Essays on Empirical Macroeconomics

by

Jose Villegas

Submitted in Partial Fulfillment of the

Requirements for the Degree

Doctor of Philosophy

Supervised by Professor Yan Bai

Department of Economics

Arts, Sciences and Engineering

School of Arts and Sciences

University of Rochester

Rochester, New York

2022
 ii

TABLE OF CONTENTS

Abstract iv

List of Tables vi

List of Figures viii

1 Labor Fluctuations, Real Estate Prices, and Property Taxes 1

1.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1
1.2 About Property Tax Changes . . . . . . . . . . . . . . . . . . . . . . . 9
1.3 Data . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 16
1.4 Empirical Strategy . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 22
1.5 Estimation Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 28
1.6 Model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 50
1.7 Calibration Strategy . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 61
1.8 Measuring Collateral Channels: Quantitative Results . . . . . . . . . 65
1.9 Conclusions and Future Work . . . . . . . . . . . . . . . . . . . . . . . 66

2 Investment, Capital Structure and Default Risk 70

2.1 Data . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 76
2.2 Capital Structure and Firm Characteristics . . . . . . . . . . . . . . . . 78
TABLE OF CONTENTS iii

2.3 Empirical Strategy . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 80

2.4 Estimation Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 82
2.5 Model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 85
2.6 Conclusions and Further Work . . . . . . . . . . . . . . . . . . . . . . 106

Bibliography 109

Appendix 114
 iv

ABSTRACT

This dissertation consists of two chapters related to empirical macroeconomics and

financial frictions.
The first chapter focus on the study the role of real estate prices on employment
fluctuations. We focus on the relative role of collateral channels on both house-
hold and firm sides. To quantify the importance of each channel, we use empirical
evidence from Italian municipality data and a quantitative model with financial
frictions. First, we exploit variation in property tax changes across municipalities
during Italy’s 2012 property tax reform. We then use these empirical estimates
to calibrate a quantitative model that includes houses and commercial real estate
charged with a different property tax rate. To discipline the model we use the re-
duced form estimated obtained with Italian data. The calibrated model shows that
both collateral channels explain at least 80% of the decline in employment due to
lower real estate prices induced by an increase in property taxes.
The second chapter studies the effect of maturity and leverage decisions on the
response of firms’ investment to changes in the sovereign spread. The paper use
firm-level data for Italy during the period 2007-2015. The empirical results show
that the response of firms’ investment to changes in the sovereign spread will differ
depending on the choices of leverage and maturity made by the firm. For high
leverage firms, an increase in the sovereign spread by 100 basis points is associated
TABLE OF CONTENTS v

with a decrease of 0.59% in the growth rate of capital for high maturity firms relative
to low maturity firms. For low leverage firms, the same increase in the sovereign
spread is associated with an increment of 0.38% in the growth rate of capital for
high maturity firms relative to low maturity firms. For firms with low maturity,
an increase in the sovereign spread by 100 basis points is associated with a drop
of 1.41% in the growth rate of capital for firms with high leverage relative to firms
with low leverage. For firms with high maturity, the same increase in the sovereign
spread is associated with a drop of 2.41% for the growth rate of capital for firms with
high leverage relative to low leverage firms. In order to understand the empirical
results, the paper builds a model where firms can issue risky short-term and long
term debt.
 vi

LIST OF TABLES

1.1 Summary Statistics - 2012: Municipal Level Variables . . . . . . . . . 21

1.2 Baseline Estimation Results: 2012 Property Tax Reform . . . . . . . . 30
1.3 Covariate Balance: Local Economic and Financial Conditions . . . . . 41
1.4 Covariate Balance: Migration Patterns and Supply Side Controls . . . 42
1.5 Covariate Balance: Financial Situation of Local Governments . . . . . 43
1.6 Externally Defined Parameters . . . . . . . . . . . . . . . . . . . . . . 64
1.7 Calibrated Parameters . . . . . . . . . . . . . . . . . . . . . . . . . . . 64
1.8 Labor Response: Model vs Data . . . . . . . . . . . . . . . . . . . . . . 65
1.9 Household Wealth and Firm Collateral Chanel . . . . . . . . . . . . . 66

