility growth when the markets are complete. Specifically,
when the home households have a higher marginal utility growth
mt+1 , which usually indicates a recession in the home country, the

Electronic copy available at: https://ssrn.com/abstract=4668578


a benchmark two-country economy 29

home currency becomes stronger in real terms. Intuitively, this is


when the home households become less willing to hold the foreign
currency.
Moreover, the expected excess return on the home currency can be
expressed as

1 1 1
Et [rxt+1 ] = −covt (m∗t+1 , ∆et+1 ) − vart (∆et+1 ) = vart (m∗t+1 ) − vart (mt+1 ),
2 2 2
(1.17)

which implies that the home currency’s risk premium is decreasing


in home SDF volatility and increasing in foreign SDF volatility. As
a result, to generate reasonable variations in either conditional or
unconditional currency expected returns, the variances of the SDFs
play a central role.
Figure 1.3 plots the equilibrium real exchange rate et , which mea-
sures the strength of the home currency, as we vary the home endow-
ment yt . In states with a low home endowment, the home goods are
relatively scarce, leading to stronger terms of trade and a real appre-
ciation of the home currency. This expenditure switching effect arises
in a large class of international models, which describes how supply
shocks affect the exchange rate determination. As we will see in later
sections, this supply effect may be offset by demand shocks.
Another way to understand this relationship is via international
risk-sharing: because of the home bias in consumption, the home
households’ consumption loads more on the home endowment.
When the home endowment is low, the home households’ marginal
utility is higher than the foreign households’, and, through the opti-
mal international risk-sharing terms described by Eq. (1.16), requires
a real appreciation of home currency to equilibrate their demand for
financial assets.

1.C.2 Special Case: Log Utility


We consider a special case in which home and foreign households
have the log utility, i.e. u(c) = log(c). In this case, the parameter k is
a constant,
π
kt = k = ,
1−π
and Proposition 1.3 implies a simple rule for consumption alloca-
tions:

απ (1 − α)(1 − π )
c H,t = yt , c∗H,t = yt ,
απ + (1 − α)(1 − π ) απ + (1 − α)(1 − π )
(1 − α ) π α (1 − π )
c F,t = y∗ , c∗F,t = y∗ .
(1 − α ) π + α (1 − π ) t (1 − α ) π + α (1 − π ) t

Electronic copy available at: https://ssrn.com/abstract=4668578


30 jiang – lecture notes on international finance

Figure 1.3: Equilibrium Exchange Rate.

2.5

1.5

0.5

-0.5
0 0.5 1 1.5 2 2.5 3 3.5 4

This allocation rule has an intuitive interpretation: home and for-


eign households agree to split the endowments according to their
Pareto weights π and 1 − π and their home bias α. If the home
households have a higher Pareto weight (i.e., π > 0.5), then, the
home households always receive more allocations. Moreover, regard-
less of the Pareto weights, home households always receive more
home goods, and foreign households always receive more foreign
goods due to their home bias in consumption (i.e., α > 0.5). As such,
while this allocation rule allows the households in both countries to
share their risks internationally,2 they may still have different expo- 2
While this statement is true in general,
sures to the home and foreign endowment shocks due to their home in this case of log preference, the terms-
of-trade responses alone provide perfect
bias in consumption. insurance against output shocks. As
Since the markets are complete, this allocation can be implemented a result, the allocation under financial
autarky is not far worse off. See [Cole
by a number of financial contracts. For example, the home house- and Obstfeld, 1991] for a detailed
holds hold απ +(1−απα)(1−π ) unit of the home endowment claim and discussion.
(1− α ) π
unit of the foreign endowment claim, and the foreign
(1− α ) π + α (1− π )
households hold the remaining fractions. In the symmetric case of
π = 1/2 and α > 1/2, the home households hold more than half of
the home endowment claim, and less than half of the foreign endow-
ment claim. Therefore, home bias in consumption preference begets
home bias in international portfolios. We will discuss the imple-
mentation of the complete-market allocation and the corresponding
portfolio shares in an example with richer preferences in Section 7.A.
Given these equilibrium consumption allocations, the home and

Electronic copy available at: https://ssrn.com/abstract=4668578


a benchmark two-country economy 31

foreign SDFs can be expressed as

mt+1 = log δ − ∆ log ct+1 = log δ − α∆ log yt+1 − (1 − α)∆ log y∗t+1 ,
m∗t+1 = log δ − ∆ log c∗t+1 = log δ − (1 − α)∆ log yt+1 − α∆ log y∗t+1 ,

and the exchange rate movement is driven by the fundamental


shocks to the endowments:

∆et+1 = mt+1 − m∗t+1 = (2α − 1)(∆ log y∗t+1 − ∆ log yt+1 ).

This exchange rate expression again highlights the result in Fig-


ure 1.3. Under the social planner’s optimal allocation scheme, the
home households’ allocation is exposed to both home and foreign
endowment shocks, with a higher exposure to the home endowment
shock. As a result, a negative home endowment shock raises the
home households’ marginal utility relative to the foreign households’,
and leads to a real appreciation of the home currency.

1.D Asset Market and Goods Market Views of Exchange Rates

In this section, we re-examine the two sets of equilibrium conditions


that characterize the exchange rate.

1.D.1 The Asset Market View


First, the Euler equations from the intertemporal solutions, repro-
duced below,

1 = Et [exp(mt+1 + rt )] ,
1 = Et [exp(mt+1 − ∆et+1 + rt∗ )] ,
1 = Et exp(m∗t+1 + rt∗ ) ,
 

1 = Et exp(m∗t+1 + ∆et+1 + rt ) ,
 

impose restrictions on the exchange rate dynamics. By relating the


exchange rate movement to the SDFs and interest rates, they rep-
resent the asset market view of exchange rates. As we discussed in
Section 1.B, these restrictions also characterize the exchange rate level
and the currency expected return.
In the complete-market case we discussed in Section 1.C, these
restrictions can be further simplified. The exchange rate movement is
equal to the SDF differential:

∆et+1 = mt+1 − m∗t+1 = ∆ log u′ (ct+1 ) − ∆ log u′ (c∗t+1 ),

which links the exchange rate movement directly to marginal utility


growth. In this case, the real exchange rate movement describes how

Electronic copy available at: https://ssrn.com/abstract=4668578


32 jiang – lecture notes on international finance

the households trade off marginal utilities across time. Specifically,


the hypothetical social planner equates the home and foreign house-
holds’ marginal utility growth rates after converting their numéraires
to the same consumption bundle.
Under this view, to make sense of the exchange rate dynamics,
our task is to understand the underlying economic structures, pref-
erences, and frictions that give rise to the SDF movements as well as
possible deviations from the perfect risk-sharing condition.

1.D.2 The Goods Market View


Next, we consider the within-period consumption choices, repro-
duced below,
α p∗t c F,t
exp(et ) = ,
1 − α pt c H,t

α pt c H,t
exp(−et ) = .
1 − α p∗t c∗F,t

More generally, if we use u(c H,t , c F,t ) and u∗ (c∗F,t , c∗H,t ) to denote the
home and foreign households’ utilities as functions of their goods-
specific consumption, we can express the within-period solutions
as
∂u(c H,t , c F,t )/∂c H,t 2α−1
 
exp(et ) = ,
∂u(c H,t , c F,t )/∂c F,t
!2α−1
∂u∗ (c∗F,t , c H,t )/∂c∗F,t
exp(−et ) = .
∂u∗ (c∗F,t , c H,t )/∂c∗H,t

This set of equations relates the exchange rate level to how the
households trade off consumption expenditures between different
goods within the same period, which gives rise to the goods market view
of exchange rates. Absent frictions in international trade and con-
sumption, a stronger home currency in real terms must correspond to
a lower consumption share in the home goods. Conversely, if house-
holds experience shocks to how they value home vs. foreign goods,
this preference shock also impacts the equilibrium exchange rate.
Moreover, if the economy is stationary and we log-linearize around
a symmetric steady state with ȳ = ȳ∗ , c̄ = c̄∗ , and ē = 0, we can com-
bine the within-period solutions with the market clearing conditions
in the goods market to obtain the following result:

Proposition 1.4. After a first-order approximation, the relative consump-


tion level between home and foreign households is determined by the equilib-
rium real exchange rate and the relative endowment level:

(2α − 1)(log ct − log c∗t ) = (log yt − log y∗t ) + 2(1 − α) et . (1.18)
2α − 1

Electronic copy available at: https://ssrn.com/abstract=4668578


a benchmark two-country economy 33

The proof is presented in Appendix A.2. This proposition shows


that, holding the endowments fixed, a strong home currency in real
terms (i.e., higher et ) leads to a higher home consumption relative to
foreign consumption, because the home endowment offers the home
households stronger purchasing power.

1.D.3 Harmonizing the Two Views


These two views play fundamental roles in understanding the ex-
change rate dynamics. A fully specified model needs to take a stance
on both aspects. For example, the model we develop in Section 1.A
assumes frictionless trading in both financial assets and goods. A
more sophisticated model can enrich the asset markets, by introduc-
ing richer preferences and shocks to the SDFs as we will discuss in
Chapter 3, by introducing deviations from the Euler equations (1.6)
to (1.9) as we will discuss in Chapters 4 and 10, or by introducing
incomplete asset market spanning as we will discuss in Chapter 5.
The effects of these additional ingredients in the asset markets can
be understood in reduced form as wedges. For example, suppose
the households have power utilities with a relative risk aversion of
γ. Then, the real exchange rate can be written as its complete-market
solution plus an asset market wedge τtAM :

et = −γ(log ct − log c∗t ) + τtAM .

This wedge captures the deviations from this benchmark characteri-


zation of the households’ intertemporal substitution.
Similarly, a more sophisticated model can enrich the goods market
by introducing frictions in the international trade of the goods, such
as invoicing frictions and producer mark-ups, by introducing costs
in the international transportation of the goods, or by introducing
shocks to household preferences over different types of goods.
The real exchange rate can be written as the frictionless goods
market solution (1.18) plus a goods market wedge τtGM :

(2α − 1)(log ct − log c∗t ) = (log yt − log y∗t ) + 2(1 − α) et + τtGM .
2α − 1
This wedge captures the deviations from the within-period consump-
tion allocation in frictionless goods market.
Combining these two equations, we obtain the following approxi-
mate expression for the real exchange rate:
2α − 1 2α − 1 AM
   
2α ∗ GM
+ 2(1 − α ) et = − log yt − log yt − τt + τt ,
γ 2α − 1 γ
(1.19)
which shows that the exchange rate is driven by the benchmark so-
lution in complete financial markets and frictionless goods market,

Electronic copy available at: https://ssrn.com/abstract=4668578


34 jiang – lecture notes on international finance

which depends on the relative endowments (log yt − log y∗t ), plus the
asset market wedge τtAM and the goods market wedge τtGM .
Any model we consider in this note can be characterized by its
stance on the supply of consumption goods and the two wedges.
While the goods market issues are equally important for understand-
ing the international macro and financial outcomes, we focus more
on the asset market view in this note.

Electronic copy available at: https://ssrn.com/abstract=4668578


2
Puzzles: Challenges to Making Sense of Data

Summary

• The stochastic properties of the exchange rate movement—cyclicality, volatility, expected re-
turn, and comovement—exhibit puzzling patterns that are difficult to explain by the complete-
market benchmark model.

• Combining bond and currency forward positions generates non-zero risk-free returns. These
near-arbitrage spreads also require significant modifications on the benchmark model.

• International portfolio quantities and capital flows exhibit strong asymmetry and cyclicality,
which also impose important restrictions on how we specify the model.

2.A Challenges to Making Sense of Exchange Rates

The two-country economy we derived in the previous chapter is


a direct extension of the standard one-country real business cycle
model. When pioneers in this field confront the model’s implications
with the exchange rate data, many puzzles emerge. Each puzzle
sheds light on a dimension of the exchange rate data that is at odds
with the benchmark model, and serves as a useful landmark to guide
us to think about how to make progress in the theoretical literature.
In this section, we build on the complete-market solution derived
from the previous chapter. In particular, we are going to focus on Eq.
(1.16), reproduced below, which relates the equilibrium exchange rate
movement to the SDFs:

∆et+1 = mt+1 − m∗t+1 .

As this lecture note focuses on the theoretical foundation, we only


provide a brief overview of the empirical puzzles. For more com-
prehensive reviews of the empirical literature, see the lecture notes
by Ralph Koijen and Stijn Van Nieuwerburgh1 and the lecture notes 1
https://www.koijen.net/phd-notes-
from Stanford Big-Data Initiative in International Macro-Finance.2 empirical-asset-pricing.html
2
https://www.gsb.stanford.edu/faculty-
research/faculty/conferences/big-
data-initiative-international-macro-
finance/videos-codes

Electronic copy available at: https://ssrn.com/abstract=4668578


36 jiang – lecture notes on international finance

2.A.1 Volatility Puzzle


If we take an unconditional variance on both the left- and right-hand
sides of Eq. (1.16), we obtain

var (∆et+1 ) = var (mt+1 − m∗t+1 ) = var (mt+1 ) + var (m∗t+1 ) − 2cov(mt+1 , m∗t+1 ),

which states that the exchange rate variance should be equal to the
variance of the SDF differential, which can be decomposed to the
sum of SDF variances minus two times the SDF covariance.
By the Hansen and Jagannathan [1991] bound, we can derive a
lower bound on the SDF volatility based on the Sharpe ratio of any
risky asset with return r̃:

E[r̃ − r f ]
std(mt+1 ) ≥ exp(−r f ) .
std(r̃ )

For example, if the stock market has a Sharpe ratio of 0.6 per annum,
then, the SDF’s volatility must be at least roughly 60% per annum.
There may be other trading strategies that produce higher Sharpe
ratios, implying an even higher SDF volatility.
Brandt, Cochrane, and Santa-Clara [2006] compares the implied
SDF volatility with the exchange rate movement’s volatility. Between
developed economies, the exchange rate movement’s volatility is
roughly 10% per annum, which is much lower than the SDF volatility.
Then, if markets are complete, either the exchange rate movement’s
volatility is anomalously low, or the correlation between home and
foreign SDFs is incredibly high, which implies a high degree of inter-
national risk-sharing.
This puzzle triggers two responses in the subsequent literature.
First, many papers stay within the complete-market benchmark but
posit mechanisms that generate a high correlation between home
and foreign SDFs. Chapter 3 develops this idea in detail. Second,
other papers deviate from the complete-market benchmark and study
alternative settings in which Eq. (1.16) do not hold. Chapters 4 and 5
develop some of these ideas in detail.

2.A.2 Exchange Rate Disconnect


Use the definition of the SDFs in Section 1.A, we can rewrite Eq.
(1.16) as

∆et+1 = ∆ log u′ (ct+1 ) − ∆ log u′ (c∗t+1 ),

which implies that the exchange rate movement should be closely


related to the consumption growth or other business cycle variables.
For example, if we assume a power utility, i.e., u(c) = c1−γ /(1 − γ),

Electronic copy available at: https://ssrn.com/abstract=4668578


puzzles: challenges to making sense of data 37

then, the exchange rate movement should be perfectly correlated


with the consumption growth differential between home and for-
eign countries. In other words, the exchange rate has to be strongly
countercyclical with respect to the local consumption growth.
This implication is at odds with the data. Meese and Rogoff [1983],
Backus and Smith [1993] are among the earlier papers that observe a
lack of correlation between exchange rate movements and economic
fundamentals such as the consumption growth3 . Backus and Smith 3
Also see Kollmann [1995], Engel and
[1993] directly compare the exchange rate movement and the con- West [2005].

sumption growth differential between OECD countries, which gives


rise to the Backus-Smith regression that is commonly used by subse-
quent papers:

∆et+1 = α + β∆(c∗t+1 − ct+1 ) + ε t+1 .

While the complete-market benchmark model with power utility


predicts a positive regression coefficient β, empirical estimates of
β are usually around zero or even negative, which implies that the
exchange rate movement is acyclical or even mildly procyclical with
respect to the local consumption growth.
In addition to consumption and fundamental variables, the ex-
change rate movements are not easily spanned by bond returns,
either [Chernov and Creal, 2023].
Attempts to theoretically resolve this disconnect between exchange
rates and economic fundamentals again fall under the two categories
above. First, within the complete-market benchmark, we can posit
mechanisms that generate volatile exchange rates while keeping their
correlation with economic fundamentals low. This can be done by
considering richer preferences or additional (and usually unobserv-
able) drivers of the SDFs, or both. Second, we consider deviations
from the complete-market benchmark, which generate wedges in
the Euler equations that disentangle the exchange rates from the
economic fundamentals that ought to drive the SDFs.

2.A.3 Currency Risk Premium in the Cross-section


Besides exchange rate volatility and correlation with economic funda-
mentals, currency returns also have several salient patterns. Consider
first the unconditional expected returns in the cross-section of curren-
cies. According to the derivation in Section 1.B,
 
def 1
E[rxt+1 ] = E[rpt ] = E −covt (m∗t+1 , ∆et+1 ) − vart (∆et+1 ) .
2

Among developed countries, some currencies like Australian dol-


lar and New Zealand dollar have persistently high excess returns

Electronic copy available at: https://ssrn.com/abstract=4668578


38 jiang – lecture notes on international finance

against the U.S. dollar, whereas some currencies like Japanese Yen
and Swiss Francs have persistently low excess returns against the
U.S. dollar.
Moreover, currencies’ average returns are also correlated with their
interest rates. Low-return currencies like Japanese Yen and Swiss
Francs tend to have low interest rates, while high-return currencies
like Australian dollar and New Zealand dollar tend to have high
interest rates. Let rxti +1 denote the log excess return of currency i
against the dollar. Then, if we regress the realized currency return on
ex-ante interest rates in the cross-section of currencies:

rxti +1 = αit + β t (rti − rt$ ) + εit+1 ,

the slope coefficient β t tends to be positive. As a result, if we sort


currencies into portfolios based on their interest rates and buy high
interest rate currencies against low interest rate currencies, which
may produce more stable differences in currency risk premia [Lustig
and Verdelhan, 2007], we obtain a carry trade portfolio that offers high
average returns.

2.A.4 Currency Risk Premium in the Time Series


The positive association between interest rates and currency returns
applies to not only the cross-section of currencies but also the time
series of a specific currency pair. Fama [1984] first shows this result
using the forward premium, which is the difference between the cur-
rency forward rate and the spot exchange rate. A forward contract
allows investors to lock in the foreign exchange rate at a fixed rate
f ti in the next period. Fama [1984] regresses the currency excess re-
turn on the ex-ante forward premium in the times series of currency i
against the U.S. dollar:

rxti +1 = αi + β( f ti − eit ) + εit+1 ,

and obtain a positive regression coefficient β. This result is known


as the Forward Premium Puzzle, because in models with risk-neutral
households, the expected currency excess return should be zero and
uncorrelated with the forward premium.
When there is no arbitrage, the covered interest rate parity (CIP)
holds (see Section 2.A.6 for more details):

f ti − eit = rti − rt$ . (2.1)

As a result, we obtain very similar results if we regress the currency


excess return on interest rate differential instead of the forward pre-
mium:

rxti +1 = αi + β(rti − rt$ ) + εit+1 ,

Electronic copy available at: https://ssrn.com/abstract=4668578


puzzles: challenges to making sense of data 39

and the slope coefficient β tends to be positive. This result is of-


ten framed as the violation of the uncovered interest rate parity (UIP),
which holds when the interest rate differential does not predict the
currency return:

Et [rxti +1 ] = 0 · (rti − rt$ ).

Equivalently, since the expected excess return is equal to the ex-


change rate movement plus the interest rate differential, the UIP also
implies that the interest rate differential will be exactly offset by the
exchange rate movement:

Et [∆eit+1 ] = −1 · (rti − rt$ ).

As such, it is equivalent to test whether the interest rate differential


predicts future exchange rate movement or whether it predicts future
currency excess return.
A model that seeks to tackle the risk premium puzzles should
be able to explain (1) why there are unconditional differences be-
tween currencies such as the Japanese Yen vs. the Australian dollar,
and (2) why the interest rates vary over time and are correlated with
non-trivial variations in the conditional currency expected returns,
in particular in the U.S. dollar. To explore the relation between the
currency return predictability in the cross-section and the currency
return predictability in the time series, Hassan and Mano [2019] con-
duct a decomposition of currency risk premia into a cross-currency
component, a cross-time component, and a residual component. They
show that the cross-sectional and time-series patterns may be driven
by different forces:

The simplest model that the data do not reject features a cross-sectional asym-
metry that makes some currencies pay permanently higher expected returns
than others, and larger time series variation in expected returns on the U.S.
dollar than on other currencies.

2.A.5 Exchange Rate Comovements


While the exchange rate movements are disconnected from the eco-
nomic fundamentals, they are highly correlated across countries and
exhibit a factor structure. According to Verdelhan [2018], two com-
mon factors, carry and dollar, account for 18% to 80% of the monthly
exchange rate movements.
Moreover, these common factors also price the cross-section of
currency risk premia [Lustig, Roussanov, and Verdelhan, 2011]. More
precisely, let f t denote a common factor (which can be easily ex-
tended to a vector of factors). We regress the currency excess return

Electronic copy available at: https://ssrn.com/abstract=4668578


40 jiang – lecture notes on international finance

or the exchange rate movement on the common factor:

rxti = αi + βi f t + εit .

The slope coefficient βi measures the currency’s exposure to the fac-


tor. To the extent that the factor proxies for systematic risks, investors
require a higher compensation if the currency has a higher risk ex-
posure. Then, we expect to find a positive association between the
currency’s expected excess return and its factor loading:

E[rxti +1 ] ∝ βi .

This alignment of risk exposures and risk premia, central to all asset
pricing, is also confirmed in the currency market.

2.A.6 Covered Interest Rate Parity Violation and Convenience Yields

The exchange rate puzzles above are all related to the stochastic
properties of exchange rates. A more recent literature examines re-
turns from currency market strategies that have no risks, at least in
theory. The fact that these risk-free strategies earn non-zero returns
suggests violations of the no-arbitrage condition, and could therefore
shed light on the frictions and preferences faced by the investors in
the international financial markets.
The covered interest rate parity (CIP) describes the relation be-
tween the forward premium and the interest rate differential, which
is given by Eq. (2.1), reproduced below,

f ti − eit = rti − rt$ .

This parity holds in frictionless markets because the investors can get
access to the home risk-free rate by either investing in the home bond
with a return of rt$ , or investing in the foreign bond and hedging the
exchange rate risk using the forward contract, which has a net return
of rti − f ti + eit . In the absence of arbitrage, these two risk-free rates in
home currency units must be equal.
Traditionally, this is regarded as an identity since its deviation,
known as the CIP deviation, will be quickly exploited by the arbi-
trageurs who actively trade in both the currency forward market and
the interest rate market. Indeed, if we use the home and foreign Li-
bor rates as proxies for the home and foreign interest rates rt and rt∗ ,
this condition holds quite tightly before the Global Financial Crisis.
However, Du, Tepper, and Verdelhan [2018b] shows that, after the cri-
sis, this condition breaks down and there is non-trivial variations in
both the time series and the cross-section of the Libor CIP deviation.

Electronic copy available at: https://ssrn.com/abstract=4668578


puzzles: challenges to making sense of data 41

We use the Libor CIP basis to measure the severity of the Libor CIP
deviation, defined as
def
xtLibor = libort$ − liborti + ( f ti − eit ).

Figure 2.1 plots the Libor CIP basis between the U.S. and the average
developed countries, which is very close to 0 before the Global Finan-
cial Crisis, but has been persistently negative since then. A negative
Libor CIP basis means that the dollar Libor rate is below the foreign
Libor rate after adjusting for the exchange rate risk, which makes the
dollar cheaper to borrow.
Similarly, the Treasury CIP basis is defined by using the Treasury
yields as proxies for the interest rates rt and rt∗ :
def
xtTreas = rt$ − rti + ( f ti − eit ).

Figure 2.1 also plots the Treasury CIP basis. Unlike the Libor CIP
basis, the Treasury CIP basis has been negative both before and after
the Global Financial Crisis [Du, Im, and Schreger, 2018a, Jiang, Kr-
ishnamurthy, and Lustig, 2018]. In other words, the U.S. government
has been able to borrow at a lower interest rate than the foreign gov-
ernments, after adjusting for the exchange rate risk. The Treasury CIP
deviation will play a central role when we discuss bond convenience
yields in Chapter 4.

Figure 2.1: U.S. Treasury and Libor


CIP Basis. Data source: Jiang, Krishna-
40 murthy, and Lustig [2021a].

20

-20

-40

-60

-80

-100

-120

-140
1990 1995 2000 2005 2010 2015

2.A.7 Currency Risk Premia across Horizons


We can also consider the currency returns across multiple horizons,
which can be achieved by either rolling over short-term debt posi-
tions in home and foreign currencies, or by investing in long-term

Electronic copy available at: https://ssrn.com/abstract=4668578


42 jiang – lecture notes on international finance

debt positions. Lustig, Stathopoulos, and Verdelhan [2019] show that


the long-run UIP condition, constructed from long-run exchange
rate movements and long-term bond yields, holds in complete mar-
kets with stationary exchange rates, because the currency risks are
correctly priced in the long-term bond yields.
Engel [2016] shows that high interest rate currencies not only have
higher expected returns in the short term, but also are stronger than
can be accounted for by the path of expected real interest differen-
tials. In other words, high interest rates must predict lower currency
returns in the long term. Relatedly, Dahlquist and Pénasse [2022],
Chernov and Creal [2023] show that the exchange rate level predicts
currency returns. Thus, a coherent account of the multi-horizon cur-
rency returns involves the interest rate predicting higher currency
returns and higher exchange rate levels in the short term, and the
elevated exchange rate levels predicting lower currency returns in the
medium-to-long term.

2.B Challenges to Making Sense of Quantities and Flows

2.B.1 Global Imbalances

We start with the U.S. net external imbalances vis-à-vis the rest of the
world. We consider the equity-like, riskier asset classes and the debt-
like, safer asset classes separately. Let us define the net risky position
as the sum of portfolio equity assets and foreign direct investment
(FDI) assets minus portfolio equity liabilities and FDI liabilities, and
define the net safe position as the reserve assets plus debt assets
minus debt liabilities [Gourinchas, Rey, and Sauzet, 2019].
Figure 2.2 shows the U.S. net risky and safe positions normalized
by GDP, from 1970 to 2021. We make three observations. First, on
average, the U.S. net risky position is positive and the U.S. net safe
position is negative. In other words, the U.S. holds more foreign risky
assets than it issues to the rest of the world, which implies that the
U.S. is a net lender of risky assets. On the other hand, the U.S. holds
less foreign safe assets than it issues, which implies that the U.S. is a
net borrower of safe assets.
Second, these net asset positions have notable cyclical properties.
Risky assets tend to depreciate during recessions, whereas safe assets
tend to appreciate. As a result, during the 2008 Global Financial Cri-
sis and the early-2000 stock market crash, the U.S. net risky position
contracted while its net safe position expanded.
Third, the U.S. net safe position has been widening steadily since
mid-1980s. Currently (end of 2021), it is about 50% of the U.S. GDP,
with a large fraction attributable to the U.S. government debt. In

Electronic copy available at: https://ssrn.com/abstract=4668578


puzzles: challenges to making sense of data 43

Figure 2.2: U.S. Net External Imbal-


ances. The shaded areas represent
30 NBER recessions. Data source: Lane
and Milesi-Ferretti [2007].
20

10

-10

-20

-30

-40

-50

-60
1980 1990 2000 2010 2020

comparison, the U.S. net risky position widened before 2008, but it
has been contracting since 2010. This contraction is driven by a rising
foreign demand for the U.S. equity assets.
We also sum up the U.S. net risky and safe positions to obtain
the U.S. net foreign assets (NFA). This is equal to the difference be-
tween the foreign assets held by the U.S. investors (i.e., the U.S. ex-
ternal assets) and the U.S. assets held by the foreign investors (i.e.,
the U.S. external liabilities). Figure 2.3 shows that the U.S. NFA has
been widening steadily since the 1980s, and the widening tends to
accelerate during recessions. The widening in the past decade is also
notable: the U.S. NFA expanded from −30% in early 2010s to −80%
right now, at a speed never seen before.
As we will see in Chapter 7, these imbalances reflect a fundamen-
tal asymmetry between the U.S. and the rest of the world. They play
an important role in our understanding of the international monetary
system, as they are closely tied to the global risk-sharing arrange-
ments and the unique position of the U.S. and the dollar.

2.B.2 The Exorbitant Privilege and Duty


Changes in the U.S. external imbalances are driven by both the quan-
tity of capital flows into and out of the U.S. and the returns on the
existing assets and liabilities. When we examine the latter compo-
nent, we find that the U.S. tends to earn a higher return on its exter-
nal assets relative to what it pays to foreigners on its external liabil-
ities. To describe this phenomenon, the French Minister of Finance
Valéry Giscard d’Estaing coined the term “exorbitant privilege” in
1965 [Gourinchas and Rey, 2007b, Gourinchas, Rey, and Govillot,

Electronic copy available at: https://ssrn.com/abstract=4668578


44 jiang – lecture notes on international finance

Figure 2.3: U.S. Net Foreign Assets.


The shaded areas represent NBER
10 recessions. Data source: Lane and
Milesi-Ferretti [2007].
0

-10

-20

-30

-40

-50

-60

-70

-80
1980 1990 2000 2010 2020

2010, Gourinchas, Rey, and Truempler, 2012]. This exorbitant priv-


ilege has been allowing the U.S. to run persistent current account
deficits despite being a net debtor to the rest of the world as we saw
in Figure 2.3.
The superior return earned by the U.S. can be attributed to two
factors. First, as we noted in the discussion of the global imbalances,
the U.S. holds more foreign risky assets that have higher average re-
turns, whereas the foreigners hold more U.S. safe assets that have
lower average returns. As a result, despite the fact that the U.S. hold-
ings of foreign assets are lower than the foreign holdings of the U.S.
assets, the U.S. manages to earn a risk premium on average because
of the return differential between the riskier and the safer assets.
Second, when we compare within the class of safe assets, the U.S.
is able to borrow at lower interest rates. The U.S. safe assets earn
convenience yields, which are often attributed to the U.S. dollar being
the dominant reserve currency.
These two factors both contribute to the U.S. earning a net portfo-
lio income from the foreigners. Figure 2.4 plots the annual valuation
change and dividend payouts from the U.S. external assets and lia-
bilities, normalized by the U.S. GDP. These series describe the total
amount of investment profits that U.S. investors receive abroad and
vice versa. Their difference is the net financial gains or losses be-
tween the U.S. and the rest of the world.
We make several observations about the U.S. net financial payoff.
First, on average, the payoff that U.S. receives from foreign assets
exceeds the payoff that U.S. pays to foreigners on its external liabili-
ties. The payoff includes both the dividend or coupon payments and

Electronic copy available at: https://ssrn.com/abstract=4668578


puzzles: challenges to making sense of data 45

the capital gains or losses. The U.S. has been earning consistently
positive payoffs from the 1960s to the mid 1990s, in early 2000s, and
in some years after 2008. Moreover, these magnitudes have been in-
creasing since the 1960s, possibly reflecting increases in international
portfolio positions and developments in international financial assets.

Figure 2.4: U.S. Exorbitant Privilege.


The figure plots the annual valuation
30 change and dividend payouts from the
U.S. external assets and liabilities, nor-
malized by the U.S. GDP. The shaded
20
areas represent NBER recessions. Data
source: Bureau of Economic Analysis.
10

-10

-20

-30
1970 1980 1990 2000 2010 2020

Second, the U.S. net financial payoff is pro-cyclical. It tends to


be positive during global expansions and negative during global
recessions. For example, the U.S. suffered a net loss equal to 30% of
the U.S. GDP on its external assets in 2008, whereas the foreigners
suffered a net loss equal to 13% of the U.S. GDP on their U.S. assets.
This cyclical property is again due to the portfolio asymmetry: the
U.S. tends to hold more foreign risky assets, whereas the foreigners
tend to hold more U.S. safe assets. The risky assets depreciate more
during recessions.
Gourinchas and Rey [2022] refer to this cyclical property as the
exorbitant duty. As we will see in Section 7.A, we can interpret the
asymmetry in portfolio holdings between the U.S. and the rest of
the world as a risk-sharing or insurance arrangement, with the U.S.
earning higher returns in good times and paying off the foreigners in
bad times.
Third, the nature of the exorbitant privilege and duty seems to
have changed in the past decade, in which period the U.S. external
assets and liabilities earned similar payoffs. In the past two years af-
ter the pandemic, the payoff on the U.S. external liabilities surpassed
that on the U.S. external assets. This is driven by a significant out-
performance of the U.S. equity assets relative to the foreign equity
assets, which raised the capital gains that foreigners earned from the

Electronic copy available at: https://ssrn.com/abstract=4668578


46 jiang – lecture notes on international finance

U.S. equity assets they held. As this difference in return is not driven
by the U.S. paying off foreign investors in recessions, this potentially
represents a loss of exorbitant privilege as opposed to the exorbitant
duty [Jiang, Richmond, and Zhang, 2022c, Atkeson, Heathcote, and
Perri, 2022].
There is also a trade side that parallels the fluctuations on the
portfolio side. As we will see in Section 9.A, the trade balance (TB)
and the portfolio returns from the income balance (IB) and capital
gains (CG) jointly determine the evolution of the national accounts
captured by the net foreign assets (NFA):

∆NFAt = TBt + CGt + IBt .

An increase in the U.S. net foreign assets can be attributed to either


national saving by running a higher trade surplus and saving the
income abroad, or earning a higher capital gains or incomes from
foreign financial assets. Conversely, holding the NFA constant, a
trade deficit can be financed by profits from the portfolio side.

2.B.3 Countercyclical Flight To Safety


There is also a quantity aspect to the exorbitant privilege, which can
be illustrated by the following hypothetical example that I considered
in Jiang, Krishnamurthy, and Lustig [2022a]. There are two invest-
ment periods. In period 1 the U.S. Treasury yield is high at 5%. In
period 2 the U.S. Treasury yield is low at 1%. Table 2.1 illustrates the
returns and the holdings of home and foreign investors. The foreign
investors buy more Treasurys when the yield is lower in period 2,
whereas the home investors buy more Treasurys when the yield is
higher in period 1. In a crude way, this assumption captures the for-
eigners’ countercyclical flight to safety provided by the U.S. Treasury
market: they buy U.S. Treasurys when they are expensive and offer
low returns, and they exit their positions when Treasury bonds are
cheap and offer high returns.
In this example, the time-weighted average as measured by the
geometric mean of the returns over the two years is exactly the same
2.98% for the U.S. and the foreign investors. In other words, the
home and the foreign investors hold the same asset and receive the
same return from their Treasury portfolios in each period.
In comparison, the dollar-weighted average as measured by the
internal rate of return (IRR) is 1.37% for the foreign investors and
4.65% for the home investors–leading to a difference of 3.28% per
annum between home and foreign investors. This IRR measure in-
corporates how well the investors time the market, and suggests that
the foreigners earn a much lower return from their dynamic trading

Electronic copy available at: https://ssrn.com/abstract=4668578


puzzles: challenges to making sense of data 47

strategy even when the underlying asset offers the same return.

Panel A: Description of the Example Table 2.1: Example of Cash Flows and
the IRR Computation
Foreign Investors Home Investors
Year Holdings Yield Holdings Yield
1 1 5% 10 5%
2 10 1% 1 1%
Time-Weighted Return 2.98% 2.98%
Panel B: IRRs of Cash Flows
Foreign Investors Home Investors
Year Holdings Cash Holdings Cash
Flows Flows
1 1 −1 10 −10
2 10 −8.95 1 9.5
3 10.10 10.10 1.01 1.01
Dollar-Weighted Return 1.37% 4.65%

Note that this IRR comparison captures a timing dimension of the


U.S. funding advantage that is different from the difference in the
interest rates as we discussed in the context of CIP deviations. In this
example, we can additionally introduce foreign bonds and assume
that they are always traded at 1% higher yields than the U.S. bond,
which means their yields are 6% in year 1 and 2% in year 2. Then, the
U.S. Treasury also enjoys an additional convenience yield regardless
of the timing in the bond market. As such, this stylized example
illustrates the importance of understanding the U.S. specialness from
both the quantity and the pricing sides.
This countercyclical flight to U.S. safe assets benefits their issuers
such as the U.S. government. A low dollar-weighted return earned by
the foreign investors due to their poor market timing implies a low
funding cost faced by the U.S. government due to its great market
timing. As a result, in bad times, the U.S. government can borrow not
just at a lower interest rate, but the fact that they can borrow a lot is
also important. As Gourinchas, Rey, and Sauzet [2019] observe,
In our fiat currency system, being the hegemon confers a specific ability to
issue large amounts of nominally safe liabilities (dollar securities), which are
happily absorbed by the rest of the world. Thus, the view is that, in case of a
deficit, the United States does not have to take restrictive measures, so that the
dollar is not an impartial means of international exchange. This is the essence
of the exorbitant privilege.

2.B.4 Twin Deficits


The U.S. external imbalances are also closely related with the U.S.
government budget deficits. Figure 2.5 plots the U.S. government’s

Electronic copy available at: https://ssrn.com/abstract=4668578


48 jiang – lecture notes on international finance

total government debt held by public/GDP ratio and the U.S. exter-
nal liability/GDP. The first series represents the amount of debt owed
by the U.S. government to the U.S. private sector and the rest of the
world, whereas the second series represents the amount of debt owed
by the U.S. to the rest of the world. Since the 2008 global financial
crisis, both the U.S. government and the U.S. as a whole have been
borrowing a lot more.
This comovement shows that international finance and govern-
ment finance are closely intertwined. On the one hand, the foreign
investors, private and official, have been financing a large portion of
the U.S. external liabilities. To understand why this is the case, it is
important to understand why they find the U.S. government debt
particularly desirable. On the other hand, a large portion of the U.S.
government debt has been financed by foreign investors. To evalu-
ate the U.S. fiscal sustainability, it is also important to take a global
perspective. We will consider these issues in detail in Chapters 7 and
8.

2.B.5 Other Empirical Patterns

There are many more empirical patterns that this note does not have
space to cover. For example, home bias is a salient pattern in inter-
national asset allocation. Lewis [1999] provides a summary of the
literature. More recently, micro-data shed new light on cross-border
portfolio positions and global capital allocations. Florez-Orrego,
Maggiori, Schreger, Sun, and Tinda [2023] provide a review of this
literature.

Figure 2.5: U.S. Public and External


Liabilities. The figure plots the U.S.
100 public liability/GDP ratio and the
U.S. external liability/GDP ratio.
The shaded areas represent NBER
80
recessions. Data source: FRED and
Lane and Milesi-Ferretti [2007].
60

40

20

-20
1980 1990 2000 2010 2020

Electronic copy available at: https://ssrn.com/abstract=4668578


puzzles: challenges to making sense of data 49

Miranda-Agrippino and Rey [2022] provide a summary of the


literature on the global financial cycle, which refers to the strong
comovements between exchange rates, asset prices, capital flows, and
the U.S. monetary policy. Du and Schreger [2022a] provide a review
of the literature on the CIP deviation and frictions in international
capital markets. Hassan and Zhang [2021] provide a review of the
literature on currency risks and returns.

Electronic copy available at: https://ssrn.com/abstract=4668578


Part II

Understanding the
Exchange Rates

Electronic copy available at: https://ssrn.com/abstract=4668578


3
Risk Premia and Factor Structure

Summary

• We first investigate the role of currency risk premium rpt , which drives both the currency
expected return:

def 1
Et [rxt+1 ] = rpt = −covt (m∗t+1 , ∆et+1 ) − vart (∆et+1 ),
2
and the exchange rate level:
∞ ∞
et = ∑ Et [rt+ j − rt∗+ j ] − ∑ Et [rpt+ j ] + ē.
j =0 j =0

• The currency risk premium is related to the currency’s loadings on the risk factors, which give
rise to a factor structure in currency returns:

∆eit+1 = αi + βi f t+1 + εit+1 ,


Et [rxti +1 ] = βit Et [ f t+1 ].

We first adopt a no-arbitrage approach to organize the currency risk premium and the corre-
sponding factor structure.

• A natural question is what drives the time-series and cross-sectional variations in the currency
risk premium, which is to be answered by general equilibrium models.

• Finally, we consider the long-term currency risk premia. When markets are complete and the
exchange rate is stationary, the long-run UIP condition holds:

et − ē = lim h(rt (h) − rt∗ (h)).


h→∞

To make sense of the exchange rate puzzles and other salient pat-
terns in international financial markets, we first enrich the risk pre-
mium in the baseline model. To be clear, as long as the households’
utility function u(c) is non-linear, the currency risk premium is al-

Electronic copy available at: https://ssrn.com/abstract=4668578


52 jiang – lecture notes on international finance

ready present in the baseline model in Chapter 1. As the exchange


rate accounting in Section 1.B shows, the currency risk premium rpt
affects both the currency expected return and the exchange rate level.
In this chapter, we consider specific models that account for the
time-series and the cross-sectional variations in the currency risk
premium, and understand the properties of the implied exchange
rate dynamics. As an initial step, all models presented in this chapter
assume complete markets.
Before we start, it is useful to understand what happens in the ab-
sence of the risk premium. Then, Eq. (1.13) implies that the exchange
rate movement is determined entirely by the current interest rate
differential,

Et [∆et+1 ] = rt∗ − rt ,

and, equivalently, the expected currency excess return is zero,

Et [rxt+1 ] = 0.

These conditions are known as the Uncovered Interest Rate Parity. As


we discussed in Section 2, these predictions are rejected by the data.
Instead, currency expected returns have large variations both in the
time series and in the cross-section.
Alvarez, Atkeson, and Kehoe [2007] lay out a related calculation.
Eq. (1.13) implies that the interest rate differential is equal to the
expected exchange rate movement plus the risk premium term:

rt − rt∗ = Et [−∆et+1 ] + rpt ,

Empirically, there are large variations in the interest rate differential,


but the exchange rate appears to be a near random-walk process,
which means the exchange rate movement is largely unexpected. As
a result, the risk premium term rpt must have significant variations.
This means, to the extent that monetary policies drive a significant
fraction of the variations in the interest rates, it must operate not
through the first-order moments (i.e., the expected exchange rate
movement), but through the second-order moments (i.e., the risk
premium).

3.A The No-Arbitrage Approach

Our first approach to modeling the currency risk premia starts di-
rectly from the SDFs. Because this approach only relies on investors’
ability to correctly price the assets based on the covariances between
asset returns and their SDFs, it is known as the no-arbitrage ap-
proach. This approach asks the following question: what properties

Electronic copy available at: https://ssrn.com/abstract=4668578


risk premia and factor structure 53

do the SDFs need to have in order to match the observed exchange


rate dynamics, such as the time-series and the cross-sectional varia-
tions in the currency risk premia, and the relationship between the
interest rates and the currency risk premia? Once we understand
the answer to this question, we can then turn to general equilibrium
models and figure out the economic primitives and frictions that are
necessary to generate such SDF properties.
For example, Lustig, Roussanov, and Verdelhan [2011] consider the
following specification. The SDF of country i is
q q
mit+1 = −µit − δi zw ε w
t t +1 − γzit εit+1 ,

where εw i
t+1 is a global shock and ε t+1 is a country-specific shock.
Both shocks are i.i.d. standard normal. When the number of coun-
tries N is large, the country-specific shocks average to zero:
N q
∑ γzit εit+1 = 0.
i =1

While we take the SDF dynamics as given, it is important to know


that the SDFs ultimately arise from the underlying economy. For ex-
ample, the SDF shocks εw i
t+1 and ε t+1 could be mapped to the global
and country-specific shocks in the production process. They could
also arise from shocks to investor preferences. We will discuss some
general equilibrium models in which the SDF dynamics are endoge-
nously determined in the following sections.
For expositional convenience, we regard the U.S., denoted by $, as
the home country. Let ∆ei/$
t+1 denote the value of currency i in the unit
of the home country. Then, assuming complete markets, we can write
down the exchange rate movement directly from the SDF dynamics:

∆ei/$ i $
t +1 = m t +1 − m t +1
  √ √ q q q 
i $ i $ w w i i $ $
= − µt − µt − δ − δ z t ε t +1 − γzt ε t+1 − γzt ε t+1 .

Lustig, Roussanov, and Verdelhan [2011] assume that the time-


varying volatilities zw i
t and zt are also driven by the global and the
country-specific shocks, and they follow autoregressive square-root
processes:
q
zw w w w
t = (1 − ϕ ) θ + ϕzt−1 + σ zt−1 ε t ,
w
(3.1)
q
zit = (1 − ϕ)θ + ϕzit−1 + σ zit−1 εit . (3.2)

As a result, when either the global shock or the country-specific


shock is negative, the SDF becomes not only higher (indicating a
higher marginal utility) and drives the exchange rate movement, it
also becomes more volatile and affects the currency risk premium.

Electronic copy available at: https://ssrn.com/abstract=4668578


54 jiang – lecture notes on international finance

3.A.1 Currency Return Dynamics


Let us characterize the log currency excess return in this set-up,
which is defined as
def
rxti/$ i/$ i $
+1 = ∆et+1 + rt − rt .

First, the innovation in the currency excess return is given by


√ √ q √
q q 
rxti/$
+1 − Et [rxti/$
+1 ] =− δi − δ $ w w
z t ε t +1 − γ i i $ $
z t ε t +1 − z t ε t +1 .

Since the currency excess return is equal to the exchange rate move-
ment plus the interest rate differential, which is known ex-ante, the
innovation in the currency excess return is also equal to the innova-
tion in the exchange rate movement:

rxti/$ i/$ i/$ i/$


+1 − Et [rxt+1 ] = ∆et+1 − Et [ ∆et+1 ].

Thus, the conditional variance of the exchange rate movement in


log can be expressed as

1  √ i √ $ 2 w 1
vart (∆ei/$
t +1 ) = δ − δ zt + γ(zit + z$t ), (3.3)
2 2
which shows that the variance zw t of the global shock and the vari-
i $
ances zt and zt of the country-specific shocks both contribute to the
exchange rate variance. Moreover, for two countries whose SDFs have
very different loadings on the global shock, their bilateral exchange
rate movement loads heavily on the global shock and is therefore
more volatile.
Next, we consider the currency risk premium. Plugging the SDF
dynamics into Eq. (1.17), we obtain

def
h i 1 1
rpi/$
t = E t rx i/$ $ i
t+1 = 2 vart ( mt+1 ) − 2 vart ( mt+1 )
(3.4)
1 $  1  $ 
= δ − δi z w t + γ z t − z i
t .
2 2
The first term is easy to interpret: if two countries have different
loadings on the global shock, i.e., |δ$ − δi | > 0, then, their bilateral
exchange rate movement is exposed to the global shock and requires
a higher magnitude of risk premium.
The second term is increasing in the difference between the volatil-
ities of the two countries’ idiosyncratic shocks, i.e., z$t − zit , which
is more subtle. The readers might wonder why idiosyncratic risks
appear in the risk premium. For example, from the U.S. households’
perspective, should they not care about the foreign country’s idiosyn-
cratic risk zit since it is uncorrelated with the U.S. SDF?

Electronic copy available at: https://ssrn.com/abstract=4668578


risk premia and factor structure 55

To answer this question, let us first consider the currency risk


premium in level:

exp(ei/$ $
" #
i
i/$ def t+1 + rt ) − exp(rt )
Et [exp(rxt+1 )] = Et 1 + ,
exp(rt$ )

which captures the expected profit of a trading strategy that takes a


long position on the foreign bond and a short position on the U.S.
bond. By Eq. (1.11),

Et [exp(rxti/$ $ i/$
+1 )] = exp(− covt ( mt+1 , ∆et+1 ))
√ √ √  
$
= exp δ$ δ $ − δi z w t + γzt ,

√ √ √
which does not contain the zit term. The term δ$ ( δ$ − δi )zw t
comes from the U.S. SDF’s and the exchange rate movement’s load-
$
ings on the global shock εw t+1 , and the term γzt comes from their
$
loadings on the U.S.-specific shock ε t+1 . As the global shock and
the U.S.-specific shock both affect the U.S. SDF, they are priced risk
factors from the U.S. perspective.1 1
That said, the bilateral exchange rates
Now, let us revisit the currency risk premium in log, which is between different foreign currencies
and the dollar only have heterogeneous
related to the levels by loadings on the global shock, but not
on the U.S.-specific shock. Therefore,
1 the U.S.-specific shock leads to iden-
rpi/$
t = log Et [exp(rxti/$ i/$
+1 )] − 2 vart ( ∆et+1 ). tical risk premia across all foreign
currencies, which is γz$t .
Similarly, if we take the foreign perspective,

1 i/$
rp$/i
t = log Et [exp(rxt$/i
+1 )] − 2 vart ( ∆et+1 ).

Combine these two expressions and use the definition that implies
the log risk premia are symmetric, i.e., rpi/$
t = −rp$/i
t . We have

1
rpi/$
t = (log Et [exp(rxti/$ $/i
+1 )] − log Et [exp(rxt+1 )]).
2
Therefore, the log risk premium can be thought of as an average
between the level risk premia from the U.S. and the foreign perspec-
tives, which care about the U.S.-specific and the foreign-specific risks,
respectively.
It is also worth noting that the currency risk premia in level are
not symmetric:

log Et [exp(rxti/$ $/i


+1 )] ̸ = − log Et [exp(rxt+1 )],

which leads to the following result commonly known as the Siegel’s


paradox. For example, assume rt$ = rti = 0 and Et [∆ei/$t+1 ] = 0.
Then, the currency risk premium in log is zero: rpi/$
t = −rp$/i
t = 0.
However, given that the exchange rate is volatile, Jensen’s inequality

Electronic copy available at: https://ssrn.com/abstract=4668578


56 jiang – lecture notes on international finance

implies that the expected currency excess return in level is above 1


from both the U.S. and the foreign perspectives:

Et [exp(rxti/$ i/$ $/i $/i


+1 )] > exp(Et [rxt+1 ]) = 1 and Et [exp(rxt+1 )] > exp(Et [rxt+1 ]) = 1.

Because of this property, we use the currency risk premium in log for
most parts of this paper.
These results connect currency risk premia and realized currency
excess returns to the factor structure of the SDFs. We make three
more observations. First, for a country whose loading δi on the global
shock is high, its currency tends to appreciate during global reces-
sions (i.e., rxti/$ w
+1 increases when ε t+1 declines), and have a lower
currency risk premium rpi/$ t . Therefore, currencies with higher load-
ing δi are considered as safe currencies, and currencies with lower
loading δi are considered as risky currencies.
Second, as for the country-specific shocks, even though they are
idiosyncratic from the aggregate perspective and there is full risk-
sharing between countries, they still affect currency returns and risk
premia. This feature can arise from general equilibrium models if the
local households have a consumption bias towards home goods, so
that their consumption allocation and marginal utility tilt towards
the local shock despite full risk-sharing. Specifically, when zit is high,
country i’s SDF is volatile, and its local households require a higher
risk premium to hold foreign currencies.
Third, if two countries have high and identical idiosyncratic
volatilities, i.e., zit = z$t ≫ 0, then, their bilateral exchange rate move-
ment is volatile, while the currency risk premium can remain low.
Conversely, we can generate a high currency risk premium without
a volatile exchange rate movement. To do so, notice that the global
risk loadings enter the exchange rate variance expression (3.3) as
√ √ 2
δi − δ$ , whereas they enter the currency risk premium ex-
√ √  √ √ 
pression (3.4) as δ$ − δi = δi − δ $ δi + δ$ . Then, high
loadings on the global shock, i.e., δ$ ≈ δi ≫ 0, can generate a high
currency risk premium without a volatile exchange rate movement.
√ √ 2
For example, if δ$ = 10 and δi = 9, δi − δ$ = 0.03 implies a
low exchange rate volatility, whereas δ$ − δi = 1 implies a high cur-
rency risk premium. As such, this no-arbitrage framework provides
a flexible way to match different patterns in exchange rate volatilities
and currency risk premia.

3.A.2 Interest Rate Dynamics


This no-arbitrage framework also provides a simple characterization
of the interest rate dynamics. From the Euler equation (1.6), repro-

Electronic copy available at: https://ssrn.com/abstract=4668578


risk premia and factor structure 57

duced below,
h i
1 = Et exp(mit+1 + rti ) ,

and under joint normality, the interest rate in country i can be ex-
pressed as

1 1 1
rti = −Et [mit+1 ] − vart (mit+1 ) = µit − δi zw − γzit .
2 2 t 2

In this expression, the quadratic terms − 12 δi zw 1 i


t − 2 γzt are decreas-
ing in the volatility of the SDF. In other words, when the households’
marginal utility growth is more volatile, they will impute a higher
value to the risk-free asset and accept a lower risk-free interest rate.
This effect is usually referred to as the precautionary saving motive.
The linear term µit is a free parameter because the no-arbitrage
approach does not take a stance on the investors’ preferences. In
consumption-based models, this term usually reflects the investors’
subjective discount rate and the expected growth rate of their con-
sumption. Specifically, if the expected growth rate of consumption is
higher, then, investors tend to require a higher risk-free rate to save.
This effect is usually referred to as the intertemporal smoothing motive.
As we will see in Chapter 6, this linear term also reflects monetary or
fiscal policies.
To operationalize this model, we need to take a stance on the µit
term. Motivated by the connection between the interest rates with
currency risk premia, we further assume

µit = α + χzit + χzw


t .

Then, the interest rate differential can be expressed as

1 $ 1
rti − rt$ = ( δ − δi ) z w i $
t − ( γ − χ )( zt − zt ), (3.5)
2 2

which implies that the interest rate differential also comoves with
the global and country-specific risk premia. A country with a higher
loading δi on the global shock has a lower interest rate, in addition
to having a lower risk premium. Moreover, if 21 γ − χ > 0, which
means that the interest rate is not set high enough to offset the pre-
cautionary saving motive, then, a country with a higher idiosyncratic
volatility zit also has a lower interest rate, in addition to having a
lower risk premium.
This positive relationship between the interest rate differential and
the currency risk premium holds both across countries and over time.
We next explore these two dimensions in detail.

Electronic copy available at: https://ssrn.com/abstract=4668578


58 jiang – lecture notes on international finance

3.A.3 Carry Trade and Cross-Sectional Variations in Currency Risk Pre-


mium
To begin with, let us write down the log currency risk premium from
Eq. (3.4) and the interest rate differential from Eq. (3.5):
1 $  1  $ 
rpi/$
t = δ − δ i
z w
t + γ z t − z i
t ,
2 2
1 1
rti − rt$ = (δ$ − δi )zw i $
t − ( γ − χ )( zt − zt ).
2 2
In this subsection, we focus on the cross-country heterogeneity
in the parameter δi , which captures each foreign currency’s loading
on the global shock. This parameter appears in both expressions
with the same sign: if a country has a higher loading δi on the global
shock, then, its currency risk premium and its interest rate are both
lower.
Empirically, we need to find a way to estimate the systematic risk
loading parameter δi . To do so, we construct the carry trade. We sort
currencies into portfolios based on their interest rates. We use H
to denote the basket of currencies with high interest rates, and L to
denote the basket of currencies with low interest rates. The portfolio
composition can change over time. The carry trade factor hml is
defined as
1 1
hmlt+1 =
NH ∑ rxti/$
+1 − N ∑ rxt+1 ,
L
i/$

i∈ H i∈ L

where NH and NL denote the number of currencies in each portfolio.


q q
We let δtI denote the equal-weighted average of δti across all
currencies i in portfolio I.
This carry trade factor has three notable properties. First, the in-
novation to the carry trade factor is a scaled version of the global
shock:
q q q
w
hmlt+1 − Et [hmlt+1 ] = δtL − δtH zw
t ε t +1 .

If we assume that the idiosyncratic volatility zit is independent across


countries and has the same time-series dynamics (for example,
AR(1)), then, Eq. (3.5) implies that higher interest rate currencies
q q
tend to have lower δi , and vice versa. Then, δtL − δtH > 0, which
implies that the carry trade factor hmlt+1 has a positive loading on
the global shock.
Second, by being exposed to the global shock, the carry trade
factor earns a positive risk premium on average:
1 L 
Et [hmlt+1 ] = δt − δtH zwt > 0.
2

Electronic copy available at: https://ssrn.com/abstract=4668578


risk premia and factor structure 59

In other words, the currency trading strategy that buys high interest
rate currencies and sells low interest rate currencies earns a positive
risk premium on average.
Third, each currency’s loading on the carry trade factor informs us
about its systematic risk exposure. Consider the conditional regres-
sion

rxti/$ i i i
+1 = αt + β t · hmlt+1 + ε t+1 ,

with βit defined as

def covt (rxti/$


+1 , hmlt+1 )
βit = .
vart (hmlt+1 )

We have the following result:

Proposition 3.1. Currency i’s carry trade beta is


√ √
i δ $ − δi
βt = q q .
δtL − δtH
q q
The proof is presented in Appendix A.3. Since δtL and δtH are
q q
common to all currencies, and δtL > δtH , then, the carry trade beta
βit is decreasing in the country’s loading δi on the global shock.
This proposition connects the currencies’ risk exposures to their
risk premia, incarnating the standard asset pricing logic that each
asset’s risk exposure can be measured from its covariance with risk
factors, i.e.,

rxti +1 = αi + βit · f t+1 + εit+1 ,

and that the risk exposure determines the asset’s risk premium, i.e.,

Et [rxti +1 ] = βit · Et [ f t+1 ],

assuming that the factors are tradable. In the context of the currency
market, the carry trade captures a risk factor: f t+1 = hmlt+1 . The
carry trade beta βit captures the risk exposure of currency i, and it re-
lates to the currency’s risk premium via Eq. (3.4), which is increasing
in βit .

3.A.4 Forward Premium Puzzle and Time-Series Variations in Currency


Risk Premium
Let us return to the expressions for the log currency risk premium
and the interest rate differential between the dollar and a given for-

Electronic copy available at: https://ssrn.com/abstract=4668578


60 jiang – lecture notes on international finance

eign currency i:

1 $  1  $ 
rpi/$
t = δ − δi z w t + γ z t − z i
t ,
2 2
1 1
rti − rt$ = (δ$ − δi )zw i $
t − ( γ − χ )( zt − zt ).
2 2
In this subsection, we turn to the time-series variation in the vari-
ances of the SDF shocks zw i
t and zt , which drive variations in both the
currency risk premium and the interest rate differential over time.
As we will show in the following proposition, this common variation
gives rise to predictive power of the interest rate differential for the
currency excess return. Specifically, we compute the slope coefficient
in the forecasting regression of the currency excess return on the
interest rate differential:

rxti/$ i i i $ i
+1 = α + φ ( r t − r t ) + ε t +1 ,

with the Fama regression coefficient φi defined as

def cov(rxti/$ i $
+1 , r t − r t )
φi = .
var (rti − rt$ )

Proposition 3.2. The slope coefficient from the regression of the currency
excess return on the interest rate differential is given by

1 1 i $ 1 $ i 2 w
2 γ ( 2 γ − χ )( var ( zt ) + var ( zt )) + 4 ( δ − δ ) var ( zt )
φi = $
,
(χ − 12 γ)2 (var (zit ) + var (zt )) + 14 (δ$ − δi )2 var (zwt )

which is positive if 12 γ − χ > 0.

The proof is presented in Appendix A.4. This proposition shows


that the exposures to both the global and the country-specific SDF
shocks induce comovements between the interest rate differential
and the currency excess return. Specifically, a country’s interest rate
declines when the global SDF volatility goes up and, if we assume
1
2 γ − χ > 0, when its idiosyncratic SDF volatility goes up. Both SDF
components generate a positive slope coefficient in this regression,
which rationalizes the UIP violation in the data.
Similarly, we can replace the currency excess return on the left-
hand side of the regression with the exchange rate movement:

∆ei/$ i i i $ i
t +1 = α + ψ ( r t − r t ) + ε t +1 .

Its slope coefficient is given by

def cov(∆ei/$ i $
t +1 , r t − r t ) χ( 12 γ − χ)(var (zit ) + var (z$t ))
ψi = = ,
var (rti − rt$ ) (χ − 12 γ)2 (var (zit ) + var (z$t )) + 14 (δ$ − δi )2 var (zw
t )

Electronic copy available at: https://ssrn.com/abstract=4668578


risk premia and factor structure 61

which is also positive if 12 γ − χ > 0 and χ > 0. In other words, when


χ > 0, the interest rate differential does not fully reflect the currency
risk premium that is driven by the idiosyncratic SDF volatility in this
model, so that the expected exchange rate movement also co-moves
with the risk premium.
In models with richer structures, it is possible that the global SDF
volatility zw
t also drives the expected exchange rate movement. For
example, this happens when the drift term in the SDF is

µit = α + χzit + ζδi zw


t ,

for 0 < ζ < 1/2.


 In this case, while the currency risk premium loads
1 i
on $
2 δ −δ times the global shock’s volatility zw
t , i.e.,

def
h i 1 $  1  $ 
rpi/$
t = Et rxti/$ i
+1 = 2 δ − δ z t +
w
γ zt − zit ,
2
  
1 i − δ$ times the
the interest rate differential only loads on ζ − 2 δ
global shock’s volatility zwt :
   
1  1 
rti − rt$ = ζ− i
δ −δ $
zw
t + χ− γ zit − z$t ,
2 2

which does not fully reflect this global component of the currency
risk premium.
As a result, the expected exchange rate movement also loads on
the global SDF volatility zw
t . This might be a more realistic speci-
fication since the global SDF volatility also induces exchange rate
predictability by the interest rate differential:
    2
cov(∆ei/$ i $ χ 21 γ − χ (var (zit ) + var (z$t )) + ζ 12 − ζ δi − δ$ var (zw t )
t +1 , r t − r t )
ψi = =  2 2 > 0.
var (rti − rt$ )
 2
χ − 12 γ (var (zit ) + var (z$t )) + ζ − 12 δi − δ$ var (zw
t )

3.A.5 Currency Base Factor


Another convenient construct to measure risk exposures and cur-
rency risk premia is the currency base factor [Lustig and Richmond,
2020, Aloosh and Bekaert, 2021]. For country i, its currency base fac-
tor is defined as the average of its bilateral exchange rate movement
against foreign currencies2 : 2
Some authors use an alternative
i/j
definition ∆eit = N1−1 ∑i̸= j ∆et . Since
def 1 N ∆ei/i = 0, these two definitions of
∑ ∆et
i/j t
∆eit = . currency base factors only differ by
N i =1 a constant scalar N/( N − 1). We use
the definition in the main text since it
This construct conveniently captures a currency’s average ex- computes simpler averages.
change rate movement. For example, the dollar index is commonly

Electronic copy available at: https://ssrn.com/abstract=4668578


62 jiang – lecture notes on international finance

used in practice to reflect the overall strength of the U.S. dollar. The
dollar’s base factor ∆e$t is the unweighted version of the dollar index.
By the law of large numbers, the SDF structure in Section 3.A
implies
  √ √ q w q
∆eit+1 = mit+1 − mt+1 = −χ zit − zt − δi − δ zw
t ε t + 1 − γzit εit+1 ,

which measures currency i’s exposures to the global shock εw t+1 and
i
to its own idiosyncratic shock ε t+1 , in a way that is invariant to the
choice of the base currency j. As such, if we want to study the varia-
tion in the dollar, the dollar’s base factor might be more informative
than the dollar’s bilateral exchange rate against an individual foreign
currency.
For example, the conditional and the unconditional variances of
the currency base factor can be expressed as
√ √ 2 w
vart (∆eit+1 ) = δ zt + γzit ,
δi −
√ √ 2
var (∆eit+1 ) = χ2 var (zit ) + δi − δ θ w + γθ,

which uncovers this currency’s idiosyncratic variance zit and its expo-
sure to the global shock δi .
Moreover, Lustig and Richmond [2020] also consider the following
regression:

i/j
∆et+1 = α + φi/j ∆eit+1 + ε t+1 .

If the idiosyncratic volatility z j follows the same distribution specified


in Eq. (3.2), then, zt = N1 ∑iN=1 zit is time-invariant, and we have the
following result characterizing the slope coefficient φi/j :

Proposition 3.3. The slope coefficient φ in this regression is given by


√ √  √ √ 
i/j
cov(∆et+1 , ∆eit+1 ) χ2 var (zit ) + δi − δ j δi − δ θ w + γθ
i/j def
φ = = √ √ 2 .
var (∆eit+1 ) χ2 var (zit ) + δi − δ θ w + γθ

The proof is presented in Appendix A.5. We note that the only


foreign country j-specific variable in this expression is its exposure
j
to the systematic risk, δ j . In particular, its idiosyncratic risk zt does
not show up in this expression, because the currency base factor
∆eit+1 is not exposed to any individual foreign country’s idiosyncratic
shock. Hence, for a given home currency i, the regression coefficient
φi/j measures the difference in the systematic risk exposure between
√ √ 
countries i and j, i.e., i
δ − δ . j

Electronic copy available at: https://ssrn.com/abstract=4668578


risk premia and factor structure 63

3.A.6 Other Works


This no-arbitrage approach provides a useful framework to account
for the time-series and the cross-sectional variations in exchange rate
movements and currency risk premia using global and idiosyncratic
SDF shocks in complete markets. As such, it could be applied to all
complete-market models that we study in this chapter.
We can and should introduce additional systematic risk factors
in this no-arbitrage framework. Lustig, Roussanov, and Verdelhan
[2014], Verdelhan [2018] study a richer factor model of exchange rates
with two common factors, one of which is related to the currency
carry trade that we study in this section, and the other is related to
the U.S. dollar. To generate this structure, the SDF is exposed to both
common factors:
q q q
mit+1 = −µit − δi zw ε w
− κ i zi ε g − γzit εit+1 ,
t t +1 t t +1

where the common shock εw t+1 is tied to the currency carry trade re-
g
turn and the other common shock ε t+1 is tied to the U.S. dollar. The
loadings are time-varying, with the loadings on the dollar shock and
on the idiosyncratic shock both tied to a local state variable zit and the
loading on the carry trade shock tied to a global state variable zw t .
We can also bring the SDF dynamics closer to the data. If the
global and the country-specific shocks load on a vector of observable
variables, we can model the SDF process using a VAR representation.
For example, suppose the state vector xti follows a VAR(1) process:

xti +1 = Ψxti + Σ1/2 εit+1 ,

and the SDF innovations depend on the same VAR shocks εit+1 :
1
mit+1 = −rti − (λit )′ λit − (λit )′ εit+1 .
2
q
The vector of loadings λit generalizes the scalar loadings δi zw t
q
and γzit in the previous case, and, for tractability, they are linear in
the state vector xti :

λit = λ0i + λ1i xti .

This affine structure allows us to derive the asset prices in do-


mestic bond and equity markets in closed form. Assuming complete
markets, the exchange rate movement can be expressed as

∆ei/$ i $
t +1 = m t +1 − m t +1 .

In this way, we can jointly model exchange rates and asset prices
using the same set of observable variables [Chernov and Creal, 2023].
We will discuss this class of affine models in detail in Section 8.E.

Electronic copy available at: https://ssrn.com/abstract=4668578


64 jiang – lecture notes on international finance

3.B Currency Risk Premia in the Time Series

The no-arbitrage approach offers us an accounting framework to


break down the variations that come from country-specific and global
shocks. Using the Euler equations, it imposes discipline on how these
shocks give rise to variations in exchange rates and currency risk pre-
mia, and helps us organize the exchange rate puzzles. However, it
does not answer where these shocks come from. We need additional
structures to understand the origins of these shocks. In this section
and the next section, we consider two models that offer such struc-
tures. One model is based on the long-run risks and focuses on the
time-series variations in currency risk premia, and the other model is
based on the trade network and focuses on the cross-sectional varia-
tions in currency risk premia.

3.B.1 Households
There are two countries, home and foreign. The households have the
Epstein and Zin [1989] recursive preferences. For the home house-
holds,
1
( 1
1− ψ
) 1
1− ψ
1− ψ1
vt = (1 − δ)(ct ) + δEt [(vt+1 )1−γ ] 1−γ , (3.6)

where γ is the coefficient of risk aversion and ψ is the intertemporal


elasticity of substitution, and δ is the coefficient of time-preference.
Ct is consumption at time t. It is convenient to define θ = 1−γ1 and
1− ψ
θ
1− γ
 
1 1− γ
simplify Eq. (3.6) as vt = (1 − δ)ct θ + Et [(ut+1 )1−γ ] θ .
Consumption growth has a long-run component xt , which is small,
very persistent, and has stochastic volatility:

∆ct+1 = µ g + xt + σt ε g,t+1 ,
xt+1 = ρxt + φe σt ε x,t+1 ,
σt2+1 = σ2 + ϕ(σt2 − σ2 ) + ωε w,t+1 .

The consumption growth shock ε g,t+1 , the long-run growth shock


ε x,t+1 , and the long-run growth volatility shock ε w,t+1 are jointly
normal. We assume that long-run and short-run consumption shocks
are correlated: covt (ε g,t+1 , ε x,t+1 ) = ν. For simplicity, we assume
covt (ε x,t+1 , ε w,t+1 ) = 0, covt (ε g,t+1 , ε w,t+1 ) = 0.
Let wt denote the aggregate wealth, which is the claim to cur-
rent and future consumption. Let wct = wt /ct denote the wealth-
consumption ratio. We can define the cum-dividend return on this

Electronic copy available at: https://ssrn.com/abstract=4668578


risk premia and factor structure 65

claim as
w t +1 c wct+1
exp(rtc+1 ) = = t +1 .
wt − ct ct wct − 1
Using the Campbell-Shiller log-linearization of the log total wealth
return around the long-run average log wealth consumption ratio
µwc = E[log(wct )]:

rtc+1 = κ0c + ∆ct+1 + log(wct+1 ) − κ1c log(wct ), (3.7)


µwc µwc
with κ1c = eµewc −1 > 1 and κ0c = − log(eµwc − 1) + eµewc −1 µwc .
Epstein and Zin [1989] show that the log SDF mt+1 can be ex-
pressed as

θ
mt+1 = θ log δ − ∆c + (θ − 1)rtc+1 ,
ψ t +1

which depends on not only the log consumption growth ∆ct+1 , but
also the log return on the wealth portfolio rtc+1 . The following propo-
sition shows that the log wealth-consumption ratio and the log SDF
are linear functions of the state variables after the log-linearization.

Proposition 3.4. The log wealth-consumption ratio log(wct ) is affine in xt


and σt2 :

log(wct ) = µwc + Wx xt + Wσ (σt2 − σ2 ), (3.8)

where
1
1− ψ
Wx = ,
κ1c − ρ
 
(1 − γ) 1 − ψ1  φ2e 2νφe

Wσ = + +1 ,
2(κ1c − ϕ) (κ1c − ρ)2 κ1c − ρ

and the log SDF can be expressed as

mt+1 − Et [mt+1 ] = −γσt ε g,t+1 + (θ − 1) Wx σt φe ε x,t+1 + (θ − 1)Wσ ωε w,t+1 ,


 
θ θ
Et [mt+1 ] = θ log δ + (θ − 1) r0 − µ g + − + (θ − 1)[1 + Wx (ρ − κ1 )] xt + {Wσ (ϕ − κ1c )(θ − 1)} (σt2 − σ2 ).
c c
ψ ψ

The proof is presented in Appendix A.6. To interpret this result,


we provide a simple calibration of the model following Bansal and
Shaliastovich [2007]. The parameter values are reported in Table 3.1.
The only modification is that we assume the correlation between
long-run and short-run consumption shocks, i.e., covt (ε g,t+1 , ε x,t+1 ) =
ν, to be negative.
At the bottom of the table, we report the values of the key equi-
librium objects in this model. Given that Wx > 0 and Wσ < 0, the

Electronic copy available at: https://ssrn.com/abstract=4668578


66 jiang – lecture notes on international finance

wealth-consumption ratio is increasing in the long-run growth and


decreasing in the volatility. Similarly, we can show that the SDF mt+1
is decreasing in the short-run consumption shock ε g,t+1 , decreas-
ing in the long-run consumption shock ε x,t+1 , and increasing in the
volatility shock ε w,t+1 .

Parameter Value Table 3.1: Model Parameter Values


Preference Parameters:
Subjective discount factor δ 0.9987
Intertemporal elasticity of substitution ψ 1.5
Risk aversion coefficient γ 8
Implied theta θ −21

Consumption Growth Parameters:


Mean of consumption growth µg 0.0016
Long-run risks persistence ρ 0.991
Long-run risks volatility loading φe 0.055
Long-run and short-run shock correlation ν −0.25
Volatility level σ 0.0032
Volatility persistence ϕ 0.996
Volatility of volatility ω 1.15 × 10−6

Implied Values:
Wealth-cons. loading on long-run growth Wx 31.07
Wealth-cons. loading on variance Wσ −5031
Backus-Smith coefficient β BS −1.3975
Mean wealth-cons. ratio exp(µwc ) 579.1
Campbell-Shiller constant κ0c 0.0127
Campbell-Shiller constant κ1c 1.0017
SDF loading on short-run growth shock −0.0256
SDF loading on long-run growth shock −0.1203
SDF loading on volatility shock 0.1273

3.B.2 Macro Synthesis


It is worth noting that this model does not start with endowment
or production processes and household optimization that gener-
ate endogenous consumption processes. Instead, the consumption
processes are assumed to be exogenous. Subsequent works such as
Colacito and Croce [2013], Colacito, Croce, Ho, and Howard [2018b]
embed the long-run risk in the endowment or production processes
instead of the consumption processes. While these alternative set-
tings bring the model closer to the baseline model we considered
in Section 1.A, it also makes the model more difficult to solve and

Electronic copy available at: https://ssrn.com/abstract=4668578


risk premia and factor structure 67

requires us to rely on numerical solutions.


In this long-run risk model we considered, the exogenous variables
include the short-run consumption growth shocks, the long-run
consumption growth shocks, and the variance shocks:

(ε g,t , ε∗g,t , ε x,t , ε∗x,t , ε w,t , ε∗w,t )∞


t =0 .

There are 7 endogenous variables in each period t:

(mt , rtc , wct , m∗t , rtc∗ , wc∗t , et )∞


t =0 .

The model implies the following 6 equations in each period, in-


cluding 3 for the home country in period t,

rtc+1 = κ0c + ∆ct+1 + log(wct+1 ) − κ1c log(wct ),


θ
mt+1 = θ log δ − ∆c + (θ − 1)rtc+1 ,
ψ t +1
1 = Et [exp(mt+1 + rtc+1 )],

and 3 for the Foreign country in period t,


∗ c∗ ∗ ∗ c∗ ∗
rtc+ 1 = κ0 + ∆ct+1 + log( wct+1 ) − κ1 log( wct ),
θ ∗
m∗t+1 = θ log δ − ∆c ∗
+ (θ − 1)rtc+ 1,
ψ t +1
1 = Et [exp(m∗t+1 + rtc+

1 )],

Moreover, we assume that the markets are complete, so we have


one more equation that relates the exchange rate movement to the
SDFs:

∆et+1 = mt+1 − m∗t+1 .

3.B.3 Exchange Rate Dynamics


Now that we obtain a simple solution of the SDF, we are ready to
characterize the exchange rate dynamics.

Proposition 3.5. Assuming complete markets, the real exchange rate


movement is
 
θ
∆et+1 = − + (θ − 1)[1 + Wx (ρ − κ1c )] ( xt − xt∗ ) + {Wσ (ϕ − κ1c )(θ − 1)} (σt2 − σt∗2 )
ψ
− γ(σt ε g,t+1 − σt∗ ε∗g,t+1 ) + (θ − 1)Wx φe (σt ε x,t+1 − σt∗ ε∗x,t+1 ) + (θ − 1)Wσ ω (ε w,t+1 − ε∗w,t+1 ),

the currency risk premium is


def 1
rpt = Et [rxt+1 ] = (vart (m∗t+1 ) − vart (mt+1 ))
2
1h 2 i
= γ + (θ − 1)2 Wx2 φ2e − 2νγ(θ − 1)Wx φe (σt∗2 − σt2 ),
2

Electronic copy available at: https://ssrn.com/abstract=4668578


68 jiang – lecture notes on international finance

the interest rate differential is


   
∗ 1 ∗ 1 ∗
rt − rt = −Et [mt+1 ] − vart (mt+1 ) − −Et [mt+1 ] − vart (mt+1 )
2 2
 
θ
= − + (θ − 1)[1 + Wx (ρ − κ1c )] ( xt∗ − xt )
ψ
 
1 2 1
+ Wσ (ϕ − κ1 )(θ − 1) + γ + (θ − 1) Wx φ − νγ(θ − 1)Wx φ (σt∗2 − σt2 ),
c 2 2 2
2 2

and the Backus-Smith coefficient is

covt (∆et+1 , ∆c∗t+1 − ∆ct+1 )


β BS
t =
vart (∆c∗t+1 − ∆ct+1 )
φe (1 − θ )Wx  
= γ+ν ∗ ∗ σt2 + σt∗2 .
vart (σt ε g,t+1 − σt ε g,t+1 )

The proof is presented in Appendix A.7. When the markets are


complete, we have a simple mapping from the SDF shocks to ex-
change rate movements. Specifically, as negative home short-run and
long-run consumption shocks and positive home volatility shocks
raise the marginal utilities of home households, they lead to home
currency appreciation. In comparison, the home currency’s risk pre-
mium only depends on the second-order moments, and it is de-
creasing in home consumption volatility and increasing in foreign
consumption volatility.
The interest rate differential also loads negatively on home con-
sumption volatility, which is also driven by a higher precautionary
saving when home consumption volatility is high. In this way, the
stochastic volatility drives common movements in currency risk pre-
mia and interest rate differentials, giving rise to an explanation of the
forward premium puzzle.
Finally, the correlation between long-run and short-run con-
sumption shocks ν plays a key role in the Backus-Smith coeffi-
cient. Consider the following benchmark: suppose the consump-
tion growth process has no long-run growth or stochastic volatility:
∆c = µ g + σg ε g,t+1 . Then, the Backus-Smith coefficient is
 
−(θ 1 − ψ1 − 1)σg2 vart ((ε∗g,t+1 − ε g,t+1 )))
β BS
t = = γ,
σg2 vart ((ε∗g,t+1 − ε g,t+1 ))

which is always positive.


φe (1−θ )Wx
σt2 + σt∗2 is always

In our calibration, the term vart (σt∗ ε∗g,t+1 −σt ε g,t+1 )
positive. So, a negative correlation between long-run and short-run
consumption shocks ν lowers the Backus-Smith coefficient below γ.
In fact, this correlation is negative enough in our calibration, which
gives rise to a negative Backus-Smith coefficient. In other words,

Electronic copy available at: https://ssrn.com/abstract=4668578


risk premia and factor structure 69

when the long-run consumption shock goes in the opposite direction


than the short-run consumption shock, it disentangles the link be-
tween the spot consumption growth and marginal utilities: a higher
short-run consumption growth at the expense of lowering long-run
growth is bad news. It raises marginal utilities and leads to currency
appreciation.
This is the key mechanism emphasized by Colacito and Croce
[2013], who also provide a reason for the negative correlation: an
increase in the home households’ long-run consumption growth
leads to an increase in their continuation utility and a decline in their
marginal utility. With perfect risk-sharing, the social planner should
reduce the spot consumption allocated to the home country, leading
to a decline in the home households’ short-run consumption growth.
At the same time, these resources flow to the foreign country that
receives relatively worse news.

3.B.4 Other Works


Colacito and Croce [2011], Bansal and Shaliastovich [2013] consider
similar settings with exogenously given consumption dynamics.
Bansal and Shaliastovich [2013] further consider the inflation dynam-
ics, which allows them to price nominal bonds and exchange rates.
Colacito and Croce [2013], Colacito, Croce, Ho, and Howard [2018b]
embed this setting in endowment economies with long-run risks and
recursive preferences. Colacito, Croce, Gavazzoni, and Ready [2018a]
introduce heterogeneity in the risk loadings of different countries’
long-run consumption shocks.
Parallel to the long-run risk model, other models have also been
proposed to describe the dynamics of currency risk premia, such as
habit formation [Verdelhan, 2010, Heyerdahl-Larsen, 2014, Stathopou-
los, 2017], demand shocks [Pavlova and Rigobon, 2007], rare disasters
[Farhi and Gabaix, 2016].

3.C Currency Risk Premia in the Cross-Section

The literature on international finance has also provided several


mechanisms that generate persistent differences in currency risk
premia across currencies. These mechanisms help explain the carry
trade and the factor structure in currency returns. In this section,
we consider such a mechanism based on the global trade network
[Richmond, 2019] and Jiang and Richmond [2023b] in particular.
This approach is based on a large literature in macro that studies
production networks [Long Jr and Plosser, 1983, Foerster, Sarte, and
Watson, 2011, Gabaix, 2011, Acemoglu, Carvalho, Ozdaglar, and

Electronic copy available at: https://ssrn.com/abstract=4668578


70 jiang – lecture notes on international finance

Tahbaz-Salehi, 2012, Chaney, 2014, Baqaee and Farhi, 2019].


Time is discrete and infinite, indexed by t. There is no storage
technology that allows agents to transfer goods across periods. There
are N countries, each populated by a unit mass of households and
indexed by i.
There are two types of goods: intermediate and final consumption
goods. Each country has a distinct intermediate good that is used for
the production of other intermediate goods. The active households
undertake the production of intermediate goods in each country.
Consumption goods are assembled as a basket of intermediate goods.
We first turn to the production of intermediate goods.

3.C.1 Intermediate Goods and the Production Process

Households in each country produce a distinct intermediate tradable


good using labor and tradable intermediate goods from different
countries. Each active household in country i has an identical Cobb-
Douglas production function of the form
!
N
i
xit = ait (ℓit )θ ∏ ( xijt )wij , (3.9)
j =1

where ait is the productivity level, ℓit is the labor input, and xijt is the
quantity of the intermediate goods produced by country j that are
used as production inputs in country i. The parameter θ i measures
the contribution of country i’s labor, and the parameters wij measure
the contribution of each country j’s input. We assume that

N
θi + ∑ wij = 1 and θ i , wij > 0,
j =1

so that the production function has constant returns to scale. We


collect the production parameters for intermediate goods wij into a
matrix W.

3.C.2 Final Consumption Goods and Household Preferences

Households in country i assemble their final consumption good from


different countries’ intermediate goods. Let cit denote the active
households’ quantity of the final consumption good, which is derived
from a Cobb-Douglas production function:

N
cit = ∏(cijt )vij , (3.10)
j =1

Electronic copy available at: https://ssrn.com/abstract=4668578


risk premia and factor structure 71

where cijt is the quantity of the intermediate goods produced by


country j used as the production inputs by i. The parameters satisfy

N
∑ vij = 1 and vij > 0.
j =1

We collect the production parameters for final goods vij into a matrix
V. Along with the intermediate goods’ production weights W, the
matrix V will be a key source of international comovements in the
model.
The households have log preferences over their aggregate con-
sumption, and discount future utility at rate β. The utility function
is

∑ δt log cit .
t =1

The households have access to complete financial markets, where


they can trade contingent claims with the households in any country.

3.C.3 Market Clearing


We study the competitive equilibrium defined in the usual fashion:
All households maximize their utilities taking prices as given, and
market clearing conditions for each good and labor are satisfied.
The market clearing condition for country i’s intermediate good is

N  

j j
xit = cit + xit + dit .
j =1

On the left-hand side, we have the total output of country i’s inter-
mediate good. On the right-hand side, we have the demand for this
good from each country’s consumption and production sectors. We
have an additional term dit as a reduced-form proxy for demand
shocks. This term can represent government taxation and spending
as in Acemoglu, Akcigit, and Kerr [2016], or within-country trans-
fer to inactive/hand-to-mouth investors as in Jiang and Richmond
[2023b].
Labor supply is fixed. The market clearing condition for country
i’s labor is
i
ℓit = ℓ .

3.C.4 Macro Synthesis


The competitive equilibrium solution in complete market is character-
ized by the solution to a social planner’s problem. The Lagrangian is

Electronic copy available at: https://ssrn.com/abstract=4668578


72 jiang – lecture notes on international finance

given by
! !
N N N
i
 
∑ πi log cit + φit θi
∏(xijt )wij −∑
j j
ait (ℓit ) cit + xit − dit + χi (ℓ − ℓit ),
i =1 j =1 j =1

The exogenous variables are the productivity shocks and the de-
mand shocks:

( ait , dit )∞
t =0 , i = 1, 2, . . . , N.

There are 2N 2 + 5N endogenous variables in each period t:

(cit , cijt , xit , xijt , ℓit , φit , χit )∞


t =0 , i, j = 1, 2, . . . , N.

The model implies the following 2N 2 + 5N equations in each


period, which includes N consumption aggregation equations
N
cit = ∏(cijt )vij ,
j =1

N production equations
!
N
θi
xit = ait (ℓit ) ∏(xijt )wij ,
j =1

2N 2 + N first-order conditions
j j
w.r.t. cit : π j v ji (cit )−1 = φit ,
j j j j
w.r.t. xit : φt x̄t w ji ( xit )−1 = φit ,
w.r.t. ℓit : φit x̄ti θ i (ℓit )−1 = χit ,

N intermediate good market clearing conditions


N  

j j
xit = cit + xit + dit ,
j =1

N labor market clearing conditions


i
ℓit = ℓ .

3.C.5 Trade Network and Shock Transmission


We next characterize the equilibrium of the model and show how the
structure of global production transmits shocks across countries. A
variable with its country index omitted is a vector. For example, ct
is the vector where each element is cit . A capitalized parameter with
two country indices omitted is a matrix. For example, W is the matrix
with each element being wij .

Electronic copy available at: https://ssrn.com/abstract=4668578


risk premia and factor structure 73

Since the households have access to the complete markets, the


competitive equilibrium can be characterized by the solution to a
social planner’s problem. Within period t, the social planner’s La-
grangian is
! !
N N N 
i

i i θi
∑ π log ct + φt at (ℓt ) ∏(x jt ) − ∑ cit + xit − dt + χit (ℓ − ℓit ),
i i i i wij j j i
i =1 j =1 j =1

where π i is the Pareto weight that the social planner assigns to coun-
try i. This Pareto weight is determined by the initial level of wealth
held by each country’s households.
A key variable that emerges from the model is a function of trade
network parameters for intermediate usages, W, and consumption,
V. We define the network profile matrix as:

H ≡ V ( I − W ) −1 .

Solving the planner’s Lagrangian, we derive the following lemma


to characterize the real quantities in equilibrium. We omit the time
subscript for notational convenience.

Proposition 3.6. In equilibrium, the vector of each country’s active house-


holds’ log consumption is

log c = κ c + H (log a − θ log( H ′ π + ( I − W ′ )−1 d)), (3.11)

for some vector of constants κ c .