2.1 Summary Statistics - 2007 : Firm Level Variables . . . . . . . . . . . . 78

2.2 Balance across Capital Structure Groups . . . . . . . . . . . . . . . . . 80
2.3 Response of Investment and Capital Structure . . . . . . . . . . . . . . 82
2.4 Baseline Estimation Results . . . . . . . . . . . . . . . . . . . . . . . . 83
2.5 Alternative Specifications . . . . . . . . . . . . . . . . . . . . . . . . . 84
A.1 Non-Tradable NACE Industries . . . . . . . . . . . . . . . . . . . . . . 115
A.2 Tradable NACE Industries . . . . . . . . . . . . . . . . . . . . . . . . . 116
B.3 Non-Tradable Employment: Robustness Checks . . . . . . . . . . . . 118
B.4 Consumption Expenditure: Robustness Checks . . . . . . . . . . . . . 119
LIST OF TABLES vii

B.5 Residential Prices: Robustness Checks . . . . . . . . . . . . . . . . . . 120

B.6 Commercial Real Estate Prices: Robustness Checks . . . . . . . . . . . 121
 viii

LIST OF FIGURES

1.1 Average Property Tax Rates 2008-2015. . . . . . . . . . . . . . . . . . . 15

1.2 Property Tax Rate Changes - 2012 Tax Reform . . . . . . . . . . . . . . 15
1.3 Event Study: 2012 Change in the Tax rate for Residential Properties . 37
1.4 Event Study: 2012 Change in the Tax rate for Commercial Real Estate
Properties . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 38
1.5 Robustness of Baseline Estimates . . . . . . . . . . . . . . . . . . . . . 49

2.1 Sovereign Spread: Euro Countries . . . . . . . . . . . . . . . . . . . . 74

2.2 Non-financial Corporations Investment . . . . . . . . . . . . . . . . . 74
2.3 Non-financial Corporations Leverage . . . . . . . . . . . . . . . . . . . 75
2.4 Non-financial Corporations Maturity . . . . . . . . . . . . . . . . . . . 75
2.5 Average Leverage and Maturity for Italian Firms . . . . . . . . . . . . 79
 1

CHAPTER 1

LABOR FLUCTUATIONS, REAL ESTATE PRICES,

AND PROPERTY TAXES

1.1 INTRODUCTION

What are the employment consequences of a drop in real estate prices? This ques-
tion gained considerable interest, especially after the 2008-09 US financial crisis,
with the latter tracing its origins to the 2007 housing market bust. Previous work
found that a drop in residential and commercial real estate prices is associated with
an employment decline. To rationalize this empirical regularity, the literature uses
two narratives: (i) Housing wealth channel (Mian and Sufi, 2014) and (ii) Firm col-
lateral channel (Bahaj et al., 2022). For both channels, lower real estate prices reduce
labor demand through a balance sheet effect on households or firms.
 2

According to the housing wealth effect, labor demand reduces because lower res-
idential prices decrease household consumption of financially constrained house-
holds,1 resulting in lower sales for firms dependent on local market demand. On
the other hand, the firm collateral channel causes a drop in labor demand because
lower prices for commercial real estate reduce the value of assets firms use as col-
lateral to get loans. As a result, financially constrained firms cut labor demand due
to the lower available financial resources to fund regular operations.
This paper studies the employment consequences of a simultaneous drop in
residential and commercial real estate prices. We are particularly interested in the
differential effect of a decline in residential and commercial real estate prices on
labor demand through the housing wealth and the firm collateral channel, respec-
tively.
The importance of studying the effect of these channels on labor is mainly due
to its policy implications. During the 2007-08 US housing crisis, a significant drop
in residential and commercial real estate prices led to an increase in defaulting rates
for real estate-backed loans and an increase in Loan-to-Value ratios for house own-
ers and owners of commercial real estate. US policymakers’ debate focused on the
appropriate response to stabilize real estate prices, minimize employment losses,
and speed up economic recovery (Hubbard and Mayer, 2009). One of the discus-
sions was whether policy actions aimed at helping homeowners should also target
commercial real estate owners facing foreclosure.2 Measuring the relative impor-
1
Consumption expenditure reduce due to a: (i) wealth effect, (ii) housing collateral effect and
(iii) substitution effect between housing and non-housing goods. The wealth effect, in turn, is de-
fined by changes in future implicit rental costs and the change in initial housing endowment value.
Berger et al. (2018) show that the housing collateral effect, substitution effect, and the wealth effect
through changes in expected rental cost cancel each other out under certain conditions. Therefore, a
drop in housing prices uniquely reduces consumption due to a lower value of housing endowment.
In this paper, the housing wealth channel captures either (i), (ii) or a combination of both.
2
See Panel (2010), Levitin (2009), and Marsh (2011).
 3

tance of the housing wealth and firm collateral channel for employment during a
drop in real estate prices should provide better guidance to answer policy questions
of this nature.
However, quantifying both channels is not an easy task. In particular, we need
to overcome two main challenges. First, as residential and commercial real estate
prices are positively correlated (Piazzesi and Schneider, 2016), it is unclear how we
can separate both channels affecting labor demand similarly. Second, a drop in real
estate prices produces a labor market response from other mechanisms unrelated to
the housing wealth or firm collateral channel; for example, the potential adjustment
in labor supply from a wealth effect after the drop in residential prices.
This paper addresses these issues by combining a quantitative model with reduced-
form empirical evidence. On the one hand, we exploit a differential increase in
residential and commercial real estate property taxes to estimate its effect on em-
ployment and other outcome variables. The idea behind the latter is that a change
in the tax rate act as a specific shock to the real estate asset being taxed more heavily,
ultimately causing a drop in its demand and equilibrium price of, all else constant.
On the other hand, using a quantitative model with the appropriate structure, we
can measure each channel as long as we have a separate estimate for the reduced
form effect of a change in taxes for residential and commercial real estate properties.
The empirical portion of this paper relies on novel municipal-level data for Italy
during 2008-2014. Using this data, we estimate the differential effect of higher
taxes for residential and commercial real estate properties during the 2012 prop-
erty tax reform. Regarding the parametric specification, we employ a Difference-
in-Difference strategy with the change in residential and commercial real estate
property tax rates as treatment intensity variables. The outcome variables of inter-
est are employment, consumption expenditure, residential prices, and commercial
 4

real estate prices. The reason to focus on these variables is their importance in the
internal workings of the housing wealth and firm collateral channel. Our empiri-
cal results show that the higher property taxes in Italy during the 2012 tax reform
are associated with a drop in non-tradable employment, consumption expenditure,
and real estate prices.
Given the importance of our estimates for the quantitative model, our empirical
strategy needs to identify the reduced-form effect of higher property taxes correctly.
The research design in this paper points towards two conditions that identify the
average treatment effect of higher property taxes on treated municipalities. First,
municipalities should not anticipate the tax reform, so self-selection into different
property tax treatment intensities should be ruled out. Second, in the counterfac-
tual scenario of an equivalent increase in property taxes, outcomes’ trends should
be parallel, irrespective of the actual property tax increase in each municipality
during 2012.
Both assumptions seem to hold in our case. On the one hand, Alesina and Par-
adisi (2017) shows that the timing of municipal elections explains most of the dif-
ferences in property tax changes across municipalities.3 Moreover, the staggered
timing of local elections in Italy is explained by historical reasons entirely unrelated
to any business cycle unobserved factor; therefore, being in an electoral year is as
good as a random assignment. Additionally, we find no evidence for systematic
differences in outcome variable trends across municipalities with different prop-
erty tax changes during 2008-2011.
TBoth assumptions seem to hold in our case. On the one hand, Alesina and
Paradisi (2017) shows that the timing of municipal elections explains most of the
3
The authors find that municipalities with coming elections in 2013 chose a lower increase in
property tax rates during the 2012 property tax reform.
 5

differences in property tax changes across municipalities.4 Moreover, the staggered

timing of local elections in Italy is explained by historical reasons entirely unrelated
to any business cycle unobserved factor; therefore, being in an electoral year is as
good as a random assignment. Additionally, we find no evidence for systematic
differences in outcome variable trends across municipalities with different property
tax changes during 2008-2011.
The diff-in-diff estimates for employment do not provide a direct measure of
the housing wealth/firm collateral channel because other mechanisms that affect
the labor market after a change in property taxes could also be present. However,
the reduced form results can be crucial to discipline a model with a structure that
reproduces the main features of the Italian property tax reform and includes both
channels of interest. In line with the latter, the next step of the paper is to build a
quantitative model that includes two types of real estate assets, houses, and com-
mercial real; owners of real estate assets pay different property tax rates set ex-
ogenously by the government. Lastly, the model includes frictions in the financial
market by requiring real estate collateral for households and firms to get loans.
The model provides a closed-form characterization of the employment response
to a drop in real estate prices induced by an exogenous and differential increase in
property taxes. Using the latter, we can separate the contribution of the housing
wealth/firm collateral channel from the general equilibrium employment response.
Then, once we discipline the model with the reduced form estimations found with
Italian data, we can quantify each channel’s relative importance on employment
after a decline in real estate prices.
The model predicts that a one percentage point (pp) increase in the tax rate for
4
The authors find that municipalities with coming elections in 2013 chose a lower increase in
property tax rates during the 2012 property tax reform.
 6