The proof is presented in Appendix A.8. This result shows that


the structure of the global trade network, summarized by the net-
work profile matrix H = V ( I − W )−1 , plays an important role
in transmitting both the supply shocks (a) and the demand shocks
(d). The network profile matrix combines the production trade net-
work W and the consumption trade network V. The term ( I − W )−1
is commonly known as the Leontief inverse, which summarizes
the transmission of the productivity shocks. Loosely speaking,
( I − W )−1 = I + W + W 2 + W 3 + . . ., where W reflects the trans-
mission from the exporting country to the importing country, and
W k for k > 1 reflects the transmission of shocks due to higher order
linkages.
Moreover, in this model, the supply shocks a are transmitted via
the network matrix H = V ( I − W )−1 , whereas the demand shocks
d are first transmitted via the transpose of the production network,
i.e., ( I − W ′ )−1 . This is a general feature of network transmission.
Intuitively, supply shocks propagate downstream, meaning that a
productivity shock in one country impacts the production in other
countries which import from this country. In comparison, demand

Electronic copy available at: https://ssrn.com/abstract=4668578


74 jiang – lecture notes on international finance

shocks propagate upstream, meaning that a demand shock in one


country impacts the production in other countries which export to
this country. The transpose operation reverses the direction of the
trade linkages, so that the demand shocks are transmitted in the
opposite direction of the production linkages.
Taking a first-order approximation, we can express the active
households’ log consumption growth as a linear function of the sup-
ply shocks ∆ log a and the demand shocks ∆d:
 
θ
∆ log c ≈ H ∆ log a − ′ ( I − W ′ )−1 ∆d , (3.12)

where H1′ π is a diagonal matrix whose element (i, i ) is { H1′ π } . This


i
expression shows that each country’s consumption growth is driven
by a combination of demand and supply shocks of, potentially, all
other countries globally. The specific combination of shocks that a
country is exposed to is primarily determined by the structure of
global trade linkages summarized by the network profile matrix H.
This common exposure to country-level shocks that arises due to
trade linkages gives rise to international comovements.

3.C.6 Characterizing Consumption and Exchange Rate Comovements


We are interested in characterizing international comovements in
consumption growth and exchange rates. We define the bilateral log
i/j
real exchange rate between countries i and j, et , as the log price of
country j’s consumption bundle per unit of country i’s consumption
i/j
bundle. An increase in et implies an appreciation of country i’s real
exchange rate relative to country j’s. Since markets are complete, the
bilateral exchange rate movement is determined by the difference in
the growth rates of marginal utilities:
i/j j
∆et = ∆ log ct − ∆ log cit .

We define each country’s currency base factor, ∆eit , as the equal-


weighted average log real exchange rate change against all countries,
including itself:
N
1
∑ ∆et
i/j
∆eit = .
N j =1

Let Var [z] denote the variance-covariance matrix for a vector of


stochastic variables z. We define a general measure of closeness be-
tween two countries i and j as
   
θ ′ −1 ′
C(i, j) ≡ H · Var ∆ log a − ′ ( I − W ) ∆d · H , (3.13)
Hπ ij

Electronic copy available at: https://ssrn.com/abstract=4668578


risk premia and factor structure 75

which measures the pairwise similarity of the shocks which drive


countries’ quantities. Then, we obtain the following proposition:
Proposition 3.7. (a) Closer countries have more correlated active household
consumption growth:
 
j
cov ∆ log cit , ∆ log ct = C(i, j).

(b) Closer countries have more correlated currency base factors and less
volatile bilateral real exchange rate movements:
 
j
cov ∆eit , ∆et = C(i, j) − C(i ) − C( j) + κ e ,
i/j
var (∆et ) = −2C(i, j) + C(i, i ) + C( j, j),

where κ e is a constant:
N N
1
κe =
N2 ∑ ∑ C(k, ℓ),
k =1 ℓ=1

and C(i ) is the average closeness between country i and all countries:
N
1
C(i ) =
N ∑ C(i, j).
j =1

(c) The variance of country i’s currency base factor is


 
var ∆eit = −2C(i ) + C(i, i ) + κ e .

Fixing its closeness to itself C(i, i ), the country’s currency base factor is
less volatile if it has a higher average closeness.
The proof is presented in Appendix A.9. This proposition shows
that the consumption and exchange rate covariance is directly related
to the closeness matrix C , which depends on both the covariance
structures of supply and demand shocks as well as the structure of
the trade network that propagates these shocks.

3.C.7 Currency Risk Premia and Endogenous Common Factors


We set the U.S. dollar as the base currency.
Proposition 3.8. Currency i’s expected excess return is
def
h i 1
rpi/$
t = Et rxti/$ $ i
+1 = 2 ( vart ( mt+1 ) − vart ( mt+1 ))
1
= (C($, $) − C(i, i )) ,
2
and its interest rate is
1
rti = −Et [mit+1 ] − vart (mit+1 )
2
1
= − log δ − C(i, i )
2

Electronic copy available at: https://ssrn.com/abstract=4668578


76 jiang – lecture notes on international finance

The proof is presented in Appendix A.10. This result shows that,


fixing the base currency, a currency i’s risk premium is decreasing in
its closeness to itself, i.e., C(i, i ). Following Richmond [2019], we may
call this object country i’s trade centrality. The trade centrality sum-
marizes how this country’s position and relation to other countries in
the trade network affects its currency risk premium. A country with
a large C(i, i ) is a central country and has a low currency risk pre-
mium, whereas a country with a small C(i, i ) is a peripheral country
and has a high currency risk premium.
We consider the following numerical example to understand what
centrality captures. There are 4 countries. We assume the home bias
in consumption tends to 1 in the limit, i.e., V → I, and that the
combined
h supply and demand i shocks are i.i.d. across countries,
Var εS − Hθ′ π ( I − W ′ )−1 ε D = Var [ε] = I. These simplifying assump-
tions allow us to focus on the production trade network W, which is
given by
 
1 0 0 0
0.25 0.75 0 0 
W = (1 − θ )  ,
 
 0.5 0 0.5 0 
0.75 0 0 0.25

with θ = 0.5.
In this trade network, countries 2, 3, and 4 rely on country 1’s
export as intermediate input. The dependency is increasing from
country 2 to 4. The implied centrality measure is
   
C(1, 1) 4.00
C(2, 2) 2.72
= ,
   
C(3, 3) 2.22

C(4, 4) 2.04

which implies that country 1 is the most central country because


it provides essential production input to all other countries, and
country 4 is the most peripheral country because it is very reliant
on country 1’s export. Accordingly, country 1’s shock is the most
systematic and as a result it has the lowest currency risk premium.
This simple network structure also gives rise to a factor structure
(i )
in exchange rate movements. Let ε t+1 denote the combined supply
and demand shock in country i. Then, the bilateral exchange rate
movements also load on these shocks:
(1) (2)
∆e1/2
t+1 = −1.60ε t+1 + 1.60ε t+1 ,
(1) (3)
∆e1/3
t+1 = −1.33ε t+1 + 1.33ε t+1 ,
(1) (4)
∆e1/4
t+1 = −1.14ε t+1 + 1.14ε t+1 ,

Electronic copy available at: https://ssrn.com/abstract=4668578


risk premia and factor structure 77

(1)
and the central country’s shock ε t+1 becomes the common factor in
the cross-section of exchange rates. In fact, even when we consider
the bilateral pair between countries 2 and 3 that does not involve the
(1)
central country 1, ε t+1 still shows up:

(1) (2) (3)


∆e2/3
t+1 = 0.27ε t+1 − 1.60ε t+1 + 1.33ε t+1 .

In comparison, all other (peripheral) countries’ consumption load


on this common factor, and have no further pairwise comovements
beyond that induced by their exposure to the common factor. The
(2) (3) (4)
shocks ε t+1 , ε t+1 , and ε t+1 are idiosyncratic in the sense that only one
country is exposed to each of them.
This expression demonstrates how our general equilibrium model
produces exchange rate dynamics like those found in factor mod-
els of exchange rates in Section 3.A. Specifically, country 1’s shock
becomes the global shock εw t+1 , and other countries’ shocks become
i
country-specific shocks ε t+1 . In this way, common factors in exchange
rates arise from the network propagation of independent shocks
across countries. This example could also be generalized to allow for
multiple factors once we impose a richer dependence structure in the
consumption and production networks.
In the general model with dense input-output linkages in W and
V, there are no truly idiosyncratic shocks, but some shocks are more
important for the variation of quantities and asset prices, and this
information can be summarized by the closeness measure.

3.C.8 Other Works


Other works explain cross-country differences in currency risk pre-
mia based on different mechanisms, such as size [Martin, 2011, Has-
san, 2013], commodity goods [Ready, Roussanov, and Ward, 2017b,a],
and financial intermediation and risk-bearing capacity [Maggiori,
2017, Wiriadinata, 2018, Fang, 2021].

3.C.9 A Note on Log Preference


In this model, we assume that agents have log preferences: u(c) =
log(c). As with all modeling assumptions, this has pros and cons.
In this context with Cobb-Douglas aggregation of goods, the key
implication of the log preferences is that the households will allocate
a constant share of their budget to each consumption goods. That is,
j
cit π j v ji
=
ck jit π k vki
is a constant.

Electronic copy available at: https://ssrn.com/abstract=4668578


78 jiang – lecture notes on international finance

If we aggregate this result across all countries, we can show that


the value of a country’s aggregate output in the common numéraire
is

φ t x t = H ′ π + ( I − W ′ ) −1 d t ,

where φt x t is a vector whose i-th element is φit xit . That is, supply
shocks at do not affect the value of a country’s aggregate output;
only demand shocks do. This is because, while a higher supply shock
raises the quantity of a country’s output, it proportionally lowers
its price, so that the value of the output is unchanged [Cole and
Obstfeld, 1991]. In contrast, a higher demand shock raises the price
of the local goods while leaving the quantity unchanged, so that the
value of the output increases.
While this is a strong assumption, it is technically convenient.
From the production equations (3.9) and the social planner’s first-
order conditions, the country-level production quantities satisfy
!
N φit x̄ti wij
log x t = log at + θ log ℓt + ∑ wij log
i i i i
j j j
,
j =1 φt x̄t / x̄t

which allows us to solve the vector of the production quantity log x t


as a linear transformation of the vector of the production value
log( φt x̄t ), and obtain a closed-form characterization of the equilib-
rium allocations.
This discussion centers on the implication of log preferences for
the households’ within-period choices. Log preferences also have
important implications for the households’ intertemporal choices. As
we will see in Section 9.B, the households with log preferences are
“myopic” because they do not pay attention to the time variations in
investment opportunities.

3.D Currency Risk Premia in the Long Run

So far we have been focusing on the currency risk premium and


currency factor structure in one-period returns. In this section, we
turn to the long-run currency risk premium and the long-term bond
yields. This approach builds on works related to the permanent and
the transitory components of the SDFs, such as Alvarez and Jermann
[2005], Hansen and Scheinkman [2009], Backus, Boyarchenko, and
Chernov [2018], Lustig, Stathopoulos, and Verdelhan [2019]. We
start with some additional notations. First, we introduce the multi-
period bonds. Let rt (h) denote the log yield on the risk-free bond
with maturity h. The bond price is given by

pt (h) = exp(−rt (h)h).

Electronic copy available at: https://ssrn.com/abstract=4668578


risk premia and factor structure 79

If we allow the home and foreign households to trade this bond in


the baseline economy, we can extend the household budget constraint
(1.1) to

pt yt + ∑ b H,t−1 (h) exp(rt−1 (h)) + ∑ bF,t−1 (h) exp(rt∗−1 (h) − et ) = ct + ∑ b H,t (h) + ∑ bF,t (h) exp(−et ).
h h h h

Then, the Lagrangian implies the following Euler equations


 ′ 
h u (ct+h )
1 = Et δ exp(rt (h)h) ,
u′ (ct )
u′ (c∗t+h )
" #
h
1 = Et δ exp(−∆et,t+h + rt (h)h) ,
u′ (c∗t )

def
where ∆et,t+h = et+h − et denotes the h-period log exchange rate
movement.
If we define the multi-horizon SDFs as
def u′ (ct+h )
exp(mt,t+h ) = δh ,
u′ (ct )
def u′ (c∗ )
exp(m∗t,t+h ) = δh ′ t+∗ h ,
u (ct )
then, the Euler equations can be written as

1 = Et [exp(mt,t+h ) exp(rt (h)h)] ,


h i
1 = Et exp(m∗t,t+h ) exp(−∆et,t+h + rt (h)h) .

It is useful to compare two different multi-horizon currency trad-


ing strategies from period 0 to period h. In the first strategy, the
home households buy the foreign one-period bond and sell the home
one-period bond in each period t. The cumulative log return from
this strategy is
h h
def
∑ (∆et + rt−1 (1) − rt∗−1 (1)) = ∆e0,h + ∑ (rt−1 (1) − rt∗−1 (1)).
(1)
rxh (h) =
t =1 t =1

In the second strategy, the home households buy the foreign h-


period bond and sell the home h-period bond in period 0, and hold
both positions until maturity. The cumulative log return from this
strategy is
(h) def
rxh (h) = ∆e0,h + h(r0 (h) − r0∗ (h)).

While both strategies are exposed to the same exchange rate risk
captured by ∆e0,h , the cumulative interest rate differential between
one-period bonds, i.e., ∑th=1 (rt−1 (1) − rt∗−1 (1)), can be different from
the interest rate differential between h-period bonds, i.e., h(r0 (h) −
r0∗ (h)). So, these two strategies generate different returns.

Electronic copy available at: https://ssrn.com/abstract=4668578


80 jiang – lecture notes on international finance

3.D.1 Economic Set-up


We begin with an economic set-up to motivate the SDF decompo-
sition that we will consider next. We begin with a one-country en-
dowment economy with CRRA preferences. The endowment yt has
a permanent component ytP and a transitory component ytT , which
satisfy

log yt = log ytP + log ytT ,


log ytP = log ytP−1 + µ + νP ε Pt ,
log ytT = ϕ log ytT−1 + νT ε Tt ,

where 0 ≤ ϕ < 1 and ε Pt and ε Tt are serially independent and nor-


mally distributed random variables with mean zero and standard
deviation of one.
The representative households’ preference is
1− γ
ct
u(ct ) = δt .
1−γ

With market clearing yt = ct , the log marginal utility can be written


as
def
mct = log u′ (ct ) = t log δ − γ log yt ,

which we decompose into a permanent component mctP and a transi-


tory component mctT :

mct = mctP + mctT ,

where
mc T
 
ϕ log β log β
mctP = − + + log δ − t − γ log ytP ,
1−ϕ 1−ϕ1−ϕ 1−ϕ
mc T ϕ log β log β
mctT = − + t − γ log ytT ,
1−ϕ 1−ϕ1−ϕ 1−ϕ

where the constant β and mc T satisfies

log β 1
= (γνP )2 + log δ − µγ,
1−ϕ 2
1 ϕ
mc T = − γ2 νT2 − γ2 νP νT ρ TP + log β.
2 + 2ϕ 1−ϕ

The permanent component mctP contains the shock ε Pt to the per-


manent component of the endowment, whereas the transitory com-
ponent mctT contains the shock ε Tt to the transitory component of the
endowment. The two marginal utility components also contain addi-
tional constants and time trends, such that the permanent component

Electronic copy available at: https://ssrn.com/abstract=4668578


risk premia and factor structure 81

of the marginal utility, i.e., exp(mctP ), is a Martingale:


1
mctP = mctP−1 − σP2 + σP ε Pt , (3.14)
2
mctT = mc T + ϕmctT−1 + t log β + σT ε Tt , (3.15)

where

σP = −γνP ,
σT = −γνT .

In this way, our simple endowment economy motivates a decom-


position of the marginal utility into the permanent and transitory
components, which we study further in the next subsection.

3.D.2 SDF Decomposition


Alvarez and Jermann [2005] propose a general decomposition of the
marginal utility:

mct = mctT + mctP ,

where mctT denotes the transitory component of the SDF,

δ̃t+h
exp(mctT ) = lim , (3.16)
h→∞ pt ( h)

where the constant δ̃ is chosen to satisfy the regularity condition


0 < limh→∞ pt (h)/δ̃h < ∞. The transitory component is closely
related to the return to the infinite-horizon bond:
p t +1 ( h − 1 )
lim = exp(mctT − mctT+1 ). (3.17)
h→∞ pt ( h)
The permanent component of the marginal utility is a martingale
by construction:
pt ( h) Et [exp(mct+h )]
exp(mctP ) = lim exp(mct ) = lim ,
h→∞ δ̃t+h h→∞ δ̃t+h
where we used pt (h) = Et [exp(mt,t+h )] = Et [exp(mct+h − mct )].
The marginal utility components mctT and mctP we considered in
Eq. (3.14) and (3.15) provide a concrete example of the decomposi-
tion. Let us use ρ TP to denote the correlation between the permanent
shock ε Pt and the transitory shock ε Tt . Let σTP = σT σP ρ TP .
In this economy, the bond price satisfies

1 = Et [exp(mt,t+h ) pt (h)].

Evaluating this conditional expectation and take the horizon h to


infinity, we obtain the following proposition about the long-term
yield.

Electronic copy available at: https://ssrn.com/abstract=4668578


82 jiang – lecture notes on international finance

Proposition 3.9. (a) The long term bond price is given by


!
def 1 − ϕh T t + h − ϕ h ( t + 1) ϕ − ϕh
pt (h) = exp(−rt (h)h) = exp mc + − log β + (ϕh − 1)mctT
1−ϕ 1−ϕ (1 − ϕ )2
!
1 − ϕ2h 2 1 − ϕh
+ σ + σTP .
2 − 2ϕ2 T 1−ϕ

def
(b) Let δ̃ = β1/(1−ϕ) . Then, the long-term bond yield satisfies

lim rt (h) + log δ̃ = 0, (3.18)


h→∞

and

1 1
lim (rt (h) + log δ̃)h =
h→∞
∑ ϕi σT εTt−i − 2 − 2ϕ2 σT2 − 1 − ϕ σTP . (3.19)
i =0

The proof is presented in Appendix A.11. This result offers sev-


eral insights for thinking about the valuation of the long-term bond.
Part (a) of the proposition shows that the variations in the bond yield
rt (h) only depend on the variations in the transitory SDF component
mctT , both when h is finite and when we take h to infinity. This is be-
cause movements in the transitory SDF component will mean-revert
in the future, which contains information about the future SDF move-
ment mt,t+h and therefore affects the bond yield. In comparison, the
permanent SDF component does not affect the bond yield, because
they permanently shift the marginal utility level mct , and contain no
information about the future SDF movement mt,t+h by construction.
While the permanent SDF component does not affect the bond
yield, it manifests itself in the exchange rate. When markets are com-
plete, the exchange rate (in log) is equal to the log SDF differential.
As a result, a shock to the permanent component of a country’s SDF
will have a permanent impact on the bilateral exchange rate unless it
is offset by a shock to the permanent component of the other coun-
try’s SDF.
In part (b) of the proposition, Eq. (3.18) shows that the long-term
bond yield rt (h) converges to
log β 1
− log δ̃ = − = − σP2 − log δ + µγ
1−ϕ 2
as the horizon h approaches infinity. This value depends on the
volatility of the permanent SDF shock, σP , but it does not depend
on the volatility of the transitory SDF shock, σT , or the realization
of the transitory SDF shocks, ε Tt . In particular, this value is equal
to the one-period yield in the absence of the transitory endowment
shocks. So, while the permanent SDF component does not affect the
variations in the bond yield, it affects its unconditional level.

Electronic copy available at: https://ssrn.com/abstract=4668578


risk premia and factor structure 83

Eq. (3.19) shows that the long-term bond price, after taking out
the deterministic trend (log δ̃)h, converges to a function of the past
transitory shocks ε Tt−i . This quantity is closely related to the exchange
rate level, as we will see next. Moreover, using the definition of the
def
bond price pt (h) = exp(−rt (h)h), we can show that the bond price
scaled by δ̃ is finite:

0 < lim pt (h)/δ̃h < ∞.


h→∞

Before we move on to the two-country setting, let us consider two


special cases. The first special case is one in which the transitory
endowment shock ε Tt is always zero. In this case, the SDF is a martin-
gale, and the bond yields are constant across time and across tenors:

1
rt (h) = − σP2 − log δ + µγ.
2
The second special case is one in which the permanent endow-
ment shock ε Pt is always zero. In this case, according to Eq. (3.17),
reproduced below,

p t +1 ( h − 1 )
lim = exp(mctT − mctT+1 ),
h→∞ pt ( h)

the return to the infinite-horizon bond is exposed to the shock to the


transitory component of marginal utility, which contains all the SDF
risk there is. As a result, the infinite-horizon bond is the riskiest asset
in this economy, and should therefore earn the highest Sharpe ratio.
If the 30-year bond is a good proxy for the infinite-horizon bond, its
low return in the data suggests that this special case is not a good
description of the world.

3.D.3 Two Countries


Now, we consider the pair of SDFs between the home and the foreign
countries. We also decompose the foreign marginal utility as

mc∗t = mctT,∗ + mctP,∗ .

When markets are complete, the exchange rate is equal to the ratio
of the two countries’ SDFs:

∆et = ∆mct − ∆mc∗t = (∆mctP − ∆mctP,∗ ) + (∆mctT − ∆mctT,∗ ).

For notational convenience, we define


def
∆etP = ∆mctP − ∆mctP,∗ ,
def
∆etT = ∆mctT − ∆mctT,∗ .

Electronic copy available at: https://ssrn.com/abstract=4668578


84 jiang – lecture notes on international finance

From the definition of the transitory SDF component in Eq. (3.16),


the transitory component of the exchange rate movement is
def
∆etT = ∆mctT − ∆mctT,∗
δ̃t+h (δ̃∗ )t+h
= lim ∆ log − ∆ log ∗ = log δ̃ − log δ̃∗ + lim h(∆rt (h) − ∆rt∗ (h)).
h→∞ pt ( h) pt ( h) h→∞

If two countries share the same martingale component of the pric-


ing kernel, then the resulting exchange rate is stationary up to a
deterministic time trend. In this case,
∆et = ∆etT = log δ̃ − log δ̃∗ + lim h(∆rt (h) − ∆rt∗ (h)).
h→∞

For the exchange rate movement to be finite, limh→∞ h(∆rt (h) −


∆rt∗ (h)) must be finite, which means that a stationary exchange rate
in a complete-market economy requires the home and foreign long-
term bond yields to be identical in the limit:
lim (∆rt (h) − ∆rt∗ (h)) = 0. (3.20)
h→∞

Finally, assuming log δ̃ = log δ̃∗ and iterating this equation for-
ward, we obtain the following result [Lustig, Stathopoulos, and
Verdelhan, 2019].
Proposition 3.10. When markets are complete and the two countries’ SDFs
share the same permanent component, the deviation of the current exchange
rate level et from its long-run mean ē = lim j→∞ Et [et+ j ] is purely driven
by the infinite-horizon bond yield differential:
et − ē = lim h(rt (h) − rt∗ (h)). (3.21)
h→∞

We can think of Eq. (3.21) as the long-run uncovered interest rate


parity (long-run UIP) condition. This condition is related to the one-
period UIP condition:
et − Et [et+1 ] = rt − rt∗ .
In Section 1.B Eq. (1.13), we show that the one-period UIP condition
fails, and an additional risk premium term rpt is required on the
right-hand side. In comparison, Eq. (3.21) shows that, when the ex-
change rate is stationary, the uncertainty in the exchange rate level
over the infinite horizon vanishes from the pricing perspective, lead-
ing to zero risk premium. As a result, the exchange rate’s deviation
from the long-run mean is only determined by the long-term bond
yield differential.
Moreover, if we compare this result with the exchange rate ac-
counting formula in Proposition 1.2, reproduced below,
∞ ∞
et − ē = ∑ Et [rt+ j − rt∗+ j ] − ∑ Et [rpt+ j ],
j =0 j =0

Electronic copy available at: https://ssrn.com/abstract=4668578


risk premia and factor structure 85

we can see that the infinite-horizon bond yield incorporates infor-


mation about both the future short rates and the future currency risk
premia, i.e.,
∞ ∞
lim h(rt (h) − rt∗ (h)) =
h→∞
∑ Et [rt+ j − rt∗+ j ] − ∑ Et [rpt+ j ].
j =0 j =0

If we further decompose the infinite-horizon bond yield into an ex-


pected short rate component and a yield spread component, the bond
yield spread differential is related to the currency risk premium:

! !
h −1 h −1
lim
h→∞
hrt (h) − ∑ Et [ r t + j ] − lim
h→∞
hrt∗ (h) − ∑ Et [rt∗+ j ] = − ∑ Et [rpt+ j ].
j =0 j =0 j =0

In this equation, the left-hand side is the cumulative excess return


of holding the home bond until maturity minus that of holding the
foreign bond, and the right-hand side is the currency risk premium
derived from rolling over one-period bond positions across countries.
This equation shows that these two risk premium measures in the
bond market and in the currency market are closely related, provided
that markets are complete and the exchange rate is stationary.
Finally, the exchange rate stationarity can be imposed implicitly
by monetary policies. Let us consider the following special case.
Suppose the home and foreign SDFs have the following dynamics:

mt,t+1 = −µ − σε t+1 ,
m∗t,t+1 = −µ + ϕet .

While the home SDF is a random walk (plus a deterministic drift),


the foreign SDF is set up so that the foreign short rate responds to
the exchange rate level. Assuming ϕ > 0, the foreign central bank
lowers the local one-period interest rate rt∗ (1) = µ − ϕet when the
home currency is strong, which creates an expected appreciation of
the foreign currency.
In complete markets, the exchange rate movement is

∆et+1 = mt,t+1 − m∗t,t+1 = −ϕet − σε t+1 ,

which is a stationary AR(1) process. In other words, the foreign in-


terest rate’s response to the exchange rate level guarantees a stable
exchange rate in the long run. The long-run exchange rate is
def
ē = lim Et [et+ j ] = 0.
j→∞

This model allows us to solve the long-term bond yields in closed


form. The home country’s bond yields have a flat term structure:
1
rt ( h ) = µ − σ2 .
2

Electronic copy available at: https://ssrn.com/abstract=4668578


86 jiang – lecture notes on international finance

Conjecture that


Et [ Mt,t +h ] = exp( h ( a h + bh et ))

with the boundary condition a0 = b0 = 0. By iteration,

∗ ∗
Et [ Mt,t +h ] = Et [exp(− µ + ϕet )Et+1 [ Mt+1,t+h ]]
= Et [exp(−µ + ϕet ) exp((h − 1)( ah−1 + bh−1 et+1 ))]
exp(h( ah + bh et )) = Et [exp(−µ + ϕet + (h − 1) ah−1 + (h − 1)bh−1 (1 − ϕ)et − (h − 1)bh−1 σε t+1 )]
1
= exp(−µ + (h − 1) ah−1 + (h − 1)2 bh2−1 σ2 + ((h − 1)(1 − ϕ)bh−1 + ϕ)et )
2
which implies
!
σ2 1 − (1 − ϕ)2h 2 − 2(1 − ϕ ) h
ah = −µ + 2
− +h
2h 2ϕ − ϕ ϕ
1 − (1 − ϕ ) h
bh = .
h

As h → ∞, bh converges to 0, and ah converges to −µ + σ2 /2,


which implies

1
lim rt∗ (h) = lim (− ah + bh et ) = µ − σ2 = lim rt (h).
h→∞ h→∞ 2 h→∞

That is, the foreign long-term yield rt∗ (h) converges to the home long-
term yield rt (h) for large h, even when the next-period marginal
utility is stochastic in the home country (i.e., vart (mct+1 ) = σ2 ) while
it is fully predictable in the foreign country (i.e., vart (mc∗t+1 ) = 0).
This result confirms our earlier result in Eq. (3.20) that a stationary
exchange rate implies that the two countries’ long-term bond yields
converge to the same level.

3.D.4 Pricing Permanent Cash Flows


Let us return to the one-country setting with the SDF specification
y
according to Eq. (3.14) and (3.15). Let pt (h) denote the price of the
endowment yt+h in period t, which is equal to its present value:
y
pt (h) = Et [exp(mt,t+h )yt+h ].

We choose to study the claim to endowment because its cash flow


in log is cointegrated with the permanent component of the log SDF
after removing a deterministic trend. Note that

1
mctP = mctP−1 − σP2 + σP ε Pt ,
2
log yt = log ytP−1 + µ + νP ε Pt + log ytT ;

Electronic copy available at: https://ssrn.com/abstract=4668578


risk premia and factor structure 87

we can show that mctP /σP − (log yt − µt)/νP is a stationary process.


In comparison, the cash flow of a long-term bond is not cointegrated
with the permanent component of the log SDF.
The risk premia of the endowment strip and the long-term bond
is defined as the expected excess return of holding the asset for one
period. They are characterized by the following proposition.

Proposition 3.11. The risk premium of the h-period risk-free bond is


p ( h − 1)
 
def
rxt (h) = Et log t−1 − rt
pt ( h)
 
1 1 2( h −1)
= − ϕ σT2 + (1 − ϕh−1 )σTP .
2 2
The risk premium of the claim to the h-period endowment claim is
" y #
y def p t +1 ( h − 1 )
rxt (h) = Et log y − rt
pt ( h)
"  # " 2 #
2γ − 1 2 1 1 2( h −1) γ − 1 2 2 h −1 γ − 1
 
= σ + − ϕ σT + 1 − ϕ σTP .
2γ2 P 2 2 γ γ

As h → ∞, these risk premia converge to


1 2
lim rxt (h) = σ + σTP ,
h→∞ 2 T
y 2γ − 1 2 1 2
lim rxt (h) = σ + σ + σTP .
h→∞ 2γ2 P 2 T
The proof is presented in Appendix A.12. By comparing the
infinite-horizon risk premia on the bond and the endowment strip,
we can see that the bond risk premium does not load on the quantity
of the permanent risk, σP , in the limit, whereas the endowment risk
premium does. In most asset pricing models, the quantity of per-
manent risk is very large [Alvarez and Jermann, 2005] and the risk
aversion parameter γ is above 1/2, which generate a much higher
discount rate on the endowment strip than on the bond. As we will
see in Section 8.E, if other cash flows (such as the equity dividend
or government tax) are cointegrated with the endowment and hence
with the permanent component of the SDF, then, their risk premia
should converge to that of the endowment strip in the limit as well.
We notice that this risk premium does not depend on the transi-
tory component of the SDF as well as the transitory component of
the cash flow. In fact, for any cash flow stream that is cointegrated
with the endowment, the risk premium on its infinite-horizon strip is
y
also equal to limh→∞ rxt (h). Similarly, for any cash flow stream that
is stationary, the risk premium on its infinite-horizon strip is equal
to that of the infinite-horizon bond. We will use this result when
considering the government’s fiscal cash flows in Chapter 8.

Electronic copy available at: https://ssrn.com/abstract=4668578


4
Convenience Yields

Summary

• We deviate from the complete-market benchmark and introduce the bond convenience yield
λ∗t as a wedge in the households’ Euler equations for holding the home bond:

exp(−λ∗t ) = Et exp(m∗t+1 + ∆et+1 + rt ) .


 

• This wedge affects both the currency expected return:

Et [rxt+1 ] = rpt − λ∗t ,

and the exchange rate level:


∞ ∞ ∞
et = ∑ Et [rt+ j − rt∗+ j ] + ∑ Et [λ∗t+ j ] − ∑ Et [rpt+ j ] + ē.
j =0 j =0 j =0

• Under additional assumptions about the liquidity benefits of the currency forward, the conve-
nient yield can be measured from the Treasury CIP deviation xtTreas :

1
λ∗t = − x Treas .
1 − β∗ t

In this chapter, we consider an additional ingredient that drives


currency returns: the bond convenience yields, which reflect the
currencies’ non-pecuniary qualities including liquidity, safety, and the
ability to pledge as collateral.
This ingredient is motivated by the near-arbitrage spreads ob-
served in the data, which show that certain currencies and assets
earn different returns despite having similar payoffs. For example,
a large literature has found that the yields on U.S. Treasury debt are
traded below other benchmark interest rates [Longstaff, 2004, Kr-
ishnamurthy and Vissing-Jorgensen, 2012, Nagel, 2016, Van Binsber-
gen, Diamond, and Grotteria, 2022]. The U.S. Treasury also appears

Electronic copy available at: https://ssrn.com/abstract=4668578


convenience yields 89

more expensive than TIPS [Fleckenstein, Longstaff, and Lustig, 2014],


corporate bonds [Bai and Collin-Dufresne, 2019], foreign sovereign
bonds [Du, Im, and Schreger, 2018a, Jiang, Krishnamurthy, and
Lustig, 2018, 2021a, Koijen and Yogo, 2020], and duration-matched
stocks [Van Binsbergen, 2020]. We refer to the lower yield earned
by the U.S. Treasury as the convenience yield. Moreover, the conve-
nience yield is not specific to the U.S. Treasury, as other safe assets
such as high-grade dollar corporate bonds and bank deposits also
appear to earn lower yields [Liao, 2020, Gutierrez, Ivashina, and Salo-
mao, 2021].
To model these convenience yields, let rt denote the log Treasury
yield in the home country, and let ρt denote the log yield of a bench-
mark risk-free bond that does not contain any non-pecuniary bene-
fits. Both rates are one-period. The benchmark risk-free rate exactly
satisfies the households’ Euler equation:

1 = Et [exp(mt+1 + ρt )] .

As the Treasury is more expensive, the Treasury yield is lower than


the benchmark risk-free rate:

rt < ρt ,

which implies that the Euler equation fails for the Treasury yield:

1 = Et [exp(mt+1 + ρt )] > Et [exp(mt+1 + rt )] .

In other words, the Treasury yield gives rise to a wedge in the


Euler equation. To capture this wedge, it is useful to conceptualize a
convenience yield λt > 0, such that

exp(−λt ) = Et [exp(mt+1 + rt )] . (4.1)

This convenience yield describes the amount of risk-adjusted ex-


pected return that the households are willing to give up in order to
hold the Treasury debt. This equation will be central to organize our
discussion in this chapter.
What drives this convenience yield or, equivalently, the house-
holds’ willingness to give up some pecuniary return to hold the
convenience asset? In the next section, we consider a setting in which
this convenience yield arises because the households derive utility
directly from holding the Treasury debt. While this modeling device
is still too stylized to shed light on the underlying mechanism, it pro-
vides a useful starting point and illustrates some relevant properties
of the convenience yield. We discuss more elaborate microfounda-
tions proposed in the literature in Section 4.E.
Before we proceed to set up the model, we note that the conve-
nience yield is conceptually distinct from a risk premium, because

Electronic copy available at: https://ssrn.com/abstract=4668578


90 jiang – lecture notes on international finance

a risk premium does not generate a wedge in the Euler equation.


Consider, for example, a risky return r̃t+1 that does not carry a con-
venience yield. While this asset can earn an expected return that
reflects its cyclical properties, it must satisfy the standard Euler equa-
tion,

1 = Et [exp(mt+1 + r̃t+1 )] .

All models we considered in Chapter 3 fall into this category, al-


though they differ in the specification of the SDF mt+1 that deter-
mines the risk premium.
In contrast, a special asset that carries a convenience yield gen-
erates a wedge in the Euler equation as in Eq. (4.1). In our analysis
below, we will show that the convenience yield and the currency
risk premium will drive variations in the exchange rate level and the
currency expected return through different channels.

4.A An Illustrative Model

We assume that home bond generates non-pecuniary utilities when it


is held by either home or foreign households. As a concrete example,
we think about the home country as the U.S., the home risk-free
bond as the U.S. Treasury, and the home currency as the dollar in this
model.
Specifically, recall that b H,t is the market value of the bond and
rt is the risk-free rate. The home investors’ utility is derived over
consumption and the market value of home bond holdings:

" #
E0 ∑ δt (u(ct ) + v(bH,t ; θt )) ,
t =0

where θt captures a time-varying demand shifter for U.S. bonds.


We assume that the utility is increasing in the consumption and the
holding in the U.S. bonds, i.e. u′ (ct ) > 0 and v′ (b H,t ; θt ) > 0. In
this way, the U.S. risk-free bond carries a convenience yield, which
captures its non-pecuniary benefits to U.S. and foreign investors. We
assume that the marginal utility for holding U.S. bonds is decreasing
in quantity, i.e., v′′ (b) < 0, so that the convenience yield is decreasing
in the quantity held.

4.A.1 Issuer of the Safe Asset


We assume that households face the same endowment processes and
the same financial markets as in Section 1.A. The only difference is
the presence of a government in each country, which plays the role of
the safe asset issuer. Specifically, we assume that the home country’s

Electronic copy available at: https://ssrn.com/abstract=4668578


convenience yields 91

government borrows a one-period risk-free bond with par value b̄t


in period t, and pays it back in period t + 1. The government pays
the proceeds from the bond issuance as a transfer to the domestic
households in period t, and raises tax from the domestic households
to pay off the debt in period t + 1. The household budget in period t
is

pt yt + b H,t−1 exp(rt−1 ) + bF,t−1 exp(rt∗−1 − et ) + b̄t = ct + b H,t + bF,t exp(−et ) + b̄t−1 exp(rt−1 ),

where b̄t on the left-hand side is the proceeds from the home gov-
ernment’s debt issuance in period t, which is transferred to the home
households, and b̄t−1 exp(rt−1 ) on the right-hand side is the home
government’s debt repayment for its borrowing in period t − 1, which
is financed by a tax on the home households.
In this way, we set up a simple structure that generates a positive
supply of risk-free debt that home and foreign households can hold.
This setting implicitly assumes the government is a natural manufac-
turer of risk-free debt, a theme that we will return to in this chapter
and in Chapters 7 and 8. In our setting, consider any bond utility
function v that satisfies the Inada condition, such that the marginal
utility of bond holding approaches infinity when the bond quantity
approaches zero, e.g., v(b) = log(b). Then, we can show that home
and foreign households have to hold positive quantities of home risk-
free bonds in equilibrium. In this case, the home government is the
only provider of safe assets.

4.A.2 Intertemporal Solution


The Lagrangian is

"
E0 ∑ δt (u(ct ) + v(bH,t ; θt ))
t =1

#
+ ∑ ζ t ( pt yt + b̄t + bH,t−1 exp(rt−1 ) + bF,t−1 exp(rt∗−1 − et ) − ct − b̄t−1 exp(rt−1 ) − bH,t − bF,t exp(−et )) .
t =1

The first-order conditions w.r.t. ct , b H,t and bF,t are

δt u′ (ct ) − ζ t = 0,
Et δt v′ (b H,t ; θt ) − ζ t + ζ t+1 exp(rt ) = 0,
 

Et [−ζ t exp(−et ) + ζ t+1 exp(rt∗ − et+1 )] = 0.

The implied Euler equations for home households are


v′ (b H,t ; θt )
 ′ 
u ( c t +1 )
1 = Et δ ′ exp(rt ) + ,
u (ct ) u′ (ct )
 ′ 
u (c )
1 = Et δ ′ t+1 exp(−∆et+1 + rt∗ ) .
u (ct )

Electronic copy available at: https://ssrn.com/abstract=4668578


92 jiang – lecture notes on international finance

If we define the home households’ convenience yield as

def v′ (b H,t ; θt )
exp(−λt ) = 1 − , (4.2)
u′ (ct )

which is determined by the ratio between the marginal utility of


bond holding and the marginal utility of consumption. This equilib-
rium convenience yield λt is driven by both demand shocks captured
by θt and supply shocks which affect the equilibrium bond holdings
b H,t .
Then, we can express the Euler equations as

exp(−λt ) = Et [exp(mt+1 + rt )] ,
1 = Et [exp(mt+1 − ∆et+1 + rt∗ )] .

In doing so, we offer a more structural interpretation for the conve-


nience yields that we defined in Eq. (4.1).
Similarly, we modify the foreign utility as


" #
E0 ∑ δt (u∗ (ct ) + v(b∗H,t ; θt∗ )) ,
t =0

where the market value b∗H,t of foreign households’ bond holding is


also in the home (real) currency units. We define the foreign house-
holds’ convenience yield as

v′ (b∗H,t ; θt∗ ) exp(−et )


exp(−λ∗t ) = 1 − . (4.3)
u′ (c∗t )

Then, the foreign households’ Euler equations are

exp(−λ∗t ) = Et exp(m∗t+1 + rt + ∆et+1 ) ,


 

1 = Et exp(m∗t+1 + rt∗ ) .
 

4.A.3 Market Clearing


In goods market, the endowment is equal to the sum of consumption:

yt = c H,t + c∗H,t ,
y∗t = c F,t + c∗F,t .

In bonds market, the home and foreign bonds are in positive sup-
ply that is provided by their governments:

bt = b H,t + b∗H,t ,
bt∗ = bF,t + b∗F,t .

Electronic copy available at: https://ssrn.com/abstract=4668578


convenience yields 93

4.A.4 Macro Synthesis


The exogenous variables now include the endowments and the safe-
asset demand shifters:

(yt , y∗t , b̄t , b̄t∗ , θt , θt∗ )∞


t =0

There are 13 endogenous variables in each period t:

(c H,t , c F,t , b H,t , bF,t , pt , c∗H,t , c∗F,t , b∗H,t , b∗F,t , p∗t , rt , rt∗ , et )∞
t =0 .

u′ (c ) u′ (c∗t+1 )
Let exp(mt+1 ) = δ u′ (tc+)1 and exp(m∗t+1 ) = δ u′ (c∗t )
denote the
t
home and foreign SDFs. Let

v′ (b H,t ; θt )
exp(−λt ) = 1 − ,
u′ (ct )
v′ (b∗H,t ; θt∗ ) exp(−et )
exp(−λ∗t ) = 1 − ,
u′ (c∗t )

denote the convenience yields that home and foreign households


impute to the home bond.
The model implies the following 14 equations in each period, one
of which is redundant since the market clearing adds up to the sum
of households’ budget constraints. For the Home country in period t,

pt yt + b̄t + b H,t−1 exp(rt ) + bF,t−1 exp(rt∗−1 − et ) = (c H,t )α (c F,t )1−α + b̄t−1 exp(rt−1 ) + b H,t + bF,t exp(−et ),
(c H,t )α (c F,t )1−α = pt c H,t + p∗t c F,t exp(−et ),
yt = c H,t + c∗H,t ,
b̄t = b H,t + b∗H,t ,
exp(−λt ) = Et [exp(mt+1 ) exp(rt )] ,
1 = Et [exp(mt+1 ) exp(−∆et+1 + rt∗ )] ,

For the Foreign country in period t,

p∗t y∗t + b̄t∗ + exp(rt−1 + et )b∗H,t−1 + b∗F,t−1 exp(rt∗−1 ) = (c∗F,t )α (c∗H,t )1−α + b̄t∗−1 exp(rt∗−1 ) + exp(et )b∗H,t + b∗F,t ,
(c∗F,t )α (c∗H,t )1−α = pt c∗H,t exp(et ) + p∗t c∗F,t ,
y∗t = c F,t + c∗F,t ,
b̄t∗ = bF,t + b∗F,t ,
1 = Et exp(m∗t+1 ) exp(rt∗ ) ,
 

exp(−λ∗t ) = Et exp(m∗t+1 ) exp(∆et+1 + rt ) .


 

The within-period problem is the same as in the benchmark


model. The prices and exchange rates can be pinned down by:

pt α c F,t 1 − α c F,t
= = .
p∗t exp(−et ) 1 − α c H,t α c∗H,t

Electronic copy available at: https://ssrn.com/abstract=4668578


94 jiang – lecture notes on international finance

Compared to the macro synthesis in the baseline model in Section


1.A, the main difference is that the convenience yields are introduced
in the home and foreign households’ Euler equations for the home
bond. We also make an auxiliary assumption that the bonds are in
positive net supply, so that the utilities that the home and foreign
households derive from holding the home bond are well-behaved.

4.B Exchange Rate Accounting

Next, we repeat the exchange rate accounting exercise in this model


with bond convenience yields. The derivation is taken from Jiang,
Krishnamurthy, and Lustig [2021a], Jiang, Krishnamurthy, Lustig, and
Sun [2021b]. The model with bond convenience yields shows that
the four Euler equations for bond holdings now contain additional
convenience yield terms:

exp(−λt ) = Et [exp(mt+1 + rt )] , (4.4)


1= Et [exp(mt+1 − ∆et+1 + rt∗ )] , (4.5)
Et exp(m∗t+1 + rt∗ ) ,
 
1= (4.6)
exp(−λ∗t ) Et exp(m∗t+1 + ∆et+1 + rt ) .
 
= (4.7)

We can also express these Euler equations (4.4)–(4.7) as

1 = Et [exp(mt+1 + (rt + λt ))] ,


1 = Et [exp(mt+1 − ∆et+1 + rt∗ )] ,
1 = Et exp(m∗t+1 + rt∗ ) ,
 

1 = Et exp(m∗t+1 + ∆et+1 + (rt + λ∗t )) ,


 

which look like the original Euler equations (1.6)–(1.9) that we de-
rived in Section 1.B for risk-free bonds without convenience yields.
However, a key difference makes these two cases not observation-
ally equivalent: the home and foreign investors may have different
convenience yields λt and λ∗t . In this case, the home investors effec-
tively perceive a home risk-free rate without convenience yield equal
to ρt = rt + λt , whereas the foreign investors effectively perceive
a different home risk-free rate without convenience yield equal to
ρ̃t = rt + λ∗t . Therefore, we cannot map the setting with bond conve-
nience yields back to the baseline model without bond convenience
yields by modifying the bond yields.

4.B.1 Equilibrium Forces


Let us first work through a thought experiment that helps eluci-
date the restrictions these Euler equations impose on equilibrium
exchange rates.

Electronic copy available at: https://ssrn.com/abstract=4668578


convenience yields 95

Suppose that at time t, there is an exogenous increase in λ∗t , i.e.,


the foreign households’ convenience yield on the home safe assets.
For the sake of this argument, we will assume the home and foreign
SDFs and the home households’ convenience yield remain unaf-
fected. Then, this increase in foreign households’ convenience yield
sets off the following chain of events.
First, consider the home households’ Euler equation for holding
domestic bonds, Eq. (4.4), reproduced below,

exp(−λt ) = Et [exp(mt+1 + rt )] .

Since the home households’ SDF and convenience yield are assumed
to be unaffected, this Euler equation implies that the dollar risk-free
rate rt does not change.
Second, from the foreign households’ Euler equation for holding
home bonds, Eq. (4.7), reproduced below,

exp(−λ∗t ) = Et exp(m∗t+1 + rt + ∆et+1 ) ,


 

an increase in their convenience yield λ∗t lowers their risk-neutral


expected return on holding home risk-free bonds. Since the dollar
risk-free rate does not change, the exchange rate has to adjust to
equilibrate this Euler equation. In particular, the dollar needs to
appreciate today and create an expected depreciation to generate the
lower expected return.
Lastly, if we examine the home households’ Euler equation for
holding foreign bonds, Eq. (4.5), reproduced below,

1 = Et [exp(mt+1 + rt∗ − ∆et+1 )] ,

we learn that the dollar exchange rate movement also raises the ex-
pected return on purchasing foreign currency bonds from the home
perspective. Since the home households do not derive a convenience
yield on foreign bonds that can adjust, all adjustment must happen in
the dollar’s currency risk premium. In our equilibrium, this happens
via endogenous changes in the cyclicality and volatility of the dollar.
Thus, these four Euler equations require endogenous responses in
both first moments (i.e., exchange rate level and expected return) as
well as second moments (i.e., currency cyclicality and volatility) in
response to the shock to the convenience yield. As such, although the
convenience yield and the risk premium are conceptually different,
they could be correlated in practice.

4.B.2 Accounting for the Currency Expected Return


For the discussion in this section, we assume the random variables
are jointly normally distributed. First, we can derive the home inter-

Electronic copy available at: https://ssrn.com/abstract=4668578


96 jiang – lecture notes on international finance

est rate expression:


1
rt = −Et [mt+1 ] − vart (mt+1 ) − λt ,
2
which now contains the convenience yield term λt . When the home
households’ convenience yield is higher, the home bonds’ interest
rate is lower.
Next, the foreign households’ Euler equations can be expressed as
1
0 = Et [m∗t+1 ] + vart (m∗t+1 ) + rt∗ ,
2
1 1
−λ∗t = Et [m∗t+1 ] + vart (m∗t+1 ) + Et [∆et+1 ] + vart (∆et+1 ) + covt (m∗t+1 , ∆et+1 ) + rt .
2 2
Recall that the expected log excess return of the home currency
against the foreign currency is defined as

rxt+1 = ∆et+1 + rt − rt∗ .

Then, the Euler equations imply the following result:

Proposition 4.1. The home currency’s expected log excess return is de-
termined jointly by the home currency’s risk premium and the home bond’s
convenience yield:

Et [rxt+1 ] = rpt − λ∗t , (4.8)

def
where rpt = −covt (m∗t+1 , ∆et+1 ) − 21 vart (∆et+1 ) is the currency risk
premium.

If we compare this result to the benchmark case in Proposition


1.10, we note that the currency expected excess return now depends
on both the currency risk premium and the convenience yield. These
two components of the currency expected excess return are con-
ceptually different: the risk premium is the compensation for the
currency’s risk exposures, whereas the convenience yield reflects the
non-pecuniary benefits that investors derive from holding the bonds
issued in this currency. Of course, the currency risk premium and the
convenience yield could be correlated, as recessions tend to increase
risk premia as well as demand for safe assets. That said, they are
two conceptually distinct channels that affect the currency expected
return and, as we will see next, the exchange rate level.
Similarly, there is an expression from the home households’ per-
spective:
1
Et [−rxt+1 ] = −covt (mt+1 , −∆et+1 ) − vart (∆et+1 ) + λt .
2
Combined together, these expressions imply that the home and for-
eign investors need to agree on the combination of the equilibrium

Electronic copy available at: https://ssrn.com/abstract=4668578


convenience yields 97

currency risk premium and the convenience yield from their different
perspectives:
1 1
−covt (m∗t+1 , ∆et+1 ) − vart (∆et+1 ) − λ∗t = covt (mt+1 , −∆et+1 ) + vart (∆et+1 ) − λt ,
2 2
(4.9)

which is a direct extension of Eq. (1.12) which does not consider


convenience yields.

4.B.3 Accounting for the Exchange Rate Level


Moreover, assuming the real exchange rate is stationary, then, we
obtain a more general decomposition of the exchange rate level that
extends Proposition 1.2:

Proposition 4.2. The exchange rate level is equal to the sum of expected
future interest rate differentials, the sum of expected future convenience
yields, the sum of expected future currency risk premia, and the long-run
exchange rate level:
∞ ∞ ∞
et = ∑ Et [rt+ j − rt∗+ j ] + ∑ Et [λ∗t+ j ] − ∑ Et [rpt+ j ] + ē.
j =0 j =0 j =0

In other words, if today’s home currency is stronger than its long-


run mean, it is either because the home currency is offering a higher
interest rate than the foreign currency, or it has a higher convenience
yield, or because it has a lower risk premium (i.e., a lower discount
rate).
Similar to the exchange rate accounting in Section 1.B, we can also
use this formula to decompose the exchange rate innovation:
∞ ∞ ∞
(Et − Et−1 )[et ] = ∑ (Et − Et−1 )[rt+ j − rt∗+ j ] + ∑ (Et − Et−1 )[λ∗t+ j ] − (Et − Et−1 ) ∑ Et [rpt+ j ],
j =0 j =0 j =0

which likewise contains a convenience yield component.


Moreover, consider the risk-free rates ρt and ρ∗t without conve-
nience yields, which satisfy the standard Euler equations:

1 = Et [exp(mt+1 + ρt )] ,
1 = Et exp(m∗t+1 + ρ∗t ) .
 

We can relate these two risk-free rates to the bond yields rt and rt∗ by

ρt = rt + λt ,
ρ∗t = rt∗ .

As the home bond carries a convenience yield, the home households’


Euler equation implies that it is lower than the home risk-free rate

Electronic copy available at: https://ssrn.com/abstract=4668578


98 jiang – lecture notes on international finance

ρt by exactly λt . In comparison, the foreign bond does not carry a


convenience yield, so the foreign bond yield rt∗ is equal to the foreign
risk-free rate ρ∗t .
Then, we can reorganize the terms in Proposition 4.2 and show
∞ ∞ ∞
et = ∑ Et [ρt+ j − ρ∗t+ j ] + ∑ Et [λ∗t+ j − λt+ j ] − ∑ Et [rpt+ j ] + ē,
j =0 j =0 j =0

which implies that the variations in the exchange rate level are driven
by the variations in the risk-free rate differentials (which carry no
convenience yields), in the currency risk premia, and in the difference
between the convenience yields that home and foreign households
derive from holding the home bond.
This expression shows why it is natural to focus on the case in
which home and foreign households derive different convenience
yields, i.e., λt ̸= λ∗t . If the home and foreign households have the
identical convenience yield, i.e., λt = λ∗t , then, the exchange rate level
is determined only by the risk-free rate differentials and the currency
risk premia, both of which are driven by the SDFs via

ρt − ρ∗t = − log Et [exp(mt+1 )] + log Et exp(m∗t+1 ) ,


 

1
rpt = −covt (m∗t+1 , ∆et+1 ) − vart (∆et+1 ).
2
In other words, only the home and foreign households’ convenience
yield differential, rather than their convenience yield levels, matters
for the exchange rate level.

4.B.4 Comparison with the Complete-Market Case


Building on the discussion of the exchange rate determination, let us
consider complete markets as a special case. In the model we spec-
ified in Section 4.A, we additionally assume that the home and for-
eign households can trade the complete set of contingent claims. In
the meanwhile, these households still derive non-pecuniary benefits
from holding the dollar risk-free bond, so the model is still different
from the baseline model in Section 1.C.
In this case, we can solve the model by assuming a social planner
who maximizes the sum of home and foreign utilities with weights
π and 1 − π, subject to the market clearing conditions for goods
and bonds. The equilibrium convenience yields satisfy the following
relationships [Jiang, 2023a].

Proposition 4.3. When the markets are complete, the home and foreign
households derive the same convenience yields from holding the dollar bond:

λt = λ∗t ,

Electronic copy available at: https://ssrn.com/abstract=4668578


convenience yields 99

and the exchange rate is determined only by the marginal utility differential:

∆etcm = mt − m∗t .

The proof is in Appendix A.13. This proposition shows that, while


the convenience yields can be nonzero in complete markets, they do
not directly affect the equilibrium exchange rate.1 More precisely, the 1
In the complete-market world, conve-
nience yields may still affect exchange
convenience yields λt and λ∗t are equalized to generate zero differ-
rates indirectly via their effect on rel-
ential. Given the definitions of these convenience yields in Eq. (4.2) ative marginal utilities, for example,
and (4.3), this means that the home and foreign households’ marginal by affecting the natural rate in the
home country [Caballero, Farhi, and
utilities derived from holding the dollar bond are equalized (after Gourinchas, 2021, Kekre and Lenel,
multiplication with the Pareto weights π and 1 − π): 2021].

πv′ (b H,t ; θt ) = (1 − π )v′ (b∗H,t ; θt∗ ),

which is similar to the equilibrium condition under complete markets


that the home and foreign households’ marginal utilities over con-
sumption are also equalized (after unit conversion by the exchange
rate exp(et ) and multiplication with the Pareto weights π and 1 − π):

πu′ (ct ) = (1 − π )u′ (c∗t ) exp(et ).

As such, incomplete markets are an essential ingredient to make


the convenience yields matter for exchange rates. We will further
explore the interaction between incomplete markets and convenience
yields in Section 5.D.

4.B.5 The Yield-Exchange Rate Connection


The introduction of the convenience yield also broadens how we
think about the relationship between the interest rate in a country
and its currency strength. Under the uncovered interest rate parity
(UIP) condition, a higher interest rate is associated with a stronger
currency. We can see this from Proposition 4.2, which predicts a pos-
itive relationship between the currency value et of the home country
and its interest rate rt , holding other things constant:
∞ ∞ ∞
et = ∑ Et [rt+ j − rt∗+ j ] + ∑ Et [λ∗t+ j ] − ∑ Et [rpt+ j ] + ē.
j =0 j =0 j =0

However, when we introduce the convenience yield, the relation-


ship between the interest rate and the currency value becomes more
complicated. By the home households’ Euler equation, the home
interest rate level is determined by the home households’ SDF and
convenience yield:

1
rt = −Et [mt+1 ] − vart (mt+1 ) − λt .
2

Electronic copy available at: https://ssrn.com/abstract=4668578


100 jiang – lecture notes on international finance

Holding other things constant, a higher convenience yield λt low-


ers the home interest rate rt . During global recessions, it is plausible
that the flight to safety raises the convenience yields that both home
and foreign households impute to the U.S. bond, i.e., λt and λ∗t .
Then, the home interest rate rt decreases, while the convenience yield
component strengthens the home currency, leading to a negative re-
lationship between the home interest rate and the home currency
value.
This discussion shows that the yield-exchange rate relationship
depends on the types of shocks we are considering [Jiang, Krish-
namurthy, Lustig, and Sun, 2021b]. When we consider a monetary
shock εmt whose primary effect is to raise the home interest rate, we
expect a positive yield-exchange rate relationship:

def
covt (et , rt |εm m m
t ) = covt ( proj [ et | ε t ], proj [rt | ε t ]) > 0.

When we consider a demand shock εdt whose primary effect is to


raise the convenience yield, which lowers the bond yield while
strengthening the currency, we expect a negative yield-exchange
rate relationship:

def
covt (et , rt |εdt ) = covt ( proj[et |εdt ], proj[rt |εdt ]) < 0.

The overall relationship between the interest rate and the exchange
rate depends on the relative strength of these two effects.
This intuition also carries over to the case of long-term debt,
whose yield also depends on the expected interest rates and the
convenience yields. In particular, an increase in the expected path of
interest rates raises the long-term bond yield and appreciates the cur-
rency, whereas a flight-to-safety shock can lower the long-term bond
yield if it also carries some convenience benefits, while appreciating
the currency. However, as we see in Section 3.D, when markets are
complete and exchange rates are stationary, the presence of currency
risk premium does not break the long-run UIP condition (3.21). As a
result, like interest rate shocks, risk premium shocks affect the long-
run bond yield and the exchange rate in the same direction, which
also shows a stark difference between currency risk premium and
convenience yield.

4.C Measuring the Convenience Yields

Having described the role convenience yields play in exchange rate


dynamics, we face the obvious question of how they are measured
empirically. As indicated in Section 2.A.6, the convenience yields
manifest themselves in many interest rate spreads. In this section, we

Electronic copy available at: https://ssrn.com/abstract=4668578


convenience yields 101

discuss this issue in the context of the Treasury basis, which follows
Jiang, Krishnamurthy, and Lustig [2021a].
Recall that a dollar forward allows investors to lock in the dollar
exchange rate at a fixed rate f t in the next period. Now, consider the
following trading strategy. In period t, an investor converts 1 unit of
dollar to exp(et ) units of the foreign currency, and purchase foreign
risk-free bonds. At the same time, the investor enters a currency
forward contract to lock in the dollar exchange rate at f t in period
t + 1 for exp(et + rt∗ ) units of foreign currency.
Then, in period t + 1, the investor has exp(et + rt∗ ) units of foreign
currency, which are then converted back to exp(− f t + et + rt∗ ) units
of dollar using the forward contract. This strategy generates a return
of exp(− f t + et + rt∗ ) per dollar invested. Since the investor does not
hold the actual dollar risk-free bond but still earns a risk-free return
in dollar terms, we regard this combination of foreign risk-free bond
and currency forward contract as a synthetic dollar bond.
If this synthetic dollar bond does not carry a convenience yield,
then, its Euler equation can be expressed as

1 = Et [exp(mt+1 − f t + et + rt∗ )]

from the home households’ perspective, and

1 = Et exp(m∗t+1 + ∆et+1 − f t + et + rt∗ )


 

from the foreign households’ perspective.


Combine with the derivation in Section 4.B. We obtain

−λ∗t = rt − (rt∗ − f t + et ),
−λt = rt − (rt∗ − f t + et ),

which implies that the convenience yields that foreign and home
households impute to the dollar risk-free bonds are equal to each
other, and equal to the negative of the dollar Treasury basis, defined
as
def
xtTreas = rt − (rt∗ − f t + et ).

When the dollar risk-free bond has a convenience yield, its risk-free
rate rt is lower than the synthetic dollar yield, i.e., (rt∗ − f t + et ),
leading to a negative dollar Treasury basis:

def
xtTreas = rt − (rt∗ − f t + et ) = −λ∗t < 0.

We obtain a more realistic case by assuming that the synthetic dol-


lar bond also carries a convenience yield, albeit smaller in magnitude
than the actual dollar risk-free bond. More precisely, we assume that

Electronic copy available at: https://ssrn.com/abstract=4668578


102 jiang – lecture notes on international finance

the convenience yield derived from the synthetic dollar bond is a


fraction β∗ of the convenience yield derived from the actual dollar
bond from foreign households’ perspective:

exp(− β∗ λ∗t ) = Et exp(m∗t+1 + ∆et+1 − f t + et + rt∗ ) .


 

If this synthetic dollar bond is representative of other non-Treasury


dollar safe assets, a point we will next discuss in 4.C.1, then, concep-
tually we can say that for the convenience yield on the U.S. Treasury,
β∗ of it is about the dollar and 1 − β∗ of it is specifically about the
U.S. Treasury. Indeed, combining with the derivation in Section 4.B,
we obtain
def
xtTreas = rt − (rt∗ − f t + et ) = −(1 − β∗ )λ∗t < 0,

which states that the Treasury basis captures the fraction of the con-
venience yield that the U.S. Treasury earns on top of the synthetic
dollar bond.
Similarly, from the home households’ perspective, we assume that
the convenience yield derived from the synthetic dollar bond is a
fraction β of the convenience yield derived from the actual dollar
bond:

exp(− βλt ) = Et [exp(mt+1 − f t + et + rt∗ )] ,

which implies

−(1 − β)λt = rt − (rt∗ − f t + et ) = xtTreas .

As such, we can express the convenience yield levels λ∗t and λt as


functions of the Treasury basis xtTreas :

1
λ∗t = − x Treas ,
1 − β∗ t
1
λt = − x Treas .
1−β t

If β ̸= β∗ , the foreign and home households’ convenience yields on


the dollar risk-free bond would be different.
To restore the actual convenience yield level, we need to estimate
the scaling coefficients β and β∗ . Jiang, Krishnamurthy, and Lustig
[2021a] provide an estimate based on the implied exchange rate dy-
namics derived in Section 4.B. Assuming that currency risk premium
does not respond to a convenience yield shock, they find that the
β∗ coefficient has to be around 0.9 to match the variations in the
Treasury basis with those in the observed exchange rate movement.
This parameter estimate means that 90% of the convenience yield is
about the dollar, and 10% is about the U.S. Treasury. In other words,

Electronic copy available at: https://ssrn.com/abstract=4668578


convenience yields 103

the synthetic dollar position contains 90% of the convenience yield


on the dollar risk-free bond, and the Treasury basis only measures
1 − 0.9 = 10% of the convenience yield. To recover the convenience
yield level, we need to lever up the observed Treasury basis by 10
times. As the Treasury basis is on average −20 basis points per an-
num in the sample, the implied convenience yield has to be 2% per
annum on average, which is much greater than any near-arbitrage
interest rate spreads.

4.C.1 Treasury CIP vs. Libor CIP


This argument also helps us understand the Libor basis, which is the
spread between the dollar Libor rate and the synthetic dollar rate
based on the foreign Libor rate and the currency forward:

def
xtLibor = libort − (libort∗ − f t + et ).

Unlike the Treasury basis, the Libor basis was very close to zero
before the Global Financial Crisis, and widened afterwards [Du,
Tepper, and Verdelhan, 2018b]. This pattern suggests that the dollar
Libor and the synthetic position based on the foreign Libor and the
currency forward have a similar level of convenience yields before the
crisis, i.e.,

exp(− β∗ λ∗t ) = Et exp(m∗t+1 + ∆et+1 − libort ) ,


 

= Et exp(m∗t+1 + ∆et+1 − f t + et + libort∗ ) .


 

This is not a coincidence if there are investment banks who are indif-
ferent between the two positions and actively arbitrage any spread
between the Libor market and the currency forward market. As such,
both the dollar Libor and the synthetic dollar bond earn the same,
non-zero convenience yield β∗ λ∗t , so that their spread, i.e., the Libor
basis xtLibor , is zero.
After the crisis, likely due to regulatory constraints, the dollar
Libor now contains a higher level of convenience yield, and banks
face higher costs to arbitrage this spread.2 As a result, the dollar 2
There is a debate about whether
the post-crisis regulations distort the
Libor has a higher level of convenience yield than the synthetic dollar
interest rate market or the currency
bond, leading to a negative Libor basis: forward market [Augustin, Chernov,
Schmid, and Song, 2020].
exp(−γ∗ λ∗t ) = Et exp(m∗t+1 + ∆et+1 − libort ) ,
 

exp(− β∗ λ∗t ) = Et exp(m∗t+1 + ∆et+1 − f t + et + libort∗ ) ;


 

with γ∗ > β∗ ,

def
xtLibor = libort − (libort∗ − f t + et ) = −(γ∗ − β∗ )λ∗t < 0.

Electronic copy available at: https://ssrn.com/abstract=4668578


104 jiang – lecture notes on international finance

Finally, if the same marginal investors trade in the Libor market


and the Treasury market, then, we should expect that the Treasury
basis and the Libor basis are correlated with each other, with

xtTreas = −(1 − β∗ )λ∗t ,


xtLibor = −(γ∗ − β∗ )λ∗t .

The same argument can be applied to other types of interest rate


spreads. The key point is that, when both interest rates contain cer-
tain amount of convenience yields, the spread only measures a frac-
tion of the convenience yields. It is useful to think about these bonds
as containing different shades of dollarness.

4.D Connecting the Short Term with the Long Term

The convenience yields accrue to bonds that are safe and liquid,
which is why we begin our analysis with one-period Treasury bill.
In this section, we broaden our perspective by considering how the
convenience yield on the short-term bond also affects the pricing of
the long-term bond. To do so, we first focus on the domestic house-
holds’ perspective and understand how the convenience yields affect
the long-term bond prices. Then, we study how the home and for-
eign households trade the long-term bonds with each other and
understand how the convenience yields and bond risk premia jointly
determine the exchange rate level.

4.D.1 The Term Structure of Bond Yields with Convenience Yields

We start with the model developed in Section 4.A, and additionally


allow the households to trade long-term bonds. Specifically, let qt (h)
denote the book value of the home households’ holding of the bond
with maturity h and let pt (h) denote the price of the bond with ma-
turity h. All bonds are risk-free. We still assume that the investors
derive utility only from holding the 1-period bond:


" #
E0 ∑ δ (u(ct ) + v(qt (1) pt (1); θt ))
t
.
t =0

However, because the long-term Treasury bonds will eventually be-


come short-term bonds as they mature, if the investors do not differ-
entiate between on-the-run and off-the-run bonds, then, the conve-
nience yield on the short-term bonds will also affect the pricing of the
long-term bonds.
Similar to the model in Section 4.A, we can derive the households’

Electronic copy available at: https://ssrn.com/abstract=4668578


convenience yields 105

Euler equation for holding the 1-period bond as

v ′ ( q t (1); θ t )
 
def 1
exp(−λt ) = 1 − = E t exp ( m t,t+1 ) ,
u′ (ct ) p t (1)

where the convenience yield λt reflects the marginal utility derived


from the convenience benefits of the 1-period bond. Rearranging
terms, we obtain

pt (1) = Et [exp(mt,t+1 )] exp(λt ).

Now, consider the 2-period bond. We can write the households’


Euler equation for this bond as

pt (2) = Et [exp(mt,t+1 ) pt+1 (1)] ,

which does not contain a convenience yield term since the 2-period
bond does not offer any immediate convenience benefits. However, if
we substitute the 1-period bond price into the equation, we obtain

pt (2) = Et [exp(mt,t+1 + λt+1 )Et+1 [exp(mt+1,t+2 )]] ,

which implies that the bond price today reflects the expectation of the
bond’s convenience yield when it becomes 1-period bond in the next
period. In this way, the convenience yield that only accrues to the
short-term bond can affect the entire term structure of bond prices.
Moreover, the next period’s convenience yield λt+1 is priced by
the SDF mt,t+1 in the sense that their covariance also matters. We can
rewrite the 2-period bond price as

pt (2) = Et [exp(mt,t+1 )Et+1 [exp(mt+1,t+2 )]] Et [exp(λt+1 )]


+ covt (exp(mt,t+1 )Et+1 [exp(mt+1,t+2 )] , exp(λt+1 )).

On the right-hand side, the first term shows that a higher expected
convenience yield increases the price of the 2-period bond today, and
the second term shows that a more counter-cyclical convenience yield
can additionally raise the bond price by lowering its risk premium.
We can iteratively derive the valuation of the longer-term bonds
from that of the nearer-term bonds. Jiang and Richmond [2022] pro-
vide a general affine expression for the bond price in this setting
under the additional assumption that the SDF volatility and the con-
venience yield jointly follow an affine process. As we expected, the
1-period bond’s convenience yield and its cyclical property affects not
only the long-term bonds’ expected service flows and prices, but also
their cyclical properties and hence risk premia.
More realistically, we can additionally assume that the investors
derive utility from holding long-term bonds as well. In this case, we

Electronic copy available at: https://ssrn.com/abstract=4668578


106 jiang – lecture notes on international finance

can generalize the function v as



" #
H
E0 ∑ δ (u(ct ) + v( ∑ wt (h)qt (h) pt (h); θt ))
t
,
t =0 h =1

where the weight wt (h) describes the relative amount of convenience


generated by the bond with maturity h. For example, if the non-
pecuniary utility is generated by the bonds’ plegeability in the repo
market, then, wt (h) describes the different hair cuts on the bonds
with different maturities.
In this more general setting, the long-term bonds price in both
their own convenience yields in the current period and the expected
convenience yields of the shorter-term bonds in the future. Interest-
ingly, after the 2008 Global Financial Crisis, the long-term Treasury
bonds are traded at a discount relative to other safe assets, reflecting
an “inconvenience yield” [Duffie, 2020, He, Nagel, and Song, 2022,
Du, Hébert, and Li, 2023]. This negative yield spread can be under-
stood as a negative marginal utility investors derive from holding
the long-term Treasury bonds, or, equivalently, a positive balance
sheet cost. Moreover, Joslin, Li, and Song [2021] measures the term
structure of the convenience yield using the spread between Treasury
bonds and Refcorp bonds.

4.D.2 Long-Run UIP Condition


Next, we consider both home and foreign households’ perspectives.
When there is a convenience yield on the one-period bond and when
markets are incomplete, the equilibrium exchange rate movement
deviates from complete-market formula we derived in Proposition
4.3, i.e.,

∆etcm
+1 = m t +1 − m t +1 .

As such, we can express the exchange rate movement as

∆et+1 = mt+1 − m∗t+1 + ηt+1 ,

where the new term ηt+1 introduces a wedge in the exchange rate
expression. We will discuss this term in greater detail in the next sec-
tion. For now, it is sufficient to note that this term could be correlated
with the convenience yield λ∗t .
We can express this equation iteratively from period 0:
t t
e t − e0 = ∑ (mk − m∗k ) + ∑ ηk .
k =1 k =1

Jiang, Krishnamurthy, Lustig, and Sun [2021b] show that, if the cu-
mulative wedge ∑tk=1 ηk is a random walk, the cumulative SDF dif-
ferential ∑tk=1 (mk − m∗k ) needs to have a permanent component that

Electronic copy available at: https://ssrn.com/abstract=4668578


convenience yields 107

exactly offsets the random walk component in the cumulative wedge


in order to guarantee exchange rate stationarity. This means that the
home and foreign SDFs will not share the same permanent compo-
nent. Instead, they have to diverge from each other for the exchange
rate to remain stationary.
As a result, the long-term bond yields, which are closely related
to the long-run properties of the home and foreign SDFs, no longer
converge between the two countries:

lim (rt (h) − rt∗ (h)) ̸= 0.


h→∞

Then, the long-run UIP condition (3.21) that we derived in Proposi-


tion 3.10 under the assumption of identical permanent SDF compo-
nents fails:

et − ē ̸= lim h(rt (h) − rt∗ (h)).


h→∞

In fact, the left-hand side is finite whereas the right-hand side can be
infinite in this case. The convenience yield also introduces a wedge
in this long-run UIP condition. A full characterization of this wedge
remains an open question.

4.E Discussions (TODO)

Electronic copy available at: https://ssrn.com/abstract=4668578


5
Incomplete Markets

Summary

• Incomplete markets also lead to a wedge between the exchange rate movement and the SDF
differential:

∆et+1 = mt+1 − m∗t+1 + ηt+1 .

• This wedge affects the exchange rate volatility, cyclicality, and currency risk premium.

• This wedge also gives rise to international spill-over across countries.

• Market incompleteness is a precondition for convenience yields to impact the exchange rate
dynamics, and convenience yields enrich the effects of market incompleteness on the exchange
rate dynamics.

In general, the absence of complete spanning in the asset market


means that the home and foreign households do not equate their
SDFs state by state. So the exchange rate expression (1.16) in the
complete-market case is violated:

∆et+1 ̸= mt+1 − m∗t+1 .

Backus, Foresi, and Telmer [2001] introduces a wedge η to describe


this violation:

∆et+1 = mt+1 − m∗t+1 + ηt+1 . (5.1)

In this section, we first develop a full general equilibrium model


with incomplete markets that gives rise to the above wedge ηt+1 . We
then take a different, no-arbitrage approach that focuses only on the
Euler equations. While incomplete spanning leads to deviations from
the complete-market characterization of the exchange rate dynamics
(1.16), it still respects the Euler equations for the assets that are trad-
able by both countries’ households. For example, if home and foreign

Electronic copy available at: https://ssrn.com/abstract=4668578


incomplete markets 109

risk-free bonds are tradable, the original Euler equation (1.6) holds,
i.e.,

1 = Et [exp(mt+1 + rt )] ,

which does not require us to introduce additional Euler equation


wedges as we did in our discussion of convenience yields in Chapter
4, i.e.,

exp(−λt ) = Et [exp(mt+1 + rt )] .

In this sense, the incomplete-market wedge η in Eq. (5.1) is con-


ceptually distinct from the Euler equation wedge λt . We will further
develop these model ingredients throughout this section.
Conceptually, there are two types of market incompleteness, one
at the individual level and one at the country level. Markets can be
incomplete at the individual level, which has been studied in a large
macro literature starting from Aiyagari-Bewley-Huggett. Markets
can also be incomplete between countries, leading to incomplete
risk-sharing at the country level. In this section, we will focus on the
latter form of market incompleteness. We allow the individual-level
risk to be fully shared within countries, and model the representa-
tive household in each country. At the same time, the representative
households in different countries cannot trade the full set of contin-
gent claims to share risks internationally, leading to richer exchange
rate and capital flow dynamics.

5.A An Illustrative Model

Again we begin with a fully specified model. This is particularly


useful in the context of incomplete markets, as the incomplete-market
wedge ηt+1 in Eq. (5.1) is particularly difficult to conceptualize in the
absence of such a model.
As the discussion of incomplete-market exchange rate dynamics
is necessarily abstract, we begin by presenting a concrete general
equilibrium model with incomplete markets. This model is a variant
of Pavlova and Rigobon [2012], with the key difference being that
Pavlova and Rigobon [2012] has one international bond whose pay-
off is a weighted average of home and foreign goods, whereas our
model has a home bond and a foreign bond that we considered in the
previous sections.

5.A.1 The Economic Setting


For tractability, we now use some continuous-time math. There is a
finite horizon T. The home and foreign endowment processes are

Electronic copy available at: https://ssrn.com/abstract=4668578


110 jiang – lecture notes on international finance

given by
y
dyt = µy yt dt + σy yt dZt ,
y∗
dy∗t = µy∗ y∗t dt + σy∗ y∗t dZt .

The home households’ utility is given by


Z T 
E exp(−δt)u(ct )dt ,
0

where u(ct ) = γt log(ct ) and the aggregate consumption is again


a Cobb-Douglas aggregation of the home and foreign goods: ct =
(c H,t )α (c F,t )(1−α) . Relative to the baseline model, here we additionally
have a stochastic term γt which captures demand shocks. We assume
that γt follows
γ
dγt = 0dt + γt ωdZt ,

which guarantees that γt is a strictly positive adapted stochastic


process and a martingale.
The foreign households’ utility is given by
Z T 
∗ ∗
E exp(−δt)u (ct )dt ,
0

where u∗ (c∗t ) = γ∗ log(c∗t ) and c∗t = (c∗F,t )α (c∗H,t )(1−α) . We assume that
there are no time-varying demand shocks for foreign household, i.e.
γ∗ is constant.
More formally, we fix a probability space (Ω, F , P) and a given
filtration F = {F0 : t ≥ 0} satisfying the usual conditions. We assume
that all stochastic processes are adapted to this filtration. Specifically,
uncertainty is represented by a standard three-dimensional Brownian
y y∗
motion Zt = [ Zt , Zt , Zt ]′ .
γ

The only tradable assets are the two countries’ risk-free bonds,
which are denominated in the local consumption bundles and have
interest rates rt and rt∗ , respectively. The real exchange rate exp(et )
is the conversion ratio between the home and foreign consumption
bundles. 1 unit of the home bundle is worth exp(et ) units of the
foreign bundle. Furthermore, assume that the log exchange rate
follows

det = κt dt + σt dZt ,

where σt = [σ1,t , σ2,t , σ3,t ] is a 1 by 3 vector. By Ito’s lemma, we have


 
1 ′
d exp(et ) = exp(et ) κt + σt σt dt + exp(et )σt dZt ,
2
 
1 ′
d exp(−et ) = exp(−et ) −κt + σt σt dt − exp(−et )σt dZt .
2

Electronic copy available at: https://ssrn.com/abstract=4668578


incomplete markets 111

From the perspective of the home household, the instantaneous re-


turn on foreign bonds is:
 
∗ 1 ′
rt − κt + σt σt dt − σt dZt .
2
In addition, from the viewpoint of the foreign household, the instan-
taneous return on home bonds is:
 
1
rt + κt + σt σt′ dt + σt dZt .
2
The derivation of the within-period solution is identical to Section
1.A.3 in the baseline model. Let pt denote the price of the home
goods in the numéraire of the home consumption bundle, and let p∗t
denote the price of the foreign goods in the numéraire of the foreign
consumption bundle. As a special case of the baseline model, the
home households’ within-period allocation satisfies
α (1 − α )
c H,t = ct , c F,t = ct ,
pt p∗t exp(−et )
and the terms of trade are simply the ratio of prices between home
and foreign goods, which is related to the real exchange rate by et =
(2α − 1)tott .
Let wt denote the wealth of the home households in the home
numéraire. The dynamic budget constraint of the home households is
  
∗ 1 ′
dwt = wt rt + xt rt − κt + σt σt − rt dt − wt xt σt dZt + pt yt dt − [ pt c H,t + p∗t exp(−et )c F,t ] dt,
2
where xt is the fraction of wealth that is invested in foreign bonds.
Similarly, let wt∗ denote the wealth of the foreign households in
the foreign numéraire. The dynamic budget constraint of the foreign
households is
  
1
dwt∗ = wt∗ rt∗ + xt∗ rt + κt + σt σt′ − rt∗ dt + wt∗ xt∗ σt dZt + p∗t y∗t dt − pt exp(et )c∗H,t + p∗t c∗F,t dt,
 
2
where xt∗ is the fraction of wealth that is invested in home bonds.
def
To simplify the derivation, let w̃t∗ = wt∗ exp(−et ) denote the
wealth of the foreign households in the home numéraire. Then, we
can express foreign household’s dynamic budget constraint in the
home numéraire as
  
1
dw̃t∗ = w̃t∗ rt + (1 − xt∗ ) rt∗ − κt + σt σt′ − rt dt − w̃t∗ (1 − xt∗ )σt dZt + p∗t exp(−et )y∗t dt − pt c∗H,t + p∗t exp(−et )c∗F,t dt.
 
2

5.A.2 Macro Synthesis


There are 3 exogenous processes

(yt , y∗t , γt )tt=0

Electronic copy available at: https://ssrn.com/abstract=4668578


112 jiang – lecture notes on international finance

and 15 endogenous processes:

(ct , c H,t , c F,t , wt , xt , pt , c∗t , c∗H,t , c∗F,t , wt∗ , xt∗ , p∗t , rt , rt∗ , et )tT=0 ,
The model implies the following 15 equations in each period t,
which are very similar to the equilibrium conditions in the baseline
model listed in Section 1.A.6. They include 2 consumption aggrega-
tion equations,

ct = (c H,t )α (c F,t )1−α ,


c∗t = (c∗F,t )α (c∗H,t )1−α ,

4 household budget constraints,


  
1
dwt = wt rt + xt rt∗ − κt + σt σt′ − rt dt − wt xt σt dZt + pt yt dt − [ pt c H,t + p∗t exp(−et )c F,t ] dt,
2
ct = pt c H,t + p∗t c F,t exp(−et ),
  
∗ ∗ ∗ ∗ 1 ′ ∗
dt + wt∗ xt∗ σt dZt + p∗t y∗t dt − pt exp(et )c∗H,t + p∗t c∗F,t dt,
 
dwt = wt rt + xt rt + κt + σt σt − rt
2
c∗t = pt c∗H,t exp(et ) + p∗t c∗F,t ,

2 equations that describe the households’ within-period consumption


choices,

pt α c F,t 1 − α c F,t
= = ,
p∗t exp(−et ) 1 − α c H,t α c∗H,t
2 goods market clearing conditions,

yt = c H,t + c∗H,t ,
y∗t = c F,t + c∗F,t ,

1 bond market clearing condition,

wt xt = wt∗ exp(−et ) xt∗ ,

and 4 Euler equations,


  Z t 
0 = Et d exp mt + ru ,
0
  Z t 
0 = Et d exp mt − et + ru∗ ,
0
  Z t 
0 = Et d exp m∗t + ru∗ ,
0
  Z t 
0 = Et d exp m∗t + et + ru ,
0

where the consumption-based SDFs are defined as


def γt c 0
exp(mt ) = exp(−δt) ,
γ0 ct
def c∗
exp(m∗t ) = exp(−δt) 0∗ .
ct

Electronic copy available at: https://ssrn.com/abstract=4668578


incomplete markets 113

5.A.3 Portfolio Choices


The home household’s portfolio choice problem is
Z T 
max E exp(−δt)γt log ct dt
ct ,xt 0
  
1
s.t. dwt = wt rt + xt rt∗ − κt + σt σt′ − rt dt + ( pt yt − ct )dt − wt xt σt dZt .
2

Similarly, noting that the log preference allows separation between


the consumption choice and the numéraire, i.e.,

log c∗t = log c̃∗t + et ,

where the exchange rate et is unaffected by the decision of the com-


petitive households, we can express foreign household’s portfolio
choice problem as
Z T 
∗ ∗
max
∗ ∗
E exp (− δt ) γ log c̃ t dt
c̃t ,xt 0
  
1
s.t. dw̃t∗ = w̃t∗ rt + (1 − xt∗ ) rt∗ − κt + σt σt′ − rt dt + ( p∗t exp(−et )y∗t − c̃∗t )dt − w̃t∗ (1 − xt∗ )σt dZt .
2

In this way, under the log preference, the optimization problems


are formally equivalent to a familiar single-good consumption-
investment problem [Pavlova and Rigobon, 2012].
Next, to deal with the market incompleteness, we follow He and
Pearson [1991] to convert the above home household’s dynamic port-
folio choice problem to a static variational problem:
Z T 
max E exp(−δt)γt log ct dt (5.2)
ct 0
Z T 
s.t. E exp(mν,t )(ct − pt yt )dt ≤ W0 ,
0

where exp(mν,t ) denotes an appropriate home SDF. In incomplete


markets, multiple SDFs are consistent with no arbitrage, giving rise
to an infinite number of the static budget constraint. To characterize
this family of SDFs, let λt denote the home country market price of
risk,

−σt′
 
∗ 1 ′
λt = r − κ t + σt σ − r t ,
||σt ||2 t 2 t

which is a 3 by 1 vector. Then, the set of home SDFs can be repre-


sented as

d exp(mν,t ) = −rt exp(mν,t )dt − (λt + νt )′ exp(mν,t )dZt ,


RT
where νt ∈ R3 satisfies σt νt = 0 for any t ∈ [0, T ] and 0 ||νt ||2 dt < ∞.

Electronic copy available at: https://ssrn.com/abstract=4668578


114 jiang – lecture notes on international finance

It is worth noting that this family of SDFs correctly price the home
and foreign bonds:
  Z t 
A exp mν,t + rt dt = 0,
0
  Z t 
A exp mν,t − et + rt∗ dt = 0,
0

where A[wt ] is the infinitesimal generator of the stochastic process


wt .
When we have three non-redundant risky assets, the markets are
complete, and the volatility matrix σt is a non-degenerate square
matrix. In this case, the equation σt νt = 0 only has one solution:
νt = 0. In contrast, when the markets are incomplete, the volatility
matrix σt has fewer rows than the number of uncertainty shocks, and
many νt can satisfy σt νt = 0.
Similarly, for foreign households, the static problem is
Z T 
∗ ∗
max

E exp (− δt ) γ log c̃ t dt (5.3)
c̃t 0
T
Z 
s.t. E exp(mν̃,t )(c̃∗t − p∗t exp(−et )y∗t )dt ≤ W̃0∗ ,
0

where

d exp(mν̃,t ) = −rt exp(mν̃,t )dt − (λt + ν̃t )′ exp(mν̃,t )dZt .

He and Pearson [1991] prove that there exists a unique individual-


specific νt , which we denote by νH,t for home households, that mini-
mizes their maximum expected utility in Eq. (5.2). This νH,t is the one
that makes the constraint bind. Pavlova and Rigobon [2012] refer to
the home country SDF related to this νH,t as the personalized home
country SDF of the home households. Similarly, let the home country
SDF related to ν̃F,t denote the personalized home country SDF of the
foreign households, which is also in the unit of the home numéraire.
It is no coincidence that the home and foreign countries’ personal-
ized SDFs are exactly their households’ consumption-based SDFs, as
we summarize in the following lemma.
Lemma 5.1. (1) The consumption-based SDF of the home households is
exp(−δt) γγ0t cc0t , and it is equal to the personalized home country SDF of the
home households, exp(mνH ,t ).
c∗
(2) The consumption-based SDF of the foreign households is exp(−δt) c0∗ ,
t
and it is equal to the personalized foreign country SDF of the foreign house-
holds, exp(m∗ν∗ ,t ).
F

The proof is presented in Appendix A.14.


We report the portfolios of both home and foreign consumers in
the following proposition.