commercial real estate reduces non-tradable employment growth by 0.061 pp with

84% (0.052 pp) of this decline explained by the collateral channel. In comparison, a
one pp increase in the tax rate for houses reduces non-tradable employment growth
by 0.074 pp with 98% (0.073 pp) of this decline explained by the housing wealth
channel. The model results clearly show that both channels are crucial to under-
standing the employment response after a decline in real estate prices induced by
an increase in property taxes.
Contribution to the Literature. This paper is mainly related to recent empirical
work in macroeconomics studying the housing cycle’s impact on employment by
exploiting regional variation in real estate wealth fluctuations. About the housing
wealth channel on employment, this paper is related to the seminal work of Mian
and Sufi (2014) and Guren et al. (2021). On the importance of the firm collateral
channel on employment, our work is linked to Adelino et al. (2015), Giroud and
Mueller (2017), and Bahaj et al. (2022).
In the context of the recent US housing bust, Mian and Sufi (2014) show that
non-tradable employment decreased relatively more in counties with a significant
decline in housing prices. The previous analysis is extended and improved in
Guren et al. (2021), they find that the importance of the housing wealth channel
on employment during the US great recession is not explained by the higher sen-
sitivity of economic activity to house prices but by the magnitude of house price
movements in that period.
On the other hand, Adelino et al. (2015) shows that during 2002-07 employ-
ment growth is more sensitive to housing price changes for small firms, which are
more dependent on collateral to obtain loans from banks. Similarly, Bahaj et al.
(2022) focuses on the importance of the firm owner’s real estate assets as a source
for collateral of the firms to estimate the heterogeneous effect of monetary policy on
 7

employment. The authors find that employment at younger, more-levered firms is

most sensitive to monetary policy shocks. Finally, Giroud and Mueller (2017) use
micro-level data for the US establishment and find that housing price declines pro-
duce a higher negative effect on employment in establishments of highly levered
firms and counties with a large share of highly levered firms.
Relative to the previous papers, our main contribution is to provide a unifying
approach that measures both channels simultaneously. We combine a novel iden-
tification strategy based on property tax changes to obtain reduced form estimates
to discipline a quantitative model to quantify each channel.
On the other hand, this paper also contributes to the literature that incorporates
reduced form estimates in the calibration of general equilibrium models to quan-
tify important unobservable channels affecting the real economy. In particular, our
paper is closely connected to Chodorow-Reich et al. (2021); the latter measures the
wealth effect on household consumption expenditure by using reduced form esti-
mates of employment response to stock market wealth changes to inform a dynamic
general equilibrium model. In contrast to the latter, our paper provides a closed-
form decomposition linking employment elasticities to local property tax changes
with the housing wealth and firm collateral channel.
Finally, this paper contributes to previous empirical work on the macroeco-
nomic effect of property taxes. The 2012 Italian property tax reform has been used
extensively in this literature. Two examples of this are Oliviero and Scognamiglio
(2019) and Surico and Trezzi (2019). The former uses housing prices as the main
outcome variable, while the latter focus on the response of consumption expendi-
ture using Italian household survey data. This paper’s empirical results for housing
prices and consumption expenditure align with the previous empirical work. In
addition, we provide new evidence on the effect of property taxes on non-tradable
 8

employment, tradable employment, and commercial real estate prices.

Layout. This organization of the paper is as follows. Section provides the intu-
ition behind the link of property taxes changes with the housing wealth and firm
collateral channel on employment and summarizes the relevant particulars of the
2012 property tax reform in Italy. Section 1.3 examines the construction of our
novel municipal-level dataset for Italy. Section 1.4 focuses on the parametric speci-
fication and the conditions for identification behind the diff-in-diff research design
employed to estimate the effect of the 2012 tax reform in Italy. Section 1.5 reports
the baseline estimation results, discuss the potential threats to identification in our
diff-in-diff strategy, and presents the evidence supporting its use in the particular
case of the 2012 tax reform in Italy. Section 1.6 describes the structure of the model
and the main decomposition linking the elasticities for the outcome variables of
interest to local property tax changes with the housing wealth and firm collateral
channel for employment. Section ?? examines the calibration procedure and shows
the model’s validation test results. Section 1.8 presents the model’s predictions for
the housing wealth and firm collateral channel on employment. Finally, section 1.9
concludes.
 9

1.2 ABOUT PROPERTY TAX CHANGES

We start the empirical portion of this paper by discussing the link between property
tax changes and the housing wealth and firm collateral channel on employment.
Next, we will provide all the relevant details characterizing the 2012 property tax
reform, which will be useful once we formulate the diff-in-diff research design used
later in the paper.
The empirical analysis of this paper focuses entirely on the 2012 property tax re-
form in Italy and its effect on non-tradable employment, consumption expenditure,
and residential and real estate prices. In the following sections, we will describe the
particularities of the empirical strategy, the conditions to achieve identification, the
issues that could bias our estimated results, and the evidence providing credibility
to our empirical strategy.
The reduced-form estimates discipline our quantitative model, providing a mea-
sure for the housing wealth and firm collateral channel. However, as we will ex-
amine later, this last step implicitly assumes that our empirical strategy correctly
identifies the reduced form effect of an increase in property taxes. In other words,
we need to thoroughly examine our research design before using its results as in-
put for the quantitative model, even if this step temporarily deviates from the main
subject of our paper.

1.2.1 PROPERTY TAX CHANGES AND THE HOUSING WEALTH AND FIRM

COLLATERAL CHANNEL

The final goal of this paper is to measure the relative importance of housing wealth
and firm collateral channels on employment. Initially, we will exploit the variation
 10

in property tax changes produced by the 2012 tax reform in Italy. The logic behind
this approach lies in the connection between property taxes and real estate market
prices. Intuitively, property taxes are part of the cost of acquiring and holding real
estate assets. Therefore, an increase in property taxes reduces demand and prices
in the real estate market. The price decline in the real estate market affects labor
demand through the housing wealth and firm collateral channel. In particular, an
increase in the tax rate for houses -all else constant- causes a drop in the demand
for employment through the housing wealth effect because higher housing taxes
produce a decline in the market price for this asset. In contrast, an increase in the
tax rate for commercial real estate -all else constant- reduces labor demand via the
firm collateral effect due to the negative impact of higher taxes on the price of com-
mercial real estate assets.
However, a drop in housing and commercial real estate prices should produce
additional mechanisms affecting the labor market that are not part of the two chan-
nels of interest. First, lower real estate prices should reduce employment due to
the effect of wealth on labor supply. On the other hand, a substitution effect be-
tween labor and commercial real estate -both inputs for production- should reduce
the demand for labor as commercial real estate becomes relatively cheaper. More-
over, a drop in household consumption expenditure due to a substitution effect
between housing and non-housing goods should reduce labor demand. Finally,
employment levels also change due to a general equilibrium adjustment of wages
and prices.
The use of variation in property tax changes to estimate its employment reduced-
form effect should capture all the mechanisms mentioned above, including the
housing wealth and firm collateral channel. The empirical strategy in this paper
aims to provide a credible identification strategy to estimate the reduced form ef-
 11

fect of changes in property taxes on employment and other key variables that also
adjust through the two channels of interest. The reduced form estimates, partially
explained by the housing wealth and firm collateral channel, measure the overall
effect of an increase in property tax rates. These estimates should provide crucial
information that can be used in a quantitative model to disentangle both channels
of interest.

1.2.2 THE ITALIAN PROPERTY TAX REFORM: INSTITUTIONAL SETTINGS

In 1992, Italian legislation introduced the ICI5 or ”Municipal tax on Properties”.