Electronic copy available at: https://ssrn.com/abstract=4668578


incomplete markets 115

Proposition 5.1. (1) The fraction of wealth xt invested in the foreign bonds
by the home households is given by
!
wt + st ωσ3,t (rt∗ − κt + 12 σt σt′ − rt ) st σts σt′
xt = − 2
− 2
+ , (5.4)
wt ∥σt ∥ ∥σt ∥ wt ∥σt ∥2

and the fraction of wealth xt∗ invested in the home bonds by the foreign
households is given by
 
r ∗ − κ + 1 σ σ′ − r ∗

w̃ + s̃ ∗ t t 2 t t t s̃∗t σts̃ σt′
xt∗ = 1 − t ∗ t − . (5.5)
w̃t ∥σt ∥2 w̃t∗ ∥σt ∥2

(2) The processes νH and ν̃F , which enter the specification of the home
state price densities personalized for the home and foreign households respec-
tively, are given by:

σt′ σt
  
′ st s ′
νH,t = I3 − −ωi3 + (σ ) ,
∥σt ∥2 wt + st t
s̃∗ σ′ σt
 

ν̃F,t = ∗ t ∗ I3 − t 2 (σts̃ )′ ,
w̃t + s̃t ∥σt ∥

where I3 is a 3-dimensional identity matrix.

The proof is presented in Appendix A.15. The first term on the


right-hand side of Eq. (5.4) as is the mean-variance efficient portfolio,
because it is the solution of a myopic mean-variance investor. The
second term is referred to as the hedging portfolio, because, as we
will see in Section 9.B, it hedges the variations in the investment op-
portunities. Without the demand shock γt , σ3,t = 0, and the hedging
portfolio is zero as in the standard log cases.

5.A.4 Characterization of Equilibrium


An equilibrium in our economy is defined in a standard way: it is a
collection of goods and asset prices that satisfy certain conditions

( pt , p∗t , et , rt , rt∗ )

and consumption-investment policies

(c H,t , c F,t , c∗H,t , c∗F,t , xt , xt∗ )

such that (i) each consumer-investor maximizes their utility subject to


the budget constraint, and (ii) goods and bond markets clear.
In the presence of incomplete markets, the equilibrium allocation
is Pareto optimal. We need to employ a fictitious representative agent
with stochastic weights [Cuoco and He, 1994], which reflect the ef-
fects of market incompleteness. This fictitious representative agent

Electronic copy available at: https://ssrn.com/abstract=4668578


116 jiang – lecture notes on international finance

maximizes its utility subject to the resource constraints:


Z T 
∗ ∗
max∗ ∗
E exp (− δt ) ( u ( c t ) + π t u ( c t )) dt
c H,t ,c F,t ,c H,t ,c F,t 0

s.t. c H,t + c∗H,t = yt , c F,t + c∗F,t = y∗t ,

where the Pareto weight πt is stochastic. It can be shown that πt is


simply the ratio of the marginal utilities of either good of the two
countries’ households, which reflects the relative importance the
planner assigns to the foreign households relative to the home house-
holds.
The social planner’s optimization problem implies
γt α γt ( 1 − α )
c H,t = yt , c F,t = y∗ ,
γt α + π t γ ∗ ( 1 − α ) γt ( 1 − α ) + π t γ ∗ α t
(5.6)
π t γ ∗ (1 − α ) πt γ∗ α
c∗H,t = yt , c∗F,t = y∗ .
γt α + π t γ ∗ ( 1 − α ) γt ( 1 − α ) + π t γ ∗ α t
The equilibrium terms of trade that arise in the competitive equi-
librium is equal to the ratio of the marginal utilities of the home and
foreign goods for either country:
γt ( 1 − α )
       
∂u(c H,t , c F,t ) ∂u(c H,t , c F,t ) γt α
exp(tott ) = / = / ,
∂c H,t ∂c F,t c H,t c F,t
∂u∗ (c∗H,t , c∗F,t ) ∂u∗ (c∗H,t , c∗F,t )
! ! ! !
γ ∗ (1 − α ) γ∗ α
exp(tott ) = / = / .
∂c∗H,t ∂c∗F,t c∗H,t c∗F,t
From either equation, plug in Eq. (5.6) and we have:
γt α + πt γ∗ (1 − α) y∗t
exp(tott ) = .
γt ( 1 − α ) + π t γ ∗ α y t
We next use the no-arbitrage valuation principle to obtain the
valuation of the untraded stocks and the equilibrium wealth.
Lemma 5.2. Equilibrium home and foreign stock prices in their respective
local numéraires are given by
1 − exp(−δ( T − t)) 1 − exp(−δ( T − t)) ∗ ∗
st = pt yt , s∗t = pt yt ,
δ δ
and the wealth of the countries in their local numéraires are given by:
γt πt γ∗
wt + st = st , wt∗ + s∗t = s∗ .
γt α + π t γ ∗ ( 1 − α ) γt ( 1 − α ) + π t γ ∗ α t
The proof is presented in Appendix A.16. By Lemma 5.2, we can
express the home and foreign wealth in the home numéraire as
γt 1 − exp(−δ( T − t))
wt + st = ∗
pt yt ,
γt α + π t γ ( 1 − α ) δ
πt γ∗ 1 − exp(−δ( T − t)) ∗
w̃t∗ + s̃∗t = ∗
pt exp(−et )y∗t ,
γt ( 1 − α ) + π t γ α δ
which yield a simple interpretation of the stochastic Pareto weight πt .

Electronic copy available at: https://ssrn.com/abstract=4668578


incomplete markets 117

Lemma 5.3. The Pareto weight is determined by the relative wealth of the
two countries’ households, i.e., wt + st and w̃t∗ + s̃∗t , scaled by their discount
rates γt and γ∗ :

w̃t∗ + s̃∗t γt
πt = . (5.7)
wt + st γ ∗

As Eq. (5.6) shows, incomplete markets enrich the dynamics of the


economy with an additional state variable πt , which is determined
by the wealth distribution and the discount rates. As such, path
dependence naturally arises from incomplete-market models, as
the wealth distribution is not easily determined by the current state
variables, but rather reflects the history of the shocks to the economy.
With these results, we are now ready to characterize the equilib-
rium dynamics of the economy.

Proposition 5.2. (1) In an equilibrium, the stochastic weight that the


fictitious social planner assigns to the Foreign country, πt , has the following
dynamics:

dπt = πt (ν̃F,t − νH,t )′ dZt .

If both countries’ households hold their own endowment tree in period 0, we


can determine π0 as:
γ0
π0 = .
γ∗

(2) When such equilibrium exists, the log exchange rate dynamics follow

det = κt dt + σt dZt ,

where
2
1 1 1 1 ∥αγt ωi3 + γ∗ (1 − α)πt (ν̃F,t − νH,t )′ ∥2 1 ∥(1 − α)γt ωi3 + γ∗ απt (ν̃F,t − νH,t )′ ∥
κt = µy∗ − µy + σy2 − σy2∗ − +
2α − 1 2 2 2 (γt α + πt γ∗ (1 − α))2 2 ( γt ( 1 − α ) + π t γ ∗ α ) 2
1 αγt ωi3 + γ∗ (1 − α)πt (ν̃F,t − νH,t )′ (1 − α)γt ωi3 + γ∗ απt (ν̃F,t − νH,t )′
σt = −σy i1 + σy∗ i2 + ∗
− .
2α − 1 γt α + π t γ ( 1 − α ) γt ( 1 − α ) + π t γ ∗ α

(3) The log home state price density based on the home household’s con-
sumption, mνH ,t , follows

dmνH ,t = µm,t dt + σm,t dZt ,

where

α−1 1 1 ||αγt ωi3 + (1 − α)γ∗ πt (ν̃F,t − νH,t )′ ||2


µm,t = −δ + κt − µy + σy2 − ,
2α − 1 2 2 (γt α + πt γ∗ (1 − α))2
α−1 αγt ωi3 + (1 − α)γ∗ πt (ν̃F,t − νH,t )′
σm,t = σt − σy i1 + ,
2α − 1 γt α + π t γ ∗ ( 1 − α )

Electronic copy available at: https://ssrn.com/abstract=4668578


118 jiang – lecture notes on international finance

and the log foreign state price density based on the foreign household’s
consumption, m∗ν∗ ,t , follows
F

dm∗ν∗ ,t = µm∗ ,t dt + σm∗ ,t dZt ,


F

where

1−α 1 1 1 ||(1 − α)γt ωi3 + αγ∗ πt (ν̃F,t − νH,t )′ ||2


µm∗ ,t = −δ + κt − µy∗ + σy2∗ + (∥νH,t ∥2 − ∥ν̃F,t ∥2 ) − ,
2α − 1 2 2 2 ( γt ( 1 − α ) + π t γ ∗ α ) 2
1−α (1 − α)γt ωi3 + αγ∗ πt (ν̃F,t − νH,t )′
σm∗ ,t = σt − σy∗ i2 − (ν̃F,t − νH,t )′ + .
2α − 1 γt ( 1 − α ) + π t γ ∗ α

Then, we can introduce a stochastic wedge ηt that reconciles the log


exchange rate movement with the domestic and foreign consumption-based
SDFs:

det = dmνH ,t − dm∗ν∗ ,t + dηt , (5.8)


F

where
1
dηt = (∥νH,t ∥2 − ∥ν̃F,t ∥2 )dt + (νH,t − ν̃F,t )′ dZt .
2
(4) We can then express the market price of risk λt , and the interest rates
rt and rt∗ as functions of exogenous state variables.


λt = −σm,t − νH,t ,
1
rt = −µm,t − ∥σm,t ∥2 ,
2
1
rt∗ = −σt λt + κt − σt σt′ + rt .
2
The proof is presented in Appendix A.17. This proposition solves
the equilibrium exchange rate, SDF, and risk premium dynamics im-
plicitly as functions of νH,t and ν̃F,t , which are themselves functions
of the exchange rate dynamics and wealth distribution. The central
equation is (5.8), which shows that the exchange rate movement is
no longer equal to the SDF differential as we saw in the complete-
market model in Section 1.C. Instead, an additional wedge dηt arises
and gives to additional variations in the exchange rate. As such, we
refer to the dηt term as the incomplete-market wedge. However, this
wedge is driven by the same fundamental shocks dZt that drive the
SDF processes. In this sense, the additional exchange rate variations
in incomplete-market models are not necessarily non-fundamental.

5.A.5 Comparison with the Complete-Market Case


When the markets are complete, the model we consider reduces to
a special case of Pavlova and Rigobon [2007]. The social planner

Electronic copy available at: https://ssrn.com/abstract=4668578


incomplete markets 119

maximizes the following objective function with a constant Pareto


weight:

T
Z 

max E exp(−δt) (u(ct ) + πu (c∗t )) dt
c H,t ,c F,t ,c∗H,t ,c∗F,t 0

s.t. c H,t + c∗H,t = yt , c F,t + c∗F,t = y∗t ,

which implies

γt α γt ( 1 − α )
c H,t = yt , c F,t = y∗ ,
γt α + πγ∗ (1 − α) γt (1 − α) + πγ∗ α t
πγ∗ (1 − α) πγ∗ α
c∗H,t = yt , c∗F,t = y∗ .
γt α + πγ∗ (1 − α) γt (1 − α) + πγ∗ α t

The equilibrium terms of trade that arise in the competitive equi-


librium is equal to the ratio of the marginal utilities of the home and
foreign goods for either country:

γt ( 1 − α )
       
∂u(c H,t , c F,t ) ∂u(c H,t , c F,t ) γt α
exp(totcm
t )= / = / ,
∂c H,t ∂c F,t c H,t c F,t
∂u∗ (c∗H,t , c∗F,t ) ∂u∗ (c∗H,t , c∗F,t )
! ! ! !
γ ∗ (1 − α ) γ∗ α
exp(totcm
t )= / = / .
∂c∗H,t ∂c∗F,t c∗H,t c∗F,t

From either equation, plugging in the consumption solutions, we


have:

γt α + πγ∗ (1 − α) y∗t
exp(totcm
t )= .
γt (1 − α) + πγ∗ α yt

Compared to the incomplete-market solution, the Pareto weight


π in the complete-market case is a constant. After further simplifi-
cation, we characterize the exchange rate dynamics in the following
proposition.

Proposition 5.3. In the complete-market case, the log exchange rate follows

detcm = κtcm dt + σtcm dZt ,

where

1 1 1 1 α2 γt2 ω 2 1 (1 − α)2 γt2 ω 2


κtcm = µy∗ − µy + σy2 − σy2∗ − + ,
2α − 1 2 2 2 (γt α + πγ∗ (1 − α))2 2 (γt (1 − α) + πγ∗ α)2
( 1 − α ) γt ω
 
1 αγt ω
σtcm = −σy i1 + σy∗ i2 + − i3 .
2α − 1 γt α + πγ∗ (1 − α) γt (1 − α) + πγ∗ α

Moreover, the SDF dynamics satisfy

∗,cm
detcm = dmcm
t − dmt .

Electronic copy available at: https://ssrn.com/abstract=4668578


120 jiang – lecture notes on international finance

The proof is presented in Appendix A.18. This complete-market


case restores a continuous-time version of our standard risk-sharing
condition (1.16) under complete markets, i.e., ∆et+1 = mt+1 − m∗t+1 ,
which is in stark contrast to Eq. (5.8) under the incomplete-market
case.
Moreover, Eq. (5.7), reproduced below, also holds in the complete-
market case:

w̃t∗ γt
π= . (5.9)
wt γ∗

With a constant Pareto weight π, the equilibrium wealth distribution


is equal to the ratio of the discount rates. As such, when markets are
complete, the wealth distribution does not introduce additional state
variables.

5.B The No-Arbitrage Approach

The general equilibrium model above demonstrates one particular


case in which the exchange rate obeys Eq. (5.1), reproduced below,

∆et+1 = mt+1 − m∗t+1 + ηt+1 .

If we relax the assumption of log preference, the model becomes


much more difficult to solve in closed form. In this section, we de-
velop an alternative, no-arbitrage approach to study the proper-
ties of the exchange rate dynamics under incomplete markets. This
approach is based on Backus, Foresi, and Telmer [2001], Brandt,
Cochrane, and Santa-Clara [2006], Lustig and Verdelhan [2019].
This approach does not take a stance on the specification of pref-
erences and economic shocks. We only require risk-free bonds to
be freely traded across countries, which seems to be a reasonable
assumption at least for certain agents in developed economies. We fo-
cus on the four Euler equations (1.6)-(1.9) that characterize the home
and foreign investors’ valuation of the home and foreign risk-free
bonds. The financial markets for other assets, such as the equities
and long-term bonds, may or may not be traded. In this sense, this
no-arbitrage approach provides a general characterization of the
exchange rate dynamics under incomplete spanning.
As this approach leaves other markets unspecified, there is a fam-
ily of exchange rate solutions that are consistent with the Euler equa-
tions (1.6)-(1.9). However, this fact does not necessarily imply multi-
ple equilibria. A full specification of goods and asset market clearing
usually pins down a unique equilibrium within the family of equilib-
ria that we will study.

Electronic copy available at: https://ssrn.com/abstract=4668578


incomplete markets 121

Moreover, for any given home SDF mt+1 ,


def
m∗∗
t+1 = mt+1 − ∆et+1

is always a valid foreign SDF that is consistent with the asset market
equilibrium. While this SDF correctly prices the tradable assets, it
does not necessarily coincide with the foreign households’ marginal
utility growth. To the extent that we are interested in understanding
the relationship between the exchange rate dynamics and the macroe-
conomy, we would like to focus on the home and foreign SDFs that
reflect the marginal utility growth, i.e.,

u′ (c ) u∗ (c∗t+1 )
m t +1 = δ ′ t +1 , m∗t+1 =δ ′ .
u (ct ) u∗ (c∗t )

5.B.1 The η Algebra


To explain the restrictions implied by the Euler equations, we plug in
∆et+1 = mt+1 − m∗t+1 + ηt+1 to the Euler equation (1.9), reproduced
below,

1 = Et exp(m∗t+1 + ∆et+1 + rt ) ,
 

to obtain

1 = Et [exp(mt+1 + ηt+1 + rt )] .

In this equation, mt+1 + ηt+1 expresses the foreign households’ SDF


in the home numéraire, which prices the home risk-free rate.
Together with the Euler equation (1.6), reproduced below,

1 = Et [exp(mt+1 + rt )] ,

and assuming joint normality, they impose a restriction on the joint


distribution of mt+1 and ηt+1 :
1
Et [ηt+1 ] + vart (ηt+1 ) + covt (mt+1 , ηt+1 ) = 0. (5.10)
2
In the language of closed-economy incomplete-market models, we
can say that incomplete markets allow multiple SDFs to be consistent
with the asset market equilibrium, and mt+1 and mt+1 + ηt+1 are both
valid SDFs. However, mt+1 and mt+1 + ηt+1 cannot be arbitrary. The
fact that they both price the home risk-free bond implies a restriction,
which is what we derived in Eq. (5.10).
Similarly, home and foreign SDFs satisfy the following Euler equa-
tions,

1 = Et exp(m∗t+1 + rt∗ ) ,
 

1 = Et exp(m∗t+1 − ηt+1 + rt∗ ) ,


 

Electronic copy available at: https://ssrn.com/abstract=4668578


122 jiang – lecture notes on international finance

which imply a restriction on the joint distribution of m∗t+1 and ηt+1 :

1
Et [ηt+1 ] − vart (ηt+1 ) + covt (m∗t+1 , ηt+1 ) = 0. (5.11)
2

The restrictions (5.10) and (5.11) imply

0 = covt (mt+1 − m∗t+1 + ηt+1 , ηt+1 ) = covt (∆et+1 , ηt+1 ). (5.12)

This is the key implication of the Euler equations. The incomplete-


market wedge ηt+1 has to exist in the space orthogonal to the ex-
change rate movement ∆et+1 . Otherwise, it will affect how the ex-
change rate is priced, which makes the two SDFs mt+1 and mt+1 +
ηt+1 disagree on the pricing of the foreign currency.
Rearranging (5.12), we obtain

covt (mt+1 − m∗t+1 , ηt+1 ) = −vart (ηt+1 );

that is, in order for the incomplete-market wedge ηt+1 to be orthog-


onal to the exchange rate movement ∆et+1 , it has to be negatively
correlated with the SDF differential mt+1 − m∗t+1 . Then, the exchange
rate variance becomes

vart (∆et+1 ) = vart (mt+1 − m∗t+1 + ηt+1 ) = vart (mt+1 − m∗t+1 ) + vart (ηt+1 ) + 2covt (mt+1 − m∗t+1 , ηt+1 )
= vart (mt+1 − m∗t+1 ) − vart (ηt+1 ).

In other words, since the incomplete-market wedge ηt+1 is negatively


correlated with the SDF differential mt+1 − m∗t+1 , it reduces the ex-
change rate volatility relative to the volatility of the SDF differential.
In this way, using this result derived from the Euler equations
(1.6)-(1.9), Lustig and Verdelhan [2019] characterize the exchange
rate volatility, cyclicality, and risk premium in the presence of the
incomplete-market wedge ηt+1 .

Proposition 5.4. (a) The conditional exchange rate volatility can be ex-
pressed as

vart (∆et+1 ) = vart (mt+1 − m∗t+1 ) − vart (ηt+1 ).

(b) The conditional exchange rate cyclicality can be expressed as

covt (mt+1 − m∗t+1 , ∆et+1 ) = vart (∆et+1 ).

(c) The conditional currency risk premium can be expressed as

1
Et [rxt+1 ] = (vart (m∗t+1 ) − vart (mt+1 )) + Et [ηt+1 ].
2

Electronic copy available at: https://ssrn.com/abstract=4668578


incomplete markets 123

The proof is presented in Appendix A.19. When the markets are


complete, ηt+1 = 0, and the above results reduce to the complete-
market solution we considered in Section 1.C, reproduced below:

vart (∆et+1 ) = vart (mt+1 − m∗t+1 ),


covt (mt+1 − m∗t+1 , ∆et+1 ) = vart (∆et+1 ),
1
Et [rxt+1 ] = (vart (m∗t+1 ) − vart (mt+1 )).
2
Relative to this complete-market benchmark, the presence of the
incomplete-market wedge ηt+1 always reduces the exchange rate
volatility vart (∆et+1 ) relative to the variance of the SDF differential
vart (mt+1 − m∗t+1 ). This result shows that market incompleteness is
a useful ingredient for resolving the exchange rate puzzles, as we
expect the SDF volatility to be much higher than the exchange rate
volatility, while the cross-country SDF correlation is not high enough
to reduce vart (mt+1 − m∗t+1 ) [Brandt, Cochrane, and Santa-Clara,
2006]. In the presence of a volatile incomplete-market wedge ηt+1 , the
exchange rate volatility can be much lower than the SDF volatility.
Incomplete markets also shift the currency risk premium Et [rxt+1 ]
by Et [ηt+1 ]. The Euler equations do not rule out the possibility that
Et [ηt+1 ] is nonzero, as long as it satisfies Eq. (5.10) and (5.11). More-
over, because Et [ηt+1 ] is known ex-ante in period t, increasing the
currency risk premium comes at the cost of introducing exchange
rate predictability.
Finally, in incomplete markets, the exchange rate cyclicality is still
equal to the exchange rate variance: covt (mt+1 − m∗t+1 , ∆et+1 ) =
vart (∆et+1 ). However, because the exchange rate variance can be
lower, market incompleteness can shrink the exchange rate cyclicality
towards zero as well. That said, as the exchange rate cyclicality has to
be positive, incomplete markets alone cannot generate a procyclical
exchange rate.
Lustig and Verdelhan [2019] further point out a trade-off between
resolving the exchange rate volatility puzzle and the currency risk
premium puzzle. The Euler equations impose restrictions on the joint
dynamics of the exchange rate movement, the SDF, and ηt+1 . When
we rely on a volatile ηt+1 to reduce the exchange rate volatility and
cyclicality, we also shrink the currency risk premium. To see this
dilemma, recall from Proposition 1.1, which is valid in complete and
incomplete markets, that the currency risk premium can be expressed
as

log Et [exp(rxt+1 )] = −covt (m∗t+1 , ∆et+1 ).

When a volatile ηt+1 term lowers the exchange rate volatility and the
covariance between the SDF and the exchange rate movement, it also

Electronic copy available at: https://ssrn.com/abstract=4668578


124 jiang – lecture notes on international finance

reduces the currency risk premium.

5.B.2 Exchange Rate Projection


The η algebra above decomposes the exchange rate movement to the
SDF differential and the incomplete-market wedge η. From our dis-
cussion above, we know that a volatile η term is required to reduce
the exchange rate volatility, and η’s drift modifies the currency risk
premium.
In this subsection, we develop a different decomposition of the
exchange rate movement that is useful for interpreting the effects
of market incompleteness on the exchange rate dynamics, which
is based on Jiang [2023b]. For expositional convenience, we again
assume that all shocks are jointly normal.
Without loss of generality, we can express the exchange rate move-
ment as

∆et+1 = xt + yt mt+1 + zt m∗t+1 + wt ε t+1 , (5.13)

where ε t+1 is a standard normal variable with a mean of 0 and a


volatility of 1. This expression simply rewrites Eq. (5.1), with

ηt+1 = xt + (yt − 1)mt+1 + (zt + 1)m∗t+1 + wt ε t+1 .

We identify the coefficients xt , yt , zt , and wt by assuming that


the covariance between the SDFs and the residual ε t+1 is zero, i.e.,
covt (mt+1 , ε t+1 ) = covt (m∗t+1 , ε t+1 ) = 0, which uniquely pins down

covt (∆et+1 , mt+1 )vart (m∗t+1 ) − covt (∆et+1 , m∗t+1 )covt (mt+1 , m∗t+1 )
yt = ,
vart (mt+1 )vart (m∗t+1 ) − covt (mt+1 , m∗t+1 )2
covt (∆et+1 , m∗t+1 )vart (mt+1 ) − covt (∆et+1 , mt+1 )covt (mt+1 , m∗t+1 )
zt = ,
vart (mt+1 )vart (m∗t+1 ) − covt (mt+1 , m∗t+1 )2
xt = Et [∆et+1 ] − yt Et [mt+1 ] − zt Et [m∗t+1 ],
wt ε t+1 = ∆et+1 − xt − yt mt+1 − zt m∗t+1 .

Intuitively, we can think of this decomposition formula as a projec-


tion of the exchange rate movement on the home and foreign SDFs.
The terms yt mt+1 and zt m∗t+1 describe how the bilateral exchange
rate loads on the SDFs in the two countries, with the coefficients yt
and zt interpreted as the SDF-FX pass-through. When the markets
are complete, the SDFs comove with the exchange rate movement
one-for-one (i.e., ∆et+1 = mt+1 − m∗t+1 ), implying a full pass-through
of yt = 1 and zt = −1. As we will see later, when the markets are
incomplete, |yt | and |zt | can take values lower than 1, reflecting a
partial pass-through from the SDFs to the exchange rate.

Electronic copy available at: https://ssrn.com/abstract=4668578


incomplete markets 125

The last term wt ε t+1 captures the residual in the bilateral exchange
rate movement that is not explained by the SDFs. Compared to the
ηt+1 term in Eq. (5.1), the residual wt ε t+1 is orthogonal to the SDFs,
whereas ηt+1 can have arbitrary correlations with both countries’
SDFs.
Under joint normality, the exchange rate dynamics can be repre-
sented by ( xt , yt , zt , wt ). Alternatively, they can also be represented
by the expected exchange rate movement Et [∆et+1 ], the bilateral ex-
change rate movement’s variance vart (∆et+1 ), and its covariances
with the two countries’ SDFs covt (∆et+1 , mt+1 ) and covt (∆et+1 , m∗t+1 ).
Using this exchange rate decomposition, we can provide a differ-
ent characterization of the exchange rate volatility, cyclicality, and risk
premium [Jiang, 2023b].

Proposition 5.5. (a) The conditional exchange rate volatility can be ex-
pressed as

vart (∆et+1 ) = vart (yt mt+1 + zt m∗t+1 ) + vart (wt ε t+1 ).

(b) The conditional exchange rate cyclicality can be expressed as

covt (mt+1 − m∗t+1 , ∆et+1 ) = covt (yt mt+1 + zt m∗t+1 , mt+1 − m∗t+1 ).

(c) Using the Euler equations (1.6)–(1.9), the conditional currency risk
premium can be expressed as

def 1
rpt = Et [rxt+1 ] = − covt (yt mt+1 + zt m∗t+1 , mt+1 + m∗t+1 ).
2
The proof is in Appendix A.20. As a special case, when markets
are complete, yt = 1, zt = −1 and ε t+1 = 0. These exchange rate
moments simplify to

vart (∆etcm
+1 ) = vart ( mt+1 − mt+1 ),
covt (mt+1 − m∗t+1 , ∆etcm ∗
+1 ) = vart ( mt+1 − mt+1 ),
1 ∗ ∗ 1 ∗ 1
rpcm
t = − covt ( mt+1 − mt+1 , mt+1 + mt+1 ) = vart ( mt+1 ) − vart ( mt+1 ),
2 2 2
which recovers the complete-market solution in Section 1.C.
When markets are incomplete, if the magnitudes of the SDF-FX
pass-through coefficients yt and zt are smaller than 1, then, we obtain
a less volatile SDF term yt mt+1 + zt m∗t+1 , which reduces the exchange
rate’s variance and covariance with the SDF differential towards 0.
At the same time, a smaller SDF term also reduces the currency risk
premium. In this way, we recover the dilemma in Lustig and Verdel-
han [2019] that, as market incompleteness reduces the exchange rate
volatility, it also shrinks the currency risk premium towards zero. By
teasing out the SDF term yt mt+1 + zt m∗t+1 from the residual wt ε t+1 in

Electronic copy available at: https://ssrn.com/abstract=4668578


126 jiang – lecture notes on international finance

the exchange rate movement, our decomposition formula shows how


this SDF term ties together the exchange rate variance, cyclicality, and
the currency risk premium in the general incomplete-market setting.
Moreover, the role of the residual term wt ε t+1 is also simple to in-
terpret: while a volatile residual increases the exchange rate variance,
it has no effect on the exchange rate cyclicality and the currency risk
premium. It can be thought of as a truly idiosyncratic term.
Conversely, given the fact that the exchange rate variance is much
lower than the variance of the SDF differential [Brandt, Cochrane,
and Santa-Clara, 2006], can we derive a bound on the magnitude of
the coefficients yt and zt ? More specifically, at the annual frequency,
the exchange rate volatility between developed countries is in the
order of 10%, implying vart (∆et+1 ) = 1%. In comparison, given the
existence of financial assets or trading strategies that generate Sharpe
ratios above 1, the Hansen and Jagannathan [1991] bound implies
vart (mt+1 ) ≥ 1. Together, these empirical observations imply the
following restriction:

vart (∆et+1 ) ≤ 1% · vart (mt+1 ). (5.14)

Then, using the Euler equations (1.6)–(1.9), we can derive the fol-
lowing bound on the exchange rate variance.

Proposition 5.6. The Euler equations impose the following constraints on


the pass-through coefficients yt and zt :

yt (yt − 1)vart (mt+1 ) + zt (zt + 1)vart (m∗t+1 )+


[yt (zt + 1) + zt (yt − 1)]covt (mt+1 , m∗t+1 ) ≤ 0 (5.15)
yt vart (mt+1 ) − zt vart (m∗t+1 ) + (zt − yt )covt (mt+1 .m∗t+1 ) = vart (∆et+1 )
(5.16)

The proof is in Appendix A.21. This result shows that, if we take


a stance on the variances of the SDFs, we can reject some values of
the pass-through coefficients by noting that they generate too much
exchange rate volatility. For example, if yt = 1 and zt = −1 as in
the complete-market case, Eq. (5.15) becomes 0 ≥ 0 and Eq. (5.16)
becomes vart (mt+1 − m∗t+1 ) = vart (∆et+1 ), which recovers the key
result in Brandt, Cochrane, and Santa-Clara [2006].
This proposition generalizes Brandt, Cochrane, and Santa-Clara
[2006] by asking how much lower the pass-through coefficients yt
and zt have to be in order to match the exchange rate volatility. If we
plug the empirical restriction (5.14) into Eq. (5.16), and additionally
assume that the home and foreign SDFs have the same variance,
i.e., σt2 = vart (mt+1 ) = vart (m∗t+1 ), and that the cross-country
SDF correlation is corrt (mt+1 , m∗t+1 ) = 0.3, which is close to the
average cross-country correlation in consumption growth among

Electronic copy available at: https://ssrn.com/abstract=4668578


incomplete markets 127

developed countries, then, we can graphically represent the two


equations in Proposition 5.6 in Figure 5.1. Specifically, Eq. (5.15)
requires the coefficients to lie within the red ellipse, and Eq. (5.16)
requires the coefficients to lie above the blue line. The intersection of
these two regions is very small and close to (0, 0).

Figure 5.1: Bounds on the Magnitude of


the Pass-Through Coefficients
1.5

0.5

-0.5

-1

-1.5
-1.5 -1 -0.5 0 0.5 1 1.5

This discussion shows that the pass-through coefficients yt and


zt have to be very small in order to produce empirically plausible
exchange rate volatility. In the numerical example we consider here,
the greatest possible magnitude of either coefficient is 0.068, which
is much smaller than the complete-market benchmark of 1. As such,
this exchange rate decomposition takes one step further and sheds
light on the economic interpretation of the incomplete-market wedge
ηt+1 in relation to the SDFs.

5.C Multi-Currency Dynamics and International Spill-Over

Market incompleteness also has implications for multi-currency dy-


namics. As a benchmark, let us first understand the intuition from
complete-market models. It suffices to consider only three countries,
which we denote by 0, 1, 2. We refer to country 0 as the home coun-
try, and countries 1 and 2 as the two foreign countries. The bilateral
exchange rate movements between the home country and the two
foreign countries are given by
(0) (1)
∆e0/1
t +1 = m t +1 − m t +1 ,
(0) (2)
∆e0/2
t +1 = m t +1 − m t +1 .

First, Triangular arbitrage means that the bilateral exchange rates


between two foreign countries are directly implied from their bilat-

Electronic copy available at: https://ssrn.com/abstract=4668578


128 jiang – lecture notes on international finance

eral exchange rates with the home country:


(1) (2)
∆e1/2 0/2 0/1
t+1 = ∆et+1 − ∆et+1 = mt+1 − mt+1 . (5.17)

As a result, the three bilateral exchange rates are not mutually inde-
pendent. Given the simplicity of Eq. (5.17), it suffices to understand
the dynamics of the bilateral exchange rates between the home coun-
try and each foreign country. The dynamics between the two foreign
countries are easily implied.

5.C.1 Restrictions from the Euler Equations


When the markets are incomplete, the simple mapping from the
SDFs to the bilateral exchange rates no longer holds. It is no longer
sufficient to only study the dynamics between the home country and
each foreign country. To see this point, consider the Euler equations
(1.6)–(1.9) for holding the home and foreign risk-free bonds. Rewrit-
ing Eq. (1.12) for countries 0 and 1, we obtain
(0) (1)
covt (mt+1 − mt+1 − ∆e0/1 0/1
t+1 , ∆et+1 ) = 0. (5.18)

Similarly, for countries 0 and 2, we obtain


(0) (2)
covt (mt+1 − mt+1 − ∆e0/2 0/2
t+1 , ∆et+1 ) = 0. (5.19)

These equations arise in a large class of two-country models in


which risk-free bonds are traded. They are also part of Proposition
5.4 which characterizes the exchange rate cyclicality.
Next, we consider the Euler equations for risk-free bonds between
countries 1 and 2, which are
h i
(1) (1)
1 = Et exp(mt+1 + rt ) ,
h i
(1) (2)
1 = Et exp(mt+1 − ∆e1/2 t +1 + r t ) ,
h i
(2) (2)
1 = Et exp(mt+1 + rt ) ,
h i
(2) (1)
1 = Et exp(mt+1 + ∆e1/2 t +1 + r t ) .

The following proposition shows that these equations give rise to


additional restrictions on the exchange rate dynamics [Jiang, 2023b].

Proposition 5.7. The Euler equations for risk-free bonds between countries
1 and 2 imply
(0) (1) (0) (2)
covt (mt+1 − mt+1 − ∆e0/1 0/2 0/2 0/1
t+1 , ∆et+1 ) + covt ( mt+1 − mt+1 − ∆et+1 , ∆et+1 ) = 0.
(5.20)

The proof is presented in Appendix A.22. This proposition shows


that, while the bilateral exchange rate movements ∆e0/1 0/2
t+1 and ∆et+1

Electronic copy available at: https://ssrn.com/abstract=4668578


incomplete markets 129

between the home country and each foreign country directly imply
the bilateral exchange rate movement ∆e2/1
t+1 between the two foreign
countries, the Euler equations (5.18) and (5.19) between the home
country and each foreign country do not imply the Euler equations
(5.20) between the two foreign countries in the general case. As a
result, when we extend our analysis from two to three countries, the
number of independent exchange rate movements increases from 1
to 2, while the number of unique sets of Euler equations for cross-
country bond holdings increases from 1 to 3. The additional Euler
equations impose restrictions on the exchange rate dynamics that are
absent in the two-country setting.
(0) (i )
The complete-market case is an exception: mt+1 − mt+1 − ∆e0/i t +1 =
0 satisfies all the conditions above trivially. Perhaps this is why we
are used to studying only the bilateral dynamics between the home
country and each foreign country in the complete-market setting,
which is no longer sufficient when markets are incomplete, as the
Euler equations between the two foreign countries impose additional
restrictions on the exchange rate dynamics.
What do these additional restrictions imply for the exchange rate
dynamics? To answer this question, we first study a specific three-
country model, and then we generalize the results using the no-
arbitrage approach.

5.C.2 Motivating Example for the Spill-Over Effects


Let us consider a three-country extension of the Pavlova and Rigobon
[2012] model we considered in Section 5.A. Again, we consider three
countries labeled as 0, 1 and 2. Each country i has a unique variety of
local goods, and a representative agent with log preference scaled by
(i )
a demand shock γt :
T
Z 
(i ) (i )
E exp(−ρt)γt log ct ,
0

and the aggregate consumption is a Cobb-Douglas aggregate of the


(i )
three countries’ local goods. Specifically, let c j,t denote country i’s
agent’s consumption of country j’s goods, then,

1−α
log ct = α log ci,t + ∑
(i ) (i ) (i )
log c j,t .
j ̸ =i
2

Reflecting home bias towards domestic goods, α > 1/3. Moreover,


(i )
each country’s agent receives a stochastic endowment yt that fol-
lows a simple diffusion process.
The agents in three countries have access to the same asset space.
Since the markets are incomplete, the set of tradable assets does

Electronic copy available at: https://ssrn.com/abstract=4668578


130 jiang – lecture notes on international finance

not span the shocks. The risk-free bonds are freely tradable, but the
existence of other assets such as equities and long-term bond is not
central to our result. For simplicity, we assume that the agents can
also trade equities which are claims to the country-level endowment
claims.
(i )
Let w̃t denote the wealth of country i’s agent, expressed in a
common numéraire. We have the following result characterizing the
equilibrium exchange rate.
Proposition 5.8. The equilibrium log real exchange rate between the foreign
countries 1 and 2 can be expressed as
 (1) (2) 
1− α w̃t w̃t 1−α
2 + (0) α + (0) 2 ( 2 )
3α − 1  w̃t w̃t yt 
e1/2
t = log
( 1 ) ( 2 )
+ log (1) 
,
2  1− α w̃t 1−α w̃t y
2 + (0) 2 + (0) α
w̃t w̃t
t

(1) (2)
which depends not only on the endowments yt and yt , but also on
(1) (0)
the endogenous wealth distribution between agents, i.e., w̃t /w̃t and
(2) (0)
w̃t /w̃t .
The proof is presented in Appendix A.23. This expression relates
the bilateral exchange rate between the foreign countries 1 and 2
to the wealth of country 0’s agent. In particular, a shock to country
(0)
0’s endowment yt could directly impact the foreign countries’ bi-
lateral exchange rate by affecting the wealth distribution. This can
happen when the agents in countries 1 and 2 hold different portfolios
according to their heterogeneous wealth and preferences, which ex-
pose them differently to country 0’s endowment shock. In this way,
a shock originating from country 0 affects the bilateral exchange rate
between countries 1 and 2, which gives us a concrete example of the
spill-over effect.
In stark contrast, when the markets are complete, full risk-sharing
implies the following wealth distribution:
(i ) (i )
π (i ) w /γ
= (t0) t(0) ,
π (0) wt /γt

where π (i) is the time-invariant Pareto weight in the social planner’s


optimization. Then, the log exchange rate can be expressed as
 (1) (2) 
1− α γt π (1 ) γt π (2 ) 1 − α
2 + (0) (0) α + (0) (0) 2 ( 2 )
3α − 1  γt π γt π y 
e1/2,cm
t = log
(1) (2)
+ log t(1)  , (5.21)
2 
1− α γt π (1 ) 1 − α γt π (2 ) y
2 + (0) (0) 2 + (0) (0) α
γt π γt π
t

(1)
which is only a function of the country-specific endowments yt and
(2) (0) (1) (2)
yt and demand shocks γt , γt , and γt .

Electronic copy available at: https://ssrn.com/abstract=4668578


incomplete markets 131

In this way, this specific setting shows a clear distinction between


incomplete- and complete-market outcomes: while a shock to coun-
(0)
try 0’s endowment yt has no impact on the bilateral exchange rate
between foreign countries 1 and 2 when markets are complete, it can
directly affect the bilateral exchange rate between foreign countries
when markets are incomplete. In other words, when markets are
incomplete, not only does the bilateral exchange rate movement de-
viate from the two countries’ marginal utility growth differential, i.e.,
∆et+1 ̸= mt+1 − m∗t+1 , but the bilateral exchange rate also depends on
shocks to the third country. A natural question is whether this inter-
national spill-over effect is a general feature of incomplete markets,
which is explored in the next section.

5.C.3 General Characterization of the Spill-Over Effects (TODO)

5.D Incomplete Markets vs. Convenience Yields (TODO)

Electronic copy available at: https://ssrn.com/abstract=4668578


6
Monetary and Fiscal Policies

Summary

• We develop a stylized model in the New Keynesian tradition. In this model, when prices are
flexible, monetary and fiscal policies have no real effects: they do not affect consumption and
real exchange rate. They only affect inflation, which affects the nominal interest rate and the
nominal exchange rate.

• When prices are sticky, monetary and fiscal policies have real effects: they influence the real
interest rate, which affects consumption and the real exchange rate.

• Monetary and fiscal policies are not necessarily independent. Taking a strong stance on one
policy can affect the other policy.

• While it is beyond the scope of this chapter, monetary and fiscal policies can affect the ex-
change rate dynamics through other channels such as the risk premium and the convenience
yield as well.

The monetary policies refer to the actions taken by the central


bank, which is the Federal Reserve in the U.S. The policy tool in-
cludes setting interest rates, controlling the money supply, and con-
ducting other quantity-based operations such as security purchases
and sales. The fiscal policies refer to the government’s taxation and
spending decisions, and the corresponding debt issuance and retire-
ment.
These monetary and fiscal policies play important roles in shap-
ing the exchange rate dynamics. Studying their effects on the ex-
change rate marks the birthplace of the international macro literature
[Mundell, 1963, Fleming, 1962, Dornbusch, 1976, Frankel, 1979]. In
this chapter, we develop a stylized but unified model of monetary
and fiscal policies to capture the baseline channels, in which mone-
tary and fiscal policies affect the exchange rate dynamics by shaping
the real risk-free rates. It represents a standard way of modeling
these policies in the macro and monetary literature. However, given

Electronic copy available at: https://ssrn.com/abstract=4668578


monetary and fiscal policies 133

our discussion in the beginning of Chapter 3, it is likely that mone-


tary and fiscal policies also affect the exchange rates through the risk
premium and the convenience yield channels. We return to this issue
at the end of this chapter.

6.A Introducing the Nominal Layer

In this chapter, we will consider a model with the nominal layer for
the first time. Let us begin with some basic definitions and character-
izations, which apply to not only this model but most models with
a nominal layer. In particular, we may introduce home and foreign
currencies to the baseline model in Section 1.A and consider it as a
special case.
Let Pt denote the price index in the home country, which measures
the price of the home consumption bundle in the unit of the local
currency. For example, if one unit of the U.S. home consumption
bundle costs 100 dollars, then the U.S. price level Pt = 100. Inflation
is defined as the change in the price level:
def
πt = ∆ log Pt .
Recall that ct is the real consumption. We use Ct = Pt ct to denote
the nominal consumption. In the same example, if the U.S. house-
holds consume 10 units of the home consumption bundle, then the
real consumption is ct = 10 units of the consumption bundle and
the nominal consumption is Ct = 10 × 100 = 1000 dollars. More
generally, we use uppercase letters to denote nominal prices and
quantities, and we use lowercase letters to denote real prices and
quantities. The nominal interest rate it and the inflation rate πt are
the exceptions to this notation, as it is common to express them in
lowercase.1 1
In our derivation, we will also use the
Recall that the log real exchange rate et measures the conversion lower case pt (h) to denote the price of a
specific variety of goods, which helps to
ratio between the home and foreign countries’ consumption baskets, distinguish it from the aggregate price
and a higher value means that the home consumption basket is more index Pt .

expensive. We similarly define the nominal exchange rate exp(Et ) as


the conversion ratio between the home and foreign currencies. It is
related to the real exchange rate et and the price levels Pt and Pt∗ via
Pt
exp(et ) = exp(Et ) .
Pt∗
If we fix the real exchange rate et , a higher home inflation Pt is associ-
ated with a lower nominal value of the home currency Et .
Finally, we use the nominal SDF to discount nominal returns and
nominal cash flows. The home nominal SDF Mt+1 is defined as
def
Mt + 1 = m t + 1 − π t + 1 . (6.1)

Electronic copy available at: https://ssrn.com/abstract=4668578


134 jiang – lecture notes on international finance

If an asset with real return r̃t+1 , its nominal return is r̃t+1 + πt+1 . We
can price this asset either by

1 = Et [exp(mt+1 + r̃t+1 )],

or equivalently by

1 = Et [exp( Mt+1 + r̃t+1 + πt+1 )].

6.A.1 Exchange Rate Accounting


Next, let us relate these concepts to the exchange rate accounting
exercises in Section 1.B and 4.B. Starting from the Euler equation for
the nominal bond return, we have

1 = Et [exp( Mt+1 + it )] ,

which implies

exp(−it ) = Et [exp( Mt+1 )] = Et [exp(mt+1 − πt+1 )] .

Assuming joint normality, we can express the nominal interest rate


as
1 1
it = −Et [mt+1 ] − vart (mt+1 ) + Et [πt+1 ] − vart (πt+1 ) + covt (mt+1 , πt+1 )
2 2
= rt + Et [πt+1 ] + irpt , (6.2)

where
def 1
irpt = covt (mt+1 , πt+1 ) − vart (πt+1 )
2
denotes the inflation risk premium. This expression shows that the
nominal interest rate it can be decomposed into the real interest
rate rt , the expected inflation rate Et [πt+1 ], and the inflation risk
premium irpt .
Define the nominal currency excess return as

RXt+1 = ∆Et+1 + it − it∗ .

Then, we obtain the nominal version of Proposition 1.1, which states


that, in the absence of bond convenience yields, the home currency’s
expected nominal excess return is determined by its risk premium
captured by its covariance with the SDF plus a Jensen’s term.

Proposition 6.1. The home currency’s expected log excess return is deter-
mined by the covariance between the log foreign SDF and log exchange rate
movement minus a Jensen’s term:
def 1
RPt = Et [ RXt+1 ] = −covt ( Mt∗+1 , ∆Et+1 ) − vart (∆Et+1 ).
2

Electronic copy available at: https://ssrn.com/abstract=4668578


monetary and fiscal policies 135

However, I hesitate to generalize the real exchange rate decom-


position in Proposition 1.2 to the nominal case, because it relies on
exchange rate stationarity. Specifically, assuming that the nominal
exchange rate is stationary, we can express its level as
∞ ∞
Et = ∑ Et [it+ j − it∗+ j ] − ∑ Et [ RPt+ j ] + Ē . (6.3)
j =0 j =0

What prevents this formula from describing a realistic situation is


that the price level is usually considered to be non-stationary. For
example, suppose the real exchange rate and the foreign inflation
are fixed, and the U.S. experiences a persistent increase in inflation,
which raises the price level Pt today and increases the expected fu-
ture inflation. Holding the real rate constant, the latter effect also
raises the U.S. nominal interest rate it+ j in the future. Then, from
exp(et ) = exp(Et ) Pt /Pt∗ , we should expect the home inflation to de-
preciate the home currency in nominal terms, i.e., Et declines. On the
other hand, Eq. (6.3) implies that a higher expected path of the home
nominal interest rate it+ j should imply a stronger home currency in
nominal terms, i.e., Et increases, which leads to a contradiction.
The reason we obtain this contradiction is that Eq. (6.3) requires
the price level differential to be stationary. If the home price level
increases today, it has to decline in the future, so that the home cur-
rency’s nominal depreciation implied from exp(et ) = exp(Et ) Pt /Pt∗
and a higher Pt is consistent with lower expected home nominal in-
terest rates it+ j due to lower expected inflation.

6.B Model Set-Up

We consider a two-country model that incorporates firm mark-up


and sticky prices. Most ingredients in the model including the mon-
etary setting follow Corsetti and Pesenti [2007], and the fiscal part
follows Jiang [2022]. This model is a simpler variant of the New Key-
nesian/DSGE models.
In this section, we will first characterize the model under flexible
prices, which generates the results that are similar to the model de-
veloped in Section 1.A. In the next section, we will characterize the
model under sticky prices and explore its implications.

6.B.1 Households
There are two countries, home and foreign. Each country contains
a unit mass of households, a unit mass of firms, and a government.
Home households are indexed by j ∈ [0, 1], and home firms are
indexed by h ∈ [0, 1]. Foreign households are indexed by j∗ ∈ [0, 1],

Electronic copy available at: https://ssrn.com/abstract=4668578


136 jiang – lecture notes on international finance

and foreign firms are indexed by f ∈ [0, 1]. Each firm produces a
unique variety of good, which is an imperfect substitute for other
varieties.
The lifetime expected utility of home household j is
∞ ∞
ℓ t ( j )1+ ν
 
def
E0 ∑ δ ut ( j) = E0 ∑ δ log ct ( j) − κ
t t
,
t =0 t =0
1+ν

where ℓt ( j) is the labor effort, and ν > 0 is the labor curvature coef-
ficient. ct ( j) is the consumption composed of a Cobb-Douglas basket
of home and foreign bundles:

ct ( j) = c H,t ( j)α c F,t ( j)1−α ,

and c H,t ( j) and c F,t ( j) are constant-elasticity-of-substitution (CES)


bundles of home and foreign varieties:
Z 1
1/(1−1/ρ) Z 1
1/(1−1/ρ)
1−1/ρ 1−1/ρ
c H,t ( j) = ct (h, j) dh , c F,t ( j) = ct ( f , j) df .
0 0

The parameter α > 1/2 measures home bias in consumption, and


the parameter ρ is the elasticity of substitution across varieties.
Let pt (h) and pt ( f ) denote the home-currency prices of varieties h
and f . Let PH,t and PF,t denote the home-currency prices of the home
and foreign bundles c H,t ( j) and c F,t ( j), which can be shown to be the
CES indices with elasticity 1/ρ:
Z 1
1/(1−ρ) Z 1
1/(1−ρ)
1− ρ 1− ρ
PH,t = pt ( h) dh , PF,t = pt ( f ) df .
0 0

Similarly, let Pt denote the home-currency price of the aggregate


consumption basket in the home country. This utility-based CPI can
be expressed as

1 α 1− α
Pt = PH,t PF,t .
α α (1 − α )1− α

Moreover, the solution of the within-period problem is similar to


that in Section 1.A. The home household makes the following choices
for different goods:
 −ρ  −ρ
pt ( h) pt ( f )
ct (h, j) = c H,t , ct ( f , j) = c F,t ,
PH,t PF,t

and
Pt Pt
c H,t = α ct , c F,t = (1 − α) ct .
PH,t PF,t

Having specified the household preferences and the within-period


solution, we now turn to the households’ budget constraint and their

Electronic copy available at: https://ssrn.com/abstract=4668578


monetary and fiscal policies 137

intertemporal solution. Home household j owns the portfolio of


home firms and provides labor to the firms. It earns a nominal wage
Wt and receives a nominal dividend Dt ( j) from the firms. It also pays
tax τt ( j) to the government and purchases consumption ct ( j), both
denoted in real terms.
The financial markets are complete. Let σt denote the state of the
economy at time t. Let Θ(σt+1 |σt ) denote the time-t home-currency
price for one unit of home currency delivered at time t + 1 contin-
gent on the state being σt+1 . At time t, home household j holds
Ωt (σt+1 , j) unit of the Arrow-Debreu security that pays off in state
σt+1 . Θ∗ (σt+1 |σt ) is similarly defined as the time-t foreign-currency
price for one unit of foreign currency delivered at time t + 1 contin-
gent on the state being σt+1 , and Ω∗t (σt+1 , j) is the quantity of this
security held by home household j.
Let exp(Et ) denote the nominal exchange rate, in the unit of for-
eign currency per home currency. A higher nominal exchange rate
means a stronger foreign currency. Then, the individual household’s
budget constraint at time t and state σt is

∑ Ωt (σt+1 , j)Θ(σt+1 |σt ) + exp(−Et ) ∑ Ω∗t (σt+1 , j)Θ∗ (σt+1 |σt )


σt+1 σt+1

≤ Wt ℓt ( j) + Dt ( j) − Pt τt ( j) − Pt ct ( j) + Ωt−1 (σt , j) + exp(−Et )Ω∗t−1 (σt , j).

The home household’s Lagrangian is



ℓ t ( j )1+ ν
 
L( j) = max E0 ∑ δt log ct ( j) − κ
t =0
1+ν

− E0 ∑ δt ζ t ( j){ ∑ Ωt (σt+1 , j)Θ(σt+1 |σt ) + exp(−Et ) ∑ Ω∗t (σt+1 , j)Θ∗ (σt+1 |σt )
t =0 σt+1 σt+1

− Wt ℓt ( j) − Dt ( j) + Pt τt ( j) + Pt ct ( j) − Ωt−1 (σt , j) − exp(−Et )Ω∗t−1 (σt , j)}

The first-order conditions imply


1
ζ t ( j) = ,
Pt ct ( j)
Pt ct ( j)
Θ(σt+1 |σt ) = δP(σt+1 |σt ) ,
Pt+1 ct+1 ( j)
κ ℓt ( j)ν Pt ct ( j) = Wt .

We define the home country’s nominal SDF as

def Pt ct ( j)
exp( Mt+1 ) = δ ,
Pt+1 ct+1 ( j)
which is related to its real SDF via
def ct
exp(mt+1 ) = δ = exp( Mt+1 + πt+1 ).
c t +1

Electronic copy available at: https://ssrn.com/abstract=4668578


138 jiang – lecture notes on international finance

Since the markets are complete, we can show that the nominal
exchange rate movement equals the ratio between the two countries’
nominal SDFs:

exp( Mt+1 ) = exp( Mt∗+1 − Et + Et+1 ),


Pt ct ( j) exp(−Et ) Pt∗ c∗t ( j∗ )
= .
Pt+1 ct+1 ( j) exp(−Et+1 ) Pt∗+1 c∗t+1 ( j∗ )

To keep the algebra simple, we assume the initial nominal ex-


change rate level is pinned down by P0 c0 ( j) = exp(−E0 ) P0∗ c0∗ ( j) at
time 0. Since the market is complete, this assumption amounts to tak-
ing a stance on the two countries’ wealth distribution at time 0. Then,
the nominal exchange rate satisfies

Pt ct ( j) = exp(−Et ) Pt∗ c∗t ( j∗ )

for all time t.


In this way, we recover the complete-market solution of the ex-
change rate movement in nominal terms:

∆Et+1 = Mt+1 − Mt∗+1

By symmetry, all households in each country have the same con-


sumption, saving, and labor decisions. Therefore, we can drop their
indices j and j∗ .

6.B.2 Firms
Each home firm produces a variety h using labor supplied by home
households. The production technology is linear in labor input:

y t ( h ) = z t ℓ t ( h ),

where yt (h) is the output of firm h, ℓt (h) is the labor input, and zt
is a productivity process common to all home firms. To produce
yt (h) units of goods, firm h faces a wage cost of Wt ℓt (h). Similarly,
the foreign firm has a production function linear in labor input with
productivity z∗t .
Aggregating across home and foreign households, we obtain the
total demand for variety h:
Z 1 Z 1
yt ( h) = ct (h, j)dj + c∗t (h, j)dj.
0 0

The firm’s nominal dividend is


Z 1 Z 1
Dt ( h ) = p t ( h ) ct (h, j)dj + exp(−Et ) p∗t (h) c∗t (h, j)dj − Wt ℓt (h).
0 0

Electronic copy available at: https://ssrn.com/abstract=4668578


monetary and fiscal policies 139

The firm’s objective function is to maximize the present value of


this dividend stream. Let MCt denote the nominal marginal cost:
def Wt
MCt = .
zt
Then, plugging in the solution from the households’ within-period
problem, we obtain
pt ( h ) −ρ
 
Dt (h) = ( pt (h) − MCt ) c H,t
PH,t
!−ρ (6.4)
∗ p∗t (h)
+ (exp(−Et ) pt (h) − MCt ) ∗ c∗H,t ,
PH,t
which means that the firm’s profit is equal to the profit margin times
the quantity of goods sold in both countries. Since the firms pro-
duce unique goods varieties that are imperfect substitutes, the profit
margin and the profit are nonzero.

6.B.3 Fiscal Policy


We assume that each country’s fiscal authority controls its govern-
ment tax and spending processes. For tractability, we make the fol-
lowing simplifying assumptions. First, tax and spending fall on the
same basket of goods as the households’ consumption, which is an
aggregate of both home and foreign varieties. For the home govern-
ment, let τt denote its tax revenue and let gt denote its spending in
real terms.
Second, government spending is not remitted back to domes-
tic households. As will be shown in the next section, under this
assumption and flexible prices, government spending reduces the
available goods for household consumption and behaves like a pos-
itive demand shock in Section 3.C. This standard mechanism will
be contrasted with the new effects generated by the intertemporal
government budget condition and sticky prices.
Third, governments only issue one-period debt in local currency
units. The government debt does not default on its notional value,
but its real value can vary due to inflation. It is therefore equivalent
to a claim that pays off one unit of local currency in every state. Let
Qt+1 denote the quantity of outstanding home government debt
that is issued at time t and due at time t + 1. Let it denote the home
nominal interest rate. The government budget condition in nominal
terms is
Qt + Pt gt = Pt τt + Qt+1 exp(−it ). (6.5)
def
Let st = τt − gt denote the home real government surplus. Sim-
def
ilarly, let s∗t = τt∗ − gt∗ denote the foreign real government surplus.

Electronic copy available at: https://ssrn.com/abstract=4668578


140 jiang – lecture notes on international finance

Iterate forward the Home government’s intertemporal budget condi-


tion,

" #  
Qt Q
= Et ∑ exp(mt,t+k )st+k + lim Et exp(mt,t+T ) t+T .
Pt k =0
T →∞ Pt+T

The transversality condition requires that the terminal value in the


intertemporal budget condition vanishes:
 
Qt+ T
lim Et exp(mt,t+T ) = 0,
T →∞ Pt+T
which is equivalent to requiring that the present value of government
surpluses grows slower than the real discount rate:

" !#
lim Et exp(mt,t+T )
T →∞
∑ exp(mt+T,t+T+k )st+T+k = 0.
k =0

We will further discuss this transversality condition in Section 8.C.


Under this transversality condition, the government’s intertempo-
ral budget condition becomes

" #
Qt
= Et ∑ exp(mt,t+k )st+k . (6.6)
Pt k =0

This equation states that the real value of government debt is equal
to the real present value of government surpluses. In particular, if
Qt /Pt is greater than or smaller than the present value of government
surpluses, an arbitrage opportunity will exist.

6.B.4 Monetary Policy


We assume that each country’s monetary authority controls its local
one-period nominal interest rate it . The monetary authority usually
refers to the central bank. Usually, the nominal interest rate is set
according to a Taylor rule.
From the intertemporal Euler equation,
 
Pt
1 = Et [exp( Mt+1 ) exp(it )] = Et exp(mt+1 ) exp(it )
Pt+1
 
Pt
exp(−it ) = Et exp(mt+1 ) ,
Pt+1
the nominal interest rate it pins down the expected inflation under
the risk-neutral expectation. A higher nominal interest rate generates
a higher expected inflation.
How does the monetary authority implement this interest rate
target? Recall Eq. (6.5), reproduced below,

Qt + Pt gt = Pt τt + Qt+1 exp(−it ).

Electronic copy available at: https://ssrn.com/abstract=4668578


monetary and fiscal policies 141

When the monetary authority raises the nominal interest rate it , if the
fiscal policy (τt , gt ) and the current price level (Pt ) remain the same,
then, the government has to issue a higher nominal amount of debt
Qt+1 . This higher nominal amount of debt then raises the price level
Pt+1 in the next period via


" #
Q t +1
= Et+1 ∑ exp(mt+1,t+1+k )st+1+k .
Pt+1 k =0

def def
Equivalently, let qt = Qt /Pt and recall πt = ∆ log Pt . Then, we can
express the government budget condition Eq. (6.5) in real terms as

qt + gt = τt + qt+1 exp(πt+1 − it ).

If the government commits to the same τt , gt and real quantity of


debt qt+1 , then, a higher it triggers a higher inflation πt+1 in the next
period.
When prices are sticky, the interest rate policy generates real ef-
fects, as we discuss in Section 6.D.

6.B.5 Market Clearing Conditions


The market clearing condition for the home consumption good is

Pt P∗
yt = α (ct + gt ) + (1 − α) ∗t (c∗t + gt∗ ). (6.7)
PH,t PH,t

The market clearing condition for the home government bond is


Z 1 Z 1
Ωt (σt+1 , j)dj + Ωt (σt+1 , j∗ )dj∗ = Qt+1 , ∀σt+1 . (6.8)
0 0

Note that Qt+1 denotes the quantity of the home government debt
that is due at time t + 1. The market clearing condition (6.8) requires
that the total amount of home currency that both countries’ house-
holds receive from their holdings of Arrow-Debreu securities in state
σt+1 is equal to the amount of nominal debt the home government
pays back at time t + 1.

6.B.6 Macro Synthesis


The exogenous variables include the productivity levels, the interest
rates, and the government surpluses:

(zt , z∗t , it , it∗ , st , s∗t )∞


t =0 ,

which are assumed to be stochastic and i.i.d. For simplicity, we use


log z̄ = 0 to denote the mean of the log productivity log zt and log z∗t .

Electronic copy available at: https://ssrn.com/abstract=4668578


142 jiang – lecture notes on international finance

There are 17 endogenous variables in each period t:

(c H,t , c F,t , ct , PH,t , PF,t , Pt , Wt , ℓt , c∗H,t , c∗F,t , c∗t , PH,t


∗ ∗
, PF,t , Pt∗ , Wt∗ , ℓ∗t , Et )∞
t =0 ,

and we have the following 17 equations, including 5 for the home


country,

κPt ct ℓνt = Wt ,
1 α 1− α
Pt = PH,t PF,t ,
α α (1 − α )1− α
PH,t c H,t = αPt ct ,
PF,t c F,t = (1 − α) Pt ct ,
Pt P∗
zt ℓt = α (ct + gt ) + (1 − α) ∗t (c∗t + gt∗ ),
PH,t PH,t

5 for the foreign country,

κPt∗ c∗t (ℓ∗t )ν = Wt∗ ,


1
Pt∗ = ∗ α ∗ 1− α
( PF,t ) ( PH,t ) ,
α α (1 − α )1− α
∗ ∗
PH,t c H,t = (1 − α) Pt∗ c∗t ,
∗ ∗
PF,t c F,t = αPt∗ c∗t ,
Pt P∗
z∗t ℓ∗t = (1 − α) (ct + gt ) + α ∗t (c∗t + gt∗ ).
PF,t PF,t

2 equations relating the price levels to the government surpluses



" #
1 1 k ct
= Et ∑ δ s , (6.9)
Pt Qt k =0
ct+k t+k

" #

1 1 k ct
= ∗ Et ∑ δ ∗ s t + k , ∗
(6.10)
Pt∗ Qt k =0
ct+k

and 1 equation for the nominal exchange rate,

Pt ct = exp(−Et ) Pt∗ c∗t .

There are 4 additional equations governing the law of motion for


prices, which depend on the firms’ price-setting strategies. They are
Eq. (6.11) and Eq. (6.12) for the case of flexible prices that we will
consider in Section 6.C, and Eq. (6.13) and Eq. (6.14) for the case of
sticky prices that we will consider in Section 6.D.

6.C Characterizations under Flexible Prices

6.C.1 Optimal Price-Setting


We consider the economy under flexible prices as a useful bench-
mark. In this case, the firms maximize their dividends period by

Electronic copy available at: https://ssrn.com/abstract=4668578


monetary and fiscal policies 143

period. Take a home firm h as an example,

max Dt ( h ) ,
pt (h),p∗t (h)

where the firm dividend Dt (h) is given by Eq. (6.4). The solution is
ρ
pt (h) = exp(−Et ) p∗t (h) = MCt ,
ρ−1

which implies that the firm sets the same price in the home and
foreign markets, and the price is equal to a constant ρ/(ρ − 1) times
the marginal cost MCt . If we define the mark-up as the ratio between
the price and the marginal cost, the mark-up is a constant in this
case. This is a standard result in the oligopoly problem whenever the
substitution pattern is modeled by the CES aggregator.
Then, the model is closed by the following equations for prices:

∗ ρ
PH,t = exp(−Et ) PH,t = MCt , (6.11)
ρ−1
∗ ρ
PF,t = exp(Et ) PF,t = MCt∗ . (6.12)
ρ−1

Moreover, at the aggregate level, the price level satisfies


ρ
Pt = MCt ,
ρ−1

which implies that the aggregate price level is always set at the mark-
up multiplier times the nominal marginal cost.

6.C.2 Equilibrium Consumption and Exchange Rate when ν = 0


We first consider a simplification of the model by setting ν = 0. In
this case, the disutility from work is linear in the labor provided:
∞ ∞
def
E0 ∑ δt ut ( j) = E0 ∑ δt (log ct ( j) − κ ℓt ( j)) .
t =0 t =0

As we will see below, this simplifying assumption implies that labor


def
adjusts fully in response to government spending shocks. Let ℓ̄ =
(ρ − 1)/(ρκ ) denote the natural rate of employment, which prevails
in an economy without nominal rigidities and government spending.
The following result characterizes the equilibrium allocation and
exchange rate in the model.

Proposition 6.2. If ν = 0, the equilibrium labor is

g∗ g∗
   
gt gt
ℓt = ℓ̄ 1 + α + (1 − α) ∗t , ℓ∗t = ℓ̄ 1 + α ∗t + (1 − α) ,
ct ct ct ct

Electronic copy available at: https://ssrn.com/abstract=4668578


144 jiang – lecture notes on international finance

the equilibrium consumption is

log ct − log c̄ = α log zt + (1 − α) log z∗t , log c∗t − log c̄ = α log z∗t + (1 − α) log zt ,

the equilibrium real exchange rate is

et = − log ct + log c∗t = −(2α − 1)(log zt − log z∗t ),

the equilibrium price level is


Qt Q∗t
Pt = , Pt∗ = i,
Et [ ∑ ∞
h
k =0 exp (mt,t+k )st+k ] Et ∑ ∞ ∗ ∗
k =0 exp( mt,t+k ) st+k

and the (risk-neutral) expected inflation satisfies


 
Pt
exp(−it ) = Et exp(mt+1 ) .
Pt+1
The proof is presented in Appendix A.24. This proposition shows
that, if ν = 0 and prices are flexible, neither monetary nor fiscal poli-
cies affect the equilibrium consumption or the real exchange rate;
the price level adjusts to fully absorb the monetary and fiscal shocks.
To be more precise, consumption is only determined by the stochas-
tic productivity levels zt and z∗t . Due to home bias (i.e., α > 1/2),
the home households’ consumption is more sensitive to the home
productivity level and the foreign households’ consumption is more
sensitive to the foreign productivity level. This result is similar to the
result we obtained in the complete-market endowment economy in
Section 1.C, which implies that consumption is a linear combination
of home and foreign endowments.
In terms of the monetary policy, this proposition shows that a
higher nominal interest rate it leads to a higher expected inflation
πt+1 , but it does not affect the real rate rt , since the real rate only
depends on the consumption dynamics:
 
ct
exp(−rt ) = Et [exp(mt+1 )] = Et δ .
c t +1
In terms of the fiscal policy, this proposition shows that neither tax
τt nor spending gt affect consumption. For given productivity levels,
a higher real government spending gt incentivizes the households
to work more and produce just enough goods to fund the govern-
ment spending, while keeping their consumption level constant. A
higher real tax τt , on the other hand, does not affect the equilibrium
labor and output—that is, whether the government chooses to fund
its spending by raising tax or issuing new debt is irrelevant to real
allocations. Both tax and spending do affect the price level, via the
intertemporal government budget condition Eq. (6.6):
Qt
Pt = .
Et [ ∑ ∞
k =0 exp (mt,t+k )st+k ]

Electronic copy available at: https://ssrn.com/abstract=4668578


monetary and fiscal policies 145

In particular, a higher tax or a lower spending leads to a higher gov-


ernment surplus, which lowers the price level.
Finally, the real exchange rate is determined by the equilibrium
consumption, which only depends on the productivity level. There-
fore, this model with flexible prices gives rise to a clear distinction
between the real variables (i.e., consumption, real exchange rate) that
only depends on productivity shocks and the nominal variable (i.e.,
price level, nominal exchange rate) that also depend on the monetary
and policy shocks. In the next section, we will see how this distinc-
tion dissolves when prices are sticky.

6.C.3 Equilibrium Consumption and Exchange Rate in the General Case


Now, we consider the general case with ν > 0. The following result
characterizes the equilibrium allocation, exchange rate, and price
level in the model.

Proposition 6.3. For a general ν, the equilibrium labor is


1 1
g∗ g∗
   
1 gt 1+ ν 1 gt 1+ ν
ℓt = ℓ̄ 1+ ν 1 + α + (1 − α) ∗t , ℓ∗t = ℓ̄ 1+ ν 1 + α ∗t + (1 − α) ,
ct ct ct ct

the equilibrium consumption is

gt∗
 
ν 2 gt
log ct − log c̄ = α log zt + (1 − α) log z∗t
− 2
( α + (1 − α ) ) + 2(1 − α ) α ∗ ,
1+ν ct ct
 ∗ 
∗ ∗ ν 2 2 gt gt
log ct − log c̄ = α log zt + (1 − α) log zt − ( α + (1 − α ) ) ∗ + 2(1 − α ) α ,
1+ν ct ct

the equilibrium foreign real exchange rate is

g∗
 
ν gt
et = − log ct + log c∗t = −(2α − 1)(log zt − log z∗t ) + (2α − 1)2 − ∗t ,
1+ν ct ct

the equilibrium price level is


Qt Q∗t
Pt = , Pt∗ =
Et [ ∑ ∞
h i
k =0 exp (mt,t+k )st+k ] Et ∑ ∞ ∗ ∗
k =0 exp( mt,t+k ) st+k

and the (risk-neutral) expected inflation satisfies


 
Pt
exp(−it ) = Et exp(mt+1 ) .
Pt+1
The proof is presented in Appendix A.24. This proposition shows
that the real consumption and the real exchange rate depend on
not only the productivity shocks zt and z∗t , but also government
spending-to-consumption ratios gt /ct and gt∗ /c∗t . As such, while the
monetary shocks still have no effects on real variables, fiscal spending
shocks do.

Electronic copy available at: https://ssrn.com/abstract=4668578


146 jiang – lecture notes on international finance

To understand this result, which is representative of textbook


macro models, note that a positive labor curvature coefficient ν im-
plies that a higher labor effort drives up the real wage. Under this
assumption, a higher government spending requires more labor and
makes it more costly. As a result, government spending crowds out
the households’ private consumption and raises their marginal utility
over consumption, which, by the international risk-sharing condition,
also appreciates the local currency in real terms.
In this way, the flexible-price model predicts that a higher gov-
ernment spending leads to real depreciation of the local currency.
Conversely, holding tax constant, a higher government surplus is
associated with a lower government spending and hence real ap-
preciation of the local currency. This prediction is inconsistent with
the data, as countries that have worse fiscal conditions tend to have
weaker currencies.

6.D Characterizations under Sticky Prices

6.D.1 Optimal Price-Setting


If prices are sticky, some or all firms cannot adjust their prices after
the shocks are realized in each period. In this section, we consider a
simple form of nominal rigidities: firms have to set prices one period
in advance. Under this assumption, the firms’ profit maximization
problem only concerns one period. Take the home firm h as an exam-
ple,

max Et−1 [exp( Mt−1,t ) Dt (h)].


pt (h),p∗t (h)

Based on information available at time t − 1, the price at home


pt (h) is set to maximize the expected profit from the home market:

pt ( h ) −ρ
"   #
max Et−1 exp( Mt−1,t )( pt (h) − MCt ) c H,t .
pt (h) PH,t

The future profit is discounted by the domestic households’ SDF


exp( Mt−1,t ), as they are the shareholders whose interests the firms
maximize.
The first-order condition is
"  −ρ #
Pt−1 ct−1 −ρ − ρ −1 1 Pt
0 = Et −1 δ ((1 − ρ) pt (h) + ρMCt pt (h) ) α ct ,
Pt ct PH,t PH,t

which, under symmetry pt (h) = PH,t , implies


ρ
PH,t = E [ MCt ] .
ρ − 1 t −1

Electronic copy available at: https://ssrn.com/abstract=4668578


monetary and fiscal policies 147

If we compare this pricing strategy with Eq. (6.11) obtained under


flexible prices, we can see that the optimal price is now set to match
the expected marginal cost. This is because the firms cannot adjust
their prices after the shocks are realized, and the best they can do is
to be correct “on average.”
The home firms also need to set their sale price in the foreign mar-
ket. Our derivation below focuses on the case of Producer Currency
Pricing, under which case exports are priced and invoiced in the
domestic (producer’s) currency. Then, the price of the home firms’
production sold in the foreign country is set according to
 !−ρ 

pt ( h)
max Et−1 exp( Mt−1,t )(exp(−Et ) p∗t (h) − MCt ) ∗ c∗H,t  ,
exp(−Et ) p∗t (h) PH,t

which implies
∗ ρ
exp(−Et ) PH,t = E [ MCt ] .
ρ − 1 t −1
So, the entire price block can be described as
∗ ρ
PH,t = exp(−Et ) PH,t = E [ MCt ] , (6.13)
ρ − 1 t −1
∗ ρ
PF,t = exp(Et ) PF,t = E [ MCt∗ ] . (6.14)
ρ − 1 t −1
In other words, under Producer Currency Pricing, the local-currency
price of the local consumption bundle (e.g., PH,t ) is fixed and the
local-currency price of the foreign consumption bundle (e.g., PF,t )
comoves with the exchange rate.
Alternatively, some models assume Local Currency Pricing, under
which case exports are priced and invoiced in the foreign (importers’)
currency. Then, the price of the home firms’ production sold in the
foreign country is set according to
 !−ρ 
p ∗ (h)
max Et−1 exp( Mt−1,t )(exp(−Et ) p∗t (h) − MCt ) t
∗ c∗H,t  .
p∗t (h) PH,t

The first order condition is


" −ρ #
Pt∗ ∗

Pt−1 ct−1 −ρ − ρ −1 1
0 = Et −1 δ ((1 − ρ) exp(−Et ) pt (h) + ρMCt pt (h) ) α ∗ ct .
Pt ct PH,t PH,t
∗ c∗ . This implies
Note that Pt−1 ct−1 = exp(−Et ) PH,t t

∗ ρ
PH,t = E [ MCt exp(Et )] .
ρ − 1 t −1
Compared to the case of Producer Currency Pricing, in this case the
exchange rate Et goes into the expectation operator, which is consis-
tent with the assumption that the price is fixed at the consumers’ (i.e.

Electronic copy available at: https://ssrn.com/abstract=4668578


148 jiang – lecture notes on international finance

the foreign households’) currency unit. As a result, both the local-


currency price of the local consumption bundle (e.g., PH,t ) and the
local-currency price of the foreign consumption bundle (e.g., PF,t ) are
fixed. The solution in this case is discussed in Jiang [2022].

6.D.2 Equilibrium Consumption and Aggregate Demand


The sticky prices imply slow adjustments in the price level. As a re-
sult, movements in the nominal exchange rate lead to movements in
the real exchange rate. This idea has been central in many interna-
tional macro models since the seminal work by Mundell-Fleming in
the 1960s. In the context of this model, we formalize this idea by first
relating the real consumption and the real exchange rate movement
to the nominal shocks.

Proposition 6.4. The equilibrium consumption can be solved by

( Pt ct )α ( Pt∗ c∗t )1−α


ct = ℓ̄αα (1 − α)1−α ,
(Et−1 [ Pt ct ℓνt /zt ])α (Et−1 [ Pt∗ c∗t (ℓ∗t )ν /z∗t ])1−α
( Pt ct )1−α ( Pt∗ c∗t )α
c∗t = ℓ̄αα (1 − α)1−α ,
(Et−1 [ Pt ct ℓνt /zt ])1−α (Et−1 [ Pt∗ c∗t (ℓ∗t )ν /z∗t ])α

which implies

log ct = κtc−1 + α log( Pt ct ) + (1 − α) log( Pt∗ c∗t ),



log c∗t = κtc−1 + α log( Pt∗ c∗t ) + (1 − α) log( Pt ct ),
−et = κte−1 + (2α − 1) log( Pt ct ) − (2α − 1) log( Pt∗ c∗t ),

where the terms κtc−1 , κtc−1 , κte−1 are known in period t − 1.

The proof is presented in Appendix A.25. A useful way to under-


stand the equilibrium outcome under sticky prices is to regard the
equilibrium nominal consumption Pt ct and Pt∗ c∗t as the aggregate de-
mand of the home and foreign households, which directly respond to
monetary and fiscal policies. In fact, this proposition shows that they
are sufficient statistics for determining the real consumption and the
real exchange rate.
More concretely, an expansionary monetary or fiscal policy in
the home country stimulates the aggregate demand by raising the
nominal household expenditure Pt ct . As the price level is slow to
adjust, this nominal shock also raises the real household expenditure
ct . Due to the home bias in consumption (i.e., α > 1/2), even though
complete markets facilitate full risk-sharing, this proposition shows
that the home consumption ct is more exposed to the home aggregate
demand Pt ct , and the foreign consumption c∗t is more exposed to the
foreign aggregate demand Pt∗ c∗t . The real exchange rate, determined

Electronic copy available at: https://ssrn.com/abstract=4668578


monetary and fiscal policies 149

by the relative marginal utility of consumption, is also affected by the


nominal household expenditures.
Next, we consider monetary and fiscal policies separately and
trace out exactly how they impact the aggregate demand as summa-
rized by the nominal household expenditures Pt ct and Pt∗ c∗t .

6.D.3 Real Effects of Monetary Policy


Let us first consider the home country’s monetary policy. Unlike the
case of flexible prices, in which the nominal interest rate policy has
no real effects, sticky prices transmit nominal interest rate shocks
to real interest rates and affect the households’ consumption and
investment decisions. Given Eq. (6.2), reproduced below,

it = rt + Et [πt+1 ] + irpt ,

the nominal interest rate it is equal to the real interest rate plus the
expected inflation plus the inflation risk premium. When the prices
are sticky, expected inflation and inflation risk premium do not re-
spond fully to an increase in the nominal interest rate. As a result,
the real interest rate rt responds as well.
Specifically in our setting, the Euler equation

1 = Et [exp( Mt+1 + it )]

implies the following intertemporal relationship between the nominal


interest rate and the current and future aggregate demand:
 
1 1
= exp(it )Et δ . (6.15)
Pt ct Pt+1 ct+1

We consider the case


h in which
i the expected inverse future ag-
gregate demand Et δ P 1c is constant, which we will show is
t +1 t +1
without loss of generality in this model. Then, Eq. (6.15) implies that
the nominal interest rate it directly affects today’s aggregate demand
Pt ct . If the monetary authority in the home country raises the nomi-
nal interest rate by 1%, today’s aggregate demand Pt ct will shrink by
1%. Similarly, a higher nominal interest rate in the foreign country
lowers the foreign aggregate demand. Plugging in this relationship to
Proposition 6.4, we obtain the following result [Corsetti and Pesenti,
2007]:

Proposition 6.5. When the monetary authorities set the nominal interest
rates in home and foreign countries, the equilibrium consumption is

log ct = κ̃tc−1 − αit − (1 − α)it∗ ,



log c∗t = κ̃tc−1 − αit∗ − (1 − α)it ,

Electronic copy available at: https://ssrn.com/abstract=4668578


150 jiang – lecture notes on international finance

the equilibrium real interest rate is

rt = r̄ + αit + (1 − α)it∗ ,

the equilibrium real exchange rate is



et = −κ̃tc−1 + κ̃tc−1 + (2α − 1)(it − it∗ ),

and the equilibrium price levels are


 
1
c
log Pt = −κ̃t−1 − log Et δ + (1 − α)(it∗ − it ).
Pt+1 ct+1
The proof is presented in Appendix A.26. This proposition has
an intuitive interpretation. When the home country’s monetary au-
thority raises the nominal interest rate, it depresses the aggregate
demand and lowers the equilibrium consumption in the home coun-
try. Due to international risk-sharing through complete markets, the
consumption loads on both home and foreign nominal interest rates.
That said, the effect of the home nominal interest rate on home con-
sumption is still stronger due to the home bias in the goods market
(i.e., α > 1/2).
Moreover, since prices are sticky, a higher nominal interest rate
also leads to a higher real interest rate. By the following Euler equa-
tion,

rt = − log δ − log Et [ct /ct+1 ],

the real interest rate is inversely related to the expected consumption


growth. So, the increase in the real interest rate is also consistent
with an increase in the expected consumption growth, which is made
possible by a decline in today’s consumption.
Finally, as the home consumption declines when the home coun-
try’s monetary authority raises the nominal interest rate, the home
households’ marginal utility increases and appreciates the home cur-
rency in both nominal and real terms. In this way, our model repro-
duces a positive relationship between the local nominal interest rate
and the real strength of the currency, which goes back a long way
to some of the seminal papers in the international macro literature
[Mundell, 1963, Fleming, 1962, Dornbusch, 1976, Frankel, 1979].
Some readers might wonder where the productivity shocks zt and

zt go in the equilibrium consumption and the real exchange rate. In
this specific model, monetary and fiscal policies are strong enough
to fully counter the effects of productivity shocks. Specifically, there
exist nominal interest rates

it = − log δ − (log zt − Et−1 [log zt ]),


it∗ = − log δ − (log z∗t − Et−1 [log z∗t ]),

Electronic copy available at: https://ssrn.com/abstract=4668578


monetary and fiscal policies 151

such that our sticky-price model replicates the equilibrium consump-


tion and real exchange rate under the flexible-price case. As such,
we can think of these nominal interest rates as the natural nominal
rates that would prevail in the absence of monetary and fiscal poli-
cies. If the monetary authority sets the nominal interest rate above
the natural rate, then, they depress the aggregate demand and lower
the equilibrium consumption below the flexible-price level, and vice
versa.
A similar argument can be applied to the fiscal policy, which we
study in the next subsection. In more general models, monetary and
fiscal policies might be less effective in fully counteracting the effects
of productivity shocks, and the equilibrium consumption and the real
exchange rate would depend on the productivity shocks as well.

6.D.4 Real Effects of Fiscal Policy


Next, let us shift our focus on the home country’s fiscal policy. Given
the definition of the real SDF, i.e., exp(mt,t+k ) = δk ct /ct+k , it is useful
to express the present value of government surpluses as the current
surplus plus the sum of discounted expected future surpluses.

∞ ∞
" # " #
s
Et ∑ exp(mt,t+k )st+k = s t + c t · Et ∑ δk ctt++kk .
k =0 k =1

Note that the k-period real rate is equal to

rt (k) = − log Et [δct /ct+k ] = − log(ct Et [δ/ct+k ]).

All else equal, a higher current consumption ct implies a lower real


discount rate and increases the present value of future government
surpluses. As a result, the present value of current and future sur-
pluses Et [∑∞k =0 exp( mt,t+k ) st+k ] is increasing in the current consump-
tion ct .
For tractability, we consider the simple case h in whichi future con-
sumption and surpluses are i.i.d. Then, Et ∑∞ k t+k s
k =1 δ ct+k is equal to a
constant that we denote by A, which allows us to simplify Eq. (6.6)
and express the home country’s nominal aggregate demand as

Qt ct
Pt ct = . (6.16)
st + ct A
∗ i
k st+k
h
Likewise, we define A∗ = Et ∑∞
def
k =1 δ c∗t+k , and express the for-
eign country’s nominal aggregate demand as

Q∗t c∗t
Pt∗ c∗t = . (6.17)
s∗t + c∗t A∗

Electronic copy available at: https://ssrn.com/abstract=4668578


152 jiang – lecture notes on international finance

Plug the expressions (6.16) and (6.17) for the aggregate demand
into Proposition 6.4, we obtain

c∗
   
ct
log ct = κtc−1 + α log + (1 − α) log ∗ t ∗ ∗ ,
st + ct A s +c A
   t ∗t 
∗ ct c
log c∗t = κtc−1 + (1 − α) log + α log ∗ t ∗ ∗ .
st + ct A st + ct A

These equations allow us to solve the equilibrium consumption


ct and c∗t as functions of the government surpluses st and s∗t . For
expositional convenience, in the proposition below we consider a
symmetric distribution for home and foreign variables. Let s̄ denote
both countries’ mean government surplus in this distribution and let
c̄ denote both countries’ mean equilibrium consumption. Symmetry
also implies A = A∗ .
If we take a first-order approximation around st = s∗t = s̄ and
ct = c∗t = c̄, then, we obtain a linear equation system with a simple
solution [Jiang, 2022]:

Proposition 6.6. Under a first-order approximation, the equilibrium con-


sumption is

((1 − α)s̄/A + αc̄) (1 − α)(c̄ + s̄/A) ∗


log ct = κc − st − s ,
c̄( Ac̄ + (1 − α)2s̄) c̄( Ac̄ + (1 − α)2s̄) t
(1 − α)(c̄ + s̄/A) ((1 − α)s̄/A + αc̄) ∗
log c∗t = κc∗ − st − s ,
c̄( Ac̄ + (1 − α)2s̄) c̄( Ac̄ + (1 − α)2s̄) t

the equilibrium real interest rate is

((1 − α)s̄/A + αc̄) (1 − α)(c̄ + s̄/A) ∗


rt = − log δ − log Et [1/ct+1 ] − κc + st + s ,
c̄( Ac̄ + (1 − α)2s̄) c̄( Ac̄ + (1 − α)2s̄) t

the equilibrium real exchange rate is

(2α − 1)c̄
et = −κc + κc∗ + (st − s∗t ),
Ac̄ + (1 − α)2s̄

and the equilibrium price levels are


1−α
log Pt = κ P + log Qt − (st − s∗t ).
Ac̄ + (1 − α)s̄

The proof is presented in Appendix A.27. This proposition shows


that the government surplus st also has real effects: by Eq. (6.6),
reproduced below,
∞ ∞
" # " #
Qt
= Et ∑ exp(mt,t+k )st+k = st + Et ∑ exp(mt,t+k )st+k , (6.18)
Pt k =0 k =1

when the current surplus st increases, the present value of govern-


ment surpluses on the right-hand side of this equation will increase.

Electronic copy available at: https://ssrn.com/abstract=4668578


monetary and fiscal policies 153

However, on the left-hand side, the debt quantity Qt is determined


in the previous period, and the price index Pt is sticky. Therefore, to
equilibrate this intertemporal government budget constraint, the real
interest rate captured by mt,t+k on the right-hand side has to rise to
lower the present value of future surpluses and offset the increase in
the current surplus.
In this way, a higher U.S. surplus increases the U.S. real interest
rate and, by Proposition 1.2, appreciates the dollar’s real exchange
rate. Similar to the case with monetary shocks, to engineer the in-
crease in the U.S. real interest rate, the U.S. consumption has to de-
cline in this period and create an expected increase in consumption
growth. As a result, a fiscal shock that increases the government sur-
plus st in the home country leads to a higher home real interest rate
and a stronger home currency in real terms.
We can also think about the exchange rate response to the fiscal
shock from the foreign households’ perspective. Using the foreign
numéraire, we can rewrite Eq. (6.18) as


" #
Qt
exp(et )
Pt
= exp(et )st + Et ∑ exp(m∗t,t+k ) exp(et+k )st+k .
k =1

Rearranging terms, we obtain


  " #
Qt
exp(et )
Pt
− st = Et ∑ exp(m∗t,t+k ) exp(et+k )st+k .
k =1

For this thought experiment, let us assume that the home country
is very small so that its fiscal shock does not affect the foreign SDF
m∗t,t+k , and that the shock is transitory so that the future surpluses
st+k and the future real exchange rate et+k are unaffected. Then, to
equilibrate this equation in response to an increase in the current
surplus st , the current real exchange rate et has to appreciate. In this
sense, the real exchange rate behaves like the asset price for the claim
to government surpluses, which has to adjust when the fiscal cash
flows change. We can also use this logic to show that the U.S. real
exchange rate has to appreciate when the expected future surpluses
Et [st+k ] increase [Jiang, 2022].
Notably, this result is opposite to what we obtained under flexible
prices. Specifically, Proposition 6.3 shows that, with a positive cur-
vature parameter ν for labor, a higher government spending in the
home country gt , which corresponds to a lower government surplus
st , crowds out the local households’ private consumption, increases
their marginal utility, and appreciates the home currency in real
terms. Sticky prices overturn this result and generate home currency
depreciation through the valuation channel.