The ICI system allowed different tax rates for residential and commercial real es-
tate properties. On the one hand, residential properties used as the main dwelling
of the owner pay the prin tax. On the other hand, owners of properties used for pro-
ductive purposes and other residential properties not used as the owner’s principal
residence pay the sec tax rate.
The ICI system provided some autonomy to local municipal governments6 to
set the prin and sec tax rates. Each municipality’s mayor and local council must de-
liberate the property tax rates annually. However, two restrictions limit their au-
tonomy to set property taxes;7 first, property tax rates must be within a valid range
determined by an upper and lower bound; second, there is a deadline for tax rates
deliberation. The central government is in charge of changing both restrictions.
Starting with the ICI system, the tax base relevant for property taxation equals
5
From the Italian initials ”Imposta Comunale sugli Immobili”
6
The Italian territory is composed of approximately 8, 000 municipalities. Each municipality
has a decentralized government defined by a Mayor and a local council. The functions of municipal
governments are mainly related to providing local services such as education, public transportation,
wage disposal, and social aid. Municipal revenues come mainly from property taxes, a surcharge
on income taxes, and transfers from the central government.
7
The range for prin and sec tax rates under ICI was 0.4%-0.7%, municipalities could set a deduc-
tion of up to 258.22 euros and a tax reduction of up to 50% only for properties used as the principal
residence.
 12

the sum of cadastral incomes of each real estate unit located within the municipal-
ity. The cadastral income -annual rent ordinarily obtainable by a real estate unit-
is assigned to each property within Italy using a predetermined set of values es-
timated with data on real estate rental rates for 1988-1989. Therefore, there is no
relationship between the market value of a real estate unit and its cadastral income,
which implies that changes in real estate prices have no effect on revenues from
property taxes.
The ICI system remained reasonably unchanged during 1992-2010. The only
significant change occurred in March 2008 when the ICI tax for the primary res-
idence of households -excluding luxury residential properties- was canceled8 . In
April 2011, the government led by Prime Minister Berlusconi announced the cre-
ation of the IMU 9 or ”Own Municipal Tax” which was going to replace the old ICI
tax. The IMU system was part of a more extensive reform regarding fiscal feder-
alism prescribing a reduction of transfers from the general government to munici-
palities counterbalanced by an increase in local fiscal autonomy. Initially, the IMU
system proposed increasing property taxes and provided additional discretion for
local governments to determine the tax rates. However, it excluded the principal
residence of households that remained exempted from property taxation. The IMU
system’s introduction was expected to occur in January 2014.10
At the end of 2011, the spread of government bonds increased to new levels
unseen before. This event marked the start of the Italian sovereign debt crisis, fol-
lowed by a political crisis in the central government that ended with the resignation
of Prime Minister Berlusconi in November 2011. To ease the pressure of financial
8
Additionally, an earlier minor change was the introduction of an additional deduction of
0.133% in the tax rate on primary dwellings with a deduction cap of 200 euros at the end of 2007
9
From the Italian initials ”Imposta Municipale Unica”.
10
See Article 8, paragraph 2, of Legislative Decree no. 23/2011.
 13

markets on Italian sovereign debt, the newly appointed government led by Primer
Minister Monti pushed forward an emergency fiscal package named ”Save-Italy”
decree. The fiscal austerity measures with ”Save-Italy” decree relied primarily on
reforming the property tax system using the initial IMU tax renamed as ”Experi-
mental”-IMU. The latter canceled the tax exemption on the principal residence for
households, meaning that the increase in property taxes would affect all real estate
properties across Italy. Additionally, the introduction date for this new tax system
was moved to January 2012.
The ”Experimental”-IMU required that: (i) municipalities set the property taxes
within the range 0.2%-0.6% for the prin tax rate and 0.46%-1.6% for the sec tax rate,
(ii) tax rates had to be deliberated before October 31st (iii) the default rates of 0.4%
for prin and 0.76% for sec automatically applied if a municipality did not deliberate
before the deadline (iv) municipalities need to transfer back to the general govern-
ment about 50% of the IMU tax revenues.
The fiscal reforms increased property tax revenues from 9.5 billion euros in 2011
to 14.8 billion euros in 2012. Moreover, in 2012 municipalities reimbursed back 3.9
billion euros in property tax collections and received 8.14 billion euros less in trans-
fers from the central government relative to 2011. The characteristics of the fiscal
adjustment in 2012 translated into a modest 1 billion euro increase in total revenues
for municipalities, significantly lower than the average increase of 7.4 billion euros
during 2000-2011. Meanwhile, the central government deficit was reduced by al-
most 8 billion euros, amounting to 0.5% of the nominal GDP for Italy in 2012. How-
ever, the measures prescribed by the ”Experimental”-IMU were not well received by
the general population and municipal authorities. The widespread opposition to
this new tax system forced the central government to cancel it at the end of 2013.
As a result of the property tax reform, the prin and sec tax rates increased sig-
 14