Electronic copy available at: https://ssrn.com/abstract=4668578


154 jiang – lecture notes on international finance

This result is also different from the Mundell-Fleming view of the


fiscal policy, under which a fiscal expansion in the form of higher
government spending shifts up the IS curve, puts an upward pres-
sure on the local interest rate, attracts capital inflows, and appreciates
the local currency. In comparison, imposing the intertemporal gov-
ernment budget constraint (6.6) as we do in our sticky-price model
generates an opposite valuation effect that depreciates the local cur-
rency when the surplus declines or government spending increases.
Our model makes the extreme assumption that prices are fully
sticky for one period. It is possible to extend the model to allow for
a more realistic form of price stickiness. For example, if some firms’
prices are sticky for multiple periods, monetary and fiscal policies
could have persistent effects on the real economy.
Moreover, our model assumes that the government only issues
one-period debt. If the government issues longer-term debt, as we
will see in Chapter 8, the bond quantity Qt on the left-hand side of
Eq. (6.18) becomes the market value of the outstanding debt. Then,
when the fiscal shocks impact the government surpluses on the right-
hand side of this equation, the bond price can also adjust and absorb
some of the shocks, so that the consumption does not need to adjust
as much. However, adjustments in the long-term bond price have
implications for the discount rates in the future, which also transmits
the fiscal shocks to the real economy in future periods. As such, the
long-term bond could lower the real response in the current period
while magnifying the real responses in future periods. See Cochrane
[2023] for detailed discussions.

6.E Comparing Monetary and Fiscal Policies

6.E.1 Observational Equivalence


Eq. (6.15) can be expressed as
 
st + ct A s t +1 + c t +1 A
= exp(it )Et δ .
Qt ct Q t +1 c t +1
Note that the nominal quantity of government debt Qt+1 is chosen
by the government in period t, so it is known in period t. Then, we
can express the home nominal interest rate as
   
st + ct A s t +1 + c t +1 A
it = log − log Et δ + ∆ log Qt+1 . (6.19)
ct c t +1
Since the consumption
h and igovernment surplus shocks are i.i.d.
s +c A
across periods, Et δ t+1c t+1 is a constant. Then, for any given
t +1
nominal debt growth Qt+1 /Qt , the nominal interest rate it directly
affects the real consumption ct .

Electronic copy available at: https://ssrn.com/abstract=4668578


monetary and fiscal policies 155

Similarly, we can express the foreign nominal interest rate as


" #
s∗t + c∗t A∗ s∗t+1 + c∗t+1 A∗
 
it∗ = log − log Et δ + ∆ log Q∗t+1 .
c∗t c∗t+1

Along with

c∗t
   
ct
log ct = κtc−1
+ α log + (1 − α) log ∗ ,
st + ct A s + c∗ A∗
   t ∗t 
∗ ct c
log c∗t = κtc−1 + (1 − α) log + α log ∗ t ∗ ∗ ,
st + ct A st + ct A

we obtain 4 equations with 6 unknowns: the nominal interest rates it


and it∗ , the government surpluses st and s∗t , and the equilibrium con-
sumption ct and c∗t . So, the monetary policies which set the nominal
interest rates and the fiscal policies which set the government sur-
pluses cannot both be exogenous. If they are, then we have 4 equa-
tions with only 2 endogenous variables, which makes the equation
system overidentified and there might be no equilibrium.
For example, in our discussion of the monetary policy above, the
nominal interest rate policy will determine both the equilibrium
consumption and the government surpluses that the fiscal authority
needs to accommodate. In this case, the monetary authority sets
the price level and the aggregate demand, and the fiscal authority
implements the corresponding taxation and spending to ensure the
government budget constraint is satisfied.
Conversely, in our discussion of the fiscal policy above, the govern-
ment surplus policy will determine both the equilibrium consump-
tion and the nominal interest rates that the monetary authority needs
to accommodate. In this case, the fiscal authority sets the price level
and the aggregate demand, and the monetary authority implements
the corresponding nominal interest rate to ensure the government
budget constraint is satisfied.
Another way to frame this result is to note that the observed
macroeconomic data can be mapped to either a model in which the
monetary policies are active and determines the path of the nomi-
nal interest rates as well as the necessary responses in government
surpluses, or a model in which the fiscal policies are active and deter-
mines the path of the government surpluses as well as the necessary
responses in nominal interest rates. Cochrane [2021] offers a more
detailed discussion of this observational equivalence.
Finally, throughout our discussion above, we have left out one
degree of freedom, which is the growth rate of the government debt
quantity ∆ log Qt+1 . For given nominal interest rate it and real gov-
ernment surplus st , varying the nominal quantity of government debt

Electronic copy available at: https://ssrn.com/abstract=4668578


156 jiang – lecture notes on international finance

has no real effects. Given Eq. (6.19), reproduced below,


   
st + ct A s + c t +1 A
it = log − log Et δ t+1 + ∆ log Qt+1 ,
ct c t +1
and given
∞ ∞
" # " #
πt+1 = log Qt+1 − log Et+1 ∑ exp(mt+1,t+1+k )st+1+k + log Et ∑ exp(mt,t+k )st+k
k =0 k =0

= log Qt+1 − log(st+1 + ct+1 A) + log(st + ct A),


we can see that a higher nominal government debt growth ∆ log Qt+1
leads to a higher nominal interest rate by the same magnitude. It
does not affect the real consumption ct and hence the real rate rt .
Rather, it moves one-to-one with the inflation πt+1 . Since ∆ log Qt+1
is known in period t, a higher nominal debt growth creates expected
inflation.
As such, we can think of the nominal quantity of government debt
Qt as another policy instrument separate from the nominal interest
rate it and the real government surplus st . However, this additional
instrument does not affect the real economy, but only affects the
expected inflation. In comparison, the nominal interest rate it and the
real government surplus st create surprise inflation, which is what is
required to deliver real effects in this model.

6.E.2 Active vs. Passive Monetary and Fiscal Policies


The sticky-price model above generates a very stark result: in each
period, an active monetary policy requires full cooperation from the
fiscal side, and an active fiscal policy requires full cooperation from
the monetary side. In other words, implementing one policy leaves
no room for the other. A natural question is whether this result can
be relaxed when we consider more general forms of price stickiness
and government policies.
Leeper [1991] provides an answer to this question in a fairly gen-
eral class of linearized models, which allows less extreme forms of
price stickiness. He shows that the policy responses can be character-
ized by two regions: in the monetary dominance region, the monetary
shock is exogenous and requires the fiscal policy to accommodate the
shock; in the fiscal dominance region, the fiscal shock is exogenous and
requires the monetary policy to adjust to accommodate the shock. As
we relax the assumption of full price stickiness for one period, there
is some space for the monetary rate to respond to inflation, and some
space for the fiscal surplus to respond to the level of outstanding
debt. However, a strong monetary response to inflation constrains the
extent to which the fiscal authority can respond to the debt level, and
vice versa.

Electronic copy available at: https://ssrn.com/abstract=4668578


monetary and fiscal policies 157

In light of this discussion, the standard New Keynesian framework


can be understood as assuming that the monetary policy is active and
the fiscal authority accommodates the monetary shocks. As Leeper
[2010] puts it, “most macroeconomists were raised on the belief that infla-
tion is determined by monetary policy, especially in the long run. Full stop...
Central bankers need a broader perspective on price level determination—to
at least understand and acknowledge that there is another channel through
which inflation can be determined. The broader perspective is important
because the new Keynesian/old monetarist view implicitly embeds a dirty
little secret: for monetary policy to successfully control inflation, fiscal policy
must behave in a particular, circumscribed manner.”
In particular, the accommodative fiscal stance can be engineered
by the expectation that future government surpluses will adjust to
stabilize the debt level, as opposed to requiring today’s government
surplus to adjust right away in response to monetary shocks. As long
as the investors’ and households’ fiscal expectations are anchored
to this expectation, inflation and, by the same argument, the real
exchange rate can be understood in terms of the monetary shocks
as we have seen in Proposition 6.5. However, when the agents stop
believing that the fiscal authority is committed to the accommodative
stance, then, its fiscal policy can undermine the ability of monetary
policy to control inflation and drive real economic outcomes on its
own.

6.E.3 Discussions

In this chapter, we have made two extreme assumptions: (1) prices


are fully sticky for one period, and (2) monetary and fiscal authori-
ties cannot trade off their policies intertemporally. In exchange, we
are able to derive closed-form solutions to characterize the equilib-
rium responses in consumption and the exchange rate. To consider
these issues quantitatively, we may need to relax these assumptions
and work with numerical models. For example, we may consider a
setting in which prices are only partially sticky, which gives rise to
a more realistic Phillips Curve. In addition, the monetary and fiscal
responses can have richer time-series patterns. Galí [2015] provides a
good reference on the monetary side, and Cochrane [2023] provides a
good reference on the fiscal side.
The fiscal-currency linkage can be applied to studying not only the
exchange rate variations in normal times, but also currency crashes
in crisis times [Burnside, Eichenbaum, and Rebelo, 2006]. Burnside,
Eichenbaum, and Rebelo [2001] show that prospective fiscal deficits
that drive currency crisis can be in the form of implicit bailout guar-
antees to failing banking systems. See Burnside, Eichenbaum, and

Electronic copy available at: https://ssrn.com/abstract=4668578


158 jiang – lecture notes on international finance

Rebelo [2016] for a review of the currency crisis models.


Moreover, the model in this chapter only investigates how mon-
etary and fiscal shocks affect the real exchange rate by influencing
the real interest rate. It is very likely that monetary and fiscal policies
also drive variations in risk premia and convenience yields, which
lead to exchange rate variations through different channels. For ex-
ample, Caramp and Singh [2020] studies how the monetary policy
impacts liquidity premia. Croce, Nguyen, and Schmid [2012b], Croce,
Kung, Nguyen, and Schmid [2012a], Jiang [2021] study how fiscal
policies affect risk premia. Chernov, Schmid, and Schneider [2020],
Liu, Schmid, and Yaron [2020] study how fiscal policies affect default
premia and liquidity premia.
Finally, these theoretical arguments are also relevant for un-
derstanding currency unions such as the Eurozone. In a currency
union, as the nominal interest rate and the nominal exchange rate
are pegged across the member countries, they can only adjust in
response to union-wide monetary or fiscal shocks, but they cannot
simultaneously adjust in response to country-specific shocks. As
a result, the convenience yields and their cross-country variations
thereof have to play an important role as the shock absorber. Jiang,
Lustig, Van Nieuwerburgh, and Xiaolan [2020c] provide a detailed
theoretical and empirical discussion.

Electronic copy available at: https://ssrn.com/abstract=4668578


Part III

Understanding the
Quantities and Flows

Electronic copy available at: https://ssrn.com/abstract=4668578


7
Global Imbalances and the Exorbitant Privilege

Summary

• The architecture of the international monetary system has important implications for exchange
rate and capital flow dynamics, and reserve assets are at its cornerstone.

• We examine two complementary views of the architecture: the insurance provision view em-
phasizes the U.S.’ role as the world’s insurance provider, and the safe asset view emphasizes
the foreign demand for dollar safe assets. Both views emphasize the centrality of the U.S. in
the global financial markets, but they have different implications for the U.S. external imbal-
ances and the dollar exchange rate.

• We will also briefly survey topics related to the stability of the international monetary system.

Starting from this chapter, we will shift our focus from the ex-
change rates to international portfolio positions and capital flows.
For this purpose, studying the Euler equations alone is no longer suf-
ficient, and general equilibrium models become necessary to under-
stand the financial quantities. We will begin with the most important
asymmetry in the international financial system: the U.S. vs. the rest
of the world.
A prototype financial system involves households who save and
consume, firms or entrepreneurs who produce, and financiers who
intermediate the funds between the households and the firms. These
financiers can be the banks or many types of shadow banks. They
provide funds to the firms by investing in their risky projects, and
they provide saving vehicles to the households by taking safe de-
posits. In doing so, these financiers engage in the safety, liquidity,
and maturity transformations.
This summary of financial system also applies at the global level,
with the U.S. playing the central role as the financier who intermedi-
ates capital flows to the rest of the world and earns a premium from
the intermediation process. Other countries, playing the roles of the
households and the firms, invest at and get funding from the inter-

Electronic copy available at: https://ssrn.com/abstract=4668578


global imbalances and the exorbitant privilege 161

mediary. This core-periphery structure is a key feature of the modern


international monetary system.
This core-periphery structure gives rise to two salient patterns in
the data: global imbalances and exorbitant privilege. We have had a
discussion of these patterns in Chapter 2. In this chapter, we consider
two complementary ways to model these patterns.

7.A The Insurance Provision View

First, we consider the insurance provision view as in Gourinchas and


Rey [2007b, 2022]. We adapt their model to follow the baseline set-
up in Section 1.A more closely. There are two countries, home and
foreign. Each country produces a unique type of good, which is
endowed to the domestic households. The home households’ utility
function is

1
u(ct ) = ( c t )1− γ ,
1−γ

where the consumption ct is again a Cobb-Douglas aggregation of


the home and foreign goods: ct = (c H,t )α (c F,t )1−α .
We deviate from the symmetric baseline case by assuming that the
foreign households are more risk-averse. Their utility function is

1 ∗
u∗ (c∗t ) = ( c ∗ )1− γ ,
1 − γ∗ t

with γ∗ > γ. As in Section 4, we interpret the home country as the


U.S., and the foreign country as the rest of the world. This difference
in risk aversion can be motivated by difference in financial develop-
ments. For example, the more developed financial sector in the U.S.
may face weaker liquidity or financing constraints and thus have
higher risk-taking capacity [Maggiori, 2017]. Similarly, Chien and
Naknoi [2015] develop a model with heterogeneous agents in which
the U.S. has a larger mass of sophisticated traders than the foreign
countries.
The macro synthesis of this model is nearly identical to the base-
line model in Section 1.A with complete markets. The only difference
is that the home and foreign households have different risk aversion.

7.A.1 Social Planner’s Solution

We assume the markets are complete, which allows us to solve the


equilibrium using the social planner approach. More precisely, we
solve the equilibrium outcome under a social planner, who maxi-

Electronic copy available at: https://ssrn.com/abstract=4668578


162 jiang – lecture notes on international finance

mizes a weighted sum of the households’ welfare:


" #
E0 ∑ δt (πu(ct ) + (1 − π )u(c∗t )) .
t =0

The social planner tells the home and foreign households how much
to consume, subject to the resource constraints

yt = c H,t + c∗H,t ,
y∗t = c F,t + c∗F,t .

The social planner’s Lagrangian is

∞ ∞ ∞
" !#
1 1 ∗ 1− γ ∗
E0 ∑ δ π t
(ct ) 1− γ
+ (1 − π ) (c ) + ∑ ζ H,t (yt − c H,t − c H,t ) + ∑ ζ F,t (yt − c F,t − c F,t )
∗ ∗ ∗
,
t =1
1−γ 1 − γ∗ t t =1 t =1

which implies the following first-order conditions


  1− α
c F,t
w.r.t. c H,t : δt π (ct )−γ α = ζ H,t ,
c H,t

∗ c∗F,t
w.r.t. c∗H,t : δ t
(1 − π )(c∗t )−γ (1 − α) = ζ H,t ,
c∗H,t
 α
c H,t
w.r.t. c F,t : δ t π ( c t ) − γ (1 − α ) = ζ F,t ,
c F,t
!1− α
−γ∗
c∗H,t
w.r.t. c∗F,t : δt (1 − π )(c∗t ) α = ζ F,t .
c∗F,t

Let tott = et /(2α − 1) denote the log terms of trade as defined


in Section 1.A. Then, we can follow the derivation in Section 1.3 and
obtain the following equation system that pins down the equilibrium
allocations and exchange rate.

Proposition 7.1. The equilibrium is pinned down by the following simulta-


neous equations
 α −1 −α
1−α
 
α
exp(tott ) ct + exp(tott ) c∗t = yt ,
α 1−α
−α  α −1
1−α
 
α
exp(−tott ) ct + exp(−tott ) c∗t = y∗t ,
1−α α
 1
π (ct )−γ 2α−1

pt
∗ = ∗ = exp(tott ).
1 − π (c∗t )−γ pt exp(−et )

The proof is presented in Appendix A.28. To illustrate the equi-


librium dynamics, let us consider a simple example. Suppose γ = 2
and γ∗ = 8, α = 0.7, and π = 0.5. The endowment shocks yt and y∗t
are always equal. Figure 7.1 traces out the equilibrium consumption

Electronic copy available at: https://ssrn.com/abstract=4668578


global imbalances and the exorbitant privilege 163

allocations ct and c∗t for home and foreign households as we vary the
endowments. We can see that, as the endowments become higher
in both countries, the home country’s consumption increases faster
while the foreign country’s consumption increases slower. In other
words, the home country takes away a greater share of the endow-
ments in high-endowment states, while the foreign country takes
away a greater share of the endowments in low-endowment states.
In this way, the home country insures the foreign country in the bad
states in exchange for higher pay-off in good states.

Figure 7.1: Equilibrium Consumption


Allocation under the Insurance View.
3

2.5

1.5

0.5

0
0 0.5 1 1.5 2 2.5 3 3.5 4

We can frame the same pattern in terms of international transfer,


which we plot in Figure 7.2. The home country’s net transfer is de-
fined as the value of its endowment minus the value of its consump-
tion, normalized by the value of its endowment. In this model, the
only reason why the home households’ endowment and consump-
tion are different is because the home households transfer resources
to the foreign households according to the risk-sharing agreement.
The net home transfer to the foreign country is positive when the
endowments are low, and negative when the endowments are high.
In other words, the home country earns profits in good times as a
compensation for its insurance provision, and bears the losses in bad
times as the insurance pays off. The foreign country’s net transfer
has the opposite pattern, as it is the recipient of the international
insurance.
Unconditionally, because the agents are willing to pay more for
pay-offs in bad states, the U.S. receives a higher insurance premium
in good times than it pays off in bad times, making a net profit from
its insurance provision. This profit funds a persistent trade deficit on

Electronic copy available at: https://ssrn.com/abstract=4668578


164 jiang – lecture notes on international finance

average.

Figure 7.2: Equilibrium International


Transfer under the Insurance View.
1

0.5

-0.5
0 0.5 1 1.5 2 2.5 3 3.5 4

Figure 7.3 plots the equilibrium real exchange rate et , which mea-
sures the strength of the home currency (i.e., the dollar). The dollar
is stronger in high-endowment states and weaker in low-endowment
states. This is because, due to home bias in consumption, when the
home households receive a greater share of the endowments, they
prefer to consume more home goods. Their demand bids up the
price of the home goods, and generates real dollar appreciation. Con-
versely, when the aggregate endowment is low, the foreign house-
holds receive a wealth transfer from the U.S. and consume more
domestic goods, leading to real dollar depreciation.

Figure 7.3: Equilibrium Exchange Rate


under the Insurance View.
0.2

0.1

-0.1

-0.2

-0.3

-0.4
0 0.5 1 1.5 2 2.5 3 3.5 4

Electronic copy available at: https://ssrn.com/abstract=4668578


global imbalances and the exorbitant privilege 165

7.A.2 Decentralized Competitive Equilibrium


The social planner approach allows us to derive the equilibrium
allocations and prices in complete markets. We are also interested
in how the households trade to implement this equilibrium. In this
subsection, we convert the problem to a continuous-time setting in
order to derive the portfolio choice in the decentralized competitive
equilibrium. Time is finite with horizon T. The representative home
households maximize
Z T 
1
Et δt (ct )1−γ dt ,
0 1−γ
and the representative foreign households maximize
Z T 
1 ∗ 1− γ ∗
Et δt ( c ) dt .
0 1 − γ∗ t
For simplicity, we assume that the home and foreign endowments
are identical, i.e., yt = y∗t , reflecting a common global cycle. We
assume

d log yt = d log y∗t = κ (log ȳ − log yt )dt + σdWt ,

which implies

σ2
Z t  
−κt −κt −κt −κt −κt −2κt
log yt = e log y0 + (1 − e ) log ȳ + σe e dWs ∼ N
κs
e log y0 + (1 − e ) log ȳ, (1 − e ) .
0 2κ
The social planner’s problem in continuous time is
Z T   
t 1 1− γ 1 ∗ 1− γ ∗
E0 δ π (ct ) + (1 − π ) (c ) dt
0 1−γ 1 − γ∗ t
subject to c H,t + c∗H,t = yt and c F,t + c∗F,t = y∗t . As we discussed in
Proposition 1.3 in Section 1.C, we can solve for the consumption ct , c∗t
and the exchange rate et as functions of k t , which in turn depends on
yt = y∗t .
We define the home and foreign SDFs as

exp(mt ) = δt π (ct )−γ , exp(m∗t ) = δt (1 − π )(c∗t )−γ ,

which allows us to evaluate the wealth claims in local numéraires as


the present values of the consumption streams
exp(m∗k ) ∗
Z T  Z T 
exp(mk ) ∗
a t = Et ck dk , a t = Et ∗ ck dk . (7.1)
t exp( mt ) t exp( mt )

Similarly, we can evaluate the world equity claim in the home numéraire
as the present value of the endowment streams
p∗k
Z T   
exp(mk ) ∗
s t = Et pk yk + y dk .
t exp( mt ) exp(ek ) k

Electronic copy available at: https://ssrn.com/abstract=4668578


166 jiang – lecture notes on international finance

Thanks to our simplifying assumption yt = y∗t , the equilibrium


asset prices at , a∗t , and st are all functions of yt . For example, if at =
f (log yt ), then, we can express the wealth dynamics as

1 ′′
dat = [ f ′ (log yt )κ (log ȳ − log yt ) + f (log yt )σ2 ]dt + f ′ (log yt )σdWt
2
def
= µ a,t dt + σa,t dWt .
(7.2)

Likewise, we can derive the world equity price’s dynamics and


denote it as

dst = µs,t dt + σs,t dWt .

Then, to compute the home and foreign households’ portfolio


choices, we need to take a stance on the set of assets that we use to
replicate the wealth dynamics. Because the markets are complete,
there are infinitely many sets of assets that span the endowment
shock. We choose the home risk-free bond and the world equity
claim to construct the replicating portfolios. Let xt denote the weight
on world equity in home households’ portfolio, and let xt∗ denote the
weight on world equity in foreign households’ portfolio.
If we match the home households’ wealth dynamics (7.2) with the
wealth dynamics implied from consumption and portfolio holdings,
i.e.,
dst + ( pt yt + p∗t exp(−et )y∗t )dt
 
dat = at xt + (1 − xt )rt dt − ct dt,
st

we can use the Martingale representation theorem to obtain the port-


folio choice xt from
σa,t σs,t
= xt .
at st
Similarly, we match the foreign households’ wealth dynamics, i.e.,

d( a∗t exp(−et )) = µ a∗ ,t dt + σa∗ ,t dWt ,

with the wealth dynamics implied from consumption and the repli-
cating portfolio, i.e.,

dst + ( pt yt + p∗t exp(−et )y∗t )dt


 
a∗t exp(−et ) xt∗ + (1 − xt∗ )rt dt − c∗t dt,
st

to obtain the portfolio choice xt∗ .


We implement this algorithm with ȳ = ȳ∗ = 1.84, κ = 0.1,
σ = 0.05, T = 500 and δ = 0.97. We study the equilibrium allocations
and asset prices at time t = 0, as a function of the endowment level
y0 = y0∗ . Conceptually, we can think of the economy as being created

Electronic copy available at: https://ssrn.com/abstract=4668578


global imbalances and the exorbitant privilege 167

at t = −τ < 0, and the endowment shocks from time −τ to 0 push


the endowment level to y0 at time 0.
Figure 7.4 plots the equilibrium wealth share of the U.S. house-
holds, defined as a0 /( a0 + a0∗ exp(−e0 )). Consistent with the U.S.
earning a risk premium in good times and paying off the insurance
in bad times, the U.S. households’ wealth share is higher in high-
endowment states and lower in low-endowment states. In other
words, the U.S. households’ financial holdings are more exposed to
adverse endowment shocks, leading to a countercyclical wealth share.

Figure 7.4: Equilibrium Wealth Share of


U.S. Households.
0.54

0.52

0.5

0.48

0.46

0.44

0.42

0.4
1 1.5 2 2.5 3 3.5 4

Since the U.S. households have lower risk aversion, as their wealth
share declines in low endowment states, the wealth-weighted aver-
age investor’s risk aversion is also countercyclical. As a result, the
risk premium on the world equity is higher and the equity return
becomes more volatile in bad times. Figure 7.5 shows the overall neg-
ative relationship between the endowment level and the equity return
volatility.
Perhaps paradoxically, as the equity becomes riskier, it magnifies
the difference in risk aversion between the U.S. and foreign house-
holds, which leads to further divergence in their portfolio choices.
Figure 7.6 reports the equilibrium portfolio allocation towards the
world equity by the U.S. and foreign households, xt and xt∗ . The U.S.
households’ equity share is always above 1, and the foreign house-
holds’ is always below 1, which is consistent with our intuition that
the U.S. takes a more levered position on risky assets while the for-
eign households seek safety in the U.S. risk-free bond. When the
endowment level is lower, the U.S. households take a more levered
position on the world equity despite suffering greater wealth losses,

Electronic copy available at: https://ssrn.com/abstract=4668578


168 jiang – lecture notes on international finance

Figure 7.5: Volatility of World Equity


Return in Dollar.
0.155

0.15

0.145

0.14

0.135

0.13

0.125

0.12

0.115

0.11
1 1.5 2 2.5 3 3.5 4

while the foreign households further reduce their equity shares. That
said, in the region of very low endowments, the U.S. wealth share can
become so small that they offer only tiny amount of risk-free bond
despite their highly levered positions. In this case, the foreign house-
holds’ equity share increases. In the limit, as the U.S. wealth shrinks
much faster than the foreign wealth, the foreign households have to
hold the entire world equity and no risk-free bond.

Figure 7.6: Equilibrium Equity Alloca-


tion by U.S. and Foreign Households, xt
1.2 and xt∗ .

1.15

1.1

1.05

0.95

0.9

0.85
1 1.5 2 2.5 3 3.5 4

This countercyclical behavior of asset return volatility and risk


premium also provides a motivation for the more reduced-form
specification of the SDFs that we considered in Eq. (3.1) and (3.2) in

Electronic copy available at: https://ssrn.com/abstract=4668578


global imbalances and the exorbitant privilege 169

Section 3.A.

7.A.3 The Cyclicality of Bond Holdings

Finally, we plot the equilibrium U.S. bond holdings by foreign house-


holds. Unconditionally, the foreign households’ holdings are always
positive, which again reflects the leveraged nature of the U.S. port-
folios. Conditionally, while the foreign households’ portfolio share
in the U.S. bond increases when the endowment level declines, their
dollar-denominated wealth declines even faster. The dollar value of
their bond holdings, which are equal to their product, thus decline
in bad times under our model specification. In addition to the dollar
exchange rate’s cyclicality, this procyclical bond holding is another
implication that is inconsistent with the flight-to-safety pattern in
data.

Figure 7.7: Equilibrium U.S. Bond


Holding by Foreign Households.
4.5

3.5

2.5

1.5

0.5
1 1.5 2 2.5 3 3.5 4

This observation opens up a broader discussion of what is so


special about reserve assets. If the reserve assets are about providing
insurance against risks, then, investors want to hold them before
the risks materialize, and in particular when the probabilities of the
negative shocks are high. However, what we observe in the data is
flight to safety after the negative shocks have already happened,
which could have a different motivation than precautionary savings.
One way to generate the countercyclical bond demand under the
frictionless model is to assume that the bad states of the world mean
not only low endowments, but also greater risk quantities or risk
prices, which are consistent with the countercyclical behaviors of the
VIX index. For example, in Gourinchas and Rey [2022], the foreign

Electronic copy available at: https://ssrn.com/abstract=4668578


170 jiang – lecture notes on international finance

investors’ risk aversion goes up in bad times, which increases the risk
price and generates greater demand for safe assets.
An alternative approach is to introduce frictions in the model. In
Section 7.C, we will consider a reduced-form way of capturing these
frictions by introducing a countercyclical convenience yield for the
U.S. bond. In this case, non-pecuniary benefits of safe assets lead
to greater demand for the U.S. bond after the negative shocks have
already happened.

7.B The Reserve Currency Paradox

The insurance provision view provides a powerful tool for under-


standing the asymmetry in portfolio holdings and asset returns
between the U.S. and the rest of the world. It offers a rational, risk-
based explanation for why the U.S. enjoys the exorbitant privilege.
However, this view in its most basic form leads to a counterfactual
prediction. As we saw in Figure 7.2, the U.S. pays out insurance in
bad times by transferring wealth to the foreign country. As such,
being the world’s insurance provider has a cost: in the language of
Gourinchas and Rey [2022], the exorbitant privilege comes with an
exorbitant duty. As the foreign households have relatively higher
wealth during global downturns, and they have a home bias towards
foreign goods, the dollar has to depreciate in real terms as we saw in
Figure 7.3.
In the data, the dollar tends to appreciate during global down-
turns. Maggiori [2017] notes this inconsistency and names it the Re-
serve Currency Paradox. Below, we provide two complementary ways
of thinking about this paradox. In doing so, we set the stage for an
alternative view of the global imbalances that we develop in the next
section.

7.B.1 Demand vs. Supply Shocks


In Maggiori [2017]’s original analysis of the reserve currency para-
dox, he provides a simple resolution by introducing a state-dependent
trade cost. If the cost of shipping goods internationally is higher dur-
ing recessions, the U.S. economy effectively becomes more closed and
the U.S. households’ demand for the U.S. goods increases relative to
their demand for foreign goods, leading to real dollar appreciation.
We can think of this variation in the trade cost as a demand shock,
whereas the baseline setup we considered that gives rise to the re-
serve currency paradox is only about the supply shock. The U.S.
transfer to the rest of the world increases the supply of the U.S.
goods in the foreign countries. Given the foreigners’ downward-

Electronic copy available at: https://ssrn.com/abstract=4668578


global imbalances and the exorbitant privilege 171

sloping demand for the U.S. goods, the U.S. goods have to become
cheaper and generate real dollar depreciation.
Other types of demand shocks could also reverse the dollar ex-
change rate’s cyclicality. Consider, for example, the bond convenience
yields we studied in Chapter 4, which capture demand shocks for
assets instead of goods. In particular, the foreigners’ demand for U.S.
safe assets increases in bad times, which leads to dollar appreciation.
Similarly, by introducing time-varying risk aversion as another type
of demand shock, Gourinchas and Rey [2022] also generates dollar
appreciation in bad times in a slight extension of the baseline model.

7.B.2 Net Foreign Assets vs. Total Wealth


The insurance provision view emphasizes the international transfer
of wealth from the U.S. to the rest of the world that occurs during
global recessions. Figure 7.8 plots the U.S. net foreign assets (NFA)
normalized by the U.S. consumption in our model. The U.S. NFA is
defined as the U.S. holdings of foreign equity minus foreign holdings
of U.S. equity and bond:
s F,t s
n f at = at xt − a∗t exp(−et ) xt∗ H,t − a∗t exp(−et )(1 − xt∗ ),
st st
where s H,t and s F,t denote the value of the U.S. and foreign equity
in the U.S. numéraire, respectively. In bad times, the U.S. external
assets, which are the world equity, depreciate more than the U.S.
external liabilities, which include the risk-free bond. As a result, the
U.S. NFA deteriorates along with the U.S. wealth share.

Figure 7.8: Equilibrium U.S.


NFA/Consumption Ratio.
3

-1

-2

-3
1 1.5 2 2.5 3 3.5 4

In this model, low endowments are associated with low U.S.


wealth share and low U.S. NFA. Conceptually, these two quantities

Electronic copy available at: https://ssrn.com/abstract=4668578


172 jiang – lecture notes on international finance

are related by

at = s H,t + n f at , (7.3)

which states that the U.S. wealth is equal to the value of U.S.-issued
assets adjusted by the U.S. NFA. If the U.S. households hold a lot
of foreign assets, the NFA will be positive and the U.S. wealth will
be higher than the value of the domestic assets. Conversely, if the
foreign households hold a lot of U.S. assets, the NFA will be negative
and the U.S. wealth will be lower than the value of the domestic
assets. In our model, both the value of the domestic wealth s H,t and
the NFA n f at are procyclical, leading to a procyclical U.S. wealth
share that we saw in Figure 7.4. In Section 9.A, we will have a more
detailed discussion of the net foreign assets accounting.
One essential feature of this model that leads to the reserve cur-
rency paradox is the procyclical U.S. wealth share: as the foreign
households are relatively wealthier in bad times, they tend to con-
sume more and tilt their consumption towards the foreign goods,
which requires the U.S. goods to become cheaper. However, does the
U.S. wealth share have to be procyclical? In the next section, we will
examine a complementary view under which the U.S.-issued assets
are better hedged against bad states of the world. As a result, while
the U.S. NFA n f at still declines in bad times, the U.S. domestic assets
s H,t depreciate less and offset the procyclical nature of the U.S. NFA,
leading to a countercyclical U.S. wealth share.

7.C The Safe Asset View

Next, we consider a complementary safe asset view, which empha-


sizes the U.S.’ role as the sole supplier of the world’s safe assets. The
key difference from the insurance provision view is that investors
derive non-pecuniary benefits from holding dollar safe assets such
as the U.S. Treasury bonds, which leads to a convenience yield. This
convenience yield enters the standard Euler equation as a wedge,
which, as we saw in Chapter 4, makes progress in explaining the
exchange rate dynamics. Building on the analysis in Chapter 4, our
hypothesis here is that this ingredient also helps us understand the
global imbalances and international transfers.
Our derivation loosely follows a simplified version of Jiang [2023a].
Earlier works on safe assets in international macro models include
Caballero and Farhi [2018], Farhi and Maggiori [2018], Caballero,
Farhi, and Gourinchas [2021], Gopinath and Stein [2021], Jiang, Krish-
namurthy, and Lustig [2020a].
We consider an endowment economy that is very similar to the
one we considered in Section 4.A. The government in each country

Electronic copy available at: https://ssrn.com/abstract=4668578


global imbalances and the exorbitant privilege 173

issues an exogenously specified amount of risk-free bond, which is


held by the households in both countries. Let b̄t and b̄t∗ denote the
quantity of bonds issued by the home and foreign governments. The
proceeds from the issuance are transferred to the local households.
The home households’ utility is derived over consumption and the
market value of home and foreign bond holdings:

"  #
1 1− γ 1 1
E0 ∑ δ t
c + 1− σ
ω H,t b H,t + 1− σ
ω F,t bF,t ,
t =0
1−γ t 1−σ 1−σ

subject to the budget constraint

pt yt + exp(rt−1 )b H,t−1 + exp(rt∗−1 − et )bF,t−1 = ct + b H,t + exp(−et )bF,t + (exp(rt−1 )b̄t−1 − b̄t ).

The home households’ consumption is a basket of home and for-


eign goods: ct = (c H,t )α (c F,t )1−α . Following the same derivation in
Section 4.A, we obtain the following Euler equations:
ω H,t b−
" 
c t +1 − γ
 σ
#
H,t
1 = Et δ exp(rt ) + −γ ,
ct ct
ω F,t b−
" 
c t +1 − γ
 σ
#
∗ F,t
1 = Et δ exp(−∆et+1 + rt ) + −γ exp(et ) .
ct ct
Similarly, the foreign households’ utility is

"  #
1 1 1
E0 ∑ δ t ∗ 1− γ
(c ) + ∗
ω (b )∗ 1− σ ∗ −γ ∗ ∗
+ (c̄ ) θ H,t b H,t + ∗ ∗ 1− σ
ω (b ) ,
t =0
1−γ t 1 − σ H,t H,t 1 − σ F,t F,t

subject to the budget constraint

p∗t y∗t + exp(rt∗−1 )b∗F,t−1 + exp(rt−1 + et )b∗H,t−1 = c∗t + b∗F,t + exp(et )b∗H,t + (exp(rt∗−1 )b̄t∗−1 − b̄t∗ ).

Likewise, we obtain the following Euler equations for the foreign


households:
c∗t+1 −γ ω F∗ (b∗F,t )−σ
"   #

1 = Et δ exp(rt ) + ,
c∗t (c∗t )−γ
c∗t+1 −γ ω ∗H (b∗H,t )−σ + (c̄∗ )−γ θ H,t

"   #
1 = Et δ exp(∆et+1 + rt ) + exp(−et ) .
c∗t (c∗t )−γ
We have chosen a parsimonious specification of the safe asset de-
mand. The terms ω ∗H (b∗H,t )−σ and ω F∗ (b∗F,t )−σ capture the downward-
sloping demand for each type of bonds, which helps us pin down
the equilibrium quantities of bonds held by home and foreign house-
holds. The additional term (c̄∗ )−γ θ H,t
∗ captures a demand shifter,

which reflects the foreign households’ time-varying and countercycli-


cal demand for dollar safe assets. In general, we could have inserted
this demand shifter for the home households and for the foreign
bond as well, but one demand shifter is sufficient to capture the key
features of the model, which we will study in more detail below.

Electronic copy available at: https://ssrn.com/abstract=4668578


174 jiang – lecture notes on international finance

7.C.1 Macro Synthesis

We consider a simple case in which the government debt supply is


∗ is exogenous. The ex-
constant and, the bond demand shock θ H,t
ogenous variables are the endowments and the foreign households’
dollar asset preferences:

∗ ∞
(yt , y∗t , θ H,t ) t =0 .

There are 15 endogenous variables in each period t:

(ct , c H,t , c F,t , b H,t , bF,t , pt , c∗t , c∗H,t , c∗F,t , b∗H,t , b∗F,t , p∗t , rt , rt∗ , et )∞
t =0 .

The model implies the following 16 equations in each period, one


of which is redundant. These 16 equations include 2 consumption
aggregation equations,

ct = (c H,t )α (c F,t )1−α ,


c∗t = (c∗F,t )α (c∗H,t )1−α ,

4 household budget conditions are

pt yt + exp(rt−1 )b H,t−1 + exp(rt∗−1 − et )bF,t−1 = ct + b H,t + exp(−et )bF,t + (exp(rt−1 )b̄t−1 − b̄t ),
ct = pt c H,t + p∗t c F,t exp(−et ),
p∗t y∗t + exp(rt∗−1 )b∗F,t−1 + exp(rt−1 + et )b∗H,t−1 = c∗t + b∗F,t + exp(et )b∗H,t + (exp(rt∗−1 )b̄t∗−1 − b̄t∗ ),
c∗t = p∗t c∗F,t + pt c∗H,t exp(et ),

2 goods market clearing conditions

c H,t + c∗H,t = yt ,
c F,t + c∗F,t = y∗t ,

2 bond market clearing conditions

b̄t = b H,t + b∗H,t ,


b̄t∗ = bF,t + b∗F,t ,

2 within-period consumption choices


pt α c F,t 1 − α c F,t
= = ,
p∗t exp(−et ) 1 − α c H,t α c∗H,t

Electronic copy available at: https://ssrn.com/abstract=4668578


global imbalances and the exorbitant privilege 175

and 4 Euler equations

ω H b−
" 
c t +1 − γ
 σ
#
H,t
1 = Et δ exp(rt ) + −γ ,
ct ct
ω F b−
" 
c t +1 − γ
 σ
#
∗ F,t
1 = Et δ exp(−∆et+1 + rt ) + −γ exp(et ) ,
ct ct
c∗t+1 −γ ω F∗ (b∗F,t )−σ
"   #

1 = Et δ exp ( r t ) + ,
c∗t (c∗t )−γ
c∗t+1 −γ ω ∗H (b∗H,t )−σ + (c̄∗ )−γ θ H,t

"   #
1 = Et δ exp(∆et+1 + rt ) + exp(−et ) .
c∗t (c∗t )−γ

7.C.2 Model Characterization


Log-linearizing the Euler equations yields

−(λ∗H,t − λ H,t ) = Et [m∗t+1 − mt+1 ] + Et [∆et+1 ] = Et [γ(∆ log ct+1 − ∆ log c∗t+1 )] + Et [∆et+1 ].

def
Let ē = limt→∞ et denote the long-term exchange rate level, which is
well defined in a stationary economy. Iterating this equation forward,
we obtain the following result.

Proposition 7.2. The exchange rate level is equal to the expected consump-
tion growth differential and the expected convenience yield differential:
∞ ∞
et − ē = ∑ Et [γ(∆ log ct+ j − ∆ log c∗t+ j )] + ∑ Et [λ∗H,t+ j − λ H,t+ j ]
j =1 j =0
 ∗  ∞
ct ct
= −γ log

− log
c̄∗
+ ∑ Et [λ∗H,t+ j − λ H,t+ j ], (7.4)
j =0

where the convenience yield differential on the U.S. bond is approximately

ω ∗H (b∗H,t )−σ ω H b− σ
H,t
λ∗H,t − λ H,t ≈ exp(−et ) − ∗
− θ H,t exp(−et ). (7.5)
(c∗t )−γ −γ
ct

The proof is presented in Appendix A.29. This result extends the


exchange rate accounting formula in Proposition 4.2 with additional
assumptions about household preferences, which provides the key
to understand the exchange rate dynamics in response to different
types of shocks. The right-hand side of Eq. (7.4) is a sum of two
components. The first component is the marginal utility differential.
When markets are complete, as we saw in Section 4.B.4, it is the only
determinant of the exchange rate level. The second component is
the sum of expected convenience yield differentials. It arises from
the Euler equation wedges, and drives additional variations in the
exchange rate dynamics. Interestingly, it is not the convenience yield

Electronic copy available at: https://ssrn.com/abstract=4668578


176 jiang – lecture notes on international finance

per se that affects the exchange rate, but the differential between the
foreign investors’ perspective and the home investors’ perspective.
To obtain a stronger dollar (i.e., a higher et ), we need the foreign
investors’ convenience yield on the dollar bond to exceed the home
investors’ convenience yield.
Eq. (7.5) further connects the convenience yield differential to the
quantities of the U.S. bond held by foreign and U.S. households,
b∗H,t and b H,t , and the exogenous demand shifter θ H,t
∗ . If the foreign

households hold more U.S. bonds, their marginal utility from holding
the U.S. bond is lower, which reduces their convenience yield λ∗H,t
and depreciates the dollar, unless the increase in the foreign house-
holds’ holding is accompanied by a change in the demand shifter
∗ .
θ H,t
In this way, the exchange rate dynamics are driven not only by
the marginal utility differential, but also by the convenience yield
differential. Heuristically, in complete markets, the exchange rate
movement is determined by the differential in marginal utilities over
consumption:

∆e = (∆uc − ∆u∗c ).

As recessions lower the local households’ consumption, raise their


marginal utility, and appreciate the local currency, the exchange
rate is always counter-cyclical. However, in the data, exchange rates
are acyclical or even procyclical, meaning that currencies tend to
depreciate in bad times [Backus and Smith, 1993].
In comparison, after we introduce the non-pecuniary benefits
of holding bonds, the exchange rate is determined by not only the
differential in marginal utilities over consumption, but also the differ-
ential in marginal utilities over bond holding. Heuristically, we have

∆e = (∆uc − ∆u∗c ) + (∆u∗b − ∆ub ),

where ∆u∗b − ∆ub captures the marginal utility differentials over bond
holding for current and future periods as we specified in Eq. (7.5).
Take the foreign households as an example. Recessions lower their
wealth and reduce their holdings of the dollar safe assets. Given
their downward-sloping demand curve for dollar safe assets, they
impute a higher convenience yield and accept a lower expected re-
turn to hold the U.S. bond. To equilibrate their demand with the U.S.
households, the dollar has to appreciate to generate an expected de-
preciation, which leads to a lower expected return to hold the U.S.
bond from the foreign perspective. In this way, the demand for safe
assets connects wealth decline to local currency depreciation. If the
wealth declines more than consumption in recessions, this channel

Electronic copy available at: https://ssrn.com/abstract=4668578


global imbalances and the exorbitant privilege 177

based on marginal utilities over bonds could overturn the complete-


market channel based on marginal utilities over consumption, and
generate procyclical exchange rates.

7.C.3 Calibration and the Steady State


We log-linearize the model and solve the forward-looking first-order
dynamics around a non-stochastic steady state. The details of this
technique can be found in Christiano [2002], and Dynare offers a
convenient implementation.
We find the non-stochastic steady state by setting all variables to
be constant over time. For example, the home goods’ market clearing
condition becomes
c H,SS + c∗H,SS = ySS ,
and the foreign investors’ Euler equation for home bonds becomes
−γ
ω ∗H (b∗H,SS )−σ + cSS θ H,SS

1 = δ exp(rSS ) + −γ exp(−eSS ),
cSS
which uses ∆eSS = eSS − eSS = 0 and c∗SS /c∗SS = 1.
We pick the following parameter values: γ = 1, σ = 3, α = 0.85,
δ = 0.99, and ȳ = ȳ∗ = 1. We calibrate the variables ω H = ω F∗ ,
ω ∗H = ω F , and θ̄ H∗ such that in the non-stochastic steady state, the

convenience yield on the home bond λ H,SS is 1.01%, the convenience


yield on the foreign bond λ F,SS is 0.50%, the U.S. debt/output ratio
b̄/( pSS ySS ) is 100%, the foreign debt/output ratio b̄∗ /( p∗SS y∗SS ) is
79%, b∗H,SS /b̄ = 50% of the U.S. bond is held by the foreign house-
holds, and bF,SS /b̄∗ = 20% of the foreign bond is held by the U.S.
households. This set of targets implies that the U.S. net foreign asset
is −34% of the U.S. output in the steady state.
Table 7.1 reports the more details about the steady-state values.
The U.S. consumption is slightly higher than the foreign consump-
tion and on average the U.S. runs a permanent trade deficit equal to
−0.08% of the total output. This trade deficit is funded by the fact
that the U.S. pays a lower interest rate on its external borrowing than
the foreign country—a situation we refer to as the seigniorage rev-
enue. Quantitatively, the U.S. borrows 50% of its output from the
foreign households at an interest rate of 0%, whereas the foreign
country borrows 16% of its output from the U.S. households at an
interest rate of 0.5%. In the steady state, the foreign country pays the
U.S. a net interest equal to 16% × 0.5% = 0.08% of the foreign output.
In reality, this seigniorage revenue could be much bigger if the dis-
count rate on the foreign liabilities is higher than 0.5% or if the size of
the foreign liabilities is greater than 16% of the foreign output in our
calibration.

Electronic copy available at: https://ssrn.com/abstract=4668578


178 jiang – lecture notes on international finance

Variable Notation Value Table 7.1: Steady-State Values


Home consumption cSS 0.656
Foreign consumption c∗SS 0.654
Home trade balance tbSS /( pSS ySS ) −0.08%
Home bond rate rSS 0.00%
Foreign bond rate ∗
rSS 0.50%
Home bond convenience yield λ H,SS 1.01%
Foreign bond convenience yield λ F,SS 0.50%
Exchange rate eSS 0.0026
U.S. debt-output ratio b̄/( pSS ySS ) 100%
Foreign debt-output ratio b̄∗ /( p∗SS y∗SS ) 79%
Foreign holding of home bond b∗H,SS /b̄ 50%
Home holding of foreign bond bF,SS /b̄∗ 20%
Home net foreign asset n f a H,SS /( pSS ySS ) −34%

7.C.4 Flight to Safety and International Wealth Transfers


In the model we considered under the insurance provision view, we
interpret a global crisis as a low endowment state. In this model, we
are going to highlight a different aspect of a global crisis: a flight to
safety. In particular, we focus on the foreign households’ demand for
∗ . It is likely that an
the U.S. risk-free bond, which is captured by θ H,t
increase in the foreign households’ safe asset demand is correlated
with declines in home and foreign endowments. However, we first
study this shock in isolation to highlight its role in resolving the
reserve currency paradox.
Figure 7.9 reports the impulse responses to a shock to θ H,t∗ . We

assume that this shock dissipates slowly with an autocorrelation of


0.9. Upon the arrival of the shock, we see an increase in the foreign
households’ convenience yield λ∗H,t on the dollar risk-free bond,
defined as

c∗t+1 −γ
"   #

exp(−λ H,t ) = Et δ exp(∆et+1 + rt ) .
c∗t

Through the exchange rate formula we derived in Proposition 4.2,


Section 4.B, generalized below,
∞ ∞ ∞
et = ∑ Et [rt+ j − rt∗+ j ] + ∑ Et [λ∗H,t+ j − λ∗F,t+ j ] − ∑ Et [rpt+ j ] + ē,
j =0 j =0 j =0

the increase in the convenience yield on the dollar bond appreciates


the dollar. It is worth noting that the dollar appreciation requires us
to break the complete-market condition (1.16), reproduced below,

∆et+1 = mt+1 − m∗t+1 = −∆ct+1 + ∆c∗t+1 ,

Electronic copy available at: https://ssrn.com/abstract=4668578


global imbalances and the exorbitant privilege 179

1 1
2

1.5
0.5 0.5 1

0.5

0 0 0
10 20 30 40 10 20 30 40 10 20 30 40

0
2 2
1.5 1.5 -0.2
1 1
-0.4
0.5 0.5
-0.6
0 0
10 20 30 40 10 20 30 40 10 20 30 40

0 6
6
-0.5
4
4
-1
2 2
-1.5

-2 0 0
10 20 30 40 10 20 30 40 10 20 30 40

Figure 7.9: Impulse Responses to


Dollar Bond Demand Shock. The
variables are the foreign demand for the
because, as we will see below, the U.S. consumption is relatively dollar bond, the foreign convenience
yield, the dollar’s log exchange rate,
higher than the foreign consumption in the flight-to-safety episodes, the log consumption differential,
leading to a lower U.S. households’ marginal utilities and hence a the log wealth differential, the trade
balance/output ratio, the NFA/output
weaker dollar under complete markets.
ratio, the fraction of dollar bonds
On the quantity side, the foreign households’ purchase of the dol- held by foreign households, and the
lar bond is financed by reducing their consumption and their holding fraction of foreign bonds held by U.S.
households.
of the foreign bond. As a result, the transfer of the dollar bond from
the U.S. households to the foreign households is accompanied by a
transfer of consumption goods and the foreign bond from the foreign
households to the U.S. households. The U.S. households’ consump-
tion increases relative to the foreign households’ consumption.
This increase in U.S. relative consumption reflects a seigniorage
revenue that the U.S. earns from issuing bonds that carry conve-
nience yields. By issuing expensive bonds in high convenience yield
states, the U.S. receives real resources from the rest of the world. If
the convenience yield shock is persistent, the seigniorage revenue is

Electronic copy available at: https://ssrn.com/abstract=4668578


180 jiang – lecture notes on international finance

also expected to last, leading to an increase in the expected future


seigniorage revenues. This increase in the expected future seignior-
age revenues is reflected in the U.S. households’ wealth at , which is
defined as the present value of their consumption streams:
∞ −γ
" #
c
a t = Et ∑ δ s −s
γ cs .
s=t ct

This wealth effect is the key distinguishing feature of the safe asset
view, as it disentangles the response in the U.S. wealth at from the re-
sponse in the U.S. net foreign assets n f at . On the one hand, the U.S.
buys some foreign bonds from the foreign country and sells some
dollar bonds to the foreign country. In a flight-to-dollar episode, the
dollar appreciates and the U.S. suffers a loss on its external portfo-
lio, leading to a decline in the U.S. NFA. On the other hand, the U.S.
receives a higher seigniorage revenue from issuing the dollar bonds,
which increases the U.S. wealth despite its loss on the external port-
folio.
In this model, a higher U.S. wealth relative to the foreign wealth
leads to a higher U.S. consumption relative to the foreign consump-
tion. In the special case with log preferences, the U.S. households’
consumption is proportional to their wealth:

c t = (1 − δ ) a t .

Due to the home bias in the U.S. households’ consumption, the U.S.
spends more on the U.S. goods, which appreciates the dollar in real
terms. In this way, the goods market clearing also implies a stronger
dollar that is consistent with the convenience yield channel in the
asset market. As such, a decline in the U.S. NFA does not necessar-
ily imply a relative wealth gain for the foreign households and a
stronger demand for the foreign goods. The reserve currency para-
dox is resolved by engineering a countercyclical U.S. wealth share
from the seigniorage revenue.
This discussion highlights the countercyclicality of the U.S. wealth
share as the key to understand the cyclical properties of international
transfers. Jiang, Krishnamurthy, and Lustig [2020a] presents a simple
calculation around 2008, and shows that the decline in the market
value of the U.S. equities, bonds, and deposits was indeed less than
the decline in the market value of the equities, bonds, and deposits
in major developed foreign countries after converting to the dollar
units. This relative gain in market value was also higher than the loss
in the U.S. NFA. Dahlquist, Heyerdahl-Larsen, Pavlova, and Pénasse
[2022], Kim [2022] present additional empirical evidence in favor of
a countercyclical U.S. wealth share, whereas Sauzet [2022] presents a
calculation that suggests the opposite. It is possible that the wealth

Electronic copy available at: https://ssrn.com/abstract=4668578


global imbalances and the exorbitant privilege 181

calculation includes different types of agents and assets in these


studies, and more work is needed to think about the appropriate
definition of wealth and reconcile these findings.
Finally, we note that there are alternative approaches to simulta-
neously address the cyclicality of the dollar and that of the interna-
tional transfers and resolve the reserve currency paradox. Kekre and
Lenel [2021], Sauzet [2022], Devereux, Engel, and Wu [2023] engineer
greater recessions in the U.S. relative to the foreign countries dur-
ing global downturns, which raise the U.S. marginal utility relative
to the foreign marginal utility and appreciate the dollar. Dahlquist,
Heyerdahl-Larsen, Pavlova, and Pénasse [2022] consider a different
mechanism that emphasizes the time-varying demand for the U.S.
goods, which also generates a countercyclical U.S. wealth share as in
our model. Finally, the demand shift towards the U.S. goods is con-
sidered by Maggiori [2017] as a way to engineer a dollar appreciation
during global downturns.
All of these resolutions can be framed under the goods market and
the asset market views of the exchange rate in Section 1.D. Specifi-
cally, Eq. (1.19), reproduced below,
2α − 1 2α − 1 AM
   
2α ∗ GM
+ 2(1 − α ) et = − log yt − log yt − τt + τt ,
γ 2α − 1 γ
shows that both the asset market wedge τtAM and the goods market
wedge τtGM can impact the exchange rate et and its cyclicality. Each
of the above resolutions proposes a specific wedge that deviates from
the frictionless benchmark and generates the dollar exchange rate’s
countercyclicality.

7.D The Stability of the International Monetary System

Having presented the two views of the reserve currency paradigm,


we briefly survey two important issues related to the stability of the
international monetary system. This discussion is based on a large
body of literature on the international monetary system.

7.D.1 Triffin’s Dilemma


In the Bretton-Woods system with a de-jure dollar standard, foreign
countries peg their currencies to the U.S. dollar, and the U.S. in turn
pegs the dollar to the price of gold. Triffin [1960] foresaw an emerg-
ing imbalance. He argued that, as the world demand for dollar re-
serve assets grew with the world economy, the U.S. will inevitably
run out of gold backing for its supply of reserve assets. Eventually,
the erosion of backing will lead to a run on the dollar and the col-
lapse of the international monetary system. This is the famous Triffin

Electronic copy available at: https://ssrn.com/abstract=4668578


182 jiang – lecture notes on international finance

dilemma, which highlights the tension between providing liquidity


demanded by a growing world economy and maintaining the value
of the reserve currency.
In response to Triffin’s critique of the Bretton-Woods system, De-
spres, Kindleberger, and Salant [1966] propose the famous minority
view, which holds that the U.S., equipped with the deepest and most
liquid financial markets in the world, is a natural world banker which
provides liquid assets to the world. Like banks, this intermediation
process can be stable for a long period of time. While this U.S.-as-
world-banker view survives and exerts a strong influence on the
modern literature, Triffin proved to be prescient. The U.S. ran out of
its gold backing for its growing dollar liabilities in the early 1970s.
As a result, it was forced to devalue the dollar relative to gold in 1971
and eventually abandon convertibility from the dollar to gold in 1973.
Exchange rates became free-floating and the Bretton-Woods system
collapsed.
In today’s world, we live with a de-facto dollar standard. Both the
insurance provision view and the safe asset view hold that the world
has a strong demand for dollar reserve assets, whose backing is pro-
vided by revenue streams of firms and governments instead of gold.
One can in fact argue that this is an extraordinary feat of financial en-
gineering, as the U.S. managed to manufacture reserve assets backed
by risky (pro-cyclical) cash flows instead of steady collaterals such
as gold. In this sense, as Gourinchas, Rey, and Sauzet [2019] point
out, “Triffin’s analysis, however, was incomplete because, despite the aban-
donment of the dollar-gold parity, the dominance of the dollar has increased
since the collapse of the Bretton Woods system. Paradoxically, once free from
the shackle of a fixed gold parity, the use of the US dollar as an international
currency soared to unprecedented levels. Yet, as we argue, the financial
fragilities inherent in a hegemonic system have not disappeared: The Triffin
dilemma is still with us, albeit in a subtly different form.”
Indeed, we may expect two new forms of Triffin dilemma in to-
day’s world. First, the increasing demand for dollar safe assets may
be met with insufficient fiscal backing [Gourinchas and Rey, 2007b,
2022, Farhi, Gourinchas, and Rey, 2011, Obstfeld, 2011, Farhi and
Maggiori, 2018, Jiang, Lustig, Van Nieuwerburgh, and Xiaolan, 2019],
which makes the U.S. become a more leveraged safe asset supplier.
We will discuss this issue in greater details in Chapter 8.
Second, the premium on the dollar debt incentivizes foreign is-
suers to also dollarize their liabilities, while their revenues are mostly
in local currency terms and thus generate a currency mismatch on
their balance sheets. As Jiang, Krishnamurthy, and Lustig [2020a]
write, “as a result, a different form of instability appears: as the world de-
mand for dollar grows, the incentive for both U.S. and foreign issuers to

Electronic copy available at: https://ssrn.com/abstract=4668578


global imbalances and the exorbitant privilege 183

supply dollar assets will grow. In particular, if the growth in the world de-
mand for dollar safe assets exceeds the growth in U.S. supply, the result
will be growth in currency-mismatched balance sheets around the world.
The conclusion is that financial spillovers and the global financial cycle may
grow in importance.”
This currency mismatch can be incurred by foreign governments,
firms, and banks as they take advantage of the convenience yield
on dollar funding. A large literature studies this issue from both
theoretical and empirical angles. See Caballero and Krishnamurthy
[2003], Schneider and Tornell [2004], Bocola and Lorenzoni [2020],
Du, Pflueger, and Schreger [2020], Du and Schreger [2022b], Salomao
and Varela [2022], Gutierrez, Ivashina, and Salomao [2023].

7.D.2 The Global Financial Cycle and the Impossible Duo


Another form of international fluctuations arises in the form of a
global financial cycle. Rey [2015], Miranda-Agrippino and Rey [2015]
find a strong commonality in the fluctuations in capital flows, asset
prices, and credit growth around the world, and this cycle is aligned
with the VIX and the dollar. In this sense, we can think of the VIX
and the dollar as barometers of global risk appetite. Moreover, the
U.S. monetary policy plays a key role in driving this global financial
cycle, with a U.S. monetary tightening triggering collapses in asset
prices and retrenchment in capital flows. In comparison, other coun-
tries’ monetary policies appear to be less important [Gerko and Rey,
2017].
This global financial cycle challenges how we think about the inde-
pendence of monetary policy. The classical Mundell-Flemming model
presents the impossible trinity: with free capital mobility, indepen-
dent monetary policies are feasible if and only if exchange rates are
floating. This result is obtained in canonical models that focus on the
interest rate channel alone, as the exchange rate can endogenously
adjust to make independent interest rate targets possible.
The global financial cycle shows that the policy space is more con-
strained, leading to an impossible duo [Rey, 2015, Miranda-Agrippino
and Rey, 2015]: with free capital mobility, floating exchange rates do
not guarantee monetary policy independence because asset prices
and capital flows still respond to global factors. In light of our ex-
change rate accounting formula in Proposition 4.2, reproduced below,
∞ ∞ ∞
et = ∑ Et [rt+ j − rt∗+ j ] + ∑ Et [λ∗t+ j ] − ∑ Et [rpt+ j ] + ē,
j =0 j =0 j =0

the traditional impossible trinity view can be thought of as focusing


on the interest rate term, i.e., rt+ j − rt∗+ j . Ignoring the convenience

Electronic copy available at: https://ssrn.com/abstract=4668578


184 jiang – lecture notes on international finance

yield and the risk premium terms, we can expect the variations in the
exchange rate to absorb the shocks to the autonomous interest rate
policies. In comparison, the impossible duo view emphasizes the risk
premium terms rpt+ j and the convenience yield terms λ∗t+ j , which af-
fect not only exchange rates but also domestic financial and economic
conditions. Even with floating exchange rates, monetary policy re-
sponses are needed to address the influences of these external factors.
As a result, monetary policies are not truly independent if the capital
accounts are open and allow the risk premium and convenience yield
shocks to spill over.

Electronic copy available at: https://ssrn.com/abstract=4668578


8
Government Debt

Summary

• If we look backward, the market value of government debt is determined by past debt, pri-
mary surpluses, and debt returns:

t −1 S
Dt t− j D0
=−∑ exp( RtD− j→t − Xt− j→t ) + exp( R0D→t − X0→t ).
Yt Y
j =0 t − j
Y0

• If we look forward, the market value of government debt is backed by the present value of
future primary surpluses:


" #
Dt = E t ∑ exp( Mt,t+k )St+k + lim Et [exp( Mt,t+k ) Dt+k ] .
k→∞
k =1

• This forward-looking valuation equation also implies a trade-off in the risk dimension. The
government has to choose between insuring the debtholders and insuring the taxpayers.

• In the cases of the U.S. and past reserve asset issuers, the market value of government debt
may exceed its fiscal backing, leading to a public debt valuation puzzle.

In the last chapter, we saw two accounts of the international mon-


etary system that both regard safe assets as the key ingredient. One
view emphasizes the international risk-sharing arrangement that is
implemented by the U.S. supplying safe assets and taking a levered
position on riskier claims, whereas the other view focuses on the
foreigners’ countercyclical demand for safe assets. In reality, most
safe assets are government debt. In fact, when the U.S. establishes its
status as the world’s reserve asset supplier, it is conceivable that the
U.S. government debt first becomes an important reserve asset before
other dollar-denominated private debt caught up in their status.
In this chapter, we study the valuation of government debt and
issues related to fiscal sustainability in greater details. We build on
Chapters 4, 6, and 7, in which we introduced one-period risk-free

Electronic copy available at: https://ssrn.com/abstract=4668578


186 jiang – lecture notes on international finance

government bonds. For example, in Chapter 6, the government bud-


get condition in nominal terms can be expressed as

Qt = St + Qt+1 exp(−it ),

where St is the government primary surplus in nominal terms, Qt is


the nominal face value of the debt, and it is the log nominal risk-free
rate.
Now, we consider more general cases in which the government
can issue different types contingent claims, including the long-term
bonds. Let Dt denote the market value of aggregate government debt
portfolio, and let exp( RtD ) denote its cum-coupon return. Then, the
government budget condition can be expressed as

Dt−1 exp( RtD ) = St + Dt , (8.1)

with the previous special case of one-period nominal risk-free bond


attained when Qt = Dt−1 exp( RtD ) and RtD = it−1 .
Before we formally analyze this general case, let us consider a sim-
ple personal loan example. Suppose I borrow $100,000 today at the
annual interest rate of 5% to purchase a house. The loan duration
is 10 years, with constant payment each year. A simple calculation
shows that the constant payment is $12,950 each year. Table 8.1 re-
ports the detailed loan schedule.
Now, suppose I stand at the beginning of year 6. I have made 5
loan payments in the past, and my outstanding balance is $56,069.
This number has two interpretations. The first interpretation is based
on backward-looking accounting: this outstanding balance of $56,069
is a result of me repaying the loan in the past 5 years. In each year,
after I pay back the interest that is equal to 5% times the outstanding
loan balance, the remaining amount of my payment (i.e., the princi-
pal payment in the last column) contributes to the reduction of my
outstanding loan amount over time. In other words, my outstand-

Year Balance Payment Interest payment Principal payment Table 8.1: Personal Loan Example
1 $100,000 $12,950 $5,000 $7,950
2 $92,050 $12,950 $4,602 $8,348
3 $83,702 $12,950 $4,185 $8,765
4 $74,936 $12,950 $3,747 $9,204
5 $65,733 $12,950 $3,287 $9,664
6 $56,069 $12,950 $2,803 $10,147
7 $45,922 $12,950 $2,296 $10,654
8 $35,267 $12,950 $1,763 $11,187
9 $24,080 $12,950 $1,204 $11,746
10 $12,334 $12,950 $617 $12,334

Electronic copy available at: https://ssrn.com/abstract=4668578


government debt 187

ing balance of $56,069 in year 6 is equal to the initial loan amount of


$100,000 minus the sum of the first 5 principal payments.
The second interpretation is based on forward-looking valuation:
my outstanding balance of $56,069 at the beginning of year 6 also
reflects the present value of my remaining payments in the future.
From the bank’s perspective, my personal loan is valued at $56,069
because this value describes the present value of the remaining 5
payments discounted at the interest rate of 5%. This discount rate,
by the way, does not have to equal to the risk-free rate if the bank
believes that I may default on the loan. In this way, my outstanding
loan balance in a given year can be interpreted both as the result of
past borrowing and payments, and as the discounted present value of
future promised payments.
The analysis of government debt is similar to this example of
personal loan. The only key difference is that the analysis of gov-
ernment debt tends to have an infinite horizon as the government
is supposed to be long-lasting, whereas the personal finance prod-
ucts tend to have a pre-determined horizon. We next examine the
backward-looking and forward-looking approaches to government
debt following Jiang, Lustig, Van Nieuwerburgh, and Xiaolan [2023b].

8.A Backward-Looking Accounting

The backward-looking approach attributes variations in the debt/GDP


ratio to the history of primary surpluses, output growth, inflation,
and interest rates. Specifically, by iterating backward Eq. (8.1), we
can write the debt today as a function of cumulative past returns and
past primary deficits [Hall and Sargent, 2011].

Proposition 8.1. (a) The market value of government debt can be expressed
as
t −1
Dt = − ∑ St− j exp( RtD− j→t ) + D0 exp( R0D→t ), (8.2)
j =0

where RtD− j→t = RtD− j+1 + . . . RtD−1 + RtD is the cumulative debt return.
(b) We can restate this expression to obtain a backward-looking expres-
sion for the debt/GDP ratio:

t −1 S
Dt t− j D0
=−∑ exp( RtD− j→t − Xt− j→t ) + exp( R0D→t − X0→t ), (8.3)
Yt Y
j =0 t − j
Y0

where exp( Xt− j→t ) = Yt /Yt− j is the cumulative nominal GDP growth.

The proof is in Appendix A.30. As the government debt/GDP


ratio is stationary in many models, it is usually easier to focus on

Electronic copy available at: https://ssrn.com/abstract=4668578


188 jiang – lecture notes on international finance

this ratio instead of the debt level. Proposition 8.1(b) shows that the
current debt/GDP ratio depends on past primary surpluses and
deficits: a greater borrowing in the past to finance deficits −St− j
leads to a higher debt/GDP ratio today. Moreover, the borrowing is
compounded by the growth-adjusted return RtD− j→t − Xt− j→t , so that
high real debt returns, high inflation rates, and low output growth all
magnify the past borrowing and contribute to a high debt/GDP ratio
today.
In this way, the backward-looking approach provides a useful way
to attribute the increase in the government debt/GDP ratio from time
0 to t to four components: government deficits, real debt returns,
inflation, and output growth. Hall and Sargent [2011, 2022], Ander-
son and Leeper [2023] conduct such decomposition exercises in the
history of the U.S. For example, the debt/GDP ratio increased from
86% to 101% in the Covid period from 2019 to 2023. Government
deficits were the main contributor to the increase, while high infla-
tion, positive real GDP growth, and negative debt returns all helped
to partially offset the deficits.
We can also examine this result in a steady state in which debt
returns, growth rates, and surplus/GDP ratios are all constants.
Then, the steady-state debt/GDP ratio can be expressed as

D S 1 S 1
=− D
≈ D
,
Y Y 1 − exp( R − X ) YR −X
S D
= ( R D − X ).
Y Y
In real terms, we can express this relation as

s d
= (r D − x ), (8.4)
y y

where r D = R D − π denotes the real return to the bond portfolio and


x = X − π denotes the real growth rate.
First, consider the case in which the debt returns are higher than
the GDP growth, i.e., r D > x. For a positive amount of debt, i.e.,
d > 0, the government has to run a budget surplus in perpetuity, i.e.,
s > 0, to pay off the interest and sustain the same debt/GDP level.
Then, consider the case in which the debt returns are lower than
the GDP growth, i.e., r D < x. The literature usually denotes this con-
dition by r < g. In this case, the government can roll over its debt in
perpetuity while running steady-state deficits, i.e., s < 0. Blanchard
[2019] uses this logic to argue that the U.S. has an infinite debt capac-
ity as long as the interest rate is lower than the GDP growth rate. In
this environment, in fact, the government can start with an arbitrary
amount of government debt. If it just rolls over its debt by raising

Electronic copy available at: https://ssrn.com/abstract=4668578


government debt 189

more debt to pay the interests due, then, the government debt quan-
tity rises at the interest rate r D , while the GDP grows at a higher rate
of x. If we wait long enough, the government debt/GDP ratio will
decline to zero, making it easy for the government to eventually raise
a one-time tax as a small fraction of the GDP and pay down the debt.
It is also straightforward to consider bond convenience yields in
this calculation. In the absence of convenience yields, the debt return
is equal to the risk-free rate, r D = r. In the presence of convenience
yields, the government is able to finance its debt at a rate lower than
the risk-free rate ρ:

r D = ρ − λ.

Then, Eq. (8.4) becomes

s d
= ( ρ − λ − x ),
y y

which implies that, in the case of r D > x, a higher convenience


yield λ lowers the required surplus/GDP ratio to sustain a given
debt/GDP ratio.

8.A.1 Limitations to the Backward-Looking Approach


There are limits to what we learn from the backward-looking ap-
proach, which is a statement of the past. The future may or may not
be the same. Specifically, consider the task of predicting the market
value of debt in the next period t + 1, which follows

Dt+1 = −St+1 + Dt exp( RtD+1 ),

and therefore depends on not only the future government surplus


St+1 , but also the future debt return RtD+1 . While the future surplus
may be predicted following some well-defined legislative or bud-
getary rules, it is much harder to predict the future debt return,
which depends on the investors’ expectation of the government’s
fiscal conditions in the more distant future and their risk appetite.
Similarly, back in the personal loan example in Table 8.1, if the lender
is unsure of the borrower’s future income and therefore has doubt
about the borrower’s ability to repay the loan, then, the market value
of the outstanding loan needs to adjust to reflect this uncertainty.
In this sense, the law of motion for government debt is less “me-
chanical” than the law of motion for economic quantities such as the
capital stock. The capital stock depends on depreciation and invest-
ments, and large adjustments in these economic quantities are subject
to adjustment cost. Adjustments in the debt return, in comparison,
reflect changes in investors’ expectations and risk preferences which

Electronic copy available at: https://ssrn.com/abstract=4668578


190 jiang – lecture notes on international finance

can be much faster and greater. As a result, in the next section we


turn to the valuation tools developed in finance, which are exactly
about understanding how asset prices respond to uncertainty in the
future.

8.B Forward-Looking Valuation

As in the backward-looking accounting, here we also start with Eq.


(8.1). We consider a useful special case in which the government
issues nominal bonds up to tenor H. So, the market value of the
entire government debt portfolio can be expressed as

H
Dt = ∑ Qt (h) Pt (h),
h =1

where Qt (h) is the quantity of outstanding nominal zero-coupon


bonds of maturity h in period t, and Pt (h) is the price for the h-year
bonds in period t. We note that the one-period nominal risk-free debt
is a special case with H = 1, Qt (1) = Bt+1 , and Pt (1) = exp(−it ).
These bonds may be defaultable. When government default does
not happen in period t, the one-period government budget condition
becomes
H H
Dt−1 exp( RtD ) = ∑ Qt−1 (h) Pt (h − 1) = St + ∑ Qt (h) Pt (h). (8.5)
h =1 h =1

When default happens in period t, all existing bonds from period


t are wiped out.1 After the default, the government may be able to 1
It is a simple extension to consider
issue new bonds at prices Pt (h) in this current period, as well as in partial default, which we do not con-
sider in this note.
future periods:

H
0 = St + ∑ Qt (h) Pt (h).
h =1

This time, we iterate forward these equations, and derive the fol-
lowing intertemporal government budget condition. We use Mt,t+k to
denote the multi-horizon nominal SDF from period t to period t + k.

Proposition 8.2. Even when the government debt is defaultable, the market
value of government debt is equal to the present value of future government
surpluses plus a transversality term:


" #
Dt = E t ∑ exp( Mt,t+k )St+k + lim Et [exp( Mt,t+k ) Dt+k ] .
k→∞
(8.6)
k =1

Using the one-period budget constraint (8.1) again, we can also express

Electronic copy available at: https://ssrn.com/abstract=4668578


government debt 191

this intertemporal government budget condition as



" #
Dt−1 exp( RtD ) = Et ∑ exp( Mt,t+k )St+k + lim Et [exp( Mt,t+k ) Dt+k ] .
k→∞
k =0
(8.7)

The proof is given in Appendix A.31. Eq. (8.6) and (8.7) are equiv-
alent ways to express the same intertemporal government budget
condition. Eq. (8.6) equates the end-of-period market value of the
government debt after issuances and repayments, Dt , to the ex-
dividend present value of future surpluses and the transversality term,
whereas Eq. (8.7) equates the beginning-of-period market value of the
government debt, Dt−1 exp( RtD ), to the cum-dividend present value
of future surpluses and the transversality term. Our results can be
expressed in either convention we use. To avoid confusion, we stick
to the first expression in this section.
If the transversality condition holds, i.e.,

lim Et [exp( Mt,t+k ) Dt+k ] = 0, (8.8)


k→∞

then, we can express this intertemporal government budget condition


(8.6) as

Dt = PtT − PtG , (8.9)

where Dt on the left-hand side is the market value of debt at the end
of period t, and PtT and PtG on the right-hand side denote the present
value of current and future tax revenues and government spending:

" #
def
PtT = Et ∑ exp( Mt,t+k )Tt+k ,
k =1

" #
def
PtG = Et ∑ exp( Mt,t+k )Gt+k .
k =1

We refer to the present value of future government surpluses PtT −


PtG as the fiscal backing. When the transversality condition holds,
the fiscal backing determines the market valuation of the aggregate
government debt portfolio.
This equation provides a forward-looking valuation of the govern-
ment debt, which determines its value based on the expectation of
future government surpluses and their discount rates. In compari-
son, the backward-looking accounting formula (8.2) accounts for the
debt value based on the history of past government surpluses and
realized debt returns. As in the personal loan example we consid-
ered at the beginning of this chapter, these two approaches are both
valid and provide completementary perspectives on the valuation of
government debt.

Electronic copy available at: https://ssrn.com/abstract=4668578


192 jiang – lecture notes on international finance

We offer three economic interpretations of this forward-looking


relationship. First, it is worth noting that the cash flow stream of
government surpluses St can be replicated by simply trading the gov-
ernment debt: if an investor buys all new government debt issuances
and receives the government debt coupons and repayments, then,
his or her cash flow is exactly equal to the government surpluses.
As such, Proposition 8.2 shows that the aggregate government debt
portfolio can be priced by any investors who can trade this strategy,
even when the markets are incomplete and there are multiple SDFs
that price the tradable asset space.
In fact, for investors who can trade government debt, they should
agree on the valuation of government debt even when they have
i j
different SDFs. For investors i and j with SDFs Mt,t +k and Mt,t+k ,
respectively, we have
∞ ∞
" # " #
∑ exp( Mt,t ∑ exp( Mt,t+k )(Tt+k − Gt+k )
i j
Et +k )( Tt+k − Gt+k ) = Et .
k =1 k =1
(8.10)

If this equality does not hold, these two investors should trade with
each other using the government debt portfolio.
Second, this result implies that, in order to back up Dt dollars’
worth of debt, the government needs to generate a positive present
value from its future primary surpluses. Conversely, when the
present value of primary surpluses increases in period t, the gov-
ernment debt appreciates in value and generates a higher return RtD .
That is, Eq. (8.9) implies
h i h i
Dt−1 (Et − Et−1 ) exp( RtD ) = (Et − Et−1 ) ( PtT + Tt ) − ( PtG + Gt ) ,

where exp( RtD ) denotes the holding return of the aggregate govern-
ment bond portfolio:

∑hH=1 Qt−1 (h) Pt (h − 1) Dt + S t


exp( RtD ) = H
= .
∑h=1 Qt−1 (h) Pt−1 (h) Dt − 1

Third, if the transversality condition holds, we can express Eq.


(8.6) as
∞ ∞
Dt = ∑ Et [exp( Mt,t+k )] Et [St+k ] + ∑ covt (exp( Mt,t+k ), St+k ) ,
k =1 k =1

which implies that the fiscal backing can be created either by running
a higher government surpluses on average (i.e., higher Et [St+k ]), or
by making government surpluses more countercyclical (i.e., higher
covt (exp( Mt,t+k ), St+k )). The latter channel creates fiscal backing by
conditionally generating higher cash flows in high marginal utility

Electronic copy available at: https://ssrn.com/abstract=4668578


government debt 193

states, which provides an insurance to the debtholders—a point we


will revisit in our discussion of the fiscal trade-off in the risk domain.
Finally, we make two more technical observations. First, if the
transversality condition (8.8) holds, the intertemporal government
budget condition (8.6) for all periods has the same information con-
tent as the one-period budget constraint (8.1) for all periods. Specifi-
cally, imposing the intertemporal budget constraints in two adjacent
periods implies the one-period budget constraint in the first period.
To see this, simply apply Et [exp( Mt,t+1 )(·)] to both sides of the sec-
ond period’s intertemporal condition, and then subtract from the
first period’s intertemporal condition to obtain the first period’s one-
period budget constraint.
Second, the intertemporal government budget condition can also
be expressed in real terms. For example, we can rewrite Eq. (8.6) be
expressed as


" #
d t = Et ∑ exp(mt,t+k )st+k + lim Et [exp(mt,t+k )dt+k ] ,
k→∞
k =1

where dt and st are the real market value of government debt and the
real primary surplus, and mt,t+k is the real SDF which is related to
the nominal SDF via Eq. (6.1).

8.B.1 Incorporating the Convenience Yields


We can further generalize this result with convenience yields. The
convenience yield λt (h) is the expected returns on government bonds
of maturity h that investors are willing to forgo. As in Chapter 4, the
convenience yield is a wedge in the standard Euler equation:

Pt+1 (h − 1)
 
exp(−λt (h)) = Et exp( Mt+1 ) .
Pt (h)

The presence of the convenience yield implies that the government


can finance its debt at a lower interest rate or a higher bond price,
which increases the amount it can raise from debt issuance. This
additional revenue from debt issuance is a form of seigniorage rev-
enue, which we formally define as the bond market value Qt (h) Pt (h)
times the Euler equation wedge 1 − exp(−λt (h)), summed across all
maturities h:
H
def
Kt = ∑ Qt (h) Pt (h)(1 − exp(−λt (h))).
h =1

When the convenience yields are identical for bonds of all maturi-
ties, i.e., λt (h) = λt , then, we can simplify this seigniorage revenue

Electronic copy available at: https://ssrn.com/abstract=4668578


194 jiang – lecture notes on international finance

as the product between the market value of debt and the convenience
yield level:

Kt = Dt (1 − exp(−λt )) ≈ Dt λt ,

where the approximation holds when the convenience yield λt is


small.
This seigniorage revenue contributes to the government’s income
in addition to the primary surpluses and results in a higher fiscal
backing. In the following proposition, we generalize Proposition 8.2
to incorporate the convenience yields.

Proposition 8.3. When there is convenience yield on government debt and


possible government default, the market value of government debt is equal
to the present value of future government surpluses and future seigniorage
revenues plus a transversality term:

∞ ∞
" # " #
Dt = E t ∑ exp( Mt,t+k )St+k + Et ∑ exp( Mt,t+k )Kt+k + lim Et [exp( Mt,t+k ) Dt+k ] .
k→∞
k =1 k =0
(8.11)

The proof is given in Appendix A.32. In this formula, while the


sum of government surpluses starts from period t + 1 to back up the
debt outstanding at the end of period t, the sum of the seigniorage
revenues starts from period t. This is because the seigniorage revenue
Kt results from the fact that the government can pay back its debt in
period t + 1 with a lower interest, which is tied to λt (h), the conve-
nience yield in period t. Since these additional proceeds are due to
investors’ willingness to accept lower returns to hold the government
debt in exchange for its liquidity and safety, we can also interpret the
seigniorage revenue term as the valuation of the government debt’s
service flows. When the service flows become more valuable, the
value of government debt appreciates and results in a higher return
RtD .

8.B.2 Fiscal Trade-Off in the Risk Domain


The intertemporal government budget condition also has implica-
tions for how the government trades off the risk that it transfers to
the taxpayers and the debtholders. Recall that the cum-dividend debt
return is

∑∞
h=1 Pt+1 ( h − 1) Qt ( h )
exp( RtD+1 ) = .
∑∞h=1 Pt ( h ) Qt ( h )

Similarly, we can regard the claims to tax revenues and govern-


ment spending as financial securities, and define their cum-dividend

Electronic copy available at: https://ssrn.com/abstract=4668578


government debt 195

returns as
PtT+1 + Tt+1
exp( RtT+1 ) = ,
PtT
PtG+1 + Gt+1
exp( RtG+1 ) = .
PtG

Moreover, if we regard the equity market return exp( RtM+1 ) as a


systematic risk factor, we can define the debt beta, the tax beta, and
the spending beta as

covt (exp( RtD+1 ), exp( RtM+1 ))


βD
t = ,
vart (exp( RtM+1 ))
covt (exp( RtT+1 ), exp( RtM+1 ))
β Tt = ,
vart (exp( RtM+1 ))
covt (exp( RtG+1 ), exp( RtM+1 ))
βG
t = .
vart (exp( RtM+1 ))

Then, we obtain the following result characterizing these claims’


expected returns and systematic risk exposures [Jiang, Lustig, Van Nieuwer-
burgh, and Xiaolan, 2020b].

Proposition 8.4. (a) The expected returns on the aggregate government


debt portfolio, the claim to tax revenues, and the claim to government spend-
ing satisfy

Dt + PtG h i h i PG h i
Et exp( RtT+1 ) = Et exp( RtD+1 ) + t Et exp( RtG+1 ) .
Dt Dt
(b) The betas of the aggregate government debt portfolio, the claim to tax
revenues, and the claim to government spending satisfy

Dt + PtG T PtG G
βt = βD
t + β .
Dt Dt t
The proof is presented in Appendix A.33. This proposition resem-
bles the Modigliani-Miller Theorem in corporate finance, which states
that, within a firm, its equity is a levered claim on the underlying
asset. As a result, the return beta and the risk premium on the firm
asset are equal to the weighted average of those on the firm’s equity
and debt claims:
D+E A E E
βt = βD
t + β ,
D D t
where D is the market value of debt and E is the ex-dividend market
value of equity. Similarly, the government holds the claim to tax
revenues and splits up these cash flows to debtholders and recipients
of government spending. As a result, the return beta and the risk

Electronic copy available at: https://ssrn.com/abstract=4668578


196 jiang – lecture notes on international finance

premium on the tax claim are equal to the weighted average of those
on the government debt and the spending claim.
This result implies a very tight constraint on how the government
insures taxpayers and debtholders. To simplify the argument, we
assume that the taxpayers both pay tax and receive the government
spending as a transfer.
If the government decides to make its debt risk-free, which implies
a zero debt beta β D
t = 0, then, Proposition 8.4 implies

PtG
β Tt = βG
t .
Dt + PtG

With a positive amount of outstanding debt, i.e., Dt > 0, and a pro-


cyclical tax claim, i.e., β Tt > 0, then, the spending beta has to be even
higher than the tax beta. That is, the government spending has to
be more procyclical than the tax revenue in order to manufacture
risk-free government debt. In other words, the risk-free debt is manu-
factured by transferring more risk to the taxpayers.
Conversely, insuring taxpayers by lowering tax or raising govern-
ment spending in bad times implies high debt beta or low spending
beta, which renders the government debt risky by requiring a high
debt beta.
Taking a step back, this trade-off between insuring the taxpayers
and insuring the debtholders is a manifestation of the resource con-
straint. There are only so many resources that the government can
use during a recession. If it allocates these resources to the debthold-
ers, then the debt value is preserved in recessions at the expense of
the taxpayers. Conversely, if it allocates these resources to the taxpay-
ers, then the debt value has to go down in recessions.
Moreover, we can extend this trade-off result to the case of con-
venience yields. In this case, the debt return can be expressed as a
weighted average of the returns on the tax, spending, and seigniorage
claims:

Dt + 1 + S t +1 P T + PtK+1 + Kt+1 − PtG+1 + Tt+1 − Gt+1


exp( RtD+1 ) = = t +1
Dt Dt
PtT P K P G
= exp( RtT+1 ) + t exp( RtK+1 ) − t exp( RtG+1 ), (8.12)
Dt Dt Dt

where PtK denotes the ex-dividend present value of the seigniorage


claim.
While this formula looks like a simple extension of the case with-
out convenience yield, there is a subtle difference. Proposition 8.3
implies that

Dt = PtT + PtK + Kt − PtG .

Electronic copy available at: https://ssrn.com/abstract=4668578


government debt 197

We have an additional Kt term since the sum of the seigniorage rev-


enue starts from t = 0 in the intertemporal government budget
condition (8.11), which means that, unlike the case without conve-
nience yields, the weights on the right-hand side of Eq. (8.12) do not
sum up to 1:

PtT PK PG Kt
+ t − t = 1− < 1.
Dt Dt Dt Dt
This is a feature, not a bug. Suppose the tax, spending, and seignior-
age claims all have the same discount rate. Then, as long as the
weights sum up to 1, a linear combination of these claims also has
the same discount rate. The presence of the bond convenience yield
introduces a “missing weight,” which allows the debt to have a lower
discount rate RtD+1 . More precisely, we can write Eq. (8.12) as
!
D Kt PtT T PtK K PtG
exp( Rt+1 ) 1 + T = exp ( R t + 1 ) + exp ( R t + 1 ) − exp( RtG+1 ).
Pt + PtK − PtG PtT + PtK − PtG PtT + PtK − PtG PtT + PtK − PtG

On the left-hand side, the debt return is adjusted by a factor of


Kt /( PtT + PtK − PtG ), which reflects the size of the convenience yield.
On the right-hand side, the weights are normalized to sum up to 1.
When the convenience yield is higher and generates a high seignior-
age revenue Kt , the debt return RtD+1 can be lower than the weighted
average of the tax, spending, and seigniorage returns.

Proposition 8.5. (a) The expected returns on the aggregate government


debt portfolio, the claims to tax revenues, government spending, and
seigniorage revenues satisfy
!
h
D
i Kt PtT h
T
i PtK h
K
i
Et exp( Rt+1 ) 1 + T = E t exp ( R t + 1 ) + E t exp ( R t + 1 )
Pt + PtK − PtG PtT + PtK − PtG PtT + PtK − PtG
PtG h
G
i
− T E t exp ( R t +1 ) .
Pt + PtK − PtG
(b) The betas of the aggregate government debt portfolio and the claims to
tax revenues, government spending, and seigniorage revenues satisfy
!
K t PtT PtK PtG
βDt 1+ T = β T
t + β K
t − βG
t .
Pt + PtK − PtG PtT + PtK − PtG PtT + PtK − PtG PtT + PtK − PtG

The proof is presented in Appendix A.34. Part (a) shows that


the expected return on the debt is tied to a weighted average of the
expected returns on the tax, spending, and seigniorage claims. How-
ever, the presence of convenience yield introduces a wedge that is
increasing in the convenience yield and hence the seigniorage rev-
enue Kt : a higher convenience yield allows the discount rate on the
debt to be lower than the weighted average.

Electronic copy available at: https://ssrn.com/abstract=4668578


198 jiang – lecture notes on international finance

However, on the risk domain, we still obtain a tight constraint on


how the government insures the taxpayers and the debtholders. As
a special case, suppose the government commits to keeping its debt
risk-free, i.e., β D
t = 0. Then, part (b) simplifies to

PtT PtK PtG


0= G
β Tt + T G
βKt − T βG
t ,
PtT K
+ Pt − Pt K
Pt + Pt − Pt Pt + PtK − PtG
which means that a higher convenience yield per se does not help
with insuring the taxpayers. Instead, a countercyclical convenience
yield is needed to allow the government to impose lower taxes and
insure the taxpayers in high marginal utility states.
What we derived above is a positive statement of the trade-off that
the government faces between insuring the taxpayers and insuring
the debtholders. This trade-off in itself is silent about what is opti-
mal. We need to develop specific models of the government’s objec-
tive function and the taxpayers’ and the bondholders’ preferences to
answer the normative question. In some cases, this normative ques-
tion is not only economic but also political. As Keynes [1926] noted
in the context of France’s debt payments in the interbellum period,
The level of the franc is going to be settled in the long run, not by speculation
or the balance of trade, or even the outcome of the Ruhr adventure, but by the
proportion of his earned income which the French taxpayer will permit to be
taken from him to pay the claims of the French rentier...
Now it is obvious that there are two methods of attaining the desired equilib-
rium. You can increase the burdens on the taxpayer, or you can diminish the
claims of the rentier. If you choose the first alternative, taxation will absorb
nearly a quarter of the national income of France...
Since this by itself is not enough, your next business—provided you accept
my conclusion as to the mind of the French public—is to consider coolly how
best to reduce the claims of the rentier. Three methods offer themselves: first,
a general capital levy; second, a forced reduction of the rate of interest on the
public debt; third, a rise of prices which would reduce the real value of the
rentier’s money claims.

8.B.3 Connecting the Aggregate to the Disaggregate Bond Prices


Propositions 8.2 and 8.3 describe the valuation Dt of the entire gov-
ernment debt portfolio, defined as the sum of individual government
bonds’ values:
H
Dt = ∑ Qt (h) Pt (h).
h =1

Let us first consider the determination of these individual gov-


ernment bonds’ values. We continue to assume that the government
bonds are denominated in local currencies, promise nominal pay-
ments, and could potentially default. As the government bonds are

Electronic copy available at: https://ssrn.com/abstract=4668578


government debt 199

defaultable, their yields are not necessarily the same as the nomi-
nal risk-free rates, which we denote by it (h). Let us use itTreas (h) to
denote the yield of the government debt with maturity h, which satis-
fies
def 1
itTreas (h) = − log Pt (h).
h
This government debt yield can be decomposed into a risk-free
component, a default spread component, and a convenience yield
component:

itTreas (h) = ρt (h) + δt (h) − ct (h),

where the risk-free component reflects the benchmark discount rate


implied from the SDF:

def 1
ρt (h) = − log Et [exp( Mt,t+h )],
h
the default spread component captures the risk-neutral expectation of
sovereign default,
" # !
h
def 1
δt (h) = − log Et exp( Mt,t+h ) ∏ (1 − χt+ j ) − log Et [exp( Mt,t+h )] ,
i
h j =1

and the convenience yield component captures the wedge between


the bond yield and the yield of a hypothetical bond with the same
default spread but no Euler equation wedge [Jiang, Lustig, Van Nieuwer-
burgh, and Xiaolan, 2020c]:
" # " #!
h h
def 1
ct ( h) = log Et exp( Mt,t+h ) ∏ (1 − χt+ j ) exp(λt+ j−1 (h − j + 1)) − log Et exp( Mt,t+h ) ∏ (1 − χt+ j )
i i
.
h j =1 j =1

Moreover, while the risk-free rate is risk-free in local currency


units, it could carry a currency risk premium. For the case of h = 1,
according to Section 1.B applied to nominal exchange rates Et and
interest rates ρt and ρ∗t , under joint normality,

1
ρt = ρ∗t − Et [∆Et+1 ] − vart (∆Et+1 ) − covt ( Mt∗+1 , ∆Et+1 ),
2
which implies that the nominal home risk-free rate ρt relative to a
foreign benchmark ρ∗t is decreasing in the home currency’s expected
nominal appreciation Et [∆Et+1 ], and increasing in the home cur-
rency’s risk premium RPt = −covt ( Mt∗+1 , ∆Et+1 ) − 12 vart (∆Et+1 ).2 As 2
The longer-term nominal rates have
such, government bonds compensate investors for not only credit risk similar expressions that contain the
marginal utility growth and the ex-
but also exchange rate risk. change rate movement from period t to
How do we connect this determination of individual bond prices t + h.

to the valuation of the aggregate government debt portfolio? While

Electronic copy available at: https://ssrn.com/abstract=4668578


200 jiang – lecture notes on international finance

the intertemporal government budget condition in Propositions 8.2


and 8.3 imposes a restriction on the aggregate government debt port-
folio, the government can still carve out different types of liabilities
subject to this restriction in aggregate. This is again similar to the
corporate finance context, in which the company with a certain type
of assets can issue different types of equity and debt liabilities subject
to the Modigliani-Miller restriction. That said, when the government
surpluses and seigniorage revenues fluctuate, the valuation of the
individual bonds needs to respond, resulting in changes in any of the
risk-free rate ρt , the default spread δt , and the convenience yield ct
components.

8.C The Transversality Condition

All of our results above rely on the transversality condition (8.8),


reproduced below:

lim Et [exp( Mt,t+k ) Dt+k ] = 0,


k→∞

which states that the discounted value of future government debt


has to converge to zero. If this condition does not hold, then, the
market value of debt could exceed the present value of government
surpluses, and the gap could potentially be infinite. In this section,
we provide two perspectives on this condition.

8.C.1 The Transversality Condition as Discounted Debt Strip


To evaluate this condition, we may think of a hypothetical financial
claim whose payoff at time t + k is equal to the market value of out-
standing government debt in that period. Using the language from
the literature that studies dividend strips, we may call this claim the
debt strip. Importantly, this debt strip does not pay out the promised
payment of a specific debt, but instead pays out the market value of
outstanding debt. In other words, holders of this claim are betting on
a greater amount of debt outstanding.
In a large class of macro models, the amount of outstanding gov-
ernment debt Dt is cointegrated with the GDP Yt . The implicit as-
sumption is either that the government manages its debt responsibly
and does not allow the size of debt to explode, or that the investors
impose discipline so that market value of debt adjusts as the notional
amount of borrowing goes up. In either case, the risk premium on
the long-term debt strip converges to the risk premium on the long-
term GDP claim. Given that the GDP is risky, the risk premium is
high enough in many macro models to guarantee that the discounted
value of long-term GDP strip converges to zero. As a result, the value

Electronic copy available at: https://ssrn.com/abstract=4668578


government debt 201

of debt strip also converges to zero, and the transversality condition


holds.

8.C.2 The Transversality Condition as Household Optimality Condi-


tion
The transversality condition is a necessary condition for the optimal-
ity of (long-lasting) households’ decisions. To derive this result, we
consider a set-up similar to the home country in Chapter 1, in which
the households maximize expected lifetime utility:


" #
E0 ∑ δt u(ct ) .
t =0

Proving the transversality condition in a general way is technically


difficult. Here we consider a much narrower class of utility func-
tions, which is adapted from Brock [1982]: the utility function u(c)
is not only monotone increasing in c and concave, but also bounded
from below. Without loss of generality, we can normalize the utility
level and assume u(c) ≥ 0. For simplicity in exposition, we also as-
sume that the price level is fixed, so that the nominal risk-free debt
is also risk-free in real terms. For interested readers, Ekeland and
Scheinkman [1986] provides a more general characterization.
The households can trade the outstanding government debt. Let
dt denote the households’ government debt portfolio and let exp(rtD )
denote its nominal return.
def
wt = ct + dt = yt + dt−1 exp(rtD ).

The first-order condition (Euler equation) w.r.t. dt−1 is given by


h i
u′ (ct−1 ) = Et−1 δu′ (ct ) exp(rtd ) . (8.13)

The households are also subject to a no-short-sale constraint dt ≥ 0


for all t. In general, we need a lower bound on the households’ asset
position, which is not necessarily zero, so that the households cannot
borrow an infinite amount of money for consumption.
We define the households’ value function in period 0 as


" #
ψ(wt , t) = max E0 ∑ δs u(cs ) ,
s=t

which can be shown to be concave and differentiable [Benveniste and


Scheinkman, 1979, Brock, 1982].
Given this set-up, let us first consider the special case in which
the households have a finite horizon T. Then, in the last period, it is
in the households’ interest to consume all of their savings instead of

Electronic copy available at: https://ssrn.com/abstract=4668578


202 jiang – lecture notes on international finance

leaving money on the table, as the remaining wealth will be wasted.


Moreover, the no-short-sale constraint also implies that the house-
holds cannot finish with negative wealth. Therefore, we can express
the finite-period transversality condition as
h i
E0 δ T u′ (c T )d T = 0.

The infinite-horizon case has a similar interpretation, though its


derivation is more complicated. The following proposition shows
that the transversality condition is a necessary condition for the opti-
mality of households’ decisions.

Proposition 8.6. Assume that the value function satisfies limt→∞ ψ(wt , t) =
0. Then, the optimal solution {ct }∞ ∞
t=0 , { dt }t=0 to the households’ problem
implies the Euler equation (8.13) and the transversality condition

lim E0 δt u′ (ct )dt = 0.


 
t→∞

The proof is in Appendix A.35. This proposition shows that the


households optimally choose not to “leave money on the table” in
the infinite horizon, which means to increase their savings dt in the
government bond portfolio at a rate higher than their discount rate
δt u′ (ct ). The households do not borrow an infinite amount of money
by short-selling the government debt, either, which is ruled out by
the no-short-sale constraint. This result emphasizes that the transver-
sality condition is usually imposed by the households’ optimal deci-
sion, just like the Euler equation (8.13).
Moreover, while this set-up considers only one type of assets in
which households can deposit their savings, the extension to mul-
tiple assets is straight-forward. Let at denote the households’ total
savings and dt denote the households’ holdings of one asset that we
denote as the government debt. Then, our derivation shows that the
transversality condition holds at the portfolio level:

lim E0 δt u′ (ct ) at = 0.
 
t→∞

Now, suppose the households’ transversality condition for the


government debt is violated, i.e.,

lim E0 δt u′ (ct )dt > 0.


 
t→∞

This implies that the households’ holdings in other assets are nega-
tive in the infinite horizon, i.e.,

lim E0 δt u′ (ct )( at − dt ) < 0,


 
t→∞

which violates the no-short-sale constraint. So, the transversality


condition also holds at the level of individual assets.

Electronic copy available at: https://ssrn.com/abstract=4668578


government debt 203

Does the presence of bond convenience yields affects the transver-


sality condition? Following Section 4.A, we assume that the house-
holds also derive utility directly from holding the government bonds.
The utility function is


" #
E0 ∑ δt (u(ct ) + vt (dt )) .
t =0

We use λt to denote the convenience yields at time t, which, as we


see in Section 4.A, is determined by the ratio between the marginal
utility of bond holding and the marginal utility of consumption:

v′t (dt )
exp(−λt ) = 1 − .
u′ (ct )

Proposition 8.7. In the presence of the bond convenience yield, the optimal
solution {ct }∞ ∞
t=0 , { dt }t=0 to the households’ problem implies the Euler
equation (8.13) and the transversality condition

lim E0 δt u′ (ct )dt exp(−λt ) = 0.


 
t→∞

When the convenience yield λt is bounded, we recover the original


transversality condition:

lim E0 δt u′ (ct )dt = 0.


 
t→∞

The proof is in Appendix A.36. When the bond convenience yield


λt is bounded, we recover the original transversality condition in
the presence of the bond convenience yield. In other words, while
the bond convenience yield can increase the fiscal backing of the
government debt as we show in Proposition 8.3, it does not affect the
transversality condition.

8.C.3 When Does the Transversality Condition Fail?


The recent literature sees a revival of the idea that the transversality
condition may fail under certain conditions, which is used to justify
why the market value of government debt could exceed the govern-
ment’s fiscal backing, i.e., the present value of its primary surpluses
plus additional seigniorage revenue.
First, in overlapping generation (OLG) models, there are no long-
lived investors who impose the transversality condition as an opti-
mality condition as we discussed in Proposition 8.6 and Proposition
8.7. Then, the valuation of the government debt, which is usually the
only long-lasting asset in the economy, could be infinite. Blanchard
[2019] offers one such example.

Electronic copy available at: https://ssrn.com/abstract=4668578


204 jiang – lecture notes on international finance

Second, the transversality condition can fail when evaluated at


the “aggregate SDF” in incomplete markets where the representa-
tive agent does not exist. Brunnermeier, Merkel, and Sannikov [2022]
present one example, in which individual households face idiosyn-
cratic risks, and the transversality condition still holds at their indi-
vidual SDFs and their actual cash flows from trading the government
debt. In fact, not only does each individual household’s SDF look
different because of their idiosyncratic risks, their cash flows from
trading the government debt is also different from the aggregate pri-
mary surpluses because they dynamically buy and sell government
debt in response to their idiosyncratic shocks. These cash flows allow
the individual households to partially hedge against their idiosyn-
cratic shocks, and can be therefore interpreted as service flows.

8.D An Example Economy

The results we derived so far are general. Next, we illustrate their


economic meanings in the context of a simple example economy,
which largely follows from Jiang, Lustig, Van Nieuwerburgh, and Xi-
aolan [2020b]. In this economy, we only consider one-period risk-free
debt. Yet, the government surplus and the quantity of government
debt are risky and cointegrated with GDP.
Let Yt denote the nominal GDP. The GDP growth is i.i.d. and
permanent:

∆ log Yt+1 = X + σε t+1 ,

where ε t+1 denotes the innovation to GDP growth that is i.i.d. nor-
mally distributed with mean zero and standard deviation one.
The log nominal SDF is driven by the same GDP shock

1
Mt,t+1 = −i − γ2 − γε t+1 ,
2
which implies that a lower GDP growth leads to a higher marginal
utility. By the Euler equation for risk-free debt, the one-period log
risk-free rate is a constant: R f = i.
Given these specifications, the present value of a GDP strip k peri-
ods from now is
 
1
Et [exp( Mt,t+k )Yt+k ] = exp k( X − i + σ2 − γσ) Yt ,
2

which converges to zero as k → ∞ if the discount rate is higher than


the GDP growth rate, i.e.,

1
i + γσ > X + σ2 .
2

Electronic copy available at: https://ssrn.com/abstract=4668578


government debt 205

We note that the discount rate is the benchmark risk-free rate i im-
plied from the SDF plus the risk premium term γσ. The growth rate
is the log growth rate X plus the Jensen’s term 21 σ2 .
We assume that the government commits to issuing one-period
risk-free debt. The government also commits to a simple fiscal policy,
with a constant spending/GDP ratio γ̄ = Gt /Yt and a constant
debt/GDP ratio δ̄ = Dt /Yt .
Then, the government budget constraint (8.1) implies a counter-
cyclical process for the surplus/GDP ratio:
St Tt − Gt Dt D Y
= =− + exp( R f ) t−1 t−1 = −δ̄ (1 − exp (−( X − i + σε t ))) .
Yt Yt Yt Yt−1 Yt
This result implies that ∂(St /Yt )/∂ε t < 0. Therefore, to insure the
debtholders by keeping its debt risk-free, the government must gen-
erate counter-cyclical primary surpluses. In particular, when the GDP
declines (ε t < 0) in recessions, the tax revenue needs to increase as
a fraction of GDP or the government spending needs to decrease.
The magnitude of the required counter-cyclical response in primary
surpluses is increasing in the debt/GDP ratio δ̄.

8.D.1 The Transversality Condition


We can evaluate the transversality condition in this fully specified
model. In particular, we can show that if the risk premium is high
enough to guarantee a finite valuation on the GDP claim, then, the
transversality condition holds:

lim Et [exp( Mt,t+k ) Dt+k ] = lim Et δ̄ exp( Mt,t+k )Yt+k = 0.


 
k→∞ k→∞

Given the transversality condition holds, the government debt


value is the sum of the values of the surplus strips:

" #
Dt = E t ∑ exp( Mt,t+k )St+k = δ̄Yt ,
k =1

which confirms the assumption that the debt/GDP ratio is a constant.

8.D.2 Risks and Risk Premia


How do we reconcile the observation that the government surplus St
and the market value of government debt Dt are both stochastic with
the observation that the debt is risk-free? When the transversality
condition is satisfied, the debt return innovation reflects news about
the present discounted value of future government surpluses:

" #
Dt (Et+1 − Et ) exp( RtD+1 ) = (Et+1 − Et ) ∑ exp( Mt+1,t+ j )St+ j .
j =1

Electronic copy available at: https://ssrn.com/abstract=4668578


206 jiang – lecture notes on international finance

When the debt is risk-free, there is no uncertainty about the debt


return: (Et+1 − Et ) exp( RtD+1 ) = 0. Then, there is no news about the
present value of current and future surpluses; technically, we can say
that the present value of current and future surpluses in period t + 1
is measurable in period t:


" #
(Et +1 − Et ) ∑ exp( Mt+1,t+ j )St+ j = (Et+1 − Et ) [ Dt+1 + St+1 ] = 0.
j =1

In other words, the government surplus St+1 is engineered such that


it exactly offsets the variation in the market value of government
debt Dt+1 , which allows the debt to offer a constant risk-free return.
However, as both government debt strip Dt+1 and surpluses St+1 are
stochastic, their discount rates need to reflect their risk premia. In
particular, as both government surplus and the debt strip is cointe-
grated with the GDP, the long-term surplus claim or debt claim has
the same risk premium as the GDP risk premium.

8.D.3 The r < g Case


A particularly interesting case in this model is when the risk-free rate
is lower than the average GDP growth rate. In our notation, i < X. In
this case, if the Jensen’s term 12 σ2 is small enough so that i + 12 σ2 < X,
then, the government can sustain deficits on average:
    
St 1
E = −δ̄ 1 − exp − X + i + σ2 < 0.
Yt 2

That said, when the fundamental shock is negative enough (X − i <


−σε t ), the government still needs to run a primary surplus.
Even though the government runs a deficit on average, the transver-
sality condition can be satisfied if the risk premium γσ on the output
claim is large enough:

1
i + γσ > X + σ2 .
2
In this case, discounting growing surpluses and future debt at the
risk-free rate i < X would not produce the right answer for the
valuation of the current debt.
However, this is not a free lunch: the beta constraint characterized
by Proposition 8.4 still binds: in order to keep debt risk-free, the
government needs to raise taxes or lower spending in high marginal
utility states.
In other words, the average deficits reflect an insurance premium
that the government earns by extracting countercyclical cash flows
from the taxpayers. This risk-based view is very different from the

Electronic copy available at: https://ssrn.com/abstract=4668578


government debt 207

Blanchard [2019] interpretation of r < g. From our perspective, it is


precisely the positive covariance between the marginal utility and the
government surpluses that creates the fiscal backing to manufacture
risk-free debt. The r < g condition is neither necessary nor sufficient
to gauge fiscal backing.

8.E The Public Debt Valuation Puzzle

When the transversality condition (8.8) holds, the intertemporal gov-


ernment budget condition (8.6) can be expressed as
∞ ∞
" # " #
Dt = E t ∑ exp( Mt,t+k ) Tt+k − Et ∑ exp( Mt,t+k )Gt+k . (8.14)
k =1 k =1

A fundamental question in finance and macroeconomics is whether


the present values of government surpluses on the right-hand side of
this equation matches the market value of government debt on the
left-hand side. To evaluate the present values on the right-hand side
of this equation, we need to know (i) the expected future tax rev-
enues and government spending and (ii) the appropriate discount
rates that we assign to these fiscal cash flows. To figure out the ex-
pected fiscal cash flows, the data since WW-II suggest that the U.S. on
average has similar levels of tax revenue and government spending
as fractions of the GDP. If we zoom into the past 15 years, govern-
ment spending surpassed tax revenue by a large margin. Feeling
optimistic, let us assume the U.S. fiscal cash flows will revert to the
historical norm over time, which maintains roughly equal tax rev-
enues and government spending on average.
To figure out the appropriate discount rates, we note that the stan-
dard risk-free rates are not the right ones to use. Riskier cash flows
need to be discounted at higher rates, and the riskiness is deter-
mined by how the cash flows comove with the SDF Mt,t+k . Consider
the stock market for example. As corporate revenues decline dra-
matically in high marginal utility states, investors regard stocks as
risky assets and therefore require a high compensation to hold them.
This is why stocks tend to have higher returns than risk-free bonds
over a long enough time period, and why, for the purpose of valua-
tion, corporate cash flows are discounted at much higher rates than
the standard interest rates. The data since WW-II suggest that the
U.S. tax revenue also exhibits a cyclical behavior, which warrants
high discount rates. In comparison, the U.S. government spending is
counter-cyclical: it tends to increase during recessions, as unemploy-
ment benefits and other welfare payments tend to be higher in the
downturns. The government may also decide to spend more on pub-
lic projects to stimulate the economy. As recessions represent high

Electronic copy available at: https://ssrn.com/abstract=4668578


208 jiang – lecture notes on international finance

marginal utility states, the U.S. government spending deserves lower


discount rates.
To put everything together, Eq. (8.14) computes the difference
between the expected tax revenue and government spending, dis-
counted at their appropriate discount rates. These cash flows have
similar levels on average, but the tax revenue has higher discount
rates than the government spending. Then, the present value of the
tax cash flows should be lower than that of the spending cash flows,
implying a negative present value of government surpluses [Jiang,
Lustig, Van Nieuwerburgh, and Xiaolan, 2019]. This negative valua-
tion of government surpluses on the right-hand side of Eq. (8.14) is
at odds with the observed positive market value of government debt
on the left-hand side of Eq. (8.14). This gap suggests that the U.S.
government may not have enough fiscal backing for its outstanding
government debt. We call it the U.S. public debt valuation puzzle.
Related to this puzzle, a salient fact is that the U.S. sovereign credit
default swap has high premium since the financial crisis, reflecting
a rising risk-adjusted probability of U.S. sovereign default [Cher-
nov, Schmid, and Schneider, 2020, Augustin, Chernov, Schmid, and
Song, 2021, Liu, Schmid, and Yaron, 2020]. This analysis is based on
a large literature on affine sovereign default models [Pan and Single-
ton, 2008, Longstaff, Pan, Pedersen, and Singleton, 2011] and a large
literature on strategic sovereign default [Eaton and Gersovitz, 1981,
Arellano, 2008]. While they are also important issues related to gov-
ernment debt and fiscal sustainability, they are beyond the scope of
this note.
So far, we have been working with a very general argument that
is consistent with a large number of models with a government and
infinitely lived households. More concretely, we can implement this
valuation exercise in specific models of cash flows and discount rates.
Below, we discuss several valuation techniques that have been used.
The results emphasize that the U.S. finances itself at not only favor-
able rates, reflecting the convenience yield idea discussed in Chapters
4 and 7, but also favorable quantities beyond its fiscal fundamentals.

8.E.1 Affine Valuation Models


To evaluate the present value of government surpluses, i.e.,


" #
Et ∑ exp( Mt,t+k )(Tt+k − Gt+k ) ,
k =1

we need to model the joint dynamics of the SDF Mt,t+k and the fiscal
cash flows Tt+k and Gt+k . Jiang, Lustig, Van Nieuwerburgh, and
Xiaolan [2019] adopts the affine term structure framework, which

Electronic copy available at: https://ssrn.com/abstract=4668578


government debt 209

has been used in various asset pricing settings to understand stock


and bond prices [Duffie and Kan, 1996, Dai and Singleton, 2000, Ang
and Piazzesi, 2003, Lustig, Van Nieuwerburgh, and Verdelhan, 2013,
Chernov and Creal, 2023].
Specifically, let zt denote the vector of state variables demeaned by
their average value, which includes the interest rates, stock dividends
and price/dividend ratios, and the fiscal cash flows. This vector
follows a Gaussian first-order VAR:
1
zt+1 = Ψzt + ut+1 = Ψzt + Σ 2 ε t+1 ,

with the companion matrix Ψ and homoscedastic innovations ut+1 ∼


i.i.d. N (0, Σ). The Cholesky decomposition of the covariance matrix,
1
 1 ′
Σ = Σ 2 Σ 2 , has non-zero elements on and below the diagonal, so
that ε t+1 ∼ i.i.d. N (0, I ).
This VAR setting allows us to model expectations of cash flows.
We also choose a flexible SDF that assumes no arbitrage. The log
nominal SDF Mt+1 is conditionally normal:
1
Mt+1 = −it (1) − Λ′t Λt − Λ′t ε t+1 .
2
The real SDF is mt+1 = Mt+1 + πt+1 , which is also conditionally
normal. The priced sources of risk are the same as the innovations in
the state vector ε t+1 . These innovations are associated with a market
price of risk vector Λt of the affine form:

Λ t = Λ0 + Λ1 z t ,

where Λ0 collects the average prices of risk and Λ1 governs the time
variation in risk premia. Asset pricing in this model amounts to
taking a stance on the market prices of risk in Λ0 and Λ1 .
If the state vector zt contains the short rate, inflation, GDP growth,
the stock dividend/GDP ratio, and the stock price/dividend ratio,
this SDF generates affine solutions for the value of the aggregate
stock market and the term structure of nominal and real interest
rates. We also include fiscal variables in the state vector in order to
price the fiscal cash flows.
def
Let it (h) = (1/h) log Pt (h) denote the nominal risk-free rate
with maturity h. For notational convenience, let ei denote the vector
(0, . . . , 1, . . . , 0)′ that selects the variable corresponding to the short
rate, and let i0 (1) denote the mean value of the short rate. Then,

it (1) = ei′ zt + i0 (1).

Similarly, let e∆d , ex , and eπ denote vectors that select the variables
corresponding to the growth in the dividend/GDP ratio, the real
GDP growth, and the inflation, respectively.

Electronic copy available at: https://ssrn.com/abstract=4668578


210 jiang – lecture notes on international finance

Let Divt denote the stock dividend strip. The value of the divi-
dend strip of horizon h is defined as

Ptm (h) = Et [exp( Mt,t+h ) Divt+h ],

and let pdm


t ( h ) denote the price-dividend ratio of the stock market:

pdm m
t ( h ) = log( Pt ( h ) /Divt ).

Then, the cum-dividend valuation of the stock market can be ex-


pressed as

PVtm = Divt ∑ exp( Am (h + 1) + Bm (h + 1)′ zt ).
h =0

g
Similarly, let Ptτ (h) and Pt (h) denote the price of the tax and
g
spending strips of horizon h, and let pdτt (h) and pdt (h) denote the
price-dividend ratios of the tax and spending strips.

Proposition 8.8. (a) The nominal yield curve is affine in the state vector:

A(h) B( h)′
it ( h) = − − zt ,
h h

where the coefficients A(h) and B(h) satisfy the following recursions:

1 1
A(h + 1) = −i0 (1) + A(h) + B(h)′ ΣB(h) − B(h)′ Σ 2 Λ0 ,
2
1
B(h + 1)′ = −ei′ + B(h)′ (Ψ − Σ 2 Λ1 ),

initialized at A(0) = 0 and B(0) = ⃗0.


(b) The dividend strip’s log price-dividend ratio is also affine in the state
vector:


pdm m m
t ( h) = A ( h) + B ( h) zt ,

where
1
Am (h + 1) = −i0 (1) + Am (h) + x0 + π0 + ( Bm (h) + e∆d + ex + eπ )′ Σ( Bm (h) + e∆d + ex + eπ )
2
1
− ( Bm (h) + e∆d + ex + eπ )′ Σ 2 Λ0 ,
1
Bm (h + 1)′ = −ei′ + ( Bm (h) + e∆d + ex + eπ )′ (Ψ − Σ 2 Λ1 ).

(c) The tax and spending strips’ log price-dividend ratios are also affine in
the state vector:

pdτt (h) = Aτ (h) + Bτ (h)′ zt ,


g
pdt (h) = A g (h) + B g (h)′ zt ,

Electronic copy available at: https://ssrn.com/abstract=4668578


government debt 211

where
1
Aτ (h + 1) = −i0 (1) + Aτ (h) + x0 + π0 + ( Bτ (h) + e∆τ + ex + eπ )′ Σ( Bτ (h) + e∆τ + ex + eπ )
2
1
− ( Bτ (h) + e∆τ + ex + eπ )′ Σ 2 Λ0 ,
1
Bτ (h + 1)′ = −ei′ + ( Bτ (h) + e∆τ + ex + eπ )′ (Ψ − Σ 2 Λ1 ),
1
A g (h + 1) = −i0 (1) + A g (h) + x0 + π0 + ( B g (h) + e∆g + ex + eπ )′ Σ( B g (h) + e∆g + ex + eπ )
2
1
− ( B g (h) + e∆g + ex + eπ )′ Σ 2 Λ0 ,
1
B g (h + 1)′ = −ei′ + ( B g (h) + e∆g + ex + eπ )′ (Ψ − Σ 2 Λ1 ).

The proof is in Appendix A.37. Our key assumption behind this


proposition is that the SDF that prices stocks and bonds can also
price the fiscal cash flows. As we discussed earlier in Eq. (8.10), any
investors who can trade the aggregate government debt portfolio
have to agree on how they price the primary surpluses. Moreover,
by assuming that there is no market segmentation between the stock
and bond markets, we allow the market price of permanent risks
implied from the stock market to also inform the pricing of the fiscal
cash flows. In other words, we set up the VAR dynamics to allow
cointegration between the GDP, the aggregate stock market dividend,
and the government tax and spending strips. As a result, the pricing
of the long-term fiscal cash flows is governed by the same market
prices of risk that price the long-term stock dividends.
To develop this argument further, let us consider the pricing of
infinite-horizon cash flows. In the limit, the nominal yield converges
to a certain value, with
1
B(∞)′ = B(∞)′ Ψ − ei′ − B(∞)′ Σ 2 Λ1 ,

which implies
 1
 −1
B(∞)′ = −ei′ I − (Ψ − Σ 2 Λ1 ) .

1
It is useful to interpret (Ψ − Σ 2 Λ1 ) as the risk-neutral transition ma-
trix, and then the risk loadings of the infinite-horizon bond are given
by the loadings of the one-period bond, ei , multiplied with the Leon-
1
tief inverse of (Ψ − Σ 2 Λ1 ). The same Leontief inverse operator also
appeared in the trade network model in Section 3.C, which captures
the effects of higher-order connections in the trade network.
Following this derivation, we can now express the risk premium
of the long-term bond and, in a similar way, the risk premia of stock
dividend and fiscal cash flows in the infinite horizon.

Proposition 8.9. (a) In the infinite horizon, the long-run nominal yield and

Electronic copy available at: https://ssrn.com/abstract=4668578


212 jiang – lecture notes on international finance

bond expected return are


1 ′ ′ 1
lim it (h) = i0 (1) − Btrans ΣBtrans + Btrans Σ 2 Λ0 ,
h→∞ 2
1 ′ ′ 1 ′
lim Et [log Pt+1 (h − 1) − log Pt (h)] = i0 (1) − Btrans ΣBtrans + Btrans Σ 2 Λ0 + Btrans (Ψ − I )zt ,
h→∞ 2
where
def
 1
 −1

Btrans = B(∞)′ = −ei′ I − (Ψ − Σ 2 Λ1 ) .

(b) In the infinite horizon, the expected returns on the dividend, tax, and
spending strips are
1 1
lim Et [log Ptm+1 (h − 1) − log Ptm (h)] = i0 (1) − B′perm ΣB perm + B′perm Σ 2 Λ0 + B′perm (Ψ − I )zt + (e∆d + ex + eπ )′ zt ,
h→∞ 2
where
def
 1
 −1
B′perm = ( Bm (∞) + e∆d + ex + eπ )′ = (−ei + e∆d + ex + eπ )′ I − (Ψ − Σ 2 Λ1 ) .

The proof is in Appendix A.38. In the language of the permanent


and transitory SDF components that we discussed in Section 3.D, the
SDF we consider here has a permanent component. In the infinite-
horizon limit, it prices the cash flows that contain this permanent
risk, including the GDP strip, the stock dividend strip, the tax strip,
and the spending strip in a coherent fashion, by assigning the same
risk premia to these cash flows. The unconditional expected returns
of these strips are given by
1 1
lim E[log Ptm+1 (h − 1) − log Ptm (h)] − i0 (1) = B′perm Σ 2 Λ0 − B′perm ΣB perm ,
h→∞ 2
(8.15)
1
which is the product of the risk loadings B′perm Σ 2 and the market
prices of risk Λ0 , plus a second-order Jensen’s term.
In contrast, this SDF prices the cash flows that do not have the per-
manent risk, such as the nominal bond, differently. The unconditional
bond risk premia are given by

′ 1 1 ′
lim E[log Pt+1 (h − 1) − log Pt (h)] − i0 (1) = Btrans Σ 2 Λ0 − Btrans ΣBtrans ,
h→∞ 2
(8.16)
1

which is also the product of the risk loadings Btrans Σ 2 and the mar-
ket prices of risk Λ0 , plus a second-order Jensen’s term. Note that Eq.
(8.16) is a special case of Eq. (8.15), obtained when the cash flow does
not grow, i.e., by setting e∆d + ex + eπ = 0.
Figure 8.1 plots the term structures of risk premia for the spend-
ing, tax, equity dividend, and GDP claims, and the long-term nom-
inal bonds. The parameterization is taken from the homoscedastic

Electronic copy available at: https://ssrn.com/abstract=4668578


government debt 213

SDF model in Jiang, Lustig, Van Nieuwerburgh, and Xiaolan [2019].


In the short run, because the spending claim is countercyclical while
the tax and the equity dividend claims are procyclical, the spending
claim has a negative risk premium while the tax and the equity div-
idend claims have positive risk premia. In the long run, because the
spending, the tax, and the equity dividend cash flows are all cointe-
grated with the GDP, they inherit the same risk premium from the
permanent component of the GDP, which converges to about 4% in
the limit. In comparison, the nominal bond’s payoff is not cointe-
grated with the GDP, which is why the long-term nominal bonds
have a different risk premium close to 2% in the limit.

Figure 8.1: The Term Structures of Risk


Premia.
10

-2

-4
0 20 40 60 80 100

This result echoes Proposition 3.11 in Section 3.D, which shows


that the infinite-horizon risk premia of all cash flows that contain
the same permanent component as the SDF will be the same, and
reflect the magnitude of the permanent risk in the SDF, i.e., σP2 . In
comparison, the infinite-horizon nominal bond’s risk premium is
only determined by the magnitude of the transitory risk in the SDF,
and it will be lower if the risk aversion parameter is big enough.

8.E.2 Linearized Intertemporal Budget


Let vt = log( Dt /Yt ) denote the log debt/GDP ratio, and let syt =
St /Yt denote the surplus/GDP ratio. Let e rt+1 = RtD+1 − πt+1 − xt+1
denote the inflation-and-growth-adjusted debt return. Cochrane
[2022] derives a log-linearized return equation implied by the govern-
ment budget constraint:

rt+1 = κvt+1 − vt + st+1 ,


e

Electronic copy available at: https://ssrn.com/abstract=4668578


214 jiang – lecture notes on international finance

where κ = exp(−( R − x − π )) is a constant, and st+ j = κ · syt+ j / exp(v̄)


is κ times the primary surplus/GDP ratio scaled by the average
debt/GDP ratio. This formula is similar to the log-linearization for
stock returns in Campbell and Shiller [1988]:

rt+1 = κ pdt+1 − pdt + ∆dt+1 ,

but it allows us to deal with the possibility that the primary surplus
st can be negative.
By iterating this equation forward T times and taking expectations,
we obtain
" #
T
v t = Et ∑ κ j −1 s t + j − e
r t + j + Et [ κ T v t + T ],

j =1

which is a linearized version of the intertemporal government budget


condition (8.6). It provides a way to evaluate the government debt
as a fraction of GDP, i.e., vt , based on the expectations of future sur-
pluses st+ j , future adjusted debt returns e rt+ j , and future debt/GDP
level vt+T .
This relationship holds both ex-ante and ex-post. Without taking
expectations, we can express this result as a variance decomposition
exercise:
!
T
var (vt ) = cov vt , ∑ κ j −1
+ cov(vt , κ T vt+T ),

st+ j − e
rt+ j
j =1

which states that the debt/GDP ratio varies only because it predicts
future surpluses, future returns, or the future debt/GDP ratio.
If we assume that the transversality condition holds, which in this
context means limT →∞ Et [κ T vt+T ] = 0, then, we obtain

" #
v t = Et ∑ κ j −1 s t + j − e

rt+ j .
j =1

To implement this model, we could use statistical tools (such as the


VAR model or the local projection method) to construct conditional
expectations of st+ j and e
rt+ j and build the debt/GDP value from
these expectations.
This result helps us gauge whether the transversality condition
holds. The conventional wisdom from Bohn [1998] is that a higher
debt/GDP ratio vt is justified by higher surpluses st+ j in the fu-
ture. Jiang, Lustig, Van Nieuwerburgh, and Xiaolan [2021c] use a
simple local projection method to construct conditional expecta-
tions of not only future surpluses and returns (st+ j and e rt+ j ), but
also future debt/GDP ratio (vt+ j ). They find that, after addressing a
small-sample bias, the variations in the current debt/GDP ratio are

Electronic copy available at: https://ssrn.com/abstract=4668578


government debt 215

explained by the variations in the future debt/GDP ratio 10 years


later instead of by future fundamentals (i.e., future surpluses and
returns) in the next 10 years.

8.E.3 Bias in Subjective Expectations


One explanation for the result in Jiang, Lustig, Van Nieuwerburgh,
and Xiaolan [2021c] is biased beliefs. While the econometrician does
not expect higher surpluses or lower returns when the debt/GDP
ratio rises, bond investors may. More precisely, note that the same
linearized decomposition works under the subjective measure, too:
" #
T
v t = Ft ∑ κ j −1
r t + j + Ft [ κ T v t + T ]

st+ j − e
j =1
" # " #!
T T
= Et ∑κ j −1 T
+ Ft [ κ v t + T ] + (Ft − Et ) ∑κ j −1

st+ j − e
rt+ j (st+ j − ert+ j ) .
j =1 j =1

Then, the transversality term Et [κ T vt+T ] under the objective mea-


sure can be expressed as
" #
T
Et [ κ T v t + T ] = Ft [ κ T v t + T ] + (Ft − Et ) ∑ κ j−1 (st+ j − ert+ j ) ,
j =1
h i
and the second term (Ft − Et ) ∑ Tj=1 κ j−1 (st+ j − e
rt+ j ) on the right-
hand side represents the investors’ forecast error. In other words,
what the econometrician perceives as transversality condition viola-
tion can be explained by the investors’ subjective belief errors.
Specifically, if the investors systematically over-predict surpluses
or under-predict returns when the debt/GDP ratio increases, this
forecast error term can impute a unit root in the debt/GDP ratio
under the econometrician’s measure E. At the same time, under
the investors’ subjective beliefs measure F, the debt/GDP ratio may
remain stationary. In this case, we obtain
" #
T
lim Et [κ T vt+T ] = lim (Ft − Et )
T →∞ T →∞
∑ κ j−1 (st+ j − ert+ j ) .
j =1

Let us consider a concrete example that gives a full specification


of the debt dynamics under the two measures. For simplicity, we
abstract away from variations in growth-adjusted returns by setting
rt+ j = 0 under both measures and let κ = 1 which is consistent with
e
the primary surplus being zero on average.
We assume that, under the econometrician’s objective measure E,
the surplus process is stationary and follows an i.i.d. process:

s t = εE
t ,

Electronic copy available at: https://ssrn.com/abstract=4668578


216 jiang – lecture notes on international finance

which implies that the debt/output ratio is a random walk:

v t = v t −1 − ε E
t .

Therefore, under the econometrician’s measure E, the debt/output


ratio satisfies
T
v t = Et ∑ r t + j + Et v t + T = Et v t + T ,

st+ j − e
j =1

which is a unit root process. More precisely, if we repeat the regres-


sions in our baseline result, we would find that the variation in the
debt/output ratio is 100% attributable to the transversality term
Et [vt+T ], and 0% to the variations in the surplus and return processes
due to the i.i.d. assumption.
Under the investors’ subjective measure F, the surplus responds to
debt level:

st = bvt−1 + εFt ,

with a coefficient b > 0. Under this measure, the government


debt/output ratio is priced according to

T T
v t = Ft ∑ r t + j + Ft [ v t + T ] = Ft ∑ bvt+ j−1 + Ft [vt+T ]

st+ j − e (8.17)
j =1 j =1

We conjecture and verify that the law of motion for the debt/output
ratio under the subjective measure F is

vt+1 = ψvt + ηtF+1 .

Plug the investors’ subjective surplus response into the linearized


government budget constraint, we obtain the following solution

ψ = 1−b
ηtF+1 = −εFt .

In other words, due to their biased belief, the investors think that the
debt/output process is stationary with a persistence that is consistent
with their expected response of surplus to the debt/output ratio.
Therefore, if we take Eq. (8.17) to the limit,

T
vt = lim Ft
T →∞
∑ bvt+ j−1 + Tlim
→∞
Ft [ v t + T ],
j =1

we obtain that the variation in the debt/output ratio is 100% at-


tributable to the expected future surplus limT →∞ Ft ∑ Tj=1 bvt+ j−1 , and
0% to the transversality term.

Electronic copy available at: https://ssrn.com/abstract=4668578


government debt 217

These equations constitute a full specification of the debt dynamics


under the two measures, which is mean-reverting according to the
investors and has a unit root according to the econometrician. We can
rearrange the equations and obtain

T
v t = Et v t + T = 0 + (Ft − Et ) ∑ ( s t + j − e
r t + j ),
j =1

which implies that the valuation of government debt is fully driven


by the investors’ belief bias (Ft − Et ), which provides an economic
interpretation of the transversality term Et [vt+T ] under the econome-
trician’s measure.

8.E.4 Dollar-Weighted Return and Government’s Market Timing


Another way to gauge the fiscal capacity is to understand the timing
of the fiscal cash flows. As we discussed in Section 2.B.3, the market
timing of the government issuer, as well as that of the investors in
the Treasury market can be measured by the internal rate of return
(IRR). In the data, the foreign investors have been earning a low IRR,
compared to the IRR that can be achieved from a simple buy-and-
hold strategy [Jiang, Krishnamurthy, and Lustig, 2022a]. The flip side
of this observation is that the U.S. government has been enjoying a
low effective funding rate, thanks mainly to the foreign investors and
the Federal Reserve. In other words, the government tends to time
the market by issuing more debt in recessions when the interest rate
is low and the debt is expensive, and issuing less debt in expansions
when the interest rate is high and the debt is cheap.

8.E.5 International Perspectives


We also note that this public debt valuation puzzle is not necessarily
a universal phenomenon. Chen, Jiang, Lustig, Van Nieuwerburgh,
and Xiaolan [2022] study historical fiscal data in the U.S., the U.K.,
and the Netherlands. Prior to the U.S., the U.K. and the Dutch were
the dominant reserve asset issuers. We learn some common lessons
from their rises and falls: there is usually a unique hegemon who
dominates the supply of reserve assets. The hegemon enjoys a conve-
nience yield on its debt, and issues more debt than is warranted by
its fiscal backing. However, these exorbitant privileges do not last for-
ever: when the hegemon’s fiscal fundamentals deteriorate, investors
eventually withdraw the extra debt capacity, and debt market crash
ensues.
This observation raises the bar for any explanation of the public
debt valuation puzzle, because it needs to answer not only why this

Electronic copy available at: https://ssrn.com/abstract=4668578


218 jiang – lecture notes on international finance

extra fiscal capacity accrues to the U.S. or other global hegemons in


the past, and not to other countries. In this sense, these results call
for an international perspective on fiscal sustainability issues.
On the other hand, this observation connects the public debt val-
uation puzzle to the literature on the coordination on reserve assets
[Farhi and Maggiori, 2018, He, Krishnamurthy, and Milbradt, 2019,
Gopinath and Stein, 2021, Coppola, Krishnamurthy, and Xu, 2023].
This literature points out that there is strategic complementarity in
different investors’ investment decisions. The reserve asset that is
commonly chosen by the investors, which usually coincides with
the hegemon in the economic and political spheres, could enjoy not
only a convenience yield in the price dimension, but also the ability
to issue more debt beyond its fiscal backing in the quantity dimen-
sion. That said, the coordinated outcome could be broken when the
investors become concerned about the hegemon’s fiscal position, as
we have seen in the case of the Dutch and the U.K. Moreover, this
literature also explores the strategic decisions of one or more reserve
asset issuers, who take their market power into consideration when
deciding how much debt to issue [Choi, Kirpalani, and Perez, 2022,
Jiang and Richmond, 2023a].3 Their endogenous issuance decisions 3
More broadly, the strategic coordi-
also affect the valuation of the reserve assets. nation issues between governments
and policymakers enrich the possible
In conclusion, it is worth reflecting upon a quote from Wealth of outcomes in the international economy.
Nations by Adam Smith: Besides the classical tariff war studied
by the international trade literature,
A prince, who should enact that a certain proportion of his taxes be paid in a recent papers study the strategic inter-
action along multiple fronts, including
paper money of a certain kind, might thereby give a certain value to this paper
monetary policies [Egorov and Mukhin,
money. 2019, Fontanier, 2023], trade and mon-
etary policies [Auray, Devereux, and
Our discussion shows that the issue of government debt valuation Eyquem, 2019], trade and containment
and sustainability is more complicated. Investors are willing to offer policies [Acharya, Jiang, Richmond, and
von Thadden, 2020], bank regulation
extra fiscal capacity to the world’s safe asset supplier beyond what is [Clayton and Schaab, 2022], and geopol-
warranted by its fiscal fundamentals and seigniorage revenues. When itics [Clayton, Maggiori, and Schreger,
the safe asset supplier’s relative fundamentals deteriorate, this extra 2023]. A common theme is the empha-
sis on the exchange rate as both a policy
fiscal backing is withdrawn by bond investors who then focus only instrument and a coordinated outcome.
on the country’s fundamentals.

Electronic copy available at: https://ssrn.com/abstract=4668578


9
Portfolio Choice and Asset Demand

Summary

• International portfolio choices underlie each country’s external imbalance and capital flow
dynamics. A country’s net foreign assets are directly a consequence of portfolio choices:

def
n f at = at ∑ x F,t (ι) − a∗t ∑ x ∗H,t (ι),
ι ι

the capital flows depend on how the portfolio choices evolve:


!
p∗t (ι)
 
def pt (ι)
c f t = ∑ at x F,t (ι) − at−1 x F,t−1 (ι) ∗ − ∑ a∗t x ∗H,t (ι) − a∗t−1 x ∗H,t−1 (ι) ,
ι p t −1 ( ι ) ι p t −1 ( ι )

and the profits from the chosen portfolios determine how a country’s external wealth evolves:

n f at − n f at−1 = tbt + ibt + cgt .

• The standard asset pricing literature models portfolio choices as the mean-variance trade-off
plus additional dynamic hedging terms:

x t = γ −1 Σ − 1
t Et [rt+1 ] + dynamic hedging terms.

• Alternatively, the demand system approach models portfolio choices based on asset prices,
characteristics, and latent demand terms:

exp(α log pt + βΞt + κt )


xt = .
1 + ∑k exp(α log pt + βΞt + κt )

Central in any account of the international monetary system is


how the investors in different countries choose their portfolios, how
their portfolios differ in ex-ante characteristics and in ex-post returns,
and how the capital flows due to investment rebalancing and pay-
outs allocate resources across countries. Answers to these questions
inform us about how the international monetary system is organized,

Electronic copy available at: https://ssrn.com/abstract=4668578


220 jiang – lecture notes on international finance

which countries play a central role, and the benefits and costs of
being at the central location.
In this chapter, we relate the households’ equilibrium portfolio
choices on the finance side to important quantities on the macro side
such as net foreign assets and balance of payments. Then, we discuss
two complementary approaches to modeling portfolio choices. One
is the traditional mean-variance approach, which is also central in the
standard asset pricing literature. The other is the demand system ap-
proach, which results from more recent developments. We will also
discuss specifications and applications of these approaches adapted
for studying international portfolio dynamics.

9.A Net Foreign Assets Accounting

To begin with, we relate the portfolio allocation decisions to ag-


gregate international macro variables such as net foreign assets
(NFA) and balance of payments. Our setting again nests the bench-
mark model in Section 1.A with one extension: home and foreign
households can trade more assets than just the risk-free bonds. Let
ι ∈ {1, . . . , I } index the different assets issued by either country,
which we refer to asset classes. These asset classes include the risk-
free bond, but they may also include typical asset classes such as
equity and long-term debt. We work out the NFA accounting be-
tween two countries, and our approach follows Jiang, Richmond, and
Zhang [2022c] which considers a more general multi-country setting.
For discussions and decompositions of the external adjustments, also
see Gourinchas and Rey [2007a], Ghironi, Lee, and Rebucci [2015].

9.A.1 Household Portfolio and Net Foreign Assets


Let us start with some definitions on the asset pricing side. We use
x H,t (ι) and x F,t (ι) to denote the home households’ portfolio shares
allocated to home and foreign assets, which satisfy

∑ x H,t (ι) + ∑ xF,t (ι) = 1.


ι ι

Similarly, foreign households’ portfolio shares satisfy ∑ι x ∗H,t (ι) +


∑ι x ∗F,t (ι) = 1. For simplicity, we will omit the foreign households’
expressions whenever they are symmetric.
We use at and a∗t to denote the home and foreign households’
wealth. In this case, it is simpler to denote both in home currency
units. We define the home households’ net foreign assets (n f a) as the
difference between their holdings of foreign assets (i.e., their external
assets) minus the foreigners’ holdings of their domestic assets (i.e.,

Electronic copy available at: https://ssrn.com/abstract=4668578


portfolio choice and asset demand 221

their external liabilities):


def
n f at = at ∑ x F,t (ι) − a∗t ∑ x ∗H,t (ι). (9.1)
ι ι

Let pt (ι) and p∗t (ι) denote the asset prices, and let divt (ι) and
div∗t (ι) denote the dividend or interest payouts, all in home currency
terms. We use rt (ι) to denote the cum-dividend returns in log:

def pt (ι) + divt (ι)


exp(rt (ι)) = .
p t −1 ( ι )

In this section, we further break down the cum-dividend returns


into the ex-dividend component ρt (ι) (i.e., capital gains) and the
dividend yield component dt (ι)1 : 1
Throughout the note, we use rt to de-
note returns in log. Here it is easier to
exp(rt (ι)) = 1 + ρt (ι) + dt (ι), deviate from this notational convention
and instead use ρt and dt to denote
p t ( ι ) − p t −1 ( ι )
def components of returns in level.
ρt (ι) = ,
p t −1 ( ι )
def divt ( ι )
dt (ι) = .
p t −1 ( ι )

Intuitively, household wealth accumulates as a result of realized fi-


nancial returns from existing positions as well as additional financial
savings deposited to their accounts. Let st denote the financial sav-
ings from period t − 1 to t in home currency terms. Then, we obtain
the standard law of motion for home household wealth:
!
a t = a t −1 ∑ x H,t−1 (ι) exp(rt (ι)) + ∑ xF,t−1 (ι) exp(rt∗ (ι)) + st . (9.2)
ι ι

Finally, let qt (ι) denote the notional quantity of the asset, and let
mt (ι) denote the total market value (i.e., market capitalization). The
market clearing condition for each asset can be expressed as
def
mt (ι) = pt (ι)qt (ι) = at x H,t (ι) + a∗t x ∗H,t (ι). (9.3)

We can relate the market value of all home assets to the home house-
holds’ wealth by the NFA:

at = ∑ mt (ι) + n f at ,
ι

which generalizes Eq. (7.3) to the multi-asset case.


The NFA definition (9.1), wealth law of motion (9.2), and market
clearing (9.3) constitute an equation system that describes the asset
market. If we also take a stance on how the households decide their
portfolio shares x H,t and x F,t , which is to be specified in subsequent
sections, then, we can solve the equilibrium asset prices, household
wealth dynamics, and international NFA dynamics and capital flows.

Electronic copy available at: https://ssrn.com/abstract=4668578


222 jiang – lecture notes on international finance

9.A.2 Balance of Payments


Having specified the asset market space, we next turn to its con-
nection to the quantity side. In international macroeconomics, it is
common to use the following balance of payments identity to organize
the law of motion for the net foreign assets:

n f at − n f at−1 = tbt + ibt + cgt , (9.4)

where tb denotes the trade balance, ib the income balance, and cg the
capital gains.
The trade balance is defined in the standard way as the value of
exports minus the value of imports, expressed in the home currency.
The income balance is defined as the earnings on foreign invest-
ments minus the payments made to foreign investors:

ibt = at−1 ∑ x F,t−1 (ι)d∗t (ι) − a∗t−1 ∑ x ∗H,t−1 (ι)dt (ι).


ι ι

The capital gains are defined as the changes in the value of assets
that the home households hold abroad minus changes in the value of
domestic assets held by foreign investors:

cgt = at−1 ∑ x F,t−1 (ι)ρ∗t (ι) − a∗t−1 ∑ x ∗H,t−1 (ι)ρt (ι).


ι ι

The sum of income balance and capital gains (i.e., ibt + cgt ) cap-
tures the overall wealth transfer between home and foreign house-
holds due to their holdings of each other’s assets. The balance of
payments identity (9.4) states that the home country accumulates a
higher NFA position when (i) it runs a trade surplus, which results
in claims that foreigners owe to the home country, and (ii) it earns a
higher financial return than it pays to foreigners, which includes both
capital gains and dividend payouts. Conversely, if the NFA remains
constant, any trade deficit (i.e., negative tb) has to be offset by a posi-
tive financial return (i.e., positive ib + cg), as we have seen in the case
of the U.S. in Chapter 7.
In international macroeconomics, it is also common to bundle
together the trade balance and the income balance into the current
def
account: cat = tbt + ibt , which describes the net earnings of the home
country from its international trade and investments. As capital gains
cgt capture important variations in cross-border wealth transfers
from financial markets in the data, we find it more convenient to
conceptually bundle the income balance and the capital gains as total
financial transfers.
Finally, like the government budget condition in Chapter 8, we
can also think of the NFA dynamics (9.4) as a backward-looking

Electronic copy available at: https://ssrn.com/abstract=4668578


portfolio choice and asset demand 223

relationship. Assuming that n f at is stationary, the forward-looking


relationship can be expressed as
∞ ∞
n f at − lim n f at+s = −
s→∞
∑ Et [tbt+s ] − ∑ Et [ibt+s + cgt+s ].
s =1 s =1

Therefore, any deficit in the NFA relative to its long-run mean has to
be paid off by either future trade surpluses tbt+s or future financial
incomes ibt+s + cgt+s . We will see an example of this decomposition
in the context of the segmented market and safe asset models in
Section 10.B. Moreover, since the world economy grows over time,
we can also normalize the NFA by the GDP or its low-frequency
component. See Gourinchas and Rey [2007a] for details.

9.A.3 Capital Flows


Cross-border capital flows are also related to disaggregated bilateral
portfolio choices. The market value of net position changes in the
foreign assets held by the home households is
!
p∗t (ι)
∑ at xF,t (ι) − at−1 xF,t−1 (ι) p∗ (ι) .
ι t −1

Similarly, the market value of net position changes in the home assets
held by the foreign households is
 
pt (ι)
∑ t H,t
a ∗ ∗
x ( ι ) − a ∗
x ∗
t−1 H,t−1 ( ι )
p t −1 ( ι )
.
ι

These quantities capture the position changes in cross-border


portfolio holdings due to investors’ active portfolio rebalancing and
trading. Variations in asset prices and hence capital gains do not
affect these quantities. The (portfolio-driven) net capital flows are
defined as their difference:
!
p∗t (ι)
 
def pt (ι)
c f t = ∑ at x F,t (ι) − at−1 x F,t−1 (ι) ∗ − ∑ a∗t x ∗H,t (ι) − a∗t−1 x ∗H,t−1 (ι) ,
ι p t −1 ( ι ) ι p t −1 ( ι )

which is increasing in the home households’ purchase of foreign


assets and decreasing in the foreign households’ purchase of home
assets.
Using this definition, we can decompose the net foreign assets
dynamics as

n f at − n f at−1 = cgt + c f t , (9.5)

which clearly separates the valuation effects as captured by the capi-


tal gains cgt from the quantity changes as captured by the net capital
flows c f t .

Electronic copy available at: https://ssrn.com/abstract=4668578


224 jiang – lecture notes on international finance

Moreover, the income balance constitutes another form of cross-


border flows, which results from the dividend or coupon payouts
to foreign investors. Since c f t and ibt denote flows in opposite di-
rections under our notation, the aggregate financial flow is their
difference:
def
f f t = c f t − ibt . (9.6)

Combining Eq. (9.4), (9.5), and (9.6), we obtain an identity between


the trade balance and the aggregate financial flow:
def
tbt = f f t = c f t − ibt ,

which states another form of the balance of payments identity. This


identity again shows that imbalances on the trade side need to be
consistent with the adjustments on the financial side. Specifically,
a trade surplus corresponds to a net purchase of foreign assets by
home households, which results in a financial outflow.

9.A.4 Market Clearing and Financial Savings


We next consider market clearing, which connects the assets’ market
values to the households’ financial wealth. In the financial autarky
case, n f a = tb = ib = cg = 0, and we have the domestic market
clearing conditions:

at x H,t (ι) = mt (ι),


at = ∑ m t ( ι ).
ι

Then, given the definition of financial savings under autarky, i.e.,

at = ∑ mt−1 (ι)(1 + dt (ι) + ρt (ι)) + st ,


ι

we obtain the following relationship between financial savings and


asset market values:
!
st = ∑ mt (ι) − ∑ mt−1 (ι)(1 + ρt (ι)) − ∑ m t −1 ( ι ) d t ( ι ). (9.7)
ι ι ι

The first term on the right-hand side captures new asset issuances,
which channel funds from the households to the corporate sector.
The second term captures dividend payouts, which are transfers in
the opposite direction from the corporate sector to the households.
The households’ financial savings are equal to the transfers from
households to the corporate sector in the form of asset issuances
minus the transfers from the corporate sector to the households in
the form of dividend payouts.

Electronic copy available at: https://ssrn.com/abstract=4668578


portfolio choice and asset demand 225

Another way to understand this relationship is to realize that


not everyone in this economy can reinvest their dividends and hold
more shares at the same time. So, by market clearing, absent issuing
new assets and financing new projects, dividend payouts have to be
taken out of the financial market and consumed, which results in a
decrease in the households’ financial savings.
With these lessons learned from the autarky economy, we are now
ready to consider the more general case with cross-border invest-
ments. We can extend Eq. (9.7) and obtain the following result in the
open economy.

Proposition 9.1. The home households’ financial savings st are equal to


the trade balance tbt plus the net transfers between investors and the home
corporate sector:
!
st = tbt + ∑ mt (ι) − ∑ mt−1 (ι)(1 + ρt (ι)) − ∑ m t −1 ( ι ) d t ( ι ).
ι ι ι

The proof is presented in Appendix A.39. This result connects the


trade balance on the macro side to net financial savings and asset
issuance on the asset market side.
First, consider the case in which asset issuances are exactly equal
to the dividend payouts. Then, we obtain a special case:

tbt = st ,

which states that the trade balance is equal to the home households’
net financial savings. In this case, absent transfers to and from the
financial markets, savings can only result from exporting more goods
to foreigners and earning their IOUs.
Next, consider the case in which the trade balance is tbt = 0, and
the dividend payouts are dt (ι) = 0. Then, we obtain
!
st = ∑ mt (ι) − ∑ mt−1 (ι)(1 + ρt (ι)) ,
ι ι

which implies that the asset issuances in the home country are en-
tirely financed by the home households’ financial savings. If the
foreign households participate in the financing of the new issuances,
we should expect to see capital flows which, by the previous sec-
tion, should result in non-zero trade balance. Thus, the trade balance
becomes the sufficient statistics for the foreign households’ participa-
tion in the home asset issuances and financial savings process.
Finally, we can understand the general case by again understand-
ing that (∑ι mt (ι) − ∑ι mt−1 (ι)(1 + ρt (ι))) − ∑ι mt−1 (ι)dt (ι) captures
the net transfers between (home and foreign) investors and the home
corporate sector. When the home investors contribute funds to the

Electronic copy available at: https://ssrn.com/abstract=4668578


226 jiang – lecture notes on international finance

home corporate sector, we expect to see manifestations in the finan-


cial savings st . When the foreign investors contribute funds to the
home corporate sector, we expect to see manifestations in the trade
balance tbt .
Conversely, once we understand how the asset market players
in different countries make their saving and issuance decisions, we
should also expect adjustments on the trade side. In particular, for
the U.S. which runs persistent trade deficits, this approach offers an
account of the financial drivers that fund the trade deficits.

9.A.5 Relation to National Consumption


Finally, we make a connection to consumption decisions in the na-
tional accounts. We assume that the aggregate output yt is split into
payoff to the financial assets and labor income wtℓ , i.e.,

pt yt = ∑ mt (ι)dt (ι) + wtℓ .


ι

Then, we can rewrite the standard household budget constraint Eq.


(1.1) as
!
a t −1 ∑ x H,t−1 (ι) exp(rt (ι)) + ∑ xF,t−1 (ι) exp(rt∗ (ι)) + wtℓ = ct + at .
ι ι

The left-hand side is the national income for the home households,
which consists of financial asset payoff and labor income. The right-
hand side contains two types of expenditure: consumption and finan-
cial investments.
Using Eq. (9.2), we obtain

ct = wtℓ − st ,

which states that consumption is equal to labor income minus finan-


cial savings. For example, if the investors experience a high return
from the financial assets they hold, they can liquidate some of their
financial assets and increase consumption.
In this context, it is more convenient to consider a notion of net
financial savings, which includes the dividends households receive:
!
def
f t = s t + a t −1 ∑ x H,t−1 (ι)dt (ι) + ∑ xF,t−1 (ι)d∗t (ι) .
ι ι

This notion allows us to express the household wealth as the sum of


the previous period’s wealth, the capital gains, and the net financial
savings:
!
a t = a t −1 1 + ∑ x H,t−1 (ι)ρt (ι) + ∑ x F,t−1 (ι)ρ∗t (ι) + ft. (9.8)
ι ι

Electronic copy available at: https://ssrn.com/abstract=4668578


portfolio choice and asset demand 227

For example, if the net financial savings f t = 0, the households


consume all dividend payouts instead of reinvesting them in the
financial market. In this case, the households’ financial wealth accu-
mulates according to the capital gains from their existing positions.
This expression implies
!
f t − a t −1 ∑ x H,t−1 (ι)dt (ι) + ∑ xF,t−1 (ι)d∗t (ι) = st = wtℓ − ct ,
ι ι

or
!
ct = wtℓ + at−1 ∑ x H,t−1 (ι)dt (ι) + ∑ xF,t−1 (ι)d∗t (ι) − ft.
ι ι

So, the consumption is equal to labor income plus dividend payouts


minus net financial savings f t . Suppose the home households do
not buy or sell any of their financial assets, i.e., f t = 0, then, the
consumption is equal to labor income plus dividend payouts. In
contrast, if the home households decide to save more by purchasing
additional financial assets, they will have a lower consumption.

9.B The Mean-Variance Approach

This set-up follows the standard Merton [1969] portfolio problem


with multiple assets. For tractability, we assume households have
power utility and a finite horizon T. This set-up is otherwise similar
to the domestic economy considered in Chapter 1, which is pop-
ulated by a representative household. The household maximizes
expected lifetime utility:
1− γ
" #
T
t ct
E0 ∑ δ .
t =0
1−γ

The household can trade a set of financial assets. Let wt denote


the wealth before consumption, and let at denote its total asset under
management after consumption. Risky assets are indexed by ι ∈
{1, 2, . . . , N }, and there is an asset that is risk-free in real terms. Let
xt (ι) denote the households’ portfolio share on the asset ι, and let
f
exp(rt+1 (ι)) denote the asset’s real return. Let exp(rt ) denote the
risk-free asset’s real return. Then, the portfolio-level return can be
expressed as
!
def
1 − ∑ xt−1 (ι) exp(rt−1 ) + ∑ xt−1 (ι) exp(rt (ι)),
p f
exp(rt+1 ) =
ι ≥1 ι ≥1

which can be expressed more conveniently in vector form:


p f f
exp(rt+1 ) = exp(rt−1 ) + xt′ −1 (exp(rt ) − exp(rt−1 )).

Electronic copy available at: https://ssrn.com/abstract=4668578


228 jiang – lecture notes on international finance

The standard law of motion for home household wealth is


p
wt = at + ct = at−1 exp(rt+1 ).

Let zt denote the vector of state variables that determine the distri-
bution of returns. We define the value function as
1− γ
" #
T
def s cs
ψt (wt , zt ) = max Et ∑ δ ,
{cs ,xs }sT=t s=t 1−γ

which can be expressed in the following recursive form


1− γ
ct
ψt (wt , zt ) = max δt + Et [ψt+1 (wt+1 , zt+1 )] .
ct ,xt 1−γ
This is a generic set up of a portfolio choice model. See Back
[2010] for more details, or Duffie [2010] for a continuous-time for-
mulation. To fully solve this model, we need to rely on specific as-
sumptions of investor preferences and asset return dynamics. When
the investors have power utility, we obtain the following characteriza-
tions for portfolio demand.
Proposition 9.2. (a) The value function satisfies
1− γ
wt
ψt (wt , zt ) = δt f t (zt ) ,
1−γ
where the function f t (zt ) depends on the state variable and satisfies f t (zt ) =
γ
wct where wct = wt /ct denotes the wealth-consumption ratio.
(b) The optimal portfolio allocation satisfies
" p #
exp((1 − γ)rt+1 )
xt = arg max Et f t+1 (zt+1 ) , (9.9)
xt 1−γ
p
where the log portfolio return rt+1 depends on the portfolio choice xt .
The proof is in Appendix A.40. Eq. (9.9) implies that, when the
p
investors maximize their expected portfolio return rt+1 , they also pay
attention to the expected evolution of the state variable zt+1 . Con-
sequently, the optimal portfolio choice xt depends on the expected
future state.

9.B.1 The Case of Constant Investment Opportunity


To derive closed-form solutions, the literature uses either continuous-
time math or linearization to further simplify the problem. A popular
choice is to approximate the portfolio return via log-linearization as
proposed by Campbell and Viceira [2002], which approximates the
log portfolio return by
p f f 1 1
rt+1 ≈ rt + xt′ (rt+1 − rt + σt2 ) − xt′ Σt xt ,
2 2

Electronic copy available at: https://ssrn.com/abstract=4668578


portfolio choice and asset demand 229

where Σt denotes the conditional variance-covariance matrix of the


log returns, and σt2 is the vector containing the diagonal elements of
Σt .
With this approximation formula, we can now solve the portfolio
choice problem when the state variable is a constant, i.e., zt = z̄,
which implies that the asset returns are i.i.d. and f t+1 (zt+1 ) is a
constant.

Proposition 9.3. When the asset returns are i.i.d. and normally distributed,
the optimal portfolio choice is
 
1 2 f
x t = γ −1 Σ − 1
E [ r t +1 ] + σ − r . (9.10)
t
2 t t

The proof is in Appendix A.41. This formula (9.10) recovers the


classical mean-variance trade-off in portfolio choice theory. In the
f
numerator, E[rt+1 ] − rt denotes the assets’ expected excess returns
in log, and 12 σt2 captures the Jensen’s term adjustment, so that the
f
sum E[rt+1 ] + 12 σt2 − rt captures the assets’ expected excess returns
in level, which is exact when returns are normally distributed or in
continuous time. In the denominator, Σt is the covariance matrix of
asset returns and γ is the risk aversion coefficient. Therefore, the
optimal portfolio favors assets that have high expected excess returns
and low variance and covariance with other assets.

9.B.2 Special Cases with Time-varying Investment Opportunity


When the state variable zt is time-varying, the problem becomes
more difficult in general. Here we discuss two special cases. First,
Campbell and Viceira [2002] consider a case in which the risk-free
rate is time-varying but the risky assets’ variances and risk premia
are constant. In this case, the optimal portfolio choice is


    !
1 1 2 1
r t +1 , − (Et +1 − Et ) ∑
f f
xt = Σ− 1
E t r t +1 − r + σ + 1 − Σ− 1
t covt ρ j rt+ j .
γ t t 2 t γ j =1
(9.11)

This solution contains two terms. The first term is similar to the
solution under i.i.d. returns derived above, which depends on the
assets’ excess returns and variance-covariance matrix. The second
term is commonly known as the hedging term, as the investors buy
certain assets to hedge shocks to future investment opportunities.
In this case, since risk premia are constant, variations in future in-
vestment opportunities are only driven by variations in the risk-free
rate. Specifically, a surprise reduction in expected future risk-free
rates represents a negative shock to future investment opportunities.

Electronic copy available at: https://ssrn.com/abstract=4668578


230 jiang – lecture notes on international finance

When the risk aversion coefficient γ > 1, investors would like to load
up more on assets that offer higher returns when expected future
risk-free rates go down.
Second, if the investors have log utility, then, Eq. (9.9) can be sim-
plified to
h
p
i
f 1 1
xt = arg max Et rt+1 ≈ arg max xt′ (Et [rt+1 ] − rt + σt2 ) − xt′ Σt xt ,
xt xt 2 2
which does not depend on the state variable. In this sense, investors
with log utility are myopic, as they choose portfolios like one-period
investors. They do not pay attention to how different assets’ re-
turns respond to shocks to future investment opportunities. These
investors also have a predetermined wealth-consumption ratio, which
is a constant when the horizon T is infinite.
Moreover, with log utility
 (i.e., γ = 1), Eq. (9.11) in the first special
f
case becomes xt = Σ− t
1
Et [rt+1 ] − rt + 12 σt2 . The optimal portfolio
allocation no longer loads on the term that hedges the variations
in future interest rates, and the myopic investors only focus on the
mean-variance trade-off.

9.C Applications of the Mean-Variance Approach (TODO)

9.D The Demand System Approach

Eq. (9.10) offers an intuitive way to think about the portfolio alloca-
tion problem as a mean-variance trade-off. While this approach has
been powerful and universal, several challenges stand out. First, es-
timating the expectation and the variance-covariance matrix of asset
returns is empirically challenging. As the assets’ risk characteristics
and investors’ risk appetite are time-varying, the moments estimated
from the past return data do not necessarily represent the distribu-
tion of future returns.
Second, investor demand for different assets and asset classes
appears to be less elastic than what standard models imply. For ex-
ample, according to Koijen and Yogo [2021], “a calibration of the
capital asset pricing model (CAPM) implies a demand elasticity for
individual stocks that exceeds 5,000,” whereas empirical estimates
of the demand elasticity are closer to 1 or lower [Gabaix and Koijen,
2021], which is three orders of magnitude smaller.
In response to these challenges, Koijen and Yogo [2019] turn to
a complementary approach based on demand systems. Instead of
deriving the optimal portfolio choice from the mean-variance trade-
off, their starting point is that the households allocate their wealth
based on the assets’ characteristics. In this section, we study the most
basic demand system model based on logit demand functions.

Electronic copy available at: https://ssrn.com/abstract=4668578


portfolio choice and asset demand 231

9.D.1 Logit Demand


Let δt (ι) denote the desirability of asset ι:

δt (ι) = α log pt (ι) + βΞt (ι) + κt (ι),

where log pt (ι) denotes the log asset price, Ξt (ι) denotes the vector
of additional observable asset characteristics, and κt (ι) denotes an
unobservable latent demand term. We expect α < 0, in which case
the households find an asset more attractive when its price is lower.
The asset also becomes more attractive when it has some desirable
attributes captured by Ξt (ι), which for example could include a lower
price volatility, a higher credit rating, or simply a larger asset supply.
With heterogeneous investors indexed by n, the characteristics could
also be bilateral, i.e., Ξn,t (ι), which could for example capture the
investors’ home bias towards domestic assets, or their preferences for
foreign assets according to proximity in geographic distance or trade
network as we saw in Section 3.C. Finally, the latent demand term
allows us to capture the unobservable characteristics of the asset that
simultaneously generate a higher market share and a higher asset
price by making the asset more attractive [Berry, 1994].
We assume that the households’ allocation on this asset follows
exp(δt (ι))
xt (ι) = . (9.12)
1 + ∑k exp(δt (k))
We make several observations. First, if α < 0, the households have a
downward-sloping demand curve for this asset. This can be thought
of as a generalization of the downward-sloping demand curve for re-
serve assets that we considered in Section 4.A. In our current setting,
all assets face downward-sloping demand curves.
Second, the 1 in the denominator models the existence of an out-
side asset, which is a benchmark asset whose desirability δt (0) is nor-
malized to be 0. The choice of this outside asset is usually model-
specific. For example, it could be bank deposits when studying the
demand system of money market funds, or it could be the assets is-
sued by small countries with missing characteristics when studying
the demand system of international portfolio allocation.
Third, the households’ allocation to asset ι also depends on the
desirability of other assets. When some other assets become relatively
more desirable, then, the allocation to asset ι goes down. Moreover,
when all inside assets experience a uniform increase in their de-
sirabilities, the allocations to all inside assets go up relative to the
allocation to the outside asset.
Fourth, this set-up could be micro-founded in several ways. The
most classical micro-foundation is borrowed from the field of indus-
trial organization [Berry, 1994, Berry, Levinsohn, and Pakes, 1995,

Electronic copy available at: https://ssrn.com/abstract=4668578


232 jiang – lecture notes on international finance

Nevo, 2001]. Specifically, the problem of investors choosing which as-


set to hold is translated to the problem of consumers choosing which
product to purchase. Suppose there is a continuum of investors in-
dexed by n, and each investor has one dollar to invest. If investor n
picks asset i, he or she has the following utility

un,t (ι) = δt (ι) + ε n,t (ι).

The outside asset 0 has utility un,t (0) = 0.


Investor n chooses to invest in asset ι if un,t (ι) ≥ un,t ( j) for all
j. The term ε n,t (ι) captures the investor’s personal preference. If
this term is i.i.d. and follows the standard logit distribution with a
cumulative distribution function Φ(ε) = exp(− exp(−ε)), then, we
can aggregate across all investors and derive the amount of funds
invested in asset ι to be
exp(δt (ι))
at ,
1 + ∑k exp(δt (k))

where at is the aggregate funds of the investors. In this way, we


obtain the representative investor’s demand function given by Eq.
(9.12).
Alternatively, Koijen and Yogo [2019] propose a mapping from the
standard mean-variance portfolio allocation problem in Section 9.B to
the demand system. They do so by imposing a factor structure to the
variance-covariance matrix of asset returns, allowing the factor load-
ings to depend on the asset characteristics, and adopting a particular
form of approximation.
The characteristics-based demand could have other micro-foundations
as well, such as private information, heterogeneous beliefs [Pelle-
grino, Spolaore, and Wacziarg, 2021], institutional constraints, and
non-pecuniary preferences for certain traits such as ESG, the demand
for reserve assets as we studied in Section 7, and behavioral factors.2 2
For asset-level characteristics, see
Daniel and Titman [1997] and the sub-
sequent covariance vs. characteristics
9.D.2 Closing the Model with the Supply Side debate. Investor-level characteristics
such as personalities [Jiang, Peng, and
To close the model, we need to take a stance on the total amount of Yan, 2023c] and recall bias [Jiang, Liu,
Peng, and Yan, 2022b] could affect
funds the households have for investments. Let us denote this total portfolio allocations.
asset under management by at . In some models, at is exogenously
determined; in other models, at depends on previous investment
outcomes. For example, it makes sense to use the same law of motion
for the asset under management as in the traditional portfolio model
in Section 9.B:
!
a t = a t −1 ∑ xt−1 (ι) exp(rt (ι)) + st ,
ι

Electronic copy available at: https://ssrn.com/abstract=4668578


portfolio choice and asset demand 233

which requires us to model st representing how much the house-


holds save into their investment account after consumption. In the
case that the households consume more than their income, the flow
term st could be negative. Section 9.A discusses these flows in more
details.
We also need to specify asset supply and market clearing. Let qt (ι)
denote asset ι’s quantity of shares. The market clearing condition is

p t ( ι ) q t ( ι ) = a t x t ( ι ). (9.13)

In some models, the asset supply qt (ι) is assumed to be a constant or


exogenously specified. In other models, the asset supply is endoge-
nous, as the issuers maximize their profits while being aware of the
downward-sloping demand curve. Specifically, suppose the issuer of
asset ι maximizes the profit by setting the price:

max pq( p) − c(q( p)),


p

where the issuance cost c(q) is a function of the quantity q, and the
quantity q is determined by the demand curve at a given price p.
We drop the period-t subscripts for simplicity. Then, the first-order
condition implies that the marginal revenue equals the marginal cost,
which implies

p − c′ (q( p)) = q( p)/q′ ( p).

On the left-hand side, p − c′ (q( p)) is the issuer’s mark-up defined as


the asset price minus the marginal cost. This result implies that the
issuer sets the mark-up according to the demand elasticity.
Yet in other cases, the asset supply can be fully elastic, which
means that the issuer can supply an arbitrary quantity of the asset
at a given price. For example, this could apply to open-end money
market funds whose underlying holdings are liquid and safe assets
with a deep market. In this case, when there is a shock to investor
demand, the market clears by quantity adjustment instead of by price
adjustment.

9.D.3 Demand Elasticity


This demand system model also has a simple expression for demand
elasticity. The log demand by issuer ι is given by

log qdemand
t (ι) = log( at xt (ι)/pt (ι)),
∂ log xt (ι)
which implies that the demand elasticity is3 3
The derivation is simple: ∂ log pt (ι)
=
∂δt (ι) ∂ log(1+∑k exp(δt (k)))
− = α −
∂ log qdemand
t (ι) ∂ log xt (ι) ∂ log pt (ι) ∂ log pt (ι)
− = 1− = 1 − (1 − xt (ι))α. exp(δ (ι))
α 1+∑ expt (δ (k)) .
∂ log pt (ι) ∂ log pt (ι) k t

Electronic copy available at: https://ssrn.com/abstract=4668578


234 jiang – lecture notes on international finance

To interpret this result, note that we expect α < 0 as demand is


decreasing in price. Then, the demand elasticity is increasing in the
magnitude |α|, a higher value of which means that the households’
demand is more sensitive to the price. Magnitude-wise, suppose this
asset is only 1% of the total market cap, and α = −1, so that a 1%
increase in asset price roughly leads to a 1% reduction in investor
demand. Then, the demand elasticity is 1.99, which is already at the
higher end of the distribution of empirical estimates.
The demand elasticity is also decreasing in the portfolio share
xt (ι), which means an asset with a larger size has fewer alternatives
to substitute with. This is a desirable feature of the asset demand sys-
tem, because reserve assets in high demand such as the U.S. Treasury
tend to face a more inelastic demand curve [Choi, Dang, Kirpalani,
and Perez, 2023]. If our asset demand system correctly captures the
high market share in either the asset characteristics Ξt (ι) or the latent
demand κt (ι), then, we also make its demand curve more inelastic
than other assets.
So far, we have only considered one representative household. It is
straightforward to extend the model and allow for multiple investors,
which, for example, is a natural case when we consider investments
between different countries. Let n index the investors. Then, the
market clearing condition becomes
pt (ι)qt (ι) = ∑ an,t xn,t (ι),
n

and the aggregate demand elasticity is simply a weighted average of


individual investors’ demand elasticities:
!
∂ log qdemand (ι) an,t xn,t (ι) ∂ log qdemand (ι)
− t
=∑ − n,t
.
∂ log pt (ι) n ∑m am,t xm,t ( ι ) ∂ log pt (ι)

9.D.4 What If There Is No Outside Asset?


The outside asset plays an important role in pinning down the equi-
librium solution in this asset demand system. While this point is
more general, let us consider a simple case in which there is one in-
vestor, one inside asset, and no outside asset. Then, regardless of the
desirability of the asset, the portfolio share allocated towards this as-
set is always xt = 1. We also assume that the asset quantity is qt = 1
and the asset pays no dividend. Then, the market clearing condition
implies:
pt = pt qt = at xt = at ,
and the investor’s wealth dynamics follows
pt
a t = a t −1 + st .
p t −1

Electronic copy available at: https://ssrn.com/abstract=4668578


portfolio choice and asset demand 235

We assume the wealth at−1 and the asset price pt−1 in the last period
are both 1. In this case, if the saving st = 0, the market clearing
condition and the investor’s wealth dynamics are consistent with
any asset price pt . If the saving st ̸= 0, then, there will be no asset
price that is consistent with the market clearing condition and the
investor’s wealth dynamics. Both cases are pathological.
In comparison, if we add the outside option, the portfolio choice
becomes
exp(α log pt )
xt = ,
1 + exp(α log pt )

the market clearing condition for the inside asset is:

pt = pt qt = at xt ,

and, assuming that the outside option pays no return, the investor’s
wealth dynamics follows
 
pt
a t = a t −1 (1 − x t −1 ) + x t −1 + st = ((1 − xt−1 ) + xt−1 pt ) + st .
p t −1

We assume α = −1 and xt−1 = 1/ 2. Then, this system of
equations has only one root with a positive asset price, with,
√ √ √
pt = 2 − 1, at = 2 − 2, xt = 1/ 2,

when st = 0, and, consistent with our intuition, the asset price pt


is increasing in the saving st . The introduction of the outside asset
makes the demand function for the book value of the inside asset
downward-sloping in the asset price, i.e.,

1 − xt−1 + xt−1 pt + st exp(α log pt )


qdemand
t = at xt /pt = ,
pt 1 + exp(α log pt )

because the investor can substitute towards the outside asset if the
inside asset is too expensive. As we plot in Figure 9.1, this demand
supply
function crosses a flat supply curve qt = 1 and determines the
unique equilibrium price. In comparison, without the outside asset,
the demand function for the book value of the inside asset is flat, and
the equilibrium price is not determined.

9.D.5 Identifying the Elasticity


Finally, while this note deals with the theoretical issues of the asset
demand system, it is worth noting that its empirical estimation and
identification are also at the forefront of this literature. Here, we
provide a brief discussion of the identification strategies used in the
literature.

Electronic copy available at: https://ssrn.com/abstract=4668578


236 jiang – lecture notes on international finance

Figure 9.1: Market Clearing for the


Book Quantity of the Asset.
5

0
0 0.2 0.4 0.6 0.8 1

Generally speaking, to identify the shape of the demand curve,


which is parameterized using the logit functional form in this section,
we need variations in the supply curve. Therefore, the most straight-
forward instruments are the exogenous shifters of the supply curve.
For example, Xiao [2020], Chen and Jiang [2022] use non-interest
expenses of banks and funds as exogenous cost shifters to identify
the demand for their assets. Diamond, Jiang, and Ma [2020] derive
changes in the bank branches’ marginal costs due to the shocks origi-
nating from other locations transmitted via the banks’ internal capital
market. In the context of government debt, Choi, Kirpalani, and
Perez [2022] directly use plausibly exogenous changes in the debt
supply.
Second, the literature has used the non-price characteristics of
competing assets, commonly referred to as the BLP instruments
[Berry, Levinsohn, and Pakes, 1995]. The identification assumption
is that the changes in these exogenous characteristics of other as-
sets affect the residual demand curve for the asset in question. In
the same spirit, Koijen and Yogo [2020] propose a model-based pro-
cedure to calculate counterfactual asset prices based on exogenous
characteristics and use them as instruments. For example, suppose
the asset demand is stronger for country pairs which trade a lot with
each other. Then, solving the demand system model with the asset
demand δt (ι) in Eq. (9.12) being a function of the trade relationship
alone leads to exogenous variations in asset prices. Implicitly, this
procedure uses a non-linear aggregation of all assets’ characteristics
as the instrument.
Third, Gabaix and Koijen [2020] propose a granular instrumental

Electronic copy available at: https://ssrn.com/abstract=4668578


portfolio choice and asset demand 237

variable approach that exploits the idiosyncratic shocks from large


players, which presents a new strategy for demand system settings.

9.E Application to International Portfolio Allocation

In this section, we consider a two-tier demand system that Koijen and


Yogo [2020], Jiang, Richmond, and Zhang [2022c,d] use to study the
international financial market. We consider a simplified case in which
there are only two countries and two asset classes.
In this model, we assume that the investors make two sequen-
tial decisions. First, they allocate their wealth across different asset
classes indexed by ℓ, which we refer to as short-term bonds (ℓ = 1)
and equities (ℓ = 2) in this note. Second, within each asset class, they
allocate their wealth across different issuer countries indexed by ι.
As such, investor country n’s demand for the asset in class ℓ, issuer
country ι has a nested-logit structure:

xn,t (ι, ℓ) = xn,t (ℓ) · xn,t (ι|ℓ),

where xn,t (ℓ) denotes the portfolio share of the entire asset class ℓ,
and xn,t (ι|ℓ) denotes the portfolio share of the issuer country ι within
asset class ℓ. This two-tier structure offers a richer structure to model
the cross-sectional heterogeneity in the demand for different asset
classes and countries separately.
The allocation within the asset class is similar to the one we con-
sidered in the baseline case:
exp(δn,t (ι, ℓ))
xn,t (ι|ℓ) = ,
1 + ∑k exp(δn,t (k, ℓ))
where

δn,t (ι, ℓ) = α log( pt (ι, ℓ) exp(et (n) − et (ι))) + κn,t (ι, ℓ)

captures the desirability of this asset (ι, ℓ). We refer to the currency
of the home country as the dollar. We use pt (ι, ℓ) to denote the asset
price in the local currency units, and et (ι) to denote the log exchange
rate of the local currency per dollar. Then, pt (ι, ℓ) exp(et (n) − et (ι))
captures the price of the asset from the issuer country ι in the units of
the local currency in the investor country n. For simplicity, we ignore
the asset-specific characteristics Ξt (ι). We allow the latent demand
κn,t (ι, ℓ) to be specific to the investor n.
The allocation across asset classes is modeled as
(1 + ∑k exp(δn,t (k, ℓ)))λ exp(ψℓ + ξ n,t (ℓ))
xn,t (ℓ) = ,
∑m (1 + ∑k exp(δn,t (k, m)))λ exp(ψm + ξ n,t (m))
where ψℓ + ξ n,t (ℓ) captures the investor’s preference for asset class
ℓ, which is similar to the latent demand term κn,t (ι, ℓ) for a specific

Electronic copy available at: https://ssrn.com/abstract=4668578


238 jiang – lecture notes on international finance

issuer country ι. In empirical specifications, the κn,t (ι, ℓ) across the


asset classes are identified only up to a constant, which requires us
to normalize this term to be zero for one asset class. The other term
(∑k exp(δn,t (k, ℓ)))λ is known as the inclusive value, which captures
the desirability of the assets within this asset class including their
prices and latent demand terms. For microfoundation and more
discussions of this inclusive value term, please refer to industrial
organization textbooks such as Train [2009].
Let an,t denote the wealth of investor country n in the dollar units.
Then, this investor country holds an,t xn,t (ι, ℓ) dollars’ worth of the
asset (ι, ℓ). Let qt (ι, ℓ) denote asset ι’s quantity of shares. Then, in the
presence of multiple investor countries, the market clearing condition
(9.13) is generalized to

pt (ι, ℓ) exp(−et (ι))qt (ι, ℓ) = ∑ an,t xn,t (ι, ℓ),


n

for all issuer country ι representing the set of insider assets and asset
class ℓ. We do not impose a market clearing condition on the outside
asset, which is assumed to have an infinitely elastic supply at a fixed
local-currency price and desirability.
Finally, as emphasized by Jiang, Richmond, and Zhang [2022c,d],
the investor country n’s wealth responds endogenously to the asset
returns. Let at denote the dollar value of its wealth, which follows the
following law of motion:
!
an,t = an,t−1 ∑ ∑ xn,t−1 (ι, ℓ) exp(rt (ι, ℓ)) + sn,t ,
ℓ ι ≥0

where, for simplicity, we assume that the assets pay no dividends,


and their dollar-denominated returns are given by their capital gains:

pt (ι, ℓ)
exp(rt (ι, ℓ)) = exp(−∆et (ι)).
pt−1 (ι, ℓ)

We take the saving sn,t , the asset quantity qt (ι, ℓ), and the latent de-
mand κn,t (ι, ℓ) as exogenously given. The portfolio choices xn,t (ι, ℓ),
the market clearing conditions, and the wealth law of motion jointly
determine the equilibrium asset prices, exchange rates, and capital
flows.

9.E.1 Macro Synthesis

We study the competitive equilibrium in which the portfolio choices


follow the logit demand, the wealth follows the given law of motion,
and the markets for all inside assets clear.

Electronic copy available at: https://ssrn.com/abstract=4668578


portfolio choice and asset demand 239

The exogenous variables are the savings, the latent demand terms,
and the asset quantities:

(sn,t , qt (ι, ℓ), κn,t (ι, ℓ), ξ n,t (ℓ))∞


t =0 , for n, ι, ℓ = 1, 2.

There are 23 endogenous variables in each period t:

( an,t , δn,t (ι, ℓ), xn,t (ι, ℓ), pt (ι, ℓ), et )∞


t =0 , for n, ι, ℓ = 1, 2.

The model implies the following 22 equations, including 8 for


portfolio choices:

exp(δn,t (ι, ℓ))(1 + ∑k exp(δn,t (k, ℓ)))λ−1 exp(ψℓ + ξ n,t (ℓ))


xn,t (ι, ℓ) = , for n, ι, ℓ = 1, 2,
∑m (1 + ∑k exp(δn,t (k, m)))λ exp(ψm + ξ n,t (m))
8 for asset-level desirabilities:

δn,t (ι, ℓ) = α log( pt (ι, ℓ) exp(et (n) − et (ι))) + κn,t (ι, ℓ), for n, ι, ℓ = 1, 2,

2 for the law of motion of wealth:


!
pt (ι, ℓ)
an,t = an,t−1 ∑ ∑ xn,t−1 (ι, ℓ) pt−1 (ι, ℓ) exp(−∆et (ι)) + sn,t , for n = 1, 2,
ℓ ι ≥0

and 4 for market clearing:

pt (ι, ℓ) exp(−et (ι))qt (ι, ℓ) = ∑ an,t xn,t (ι, ℓ), for ι, ℓ = 1, 2.


n

Finally, we define an auxiliary variable representing the U.S. net


foreign assets (NFA):

n f at = ∑ a1,t x1,t (2, ℓ) − ∑ a2,t x2,t (1, ℓ).


ℓ ℓ

We can use the portfolio choices, asset-level desirabilities, and


the law of motion of wealth to solve for the endogenous variables
xn,t (ι, ℓ), δn,t (ι, ℓ), and an,t . With N = 2 countries and L = 2 asset
classes, NL + ( N − 1) = 5 endogenous asset prices and exchange
rates remain to be solved, but only NL = 4 market clearing con-
ditions are available. Therefore, our system is underdetermined. In
a traditional macro model, we have additional restrictions on the
exchange rate from the goods market clearing. While this is an inter-
esting component to incorporate into the asset demand system, here
we use different assumptions to close the model. We will consider
three different settings in our numeric exercise, highlighting the dif-
ference between fixed and flexible exchange rate regimes as well as
the difference between symmetric and asymmetric settings between
the U.S. and the foreign country. In each of these settings, we have
exactly the same number of restrictions as the number of endogenous
asset prices and exchange rates.

Electronic copy available at: https://ssrn.com/abstract=4668578


240 jiang – lecture notes on international finance

9.E.2 Calibration and Results (TODO)

9.E.3 Other Works


In this section, we considered a simple two-country, two-asset class
demand system model to study the effects of the demand shocks on
the exchange rate and the asset prices. The reality is much richer. For
example, one feature we omit in our model is the currency denom-
ination of the assets, which exhibits important heterogeneity across
issuers and across investors [Maggiori, Neiman, and Schreger, 2020].
Moreover, the investor base could be further decomposed by different
sectors, such as banks, mutual funds, insurance companies and pen-
sion funds, and so on [Fang, Hardy, and Lewis, 2022, Faia, Salomao,
and Veghazy, 2022, Bergant, Milesi-Ferretti, and Schmitz, 2023, Zhou,
2023]. For example, our modeling of the bilateral holdings abstracts
away from the roles of financial intermediaries, which may act more
than just a pass-through.

Electronic copy available at: https://ssrn.com/abstract=4668578


10
Market Segmentation and Financial Intermediation

Summary

• We consider two forms of asset market segmentation and financial intermediation.

• First, asset markets can be segmented between countries. Local households may not be able to
directly hold foreign assets, and they rely on financiers to intermediate the cross-border capital
flows. The financiers face financial constraints, which connect the portfolio imbalances they
have to absorb to the risk premium they charge on foreign assets and currencies for providing
this intermediation service.

• Second, asset markets can be segmented within a country. Some households may not have
access to the financial markets, which disconnects the aggregate consumption from the pricing
of financial assets and exchange rates.

10.A A Model of International Financial Intermediation

In this section, we consider an international real business cycle


(IRBC) model in which home and foreign households cannot trade
any assets directly with each other, and specialized financiers are
required to intermediate the cross-border capital flows. This model
is adapted from Itskhoki and Mukhin [2021], whose key mechanism
builds on Jeanne and Rose [2002], Gabaix and Maggiori [2015]. This
model extends our baseline model in Chapter 1 in two ways. First, it
replaces the endowment economy with a production economy, with
standard features such as capital accumulation. Second, and more
importantly, it models segmented markets and financial intermedia-
tion.

10.A.1 Households
The households’ preferences are similar to those we considered in
Section 6.B. Home and foreign households have preferences over

Electronic copy available at: https://ssrn.com/abstract=4668578


242 jiang – lecture notes on international finance

consumption and work. The home households’ expected lifetime


utility is

" #
1 1
E0 ∑ δ t
1 −
c
γ
− ℓ1+1/ν .
t =0
1−γ t 1 + 1/ν t

Here we consider a more general aggregation function for con-


sumption goods. The home households’ consumption bundle is
composed of home and foreign goods:
η 1/η
h i
ct = α1−η c H,t + (1 − α)1−η c F,t
η
.

The Cobb-Douglas aggregation function we considered in the base-


line model is a special case obtained when η = 0. Here we consider a
positive η to a more realistic degree of substitutability between home
and foreign goods.
We use the home households’ consumption bundle as the numéraire.
Recall that pt is the price of the home good in this numéraire. We use
k t to denote the capital stock, zt to denote the households’ gross
investment, qt to denote the rental rate of capital. The home house-
holds’ budget constraint is

(wt ℓt + qt k t ) + exp(rt−1 )b H,t−1 = ct + zt + b H,t ,

where the households’ labor income wt ℓt and capital income qt k t re-


place their endowment in the baseline model. Moreover, we depart
from the baseline model by assuming (1) that the markets are seg-
mented, and the households can only hold their domestic risk-free
bond, and (2) that the households can invest in their capital stock.
Capital accumulates according to the standard rule:
κ (∆k t+1 )2
k t +1 = (1 − d ) k t + z t − ,
2 kt
(∆k )2
where d denotes the depreciation rate and κ2 kt+t 1 denotes the
quadratic adjustment cost.
Similarly, the foreign households maximize their expected lifetime
utility:

" #
1 1
E0 ∑ δ t ( c ∗ )1− γ − (ℓ∗ )1+1/ν ,
t =0
1−γ t 1 + 1/ν t

where their consumption bundle is defined as


h i1/η
c∗t = α1−η (c∗F,t )η + (1 − α)1−η (c∗H,t )η .

Likewise, the foreign households can only hold their domestic


risk-free bond. Their budget constraint is

(wt∗ ℓ∗t + q∗t k∗t ) + exp(rt∗−1 )b∗F,t−1 = c∗t + z∗t + b∗F,t .

Electronic copy available at: https://ssrn.com/abstract=4668578


market segmentation and financial intermediation 243

10.A.2 Firms
We consider very simple firms in this model. In the home country,
the firms use capital k t , labor ℓt , and intermediate input which is xt
units of the home consumption bundle to produce yt units of the
home goods:

(1− ϑ ) 1− ϕ ϕ
 
yt = exp( at )kϑt ℓt xt .

The firms are competitive, unlike the firms with market power
that we considered in Section 6. Their revenue pt yt is split among the
three factors of production according to their marginal products:

ℓt wt = (1 − ϑ)(1 − ϕ) pt yt ,
k t q t = ϑ (1 − ϕ ) p t y t ,
xt = ϕpt yt .

10.A.3 Segmented Markets and Financial Intermediation


As noted in the beginning of this section, market segmentation and
the financial intermediary are the key departure from the standard
IRBC models. Itskhoki and Mukhin [2021] consider two additional
classes of agents: the financiers and the noise traders.
The noise traders buy and sell currencies to satisfy their liquidity
demand. We assume that there is a measure n of identical noise
traders. In period t, they hold nt units of the home risk-free bond and
n∗t units of the foreign risk-free bond. They have a zero-cost portfolio,
which means their positions in the home and foreign bonds have the
same market value with opposite signs:

exp(−rt )nt = − exp(−et ) exp(−rt∗ )n∗t . (10.1)

The size of their position is determined by an exogenous process:

exp(−rt )nt = n(exp(ψt ) − 1),

where

ψt+1 = (1 − ρψ )ψ̄ + ρψ ψt + ε ψ,t+1 .

There is also a measure m of identical financiers, who absorb the


excess demand or supply of bonds from the noise traders and the
households and earn a risk premium. They also hold a zero-cost
portfolio. In period t, they hold dt units of the home risk-free bond
and d∗t units of the foreign risk-free bond, which also satisfies the
zero-cost constraint:

exp(−rt )dt = − exp(−et ) exp(−rt∗ )d∗t . (10.2)

Electronic copy available at: https://ssrn.com/abstract=4668578


244 jiang – lecture notes on international finance

The financiers are myopic and maximize the CARA utility of their
return:
 
1
E0 − exp(−ω (1 − exp(rxt+1 ))d∗t ) ,
ω

def
where rxt+1 = ∆et+1 + rt − rt∗ is the home currency’s expected excess
return in log. Itskhoki and Mukhin [2021] consider the continuous-
time limit of this problem, in which the optimal solution is given
by

Et [rxt+1 ] + 21 vart (∆et+1 )


d∗t = − , (10.3)
ωvart (∆et+1 )

This solution is identical to the myopic investors’ optimal portfolio


choice we derived in Section 9.3. Eq. (10.3) connects the currency’s
expected excess return to the financiers’ portfolio position. These
financiers absorb excess positive or negative positions in the bond
market, and they charge a risk premium for providing this service.
In particular, if they have a positive position in the home bond and
a negative position in the foreign bond (i.e., d∗t < 0), they demand a
higher expected return on the home bond to earn a positive expected
return on their portfolio. Conversely, if they have a negative position
in the home bond and a positive position in the foreign bond (i.e.,
d∗t > 0), they demand a higher expected return on the foreign bond
to earn a positive expected return on their portfolio. In this way, their
optimality condition (10.3) implies a tight link between their portfolio
position and the currency’s expected return.
This mechanism of intermediation in the currency market follows
from Gabaix and Maggiori [2015], who provide a different micro-
foundation in which these liquidity traders are risk-neutral but face
a financial friction to incur portfolio imbalances, and this friction is
greater when the exchange rate is more volatile. This setting gives
rise to an expression similar to Eq. (10.3).
Using the financiers’ optimality condition and the market clear-
ing conditions that we will next specify, we can derive the currency
expected excess return in equilibrium.

Proposition 10.1. After log-linearization, the currency expected excess


return is given by

def b H,t exp(rt )


Et [rxt+1 ] = Et [∆et+1 + rt − rt∗ ] = χ1 ψt − χ2 .

The proof is in Appendix A.42. This proposition relates the re-
quired excess return to hold the home currency to the equilibrium
quantity of imbalances absorbed that the financiers need to absorb,

Electronic copy available at: https://ssrn.com/abstract=4668578


market segmentation and financial intermediation 245

which, by market clearing, is further determined by the bond de-


mand of the noise traders (ψt ) and that of the local households b H,t .
Specifically, a positive ψt shock means that the financiers have to ab-
sorb a large amount of the home bond. They would require a higher
compensation to hold the home currency, which increases the home
currency’s expected excess return. If the home households want to
take a large position on the home bond, then, the financiers can of-
fload some of their positions to the home households, which lower
the equilibrium position they have to take and thus the risk premium
they charge.
Moreover, we can derive an exchange rate accounting formula sim-
ilar to Proposition 4.2. In this log-linearized model, as we ignore the
second-order terms, we effectively attribute all currency expected ex-
cess returns to the Euler equation wedge as opposed to the currency
risk premium. We obtain
∞ ∞
et = ∑ Et [rt+ j − rt∗+ j ] + ∑ Et [wt+ j ] + ē, (10.4)
j =0 j =0

where the wedge wt+ j is equal to

wt+ j = ω · vart+ j (∆et+ j+1 )d∗t+ j ,

which can be interpreted as the financiers’ price of risk, ω, multiplied


with the quantity of risk they absorb, vart+ j (∆et+ j+1 )d∗t+ j . The home
currency is stronger (i.e., et is higher) when the financiers have to
absorb a large amount of the foreign bond (i.e., d∗t+ j is large) and
requires a higher compensation to hold the foreign currency, which
can be triggered by a negative ψt shock.

10.A.4 Market Clearing Conditions


Since the consumption ct , the investment zt , and the intermediate
input xt are all denominated in the home consumption bundle, it is
more convenient to first figure out the total amounts of home goods
(i.e., y H,t ) and foreign goods (i.e., y F,t ) that the home households and
firms acquire for these objectives. These home and foreign goods
are aggregated by the same CES function and satisfy the following
resource constraint:
η 1/η
h i
ct + zt + xt = α1−η y H,t + (1 − α)1−η y F,t
η
.

Likewise, we have the following resource constraint for the foreign


households and firms:
h i1/η
c∗t + z∗t + xt∗ = α1−η (y∗F,t )η + (1 − α)1−η (y∗H,t )η ,

Electronic copy available at: https://ssrn.com/abstract=4668578


246 jiang – lecture notes on international finance

and the home and foreign goods’ market clearing conditions can be
expressed as

yt = y H,t + y∗H,t ,
y∗t = y F,t + y∗F,t .

As for the bond market, the home and foreign bonds are in zero
net supply. Their market clearing conditions can be expressed as

0 = b H,t exp(rt ) + dt + nt ,
0 = b∗F,t exp(rt∗ ) + d∗t + n∗t .

Since the intermediary holds a zero-cost bond portfolio (see Eq.


(10.1)), and so does the noise traders (see Eq. (10.2)), and the bonds
are in zero supply, then, the market clearing condition implies that
the home and foreign households combined must also hold a zero-
cost bond portfolio:

b H,t = − exp(−et )b∗F,t .

10.A.5 Macro Synthesis


The exogenous variables are the productivity shocks and the noise
traders’ demand shocks:

( at , a∗t , ψt )∞
t =0 .

There are 27 endogenous variables in each period t:

(yt , y H,t , y F,t , ct , k t , zt , xt , ℓt , b H,t , qt , wt , rt , pt , y∗t , y∗H,t , y∗F,t , c∗t , k∗t , z∗t , xt∗ , ℓ∗t , b∗F,t , q∗t , wt∗ , rt∗ , p∗t , et )∞
t =0 ,

plus two auxiliary variables exp(mt+1 ) and exp(m∗t+1 ) that denote the
home and foreign SDFs/marginal utility growth:

u ′ ( c t +1 )
exp(mt+1 ) = δ ,
u′ (ct )
u′ (c∗ )
exp(m∗t+1 ) = δ ′ t+∗ 1 .
u (ct )

The model implies the following 28 equations in each period, one


of which is redundant because the market clearing adds up to the
sum of households’ budget constraints. These 28 equations include 2
consumption aggregation equations,

η 1/η
h i
ct + zt + xt = α1−η y H,t + (1 − α)1−η y F,t
η
,
h i1/η
c∗t + z∗t + xt∗ = α1−η (y∗F,t )η + (1 − α)1−η (y∗H,t )η ,

Electronic copy available at: https://ssrn.com/abstract=4668578


market segmentation and financial intermediation 247

2 goods market clearing conditions,

yt = y H,t + y∗H,t ,
y∗t = y F,t + y∗F,t ,

6 optimality conditions for within-period consumption and labor


choices,

η 1/η −1 1−η η −1
h i
α1−η y H,t + (1 − α)1−η y F,t
η
α y H,t = pt ,
η 1/η −1
h i
η −1
α1−η y H,t + (1 − α)1−η y F,t (1 − α)1−η y F,t = p∗t exp(−et ),
η

ℓ1/ν
t = c− γ
t wt ,
h i1/η −1
α1−η (y∗F,t )η + (1 − α)1−η (y∗H,t )η α1−η (y∗F,t )η −1 = p∗t ,
h i1/η −1
α1−η (y∗F,t )η + (1 − α)1−η (y∗H,t )η (1 − α)1−η (y∗H,t )η −1 = pt exp(et ),
(ℓ∗t )1/ν = (c∗t )−γ wt∗ ,

8 equations that govern firm production and factor prices,

(1− ϑ ) 1− ϕ ϕ
yt = (exp( at )kϑt ℓt ) xt ,
ℓt wt = pt (1 − ϑ)(1 − ϕ)yt ,
k t q t = p t ϑ (1 − ϕ ) y t ,
xt = pt ϕyt ,
y∗t = (exp( a∗t )(k∗t )ϑ (ℓ∗t )(1−ϑ) )1−ϕ ( xt∗ )ϕ ,
ℓ∗t wt∗ = p∗t (1 − ϑ)(1 − ϕ)y∗t ,
k∗t q∗t = p∗t ϑ (1 − ϕ)y∗t ,
xt∗ = p∗t ϕy∗t ,

4 equations that govern capital accumulation,

κ (∆k t+1 )2
z t = k t +1 − (1 − d ) k t + ,
2 kt
" 
c t +1 − γ
 !#
∆k t+1 ∆k t+2 κ ∆k t+2 2
 
1+κ = Et δ q t +1 + 1 − d + κ + ,
kt ct k t +1 2 k t +1
κ (∆k∗t+1 )2
z∗t = k∗t+1 − (1 − d)∗ k t + ,
2 k∗t
  !2  
 ∗ −γ
∆k∗ c t +1 ∆k∗ ∆k∗t+2
1 + κ t∗+1 = Et δ q∗t+1 + 1 − d + κ t+2 + κ  ,
kt c∗t k∗t+1 2 k∗t+1

Electronic copy available at: https://ssrn.com/abstract=4668578


248 jiang – lecture notes on international finance

3 Euler equations for households and the intermediary,


" 
c t +1 − γ
 #
1 = Et δ exp(rt ) , (10.5)
ct
c∗t+1 −γ
"   #

1 = Et δ exp(rt ) , (10.6)
c∗t
b H,t exp(rt )
Et [∆et+1 + rt − rt∗ ] = χ1 ψt − χ2 , (10.7)

2 budget constraints for households,

(wt ℓt + qt k t ) + exp(rt−1 )b H,t−1 = ct + zt + b H,t ,


(wt∗ ℓ∗t + q∗t k∗t ) + exp(rt∗−1 )b∗F,t−1 = c∗t + z∗t + b∗F,t ,

and 1 bond market clearing condition,

b H,t = − exp(−et )b∗F,t . (10.8)

10.B Comparing Segmented Markets with Convenience Yields

The market segmentation mechanism we develop in the previous


section is in fact very similar to the convenience yield model that
we considered in Section 7.C. In the model with market segmenta-
tion, the households’ and the intermediaries’ (log-linearized) Euler
equations (10.5)–(10.7) imply the following risk-sharing condition:

b H,t exp(rt )
χ1 ψt − χ2 = Et [∆et+1 − (mt+1 − m∗t+1 )], (10.9)

which deviates from the complete-market case 0 = ∆et+1 − (mt+1 −
m∗t+1 ) in Section 1.C by introducing a wedge on the left-hand side
and by introducing the expectation operator so that the condition
only holds on average.
If we replace these equations with the following Euler equations
from the model with convenience yields:

exp(−λ H,t ) = Et [exp(mt+1 + rt )] ,


exp(−λ F,t ) = Et [exp(mt+1 − ∆et+1 + rt∗ )] ,
(10.10)
exp(−λ∗F,t ) = Et exp(m∗t+1 + rt∗ ) ,
 

exp(−λ∗H,t ) = Et exp(m∗t+1 + ∆et+1 + rt ) ,


 

change the bond market clearing condition to include the govern-


ment’s supply:

b̄t = b H,t + b∗H,t ,


b̄t∗ = bF,t + b∗F,t ,

Electronic copy available at: https://ssrn.com/abstract=4668578


market segmentation and financial intermediation 249

and modify the households’ budget conditions accordingly to include


both home and foreign bonds, we are very close to recovering the
convenience yield model. More precisely, we can log-linearize the
above equations in the convenience yield model, and obtain

−(λ∗H,t − λ H,t ) = −(λ∗F,t − λ F,t ) = Et ∆et+1 − (mt+1 − m∗t+1 ) .


 
(10.11)

Both Eq. (10.9) in the segmented market model and Eq. (10.11) in
the convenience yield model introduce a wedge to the risk-sharing
condition, which breaks the tight link between the exchange rate
and the households’ marginal utilities. The difference is that, in the
model with market segmentation, the wedge is driven by the port-
folio imbalances born by the financiers, whereas in the model with
convenience yields, the wedge is driven by the households’ demand
for safe assets.

10.B.1 Comparing the Impulse Responses


To facilitate the comparison between the segmented market model
and the convenience yield model, we use similar economic set-up
and parameter choices. The two models have identical equilibrium
conditions characterizing the goods market clearing, consumption
and labor choices, production and investment decisions, and factor
prices as listed in Section 10.A.5.
Their only difference is in the asset market. The Euler equations
(10.5) to (10.7) in the segmented market model, reproduced below,
" 
c t +1 − γ
 #
1 = Et δ exp(rt ) ,
ct
c∗t+1 −γ
"   #

1 = Et δ exp(rt ) ,
c∗t
b H,t exp(rt )
Et [∆et+1 + rt − rt∗ ] = χ1 ψt − χ2 ,

are replaced by the Euler equations with convenience yields:

ω H b−
" 
c t +1 − γ
 σ
#
H,t
1 = Et δ exp(rt ) + −γ ,
ct ct
ω F b−
" 
c t +1 − γ
 σ
#
∗ F,t
1 = Et δ exp(−∆et+1 + rt ) + −γ exp(et ) ,
ct ct
− ∗ ∗ −
"  #
c∗t+1 γ
ω F (bF,t ) σ

1 = Et δ ∗ exp(rt∗ ) + ,
ct (c∗t )−γ
c∗t+1 −γ ω ∗H (b∗H,t )−σ + (c̄∗ )−γ θ H,t

"   #
1 = Et δ exp(∆et+1 + rt ) + exp(−et ) .
c∗t (c∗t )−γ

Electronic copy available at: https://ssrn.com/abstract=4668578


250 jiang – lecture notes on international finance

This set of equations, based on Jiang [2023a] that we discussed in


Section 7.C, is a simple specification of the general case described
by Eq. (10.10). It allows households to freely trade both home and
foreign bonds, but introduces the convenience yield wedges in their
bond valuations.
Moreover, the bond market clearing conditions are different: the
convenience yield model assumes positive net supplies, and replaces
Eq. (10.8) in the segmented market model by

b H,t + b∗H,t = b̄,


b∗F,t + bF,t = b̄∗ .

The household budget constraints are also slightly different because


households can trade foreign bonds in the convenience yield model,
but not in the segmented market model.
Table 10.1 reports our parameter choices in the two models. The
common parameters describe the real economy which is common
to both models. The segmented market parameters describe the pa-
rameters related to the intermediated bond market in the segmented
market model. These parameters follow from Itskhoki and Mukhin
[2021]. The convenience yield parameters describe the parameters re-
lated to the bond demand in the convenience yield model, and follow
from Jiang [2023a].
Figure 10.1 reports the impulse responses to the ψt shock in
the segmented market model and to the λ H,t shock in the conve-
nience yield model. A period represents a quarter, and we plot the
responses in the first 25 years. We calibrate the magnitude of the
shocks such that they generate the same amount of dollar apprecia-
tion in the first period. We choose a persistence parameter of 0.97 for
both the noise trading shock and the convenience yield shock, which
will make the exchange rate close to a random walk.
Let us begin with the segmented market model. We consider a
negative ψt shock, which increases the amount of foreign bond that
the financiers have to hold. As the financiers require a higher com-
pensation to hold the foreign currency, the dollar appreciates and the
dollar’s expected excess return decreases, which are shown in the
first row of Figure 10.1. The U.S. bond also becomes relatively more
expensive, which lowers its yield relative to the foreign yield.
This is good news for the U.S. households. The second row of
Figure 10.1 reports the impulse responses in consumption and pro-
duction. Real dollar appreciation increases the U.S. households’
purchasing power, which increases their relative consumption while
incentivizing the foreign households to supply more labor and pro-
duce more. As a result, the U.S. consumption increases relative to the
foreign consumption, while the U.S. labor and production decline.

Electronic copy available at: https://ssrn.com/abstract=4668578


market segmentation and financial intermediation 251

The U.S. trade balance tb also declines as a result.


The third row of Figure 10.1 reports the impulse responses in bond
holdings. Because the households can only hold the domestic bond,
the U.S. holdings of the foreign bond bF and the foreign holdings
of the U.S. bond b∗H are zero. The noise trading shock also triggers
a persistent increase in the foreign households’ savings b∗F in the
foreign bond and a persistent decline in the U.S. households’ savings
b H in the U.S. bond. This has to be understood in conjunction with
the households’ investment decisions, which we will discuss in the
fifth row of this figure. As the foreign households are investing less
in the physical capital, more of their savings are channeled into the
local bond market.
We can also back out the implied convenience yields from the seg-
mented market model that satisfy the households’ Euler equations,
which we report in the fourth row. Specifically, we can define the

Table 10.1: Parameter Values.


Parameter Notation Value
Common parameters
Subjective discount factor δ 0.99
Relative risk averse γ 2.0
Frisch elasticity ν 1
Intermediate goods share ϕ 0.5
Capital share in value added ϑ 0.3
Depreciation rate d 0.02
Capital adjustment cost κ 50
Trade home bias α 0.93
Trade elasticity of substitution η 0.33
segmented market parameters
Transmission from shock to UIP χ1 1
Coefficient of NFA on UIP χ2 0.001
Volatility of financial shock σψ 0.01
Persistence of financial shock ρψ 0.97
Convenience yield parameters
Bond demand curvature σ 3
Home investor’s utility from home bond ωH 0.0658
Home investor’s utility from foreign bond ωF 0.000264
Foreign investor’s utility from home bond ω ∗H 0.0667
Foreign investor’s utility from foreign bond ω F∗ 0.0171
Persistence of bond demand shock ρθ ∗H 0.97
Volatility of bond demand shock σθ ∗H 0.024

Electronic copy available at: https://ssrn.com/abstract=4668578


252 jiang – lecture notes on international finance

SM,∗
wedges λSM
F,t and λ H,t as
" 
c t +1 − γ
 #

exp(−λSM = Et δ
F,t ) exp(−∆et+1 + rt ) ,
ct
∗ −γ
"  #
∗ c
exp(−λSM,
H,t ) = Et δ
t +1
exp(∆et+1 + rt ) ,
c∗t

and the other two wedges for domestic bond holdings, λSM
H,t and

λSM,
F,t , are always zero because the agents can freely trade their local

3 15 0 0

-0.2
2 10
-0.5
-0.4
1 5
-0.6
-1
0 0
-0.8

-1 -5 -1.5 -1
20 40 60 80 100 20 40 60 80 100 20 40 60 80 100 20 40 60 80 100

8 2 2 2

6 1
0
0
4 0
-2
2 -1
-4 -2
0 -2
-6
-2 -3 -4
20 40 60 80 100 20 40 60 80 100 20 40 60 80 100 20 40 60 80 100

0.2 0.6 0 0.15

0.15 0.4
-0.2 0.1
0.1 0.2
-0.4 0.05
0.05 0

0 -0.2 -0.6 0
20 40 60 80 100 20 40 60 80 100 20 40 60 80 100 20 40 60 80 100

2 0.6 0
0.2
-0.2
1.5
0.15 0.4
-0.4
1
0.1 -0.6
0.2
0.5 0.05 -0.8

0 0 0 -1
20 40 60 80 100 20 40 60 80 100 20 40 60 80 100 20 40 60 80 100

3 10 0 0

8 -0.5 -20
2
6 -1 -40

4 -1.5 -60
1
2 -2 -80

0 0 -2.5 -100
20 40 60 80 100 20 40 60 80 100 20 40 60 80 100 20 40 60 80 100

Figure 10.1: Impulse Responses to Euler


Equation Wedges in Segmented Market
and Convenience Yield Models.

Electronic copy available at: https://ssrn.com/abstract=4668578


market segmentation and financial intermediation 253

bonds by construction. A noise trader shock that appreciates the U.S.


dollar increases the shadow convenience yield that the foreign house-

holds derive from the U.S. bond, λSM, H,t , and decreases the shadow
convenience yield that the U.S. households derive from the foreign
bond, λSMF,t . These shadow convenience yield adjustments imply that
the foreign households must act as if they find it highly desirable
to hold the U.S. bond in order to justify why the equilibrium excess
return on the dollar is so low.
Finally, in the fifth row of Figure 10.1, we report the impulse re-
sponses in the capital, investment, and capital rental price. In re-
sponse to the noise trading shock, the U.S. households increase their
investment in the physical capital relative to the foreign households,
which leads to a higher capital stock k relative to k∗ and a higher
investment z relative to z∗ . As the marginal product of capital is
decreasing as the capital stock increases, the capital rental price q
also declines relative to q∗ . These patterns are driven by a discount
rate channel: as the U.S. households experience a higher consump-
tion upon the arrival of the noise trading shock, they expect a lower
consumption growth rate in the future periods, which lowers their
discount rate and makes investment more attractive. As a result,
investment and consumption move in tandem.
Now, let us turn to the convenience yield model. We consider a
∗ shock, which increases the foreign households’ demand
positive θ H,t
for the U.S. bond. This shock increases the convenience yield that the
foreign households derive from the U.S. bond, which also appreciates
the dollar, lowers the dollar’s expected excess return, and lowers the
dollar bond’s yield. As we discussed in Section 7.C, this is also good
news for the U.S. households, as it increases the U.S. consumption.
When we introduce labor and production in this model, dollar ap-
preciation also increases the foreign households’ labor supply and
production in this model, and decreases the U.S. trade balance. The
consumption dynamics also leads the U.S. households to increase
investment in the physical capital, and the convenience yield shock
also increases the foreign households’ convenience yield λ∗H on the
U.S. bond and decreases the U.S. households’ convenience yield λ F
on the foreign bond.
In this way, the segmented market model and the convenience
yield model generate very similar impulse responses in the exchange
rate, interest rate, consumption, production, trade balance, and in-
vestment. The main difference lies in the bond market. In the conve-
nience yield model, the foreign households increase their holdings of
the U.S. bond b∗H , whereas the U.S. households substitute towards the
foreign bond. In the segmented market model, the households do not
have access to the bond market in the other country, and the noise

Electronic copy available at: https://ssrn.com/abstract=4668578


254 jiang – lecture notes on international finance

traders instead accumulate a long position on the U.S. bond. So, if we


want to distinguish between the two models, it is important to look
at the quantities and flows in the bond market, which is a promis-
ing direction for future research. For example, do debt holdings by
different sectors and by different countries vary systematically with
the business cycles? And do they align with the exchange rate move-
ments as predicted by either model?
Moreover, the segmented market model and the convenience yield
model also generate different long-run dynamics. For example, while
the dollar exchange rate and the U.S.-foreign consumption differ-
ential increase persistently upon the shock’s arrival in both models,
they eventually decline below the steady-state level in the segmented
market model, but not in the convenience yield model. To under-
stand this difference, let us recall the net foreign assets (NFA) dynam-
ics (9.4), reproduced below:

n f at − n f at−1 = tbt + (ibt + cgt ).

In both models, the NFA shifts upon the shock’s arrival, but it needs
to revert to the steady-state level in the long run, which requires
adjustments in either the trade balance tb or the financial flow ib + cg
in subsequent periods. Formally, since limk→∞ Et [n f at+k ] = n f a, we
can iterate the NFA dynamics (9.4) and obtain

n f at = n f a − ∑ [tbt+k + (ibt+k + cgt+k )],
k =1

which means that if the U.S. runs a negative NFA today, it has to
be offset by positive trade balances or by positive financial gains
relative to the foreign country [Gourinchas and Rey, 2007a]. In the
segmented market model, this is achieved only by the trade channel:
the negative U.S. trade balance in the near term is offset by a positive
trade balance in the long term.1 1
In Itskhoki and Mukhin [2021], the
In the convenience yield model, the valuation effects matter: the model is linearized around a steady
state with zero bond positions, which
negative U.S. NFA is offset by higher financial gains that the U.S. means that the valuation effects ib + cg
earn from holding the foreign bond relative to what the foreign are equal to zero in the first order. In
extensions with non-zero NFA or non-
country earns from holding the U.S. bond, which is precisely the zero gross positions in the steady state,
seigniorage revenue that we characterize in Section 7.C. In this case, the valuation effects can matter in the
the U.S. does not have to run positive trade balances in the long term first order.

to pay back their trade deficits in the short term. Another place to
see this difference is in the NFA dynamics that we plot in the last
panel of Figure 10.1. While the U.S. households run trade deficits
in both models, they accumulate negative NFA in the segmented
market model, which has to be paid back by trade surpluses in the
long term. In comparison, the U.S. households do not accumulate a
large negative NFA in the convenience yield model, because the trade

Electronic copy available at: https://ssrn.com/abstract=4668578


market segmentation and financial intermediation 255

deficits are funded by the seigniorage revenue. As such, the seg-


mented market model and the convenience yield model emphasize
trade vs. financial drivers of the NFA dynamics.
Finally, different microfoundations of the exchange rate wedge
can have different policy implications. Under the segmented market
view, the government can stabilize the exchange rate by affecting
the financiers’ balance sheets or their expectation of the currency
risk [Itskhoki and Mukhin, 2023]. Under the convenience yield view,
exchange rate intervention can take the form of regulating the supply
of safe assets, which can be done by either fiscal or monetary policies
[Jiang, Krishnamurthy, and Lustig, 2020a].

10.B.2 The Cyclicality of the Wedge

Given the similarity between the segmented market model and the
convenience yield model, how do we think about them? We make
three points. First, they both suggest that the wedge in the Euler
equations for cross-country bond holdings is an important ingredient
to understand exchange rate dynamics. This wedge can have differ-
ent microfoundations, but they have similar effects that disentangle
the exchange rate movement from the households’ marginal utilities.
Second, the convenience yield mechanism provides an economic
interpretation for the cyclicality of the wedge, because it tends to
move in the direction that appreciates the dollar in flight-to-safety
episodes. As a result, the wedge does not just produce exchange
rate disconnect, it also makes the dollar more counter-cyclical to the
global business cycles than other currencies. Moreover, the conve-
nience yield interpretation also offers empirical measurement of the
wedge, as we discussed in Section 4.C.
Third, is the bond convenience yield on the dollar bond all there
is to the wedge-based view of the exchange rate dynamics? Let us
return to the Euler equations (10.10) with slightly different labeling:

exp(−λt ) = Et [exp(mt+1 + rt )] ,
exp(−ξ t ) = Et [exp(mt+1 − ∆et+1 + rt∗ )] ,
exp(−ξ t∗ ) = Et exp(m∗t+1 + rt∗ ) ,
 

exp(−λ∗t ) = Et exp(m∗t+1 + ∆et+1 + rt ) ,


 

which emphasizes that λt and λ∗t are the non-pecuniary utilities that
home and foreign households derive from the U.S. bond, whereas ξ t
and ξ t∗ are the non-pecuniary utilities that home and foreign house-
holds derive from the foreign bond. Assuming joint normality, Jiang,
Krishnamurthy, and Lustig [2023a] show the following result.

Electronic copy available at: https://ssrn.com/abstract=4668578


256 jiang – lecture notes on international finance

Proposition 10.2. The exchange rate’s conditional cyclicality can be ex-


pressed as

covt (mt+1 − m∗t+1 , ∆et+1 ) = vart (∆et+1 ) + (λ∗t − λt ) − (ξ t∗ − ξ t ). (10.12)

The proof is presented in Appendix A.43. What is remarkable


about this result is that the wedges λt , λ∗t , ξ t , ξ t∗ enter the determi-
nation of the exchange rate cyclicality in their levels as opposed to
their covariances with the SDFs, so that even a constant wedge can
affect the exchange rate cyclicality. This is because the presence of the
wedges endogenously requires the exchange rate cyclicality to adjust
to be consistent with the Euler equations.
If our objective is to generate a zero or negative covariance be-
tween the SDF differential and the exchange rate movement, i.e.,
covt (mt,t+1 − m∗t,t+1 , ∆st+1 ) ≤ 0, which is motivated by the finding
that the exchange rate is acyclical or even procyclical with respect to
the local business cycles [Backus and Smith, 1993], then, our usual
specification of the dollar convenience yield, under which the foreign
households derive a higher non-pecuniary utility from the U.S. bond
than the U.S. investors, i.e.,

λ∗t − λt > 0,

will not suffice. In fact, holding the exchange rate variance constant,
a higher convenience yield differential would increase the covariance
between the SDF differential and the exchange rate movement.
We need the foreign households to also derive higher convenience
yield on the foreign bond than the U.S. households, i.e.,

ξ t∗ − ξ t > 0,

which could lower the exchange rate cyclicality and even make it
negative, so that the foreign currency depreciates when the foreign
marginal utility is high. This condition can be interpreted as the U.S.
households finding it very undesirable to hold the foreign bond,
which can result from a strong home bias. In fact, we can rewrite Eq.
(10.12) as

covt (mt,t+1 − m∗t,t+1 , ∆st+1 ) = vart (∆st+1 ) + (λ∗t + ξ t ) − (ξ t∗ + λt ),

where (λ∗t + ξ t ) are the Euler equation wedge for cross-country


investments from the perspectives of both households, and (ξ t∗ + λt )
are the Euler equation wedge for domestic investments from the
perspectives of both households. A stronger home bias in the form of
a negative convenience yield for cross-country investments in either
country can make the exchange rate more pro-cyclical with respect to
the local business cycles.

Electronic copy available at: https://ssrn.com/abstract=4668578


market segmentation and financial intermediation 257

Given this discussion, what should a realistic specification of the


Euler equation wedges look like? A positive convenience yield on
the U.S. bond is still helpful for generating a countercyclical dollar
exchange rate to global business cycles, as well as for matching other
stylized facts about the dollar’s specialness, but a higher degree of
home bias is also needed to capture the cyclicality of foreign curren-
cies with respect to their local business cycles. In other words, not
only do we need the foreign households to really like the U.S. bond,
but we also need the U.S. households to dislike the foreign bonds.
Of course, home bias is an endogenous outcome, which could
be driven by many factors in the underlying economy and financial
markets. The segmented market set-up we considered in this chapter
is potentially one such factor. Another interpretation of the home
bias is finance repression, which could also lead the domestic agents
to prefer the home bond to the foreign bond and lower the relative
desirability of the cross-country investments. For a concrete example,
if a country’s regulators change the legal framework to make its
local bond a better collateral, or a better asset for meeting liquidity
requirements, or an asset with more favorable tax treatments, then,
the local bond will become more attractive to the local agents.

10.B.3 Segmented Markets vs Incomplete Markets

Segmented markets are related to but different from incomplete


markets. The literature on the international asset market is moving
towards a more precise and refined definition of what assets are
available and which agents can trade them.
Incomplete markets usually refer to the case in which the asset
space does not span all the shocks, but all agents can trade the avail-
able assets. For this reason, we also use the term incomplete spanning
to describe this case. In Section 5.A, we considered one such example
in which home and foreign households can only trade risk-free bonds
which do not span all the shocks.
In comparison, segmented markets usually refer to the case in
which, in addition to the possibility that the asset space does not
span all the shocks, not all agents can trade the available assets. In
Section 10.A, we considered one such example in which the house-
holds can only trade the local bond, whereas the financiers can trade
both countries’ bonds. In this case, the markets are segmented and
intermediated.
Note that these two cases are not mutually exclusive. In reality,
markets can be both incomplete and segmented. In the recent litera-
ture, Sandulescu, Trojani, and Vedolin [2021], Chernov, Haddad, and
Itskhoki [2023] study the implications of market segmentation using

Electronic copy available at: https://ssrn.com/abstract=4668578


258 jiang – lecture notes on international finance

model-free approaches. Both papers show that market segmenta-


tion breaks the strong link between exchange rates and SDFs, which
helps address salient features of international asset returns. In par-
ticular, markets in which trading in various instruments (including
bonds) is intermediated by various financial actors are likely to be
a very relevant case, a point that is in broad agreement with a large
macro-finance literature.

10.B.4 Other Works

Bruno and Shin [2015a,b] emphasize the role of global banks, who
tap dollar funding from financial centers and lend to local banks in
foreign countries. When the dollar appreciates, currency mismatch
weakens the local banks’ balance sheets and contracts their risk-
taking capacity. As a result, the dollar’s strength impacts global fi-
nancial conditions and generates dollar shortage in foreign countries.
Greenwood, Hanson, Stein, and Sunderam [2020], Gourinchas, Ray,
and Vayanos [2022] develop models with financiers who trade in both
the currency market and the home and foreign long-term bond mar-
kets. The variation in their risk-taking capacity leads to comovements
in exchange rates and bond term premia.
The deviation from covered interest rate parity as we discussed in
Section 2.A.6 is also a natural place to consider market segmentation
and financial intermediation. The key mechanism usually involves
constrained financiers and other sectors’ imbalanced hedging de-
mand [Borio, Iqbal, McCauley, McGuire, and Sushko, 2018, Andersen,
Duffie, and Song, 2019, Avdjiev, Du, Koch, and Shin, 2019, Liao and
Zhang, 2020, Amador, Bianchi, Bocola, and Perri, 2020, Cenedese,
Della Corte, and Wang, 2021, Rime, Schrimpf, and Syrstad, 2022]
The presence of financial frictions also makes it possible for the
policy interventions to improve outcomes and mitigate frictions. See
Fanelli and Straub [2020], Bocola and Lorenzoni [2020], Itskhoki and
Mukhin [2022]. In the context of the model in Section 10.A, if the pol-
icymaker can lower the exchange rate volatility, or at least convince
the financiers that the policymaker is committed to interventions that
lower the exchange rate volatility, then, the financiers will require a
lower currency risk premium to absorb a given amount of portfolio
imbalances, which, by Eq. (10.4), will indeed make the exchange rate
less volatile.
Empirically, there is a large literature that shows intermediary
balance sheet quantities do matter to asset prices [Adrian, Etula,
and Muir, 2014, He, Kelly, and Manela, 2017], which is either loosely
or precisely related to the equilibrium conditions in models with
financial intermediaries.

Electronic copy available at: https://ssrn.com/abstract=4668578


market segmentation and financial intermediation 259

10.C A Model of Domestic Financial Intermediation

Next, we consider a different type of intermediation, which focuses


on the market segmentation within a country. To be clear, the 2008
global financial crisis has spurred a large macro-finance literature on
domestic financial intermediation. What we focus on here is only one
specific type that is useful for thinking about exchange rate dynam-
ics.
This model is a two-country version of Hassan [2013]. We use
country-specific tradable goods instead of non-tradable goods to
bring the model closer to the baseline model in Section 1.A. This
model can be seen as a simplified way of specifying the monetary
model by Alvarez, Atkeson, and Kehoe [2002, 2009].

10.C.1 Households
We again consider two countries: home and foreign. There is a unit
mass of households in each country. We deviate from the baseline
model in Section 1.A by assuming that some households do not have
access to the financial markets. Specifically, we assume that each
country has two types of households: active and inactive.
Active households can trade a complete set of state-contingent
claims in international markets. They behave like the households in
the complete-market version of the baseline model in Section 1.C. Let
ϕ ∈ (0, 1) denote the fraction of these households in each country. We
use ( a) to denote the active households. The active households in the
home country maximize their expected lifetime utility

" #
E0 ∑ δt u(ct (a)) ,
t =0

and their consumption is a Cobb-Douglas aggregate of home and


foreign goods:

ct ( a) = (c H,t ( a))α (c F,t ( a))1−α .

Inactive households have identical preferences. We use (i ) to de-


note the inactive households. They maximize their expected lifetime
utility

" #
E0 ∑ δt u(ct (i)) ,
t =0

and their consumption is a Cobb-Douglas aggregate of home and


foreign goods:

ct (i ) = (c H,t (i ))α (c F,t (i ))1−α .

Electronic copy available at: https://ssrn.com/abstract=4668578


260 jiang – lecture notes on international finance

They cannot trade any financial claims. Instead, they cede their en-
dowments to the active households and receive a stochastic transfer,
which is denoted as τt for each inactive household in the home coun-
try and τt∗ for each inactive household in the foreign country. Inactive
households can still participate in the goods market, where they use
their transfers to purchase home and foreign goods.
Because the active and inactive households in the home country
have identical preferences, they face the same within-period problem
and will choose the same consumption bundle, which is again iden-
tical to the solution in Section 1.A. Therefore, we can use the home
consumption bundle to refer to the numéraire shared by both types
of households.
Let pt denote the price of home goods in the home consumption
bundle, let at−1 ( a) denote the financial wealth held by the active
households, and let exp(rta ) denote the gross return on this wealth
portfolio. The home active households’ budget constraint can be
expressed as

pt yt + ϕat−1 ( a) exp(rta ) = ϕ(ct ( a) + at ( a)) + (1 − ϕ)τt ,

and the home inactive households’ budget constraint can be ex-


pressed as

(1 − ϕ)τt = (1 − ϕ)ct (i ).

The active and inactive households in the foreign country are


similarly specified. This model simplifies to the complete-market
version of the baseline model in Section 1.C when the share of active
households approaches 1, i.e., ϕ → 1.

10.C.2 The Social Planner’s Solution


Because the active households have access to the complete markets,
the First Welfare Theorem holds for them. We can again use the
social planner approach to derive the equilibrium allocation. Specif-
ically, the social planner maximizes a weighted sum of the active
households’ welfare:

" #
E0 ∑ δt (πu(ct ) + (1 − π )u(c∗t )) .
t =0

If we only consider the active agents from home and foreign coun-
tries, the transfer to the inactive households takes away resources
available to them. Effectively, the resources that the social planner al-
locates between the active households in home and foreign countries
are equal to the total endowment minus the transfer. Thus, we can

Electronic copy available at: https://ssrn.com/abstract=4668578


market segmentation and financial intermediation 261

write the social planner’s resource constraints as

yt − (1 − ϕ)(c H,t (i ) + c∗H,t (i )) = ϕ(c H,t ( a) + c∗H,t ( a)),


y∗t − (1 − ϕ)(c F,t (i ) + c∗F,t (i )) = ϕ(c F,t ( a) + c∗F,t ( a)),

where we moved the inactive households’ consumption to the left-


hand side in order to emphasize what is available to the active house-
holds.
The social planner’s Lagrangian is

"
E0 ∑ δt (πu(ct (a)) + (1 − π )u(c∗t (a)))
t =1

+ ∑ ζ H,t (yt − ϕ(c H,t ( a) + c∗H,t ( a)) − (1 − ϕ)(c H,t (i ) + c∗H,t (i )))
t =1

#
+ ∑ ζ F,t (y∗t − ϕ(c F,t ( a) + c∗F,t ( a)) − (1 − ϕ)(c F,t (i ) + c∗F,t (i ))) ,
t =1

which implies the following standard first-order conditions

c F,t ( a) 1−α
 
t ′
w.r.t. c H,t ( a): δ πu (ct ( a))α = ϕζ H,t ,
c H,t ( a)

c∗F,t ( a)
w.r.t. c∗H,t ( a): δt (1 − π )u′ (c∗t ( a))(1 − α) = ϕζ H,t ,
c∗H,t ( a)
c H,t ( a) α
 
w.r.t. c F,t ( a): δt πu′ (ct ( a))(1 − α) = ϕζ F,t ,
c F,t ( a)
!1− α
∗ t ′ ∗
c∗H,t ( a)
w.r.t. c F,t ( a): δ (1 − π )u (ct ( a))α = ϕζ F,t .
c∗F,t ( a)

10.C.3 Macro Synthesis


The exogenous variables are the endowment shocks and the transfer
shocks:

(yt , y∗t , τt , τt∗ )∞


t =0 .

There are 14 endogenous variables in each period t:

(ct ( a), c H,t ( a), c F,t ( a), ct (i ), c H,t (i ), c F,t (i ), ζ H,t , c∗t ( a), c∗H,t ( a), c∗F,t ( a), c∗t (i ), c∗H,t (i ), c∗F,t (i ), ζ F,t )∞
t =0 .

The model implies the following 14 equations in each period,


which include 4 consumption aggregation equations,

ct ( a) = (c H,t ( a))α (c F,t ( a))1−α ,


c∗t ( a) = (c∗F,t ( a))α (c∗H,t ( a))1−α ,
ct (i ) = (c H,t (i ))α (c F,t (i ))1−α ,
c∗t (i ) = (c∗F,t (i ))α (c∗H,t (i ))1−α ,

Electronic copy available at: https://ssrn.com/abstract=4668578


262 jiang – lecture notes on international finance

2 social planner’s resource constraints,

yt = ϕ(c H,t ( a) + c∗H,t ( a)) + (1 − ϕ)(c H,t (i ) + c∗H,t (i )),


y∗t = ϕ(c F,t ( a) + c∗F,t ( a)) + (1 − ϕ)(c F,t (i ) + c∗F,t (i )),

4 active households’ first-order conditions,


  1− α
c F,t ( a)
w.r.t. c H,t ( a): δt πu′ (ct ( a))α = ϕζ H,t ,
c H,t ( a)

c∗F,t ( a)
w.r.t. c∗H,t ( a): t
δ (1 − π ) u ′
(c∗t ( a))(1 − α) = ϕζ H,t ,
c∗H,t ( a)
c H,t ( a) α
 
w.r.t. c F,t ( a): δt πu′ (ct ( a))(1 − α) = ϕζ F,t ,
c F,t ( a)
!1− α
c∗H,t ( a)
w.r.t. c∗F,t ( a): t ′ ∗
δ (1 − π )u (ct ( a))α = ϕζ F,t ,
c∗F,t ( a)

2 inactive households’ budget constraints,

τt = ct (i ),
τt∗ = c∗t (i ),

2 inactive households’ within-period solutions,

c H,t ( a) c (i )
= H,t ,
c F,t ( a) c F,t (i )
c∗H,t ( a) c∗H,t (i )
= ∗ .
c∗F,t ( a) c F,t (i )

This equation system does not contain equilibrium prices such


as goods prices and the exchange rate, because the solution to the
social planner’s problem directly implies the equilibrium allocation.
Once we have the equilibrium allocation, we can then use the within-
period solutions to solve for the equilibrium prices. As shown in
Section 1.C, the optimality conditions for the active households’
optimization problem imply that the goods prices are given by
 1− α
c F,t ( a)
pt = α ,
c H,t ( a)
!1− α
c∗H,t ( a)
p∗t =α ,
c∗F,t ( a)

and the exchange rate is given by

πu′ (ct ( a)) = (1 − π )u′ (c∗t ( a)) exp(et ).

Electronic copy available at: https://ssrn.com/abstract=4668578


market segmentation and financial intermediation 263

10.C.4 Linearized Solution


We use x̂ = log( x/ x̄ ) to denote the log deviation of a variable x
from its steady state x̄. For example, ĉt ( a) = log(ct ( a)/c̄t ( a)). The
exchange rate êt = et − ē is an exception, since it is already in log.
Assume the households have CRRA utility with parameter γ.
Below, we take the steady state variables as given and log linearize
the system, which includes 4 consumption aggregation equations,

ĉt ( a) = αĉ H,t ( a) + (1 − α)ĉ F,t ( a),


ĉ∗t ( a) = αĉ∗F,t ( a) + (1 − α)ĉ∗H,t ( a),
ĉt (i ) = αĉ H,t (i ) + (1 − α)ĉ F,t (i ),
ĉ∗t (i ) = αĉ∗F,t (i ) + (1 − α)ĉ∗H,t (i ),

2 social planner’s resource constraints,

ȳŷt = ϕ(c̄ H ( a)ĉ H,t ( a) + c̄∗H ( a)ĉ∗H,t ( a)) + (1 − ϕ)(c̄ H (i )ĉ H,t (i ) + c̄∗H (i )ĉ∗H,t (i )),
ȳ∗ ŷ∗t = ϕ(c̄ F ( a)ĉ F,t ( a) + c̄∗F ( a)ĉ∗F,t ( a)) + (1 − ϕ)(c̄ F (i )ĉ F,t (i ) + c̄∗F (i )ĉ∗F,t (i )),

4 active households’ first-order conditions,

w.r.t. c H,t ( a): − γĉt ( a) + (1 − α)ĉ F,t ( a) − (1 − α)ĉ H,t ( a) = ζ̂ H,t ,


w.r.t. c∗H,t ( a): − γĉ∗t ( a) + αĉ∗F,t ( a) − αĉ∗H,t ( a) = ζ̂ H,t ,
w.r.t. c F,t ( a): − γĉt ( a) + αĉ H,t ( a) − αĉ F,t ( a) = ζ̂ F,t ,
w.r.t. c∗F,t ( a): − γĉ∗t ( a) + (1 − α)ĉ∗H,t ( a) − (1 − α)ĉ∗F,t ( a) = ζ̂ F,t ,

2 inactive households’ budget constraints,

τ̂t = ĉt (i ),
τ̂t∗ = ĉ∗t (i ),

2 inactive households’ within-period solutions,

ĉ H,t ( a) − ĉ F,t ( a) = ĉ H,t (i ) − ĉ F,t (i ),


ĉ∗H,t ( a) − ĉ∗F,t ( a) = ĉ∗H,t (i ) − ĉ∗F,t (i ).

We can easily solve this system of linear equations. The solutions


are given by the following proposition. For convenience, we define
each country’s aggregate consumption as ct = ϕct ( a) + (1 − ϕ)ct (i )
and c∗t = ϕc∗t ( a) + (1 − ϕ)c∗t (i ).

Proposition 10.3. After the log-linearization, the equilibrium consumption


can be expressed as

ȳŷt − (1 − ϕ)c̄ H (i )τ̂t − (1 − ϕ)c̄∗H (i )τ̂t∗


   
ĉ H,t ( a)
 ĉ ( a)  ȳ∗ ŷ∗ − (1 − ϕ)c̄ (i )τ̂ − (1 − ϕ)c̄∗ (i )τ̂ ∗ 
 F,t F t
 = A −1  t F t 
,
 
 ∗
ĉ H,t ( a)  0 

ĉ F,t ( a) 0

Electronic copy available at: https://ssrn.com/abstract=4668578


264 jiang – lecture notes on international finance

where A is given by

A=
−(1 − ϕ)(1 − α)c̄ H (i ) ϕc̄∗H ( a) + (1 − ϕ)αc̄∗H (i ) −(1 − ϕ)αc̄∗H (i )
 
ϕc̄ H ( a) + (1 − ϕ)(1 − α)c̄ H (i )
 −(1 − ϕ)αc̄ F (i ) ϕc̄ F ( a) + (1 − ϕ)αc̄ F (i ) −(1 − ϕ)(1 − α)c̄∗F (i ) ϕc̄∗F ( a) + (1 − ϕ)(1 − α)c̄∗F (i )
,
 
α − 1 − γα (1 − γ)(1 − α) α + γ − γα ( γ − 1) α

 
( γ − 1) α α + γ − γα (1 − γ)(1 − α) α − 1 − γα
and the total consumption is given by

ĉt = ϕc̄t ( a)ĉt ( a) + (1 − ϕ)c̄t (i )ĉt (i ),


ĉ∗t = ϕc̄∗t ( a)ĉ∗t ( a) + (1 − ϕ)c̄∗t (i )ĉ∗t (i ),

the exchange rate is given by

êt = −γ(ĉt ( a) − ĉ∗t ( a)). (10.13)

The proof is in Appendix A.44. This model separates the exchange


rate and the aggregate consumption by introducing market segmen-
tation. On the one hand, as the active households have access to the
complete markets, their consumption growth is tightly connected to
the exchange rate movement as shown in Eq. (10.13). On the other
hand, while the inactive households are insulated from the financial
markets, their consumption still accounts for a large share of the ag-
gregate consumption, leading to a disconnect between the aggregate
consumption and the exchange rate.
We consider a special case in which π = 1/2, ȳ = ȳ∗ = 1, τ̄ =
τ̄ = αα (1 − α)1−α . In this case, all active and inactive households’

steady-state consumption of domestic goods is α and that of foreign


goods is 1 − α:

c̄ H ( a) = c̄ H (i ) = c̄∗F ( a) = c̄∗F (i ) = α,
c̄ F ( a) = c̄ F (i ) = c̄∗H ( a) = c̄∗H (i ) = 1 − α.

We further assume almost perfect home bias, i.e., α → 1, and, for


simplicity, the following correlation structure:
def
ρ = corrt (τ̂t+1 , ŷt+1 ) = corrt (τ̂t∗+1 , ŷ∗t+1 ),
def
σy2 = vart (ŷt+1 ) = vart (ŷ∗t+1 ),
def
στ2 = vart (τ̂t+1 ) = vart (τ̂t∗+1 ).

Then, we obtain the following result:


Proposition 10.4. Under these additional assumptions, the exchange rate
and the aggregate consumption growth differential can be expressed as
γ γ
êt = − (ŷt − ŷ∗t ) + (1 − ϕ)(τ̂t − τ̂t∗ ),
ϕ ϕ
ĉ∗t − ĉt = −(ŷt − ŷ∗t ),

Electronic copy available at: https://ssrn.com/abstract=4668578


market segmentation and financial intermediation 265

which implies

1 − (1 − ϕ)ρ σστy
corrt (êt+1 , ĉ∗t+1 − ĉt+1 ) = r .
2
1 + (1 − ϕ)2 σστ2 − 2(1 − ϕ)ρ σστy
y

In comparison, the active households’ consumption differential can be


expressed as

1 ∗
ĉ∗t+1 ( a) − ĉt+1 ( a) = [(ŷ − ŷt+1 ) − (1 − ϕ)(τ̂t∗+1 − τ̂t+1 )],
ϕ t +1

which leads to a Backus-Smith correlation of 1 between the exchange rate


and the active households’ consumption differential:

corrt (êt+1 , ĉ∗t+1 ( a) − ĉt+1 ( a)) = 1.

The proof is in Appendix A.45. If there is no inactive households,


i.e., ϕ = 1, then, the consumption growth and exchange rate are
perfectly correlated: corrt (êt+1 , ĉ∗t+1 − ĉt+1 ) = 1. We obtain the
complete-market benchmark in which a low home consumption
growth is associated with a high home marginal utility growth and
thus a home currency appreciation.
In contrast, if there is a large share of inactive households, i.e.,
ϕ ≪ 1, then, the Backus-Smith correlation corrt (êt+1 , ĉ∗t+1 − ĉt+1 )
may become negative. The precondition is to have a sufficiently high
correlation between the transfer shock and the endowment shock,
and a sufficiently high variance of the transfer shock relative to the
endowment shock. More precisely, we may interpret ρ σστy as the beta
of the transfer shock with respect to the local endowment shock.
When this beta is large enough, i.e., (1 − ϕ)ρ σστy > 1, we obtain a
negative Backus-Smith correlation.
In this case, the model resolves the Backus-Smith puzzle by re-
interpreting the cyclicality of the households’ marginal utilities: a
high endowment shock for the entire country is actually a bad, high
marginal utility state for the active households who are the marginal
investors in the currency market, even though it is good news for
the average household which includes the inactive households. As
a result, while the Backus-Smith correlation is still positive for the
active households’ consumption, it is negative for the aggregate con-
sumption. As this difference in the cyclicality of household marginal
utility is central to the model’s mechanism, it would be interesting to
empirically identify the active and inactive households and examine
the properties of their profits, losses, and consumption.
This view also implies that the active households to have the oppo-
site cyclicality in their marginal utility growth relative to the domestic

Electronic copy available at: https://ssrn.com/abstract=4668578


266 jiang – lecture notes on international finance

business cycle. These active households usually represent financial


intermediaries who invest in the risky assets on behalf of the inactive
households. According to Hassan [2013]’s interpretation, the active
households underwrite a nominal risk-free bond denominated in the
local currency to the inactive households. In good times, low inflation
makes the nominal bond valuable, which benefits the inactive house-
holds at the expense of the active households. If the active house-
holds are sufficiently levered, then, they could lose wealth in good
times, leading to a procyclical marginal utility with respect to the
domestic business cycle. Alvarez, Atkeson, and Kehoe [2002, 2009]
provide a microfoundation for this dynamics, in which the identities
of active and inactive households are endogenously determined by
the households’ holdings.

10.C.5 Other Works

Market segmentation and financial intermediation have been studied


extensively in the macro-finance literature. He and Krishnamurthy
[2013], Brunnermeier and Sannikov [2014] present two prominent
models of financial frictions in closed economies. In the international
literature, Chien and Naknoi [2015] propose a segmented market
model in which not all households can participate in the equity mar-
ket, but the participation rate is higher in the U.S. Maggiori [2017]
also considers a model with credit constraints for financial intermedi-
aries, in which the U.S. intermediaries are less constrained. In these
models, the U.S. as a whole loads up more aggregate risk than the
foreign country, which speaks to the asymmetry in global portfo-
lio allocation we discussed in Chapter 7. Chien, Lustig, and Naknoi
[2020] use a similar segmented market mechanism to resolve the
exchange rate volatility puzzle we discussed in Section 2.A.1, that
the exchange rate movement can be smooth even when the SDFs
are volatile and the macro fundamentals are only weakly correlated,
and Dou and Verdelhan [2015] use segmented markets to explain the
volatility of international capital flows. Fang and Liu [2021] also con-
sider a two-country model with within-country financial intermedia-
tion, with the key ingredient being the value-at-risk (VaR) constraint.
The model speaks quantitatively to the exchange rate volatility, cycli-
cality, risk premium, and the deviations from the covered interest rate
parity.
Jiang and Richmond [2023b] introduce global production network
to this model with active and inactive households. The transfer shock
from active to inactive households can be interpreted as a demand
shock that is transmitted in the global trade network, which, as we
saw in Section 3.C, plays an important role in shaping the comove-

Electronic copy available at: https://ssrn.com/abstract=4668578


market segmentation and financial intermediation 267

ments of exchange rates.

Electronic copy available at: https://ssrn.com/abstract=4668578


Appendix

Electronic copy available at: https://ssrn.com/abstract=4668578


A
Proof of Selected Results

A.1 Proposition 1.3 in Section 1.C

Proof. Recall the first-order conditions


  1− α
t ′ c F,t
w.r.t. c H,t : δ πu (ct )α = ζ H,t
c H,t

c∗F,t
w.r.t. c∗H,t : t
δ (1 − π ) u ′
(c∗t )(1 − α) = ζ H,t
c∗H,t
c H,t α
 
t ′
w.r.t. c F,t : δ πu (ct )(1 − α) = ζ F,t
c F,t
!1− α
c∗H,t
w.r.t. c∗F,t : δt (1 − π )u′ (c∗t )α = ζ F,t
c∗F,t

multiply the first and the last first-order conditions and divide it by
the product of the second and third first-order conditions to get

c F,t c H,t α2
=1
c∗F,t c H,t (1 − α)2
α c F,t 1 − α c H,t
⇒ ∗ =
1 − α c F,t α c∗H,t

Let
def α c F,t 1 − α c H,t
kt = =
1 − α c∗F,t α c∗H,t
α 1−α ∗
⇒ c H,t = k t c∗ , c F,t = k t c F,t
1 − α H,t α
Plugging the equations above into the market clearing conditions
yields
αk t 1−α
c H,t = yt , c∗H,t = yt ,
(1 − α) + αk t (1 − α) + αk t
(1 − α ) k t α
c F,t = y∗ , c∗F,t = y∗ ,
α + (1 − α ) k t t α + (1 − α ) k t t

Electronic copy available at: https://ssrn.com/abstract=4668578


270 jiang – lecture notes on international finance

.
Divide the first first-order condition by the second to get

c∗H,t
1− α
π u′ (ct ) α

c F,t
= 1.
1 − π u′ (c∗t ) 1 − α c H,t c∗F,t

where c H,t , c F,t , c∗H,t and c∗F,t are functions of k t . Hence, we can solve
k t via the implicit equation above.

A.2 Proposition 1.4 in Section 1.D

Proof. Under symmetric steady state, ȳ∗ = ȳ, ē = 0, c̄ H = c̄∗F , c̄ F = c̄∗H ,


and

c̄ H + c̄∗H = ȳ, c̄ F + c̄∗F = ȳ∗ ,


p̄c̄ H = αc̄, p̄∗ c̄∗F = αc̄∗ ,

which implies c̄ H = c̄∗F = αȳ, c̄∗H = c̄ F = (1 − α)ȳ, c̄ = c̄∗ =


αα (1 − α)1−α ȳ and p̄ = p̄∗ = αα (1 − α)1−α . The goods market clearing
conditions plus the equilibrium conditions (1.4) and (1.5) imply

pt yt = αct + (1 − α)c∗t exp(−et ),


p∗t y∗t = αc∗t + (1 − α)ct exp(et ).

Log linearize around the symmetric steady state to obtain

p̄ȳ(log pt − log p̄ + log yt − log ȳ) = αc̄(log ct − log c̄) + (1 − α)c̄∗ exp(−ē)(log c∗t − log c̄∗ − et + ē)
p̄∗ ȳ∗ (log p∗t − log p̄∗ + log y∗t − log ȳ) = αc̄∗ (log c∗t − log c̄∗ ) + (1 − α)c̄ exp(ē)(log ct − log c̄ + et − ē)

subtracting the second equation from the first and plugging in steady
states yields

p̄ȳ(log yt − log y∗t ) + p̄ȳ(log pt − log p∗t ) = (2α − 1)c̄(log ct − log c∗t ) − 2(1 − α)c̄et

recall that et = (2α − 1)tott , where tott = log pt − log p∗t + et , hence
log pt − log p∗t = (2 − 2α)/(2α − 1)et , and

2 − 2α
p̄ȳ(log yt − log y∗t ) + p̄ȳ et = (2α − 1)c̄(log ct − log c∗t ) − 2(1 − α)c̄et
2α − 1
2 − 2α
(2α − 1)(log ct − log c∗t ) − 2(1 − α)et − et = log yt − log y∗t
2α − 1

(2α − 1)(log ct − log c∗t ) − 2(1 − α) et = log yt − log y∗t
2α − 1

where we plug in p̄ȳ = c̄.

Electronic copy available at: https://ssrn.com/abstract=4668578


proof of selected results 271

A.3 Proposition 3.1 in Section 3.A

Proof. Plug in
h i √ √ q q q 
rxti/$
+1 − E t rx i/$
t +1 = − δ i − δ$ z w εw −
t t +1 γz i εi
t t +1 − γz $ $
ε
t t +1

in the expression of hmlt+1 yields

1 √ √ q 1
q q 
hmlt+1 − Et [hmlt+1 ] = −
NH ∑ δi
− δ $ w w
z t ε t +1 −
NH i∑
i i
γzt ε t+1 − γzt ε t+1$ $

i∈ H ∈H
1 √ √ q q 
w εw + 1
q
NL i∑ NL i∑
i − δ$ i εi $ $
+ δ z t t +1 γz t t +1 − γz ε
t t +1
∈L ∈L
q q q
1 1
q q
NH i∑ ∑ γzit εit+1
w i εi
= δtL − δtH zw ε
t t +1 − γz t t +1 +
∈H
N L i∈ L
q q q
w
= δtL − δtH zw
t ε t +1

q q
The country-specific shock terms, N1H ∑i∈ H γzit εit+1 and N1L ∑i∈ L γzit εit+1
averages out according to the law of large numbers. Hence,
√ √
δ $ − δi
q q 
i/$ i/$ i $ $
rxt+1 = rpt + q q (hmlt+1 − Et [hmlt+1 ]) − i
γzt ε t+1 − γzt ε t+1
δtL − δtH

which shows that βit is the carry trade loading on hmlt+1

A.4 Proposition 3.2 in Section 3.A

Proof. Recall that


  √ √ q q q 
∆ei/$
t +1 = − µ i
t − µ $
t − δ i − δ$ z w εw −
t t +1 γz i εi
t t +1 − γz $ $
ε
t t +1
  √ √  q  q q 
= −χ zit − z$t − δi − δ $ zw ε w
t t +1 − γz i εi
t t +1 − γz $ $
ε
t t +1

and
1 1
rti − rt$ = (χ − γ)(zit − z$t ) − (δi − δ$ )zw
t .
2 2
$
Note that zw i
t is uncorrelated zt and zt . Hence,
2 2
1 1 i
var (rti − rt$ )
= χ − γ var (zit − z$t ) + δ − δ$ var (zw
t )
2 4
 
1
cov(∆ei/$ i $ i $
t+1 , rt − rt ) = − χ χ − 2 γ var ( zt − zt )

cov(rxti/$ i $ i/$ i $ i $
+1 , rt − rt ) = cov ( ∆et+1 + rt − rt , rt − rt )
= cov(∆ei/$ i $ i $
t+1 , rt − rt ) + var (rt − rt )

Electronic copy available at: https://ssrn.com/abstract=4668578


272 jiang – lecture notes on international finance

Assuming zit and z$t are also independent,


var (zit − z$t ) = var (zit ) + var (z$t ).

A.5 Proposition 3.3 in Section 3.A

Proof. Recall that


  √ p q q q 
i/j j j j
∆et+1 = −χ zit − zt − δi − δ j zw ε w
t t +1 − γz i εi
t t +1 − γz ε
t t +1

j j
Note that zit , zt , zw i w
t , ε t+1 , ε t+1 , ε t+1 are independent for i ̸ = j. By the
law of large numbers,
j 1 1 j
cov(zit − zt , zit − zt ) = var (zit ) − var (zit ) + var (zt ) = var (zit )
N N
2 1 N
var (zit − zt ) = var (zit ) − var (zit ) + 2 ∑ var (zt ) = var (zit )
j
N N j =1
q  q  h i q 
cov( zit εit+1 , zit ) = E zit zit εit+1 − E zit E zit εit+1 = 0
q h i  hq i2
w w w 2 w (εw )
var ( zw ε
t t +1 ) = E z t ( ε t +1 ) − E z t t +1
h i
= E [zw w
t ] E ( ε t +1 )
2

= θw
q
var ( zit εit+1 ) = θ
Hence,
√
p  √ √ 
i/j
cov(∆et+1 , ∆eit+1 ) = χ2 var (zit ) + δj δi − δ θ w + γθ
δi −
√ √ 2
var (∆eit+1 ) = χ2 var (zit ) + δi − δ θ w + γθ
√ p 2
i/j j
var (∆et+1 ) = χ2 (var (zit ) + var (zt )) + δi − δ j θ w + 2γθ
i/j
cov(∆et+1 , ∆eit+1 )
φ=
var (∆eit+1 )
√ √  √ √ 
δi − δ δj − δ θw
= 1− √ √ 2 ,
2 i
χ var (zt ) + i w
δ − δ θ + γθ
φ2 var (∆eit+1 )
R2 = i/j
var (∆et+1 )
h √ √  √ √  i2
χ2 var (zit ) + δi − δ j δi − δ θ w + γθ
=  √ √ 2
 √ √ 2 .
j
χ2 var (zit ) + δi − δ θ w + γθ χ2 (var (zit ) + var (zt )) + δi − δ j θ w + 2γθ

Electronic copy available at: https://ssrn.com/abstract=4668578


proof of selected results 273

A.6 Proposition 3.4 in Section 3.B

Proof. The Euler equation for the consumption claim implies

1 = Et [exp(mt+1 + rtc+1 )],


1 1
0 = Et [mt+1 ] + Et [rtc+1 ] + vart (mt+1 ) + vart (rtc+1 ) + covt (mt+1 , rtc+1 ).
2 2

Define r0c = κ0c + µ g + µwc (1 − κ1c ). The log-linearization of the log


total wealth return around the long-run average wealth consumption
ratio can be expressed as

rtc+1 = κ0c + ∆ct+1 + log(wct+1 ) − κ1c log(wct )


 
= κ0c + µ g + xt + σt ε g,t+1 + µwc + Wx xt+1 + Wσ (σt2+1 − σ2 ) − κ1c µwc + Wx xt + Wσ (σt2 − σ2 )
   
= r0c + xt + σt ε g,t+1 + Wx (ρxt + σt φe ε x,t+1 ) + Wσ ϕ(σt2 − σ2 ) + ωε w,t+1 − κ1c Wx xt + Wσ (σt2 − σ2 )
= r0c + [1 + Wx (ρ − κ1c )] xt + Wσ (σt2 − σ2 )(ϕ − κ1c ) + σt ε g,t+1 + Wx σt φe ε x,t+1 + Wσ ωε w,t+1

which implies

Et [rtc+1 ] = r0c + [1 + Wx (ρ − κ1c )] xt + Wσ (σt2 − σ2 )(ϕ − κ1c )


rtc+1 − Et [rtc+1 ] = σt ε g,t+1 + Wx σt φe ε x,t+1 + Wσ ωε w,t+1
vart (rtc+1 ) = Et [(σt ε g,t+1 + Wx σt φe ε x,t+1 + Wσ ωε w,t+1 )2 ]
= Et [σt2 ε2g,t+1 + 2σt ε g,t+1 Wx σt φe ε x,t+1 + 2σt ε g,t+1 Wσ ωε w,t+1 + Wx2 σt2 φe ε2x,t+1
+ 2Wx σt φe ε x,t+1 Wσ ωε w,t+1 + Wσ2 ω 2 ε2w,t+1 ]
= σt2 + Wx2 φ2e σt2 + Wσ2 ω 2 + 2νWx φe σt2

Define µs = θ log δ + (θ − 1)r0c − ψθ µ g . Recall that θ (1 − 1/ψ) =


1 − γ. Then,

θ
mt+1 = θ log δ − ∆c + (θ − 1)rtc+1
ψ t +1
θ
= µs − ( xt + σt ε g,t+1 )
ψ
n o
+ (θ − 1) [1 + Wx (ρ − κ1c )] xt + Wσ (σt2 − σ2 )(ϕ − κ1c ) + σt ε g,t+1 + Wx σt φe ε x,t+1 + Wσ ωε w,t+1
 
θ
= µs + − + (θ − 1)[1 + Wx (ρ − κ1c )] xt − γσt ε g,t+1
ψ
+ {Wσ (ϕ − κ1c )(θ − 1)} (σt2 − σ2 ) + (θ − 1) {Wx σt φe ε x,t+1 + Wσ ωε w,t+1 }

Electronic copy available at: https://ssrn.com/abstract=4668578


274 jiang – lecture notes on international finance

which implies

mt+1 − Et [mt+1 ] = −γσt ε g,t+1 + (θ − 1) {Wx σt φe ε x,t+1 + Wσ ωε w,t+1 }


 
θ
Et [mt+1 ] = µs + − + (θ − 1)[1 + Wx (ρ − κ1c )] xt + {Wσ (ϕ − κ1c )(θ − 1)} (σt2 − σ2 )
ψ
h 2 i
vart (mt+1 ) = Et −γσt ε g,t+1 + (θ − 1) {Wx σt φe ε x,t+1 + Wσ ωε w,t+1 }
n o
= γ2 σt2 + (θ − 1)2 Wx2 φ2e σt2 + Wσ2 ω 2 − 2νγ(θ − 1)Wx φe σt2
covt (mt+1 , rtc+1 ) = Et [(rtc+1 − Et [r0c ])(mt+1 − Et [mt+1 ])]
= −γσt2 + Wx2 φ2e (θ − 1)σt2 + Wσ2 (θ − 1)ω 2 + ν (−γ + θ − 1) Wx φe σt2

Now plugging all the components into the Euler equation,

0 = r0c + µs
θ2 n 2 2 2
 
o 1 θn o
+ Wx φe σ + Wσ2 ω 2 + θ − + [1 + Wx (ρ − κ1c )] xt + 2Wσ (ϕ − κ1c ) + θWx2 φ2e (σt2 − σ2 )
2 ψ 2
1
+ (1 − γ)2 σt2 + θν(1 − γ)Wx φe σt2
2

As this equation holds for any xt and σt2 ,

θ2 n 2 2 2 o
r0c + µs + Wx φe σ + Wσ2 ω 2 − Wσ θ (ϕ − κ1c )σ2 = 0 (A.1)
2  
1 c
θ − + [1 + Wx (ρ − κ1 )] = 0 (A.2)
ψ
1 θ n o
(1 − γ )2 + 2Wσ (ϕ − κ1c ) + θWx2 φ2e + θν(1 − γ)Wx φe = 0 (A.3)
2 2

which implies

1
1− ψ
Wx =
κ1c − ρ
(1 − γ)(1 − ψ1 )  φ2e 2νφe

Wσ = + c +1
2(κ1c − ϕ) c
(κ1 − ρ ) 2 κ1 − ρ

Furthermore, we can solve additionally for r0c , κ0c , κ1c and µwc , using
Eq. (A.1) and

eµwc
κ0c = − log(eµwc − 1) + µwc
eµwc−1
eµwc
κ1c =
eµwc − 1
r0 = κ0c + µ g + µwc (1 − κ1c )
c

Electronic copy available at: https://ssrn.com/abstract=4668578


proof of selected results 275

A.7 Proposition 3.5 in Section 3.B

Proof. Assuming complete markets, the real exchange rate movement


is

∆et+1 = mt+1 − m∗t+1


   
θ c θ ∗ c∗
= − ∆ct+1 + (θ − 1)rt+1 − − ∆ct+1 + (θ − 1)rt+1
ψ ψ
θ
∆c∗t+1 − ∆ct+1 + (θ − 1)(rtc+1 − rtc+ ∗

= 1)
ψ
θ n ∗ o
= ( xt − xt ) + (σt ε∗g,t+1 − σt∗ ε g,t+1 )
ψ
+ (θ − 1){[1 + Wx (ρ − κ1c )]( xt − xt∗ ) + Wσ (ϕ − κ1c )(σt2 − σx,t
∗2
) + (σt ε g,t+1 − σt∗ ε∗g,t+1 )
+ Wx (σt φe ε x,t+1 − σt∗ φe ε∗x,t+1 ) + Wσ ω (ε w,t+1 − ε∗w,t+1 )}
 
θ
= − + (θ − 1)[1 + Wx (ρ − κ1 )] ( xt − xt∗ ) + {Wσ (ϕ − κ1c )(θ − 1)} (σt2 − σt∗2 )
c
ψ
− γ(σt ε g,t+1 − σt∗ ε∗g,t+1 ) + (θ − 1)Wx φe (σt ε x,t+1 − σt∗ ε∗x,t+1 ) + (θ − 1)Wσ ω (ε w,t+1 − ε∗w,t+1 )

and the currency risk premium is

1
Et [rxt+1 ] = (vart (m∗t+1 ) − vart (mt+1 ))
2
1 h 2 ∗2 i
= γ σt + (θ − 1)2 Wx2 φ2e σt∗2 + (θ − 1)2 Wσ2 ω 2 − 2νγ(θ − 1)Wx φe σt∗2
2
1h 2 2 i
− γ σt + (θ − 1)2 Wx2 φ2e σt2 + (θ − 1)2 Wσ2 ω 2 − 2νγ(θ − 1)Wx φe σt2
2
1h 2 i
= γ + (θ − 1)2 Wx2 φ2e − 2νγ(θ − 1)Wx φe (σt∗2 − σt2 )
2

Also, from the Euler equations for risk-free bonds,

1
rt = −Et [mt+1 ] − vart (mt+1 )
  2  
θ c c 2 2
= − µs + − + (θ − 1)[1 + Wx (ρ − κ1 )] xt + {Wσ (ϕ − κ1 )(θ − 1)} (σt − σ )
ψ
1 n o
− γ2 σt2 + (θ − 1)2 Wx2 φ2e σt2 + (θ − 1)2 Wσ2 ω 2 + νγ(θ − 1)Wx φe σt2
2
1
rt∗ = −Et [m∗t+1 ] − vart (m∗t+1 )
  2  
θ c ∗ c ∗2 2
= − µs + − + (θ − 1)[1 + Wx (ρ − κ1 )] xt + {Wσ (ϕ − κ1 )(θ − 1)}(σt − σ )
ψ
1 n 2 ∗2 o
− γ σt + (θ − 1)2 Wx2 φ2e σt∗2 + (θ − 1)2 Wσ2 ω 2 + νγ(θ − 1)Wx φe σt∗2
2

Electronic copy available at: https://ssrn.com/abstract=4668578


276 jiang – lecture notes on international finance

also compute the interest rate differential as


 
θ

rt − rt = − + (θ − 1)[1 + Wx (ρ − κ1 )] ( xt∗ − xt )
c
ψ
 
1 2 1
+ {Wσ (ϕ − κ1 )(θ − 1)} + γ + (θ − 1) Wx φ − νγ(θ − 1)Wx φ (σt∗2 − σt2 )
c 2 2 2
2 2
From here, we assume the following for the derivations below:

covt (ε∗g,t+1 , ε g,t+1 ) ̸= 0


covt (ε∗x,t+1 , ε x,t+1 ) ̸= 0
covt (σt∗ , σt ) ̸= 0

Finally, the Backus-Smith coefficient:

covt (∆et+1 , ∆c∗t+1 − ∆ct+1 )


β BS
t =
vart (∆c∗t+1 − ∆ct+1 )
1
= [covt (γ(σt∗ ε∗g,t+1 − σt ε g,t+1 ), σt∗ ε∗g,t+1 − σt ε g,t+1 )
vart (σt∗ ε∗g,t+1 − σt ε g,t+1 )
− covt ((θ − 1)Wx (σt∗ φe ε∗x,t+1 − σt φe ε x,t+1 ) + (θ − 1)Wσ ω (ε∗w,t+1 − ε w,t+1 ), σt∗ ε∗g,t+1 − σt ε g,t+1 )]
(θ − 1)Wx
= γ− covt ((σt∗ φe ε∗x,t+1 − σt φe ε x,t+1 ), (σt∗ ε∗g,t+1 − σt ε g,t+1 ))
vart (σt∗ ε∗g,t+1
− σt ε g,t+1 )
νφe (θ − 1)Wx  
= γ− ∗ ∗ σt2 + σt∗2
vart (σt ε g,t+1 − σt ε g,t+1 )

where vart (σt∗ ε∗g,t+1 − σt ε g,t+1 ) = σt2 + σt∗2 − 2σt σt∗ covt (ε g,t+1 , ε∗g,t+1 ).

A.8 Proposition 3.6 in Section 3.C

Proof. Within period t, the social planner’s Lagrangian is


! ! !
N N N N  
∑ π ∑ vij log c j + φ a (ℓ ) ∏(x j ) − ∑ ci + xi − d
i i i i i θ i wij j j i
i =1 j =1 j =1 j =1
i i i
+ χ (ℓ̄ − ℓ )

which implies the following first-order conditions:


j j
w.r.t. ci : π j v ji (ci )−1 = φi (A.4)
j j
w.r.t. xi : φ j x j w ji ( xi )−1 = φi (A.5)
i i i i −1 i
w.r.t. ℓ : φ x θ (ℓ ) =χ (A.6)

Substitute into the market clearing condition:


N  
φi x i − di = ∑ π j v ji + φ j x j w ji .
j =1

Electronic copy available at: https://ssrn.com/abstract=4668578


proof of selected results 277

Define γi = φi xi . It can be interpreted as the value of country


i’s total output, because the Lagrangian multiplier φi can be inter-
preted as the price of country i’s intermediate goods in a common
numéraire. The market clearing condition becomes
N  
γi − di = ∑ π j v ji + γ j w ji
j =1

γ − d = V′π + W′γ
γ = ( I − W ′ )−1 (V ′ π + d) = H ′ π + ( I − W ′ )−1 d.

Then, the log production is


!
N
φi x i
i i
log x = log a + θ log ℓ + i
∑ wij log φj
wij
j =1
N
γi
 
= log a + θ log ℓ + ∑ wij log
i i
wij
j =1 γ j /x j
x
log x = κ + log a + (1 − θ ) log γ + W (log x − log γ)
= ( I − W )−1 (κ x + log a − θ log γ) + log γ

where κ x = θ log ℓi + ∑ N
j=1 wij log wij is a vector of constants.
Define H = V ( I − W )−1 . The log consumption of active house-
holds is
!
N π i vij j
log c = ∑ vij log
i
x
j =1 γj
N  
log c = ∑ vij log π i vij + V log x − V log γ
j =1
N  
= ∑ vij log π i vij + H (κ x + log a − θ log γ)
j =1

= κ c + H (log a − θ log( H ′ π + ( I − W ′ )−1 d))

Let L = ( I − W ′ )−1 . Denote Li the i-th row of L. Since


∂{log( H ′ π + Ld)}i ∂ log({ H ′ π }i + Li d) 1
= = ′
L
∂d ∂d { H π }i + Li d
we can apply a first-order approximation at d = 0, and obtain
1
{log(γ + Ld)}i ≈ log{ H ′ π }i + Ld
{ H ′ π }i
θ θ
We define H′ π as a diagonal matrix whose element (i, i ) is { H ′ π }i
.
Then,
 
θ
log c = κ + H log a − ′ ( I − W ′ )−1 d
c

Electronic copy available at: https://ssrn.com/abstract=4668578


278 jiang – lecture notes on international finance

A.9 Proposition 3.7 in Section 3.C

Proof. The covariance of consumption growth follows directly from


Eq. (3.12):

j
cov(∆ log cit , ∆ log ct ) = C(i, j).

The moments of exchange rates follow from the covariance of con-


sumption growth. In particular, the covariance between the changes
in currency base factors is

!
N N
1 1
∑ ∑
j j
cov(∆eit+1 , ∆et+1 ) = cov ∆ckt+1 − ∆cit+1 , ∆ckt+1 − ∆ct+1
N k =1
N k =1

= C(i, j) − C(i ) − C( j) + κ e .

A.10 Proposition 3.8 in Section 3.C

Proof. The log SDF is given by:

 
c̄t
mit+1 = log δ
c̄t+1
= log δ − ∆ log c̄t+1

As shown in Section 1.B, the log interest rate and currency risk pre-
mium are given by:

1
rti = −Et [mit+1 ] − vart (mit+1 )
2
θ 1
= − log δ − HEt [∆ log at+1 ] + H ( I − W ′ )−1 Et [∆dt+1 ] − vart (∆ log c̄it+1 )
H′ π 2
1
= − log δ − C(i, i )
2
h i
def
rpi/$
t = Et rxti/$
+1

= Et [−ei/$ i $
t +1 ] + r t − r t
1 1
= − vart (mit+1 ) + vart (m$t+1 )
2 2
1
= (C($, $) − C(i, i ))
2

where we set Et [∆ log at+1 ] = Et [∆dt+1 ] = 0

Electronic copy available at: https://ssrn.com/abstract=4668578


proof of selected results 279

A.11 Proposition 3.9 in Section 3.D

Proof. The long term bond return and price are given by

P T
exp(−rt (h)h) = pt (h) = Et [exp(mt,t +h + mt,t+h )]
= Et [exp(mctP+h − mctP + mctT+h − mctT )]
"
h
1
= Et exp ∑ (− σP2 + σP ε Pt+k )
k =1
2
!#
h h h  
+ ∑ϕ h−k T
mc + ∑ϕ h−k
(t + k) log β + (ϕ h
− 1)mctT + ∑ ϕ h−k
σT ε Tt+k
k =1 k =1 k =1
!
1 − ϕh T t + h − ϕ h ( t + 1) ϕ − ϕh
= exp mc + − log β + (ϕh − 1)mctT
1−ϕ 1−ϕ (1 − ϕ )2
!
1 − ϕ2h 2 1 − ϕh
+ σ + σTP .
2 − 2ϕ2 T 1−ϕ

Hence,
!
1 − ϕh T t + h − ϕ h ( t + 1) ϕ − ϕh h T 1 − ϕ2h 2 1 − ϕh
−r t ( h ) h = mc + − log β + ( ϕ − 1 ) mc t + σ + σTP ,
1−ϕ 1−ϕ (1 − ϕ )2 2 − 2ϕ2 T 1−ϕ
!
1 − ϕh T t − ϕ h ( t + 1) ϕ − ϕh 1 − ϕ2h 2 1 − ϕh
 
log β
rt ( h) + h=− mc − − log β − (ϕh − 1)mctT − σ − σTP ,
1−ϕ 1−ϕ 1−ϕ (1 − ϕ ) 2 2 − 2ϕ2 T 1−ϕ

which implies that the constant δ̃ = β1/(1−ϕ) satisfies 0 < limh→∞ pt (h)/δ̃h <
∞, and the long-term bond yield satisfies

lim rt (h) + log δ̃ = 0,


h→∞

and
t
lim (rt (h) + log δ̃)h = − log β + mctT .
h→∞ 1−ϕ

which confirms the definition of transitory components of the SDF,


i.e.,

mctT = lim log δ̃(t + h) − log pt (h).


h→∞

Recall that

mc T ϕ log β log β
mctT = − + t − γ log ytT
1−ϕ 1−ϕ1−ϕ 1−ϕ

mc T ϕ log β log β
= − + t + ∑ ϕi σT ε Tt−i
1−ϕ 1−ϕ1−ϕ 1−ϕ i =0

plugging mctT into the bond return immediately yields Eq. (3.19).

Electronic copy available at: https://ssrn.com/abstract=4668578


280 jiang – lecture notes on international finance

A.12 Proposition 3.11 in Section 3.D.

Proof. The one-period risk free rate is given by

exp(−rt ) = Et [exp(mt,t+1 )]
= Et [exp(mctP+1 − mctP + mctT+1 − mctT )]
  
1 2 P T T T
= Et exp − σP + σP ε t+1 + mc + (ϕ − 1)mct + (t + 1) log β + σT ε t+1
2
 
T T 1 2
= exp mc + (ϕ − 1)mct + (t + 1) log β + σT + σTP ,
2
1
−rt = mc T + (ϕ − 1)mctT + (t + 1) log β + σT2 + σTP .
2

For long-term bond, the risk premium is

Et [log pt−1 (h − 1) − log pt (h) − rt ]


!
h −1 T ( t + 1 ) ϕ h − ( t + 2 ) ϕ h −1 ϕ h −1
=−ϕ mc + + log β
1−ϕ 1−ϕ
1
+(ϕh−1 − 1)(mc T + ϕmctT + (t + 1) log β) − (ϕh − 1)mctT − ϕ2(h−1) σT2 − ϕh−1 σTP
2
T T 1 2
+mc + (ϕ − 1)mct + (t + 1) log β + σT + σTP
  2
1 1 2( h −1) 2 h −1
= − ϕ σT + (1 − ϕ )σTP
2 2

Taking h → ∞ yields

1 2
lim Et [log pt−1 (h − 1) − log pt (h) − rt ] = σ + σTP .
h→∞ 2 T

The price of gdp claim at time t is given by

y
pt (h) = Et [exp(mt,t+h + log yt+h )]
= Et [exp(mctP+h − mctP + mctT+h − mctT + log yt+h )]
"
h
1
= Et exp ∑ (− σP2 + σP ε Pt+k )
k =1
2
h h h  
+ ∑ ϕh−k mcT + ∑ ϕh−k (t + k) log β + (ϕh − 1)mctT + ∑ ϕh−k σT ε Tt+k
k =1 k =1 k =1
!#
h h
+ log ytP + hµ + ∑ νP εPt+k + ϕh log ytT + ∑ ϕh−k νT εTt+k
k =1 k =1
!
1 − ϕh T t + h − ( t + 1) ϕ h
ϕ − ϕh
= exp mc + − log β + (ϕh − 1)mctT
1−ϕ 1−ϕ (1 − ϕ )2
!
1 − 2γ 2 1 − ϕ2h γ − 1 2 2 1 − ϕh γ − 1 2
   
P h T
+ log yt + hµ + ϕ log yt + hσP + σT + σTP ,
2γ2 2(1 − ϕ2 ) γ 1−ϕ γ

Electronic copy available at: https://ssrn.com/abstract=4668578


proof of selected results 281

where we plugged in νP = −(1/γ)σP , νT = −(1/γ)σT . Hence,

y y
Et [log pt+1 (h − 1) − log pt (h) − rt ]
!
h −1 ( t + 1 ) ϕ h − ( t + 2 ) ϕ h −1 ϕ h −1
= − ϕ mc + T
+ log β + (ϕh−1 − 1)(mc T + ϕmctT + (t + 1) β) − (ϕh − 1)mctT
1−ϕ 1−ϕ
1 − 2γ 2 ϕ2(h−1) γ − 1 2 2
2
h −1 γ − 1
  
− σ − σT − ϕ σTP − rt
2γ2 P 2 γ γ
" # " 2 #
2γ − 1 2 1 ϕ 2( h −1) γ − 1 2 2 h −1 γ − 1
  
= σ + − σT + 1 − ϕ σTP .
2γ2 P 2 2 γ γ

Taking h → ∞ yields

y y 2γ − 1 2 1 2
lim Et [log pt+1 (h − 1) − log pt (h) − rt ] = σ + σ + σTP .
h→∞ 2γ2 P 2 T

A.13 Proposition 4.3 in Section 4.B

Proof. Recall the within-in period solutions

ct ct
pt = α , p∗t exp(−et ) = (1 − α)
c H,t c F,t

and

c∗t c∗t
p∗t = α , pt exp(et ) = (1 − α) ,
c∗F,t c∗H,t

which combined characterize the exchange rate as

(1 − α)c∗t /c∗H,t αc∗t /c F,t


exp(et ) = = .
αct /c H,t (1 − α)ct /c F,t

Write the Lagrangian of social planner’s problem as follows

∞ ∞
" # " #
∑δ t
(u(ct ) + v(b H,t ; θt )) + (1 − π )E0 ∑δ t
u(c∗t ) + v(b∗H,t ; θt∗ )

L = πE0
t =0 t =0
∞ ∞
" # " #
+ E0 ∑ ζ H,t yt − c H,t − c∗H,t + ζ F,t y∗t − c F,t − c∗F,t + E0 ∑ ξ H,t b̄t − b H,t − b∗H,t
  
.
t =0 t =0

Electronic copy available at: https://ssrn.com/abstract=4668578


282 jiang – lecture notes on international finance

The first order conditions w.r.t. 6 endogenous variables are


ct
w.r.t. c H,t : πδt u′ (ct )α = ζ H,t
c H,t
c∗
w.r.t. c∗H,t : (1 − π )δt u′ (c∗t )(1 − α) ∗t = ζ H,t
c H,t
ct
w.r.t. c F,t : πδt u′ (ct )(1 − α) = ζ F,t
c F,t
c∗
w.r.t. c∗F,t : (1 − π )δt u′ (c∗t )α ∗t = ζ F,t
c F,t
w.r.t. b H,t : πδt v′ (b H,t ; θt ) = ξ H,t
w.r.t. b∗H,t : (1 − π )δt v′ (b∗H,t ; θt∗ ) = ξ H,t

with 3 resource constraint conditions, we can solve the 6 endogenous


variables and the 3 lagrangian multiplers. Divide the first foc con-
dition by the second one and plug in the within-period solution to
obtain the exchange rate as follows
πu′ (ct )
exp(et ) = .
(1 − π )u′ (c∗t )
which implies
∆et = mt − m∗t .
Besides, we are typically interested in the convenience yields, which
are given by Eq. (4.2) and (4.3), i.e.,
v′ (b H,t ; θt ) ξ
exp(−λt ) = 1 − ′
= 1 − t H,t
u (ct ) πδ u′ (ct )
′ ∗ ∗
v (b H,t ; θt )
exp(−λ∗t ) = 1 − exp(−et )
u′ (c∗t )
ξ H,t
= 1− exp(−et )
(1 − π )δt u′ (c∗t )
ξ H,t (1 − π )u′ (c∗t )
= 1− ∗
(1 − π ) δ t u ′ ( c t ) πu′ (ct )
ξ
= 1 − t H,t .
πδ u′ (ct )
Hence, λt = λ∗t . Finally, the last two first-order conditions imply
πv′ (b H,t ; θt ) = (1 − π )v′ (b∗H,t ; θt∗ ). This concludes the proof.

A.14 Lemma 5.1 in Section 5.A

Proof. (1) The first-order condition for the home household’s portfo-
lio choice problem (5.2) is:
γt
exp(−δt) = ζ H exp(mνH ,t ), (A.7)
ct

Electronic copy available at: https://ssrn.com/abstract=4668578


proof of selected results 283

where the Lagrange multiplier ζ H is such that the budget constraint


evaluated at the optimal consumption expenditure, ct , is satisfied
with equality:
Z T 
E exp(mνH ,t )(ct − pt yt )dt = w0 .
0

The home country SDF based on home household’s consumption is


then
γt c 0
exp(−δt) = exp(mνH ,t − mνH ,0 ) = exp(mνH ,t ).
γ0 ct
(2) Let λ∗t denote the market price of risk from the viewpoint of a
foreign consumer in the foreign numéraire:
σt′
 
1
λ∗t = r t + κ t + σt σ ′
− r ∗
∥σt ∥2 2 t t

Then, the set of foreign country SDFs can be represented as:

d exp(m∗ν∗ ,t ) = −rt∗ exp(m∗ν∗ ,t )dt − (λ∗t + νt∗ ) T exp(m∗ν∗ ,t )dZt ,


RT
with ν∗ ∈ R3 satisfying σt νt∗ = 0 for all t ∈ [0, T ] and 0 ∥νt∗ ∥2 dt <
∞.
Then, we can write foreign household’s static variational portfolio
choice problem in their local numéraire:
Z T 
∗ ∗
max

E exp (− δt ) γ log c t dt
ct 0
T
Z 
s.t. E exp(m∗ν∗ ,t )(c∗t − p∗t y∗t )dt ≤ w0∗ .
0

The first-order condition for the foreign household’s portfolio choice


problem is:
γ∗
exp(−δt) = ζ F exp(m∗ν∗ ,t ), (A.8)
c∗t F

where exp(m∗ν∗ ,t ) denotes the foreign country state price density that
F
bounds all the budget constraints, and the Lagrangian multiplier ζ F
is such that the budget constraint evaluated at the optimal consump-
tion expenditure c∗t is satisfied with equality
Z T 
E exp(m∗ν∗ ,t )(c∗t − p∗t y∗t )dt = w0∗ .
0 F

Then, the foreign country SDF based on foreign household’s con-


sumption is:
c0∗
exp(−δt) = exp(m∗ν∗ ,t − m∗ν∗ ,0 ) = exp(m∗ν∗ ,t ).
c∗t F F F

Electronic copy available at: https://ssrn.com/abstract=4668578


284 jiang – lecture notes on international finance

A.15 Proposition 5.1 in Section 5.A

Proof. (1) By no arbitrage, home household’s wealth in the home


numéraire at time t is:
Z T 
exp(mνH ,u )
w t = Et (cu − pu yu )du .
t exp( mνH ,t )

Define
T
Z 
exp(mνH ,u )
s t = Et pu yu du .
t exp(mνH ,t )
Assume the law of motion of st is given by

dst = µst st dt + σts st dZt

Making use of (A.7) in Lemma 5.1, we have


Z T
exp(−δt) − exp(−δT )

1 γu γt
wt + st = Et exp(−δu) du = .
exp(mνH ,t ) t ζH δ ζ H exp(mνH ,t )
(A.9)
Note that wt can be interpreted as the “bond wealth” and the total
wealth is wt plus the value of future endowment flow, which together
finance the future consumption. To find the optimal portfolios, we
apply Ito’s Lemma to (A.9).
exp(−δt) − exp(−δT )
 
γt
d ( wt + st ) = d
δ ζ H exp(mνH ,t )
 
δ ′ γt ωi3
= ( wt + st ) − + rt + (λt + νH,t ) (λt + νH,t ) + (λt + νH,t ) dt
1 − exp(−δ( T − t)) γt
 
γt ωi3
+ ( wt + st ) + (λt + νH,t )′ dZt .
γt
Matching the diffusion term with that in the dynamic budget con-
straint yields:
 
γt ωi3 ′
− xt wt σt = (wt + st ) + (λt + νH,t ) − st σts . (A.10)
γt
Notice that in incomplete markets, the matrix σt′ σt is not a full-rank
square matrix, and the above system of equations contains three
equations in one unknown. It has a solution if and only if its right-
hand side lies in the Span(σt ). This implies a restriction:
σ′ σt γt ωi3′ σ′ σt
   
st
I3 − t 2 + νH,t − I3 − t 2 (σts )′ = 0,
∥σt ∥ γt wt + st ∥σt ∥
σ′ σ
 
where we apply operator I3 − ∥σt ∥t2 to both sides of (A.10). Rear-
t
ranging the terms, we have:
σt′ σt
  
′ st s ′
νH,t = I3 − −ωi3 + (σ ) . (A.11)
∥σt ∥2 wt + st t

Electronic copy available at: https://ssrn.com/abstract=4668578


proof of selected results 285

To obtain the optimal portfolio of the home country household, we


σ′
apply the operator − |σ t|2 to both sides of (A.10).
t

σt′ st σts σt′


 
wt + st γt ωi3
xt = − + (λt + νH,t )′ 2
+
wt γt ∥σt ∥ wt ∥σt ∥2
!
wt + st ωσ3,t (rt∗ − κt + 12 σt σt′ − rt ) st σts σt′
=− − + .
wt ∥σt ∥2 ∥σt ∥2 wt ∥σt ∥2

For the foreign households, the first-order condition for the foreign
household’s static variational portfolio choice problem in the home
numéraire is:
γ∗
exp(−δt) = ζ̃ F exp(mν̃F ,t ), (A.12)
c̃∗t
where exp(mν̃F ,t ) denotes an appropriate home country state price
density that bounds all the budget constraints of the foreign house-
hold, and ζ̃ F is the Lagrange multiplier such that the budget con-
straint evaluated at the optimal consumption expenditure, c̃∗t , is satis-
fied with equality:
Z T 
E exp(mν̃F ,t )(c̃t − pt exp(−et )yt )dt = w̃0∗ .
∗ ∗ ∗
0

By no arbitrage, foreign household’s wealth in the home numéraires


at time t is:
Z T 
exp(mν̃F ,u ) ∗
w̃t∗ = Et (c̃t − p∗t exp(−et )y∗t )du .
t exp( mν̃F ,t )

Similarly, denote
T
Z 
exp(mν̃F ,u ) ∗
s̃∗t = Et pt exp(−et )y∗t du .
t exp(mν̃F ,t )
Assume the law of motion of s̃∗t is given by
∗ ∗
ds̃∗t = µs̃t s̃∗t dt + σts̃ s̃∗t dZt

Making use of (A.12), we have


γ∗ γ∗
Z T
exp(−δt) − exp(−δT )

∗ ∗ 1
w̃t + s̃t = Et exp(−δu) du = .
exp(mν̃F ,t ) t ζ̃ F δ ζ̃ F exp(mν̃F ,t )
(A.13)
To find the optimal portfolios, we apply Ito’s Lemma to (A.13),
 
∗ ∗ ∗ ∗ δ
d(w̃t + s̃t ) = (w̃t + s̃t ) − + rt + (λt + ν̃F,t ) (λt + ν̃F,t ) dt + (w̃t∗ + s̃∗t )(λt + ν̃F,t )′ dZt .

1 − exp(−δ( T − t))
Matching the diffusion term with that in the dynamic budget con-
straint yields

( xt∗ − 1)w̃t∗ σt = (w̃t∗ + s̃∗t )(λt + ν̃F,t )′ − s̃∗t σts̃ (A.14)

Electronic copy available at: https://ssrn.com/abstract=4668578


286 jiang – lecture notes on international finance

In incomplete markets, the matrix σt′ σt has a solution if and only if its
right hand side lies in Span(σt ). This entails a restriction:

s̃∗ σ′ σt
 

ν̃F,t = ∗ t ∗ I3 − t 2 (σts̃ )′ , (A.15)
w̃t + s̃t ∥σt ∥
σ′ σ
 
where we apply operator I3 − ∥σt ∥t2 to both sides of (A.14).
t
To get the optimal portfolio choice for the foreign household, we
σt′
apply the operator ∥σt ∥2
from right to both sides of (A.14).

w̃t∗ + s̃∗t σt′ s̃∗t σts̃ σt′
xt∗ − 1 = [( λ t + ν̃F,t ) ′
] −
w̃t∗ ∥σt ∥2 w̃t∗ ∥σt ∥2
 
w̃ ∗ + s̃∗ rt∗ − κ t + 21 σt σt′ − rt ∗
s̃∗t σts̃ σt′
xt∗ = 1 − t ∗ t − .
w̃t ∥σt ∥2 w̃t∗ ∥σt ∥2

A.16 Lemma 5.2 in Section 5.A

Proof. The representative agent’s utility evaluated at the aggregate


output is given by
Z T 
U (yt , y∗t ; πt ) = max
∗ ∗
E exp (− δt )( u ( c , c
H,t F,t ) + π t u ∗ ∗
( c , c
H,t F,t

)) dt
c H,t ,c F,t ,c H,t ,c F,t 0

s.t. c H,t + c∗H,t = yt ,


c F,t + c∗F,t = y∗t .

Then, the marginal utilities of the representative agent and the indi-
vidual agents evaluated at the optimum are related as

∇U (yt , y∗t ; πt ) = ∇u (c H,t , c F,t ) = πt ∇u∗ c∗H,t , c∗F,t .




From the first-order conditions of the home and foreign households,


we have
γt α γt ( 1 − α )
 
∇u (c H,t , c F,t ) = , ,
c H,t c F,t
!
γ ∗ (1 − α ) γ ∗ α
∇u∗ (c∗H,t , c∗F,t ) = , ∗ .
c∗H,t c F,t

Besides, by our previous results:

α (1 − α ) γt
c H,t = ct , c F,t = ct , exp(−δt) = ζ H exp(mνH ,t ),
pt p∗t exp(−et ) ct

we have

∇u (c H,t , c F,t ) = (exp(δt)ζ H exp(mνH ,t ) pt , exp(δt)ζ H exp(mνH ,t ) p∗t exp(−et )).

Electronic copy available at: https://ssrn.com/abstract=4668578


proof of selected results 287

Similarly, for the foreign household,


∇u∗ c F,t , c∗F,t = (exp(δt)ζ̃ F exp(mν̃F ,t ) pt , exp(δt)ζ̃ F exp(mν̃F ,t ) p∗t exp(−et )).


We substitute the solutions of c H,t , c F,t , c∗H,t , c∗F,t into the above gradi-
 
ents of ∇u c H,t , c∗H,t to get the state price density that bound all the
budget constraints of the home consumers.
exp(mνH ,t ) γt c p c γt
exp(mνH ,t ) = = exp(−δt) 0 = exp(−δt) 0 H,0
exp(mνH ,0 ) γ0 ct pt c H,t γ0
p 0 y 0 γt α + π t γ ∗ ( 1 − α )
= exp(−δt) (A.16)
pt yt γ0 α + π0 γ∗ (1 − α)
Therefore, by the expression of the state-price-density in A.16, the
price of the home stock in the home numéraire is
Z T 
1
st = Et exp(mνH ,u ) pu yu du (A.17)
exp(mνH ,t ) t
Z T 
exp(δt) pt yt ∗
= E t exp (− δu )( γ u α + π u γ ( 1 − α )) du
γt α + π t γ ∗ ( 1 − α ) t
 Z T 
exp(δt) pt yt 1 ∗
= ( exp (− δt ) − exp (− δT )) γ t α + E t [ exp (− δu ) π u γ ( 1 − α ) du ]
γt α + π t γ ∗ ( 1 − α ) δ t
1 − exp(−δ( T − t)) exp(δt) pt yt γ∗ (1 − α) T
 Z 
= pt yt + E t exp (− δu )( π u − π t ) du ,
δ γt α + π t γ ∗ ( 1 − α ) t

where we used the fact that γt is a martingale.


Similarly, we can then derive the home country state price density
that bounds all the budget constraints of the foreign consumers as:
exp(mν̃F ,t ) c̃∗ p∗ exp(−e0 )c∗F,0
exp(mν̃F ,t ) = = exp(−δt) 0∗ = exp(−δt) 0∗
exp(mν̃F ,0 ) c̃t pt exp(−et )c∗F,t
p0∗ exp(−e0 )y0∗ π0 γt (1 − α) + πt γ∗ α
= exp(−δt) . (A.18)
p∗t exp(−et )y∗t πt γ0 (1 − α) + π0 γ∗ α
Analogously, plugging in the expression of the state-price-density in
(A.18), we find the price of the foreign stock in the home numéraire
to be
Z T 
∗ 1 ∗ ∗
s̃t = Et exp(mνH ,u ) pu exp(−eu )yu du (A.19)
exp(mνH ,t ) t
exp(δt) p∗t exp(−et )y∗t
 Z T 
= E t exp(−δu)(γu (1 − α) + πu γ∗ α)du
γt ( 1 − α ) + π t γ ∗ α t
exp(δt) p∗t exp(−et )y∗t 1
 Z T 

= (exp(−δt) − exp(−δT ))γt (1 − α) + Et [ exp(−δu)πu γ αdu]
γt ( 1 − α ) + π t γ ∗ α δ t
exp(δt) p∗t y∗t γ∗ α
Z T
1 − exp(−δ( T − t)) ∗ ∗
 
= exp(−et ) pt yt + Et [ exp(−δu)(πu − πt )du] .
δ γt ( 1 − α ) + π t γ ∗ α t

Therefore, the foreign stock price in the foreign numéraire is:


exp(δt) p∗t y∗t γ∗ α
Z T
1 − exp(−δ( T − t)) ∗ ∗


st = pt yt + Et exp(−δu)(πu − πt )du .
δ γt ( 1 − α ) + π t γ ∗ α t
(A.20)

Electronic copy available at: https://ssrn.com/abstract=4668578


288 jiang – lecture notes on international finance

There are two ways to proceed in evaluating the above conditional


expectations. The first is to assume that π is a martingale and then
verify that it is indeed the case in equilibrium. However, lowe can
only prove that it is a local martingale. In special cases, we can im-
pose some other regularity conditions on the preference shifts to
guarantee that πt is a martingale but, in general, it is not an immedi-
ate result. An alternative approach is to use the following, less direct,
argument based on market clearing. Plug the first-order conditions
γt
exp(−δt) = ζ H exp(mνH ,t ),
ct
γ∗
exp(−δt) ∗ = ζ̃ F exp(mν̃F ,t ),
c̃t

into (A.9) and (A.13), we have

γt exp(−δt) − exp(−δT ) exp(−δt) − exp(−δT ) 1 − exp(−δ( T − t))


wt + st = = exp(δt)ct = ct ,
ζ H exp(mνH ,t ) δ δ δ
γ∗ exp(−δt) − exp(−δT ) exp(−δt) − exp(−δT ) 1 − exp(−δ( T − t)) ∗
w̃t∗ + s̃∗t = = exp(δt)c̃∗t = c̃t .
ζ̃ F exp(mν̃F ,t ) δ δ δ

Therefore, we have
1 − exp(−δ( T − t)) 1 − exp(−δ( T − t))
wt + st + w̃t∗ + s̃∗t = (ct + c̃∗t ) = ( pt yt + p∗t y∗t exp(−et )) ,
δ δ
where we use the fact that the home and foreign households’ total
consumption expenditure at time t equals the value of their endow-
ments. On the other hand, from the resource constraint,

wt + st + w̃t∗ + s̃∗t = st + s̃∗t .

Combining the resulting restriction,

1 − exp(−δ( T − t))
st + s̃∗t = ( pt yt + p∗t y∗t exp(−et )),
δ
with (A.17) and (A.19), we have

exp(δt) p∗t y∗t γ∗ α exp(δt) pt yt γ∗ (1 − α)


  Z T 
+ E t exp (− δu )( π u − π t ) du = 0,
γt ( 1 − α ) + π t γ ∗ α γt α + π t γ ∗ ( 1 − α ) t

exp(δt) p∗ y∗ γ∗ α exp(δt) p y γ∗ (1−α)


h i
Since the coefficient term γ (1−α)+t πt γ∗ α + γ α+πt γt∗ (1−α) is non-
t t t t
zero for each t, it have to be
Z T 
Et exp(−δu)(πu − πt )du = 0 (A.21)
t

Then, we can conclude that the home stock price in the home numéraire
is:
1 − exp(−δ( T − t))
st = pt yt ,
δ

Electronic copy available at: https://ssrn.com/abstract=4668578


proof of selected results 289

and the price of the foreign stock in the foreign numéraire is:

1 − exp(−δ( T − t)) ∗ ∗
s∗t = pt yt .
δ
Therefore, the price of the foreign stock prices in the home numéraire
is:
1 − exp(−δ( T − t)) ∗ ∗
 
s̃∗t = exp(−et ) pt yt .
δ

By the previous results for good prices and the exchange rate,
 1− α  α −1
1−α
 
α
pt = α exp(tott ) , p∗t =α exp(tott ) , exp(et ) = exp(tott )2α−1 ,
α 1−α

we have
 1− α
1 − exp(−δ( T − t)) 1−α

st = α exp(tott ) yt , (A.22)
δ α
 α −1
1 − exp(−δ( T − t))

α
s∗t = α exp(tott ) y∗t ,
δ 1−α
 α −1
1 − exp(δ( T − t))

∗ α
s̃t = α exp(tott )−α y∗t , (A.23)
δ 1−α

and hence
y∗t
s∗t = st exp(−tott ) .
yt

We then calculate the wealth processes.

1 − exp(−δ( T − t)) 1 − exp(−δ( T − t))


wt + st = ct = ( pt c H,t + p∗t exp(−et )c F,t )
δ δ
1 − exp(−δ( T − t)) γt ( 1 − α )
 
γt α ∗ ∗
= pt yt + exp(−et ) p y
δ γt α + π t γ ∗ ( 1 − α ) γt ( 1 − α ) + π t γ ∗ α t t
γt α γt ( 1 − α )
= st + s̃∗ ,
γt α + π t γ ∗ ( 1 − α ) γt ( 1 − α ) + π t γ ∗ α t
1 − exp(−δ( T − t)) ∗ 1 − exp(−δ( T − t))
w̃t∗ + s̃∗t = exp(−et ) p∗t c∗F,t + pt c∗H,t

c̃t =
δ δ
1 − exp(−δ( T − t)) πt γ∗ α π t γ ∗ (1 − α )
 
∗ ∗
= p y exp (− e t ) + p t y t
δ γt ( 1 − α ) + π t γ ∗ α t t γt α + π t γ ∗ ( 1 − α )

π t γ (1 − α ) ∗
πt γ α
= st + s̃∗ ,

γt α + π t γ ( 1 − α ) γt ( 1 − α ) + π t γ ∗ α t

which lead to
γt α γt ( 1 − α ) y∗
wt + st = ∗
st + ∗
exp(tott )−1 t st ,
γt α + π t γ ( 1 − α ) γt ( 1 − α ) + π t γ α yt

π t γ (1 − α ) yt ∗
πt γ α
wt∗ + s̃∗t = exp(tott ) ∗ s∗t + s∗ .
γt α + π t γ ∗ ( 1 − α ) yt γt ( 1 − α ) + π t γ ∗ α t

Electronic copy available at: https://ssrn.com/abstract=4668578


290 jiang – lecture notes on international finance

Plugging the expression of terms of trade exp(tott ),

γt α + πt γ∗ (1 − α) y∗t
exp(tott ) = ,
γt ( 1 − α ) + π t γ ∗ α y t

into the expression of home and foreign wealth in local numéraires,


we have
γt
wt + st = st ,
γt α + π t γ ∗ ( 1 − α )
πt γ∗
wt∗ + s∗t = s∗ .
γt ( 1 − α ) + π t γ ∗ α t

A.17 Proposition 5.2 in Section 5.A

Proof. (1) By substituting (A.9) and (A.13) into (5.7), we have

w̃t∗ + s̃∗t γt exp(mνH ,t ) ζ H


πt = ∗
=
wt + st γ exp(mν̃F ,t ) ζ̃ F

Recall the representation of both countries’ home state price densi-


ties,

d exp(mνH ,t ) = −rt exp(mνH ,t )dt − (λt + νH,t )′ exp(mνH ,t )dZt (A.24)



d exp(mν̃F ,t ) = −rt exp(mν̃F ,t )dt − (λt + ν̃F,t ) exp(mν̃F ,t )dZt . (A.25)

Then, we apply Ito’s Lemma to πt and obtain:


   
d exp(mνH ,t ) d exp(mν̃F ,t ) d exp(mν̃F ,t ) d exp(mνH ,t ) d exp(mν̃F ,t ) d exp(mν̃F ,t )
dπt = πt − πt − πt , + πt ,
exp(mνH ,t ) exp(mν̃F ,t ) exp(mν̃F ,t ) exp(mνH ,t ) exp(mν̃F ,t ) exp(mν̃F ,t )
= πt (ν̃F,t − νH,t )′ ν̃F,t dt + πt (ν̃F,t − νH,t )′ dZt ,

where we use the definition of λt and the restriction that σt νH,t =


0, σt νH,t = 0. Eq. (A.21) holds for every t, which implies (ν̃F,t −
νH,t )′ ν̃F,t = 0.
If we further suppose that at both countries’ households do not
hold any bonds at the beginning, which means w0 = s0 and w0∗ = s0∗ ,
we can derive the following results:
γ0
= 1 ⇔ γ0 (1 − α) = π0 γ∗ (1 − α)
γ0 α + π0 γ∗ (1 − α)
π0 γ ∗
= 1 ⇔ π0 γ∗ (1 − α) = γ0 (1 − α).
γ0 (1 − α) + π0 γ∗ α

which both lead to


γ0
π0 = .
γ∗

Electronic copy available at: https://ssrn.com/abstract=4668578


proof of selected results 291

(2) Define
def def def
i1 = (1, 0, 0), i2 = (0, 1, 0), i3 = (0, 0, 1).

Apply Ito’s Lemma to exp(tott ), we obtain:

γt α + πt γ∗ (1 − α) y∗t
 
d exp(tott ) = d
γt ( 1 − α ) + π t γ ∗ α y t

1 1 1 1
= exp(tott ) ∗ dy∗t − dyt − [dyt , dy∗t ] + 2 [dyt , dyt ]
yt yt yt y∗t yt
1 1
+ (αdγt + γ∗ (1 − α)dπt ) − ((1 − α)dγt + γ∗ αdπt )
γt α + π t γ ∗ ( 1 − α ) γt ( 1 − α ) + π t γ ∗ α
1 [dy∗t , αdγt + γ∗ (1 − α)dπt ] 1 [dyt , αdγt + γ∗ (1 − α)dπt ]
+ ∗ −
yt γt α + π t γ ∗ ( 1 − α ) yt γt α + π t γ ∗ ( 1 − α )

1 [dyt , (1 − α)dγt + γ αdπt ]∗ 1 [dyt , (1 − α)dγt + γ∗ αdπt ]
− ∗ +
yt γt ( 1 − α ) + π t γ ∗ α yt γt ( 1 − α ) + π t γ ∗ α
∗ ∗
[(1 − α)dγt + γ αdπt , (1 − α)dγt + γ αdπt ] [αdγt + γ∗ (1 − α)dπt , (1 − α)dγt + γ∗ αdπt ]
+ −
( γt ( 1 − α ) + π t γ ∗ α ) 2 [γt α + πt γ∗ (1 − α)][γt (1 − α) + πt γ∗ α]

Recall that et = (2α − 1)tott , which implies


1 1
d(tott ) = κt dt + σt dZt .
2α − 1 2α − 1
If an equilibrium as in (1) exists, by matching the drift and diffusion
terms for exp(tott ),
κt 1 σt σt
d exp(tott ) = exp(tott )( + || ||2 )dt + exp(tott ) dZt ,
2α − 1 2 2α − 1 2α − 1
we can compute κt and σt as follows:

σt αγt ωi3 + γ∗ (1 − α)πt (ν̃F,t − νH,t )′ (1 − α)γt ωi3 + γ∗ απt (ν̃F,t − νH,t )′
= −σy i1 + σy∗ i2 + ∗

2α − 1 γt α + π t γ ( 1 − α ) γt ( 1 − α ) + π t γ ∗ α
2
κt 1 1 1 ∥αγt ωi3 + γ∗ (1 − α)πt (ν̃F,t − νH,t )′ ∥2 1 ∥(1 − α)γt ωi3 + γ∗ απt (ν̃F,t − νH,t )′ ∥
= µy∗ − µy + σy2 − σy2∗ − + .
2α − 1 2 2 2 (γt α + πt γ∗ (1 − α))2 2 ( γt ( 1 − α ) + π t γ ∗ α ) 2

Plug in Eq. (A.22) and (A.23) to obtain the drift and volatility of stock
price dynamics

κt ( α − 1)2 σt σt ′
µst = (1 − α) + ∥ ∥2 + µy + (1 − α)σy i ,
2α − 1 2 2α − 1 2α − 1 1
σt
σts = (1 − α) + σy i1 ,
2α − 1
∗ κt α2 σt σt ′
µs̃t = −α + ∥ ∥2 + µy∗ − ασy∗ i ,
2α − 1 2 2α − 1 2α − 1 2
∗ σt
σts̃ = −α + σy∗ i2 ,
2α − 1
˜
Combined with Eq. (A.11) and (A.15), we can solve κt , σt , νH,t , νF,t
simultaneously as function of exogenous state variables.

Electronic copy available at: https://ssrn.com/abstract=4668578


292 jiang – lecture notes on international finance

(3) Given the formula of the home numéraire, we have

αα (1 − α)1−α = pαt [ p∗t exp(−et )]1−α .

Therefore, the home household’s consumption-based home country


stochastic price density is:

1
exp(mνH ,t ) = pαt [ p∗t exp(−et )]1−α exp(mνH ,t )
α α (1 − α )1− α
1 α ∗ 1− α p 0 γt α + π t γ ∗ ( 1 − α ) y 0
= p [ p exp (− e t )] exp (− δt )
α α (1 − α )1− α t t pt γ0 α + π0 γ∗ (1 − α) yt
 ∗ 1− α
γt α + π t γ ∗ ( 1 − α )
  
1 pt exp(−et ) y0
= α exp(−δt) p0
α (1 − α )1− α pt γ0 α + π0 γ∗ (1 − α) yt
γt α + π t γ ∗ ( 1 − α )
   
1 y0
= α 1 −
exp (( α − 1 ) tot t − δt ) ∗
p0 .
α (1 − α ) α γ0 α + π 0 γ ( 1 − α ) yt

It follows that
 
1 2
d exp(mνH ,t ) = exp(mνH ,t ) µm,t + ∥σm,t ∥ dt + exp(mνH ,t )σm,t dZt ,
2

where
κt 1 1 ||αγt ωi3 + (1 − α)γ∗ πt (ν̃F,t − νH,t )′ ||2
µm,t = −δ + (α − 1) − µy + σy2 − ,
2α − 1 2 2 (γt α + πt γ∗ (1 − α))2
σt αγt ωi3 + (1 − α)γ∗ πt (ν̃F,t − νH,t )′
σm,t = ( α − 1) − σy i1 + .
2α − 1 γt α + π t γ ∗ ( 1 − α )

Similarly, for the foreign numéraire, we have

αα (1 − α)1−α = p∗t α [ pt exp(et )]1−α .

Then, the foreign household’s consumption-based foreign country


stochastic price density is:

1
exp(m∗ν∗ ,t ) = p∗α [ pt exp(et )]1−α exp(m∗ν∗ ,t )
F α α (1 − α )1− α t F

1 exp(e0 )
= p∗α [ pt exp(et )]1−α exp(mν̃F ,t )
α α (1 − α )1− α t exp(et )
1 ∗α 1−α exp( e0 ) p0∗ exp(−e0 )y0∗ π0 γt (1 − α) + πt γ∗ α
= p t [ p t exp ( e t )] exp (− δt )
α α (1 − α )1− α exp(et ) p∗t exp(−et )y∗t πt γ0 (1 − α) + π0 γ∗ α
pt exp(et ) 1−α p ∗ y ∗ π 0 γt ( 1 − α ) + π t γ ∗ α
 
1
= ∗ exp(−δt) 0 ∗ 0
α (1 − α )
α 1 − α pt yt πt γ0 (1 − α) + π0 γ∗ α
 ∗ 
γt ( 1 − α ) + π t γ ∗ α

1 y0 π0 ∗
= exp (( 1 − α ) tot t − δt ) p0 .
α α (1 − α )1− α γ0 (1 − α) + π0 γ∗ α y∗t πt

It follows that
 
1
d exp(m∗ν∗ ,t ) = exp(m∗ν∗ ,t ) µm∗ ,t + ∥σm ,t ∥ dt + exp(m∗ν∗ ,t )σm∗ ,t dZt

2
F F 2 F

Electronic copy available at: https://ssrn.com/abstract=4668578


proof of selected results 293

where
κt 1 1 1 ∥(1 − α)γt ωi3 + αγ∗ πt (ν̃F,t − νH,t )′ ∥2
µm∗ ,t = −δ + (1 − α) − µy∗ + σy2∗ + (∥νH,t ∥2 − ∥ν̃F,t ∥2 ) − ,
2α − 1 2 2 2 ( γt ( 1 − α ) + π t γ ∗ α ) 2
σt (1 − α)γt ωi3 + αγ∗ πt (ν̃F,t − νH,t )′
σm∗ ,t = (1 − α ) − σy∗ i2 − (ν̃F,t − νH,t )′ + .
2α − 1 γt ( 1 − α ) + π t γ ∗ α
We introduce a stochastic wedge ηt that reconciles the log change in
exchange rates with the domestic and foreign consumption-based
SDFs:

det = dmνH ,t − dm∗ν∗ ,t + dηt .


F

Then, we can back up the stochastic wedge ηt process as

dηt = (κt − µm,t + µm∗ ,t ) dt + (σt − σm,t + σm∗ ,t )dZt


1
= (∥νH,t ∥2 − ∥ν̃F,t ∥2 )dt + (νH,t − ν̃F,t )′ dZt .
2
(4) The interest rates and the home country market price of risk
processes can be derived from the dynamics of exp(mνH ,t ). We apply
Ito’s Lemma to exp(mνH ,t ) and match its drift and diffusion terms to
obtain rt and λt .
1
rt = −µm,t − ∥σm,t ∥2
2

λt = −σm,t − νH,t

Given λt and rt , we can calculate rt∗ using the definition of λt ,


1
rt∗ = −σt λt + κt − σt σt′ + rt .
2

A.18 Proposition 5.3 in Section 5.A

Proof. Apply Ito’s Lemma to exp(tott ), we obtain:


γt α + πγ∗ (1 − α) y∗t
 
d exp(totcm
t ) = d
γt (1 − α) + πγ∗ α yt

1 ∗ 1 1 1
= exp(totcm t ) dy − dyt − [dyt , dy∗t ] + 2 [dyt , dyt ]
y∗t t yt yt y∗t yt
αdγt (1 − α)dγt
+ −
γt α + πγ∗ (1 − α) γt (1 − α) + πγ∗ α
1 [dy∗t , αdγt ] 1 [dyt , αdγt ]
+ ∗ −
yt γt α + πγ∗ (1 − α) yt γt α + πγ∗ (1 − α)
1 [dy∗t , (1 − α)dγt ] 1 [dyt , (1 − α)dγt ]
− ∗ +
yt γt (1 − α) + πγ α yt γt (1 − α) + πγ∗ α

[(1 − α)dγt , (1 − α)dγt ] [αdγt , (1 − α)dγt ]



+ −
(γt (1 − α) + πγ∗ α)2 [γt α + πγ (1 − α)][γt (1 − α) + πγ∗ α]

Electronic copy available at: https://ssrn.com/abstract=4668578


294 jiang – lecture notes on international finance

Again, conjecture that the logarithm of terms of trade follows

d(totcm cm cm
t ) = At dt + Bt dZt .

If an equilibrium exists, by matching the drift and diffusion terms for


exp(tott ),
1 cm 2
d exp(totcm cm cm cm cm
t ) = exp( tott )( At + || Bt || ) dt + exp( tott ) Bt dZt ,
2
we can compute Acm cm
t and Bt as follows:

αγt ωi3 (1 − α)γt ωi3


Btcm = −σy i1 + σy∗ i2 + −
γt α + πγ∗ (1 − α) γt (1 − α) + πγ∗ α
1 2 1 2 1 α2 γt2 ω 2 1 (1 − α)2 γt2 ω 2
Acm
t = µy∗ − µy + σy − σy∗ − +
2 2 2 (γt α + πγ∗ (1 − α))2 2 (γt (1 − α) + πγ∗ α)2

Matching the drift and volatility terms of the log exchange rate, we
have

d(etcm ) = d ((2α − 1)totcm cm cm


t ) = (2α − 1) At dt + (2α − 1) Bt dZt .

Therefore,

κtcm = (2α − 1) Acm


t ,
σtcm = (2α − 1) Btcm .

Under complete markets, we have a unique home country state price


density:

d exp(mcm cm cm
t ) = −rt exp( mt ) dt − λt exp( mt ) dZt , (A.26)

and a unique home country state price density:

d exp(m∗t ,cm ) = −rt∗ exp(m∗t ,cm )dt − λ∗′ ∗,cm


t exp( mt )dZt , (A.27)

Given the formula of the home numéraire, we have

αα (1 − α)1−α = pαt [ p∗t exp(−etcm )]1−α .

Therefore, the home household’s consumption-based home country


stochastic price density is:
1
exp(mcm
t )= pαt [ p∗t exp(−etcm )]1−α exp(mcm
t )
α α (1 − α )1− α
1 α ∗ cm 1−α p0 γt α + πγ∗ (1 − α) y0
= p t [ p t exp (− e t )] exp (− δt )
α α (1 − α )1− α pt γ0 α + πγ∗ (1 − α) yt
pt exp(−etcm ) 1−α
 ∗
γt α + πγ∗ (1 − α)
   
1 y0
= α exp(−δt) p0
α (1 − α )1− α pt γ0 α + πγ∗ (1 − α) yt
γt α + πγ∗ (1 − α)
   
1 cm y0
= α exp (( α − 1 ) tot t − δt ) p0
α (1 − α )1− α γ0 α + πγ∗ (1 − α) yt

Electronic copy available at: https://ssrn.com/abstract=4668578


proof of selected results 295

It follows that
 
1 cm 2
d exp(mcm
t ) = exp(mcm
t ) µcm
m,t + ∥σm,t ∥ dt + exp(mcm cm
t ) σm,t dZt
2

where

1 2 1 α2 γt2 ω 2
µcm cm
m,t = − δ + ( α − 1) At − µy + σy −
2 2 (γt α + πt γ∗ (1 − α))2
cm αγt ωi3
σm,t = (α − 1) Btcm − σy i1 +
γt α + π t γ ∗ ( 1 − α )

Similarly, for the foreign numéraire, we have

αα (1 − α)1−α = p∗t α [ pt exp(etcm )]1−α .

Then, by the formula of the home numéraire, the foreign household’s


consumption-based foreign country stochastic price density is:

1
exp(m∗t ,cm ) = p∗α [ pt exp(etcm )]1−α exp(m∗t ,cm )
α α (1 − α )1− α t
1 exp(e0cm ) p∗ exp(−e0cm )y0∗ γt (1 − α) + πγ∗ α
= α p∗t α [ pt exp(etcm )]1−α exp(−δt) 0∗
α (1 − α ) 1 − α exp(et )cm pt exp(−etcm )y∗t γ0 (1 − α) + πγ∗ α
pt exp(etcm ) 1−α p0∗ y0∗ γt (1 − α) + πγ∗ α
 
1
= α exp (− δt )
α (1 − α )1− α p∗t y∗t γ0 (1 − α) + πγ∗ α
 ∗
γt (1 − α) + πγ∗ α

1 y0
= α exp (( 1 − α ) tot cm
t − δt ) p0∗
α (1 − α )1− α γ0 (1 − α) + πγ∗ α y∗t

It follows that
 
1 cm 2
d exp(m∗ν∗,cm ) = exp ( m ∗,cm
∗ ) µ cm

m ,t + ∥ σ ∗ ∥ dt + exp(m∗ν∗,cm )σmcm∗ ,t dZt
F ,t νF ,t 2 m ,t F ,t

where

1 (1 − α)2 γt2 ω 2
µcm cm
m∗ ,t = − δ + (1 − α ) At − µy∗ −
2 (γt (1 − α) + πγ∗ α)2
cm (1 − α)γt ωi3
σm ∗ ,t = (1 − α) Btcm − σy∗ i2 +
γt ( 1 − α ) + π t γ ∗ α

Then, the exchange rate dynamics is back in the benchmark complete


markets case:

∗,cm
detcm = dmcm
t − dmt

Electronic copy available at: https://ssrn.com/abstract=4668578


296 jiang – lecture notes on international finance

A.19 Proposition 5.4 in Section 5.B

Proof. First, we plug in ∆et+1 = mt+1 − m∗t+1 + ηt+1 to reproduce the


Euler equations (1.6)-(1.9) as follows

1 = Et [exp(mt+1 + rt )] ,
1 = Et exp(m∗t+1 − ηt+1 + rt∗ ) ,
 

1 = Et exp(m∗t+1 + rt∗ ) ,
 

1 = Et [exp(mt+1 + ηt+1 + rt )] .

By joint normality, we obtain


1
0 = Et [mt+1 ] + vart (mt+1 ) + rt ,
2
1 1
0 = Et [mt+1 ] + vart (mt+1 ) + rt + Et [ηt+1 ] + vart (ηt+1 ) + covt (mt+1 , ηt+1 ),
2 2
1
0 = Et [m∗t+1 ] + vart (m∗t+1 ) + rt∗ ,
2
1 1
0 = Et [m∗t+1 ] + vart (m∗t+1 ) + rt∗ − Et [ηt+1 ] + vart (ηt+1 ) − covt (m∗t+1 , ηt+1 ),
2 2
which implies
1
covt (mt+1 , ηt+1 ) = −Et [ηt+1 ] − vart (ηt+1 ),
2
∗ 1
covt (mt+1 , ηt+1 ) = −Et [ηt+1 ] + vart (ηt+1 ),
2
and hence

covt (mt+1 − m∗t+1 , ηt+1 ) = −vart (ηt+1 ).

Recall

∆et+1 = mt+1 − m∗t+1 + ηt+1 ,

which gives the following moments

vart (∆et+1 ) = vart (mt+1 − m∗t+1 ) + vart (ηt+1 ) + 2covt (mt+1 − m∗t+1 , ηt+1 )
= vart (mt+1 − m∗t+1 ) − vart (ηt+1 ),
covt (mt+1 − m∗t+1 , ∆et+1 ) = vart (mt+1 − m∗t+1 ) + covt (mt+1 − m∗t+1 , ηt+1 )
= vart (mt+1 − m∗t+1 ) − vart (ηt+1 )
= vart (∆et+1 ).

The excess return is given by

Et [rxt+1 ] = −rt∗ + rt + Et [∆et+1 ]


1 1
= Et [m∗t+1 ] + vart (m∗t+1 ) − Et [mt+1 ] − vart (mt+1 ) + Et [mt+1 − m∗t+1 + ηt+1 ]
2 2
1
= (vart (m∗t+1 ) − vart (mt+1 )) + Et [ηt+1 ].
2

Electronic copy available at: https://ssrn.com/abstract=4668578


proof of selected results 297

A.20 Proposition 5.5 in Section 5.B

Proof. By construction, covt (mt+1 , wt ε t+1 ) = covt (m∗t+1 , wt ε t+1 ) = 0.


Hence,

vart (∆et+1 ) = vart ( xt + yt mt+1 + zt m∗t+1 + wt ε t+1 )


= vart (yt mt+1 + zt m∗t+1 ) + vart (wt ε t+1 ),

which implies (a). Similarly,

covt (∆et+1 , mt+1 − m∗t+1 ) = covt ( xt + yt mt+1 + zt m∗t+1 + wt ε t+1 , mt+1 − m∗t+1 )
= covt (yt mt+1 + zt m∗t+1 , mt+1 − m∗t+1 ),

which implies (b). Define rpt = Et [∆et+1 + rt∗ − rt ]. Euler equations


(1.6) and (1.7) yield

1
0 = Et [mt+1 ] + vart (mt+1 ) + rt ,
2
1 1
0 = Et [mt+1 ] + vart (mt+1 ) − Et [∆et+1 ] + vart (∆et+1 ) − covt (mt+1 , ∆et+1 ) + rt∗ ,
2 2
which implies

1
rpt = −covt (mt+1 , ∆et+1 ) + vart (∆et+1 ).
2
Similarly, Eq. (1.8) and (1.9) imply

1
rpt = −covt (m∗t+1 , ∆et+1 ) − vart (∆et+1 ).
2

Adding together to cancel 12 vart (∆et+1 ) and obtain (c)

1 1
rpt = − covt (mt+1 , ∆et+1 ) − covt (m∗t+1 , ∆et+1 )
2 2
1
= − covt (mt+1 + m∗t+1 , ∆et+1 )
2
1
= − covt (yt mt+1 + zt m∗t+1 , mt+1 + m∗t+1 ).
2

A.21 Proposition 5.6 in Section 5.B

Proof. The Euler equation Eq. (5.18) implies

0 = covt (∆et+1 , ∆et+1 − (mt+1 − m∗t+1 ))


= covt (yt mt+1 + zt m∗t+1 + wt ε t+1 , (yt − 1)mt+1 + (zt + 1)m∗t+1 + wt ε t+1 )
= yt (yt − 1)vart (mt+1 ) + zt (zt + 1)vart (m∗t+1 )
+ [yt (zt + 1) + zt (yt − 1)]covt (mt+1 , m∗t+1 ) + w2t ,

Electronic copy available at: https://ssrn.com/abstract=4668578


298 jiang – lecture notes on international finance

where we use the fact that by construction, covt (wt ε t+1 , mt+1 ) =
covt (wt ε t+1 , m∗t+1 ) = 0. w2t ≥ 0 leads to constraint Eq. (5.15).
The Euler equation Eq. (5.18) also implies

vart (∆et+1 ) = covt (∆et+1 , mt+1 − m∗t+1 )


= covt (yt mt+1 + zt m∗t+1 + wt ε t+1 , mt+1 − m∗t+1 )
= yt vart (mt+1 ) − zt vart (m∗t+1 ) + (zt − yt )covt (mt+1 .m∗t+1 ),

which leads to constraint Eq. (5.16).

A.22 Proposition 5.7 in Section 5.C.

Proof. Recast the Euler equation between country 1 and 2 to obtain


(1) (2)
0 = covt [∆e1/2 1/2
t+1 , ∆et+1 − ( mt+1 − mt+1 )]
(0) (2) (0) (1)
= covt [∆e0/2 0/1 0/2 0/1
t+1 − ∆et+1 , ( ∆et+1 − ( mt+1 − mt+1 )) − ( ∆et+1 − ( mt+1 − mt+1 ))]
(0) (1) (0) (2)
= covt [∆e0/1 0/1 0/2 0/2
t+1 , ∆et+1 − ( mt+1 − mt+1 )] + covt [ ∆et+1 , ∆et+1 − ( mt+1 − mt+1 )]
(0) (1) (0) (2)
− covt [∆e0/2 0/1 0/1 0/2
t+1 , ∆et+1 − ( mt+1 − mt+1 )] − covt [ ∆et+1 , ∆et+1 − ( mt+1 − mt+1 )],

where the first two terms are zero from the Euler equations (5.18) and
(5.19). Hence,
(0) (1) (0) (2)
0 = covt [∆e0/2 0/1 0/1 0/2
t+1 , ∆et+1 − ( mt+1 − mt+1 )] + covt [ ∆et+1 , ∆et+1 − ( mt+1 − mt+1 )].

A.23 Proposition 5.8 in Section 5.C.

Proof. Assume the equity claims (i.e., endowment claims) and risk-
free bonds are tradable. Assume the endowment and demand shock
processes are exogenous and adapted to the filtration that is gener-
ated by some d-dimensional Brownian motion Zt . The set of tradable
assets does not span all the Brownian shocks. In each country i, the
households’ problem is given by
"Z ! #
T 1−α
α log ci,t + ∑
(i ) (i ) (i )
max E exp(−ρt)γt log c j,t dt
(i ) (i ) (i )
c ,c ,c 0 j ̸ =i
2
0,t 1,t 2,t

subject to the following dynamic budget constraint


(i ) (i ) (i ) (i ) (i ) (i ) (i ) (i ) (i ) ( j) i/j (i )
dwt = wt (rt + (xt )′ (µt − rt 1))dt + wt ((xt )′ σt )dZt − pt exp(−et )c j,t dt,
j∈{0,1,2}

( j)
where pt is the price of good j in the numéraire of country j’s con-
( j) (i ) 1− α (i ) i/j
sumption bundle ct = exp(α log ci,t + ∑ j̸=i 2 log c j,t ), and et

Electronic copy available at: https://ssrn.com/abstract=4668578


proof of selected results 299

denotes the bilateral exchange rate, which takes a higher value if cur-
(i )
rency in country i is stronger. xt is the column vector of portfolio
(i )
weights for investors in country i, µt is the column vector of ex-
(i )
pected return of risky assets in country i’s numéraire, and σt is the
corresponding volatility matrix.
Following He and Pearson [1991], Cuoco and He [1994] and
Pavlova and Rigobon [2012], we can solve the system by transforming
the dynamic budget constraints to a static one, which is given by
Z T 
(i ) (i ) (i )
E exp(mt )ct ≤ w0 , (A.28)
0
(i )
where exp(mt ) is the state price density in country i. The house-
holds’ Lagrangian is given by
Z T  Z T 
(i ) (i ) (i ) (i ) (i ) (i )
L=E exp(−ρt)γt log ct dt + ζ w0 − exp(mt )ct dt ,
0 0

which yields the following first-order condition


(i )
exp(−ρt)γt (i )
(i )
= ζ (i) exp(mt ), (A.29)
ct
(i )
with the normalization exp(m0 ) = 1. When markets are incomplete,
there are an infinite number of SDFs that correctly price the assets.
However, the SDF that coincides with the marginal utility is unique
and is given by (A.29).
The households’ wealth in period t is given by
T exp( m(i ) ) T exp(− ρu ) γ(i )
"Z # "Z #
(i ) u (i ) ζ (i ) (i ) u (i )
w t = Et cu du = c Et
(i ) t
cu du
t exp( m(i ) ) exp (− ρ ) γ t ζ (i ) c (i )
t t t u
1 − exp(−ρ( T − t)) (i)
= ct , (A.30)
ρ
(i )
where the demand shocks γt are assumed to be martingales. Going
back to households’ problem, the within-period first-order conditions
yields
( j) i/j 1 − α (i )
(i )
pt exp(−et )c j,t = (1{i= j} α + 1{i̸= j}
)ct .
2
Next, the equilibrium allocation is not necessarily Pareto opti-
mal with incomplete markets. However, as shown in Cuoco and He
[1994] and Pavlova and Rigobon [2012], we can still solve the social
(i )
planner’s problem with stochastic weights πt ,
 
Z T
1−α
∑ ∑ (1{i= j} α + 1{i̸= j} 2 ) log c j,t dt ,
(i ) (i ) (i ) 
max E exp(−ρt) π t γt
(i ) j,t 0
c i ∈{0,1,2} j∈{0,1,2}


(i ) ( j)
s.t. c j,t = yt , j ∈ {0, 1, 2}.
i ∈{0,1,2}

Electronic copy available at: https://ssrn.com/abstract=4668578


300 jiang – lecture notes on international finance

The first-order conditions imply


( i ) ( i ) 1− α
(i ) π t γt 2 ( j)
c j,t = ( i ) ( i ) 1− α ( j) ( j)
yt , i ̸= j
∑ i ̸ = j π t γt 2 + π t γt α
( j) ( j)
( j) π t γt α ( j)
c j,t = ( i ) ( i ) 1− α ( j) ( j)
yt .
∑ i ̸ = j π t γt 2 + π t γt α
In country 0, the within-period solution implies
(1) (0) 1 − α (0)
pt exp(−e0/1
t ) c1,t = c
2 t
(2) (0) 1 − α (0)
pt exp(−e0/2
t ) c2,t = c
2 t
Let q1/2
t denotes the terms of trade in log between countries 1
and 2, which is the ratio of goods prices in a common numéraire.
Dividing the first equation above by the second yields
(1)
pt exp(−e0/1
t )
exp(q1/2
t )= (2)
pt exp(−e0/2
t )
(0) (0) (0) 1− α (1) (1) (2) (2) 1− α (2)
c2,t π t γt 2 + π t γt α + π t γt 2 yt
= (0)
= (0) (0) 1− α (1) (1) 1− α (2) (2) (1)
. (A.31)
c1,t π t γt 2 + π t γt 2 + π t γt α yt
(i ) (i )
To make comparison across countries, define w̃t = wt exp(−e0/i t )
as country i’s wealth in country 0’s numéraire. Below, we show that
(i ) (i ) ( j) ( j) (i ) ( j)
πt γt /πt γt = w̃t /w̃t . To see this, note that
(1) (0) (1) (1) (1) (2) (1)
ct exp(−e0/1 1/0 0/1 0/1 1/2 0/1
t ) = pt exp(− et ) exp(− et ) c0,t + pt exp(− et ) c1,t + pt exp(− et ) exp(− et ) c2,t
(0) (1) (1) (1) (2) (1)
= pt c0,t + pt exp(−e0/1 0/2
t ) c1,t + pt exp(− et ) c2,t
(1) (1) (1)
c0,t (0) 1 − α c1,t (0) 1 − α c2,t (0)
=α c
(0) t
+ c + c
c0,t 2 c (0) t 2 c (0) t
1,t 2,t
(1) (1) (1) (1) (1) (1)
!
π γ 1− α 1−α π t γt α 1 − α πt γt 1−2 α (0)
= α t (0) t (0) 2 + (0) (0)
+ ct
π t γt α 2 πt γt 1−2 α 2 π (0) γ (0) 1− α
t t 2
(1) (1)
π t γt (0 )
= c .
(0) (0) t
π t γt
Hence, by Eq. (A.30),
(1) (1) (1) (1)
w̃t ct exp(−e0/1
t ) π t γt
(0)
= (0)
= (0) (0)
.
w̃t ct π t γt
Similarly,
(2) (2) (2) (2)
w̃t ct exp(−e0/2
t ) π t γt
(0)
= (0)
= (0) (0)
.
w̃t ct π t γt

Electronic copy available at: https://ssrn.com/abstract=4668578


proof of selected results 301

Plug in Eq. (A.31) to obtain

(0) (0) 1− α (1) (1) (2) (2) 1− α (2)


π t γt + π t γt α + π t γt yt
q1/2
t = log 2
(0) (0) 1− α (1) (1) 1− α
2
(2) (2)
+ log (1)
π t γt 2 + π t γt 2 + π t γt α yt
(1) (1) (2) (2)
1− α π t γt π t γt 1 − α
2 + (0) (0) α + (0) (0) 2 (2)
π t γt π t γt yt
= log (1) (1) (2) (2)
+ log (1)
(A.32)
1− α
+
π t γt 1 − α π t γt
+ yt
2 (0) (0) 2 (0) (0) α
π t γt π t γt
(1) (2)
1− α w̃t w̃t 1−α
2 + (0) α +(0) 2 (2)
w̃t w̃t yt
= log (1) (2)
log + (1)
1− α w̃t 1−α w̃t yt
2 + (0) 2 +(0) α
w̃t w̃t

When markets are complete, the Pareto weights are constant, i.e.
(i )
πt = π (i) . Plugging in Eq. (A.32) yields

(1) (2)
1− α π (1 ) γt π (2 ) γt 1− α
2 + (0) α+ (0) 2 (2)
π (0 ) γt π (0 ) γt yt
q1/2,cm
t = log ( 1) (2)
+ log (1)
.
1− α π (1 ) γt 1− α π (2 ) γt yt
2 + (0) 2 + (0) α
π (0 ) γt π (0 ) γt

Finally, we derive the exchange rate as a function of the terms of


(1) (1) (1)
trade. pt c1,t = αct implies

(1) (1) (1) (1) 1− α (1) 1− α


pt c1,t = α(c1,t )α (c2,t ) 2 (c3,t ) 2
  1− α
(1) (1) 2  1− α
c c 1−α

2
(1) 2,t 3,t
p t = α  (1) (1)  =α exp(q1/2
t ) exp ( q 1/3
t ) .
c1,t c1,t 2α

  1− α
(2) 1− α 2
Similarly, pt =α 2α exp(q2/1 2/3
t ) exp( qt ) . Recall that exp(q1/2
t )=
(1) (2)
pt exp(e1/2
t ) /pt , which implies

(1)
! 1− α
exp(q1/2 1/3 2
3(1 − α) 1/2
 
pt t ) exp( qt )
exp(q1/2 1/2
t ) exp(− et ) = = = exp qt .
pt
(2) exp(q2/1 2/3
t ) exp( qt )
2

Hence, the log bilateral exchange rate is linear in the log terms of
trade:

3α − 1 1/2
e1/2
t = qt ,
2

which concludes the proof.

Electronic copy available at: https://ssrn.com/abstract=4668578


302 jiang – lecture notes on international finance

A.24 Proposition 6.2 and 6.3 in Section 6.C

Proof. Consider the general case in Proposition 6.3. By market clear-


ing and price setting,

gt∗
 
Pt ct gt
zt ℓt = 1 + α + (1 − α ) ∗
PH,t ct ct

Pt ct∗ 
gt∗ gt

∗ ∗
zt ℓt = ∗ 1 + α ∗ + (1 − α ) .
PF,t ct ct

By Wt = κPt ct ℓνt ,
Pt ct ρ − 1 −ν
= zt ℓt
PH,t κρ
ρ −1
Let ℓ̄ = ρκ . The market clearing equations become

g∗
 
gt
ℓ1t +ν = ℓ̄ 1 + α + (1 − α) ∗t
ct ct
By the market clearing equations again,
Pt ct
zt ℓ̄ℓ− ν
t =
PH,t
PH,t PH,t
ct = zt ℓ̄ℓ−
t
ν
= αα (1 − α)1−α ℓ̄ℓ− ν
t zt α 1− α
Pt PH,t PF,t

By price setting,
ρ Wt Pt ct ℓν
PH,t ρ −1 z t zt
t
z∗t (ℓ∗t )−ν
= ∗ = ∗ ∗ ∗ =
PF,t ρ W
exp(−Et ) ρ−1 z∗t
P c (ℓ )
exp(−Et ) t tz∗ t
ν
zt ℓ−t
ν
t t

so

log ct = log αα (1 − α)1−α ℓ̄ + α(log zt − ν log ℓt ) + (1 − α)(log z∗t − ν log ℓ∗t )


1
= log αα (1 − α)1−α + log ℓ̄ + α log zt + (1 − α) log z∗t
1+ν
gt∗ gt∗
    
ν gt gt
− α log 1 + α + (1 − α) ∗ + (1 − α) log 1 + α ∗ + (1 − α)
1+ν ct ct ct ct
To solve this equation system, we linearize spending gt /ct and
gt∗ /c∗t around 0:
1
log ct = log αα (1 − α)1−α + log ℓ̄ + α log zt + (1 − α) log z∗t
1+ν
g∗
    ∗ 
ν gt g gt
− α α + (1 − α) ∗t + (1 − α) α ∗t + (1 − α)
1+ν ct ct ct ct
which simplifies to
g∗
 
ν gt
log ct − log c̄ ≈ α log zt + (1 − α) log z∗t − ( α2 + (1 − α )2 ) + 2(1 − α)α ∗t ,
1+ν ct ct

Electronic copy available at: https://ssrn.com/abstract=4668578


proof of selected results 303

Similarly, the foreign counterpart is

gt∗
 
ν gt
log c∗t − log c̄ ≈ (1 − α) log zt + α log z∗t − ( α2 + (1 − α )2 ) ∗ + 2(1 − α ) α ,
1+ν ct ct

Hence, the equilibrium foreign real exchange rate is

def Pt∗
−et = log exp(−Et ) = log ct − log c∗t
Pt
gt∗
 
∗ ν 2 gt
≈ (2α − 1)(log zt − log zt ) − (2α − 1) − ∗ ,
1+ν ct ct

Plug in ν = 0 to get Proposition 6.2.

A.25 Proposition 6.4 in Section 6.D

Proof. When prices are sticky, the model is closed by the following
equations for prices:
   
ρ Wt ∗ ρ Wt
PH,t = Et −1 exp(−Et ) PH,t = Et −1 (A.33)
ρ−1 zt ρ−1 zt
 ∗  ∗
ρ Wt ∗ ρ W
exp(Et ) PF,t = E PF,t = E t
(A.34)
ρ − 1 t−1 z∗t ρ − 1 t−1 z∗t

Just as in the flexible price case,

g∗
 
Pt ct gt
zt ℓt = 1 + α + (1 − α) ∗t (A.35)
PH,t ct ct

Also use Eq. (A.33),

gt∗ gt∗
   
Pt ct gt Pt ct /zt gt
ℓt = 1 + α + (1 − α ) ∗ = ℓ̄ 1 + α + (1 − α) ∗
PH,t zt ct ct Et−1 [ Pt ct ℓνt /zt ] ct ct

Use Eq. (A.35) again,


 −1
g∗

PH,t gt PH,t Pt ct
ct = zt ℓt · 1+α + (1 − α) ∗t = ℓ̄
Pt ct ct 1
P α P 1− α Et−1 [ Pt ct ℓνt /zt ]
αα (1−α)1−α H,t F,t
1− α
Pt ct 1− α
PH,t
= ℓ̄(α (1 − α) ) 1−α
α
Et−1 [ Pt ct ℓνt /zt ] PF,t

Use the price setting Eq. (A.33) and (A.34),

PH,t 1 Et−1 [Wt /zt ] P∗ c∗ Et−1 [ Pt ct ℓνt /zt ]


= = t t
PF,t exp(−Et ) Et−1 [Wt /zt ]
∗ ∗ Pt ct Et−1 [ Pt∗ c∗t (ℓ∗t )ν /z∗t ]

Then

( Pt ct )α ( Pt∗ c∗t )1−α


ct = ℓ̄αα (1 − α)1−α
(Et−1 [ Pt ct ℓt /zt ])α (Et−1 [ Pt∗ c∗t (ℓ∗t )ν /z∗t ])1−α
ν

Electronic copy available at: https://ssrn.com/abstract=4668578


304 jiang – lecture notes on international finance

So the equilibrium price levels and consumption can be solved by


( Pt ct )α ( Pt∗ c∗t )1−α
ct = ℓ̄αα (1 − α)1−α
(Et−1 [ Pt ct ℓt /zt ])α (Et−1 [ Pt∗ c∗t (ℓ∗t )ν /z∗t ])1−α
ν

( Pt ct )1−α ( Pt∗ c∗t )α


c∗t = ℓ̄αα (1 − α)1−α
(Et−1 [ Pt ct ℓνt /zt ])1−α (Et−1 [ Pt∗ c∗t (ℓ∗t )ν /z∗t ])α
Qt
Pt = h i,
s t + c t Et ∑ ∞
k = t +1 β
k −t sk
ck
Q∗t
Pt∗ = h s∗
i.
s∗t + c∗t Et ∑∞
k = t +1 β
k−t k
c∗ k

which implies

log ct = κtc−1 + α log( Pt ct ) + (1 − α) log( Pt∗ c∗t )



log c∗t = κtc−1 + α log( Pt∗ c∗t ) + (1 − α) log( Pt ct )
−et = log ct − log c∗t = κte−1 + (2α − 1)(log( Pt ct ) − log( Pt∗ c∗t ))

where
ℓ̄αα (1 − α)1−α
 
κtc−1 = log
(Et−1 [ Pt ct ℓνt /zt ])α (Et−1 [ Pt∗ c∗t (ℓ∗t )ν /z∗t ])1−α
ℓ̄αα (1 − α)1−α
 
κtc−1 = log
(Et−1 [ Pt ct ℓt /zt ])1−α (Et−1 [ Pt∗ c∗t (ℓ∗t )ν /z∗t ])α
ν

κte−1 = κtc−1 − κtc−1

A.26 Proposition 6.5 in Section 6.D

Proof. Plug in Eq. (6.15) into Proposition 6.4 to get


   " #!
c 1 ∗ 1
log ct = κt−1 − α it + log Et δ − (1 − α) it + log Et δ ∗ ∗
Pt+1 ct+1 Pt+1 ct+1
= κ̃tc−1 − αit − (1 − α)it∗

similarly,
" #!   
∗ 1 1
log c∗t = κtc−1 −α it∗ + log Et δ ∗ ∗ − (1 − α) it + log Et δ
Pt+1 ct+1 Pt+1 ct+1

= κ̃tc−1 − αit∗ − (1 − α)it

where
 " #
1 1
κ̃tc−1 = κtc−1
− α log Et δ − (1 − α) log Et δ ∗ ∗
Pt+1 ct+1 Pt+1 ct+1
" #  
∗ c∗ 1 1
κ̃tc−1 = κt−1 − α log Et δ ∗ ∗ − (1 − α) log Et δ
Pt+1 ct+1 Pt+1 ct+1

Electronic copy available at: https://ssrn.com/abstract=4668578


proof of selected results 305

Recall the real Euler equation:

rt = − log δ − log Et [ct /ct+1 ]


= − log δ − log ct − log Et [1/ct+1 ]

Plugging log consumption into the equation above yields:

rt = r̄ + αit + (1 − α)it∗ .

where

r̄ = − log δ − κ̃tc−1 − log Et [1/ct+1 ]

the equilibrium log real exchange rate is



−et = log ct − log c∗t = κ̃tc−1 − κ̃tc−1 − (2α − 1)(it − it∗ )

and the equilibrium price levels can be pinned down by plugging


consumption into Eq. (6.15)
 
1
log Pt = − log ct − it − log Et δ
Pt+1 ct+1
 
1
= −κ̃tc−1 − log Et δ + (1 − α)(it∗ − it )
Pt+1 ct+1

A.27 Proposition 6.6 in Section 6.D

Proof. From Proposition 6.4, plug the last two equations into the two
above:
!   1− α
α
h 1 i ct  1  c∗ 
s∗ t

s t + c t Et ∑ ∞ k−t sk
k = t +1 β ck s∗t +c∗t Et ∑∞
k = t +1 β k−t k
c ∗
ct = ℓ̄αα (1 − α)1−α !α 
k
 1− α

Et −1 [ h 1
k−t sk
i ct ℓν /zt ]
t
 Et −1 [  1
s∗
 c∗ (ℓ∗ )ν /z∗ ] 
t t t
s t + c t Et ∑ ∞
k = t +1 β c s∗t +c∗t Et ∑∞ β k−t k
k k = t +1 c∗
k
!1− α  α
h 1 i ct 1  c∗ 
s∗ t

k−t sk

s t + c t Et ∑ ∞
k = t +1 β ck s∗t +c∗t Et ∑∞ βk−t ck∗
k = t +1
c∗t = ℓ̄αα (1 − α)1−α !1− α 
k
α .

Et −1 [ h 1
k−t sk
i ct ℓν /zt ]
t
 Et −1 [  1
s∗
 c∗ (ℓ∗ )ν /z∗ ] 
t t t
s t + c t Et ∑ ∞
k = t +1 β ck s∗t +c∗t Et ∑∞ βk−t ck∗
k = t +1
k

We conjecture that ct and c∗t are functions of st and s∗t only. Then,
by the i.i.d. assumption of the
i government surpluses,i the expecta-
s∗
h h
tion terms Et ∑∞
k = t +1 β
k −t sk and E
ck

t ∑ k = t +1 β
k −t k are constants,
c∗ k

Electronic copy available at: https://ssrn.com/abstract=4668578


306 jiang – lecture notes on international finance

denoted as A and A∗ . The equations become

( st +1ct A ct )α ( s∗ +1c∗ A∗ c∗t )1−α


1− α t t
ct = ℓ̄α (1 − α)
α
(Et−1 [ st +1ct A ct ℓνt /zt ])α (Et−1 [ s∗ +1c∗ A∗ c∗t (ℓ∗t )ν /z∗t ])1−α
t t

( st +1ct A ct )1−α ( s∗ +1c∗ A∗ c∗t )α


c∗t = ℓ̄αα (1 − α)1−α t t
.
(Et−1 [ st +1ct A ct ℓνt /zt ])1−α (Et−1 [ s∗ +1c∗ A∗ c∗t (ℓ∗t )ν /z∗t ])α
t t

By the i.i.d. assumption of st , s∗t , zt , and z∗t , the expectation terms


Et−1 [ st +1ct A ct ℓνt /zt ] and Et−1 [ s∗ +1c∗ A∗ c∗t (ℓ∗t )ν /z∗t ] do not vary across
t t
periods. The two equations above can be used to solve ct and c∗t as
functions of st and s∗t , which confirms the conjecture.
Take log,

ℓ̄αα (1 − α)1−α ct c∗
log ct = log ℓν c∗ (ℓ∗ )ν
+ α log( ) + (1 − α) log( ∗ t ∗ ∗ )
(Et−1 [ st +cct t A ztt ])α (Et−1 [ s∗ +ct∗ A∗ zt∗ ])1−α st + ct A st + ct A
t t t

ℓ̄α (1 − α)
α 1 − α ct c∗
log c∗t = log ℓν c∗ (ℓ∗ )ν
+ (1 − α) log( ) + α log( ∗ t ∗ ∗ )
(Et−1 [ st +cct t A ztt ])1−α (Et−1 [ s∗ +ct∗ A∗ zt∗ ])α st + ct A st + ct A
t t t

Take first-order apprximation around st = s̄ and ct = c̄,


1 c̄A
log(st + ct A) ≈ log(s̄ + c̄A) + (st − s̄) + (log ct − log c̄)
s̄ + c̄A s̄ + c̄A
1 c̄A∗
log(s∗t + c∗t A∗ ) ≈ log(s̄ + c̄A∗ ) + ( s ∗
t − s̄ ) + (log c∗t − log c̄)
s̄ + c̄A∗ s̄ + c̄A∗
Then, the system of equations becomes

(1 − α)s̄ + c̄A (1 − α)s̄ α 1−α ∗


log ct − log c∗t = κ2 − st − s
s̄ + c̄A s̄ + c̄A∗ s̄ + c̄A s̄ + c̄A∗ t
(1 − α)s̄ + c̄A∗ (1 − α)s̄ α 1−α
log c∗t − log ct = κ2∗ − s∗ − st
s̄ + c̄A∗ s̄ + c̄A s̄ + c̄A∗ t s̄ + c̄A
where
s̄ + c̄A ct ℓνt s̄ + c̄A∗ c∗t (ℓ∗t )ν
κ2 = log(ℓ̄αα (1 − α)1−α ) − α log(Et−1 [ ]) − (1 − α) log(Et−1 [ ∗ ])
st + ct A zt st + c∗t A∗ z∗t
s̄ s̄ c̄A c̄A∗
+α + (1 − α ) + α log c̄ + ( 1 − α ) log c̄
s̄ + c̄A s̄ + c̄A∗ s̄ + c̄A s̄ + c̄A∗
s̄ + c̄A ct ℓνt s̄ + c̄A∗ c∗t (ℓ∗t )ν
κ2∗ = log(ℓ̄αα (1 − α)1−α ) − (1 − α) log(Et−1 [ ]) − α log(Et−1 [ ∗ ])
st + ct A zt st + c∗t A∗ z∗t
s̄ s̄ c̄A∗ c̄A
+α ∗
+ (1 − α ) +α log c̄ + (1 − α) log c̄
s̄ + c̄A s̄ + c̄A s̄ + c̄A∗ s̄ + c̄A
The solution is
((1 − α)s̄ + αA∗ c̄)st + (1 − α)( Ac̄ + s̄)s∗t
log ct = κc −
c̄( AA∗ c̄ + (1 − α)s̄( A + A∗ ))
(1 − α)( A∗ c̄ + s̄)st + ((1 − α)s̄ + αAc̄)s∗t
log c∗t = κC∗ −
c̄( AA∗ c̄ + (1 − α)s̄( A + A∗ ))

Electronic copy available at: https://ssrn.com/abstract=4668578


proof of selected results 307

where

( Ac̄ + s̄)( A∗ c̄κ2 + (1 − α)s̄(κ2 + κ2∗ ))


κc =
c̄( AA∗ c̄ + (1 − α)s̄( A + A∗ ))
( A c̄ + s̄)( Ac̄κ2∗ + (1 − α)s̄(κ2 + κ2∗ ))

κc∗ =
c̄( AA∗ c̄ + (1 − α)s̄( A + A∗ ))

Recall the real Euler equation:

rt = − log δ − log Et [ct /ct+1 ]


= − log δ − log ct − log Et [1/ct+1 ]

Plugging log consumption into the equation above yields:

((1 − α)s̄ + αA∗ c̄) (1 − α)( Ac̄ + s̄)


rt = − log δ − log Et [1/ct+1 ] − κc + st + s∗
∗ ∗
c̄( AA c̄ + (1 − α)s̄( A + A )) c̄( AA c̄ + (1 − α)s̄( A + A∗ )) t

The log real exchange rate −et = log ct − log c∗t . The log price level
is

log Pt = log Qt − log (st + ct A)


((1 − α)s̄ + αA∗ c̄)st + (1 − α)( Ac̄ + s̄)s∗t
 
1 c̄A
≈ log Qt − log(s̄ + c̄A) − (st − s̄) − κ− − log c̄
s̄ + c̄A s̄ + c̄A c̄( AA∗ c̄ + (1 − α)s̄( A + A∗ ))
1−α A∗ c̄ + s̄( A∗ /A) (1 − α )
= κ P + log Qt − st + ∗ s∗
s̄ + c̄A A∗ c̄ + (1 − α)s̄(1 + A∗ /A) A c̄ + (1 − α)s̄(1 + A∗ /A) t

where

1 c̄A
κ P = − log(s̄ + c̄A) − (−s̄) − (κ − log c̄) .
s̄ + c̄A s̄ + c̄A

Setting A = A∗ yields the expressions in the proposition.

A.28 Proposition 7.1 in Section 7.A

Proof. Divide the first condition by the third, and the second by the
fourth to get:

α c F,t 1 − α c F,t µ def
= = H,t = exp(tott )
1 − α c H,t α c∗H,t µ F.t

plugging in exp(tott ) into the aggregation rule of home and foreign


goods yields:
 1− α α
1−α
 
α
ct = exp(tott ) c H,t = exp(−tott ) c F,t
α 1−α
α  1− α
1−α
 
∗ α ∗
ct = exp(tott ) c H,t = exp(−tott ) c∗F,t
1−α α

Electronic copy available at: https://ssrn.com/abstract=4668578


308 jiang – lecture notes on international finance

then, plug in the equations above into the market clearing conditions
to get the consumption rule
 α −1 −α
1−α
 
α
exp(tott ) ct + exp(tott ) c∗t = yt
α 1−α
−α  α −1
1−α
 
α
exp(−tott ) ct + exp(−tott ) c∗t = y∗t
1−α α

Note that we have shown in Section 1.A that:


α c F,t pt
= ∗
1 − α c H,t pt exp(−et )

which implies that exp(tott ) is actually the relative price between two
goods. Dividing the first condition by the second one to pin down
the expression of exp(tott ):

π (ct )−γ 1−2α


∗ exp( tot t ) =1
1 − π (c∗t )−γ

A.29 Proposition 7.2 in Section 7.C

Proof. Linearize the Euler equation to get


"  
c̄ −γ
#
exp(−λ̄ H,t )(−λ H,t + λ̄ H + 1) = Et δ exp(r̄ )(1 − γ∆ log ct+1 + rt − r̄ )

⇒ −λ H,t + λ̄ H = Et [−γ∆ log ct+1 + rt − r̄ ],

where −λ̄ H = log δ + r̄, which implies

−λ H,t = Et [−γ∆ log ct+1 + rt + log δ].

Similarly,

−λ∗H,t = Et [−γ∆ log c∗t+1 + ∆et+1 + rt + log δ],

which implies

−(λ∗H,t − λ H,t ) = Et [m∗t+1 − mt+1 ] + Et [∆et+1 ] = Et [γ(∆ log ct+1 − ∆ log c∗t+1 )] + Et [∆et+1 ]
⇒ et = (λ∗H,t − λ H,t ) + Et [γ(∆ log ct+1 − ∆ log c∗t+1 )] + et+1 .

Iterate forward to get


∞ ∞
et = ∑ (λ∗H,t+ j − λ H,t+ j ) + ∑ Et [γ(∆ log ct+ j+1 − ∆ log c∗t+ j+1 )] + jlim
→∞
e t + j +1 .
j =0 j =0

Electronic copy available at: https://ssrn.com/abstract=4668578


proof of selected results 309

Let lim j→∞ et+ j+1 = ē. Plugging ∆ct+1 = log ct+1 − log ct yields

∞ ∞
et − ē = ∑ (λ∗H,t+ j − λ H,t+ j ) + ∑ Et [γ(log ct+ j+1 − log ct+ j − log c∗t+ j+1 + log c∗t+ j )]
j =0 j =0

= ∑ (λ∗H,t+ j − λ H,t+ j ) − γ(log ct − log c∗t ) + γ(log c̄ − log c̄∗ )
j =0

where we let c̄ = lim j→∞ ct+ j+1 , c̄∗ = lim j→∞ c∗t+ j+1 . This yields Eq.
(7.4). Finally, recall that

ω ∗H (b∗H,t )−σ − (c̄∗ )−γ θ H,t



exp(−λ∗H,t ) = 1 − exp(−et )
(c∗t )−γ
ω H (b H,t )−σ
exp(−λ H,t ) = 1 − −γ
ct

Plug in exp(−λ H,t ) ≈ 1 − λ H,t and exp(−λ∗H,t ) ≈ 1 − λ∗H,t yields Eq.


(7.5).

A.30 Proposition 8.1 in Section 8.A

Proof. Rearrange Eq. (8.1) to get:

Dt = −St + Dt−1 exp( RtD )

iterate backwards to get:


 
Dt = −St + −St−1 + Dt−2 exp( RtD−1 ) exp( RtD )
1
= − ∑ St− j exp( RtD− j→t ) + Dt−2 exp( RtD−2→t )
j =0
1
= − ∑ St− j exp( RtD− j→t ) − St−2 exp( RtD−2→t ) + Dt−3 exp( RtD−2 ) exp( RtD−2→t )
j =0
2
= − ∑ St− j exp( RtD− j→t ) + Dt−3 exp( RtD−3→t )
j =0

similarly,

k
Dt = − ∑ St− j exp( RtD− j→t ) + Dt−k−1 exp( RtD−k−1→t )
j =0

where RtD− j→t = RtD + RtD−1 + · · · + RtD− j+1 and RtD→t = 0. Let k = t − 1
to get Eq. (8.2):

t −1
Dt = − ∑ St− j exp( RtD− j→t ) + D0 exp( R0D→t )
j =0

Electronic copy available at: https://ssrn.com/abstract=4668578


310 jiang – lecture notes on international finance

Divide both sides by Yt to get Eq. (8.3):


t −1 S
Dt t− j D0
=−∑ exp( RtD− j→t ) + exp( R0D→t )
Yt j =0
Y t Yt

t −1 S
Dt t− j Yt− j D0 Y0
⇒ =−∑ exp( RtD− j→t ) + exp( R0D→t )
Yt j =0
Y t − j Yt Y0 Yt

t −1St − j D0
=−∑ exp( RtD− j→t − Xt− j→t ) + exp( R0D→t − X0→t )
Y
j =0 t − j
Y0

where exp( Xt− j→t ) = Yt /Yt− j .

A.31 Proposition 8.2 in Section 8.B

Proof. First, let us assume that the government never defaults. Con-
sider the time t + 1 constraints,
H H
∑ Qt (h) Pt+1 (h − 1) = St+1 + ∑ Qt+1 (h) Pt+1 (h)
h =1 h =1

multiply both sides by nominal SDF exp( Mt+1 ) and take expectations
conditional on time t:
" #
H H
∑ Qt (h) Pt (h) = Et [exp( Mt+1 )St+1 ] + Et exp( Mt+1 ) ∑ Qt+1 (h) Pt+1 (h)
h =1 h =1
(A.36)

where we use the asset pricing equations Et [exp( Mt+1 )] = Pt (1), and
Et [exp( Mt+1 ) Pt+1 (h − 1)] = Pt (h).
Similarly, the time t + 2 constraints imply:
" #
H H
∑ Qt+1 (h) Pt+1 (h) = Et+1 [exp( Mt+1,t+2 )St+2 ] + Et+1 exp( Mt+1,t+2 ) ∑ Qt+2 (h) Pt+2 (h)
h =1 h =1

Plug in Eq. (A.36) to get:


H
∑ Qt (h) Pt (h) = Et [exp( Mt+1 )St+1 ] + Et [exp( Mt+1 ) exp( Mt+1,t+2 )St+2 ]
h =1
" #
H
+ Et exp( Mt+1 ) exp( Mt+1,t+2 ) ∑ Qt+2 (h) Pt+2 (h)
h =1
" # " #
2 H
= Et ∑ exp( Mt,t+k )St,t+k + Et exp( Mt,t+2 ) ∑ Qt+2 (h) Pt+2 (h)
k =1 h =1

iterate forward to time t + n to get:


" # " #
H n H
∑ Qt (h) Pt (h) = Et ∑ exp( Mt,t+k )St,t+k + Et exp( Mt,t+n ) ∑ Qt+n (h) Pt+n (h)
h =1 k =1 h =1
(A.37)

Electronic copy available at: https://ssrn.com/abstract=4668578


proof of selected results 311

More specifically, assume that the equation above is true for n = ℓ.


For n = ℓ + 1, the constraints in time t + ℓ + 1 imply:
" #
H H
∑ Qt+ℓ (h) Pt+ℓ (h) = Et+ℓ [exp( Mt+ℓ,t+ℓ+1 )St+ℓ+1 ] + Et+ℓ exp( Mt+ℓ,t+ℓ+1 ) ∑ Qt+ℓ+1 (h) Pt+ℓ+1 (h)
h =1 h =1

it’s easy to show that the expression is also true for n = ℓ + 1 by


plugging the equation above in Eq. (A.37) (while letting n = ℓ). Let
n → ∞ and plug in budget constraint Eq. (8.5) to get Eq. (8.6).
Now, consider the case of default. We consider only full default,
without loss of generality. In case of default at t, the one-period bud-
get constraint is given by:
H
0 = St + ∑ Qt (h) Pt (h)
h =1

We use χt as an indicator for default. The budget constraints now


become:
H H
(1 − χ t ) ∑ Qt−1 (h) Pt (h − 1) = St + ∑ Qt (h) Pt (h)
h =1 h =1

Consider the time t + 1 constraint,


H H
(1 − χ t +1 ) ∑ Qt (h) Pt+1 (h − 1) = St+1 + ∑ Qt+1 (h) Pt+1 (h)
h =1 h =1

multiply both sides by nominal SDF exp( Mt+1 ) and take expectations
conditional on time t:
" #
H H
∑ Qt (h) Pt (h) = Et [exp( Mt+1 )St+1 ] + Et exp( Mt+1 ) ∑ Qt+1 (h) Pt+1 (h)
h =1 h =1

where we use the asset pricing equations Et [exp( Mt+1 )(1 − χt+1 )] =
Pt (1), and Et [exp( Mt+1 ) Pt+1 (h − 1)(1 − χt+1 )] = Pt (h).
Again, we reach the condition in Eq. (A.36). Iterate forward the
same way in no default case to time n → ∞ and we will get Eq. (8.6)
as long as the government does not default in time t. Since bond
payment is not certain any more, the money that government can
raise from 1% face value of bond drops (Pt decrease).

A.32 Proposition 8.3 in Section 8.B

Proof. We present the proof for general case with default. Consider
the time t + 1 constraint,
H H
(1 − χ t +1 ) ∑ Qt (h) Pt+1 (h − 1) = St+1 + ∑ Qt+1 (h) Pt+1 (h)
h =1 h =1

Electronic copy available at: https://ssrn.com/abstract=4668578


312 jiang – lecture notes on international finance

multiply both sides by nominal SDF exp( Mt+1 ) and take expectations
conditional on time t:
" #
H H
∑ Qt (h) Pt (h)e −λt (h)
= Et [exp( Mt+1 )St+1 ] + Et exp( Mt+1 ) ∑ Qt+1 (h) Pt+1 (h)
h =1 h =1
H H  
∑ Qt (h) Pt (h) = Et [exp( Mt+1 )St+1 ] + ∑ Qt (h) Pt (h) 1 − e−λt (h)
h =1 h =1
" #
H
+ Et exp( Mt+1 ) ∑ Qt+1 (h) Pt+1 (h) (A.38)
h =1

where we use the asset pricing equations Et [exp( Mt+1 )(1 − χt+1 )] =
Pt (1)e−λt (1) , and Et [exp( Mt+1 ) Pt+1 (h − 1)(1 − χt+1 )] = Pt (h)e−λt (h) .
Similarly, consider the time t + 2 budget constraint. Multiply
both sides by nominal SDF exp( Mt+1,t+2 ) and take expectations
conditional on time t + 1:

H H  
∑ Qt+1 (h) Pt+1 (h) = Et+1 [exp( Mt+1,t+2 )St+2 ] + ∑ Qt+1 (h) Pt+1 (h) 1 − e−λt+1 (h)
h =1 h =1
" #
H
+ Et+1 exp( Mt+1,t+2 ) ∑ Qt+2 (h) Pt+2 (h)
h =1

plug it into the RHS of Eq. (A.38) to get:


" # " #
H 2 1 H  
∑ Qt (h) Pt (h) = Et ∑ exp( Mt,t+k )St+k + Et ∑ exp( Mt,t+k ) ∑ Qt+k (h) Pt+k (h) 1−e −λt+k ( h)
h =1 k =1 k =0 h =1
" #
H
+ Et exp( Mt,t+2 ) ∑ Qt+2 (h) Pt+2 (h) .
h =1

To iterate forward, assume that


" # " #
H n n −1 H
∑ Qt (h) Pt (h) = Et ∑ exp( Mt,t+k )St+k + Et ∑ exp( Mt,t+k ) ∑ Qt+k (h) Pt+k (h)(1 − e −λt+k ( h)
)
h =1 k =1 k =0 h =1
" #
H
+ Et exp( Mt,t+n ) ∑ Qt+n (h) Pt+n (h) (A.39)
h =1

is true for n = ℓ. For n = ℓ + 1, consider the time t + ℓ + 1


budget constraint. Similarly, multiply both sides by nominal SDF
exp( Mt+ℓ,t+ℓ+1 ) and take expectations conditional on time t + ℓ:

H H  
∑ Qt+ℓ (h) Pt+ℓ (h) = Et+ℓ [exp( Mt+ℓ,t+ℓ+1 )St+ℓ+1 ] + ∑ Qt+ℓ (h) Pt+ℓ (h) 1 − e−λt+ℓ (h)
h =1 h =1
" #
H
+ Et+ℓ exp( Mt+ℓ,t+ℓ+1 ) ∑ Qt+ℓ+1 (h) Pt+ℓ+1 (h)
h =1

Electronic copy available at: https://ssrn.com/abstract=4668578


proof of selected results 313

plug it into the RHS of Eq. (A.39) to get:


" # " #
H ℓ ℓ−1 H
∑ Qt (h) Pt (h) = Et ∑ exp( Mt,t+k )St+k + Et ∑ exp( Mt,t+k ) ∑ Qt+k (h) Pt+k (h)(1 − e −λt+k ( h)
)
h =1 k =1 k =0 h =1
" #
H  
+ Et [exp( Mt,t+ℓ )Et+ℓ [exp( Mt+ℓ,t+ℓ+1 )St+ℓ+1 ]] + Et exp( Mt,t+ℓ ) ∑ Qt+ℓ (h) Pt+ℓ (h) 1 − e−λt+ℓ (h)
h =1
" " ##
H
+ Et exp( Mt,t+ℓ )Et+ℓ exp( Mt+ℓ,t+ℓ+1 ) ∑ Qt+ℓ+1 (h) Pt+ℓ+1 (h)
h =1

Rearrange the RHS to get:


" # " #
H ℓ+1 ℓ H
∑ Qt (h) Pt (h) = Et ∑ exp( Mt,t+k )St+k + Et ∑ exp( Mt,t+k ) ∑ Qt+k (h) Pt+k (h)(1 − e −λt+k ( h)
)
h =1 k =1 k =0 h =1
" #
H
+ Et exp( Mt,t+ℓ+1 ) ∑ Qt+ℓ+1 (h) Pt+ℓ+1 (h)
h =1

Hence, we can iterate the equation from n = ℓ to n = ℓ + 1 for any


ℓ ≥ 1 and prove Eq. (A.39). Recall the time t budget condition when
default does not happen in Eq. (8.5). Plugging ∑hH=1 Qt (h) Pt (h) into
the RHS and letting n → ∞ yields the result:
∞ ∞
" # " #
H H
∑ Qt−1 (h) Pt (h − 1) = Et ∑ exp( Mt,t+k )St+k + Et ∑ exp( Mt,t+k ) ∑ Qt+k (h) Pt+k (h)(1 − e−λt+k (h) )
h =1 k =0 k =0 h =1
" #
H
+ lim Et exp( Mt,t+k )
k→∞
∑ Qt+k (h) Pt+k (h)
h =1

note that we replace letter n to k for simplicity.

A.33 Proposition 8.4 in Section 8.B

Proof. By the definition of exp( RtD+1 ), exp( RtT+1 ) and exp( RtG+1 ):

PtT exp( RtT+1 ) − PtG exp( RtG+1 ) = PtT+1 + Tt+1 − PtG+1 − Gt+1 = Dt exp( RtD+1 )

Divide both sides by Dt and take expectations to get part (a). Simi-
larly,

covt ( Dt exp( RtD+1 ), exp( RtM+1 ))


Dt β D
t =
vart (exp( RtM+1 ) − exp( Rt ))
covt ( PtT+1 + Tt+1 − PtG+1 − Gt+1 , exp( RtM+1 ))
=
vart (exp( RtM+1 ) − exp( Rt ))
covt ( PtT exp( RtT+1 ) − PtG exp( RtG+1 ), exp( RtM+1 ))
=
vart (exp( RtM+1 ) − exp( Rt ))
= PtT β Tt − PtG βG
t

Again, divide both sides by Dt to get part (b).

Electronic copy available at: https://ssrn.com/abstract=4668578


314 jiang – lecture notes on international finance

A.34 Proposition 8.5 in Section 8.B

Proof. Multiply both sides of Eq. (8.12) by Dt /( PtT + PtK − PtG ) and
plug in Dt = PtT + PtK + Kt − PtG to obtain
!
D Kt
exp( Rt+1 ) 1 + T
Pt + PtK − PtG
PtT PtK PtG
= exp( RtT+1 ) + exp ( R K
t +1 ) − exp ( R G
t +1 ) .
PtT + PtK − PtG PtT + PtK − PtG PtT + PtK − PtG
Taking conditional expectation Et for both sides yields part (a).
To get part (b), take conditional covariance with market return
exp( RtM+1 ) and divide by its variance for both sides
!
covt (exp( RtD+1 , exp( RtM+1 ))) Kt
1+ T
vart (exp( RtM+1 )) Pt + PtK − PtG
covt (exp( RtT+1 , exp( RtM+1 ))) PtT
=
vart (exp( RtM+1 )) PtT + PtK − PtG
covt (exp( RtK+1 , exp( RtM+1 ))) PtK
+
vart (exp( RtM+1 )) PtT + PtK − PtG
covt (exp( RtG+1 , exp( RtM+1 ))) PtG
− ,
vart (exp( RtM+1 )) PtT + PtK − PtG
which implies
!
Kt PtT PtK PtG
βD
t 1+ T = β Tt + β K
t − β G
t ,
Pt + PtK − PtG PtT + PtK − PtG PtT + PtK − PtG PtT + PtK − PtG
where we plug in the expression of beta defined above.

A.35 Proposition 8.6 in Section 8.C

We will use the Envelope theorem as a lemma.


Lemma A.1. ψ′ (wt , t) = δt u′ (ct ).
Proof. Let ({cs }∞ ∞
s=t , { ds }s=t ) be the optimal solution from wt . Let
∆w > 0, define a feasible process ({c∗s }∞ ∗ ∞
s=t , { ds }s=t ) from wt + ∆w,
∗ ∗ ∗ ∗
such that ct = ct + ∆w, dt = dt , cs = cs , ds = ds , s = t + 1, t + 2, . . .. It
is clear that

" #
ψ(wt + ∆w, t) ≥ Et ∑ δs u(c∗s )
s=t
∞ ∞
" # " #
⇒ ψ(wt + ∆w, t) − ψ(wt , t) ≥ Et ∑δ s
u(c∗s ) − Et ∑ δ u(cs )
s
s=t s=t

= Et δt u(ct + ∆w) − δt u(ct )


 

= δt u′ (ct )∆w + o (∆w)

Electronic copy available at: https://ssrn.com/abstract=4668578


proof of selected results 315

Hence,
ψ(wt + ∆w, t) − ψ(wt , t)
lim ≥ δt u′ (ct )
∆w→0 ∆w
This implies that ψ′ (wt , t) ≥ δt u′ (ct ). Now, take ∆w sufficiently small
such that 0 < ∆w < ct . Define a feasible process ({c∗s }∞ ∗ ∞
s=t , { ds }s=t )
∗ ∗ ∗
from wt − ∆w, such that ct = ct − ∆w, dt = dt , cs = cs , ds = ds ,∗

s = t + 1, t + 2, . . .. Similarly,

" #
ψ(wt − ∆w, t) ≥ Et ∑ δs u(c∗s )
s=t
∞ ∞
" # " #
⇒ ψ(wt − ∆w, t) − ψ(wt , t) ≥ Et ∑ δs u(c∗s ) − Et ∑ δs u(cs )
s=t s=t

= Et δt u(ct − ∆w) − δt u(ct )


 

= δt u′ (ct )(−∆w) + o (∆w)

Hence,
ψ(wt − ∆w, t) − ψ(wt , t)
lim ≤ δt u′ (ct )
∆w→0 −∆w
This implies that ψ′ (wt , t) ≤ δt u′ (ct ), hence ψ′ (wt , t) = δt u′ (ct ).

Proof. First, we prove the necessity of the transversality condition.


Since ψ(wt , t) is differentiable at wt , by the concavity of u(c),

ψ(wt , t) ≥ ψ(wt , t) − ψ(wt /2, t) ≥ ψ′ (wt , t)wt /2.

By Envelope theorem, i.e., ψ′ (wt , t) = δt u′ (ct ), then the assumptions


limt→∞ ψ(wt , t) = 0 and u′ (ct ) ≥ 0 implies

lim E0 δt u′ (ct )wt = 0


 
t→∞
h  i
t ′
lim E0 δ u (ct ) yt + dt−1 exp(rtd ) = 0.
t→∞

Since yt , u′ (ct ) ≥ 0, and by the short-sale constraint, dt−1 ≥ 0,


then,
h i h  i
0 ≤ E0 δt u′ (ct )dt−1 exp(rtd ) ≤ E0 δt u′ (ct ) yt + dt−1 exp(rtd )
h i h  i
0 ≤ lim E0 δt u′ (ct )dt−1 exp(rtd ) ≤ lim E0 δt u′ (ct ) yt + dt−1 exp(rtd ) = 0
t→∞ t→∞
h i
lim E0 δt u′ (ct )dt−1 exp(rtd ) = 0
t→∞

Lastly, by the Euler equation (8.13),


h i
u′ (ct−1 )dt−1 = Et−1 δu′ (ct )dt−1 exp(rtd )
h i
E0 u′ (ct−1 )dt−1 = E0 δu′ (ct )dt−1 exp(rtd ) .
 

Electronic copy available at: https://ssrn.com/abstract=4668578


316 jiang – lecture notes on international finance

then
h i h i
0 = lim E0 δt u′ (ct )dt−1 exp(rtd ) = lim E0 δt−1 u′ (ct−1 )dt−1 ,
t→∞ t→∞

which implies the transversality condition.

A.36 Proposition 8.7 in Section 8.C

Proof. First, We derive the Euler equation with utility from holding
the bonds. The Lagrangian is:

∞ ∞
" #
E0 ∑ δt (u(ct ) + vt (dt )) + ∑ ζ t (yt + dt−1 exp(rtD ) − ct − dt )
t =0 t =0

The first order conditions are given by

δt u′ (ct ) − ζ t = 0
δt v′t (dt ) − ζ t + Et [ζ t+1 exp(rtD+1 )] = 0

which imply the Euler equation

v′t (dt ) − u′ (ct ) + Et [δu′ (ct+1 ) exp(rtD+1 )] = 0

hence,

δu′ (ct+1 ) v′ (dt )


Et [ ′
exp(rtD+1 )] = 1 − t′
u (ct ) u (ct )

we denote the RHS as exp(−λt ). Similarly, we define the households’


value function as

" #
ψ(wt , t) = max E0 ∑ δs (u(cs ) + vt (dt ))
s=t

we can show in the same way in lemma A.1 that

ψ′ ( wt , t ) = δt u′ ( ct )

by the concavity of ψ(wt , t) and that limt→∞ ψ(wt , t) = 0:


wt
0 ≤ lim ψ′ (wt , t) ≤ lim (ψ(wt , t) − ψ(wt /2, t)) ≤ lim ψ(wt , t) = 0
t→∞ 2 t→∞ t→∞

hence,

lim E0 [δt u′ (ct )wt ] = 0


t→∞
lim E0 [δt u′ (ct )(yt + dt−1 exp(rtD ))] = 0
t→∞

Electronic copy available at: https://ssrn.com/abstract=4668578


proof of selected results 317

since yt , dt , u′ (ct ) are all non-negative, plugging in the Euler equation


to get:

lim E0 [δt u′ (ct )dt−1 exp(rtD )] =0


t→∞
⇒ lim E0 [δt−1 dt−1 Et−1 [δu′ (ct ) exp(rtD )]] =0
t→∞
h i
⇒ lim E0 δt−1 dt−1 u′ (ct−1 ) − v′t (dt−1 ) =0
t→∞

replacing time subscripts to t to get:

lim E0 δt u′ (ct ) − v′t (dt ) dt = 0


  
t→∞
⇒ lim E0 δt u′ (ct )dt exp(−λt ) = 0
 
t→∞

the expression of exp(−λt ) is given above.


To obtain the original transversality condition, assume that λt is
bounded above by λ̄

lim E0 δt u′ (ct )dt = lim E0 δt u′ (ct )dt exp(−λt ) exp(λt )


   
t→∞ t→∞
≤ lim E0 δt u′ (ct )dt exp(−λt ) exp(λ̄)
 
t→∞
= exp(λ̄) lim E0 δt u′ (ct )dt exp(−λt )
 
t→∞
=0

A.37 Proposition 8.8 in Section 8.E

Proof. (a) Conjecture the log price of the nominal bond is

log Pt (h) = A(h) + B(h)′ zt .

The Euler equation implies

exp(log Pt−1 (h + 1)) = Et−1 [exp( Mt + log Pt (h))]


1
A(h + 1) + B(h + 1)′ zt−1 = −it−1 (1) − Λ′t−1 Λt−1 + A(h) + B(h)′ Ψzt−1
2
1 1 1
+ Λ′t−1 Λt−1 + B(h)′ ΣB(h) − B(h)′ Σ 2 Λt−1
2 2
1 1
= −i0 (1) + A(h) + B(h)′ ΣB(h) − B(h)′ Σ 2 Λ0
2
1
+ (−ei′ + B(h)′ Ψ − B′ (h)Σ 2 Λ1 )zt−1 .

Compare coefficients to obtain


1 1
A(h + 1) = −i0 (1) + A(h) + B(h)′ ΣB(h) − B(h)′ Σ 2 Λ0 ,
2
1
B(h + 1)′ = B(h)′ Ψ − ei′ − B(h)′ Σ 2 Λ1 ,

Electronic copy available at: https://ssrn.com/abstract=4668578


318 jiang – lecture notes on international finance

with the initial condition given by A(0) = 0 and B(0) = ⃗0.


(b) Denote the time t log price-dividend ratio on the stock divi-
dend strip that pays Divt+h at time t + h as pdmt ( h ). The Euler equa-
tion is

exp( pdm m
t−1 ( h + 1)) = Et [exp( Mt + pdt ( h ) + ∆ log Divt )],

where ∆ log Divt = ∆ log dt + xt + πt , i.e., the log nominal divi-


dend growth is the sum of the log growth of dividend-GDP ratio, log
growth of real GDP and inflation. Conjecture that

pdm m m
t ( h) = A ( h) + B ( h) zt .

Plug ∆ log Divt into the the Euler equation to obtain

1
Am (h + 1) + Bm (h + 1)′ zt−1 = −it−1 (1) − Λ′t Λt + Am (h) + Bm (h)′ Ψzt−1 + x0 + π0 + (e∆d + ex + eπ )′ Ψzt−1
2
1 1
+ Λ′t Λt + ( Bm (h) + e∆d + ex + eπ )′ Σ( Bm (h) + e∆d + ex + eπ )
2 2
1
− ( Bm (h) + e∆d + ex + eπ )′ Σ 2 Λt
1
= −i0 (1) + Am (h) + x0 + π0 + ( Bm (h) + e∆d + ex + eπ )′ Σ( Bm (h) + e∆d + ex + eπ )
2
1
− ( Bm (h) + e∆d + ex + eπ )′ Σ 2 Λ0
1
+ (−ei′ + ( Bm (h) + e∆d + ex + eπ )′ Ψ − ( Bm (h) + e∆d + ex + eπ )′ Σ 2 Λ1 )zt−1 .

By matching coefficients, we obtain

1
Am (h + 1) = −i0 (1) + Am (h) + x0 + π0 + ( Bm (h) + e∆d + ex + eπ )′ Σ( Bm (h) + e∆d + ex + eπ )
2
1
− ( Bm (h) + e∆d + ex + eπ )′ Σ 2 Λ0 ,
1
Bm (h + 1)′ = −ei′ + ( Bm (h) + e∆d + ex + eπ )′ Ψ − ( Bm (h) + e∆d + ex + eπ )′ Σ 2 Λ1 .

with initial condition Am (0) = 0, Bm (0) = 0.


(c) The derivation is very similar to that in (b).

A.38 Proposition 8.9 in Section 8.E

Proof. (a) Recall that


1 1
A(h + 1) = −i0 (1) + A(h) + B(h)′ ΣB(h) − B(h)′ Σ 2 Λ0 ,
2
1
B(h + 1)′ = B(h)′ Ψ − ei′ − B(h)′ Σ 2 Λ1 ,

which further implies


 1
 −1  1

B(h)′ = −ei′ I − (Ψ − Σ 2 Λ1 ) I − ( Ψ − Σ 2 Λ1 ) h

Electronic copy available at: https://ssrn.com/abstract=4668578


proof of selected results 319

and
1 1
lim A(h)/h = −i0 (1) + B(∞)′ ΣB(∞) − B(∞)′ Σ 2 Λ0 ,
h→∞ 2
 1
 −1
′ ′ ′
B(∞) = lim B(h) = −ei I − (Ψ − Σ 2 Λ1 )
h→∞

The expected return of holding the long-term bond from time t to


t + 1 is

Et [log Pt+1 (h − 1) − log Pt (h)] = A(h − 1) + B(h − 1)′ Ψzt − A(h) − B(h)′ zt
1 1
= i0 (1) − B(h − 1)′ ΣB(h − 1) + B(h − 1)′ Σ 2 Λ0
2
+ B(h − 1)′ Ψzt − B(h)′ zt .

Letting h → ∞ yields

1 1
lim Et [log Pt+1 (h − 1) − log Pt (h)] = i0 (1) − B(∞)′ ΣB(∞) + B(∞)′ Σ 2 Λ0 + B(∞)′ (Ψ − I )zt ,
h→∞ 2
1 1
lim it (h) = i0 (1) − B(∞)′ ΣB(∞) + B(∞)′ Σ 2 Λ0 .
h→∞ 2
(b) Recall that
1
Am (h + 1) = −i0 (1) + Am (h) + x0 + π0 + ( Bm (h) + e∆d + ex + eπ )′ Σ( Bm (h) + e∆d + ex + eπ )
2
1
− ( Bm (h) + e∆d + ex + eπ )′ Σ 2 Λ0 ,
1
Bm (h + 1)′ = −ei′ + ( Bm (h) + e∆d + ex + eπ )′ Ψ − ( Bm (h) + e∆d + ex + eπ )′ Σ 2 Λ1 .

Note that log Ptm (h) = pdm


t ( h ) + log Divt . Similarly,

Et [log Ptm+1 (h − 1) − log Ptm (h)]


=Et [ pdm m
t+1 ( h − 1) − pdt ( h ) + log Divt+1 − log Divt ]
= Am (h − 1) + Bm (h − 1)′ Ψzt − Am (h) − Bm (h)′ zt + (e∆d + ex + eπ )′ Ψzt
1
=i0 (1) − ( Bm (h − 1) + e∆d + ex + eπ )′ Σ( Bm (h − 1) + e∆d + ex + eπ )
2
1
+( Bm (h − 1) + e∆d + ex + eπ )′ Σ 2 Λ0 + Bm (h − 1)′ Ψzt − Bm (h)′ zt + (e∆d + ex + eπ )′ Ψzt .

Let h → ∞ to obtain

lim Et [log Ptm+1 (h − 1) − log Ptm (h)]


h→∞
1
=i0 (1) − ( Bm (∞) + e∆d + ex + eπ )′ Σ( Bm (∞) + e∆d + ex + eπ )
2
1
+( B (∞) + e∆d + ex + eπ )′ Σ 2 Λ0 + Bm (∞)′ (Ψ − I )zt + (e∆d + ex + eπ )′ Ψzt ,
m

where Bm (∞)′ is given by


 1
 1
 −1
Bm (∞)′ = −ei′ + (e∆d + ex + eπ )′ (Ψ − Σ 2 Λ1 ) I − (Ψ − Σ 2 Λ1 ) ,

Electronic copy available at: https://ssrn.com/abstract=4668578


320 jiang – lecture notes on international finance

from which we can further simplify ( Bm (∞) + e∆d + ex + eπ )′ to


1 1
( Bm (∞) + e∆d + ex + eπ )′ = Bm (∞)′ + (e∆d + ex + eπ )′ ( I − (Ψ − Σ 2 Λ1 ))( I − (Ψ − Σ 2 Λ1 ))−1
 1 1
 1
= −ei′ + (e∆d + ex + eπ )′ (Ψ − Σ 2 Λ1 ) + (e∆d + ex + eπ )( I − (Ψ − Σ 2 Λ1 )) ( I − (Ψ − Σ 2 Λ1 ))−1
′  1
 −1
= −ei′ + e∆d + ex + eπ I − ( Ψ − Σ 2 Λ1 ) .

A.39 Proposition 9.1 in Section 9.A

Proof. Substitute the n f at , ibt , cgt expressions into Eq. (9.4) to get:

at ∑ x F,t (ι) − a∗t ∑ x ∗H,t (ι) − at−1 ∑ x F,t−1 (ι) + a∗t−1 ∑ x ∗H,t−1 (ι)
ι ι ι ι
= tbt + at−1 ∑ x F,t−1 (ι)d∗t (ι) − a∗t−1 ∑ w∗H,t−1 (ι)dt (ι) + at−1 ∑ x F,t−1 (ι)ρ∗t (ι) − a∗t−1 ∑ x ∗H,t−1 (ι)ρt (ι)
ι ι ι ι

Re-arrange the equation and plug in Eq. (9.3):

tbt = at − ∑ mt (ι) − at−1 ∑ x F,t−1 (ι)(1 + d∗t (ι) + ρ∗t (ι)) + a∗t−1 ∑ x ∗H,t−1 (1 + dt (ι) + ρt (ι)).
ι ι ι

Recall the dynamics of home household wealth given by Eq. (9.2):


!
a t = a t −1 ∑ x H,t−1 (ι)(1 + dt (ι) + ρt (ι)) + ∑ xF,t−1 (ι)(1 + d∗t (ι) + ρ∗t (ι)) + st .
ι ι

Plug into the tbt expression to get


!
tbt = st − ∑ mt (ι) − ∑ mt−1 (ι)(1 + ρt (ι) + dt (ι)) .
ι ι

A.40 Proposition 9.2 in Section 9.B

Proof. Conjecture that


1− γ
wt
ψt (wt , zt ) = δt f t (zt ) ,
1−γ

note that households consume all their wealth in the last period, i.e.
1− γ
wt
ψt (wt , zt ) = δt
1−γ

which implies the boundary condition for f t (zt ):

f T (z T ) = 1.

Electronic copy available at: https://ssrn.com/abstract=4668578


proof of selected results 321

Plug the conjectured expression of ψt+1 (wt+1 , zt+1 ) into the recur-
sive form of ψt (wt , zt ) to get
1− γ 1− γ
( " #)
c w
ψt (wt , zt ) = max δt t + Et δ t +1 f t +1 ( z t +1 ) t +1
ct ,xt 1−γ 1−γ
1− γ
((wt − ct ) xt′ exp(rt+1 ))1−γ
( " #)
t ct t +1
= max δ + Et δ f t +1 ( z t +1 )
ct ,xt 1−γ 1−γ
1− γ 1− γ
( " #)
c ( x ′ exp(r ))
t + 1
= max δt t + δ t +1 ( w t − c t )1− γ Et f t +1 ( z t +1 ) t
ct ,xt 1−γ 1−γ
( γ −1
( xt′ exp(rt+1 ))1−γ
" #)
t 1− γ wct 1 1− γ
= δ wt max + δ (1 − ) Et f t +1 ( z t +1 )
wct ,xt 1−γ wct 1−γ

since δ(1 − wc1 t )1−γ is always positive and xt only enters into the
expectation term in the objective function, the optimal portfolio can
be solved for independently of the optimal consumption, i.e.
( xt′ exp(rt+1 ))1−γ
 
xt = arg max Et f t+1 (zt+1 )
xt 1−γ
which gives Eq. (9.9). For now, denote
( x ′ exp(rt+1 ))1−γ
 
gt (zt ) = max Et f t+1 (zt+1 ) t
xt 1−γ
such that 1 = xt′ 1. Plug gt (zt ) back into ψt (wt , zt ) to get
( γ −1 )
1 − wc 1
ψt (wt , zt ) = δt wt max
γ t
+ δ (1 − )1− γ gt ( z t )
wct 1−γ wct

the first order condition w.r.t. wct gives


γ −2 1 − γ −2
0 = −wct + δ(1 − γ)(1 − ) wct gt (zt )
wct
⇒ 1 = δ(1 − γ)(wct − 1)−γ gt (zt )
⇒ wct = 1 + (δ(1 − γ) gt (zt ))1/γ
Hence, wct depends only on t and zt but wt . Meanwhile, the first
order conditions also shows that
(wct − 1)γ
gt ( z t ) =
δ (1 − γ )
plug gt (zt ) back into the value function to get
" γ −1 #
1 − wc 1 ( wc t − 1 ) γ
ψt (wt , zt ) = δt wt
γ t
+ δ (1 − )1− γ
1−γ wct δ (1 − γ )
1− γ h
wt γ −1 γ −1
i
= δt wct + wct (wct − 1)
1−γ
1− γ
wt
= δt wc
γ
1−γ t

Electronic copy available at: https://ssrn.com/abstract=4668578


322 jiang – lecture notes on international finance

which confirms our conjecture and implies that


γ
f t (zt ) = wct

note that f T (z T ) = wc T = 1, which fulfills the boundary condition.

A.41 Propsition 9.3 in Section 9.B

Proof. When the asset returns are i.i.d., the optimal portfolio choice is
given by
h i
p
xt = arg max E exp((1 − γ)rt+1 )
xt
h i 1
p p
= arg max exp(E (1 − γ)rt+1 + var ((1 − γ)rt+1 ))
xt 2
h i 1
p p
= arg max E (1 − γ)rt+1 + var ((1 − γ)rt+1 )
xt 2
p
plug rt+1 in to get
 
h
p
i 1 p f 1 2 f 1 ′ 1
E (1 − γ ) r t +1 ′
+ var ((1 − γ)rt+1 ) = (1 − γ) xt (Et [rt+1 − rt ] + σt ) + rt − xt Σt xt + (1 − γ)2 xt′ Σt xt
2 2 2 2

the first order condition w.r.t. xt is given by

f 1
Et [rt+1 − rt ] + σt2 − γΣxt = 0
2
Hence,
 
f 1 2
x t = γ −1 Σ − 1
E [ r t +1 − r ] + σ .
t t 2 t

note that Σt must be inversible, otherwise we can synthesize another


risk-free assets with the risky assets.

A.42 Proposition 10.1 in Section 10.A

Proof. The bond market clearing condition requires that

Et [rxt+1 ] + 21 vart (∆et+1 )


exp(rt∗ )b∗F,t + exp(rt )n(exp(ψt ) − 1) − m =0
ωvart (∆et+1 )

As shocks become small, the variance term becomes second-order


and drops out. Below, we log-linearize the equation around steady
states exp(r̄ ) = exp(r̄ ∗ ) = δ−1 , ψ̄ = 0, b̄ H = b̄F = 0, ē = 0 and some ȳ.
Note that both noise traders and intermediaries are holding zero-cost

Electronic copy available at: https://ssrn.com/abstract=4668578


proof of selected results 323

portofolios, hence b H,t + exp(−et )bF,t = 0.

Et [∆et+1 + rt − rt∗ ]
= exp(r̄ )nψt − exp(ē + r̄ ∗ − r̄ )b H,t exp(rt )
ωvart (∆et+1 )/m
b exp(rt )
⇒ Et [∆et+1 + rt − rt∗ ] = χ1 ψt − χ2 H,t ,

where χ1 = nωvart (∆et+1 )/mδ, χ2 = ωvart (∆et+1 )Ȳ/m .

A.43 Proposition 10.2 in Section 10.B

Proof. Assuming log-normality, the 4 Euler equations can be written


as
1
−λt = Et [mt+1 ] + rt + vart (mt+1 ),
2
1 1
−ξ t = Et [mt+1 ] + rt∗ − Et [∆et+1 ] + vart (mt+1 ) + vart (∆et+1 ) − covt (mt+1 , ∆et+1 ),
2 2
1
−ξ t∗ = Et [m∗t+1 ] + rt∗ + vart (m∗t+1 ),
2
1 1
∗ ∗
−λt = Et [mt+1 ] + rt + Et [∆et+1 ] + vart (m∗t+1 ) + vart (∆et+1 ) + covt (m∗t+1 , ∆et+1 ).
2 2
Adding up the first and third equation and subtracting the second
and the fourth yields

covt (mt+1 − m∗t+1 , ∆et+1 ) = vart (∆et+1 ) + (λ∗t − λt ) − (ξ t∗ − ξ t ).

A.44 Proposition 10.3 in Section 10.C

Proof. First, subtract the first first-order condition by the second and
plug in the consumption aggregation equations to get:

−γ(ĉt ( a) − ĉ∗t ( a)) + (1 − α)ĉ F,t ( a) − αĉ∗F,t + αĉ∗H,t ( a) − (1 − α)ĉ H,t ( a) = 0


⇒ (α − 1 − γα)ĉ H,t ( a) + (1 − γ)(1 − α)ĉ F,t ( a) + (α + γ − γα)ĉ∗H,t ( a) + (γ − 1)αĉ∗F,t ( a) = 0
(A.40)

similarly, subtract the third first-order condition by the fourth and


plug in the consumption aggregation equation to get:

−γ(ĉt ( a) − ĉ∗t ( a)) − αĉ F,t ( a) + (1 − α)ĉ∗F,t ( a) − (1 − α)c∗H,t ( a) + αĉ H,t ( a) = 0


⇒ (1 − γ)αĉ H,t ( a) + (αγ − γ − α)ĉ F,t ( a) + (γ − 1)(1 − α)ĉ∗H,t ( a) + (γα + 1 − α)ĉ∗F,t ( a) = 0
(A.41)

Next, consider the inactive households’ consumption aggregation


equations their within-period solution and plug in their budget con-

Electronic copy available at: https://ssrn.com/abstract=4668578


324 jiang – lecture notes on international finance

straints to get:
αĉ H,t (i ) + (1 − α)ĉ F,t (i ) = τ̂t
αĉ∗F,t (i ) + (1 − α)ĉ∗H,t (i ) = τ̂t∗
ĉ H,t (i ) − ĉ F,t (i ) = ĉ H,t ( a) − ĉ F,t ( a)
ĉ∗H,t (i ) − ĉ∗F,t (i ) = ĉ∗H,t ( a) − ĉ∗F,t ( a)
taken active households’ consumption as given, the 4 simultaneous
equations solve for inactive households’ consumption:
ĉ H,t (i ) = τ̂t + (1 − α)ĉ H,t ( a) − (1 − α)ĉ F,t ( a)
ĉ F,t (i ) = τ̂t − αĉ H,t ( a) + αĉ F,t ( a)
ĉ∗H,t (i ) = τ̂t∗ + αĉ∗H,t ( a) − αĉ∗F,t ( a)
ĉ∗F,t (i ) = τ̂t∗ − (1 − α)ĉ∗H,t ( a) + (1 − α)ĉ∗F,t ( a)
plug into the social planner’s resource constraints to get
ȳŷt = ϕ(c̄ H ( a)ĉ H,t ( a) + c̄∗H ( a)ĉ∗H,t ( a))
+ (1 − ϕ)(c̄ H (i )(τ̂t + (1 − α)ĉ H,t ( a) − (1 − α)ĉ F,t ( a)) + c̄∗H (i )(τ̂t∗ + αĉ∗H,t ( a) − αĉ∗F,t ( a)))
(A.42)
ȳ∗ ŷ∗t = ϕ(c̄ F ( a)ĉ F,t ( a) + c̄∗F ( a)ĉ∗F,t ( a))
+ (1 − ϕ)(c̄ F (i )(τ̂t − αĉ H,t ( a) + αĉ F,t ( a)) + c̄∗F (i )(τ̂t∗ − (1 − α)ĉ∗H,t ( a) + (1 − α)ĉ∗F,t ( a)))
(A.43)
Recast Eq. (A.40), Eq. (A.41), Eq. (A.42) and Eq. (A.43) to solve for
ĉ H,t ( a), ĉ F,t ( a), ĉ∗H,t ( a) and ĉ∗F,t ( a) simultaneously:

ȳŷt − (1 − ϕ)c̄ H (i )τ̂t − (1 − ϕ)c̄∗H (i )τ̂t∗


   
ĉ H,t ( a)
 ĉ ( a)  ȳ∗ ŷ∗ − (1 − ϕ)c̄ (i )τ̂ − (1 − ϕ)c̄∗ (i )τ̂ ∗ 
 F,t F t
 = A −1  t F t 
 
 ∗
ĉ H,t ( a)

 0 

ĉ F,t ( a) 0
where A is given by
A=
−(1 − ϕ)(1 − α)c̄ H (i ) ϕc̄∗H ( a) + (1 − ϕ)αc̄∗H (i ) −(1 − ϕ)αc̄∗H (i )
 
ϕc̄ H ( a) + (1 − ϕ)(1 − α)c̄ H (i )

 −(1 − ϕ)αc̄ F (i ) ϕc̄ F ( a) + (1 − ϕ)αc̄ F (i ) −(1 − ϕ)(1 − α)c̄∗F (i ) ϕc̄∗F ( a) + (1 − ϕ)(1 − α)c̄∗F (i )

α − 1 − γα (1 − γ)(1 − α) α + γ − γα ( γ − 1) α
 
 
( γ − 1) α α + γ − γα (1 − γ)(1 − α) α − 1 − γα
Finally, the exchange rate is given by
êt = −γ(ĉt ( a) − ĉ∗t ( a))
meanwhile, adding up active and inactive households’ consumption
yields the total consumption, i.e.
ĉt = ϕc̄t ( a)ĉt ( a) + (1 − ϕ)c̄t (i )ĉt (i )
ĉ∗t = ϕc̄∗t ( a)ĉ∗t ( a) + (1 − ϕ)c̄∗t (i )ĉ∗t (i )

Electronic copy available at: https://ssrn.com/abstract=4668578


proof of selected results 325

A.45 Proposition 10.4 in Section 10.C

Proof. First, we show the steady state under in special case where
π = 1/2, ȳ = ȳ∗ = 1, τ̄ = τ̄ ∗ = αα (1 − α)1−α . Note that the
two countries are perfectly symmetric in the steady state. Hence,
c̄ H,t (.) = c̄∗F,t (.), c̄ F,t (.) = c̄∗H,t (.). Plug this equality into the active
households’ first-order conditions and divide the first first-order
condition by the second one to get:

c̄ F,t ( a) 1−α
= .
c̄ H,t ( a) α

As shown above, this also implies

c̄ F,t (i ) 1−α
= .
c̄ H,t (i ) α

Recall that (c̄ H,t (i ))α (c̄ F,t (i ))1−α = τ, hence


α
c̄ H (i ) = τ̄ = α,
(1 − α )1− α α α
1−α
c̄ F (i ) = τ̄ = 1 − α.
(1 − α )1− α α α

Plugging the steady state consumption into the earlier intermedi-


ate results and letting α → 1 yields:
  1 
0 0 0
 
ĉ H,t ( a) ŷ − (1 − ϕ)τ̂t
  ∗t
ϕ
γ −1 γ
 ĉ ( a)  
 F,t 
  ϕ− ϕ 0 1 ŷt − (1 − ϕ)τ̂t∗ 
= ,

− γ−

 ∗ γ 1
ĉ H,t ( a)   ϕ ϕ 1 0  0 
ĉ∗F,t ( a) 0 1
0 0 0
ϕ

which implies

1 1 ∗
ĉt ( a) = (ŷt − (1 − ϕ)τ̂t ), ĉ∗t ( a) = (ŷ − (1 − ϕ)τ̂t∗ ).
ϕ ϕ t

Note that when α → 1, c̄(i ) = c̄∗ (i ) = τ̄ → 1. Also, c̄( a) = c̄∗ ( a) →


1. Hence,

ĉt = ϕc̄( a)ĉt ( a) + (1 − ϕ)c̄(i )ĉt (i )


= ŷt − (1 − ϕ)τ̂t + (1 − ϕ)τ̂t
= ŷt ,

which implies
γ γ
êt = − (ŷt − ŷ∗t ) + (1 − ϕ)(τ̂t − τ̂t∗ ),
ϕ ϕ
ĉ∗t − ĉt = −(ŷt − ŷ∗t ).

Electronic copy available at: https://ssrn.com/abstract=4668578


326 jiang – lecture notes on international finance

And the covariance is given by


γ
covt (êt+1 , ĉ∗t+1 − ĉt+1 ) = var (ŷt+1 − ŷ∗t+1 ) − (1 − ϕ)covt (τ̂t+1 − τ̂t∗+1 , ŷt+1 − ŷ∗t+1 )

ϕ
γ 2 
= 2σy − 2(1 − ϕ)ρσy στ ,
ϕ
 
γ
2σ 2 − 2(1 − ϕ ) ρσ σ
ϕ y y τ
corrt (êt+1 , ĉ∗t+1 − ĉt+1 ) = q
γ 2 2 2
ϕ 2σy σy + (1 − ϕ ) στ − 2(1 − ϕ ) ρσy στ
1 − (1 − ϕ)ρ σστy
= r .
2
1 + (1 − ϕ)2 σστ2 − 2(1 − ϕ)ρ σστy
y

where

ρ = corrt (τ̂t+1 , ŷt+1 ) = corrt (τ̂t∗+1 , ŷ∗t+1 ),


σy2 = vart (ŷt+1 ) = vart (ŷ∗t+1 ),
στ2 = vart (τ̂t+1 ) = vart (τ̂t∗+1 ).

However, note that the Backus-Smith correlation is still 1 for active


households’ consumption. To see this, recall that the exchange rate is
given by

êt+1 = γ(ĉ∗t+1 ( a) − ĉt+1 ( a)).

Hence,

corrt (êt+1 , ĉ∗t+1 ( a) − ĉt+1 ( a)) = 1.

Electronic copy available at: https://ssrn.com/abstract=4668578


Bibliography

Daron Acemoglu, Vasco M Carvalho, Asuman Ozdaglar, and Alireza


Tahbaz-Salehi. The network origins of aggregate fluctuations.
Econometrica, 80(5):1977–2016, 2012.

Daron Acemoglu, Ufuk Akcigit, and William Kerr. Networks and


the macroeconomy: An empirical exploration. Nber macroeconomics
annual, 30(1):273–335, 2016.

Viral V Acharya, Zhengyang Jiang, Robert J Richmond, and Ernst-


Ludwig von Thadden. Divided we fall: International health and
trade coordination during a pandemic. Technical report, National
Bureau of Economic Research, 2020.

Tobias Adrian, Erkko Etula, and Tyler Muir. Financial intermediaries


and the cross-section of asset returns. The Journal of Finance, 69(6):
2557–2596, 2014.

Arash Aloosh and Geert Bekaert. Currency factors. Management


Science, 2021.

Fernando Alvarez and Urban J Jermann. Using asset prices to mea-


sure the persistence of the marginal utility of wealth. Econometrica,
73(6):1977–2016, 2005.

Fernando Alvarez, Andrew Atkeson, and Patrick J Kehoe. Money,


interest rates, and exchange rates with endogenously segmented
markets. Journal of political Economy, 110(1):73–112, 2002.

Fernando Alvarez, Andrew Atkeson, and Patrick J Kehoe. If ex-


change rates are random walks, then almost everything we say
about monetary policy is wrong. American Economic Review, 97(2):
339–345, 2007.

Fernando Alvarez, Andrew Atkeson, and Patrick J Kehoe. Time-


varying risk, interest rates, and exchange rates in general equilib-
rium. The Review of Economic Studies, 76(3):851–878, 2009.

Electronic copy available at: https://ssrn.com/abstract=4668578


328 jiang – lecture notes on international finance

Manuel Amador, Javier Bianchi, Luigi Bocola, and Fabrizio Perri.


Exchange rate policies at the zero lower bound. The Review of
Economic Studies, 87(4):1605–1645, 2020.

Leif Andersen, Darrell Duffie, and Yang Song. Funding value adjust-
ments. The Journal of Finance, 74(1):145–192, 2019.

Joe Anderson and Eric M Leeper. A fiscal accounting of covid infla-


tion. Available at SSRN 4654690, 2023.

Philip W Anderson. More is different: broken symmetry and the


nature of the hierarchical structure of science. Science, 177(4047):
393–396, 1972.

Andrew Ang and Monika Piazzesi. A no-arbitrage vector autoregres-


sion of term structure dynamics with macroeconomic and latent
variables. Journal of Monetary economics, 50(4):745–787, 2003.

Cristina Arellano. Default risk and income fluctuations in emerging


economies. American economic review, 98(3):690–712, 2008.

Andrew Atkeson, Jonathan Heathcote, and Fabrizio Perri. The end


of privilege: A reexamination of the net foreign asset position of
the united states. Technical report, National Bureau of Economic
Research, 2022.

Patrick Augustin, Mikhail Chernov, Lukas Schmid, and Dongho


Song. The term structure of cip violations. Technical report, Na-
tional Bureau of Economic Research, 2020.

Patrick Augustin, Mikhail Chernov, Lukas Schmid, and Dongho


Song. Benchmark interest rates when the government is risky.
Journal of Financial Economics, 140(1):74–100, 2021.

Stéphane Auray, Michael B Devereux, and Aurélien Eyquem. Endoge-


nous trade protection and exchange rate adjustment. Technical
report, National Bureau of Economic Research, 2019.

Stefan Avdjiev, Wenxin Du, Catherine Koch, and Hyun Song Shin.
The dollar, bank leverage, and deviations from covered interest
parity. American Economic Review: Insights, 1(2):193–208, 2019.

Kerry Back. Asset pricing and portfolio choice theory. Oxford University
Press, 2010.

David Backus, Nina Boyarchenko, and Mikhail Chernov. Term struc-


tures of asset prices and returns. Journal of Financial Economics, 129
(1):1–23, 2018.

Electronic copy available at: https://ssrn.com/abstract=4668578


bibliography 329

David K Backus and Gregor W Smith. Consumption and real ex-


change rates in dynamic economies with non-traded goods. Journal
of International Economics, 35(3-4):297–316, 1993.

David K Backus, Silverio Foresi, and Chris I Telmer. Affine term


structure models and the forward premium anomaly. The Journal of
Finance, 56(1):279–304, 2001.

Jennie Bai and Pierre Collin-Dufresne. The cds-bond basis. Financial


Management, 48(2):417–439, 2019.

Ravi Bansal and Ivan Shaliastovich. Risk and return in bond, cur-
rency and equity markets. Manuscript, Duke University, 2007.

Ravi Bansal and Ivan Shaliastovich. A long-run risks explanation of


predictability puzzles in bond and currency markets. The Review of
Financial Studies, 26(1):1–33, 2013.

David Rezza Baqaee and Emmanuel Farhi. The macroeconomic


impact of microeconomic shocks: Beyond hulten’s theorem. Econo-
metrica, 87(4):1155–1203, 2019.

Lawrence M Benveniste and Jose A Scheinkman. On the differen-


tiability of the value function in dynamic models of economics.
Econometrica: Journal of the Econometric Society, pages 727–732, 1979.

Katharina Bergant, Gian Maria Milesi-Ferretti, and Martin Schmitz.


Cross-Border Investment in Emerging Market Bonds: Stylized Facts and
Security-Level Evidence from Europe. Centre for Economic Policy
Research, 2023.

Steven Berry, James Levinsohn, and Ariel Pakes. Automobile prices in


market equilibrium. Econometrica: Journal of the Econometric Society,
pages 841–890, 1995.

Steven T Berry. Estimating discrete-choice models of product differ-


entiation. The RAND Journal of Economics, pages 242–262, 1994.

Olivier Blanchard. Public debt and low interest rates. American


Economic Review, 109(4):1197–1229, 2019.

Luigi Bocola and Guido Lorenzoni. Financial crises, dollarization,


and lending of last resort in open economies. American Economic
Review, 110(8):2524–2557, 2020.

Henning Bohn. The behavior of us public debt and deficits. the


Quarterly Journal of economics, 113(3):949–963, 1998.

Claudio EV Borio, Mobeen Iqbal, Robert N McCauley, Patrick


McGuire, and Vladyslav Sushko. The failure of covered interest
parity: Fx hedging demand and costly balance sheets. 2018.

Electronic copy available at: https://ssrn.com/abstract=4668578


330 jiang – lecture notes on international finance

Michael W Brandt, John H Cochrane, and Pedro Santa-Clara. Inter-


national risk sharing is better than you think, or exchange rates are
too smooth. Journal of Monetary Economics, 53(4):671–698, 2006.

William A Brock. Asset prices in a production economy. In The


economics of information and uncertainty, pages 1–46. University of
Chicago Press, 1982.

Markus K Brunnermeier and Yuliy Sannikov. A macroeconomic


model with a financial sector. American Economic Review, 104(2):
379–421, 2014.

Markus K Brunnermeier, Sebastian Merkel, and Yuliy Sannikov. Safe


assets: A dynamic retrading perspective. Technical report, mimeo,
2022.

Valentina Bruno and Hyun Song Shin. Capital flows and the risk-
taking channel of monetary policy. Journal of monetary economics, 71:
119–132, 2015a.

Valentina Bruno and Hyun Song Shin. Cross-border banking and


global liquidity. The Review of Economic Studies, 82(2):535–564,
2015b.

Craig Burnside, Martin Eichenbaum, and Sergio Rebelo. Prospective


deficits and the asian currency crisis. Journal of political Economy,
109(6):1155–1197, 2001.

Craig Burnside, Martin Eichenbaum, and Sergio Rebelo. Govern-


ment finance in the wake of currency crises. Journal of Monetary
Economics, 53(3):401–440, 2006.

Craig Burnside, Martin Eichenbaum, and Sergio Rebelo. Currency


crises models. In Banking Crises: Perspectives from The New Palgrave
Dictionary, pages 79–83. Springer, 2016.

Ricardo J Caballero and Emmanuel Farhi. The safety trap. The Review
of Economic Studies, 85(1):223–274, 2018.

Ricardo J Caballero and Arvind Krishnamurthy. Excessive dollar


debt: Financial development and underinsurance. The Journal of
Finance, 58(2):867–893, 2003.

Ricardo J Caballero, Emmanuel Farhi, and Pierre-Olivier Gourinchas.


Global imbalances and policy wars at the zero lower bound. The
Review of Economic Studies, 88(6):2570–2621, 2021.

John Y Campbell and Richard H Clarida. The dollar and real inter-
est rates. In Carnegie-Rochester Conference Series on Public Policy,
volume 27, pages 103–139. Elsevier, 1987.

Electronic copy available at: https://ssrn.com/abstract=4668578


bibliography 331

John Y Campbell and Robert J Shiller. The dividend-price ratio and


expectations of future dividends and discount factors. The Review of
Financial Studies, 1(3):195–228, 1988.

John Y Campbell and Luis M Viceira. Strategic asset allocation: portfolio


choice for long-term investors. Clarendon Lectures in Economic, 2002.

Nicolas Caramp and Sanjay R Singh. Bond premium cyclicality and


liquidity traps. Available at SSRN 3529769, 2020.

Gino Cenedese, Pasquale Della Corte, and Tianyu Wang. Currency


mispricing and dealer balance sheets. The Journal of Finance, 76(6):
2763–2803, 2021.

Thomas Chaney. The network structure of international trade. Ameri-


can Economic Review, 104(11):3600–3634, 2014.

Zefeng Chen and Zhengyang Jiang. The liquidity premium of digital


payment vehicle. Available at SSRN 4006574, 2022.

Zefeng Chen, Zhengyang Jiang, Hanno Lustig, Stijn Van Nieuwer-


burgh, and Mindy Z Xiaolan. Exorbitant privilege gained and lost:
Fiscal implications. Technical report, National Bureau of Economic
Research, 2022.

Mikhail Chernov and Drew Creal. International yield curves and


currency puzzles. The Journal of Finance, 78(1):209–245, 2023.

Mikhail Chernov, Lukas Schmid, and Andres Schneider. A macrofi-


nance view of us sovereign cds premiums. The Journal of Finance, 75
(5):2809–2844, 2020.

Mikhail Chernov, Valentin Haddad, and Oleg Itskhoki. What do


financial markets say about the exchange rate? Technical report,
Working Paper, UCLA, 2023.

YiLi Chien and Kanda Naknoi. The risk premium and long-run
global imbalances. Journal of Monetary Economics, 76:299–315, 2015.

YiLi Chien, Hanno Lustig, and Kanda Naknoi. Why are exchange
rates so smooth? a household finance explanation. Journal of Mone-
tary Economics, 112:129–144, 2020.

Jason Choi, Rishabh Kirpalani, and Diego J Perez. The macroeco-


nomic implications of us market power in safe assets. Technical
report, National Bureau of Economic Research, 2022.

Jason Choi, Duong Q Dang, Rishabh Kirpalani, and Diego J Perez.


The secular decrease in uk safe asset market power. Technical
report, National Bureau of Economic Research, 2023.

Electronic copy available at: https://ssrn.com/abstract=4668578


332 jiang – lecture notes on international finance

Lawrence J Christiano. Solving dynamic equilibrium models by a


method of undetermined coefficients. Computational Economics, 20:
21–55, 2002.

Christopher Clayton and Andreas Schaab. Multinational banks and


financial stability. The Quarterly Journal of Economics, 137(3):1681–
1736, 2022.

Christopher Clayton, Matteo Maggiori, and Jesse Schreger. A frame-


work for geoeconomics. Technical report, National Bureau of
Economic Research, 2023.

John H Cochrane. The fiscal theory of the price level: An introduction


and overview. Journal of Economic Perspectives, 2021.

John H Cochrane. The fiscal roots of inflation. Review of Economic


Dynamics, 45:22–40, 2022.

John H Cochrane. The fiscal theory of the price level. Princeton Univer-
sity Press, 2023.

Ric Colacito, Mariano M Croce, Federico Gavazzoni, and Robert


Ready. Currency risk factors in a recursive multicountry economy.
The Journal of Finance, 73(6):2719–2756, 2018a.

Ric Colacito, Max Croce, Steven Ho, and Philip Howard. Bkk the
ez way: International long-run growth news and capital flows.
American Economic Review, 108(11):3416–49, 2018b.

Riccardo Colacito and Mariano M Croce. Risks for the long run and
the real exchange rate. Journal of Political economy, 119(1):153–181,
2011.

Riccardo Colacito and Mariano M Croce. International asset pricing


with recursive preferences. The Journal of Finance, 68(6):2651–2686,
2013.

Harold L Cole and Maurice Obstfeld. Commodity trade and interna-


tional risk sharing: How much do financial markets matter? Journal
of monetary economics, 28(1):3–24, 1991.

Antonio Coppola, Arvind Krishnamurthy, and Chenzi Xu. Liquidity,


debt denomination, and currency dominance. Technical report,
National Bureau of Economic Research, 2023.

Giancarlo Corsetti and Paolo Pesenti. The simple geometry of trans-


mission and stabilization in closed and open economies [with com-
ments]. In NBER international seminar on macroeconomics, volume
2007, pages 65–129. The University of Chicago Press Chicago, IL,
2007.

Electronic copy available at: https://ssrn.com/abstract=4668578


bibliography 333

M Max Croce, Howard Kung, Thien T Nguyen, and Lukas Schmid.


Fiscal policies and asset prices. The Review of Financial Studies, 25(9):
2635–2672, 2012a.

Mariano M Croce, Thien T Nguyen, and Lukas Schmid. The market


price of fiscal uncertainty. Journal of Monetary Economics, 59(5):
401–416, 2012b.

Domenico Cuoco and Hua He. Dynamic equilibrium in infinite-


dimensional economies with incomplete financial markets. Unpub-
lished manuscript, 1994.

Magnus Dahlquist and Julien Pénasse. The missing risk premium in


exchange rates. Journal of financial economics, 143(2):697–715, 2022.

Magnus Dahlquist, Christian Heyerdahl-Larsen, Anna Pavlova, and


Julien Pénasse. International capital markets and wealth transfers.
Swedish House of Finance Research Paper, (22-15), 2022.

Qiang Dai and Kenneth J Singleton. Specification analysis of a ne


term structure models. Journal of Finance, 55(5):1943–1978, 2000.

Kent Daniel and Sheridan Titman. Evidence on the characteristics of


cross sectional variation in stock returns. the Journal of Finance, 52
(1):1–33, 1997.

Emile Despres, Charles P Kindleberger, and Walter S Salant. The


dollar and world liquidity: a minority view. The Economist, 6, 1966.

Michael B Devereux, Charles Engel, and Steve Pak Yeung Wu. Col-
lateral advantage: Exchange rates, capital flows and global cycles.
Technical report, National Bureau of Economic Research, 2023.

William Diamond, Zhengyang Jiang, and Yiming Ma. The reserve


supply channel of unconventional monetary policy. Jacobs Levy
Equity Management Center for Quantitative Financial Research Paper,
2020.

Rudiger Dornbusch. Expectations and exchange rate dynamics.


Journal of political Economy, 84(6):1161–1176, 1976.

Winston Wei Dou and Adrien Verdelhan. The volatility of interna-


tional capital flows and foreign assets. Unpublished working paper,
MIT, 2015.

Wenxin Du and Jesse Schreger. Cip deviations, the dollar, and fric-
tions in international capital markets. In Handbook of International
Economics, volume 6, pages 147–197. Elsevier, 2022a.

Electronic copy available at: https://ssrn.com/abstract=4668578


334 jiang – lecture notes on international finance

Wenxin Du and Jesse Schreger. Sovereign risk, currency risk, and


corporate balance sheets. The Review of Financial Studies, 35(10):
4587–4629, 2022b.

Wenxin Du, Joanne Im, and Jesse Schreger. The us treasury premium.
Journal of International Economics, 112:167–181, 2018a.

Wenxin Du, Alexander Tepper, and Adrien Verdelhan. Deviations


from covered interest rate parity. The Journal of Finance, 73(3):915–
957, 2018b.

Wenxin Du, Carolin E Pflueger, and Jesse Schreger. Sovereign debt


portfolios, bond risks, and the credibility of monetary policy. The
Journal of Finance, 75(6):3097–3138, 2020.

Wenxin Du, Benjamin Hébert, and Wenhao Li. Intermediary balance


sheets and the treasury yield curve. Journal of Financial Economics,
150(3):103722, 2023.

Darrell Duffie. Dynamic asset pricing theory. Princeton University


Press, 2010.

Darrell Duffie. Still the world’s safe haven? redesigning the us trea-
sury market after the covid-19 crisis. 2020.

Darrell Duffie and Rui Kan. A yield-factor model of interest rates.


Mathematical finance, 6(4):379–406, 1996.

Jonathan Eaton and Mark Gersovitz. Debt with potential repudiation:


Theoretical and empirical analysis. The Review of Economic Studies,
48(2):289–309, 1981.

Konstantin Egorov and Dmitry Mukhin. Optimal monetary policy


under dollar pricing. In Meeting Papers, volume 1510, 2019.

Ivar Ekeland and José Alexandre Scheinkman. Transversality con-


ditions for some infinite horizon discrete time optimization prob-
lems. Mathematics of operations research, 11(2):216–229, 1986.

Charles Engel. Exchange rates, interest rates, and the risk premium.
American Economic Review, 106(2):436–474, 2016.

Charles Engel and Kenneth D West. Exchange rates and fundamen-


tals. Journal of political Economy, 113(3):485–517, 2005.

Larry G. Epstein and Stanley E. Zin. Risk aversion, and the temporal
behavior of consumption and asset returns: a theoretical frame-
work. Econometrica, 57(4):937, 1989.

Electronic copy available at: https://ssrn.com/abstract=4668578


bibliography 335

Ester Faia, Juliana Salomao, and Alexia Ventula Veghazy. Granular


investors and international bond prices: Scarcity-induced safety.
Available at SSRN 4287955, 2022.

Eugene F Fama. Forward and spot exchange rates. Journal of monetary


economics, 14(3):319–338, 1984.

Sebastian Fanelli and Ludwig Straub. A theory of foreign exchange


interventions. Technical report, National Bureau of Economic
Research, 2020.

Xiang Fang. Intermediary leverage and the currency risk premium.


Available at SSRN 3290317, 2021.

Xiang Fang and Yang Liu. Volatility, intermediaries, and exchange


rates. Journal of Financial Economics, 141(1):217–233, 2021.

Xiang Fang, Bryan Hardy, and Karen K Lewis. Who holds sovereign
debt and why it matters. Technical report, National Bureau of
Economic Research, 2022.

Emmanuel Farhi and Xavier Gabaix. Rare disasters and exchange


rates. The Quarterly Journal of Economics, 131(1):1–52, 2016.

Emmanuel Farhi and Matteo Maggiori. A model of the international


monetary system. The Quarterly Journal of Economics, 133(1):295–355,
2018.

Emmanuel Farhi, Pierre-Olivier Gourinchas, and Hélène Rey. Reform-


ing the international monetary system. CEPR, 2011.

Matthias Fleckenstein, Francis A Longstaff, and Hanno Lustig. The


tips-treasury bond puzzle. the Journal of Finance, 69(5):2151–2197,
2014.

J Marcus Fleming. Domestic financial policies under fixed and under


floating exchange rates (politiques finacierieures interieures avec
un systeme de taux de change fixe et avec un systeme de taux de
change fluctuant)(politica financiera interna bajo sistemas de tipos
de cambio fijos o de tipos de cambio fluctuantes). Staff Papers-
International Monetary Fund, pages 369–380, 1962.

Sergio Florez-Orrego, Matteo Maggiori, Jesse Schreger, Ziwen Sun,


and Serdil Tinda. Global capital allocation. Technical report,
National Bureau of Economic Research, 2023.

Andrew T Foerster, Pierre-Daniel G Sarte, and Mark W Watson. Sec-


toral versus aggregate shocks: A structural factor analysis of indus-
trial production. Journal of Political Economy, 119(1):1–38, 2011.

Electronic copy available at: https://ssrn.com/abstract=4668578


336 jiang – lecture notes on international finance

Paul Fontanier. Dollar debt and the inefficient global financial cycle.
2023.

Jeffrey A Frankel. On the mark: A theory of floating exchange rates


based on real interest differentials. The American economic review, 69
(4):610–622, 1979.

Kenneth A Froot and Tarun Ramadorai. Currency returns, intrinsic


value, and institutional-investor flows. The Journal of Finance, 60(3):
1535–1566, 2005.

Xavier Gabaix. The granular origins of aggregate fluctuations. Econo-


metrica, 79(3):733–772, 2011.

Xavier Gabaix and Ralph SJ Koijen. Granular instrumental variables.


Technical report, National Bureau of Economic Research, 2020.

Xavier Gabaix and Ralph SJ Koijen. In search of the origins of fi-


nancial fluctuations: The inelastic markets hypothesis. Technical
report, National Bureau of Economic Research, 2021.

Xavier Gabaix and Matteo Maggiori. International liquidity and


exchange rate dynamics. The Quarterly Journal of Economics, 130(3):
1369–1420, 2015.

Jordi Galí. Monetary policy, inflation, and the business cycle: an intro-
duction to the new Keynesian framework and its applications. Princeton
University Press, 2015.

Elena Gerko and Hélene Rey. Monetary policy in the capitals of


capital. Journal of the European Economic Association, 15(4):721–745,
2017.

Fabio Ghironi, Jaewoo Lee, and Alessandro Rebucci. The valuation


channel of external adjustment. Journal of International Money and
Finance, 57:86–114, 2015.

Gita Gopinath and Jeremy C Stein. Banking, trade, and the making
of a dominant currency. The Quarterly Journal of Economics, 136(2):
783–830, 2021.

Pierre-Olivier Gourinchas and Helene Rey. International financial


adjustment. Journal of political economy, 115(4):665–703, 2007a.

Pierre-Olivier Gourinchas and Helene Rey. From world banker to


world venture capitalist: Us external adjustment and the exorbi-
tant privilege. In G7 current account imbalances: sustainability and
adjustment, pages 11–66. University of Chicago Press, 2007b.

Electronic copy available at: https://ssrn.com/abstract=4668578


bibliography 337

Pierre-Olivier Gourinchas and Helene Rey. Exorbitant privilege and


exorbitant duty. 2022.

Pierre-Olivier Gourinchas, Helene Rey, and Nicolas Govillot. Exor-


bitant privilege and exorbitant duty. Technical report, Institute for
Monetary and Economic Studies, Bank of Japan, 2010.

Pierre-Olivier Gourinchas, Helene Rey, and Kai Truempler. The


financial crisis and the geography of wealth transfers. Journal of
International Economics, 88(2):266–283, 2012.

Pierre-Olivier Gourinchas, Hélene Rey, and Maxime Sauzet. The


international monetary and financial system. Annual Review of
Economics, 11:859–893, 2019.

Pierre-Olivier Gourinchas, Walker Ray, and Dimitri Vayanos. A


preferred-habitat model of term premia, exchange rates, and mon-
etary policy spillovers. Technical report, National Bureau of Eco-
nomic Research, 2022.

Robin Greenwood, Samuel G Hanson, Jeremy C Stein, and Adi Sun-


deram. A quantity-driven theory of term premia and exchange
rates. Technical report, National Bureau of Economic Research,
2020.

Bryan Gutierrez, Victoria Ivashina, and Juliana Salomao. Why is dol-


lar debt cheaper? evidence from peru. Evidence from Peru (September
24, 2021), 2021.

Bryan Gutierrez, Victoria Ivashina, and Juliana Salomao. Why is


dollar debt cheaper? evidence from peru. Journal of Financial Eco-
nomics, 148(3):245–272, 2023.

George J Hall and Thomas J Sargent. Interest rate risk and other
determinants of post-wwii us government debt/gdp dynamics.
American Economic Journal: Macroeconomics, 3(3):192–214, 2011.

George J Hall and Thomas J Sargent. Three world wars: Fiscal–


monetary consequences. Proceedings of the National Academy of
Sciences, 119(18):e2200349119, 2022.

Lars Peter Hansen and Ravi Jagannathan. Implications of security


market data for models of dynamic economies. Journal of political
economy, 99(2):225–262, 1991.

Lars Peter Hansen and José A Scheinkman. Long-term risk: An


operator approach. Econometrica, 77(1):177–234, 2009.

Tarek A Hassan. Country size, currency unions, and international


asset returns. The Journal of Finance, 68(6):2269–2308, 2013.

Electronic copy available at: https://ssrn.com/abstract=4668578


338 jiang – lecture notes on international finance

Tarek A Hassan and Rui C Mano. Forward and spot exchange rates
in a multi-currency world. The Quarterly Journal of Economics, 134
(1):397–450, 2019.

Tarek A Hassan and Tony Zhang. The economics of currency risk.


Annual Review of Economics, 13:281–307, 2021.

Hua He and Neil D Pearson. Consumption and portfolio policies


with incomplete markets and short-sale constraints: The infinite
dimensional case. Journal of Economic Theory, 54(2):259–304, 1991.

Zhiguo He and Arvind Krishnamurthy. Intermediary asset pricing.


American Economic Review, 103(2):732–770, 2013.

Zhiguo He, Bryan Kelly, and Asaf Manela. Intermediary asset pric-
ing: New evidence from many asset classes. Journal of Financial
Economics, 126(1):1–35, 2017.

Zhiguo He, Arvind Krishnamurthy, and Konstantin Milbradt. A


model of safe asset determination. American Economic Review, 109
(4):1230–62, 2019.

Zhiguo He, Stefan Nagel, and Zhaogang Song. Treasury inconve-


nience yields during the covid-19 crisis. Journal of Financial Eco-
nomics, 143(1):57–79, 2022.

Christian Heyerdahl-Larsen. Asset prices and real exchange rates


with deep habits. The Review of Financial Studies, 27(11):3280–3317,
2014.

Oleg Itskhoki and Dmitry Mukhin. Exchange rate disconnect in


general equilibrium. Journal of Political Economy, 129(8):2183–2232,
2021.

Oleg Itskhoki and Dmitry Mukhin. Optimal exchange rate policy.


Techn. Rep, 1270, 2022.

Oleg Itskhoki and Dmitry Mukhin. Optimal exchange rate policy.


Technical report, National Bureau of Economic Research, 2023.

Olivier Jeanne and Andrew K Rose. Noise trading and exchange rate
regimes. The Quarterly Journal of Economics, 117(2):537–569, 2002.

Zhengyang Jiang. Us fiscal cycle and the dollar. Journal of Monetary


Economics, 124:91–106, 2021.

Zhengyang Jiang. Fiscal cyclicality and currency risk premia. The


Review of Financial Studies, 35(3):1527–1552, 2022.

Zhengyang Jiang. Exorbitant privilege: A safe-asset view. Available at


SSRN 4600477, 2023a.

Electronic copy available at: https://ssrn.com/abstract=4668578


bibliography 339

Zhengyang Jiang. Market incompleteness and exchange rate spill-


over. Technical report, National Bureau of Economic Research,
2023b.

Zhengyang Jiang and Robert Richmond. Convenience yields and


asset pricing models. Available at SSRN 4007377, 2022.

Zhengyang Jiang and Robert Richmond. Reserve asset competition


and the global fiscal cycle. Available at SSRN, 2023a.

Zhengyang Jiang and Robert J Richmond. Origins of international


factor structures. Journal of Financial Economics, 147(1):1–26, 2023b.

Zhengyang Jiang, Arvind Krishnamurthy, and Hanno Lustig. Foreign


safe asset demand for us treasurys and the dollar. In AEA Papers
and Proceedings, volume 108, pages 537–41, 2018.

Zhengyang Jiang, Hanno Lustig, Stijn Van Nieuwerburgh, and


Mindy Z Xiaolan. The us public debt valuation puzzle. Techni-
cal report, National Bureau of Economic Research, 2019.

Zhengyang Jiang, Arvind Krishnamurthy, and Hanno Lustig. Dollar


safety and the global financial cycle. Technical report, National
Bureau of Economic Research, 2020a.

Zhengyang Jiang, Hanno Lustig, Stijn Van Nieuwerburgh, and


Mindy Z Xiaolan. Manufacturing risk-free government debt. Tech-
nical report, National Bureau of Economic Research, 2020b.

Zhengyang Jiang, Hanno N Lustig, Stijn Van Nieuwerburgh, and


Mindy Z Xiaolan. Bond convenience yields in the eurozone cur-
rency union. Columbia Business School Research Paper Forthcoming,
2020c.

Zhengyang Jiang, Arvind Krishnamurthy, and Hanno Lustig. Foreign


safe asset demand and the dollar exchange rate. The Journal of
Finance, 76(3):1049–1089, 2021a.

Zhengyang Jiang, Arvind Krishnamurthy, Hanno N Lustig, and


Jialu Sun. Beyond incomplete spanning: Convenience yields and
exchange rate disconnect. 2021b.

Zhengyang Jiang, Hanno Lustig, Stijn Van Nieuwerburgh, and


Mindy Z Xiaolan. What drives variation in the us debt/output
ratio? the dogs that didn’t bark. Technical report, National Bureau
of Economic Research, 2021c.

Zhengyang Jiang, Arvind Krishnamurthy, and Hanno Lustig. The


rest of the world’s dollar-weighted return on us treasurys. Techni-
cal report, National Bureau of Economic Research, 2022a.

Electronic copy available at: https://ssrn.com/abstract=4668578


340 jiang – lecture notes on international finance

Zhengyang Jiang, Hongqi Liu, Cameron Peng, and Hongjun Yan.


Investor memory and biased beliefs: Evidence from the field. Avail-
able at SSRN, 2022b.

Zhengyang Jiang, Robert J Richmond, and Tony Zhang. A portfolio


approach to global imbalances. Technical report, National Bureau
of Economic Research, 2022c.

Zhengyang Jiang, Robert J Richmond, and Tony Zhang. Understand-


ing the strength of the dollar. Technical report, National Bureau of
Economic Research, 2022d.

Zhengyang Jiang, Arvind Krishnamurthy, and Hanno Lustig. Impli-


cations of asset market data for equilibrium models of exchange
rates. Technical report, National Bureau of Economic Research,
2023a.

Zhengyang Jiang, Hanno Lustig, Stijn Van Nieuwerburgh, and


Mindy Z Xiaolan. Fiscal capacity: An asset pricing perspective.
Annual Review of Financial Economics, 15, 2023b.

Zhengyang Jiang, Cameron Peng, and Hongjun Yan. Personality


differences and investment decision-making. Technical report,
National Bureau of Economic Research, 2023c.

Scott Joslin, Wenhao Li, and Yang Song. The term structure of liq-
uidity premium. USC Marshall School of Business Research Paper,
2021.

Rohan Kekre and Moritz Lenel. The flight to safety and interna-
tional risk sharing. Technical report, National Bureau of Economic
Research, 2021.

JM Keynes. An open letter to the french minister of finance. Essays


in Persuasion, The Collected Writing of John Maynard Keynes, 9:76–82,
1926.

Sun Yong Kim. The dollar, fiscal policy and the us safety puzzle.
Available at SSRN 4204972, 2022.

Ralph SJ Koijen and Motohiro Yogo. A demand system approach to


asset pricing. Journal of Political Economy, 127(4):1475–1515, 2019.

Ralph SJ Koijen and Motohiro Yogo. Exchange rates and asset prices
in a global demand system. Technical report, National Bureau of
Economic Research, 2020.

Ralph SJ Koijen and Motohiro Yogo. Asset demand systems in


macro-finance. NBER Reporter, (4):21–24, 2021.

Electronic copy available at: https://ssrn.com/abstract=4668578


bibliography 341

Robert Kollmann. Consumption, real exchange rates and the struc-


ture of international asset markets. Journal of International money
and finance, 14(2):191–211, 1995.

Arvind Krishnamurthy and Annette Vissing-Jorgensen. The aggre-


gate demand for treasury debt. Journal of Political Economy, 120(2):
233–267, 2012.

Philip R Lane and Gian Maria Milesi-Ferretti. The external wealth of


nations mark ii: Revised and extended estimates of foreign assets
and liabilities, 1970–2004. Journal of international Economics, 73(2):
223–250, 2007.

Eric M Leeper. Equilibria under ‘active’and ‘passive’monetary and


fiscal policies. Journal of monetary Economics, 27(1):129–147, 1991.

Eric M Leeper. Monetary science, fiscal alchemy. In Proceedings-


Economic Policy Symposium-Jackson Hole, pages 361–434. Federal
Reserve Bank of Kansas City, 2010.

Karen K Lewis. Trying to explain home bias in equities and con-


sumption. Journal of economic literature, 37(2):571–608, 1999.

Gordon Liao and Tony Zhang. The hedging channel of exchange rate
determination. International finance discussion paper, (1283), 2020.

Gordon Y Liao. Credit migration and covered interest rate parity.


Journal of Financial Economics, 138(2):504–525, 2020.

Yang Liu, Lukas Schmid, and Amir Yaron. The risks of safe assets.
Available at SSRN 3699618, 2020.

John B Long Jr and Charles I Plosser. Real business cycles. Journal of


political Economy, 91(1):39–69, 1983.

Francis A Longstaff. The flight-to-liquidity premium in us treasury


bond prices. The Journal of Business, 77(3):511–526, 2004.

Francis A Longstaff, Jun Pan, Lasse H Pedersen, and Kenneth J Sin-


gleton. How sovereign is sovereign credit risk? American Economic
Journal: Macroeconomics, 3(2):75–103, 2011.

Hanno Lustig and Robert J Richmond. Gravity in the exchange rate


factor structure. The Review of Financial Studies, 33(8):3492–3540,
2020.

Hanno Lustig and Adrien Verdelhan. The cross section of foreign


currency risk premia and consumption growth risk. American
Economic Review, 97(1):89–117, 2007.

Electronic copy available at: https://ssrn.com/abstract=4668578


342 jiang – lecture notes on international finance

Hanno Lustig and Adrien Verdelhan. Does incomplete spanning


in international financial markets help to explain exchange rates?
American Economic Review, 109(6):2208–2244, 2019.

Hanno Lustig, Nikolai Roussanov, and Adrien Verdelhan. Common


risk factors in currency markets. The Review of Financial Studies, 24
(11):3731–3777, 2011.

Hanno Lustig, Stijn Van Nieuwerburgh, and Adrien Verdelhan. The


wealth-consumption ratio. The Review of Asset Pricing Studies, 3(1):
38–94, 2013.

Hanno Lustig, Nikolai Roussanov, and Adrien Verdelhan. Counter-


cyclical currency risk premia. Journal of Financial Economics, 111(3):
527–553, 2014.

Hanno Lustig, Andreas Stathopoulos, and Adrien Verdelhan. The


term structure of currency carry trade risk premia. American Eco-
nomic Review, 109(12):4142–4177, 2019.

Matteo Maggiori. Financial intermediation, international risk sharing,


and reserve currencies. American Economic Review, 107(10):3038–
3071, 2017.

Matteo Maggiori, Brent Neiman, and Jesse Schreger. International


currencies and capital allocation. Journal of Political Economy, 128(6):
2019–2066, 2020.

Ian Martin. The forward premium puzzle in a two-country world.


Technical report, National Bureau of Economic Research, 2011.

Richard A Meese and Kenneth Rogoff. Empirical exchange rate


models of the seventies: Do they fit out of sample? Journal of inter-
national economics, 14(1-2):3–24, 1983.

Robert C Merton. Lifetime portfolio selection under uncertainty: The


continuous-time case. The review of Economics and Statistics, pages
247–257, 1969.

Silvia Miranda-Agrippino and Hélene Rey. World asset markets and


the global financial cycle. Technical report, National Bureau of
Economic Research, 2015.

Silvia Miranda-Agrippino and Hélène Rey. The global financial


cycle. In Handbook of international economics, volume 6, pages 1–43.
Elsevier, 2022.

Brian R Mitchell. International historical statistics 1750-2005: Americas.


Springer, 2007.

Electronic copy available at: https://ssrn.com/abstract=4668578


bibliography 343

Robert A Mundell. Capital mobility and stabilization policy under


fixed and flexible exchange rates. Canadian Journal of Economics and
Political Science/Revue canadienne de economiques et science politique, 29
(4):475–485, 1963.

Stefan Nagel. The liquidity premium of near-money assets. The


Quarterly Journal of Economics, 131(4):1927–1971, 2016.

Takashi Negishi. Welfare economics and existence of an equilibrium


for a competitive economy. Metroeconomica, 12(2-3):92–97, 1960.

Aviv Nevo. Measuring market power in the ready-to-eat cereal indus-


try. Econometrica, 69(2):307–342, 2001.

Maurice Obstfeld. International liquidity: the fiscal dimension. Tech-


nical report, National Bureau of Economic Research, 2011.

Jun Pan and Kenneth J Singleton. Default and recovery implicit in the
term structure of sovereign cds spreads. The Journal of Finance, 63
(5):2345–2384, 2008.

Anna Pavlova and Roberto Rigobon. Asset prices and exchange rates.
The Review of Financial Studies, 20(4):1139–1180, 2007.

Anna Pavlova and Roberto Rigobon. Equilibrium portfolios and


external adjustment under incomplete markets. In AFA 2009 San
Francisco Meetings Paper, 2012.

Bruno Pellegrino, Enrico Spolaore, and Romain Wacziarg. Barriers


to global capital allocation. Technical report, National Bureau of
Economic Research, 2021.

Robert Ready, Nikolai Roussanov, and Colin Ward. After the tide:
Commodity currencies and global trade. Journal of Monetary Eco-
nomics, 85:69–86, 2017a.

Robert Ready, Nikolai Roussanov, and Colin Ward. Commodity trade


and the carry trade: A tale of two countries. The Journal of Finance,
72(6):2629–2684, 2017b.

Hélène Rey. Dilemma not trilemma: the global financial cycle and
monetary policy independence. Technical report, National Bureau
of Economic Research, 2015.

Robert J Richmond. Trade network centrality and currency risk


premia. The Journal of Finance, 74(3):1315–1361, 2019.

Dagfinn Rime, Andreas Schrimpf, and Olav Syrstad. Covered interest


parity arbitrage. The Review of Financial Studies, 35(11):5185–5227,
2022.

Electronic copy available at: https://ssrn.com/abstract=4668578


344 jiang – lecture notes on international finance

Juliana Salomao and Liliana Varela. Exchange rate exposure and firm
dynamics. The Review of Economic Studies, 89(1):481–514, 2022.

Mirela Sandulescu, Fabio Trojani, and Andrea Vedolin. Model-free


international stochastic discount factors. The Journal of Finance, 76
(2):935–976, 2021.

Maxime Sauzet. Asset prices, global portfolios, and the international


financial system. Global Portfolios, and the International Financial
System (February 2, 2023), 2022.

Martin Schneider and Aaron Tornell. Balance sheet effects, bailout


guarantees and financial crises. The Review of Economic Studies, 71
(3):883–913, 2004.

Andreas Stathopoulos. Asset prices and risk sharing in open


economies. The Review of Financial Studies, 30(2):363–415, 2017.

Kenneth E Train. Discrete choice methods with simulation. Cambridge


university press, 2009.

Robert Triffin. Gold and the dollar crisis: the future of convertibility,
volume 39. New Haven: Yale University Press, 1960.

Jules H Van Binsbergen. Duration-based stock valuation: Reassessing


stock market performance and volatility. Technical report, National
Bureau of Economic Research, 2020.

Jules H Van Binsbergen, William F Diamond, and Marco Grotteria.


Risk-free interest rates. Journal of Financial Economics, 143(1):1–29,
2022.

Adrien Verdelhan. A habit-based explanation of the exchange rate


risk premium. The Journal of Finance, 65(1):123–146, 2010.

Adrien Verdelhan. The share of systematic variation in bilateral


exchange rates. The Journal of Finance, 73(1):375–418, 2018.

Ursula Wiriadinata. External debt, currency risk, and international mone-


tary policy transmission. The University of Chicago, 2018.

Kairong Xiao. Monetary transmission through shadow banks. The


Review of Financial Studies, 33(6):2379–2420, 2020.

Haonan Zhou. The fickle and the stable: Global financial cycle trans-
mission via heterogeneous investors. Available at SSRN 4616182,
2023.

Electronic copy available at: https://ssrn.com/abstract=4668578

You might also like