nificantly across the Italian territory, a feature only observed in 2012. This can be
depicted in Figure 1.1 which shows a plot for the yearly average prin (blue line)
and sec (red line) tax rates across municipalities. We can see that during 2012 there
was a significant spike in property taxes. The average prin tax rate increased by 0.43
pp., while the average sec tax rate increased by 0.15 pp. Except for 2012, the average
rates remained constant even though municipalities could change property taxes
during the entire 2008-2015 period.
Another salient feature of the tax reform is related to the variation across munic-
ipalities in prin and sec tax rate changes observed in 2012. Figure 1.2 shows a heat
map for the Italian territory. The heat map on the left corresponds to the increase
in prin tax rates, while the one on the right captures the increase in the sec tax rate
across municipalities. The 2012 property tax reform produced considerable varia-
tion in tax rate changes across the Italian territory. In particular, the standard devi-
ation for prin and sec tax rate changes in 2012 are 0.096 pp and 0.10 pp respectively,
these numbers are approximately 5 times larger to the ones observed in 201111 .
11
In 2011 the standard deviation for the change in prin and sec rates are 0.019 pp. and 0.021 pp
respectively.
 15

Note: The figure shows municipalities’ average property tax rates from 2008-2015. The
red line represents the average tax rate for residential properties (prin), and the blue
line depicts the mean tax rate for commercial real estate properties (sec)

FIGURE 1.1: Average Property Tax Rates 2008-2015.

Note: The figure presents a heat-map plot of the change in property tax rates across
municipalities in Italy during the 2012 property tax reform. The figure in panel (a) is
for the residential tax rate, figure in panel (b) is for the commercial real estate tax rate.

FIGURE 1.2: Property Tax Rate Changes - 2012 Tax Reform

 16

1.3 DATA

This section presents the relevant details related to data construction. In order to
to take advantage of the significant increase in property taxes and the considerable
variation in tax rate changes across municipalities observed in 2012, the empirical
analysis builds a novel panel data set of municipal level variables. The main vari-
ables of interest are property tax rates, employment, residential and commercial
real estate prices, and consumption expenditures.

1.3.1 PROPERTY TAX RATES

The Institute for Finance and Local Economy (IFEL) provides data on property
taxes. Among its roles, the IFEL is in charge of the systematic collection, process-
ing, and dissemination of data relating to taxes for local governments. The IFEL
gathers the data on property taxes from official acts issued by municipal govern-
ments regarding the adopted tax rates for a given year. The IFEL data provides
different values for the prin and sec tax rates which are available for all municipal-
ities during 2004-2015. We will only consider municipalities created on or before
2004. Additionally, we exclude municipalities with missing data during 2008-2014.

1.3.2 EMPLOYMENT

Data on employment comes from the yearly census of active establishments (ASIA)
provided by the Italian National Statistics Institute (ISTAT). The measure of em-
ployment represents a two-digit industry aggregate (NACE Rev. 2) of employees
working in active establishments located in the municipality. For our empirical
analysis, we only keep employment of privately owned industries unrelated to con-
 17

struction.12 .
The empirical analysis focuses on municipal level employment for tradable and
non-tradable sectors. To classify industries into tradable and non-tradable sectors,
we employ two methods.13 The first classification scheme is based on aggregate
sectoral world trade. We define a two-digit NACE sector as tradable if the aggre-
gate industry ratio of total trade (exports plus imports) to gross output is higher
than 0.16 or if total trade per worker for that industry is higher than 56,000 euros.14
The second classification criteria uses the geographical concentration of employ-
ment across sectors. The idea for the method is that industries producing tradable
goods are concentrated geographically to take advantage of economies of scale; in-
stead, industries producing non-tradable goods are more dispersed to satisfy the
local demand for these goods. To proxy concentration across sectors, we compute
a Herfindahl index using each municipality’s share of the industry’s employment.
Finally, we categorize the remaining industries not classified as tradable by the pre-
vious scheme using their geographical concentration index. A two-digit NACE in-
dustry is defined as tradable if the concentration index is above the 75𝑡ℎ percentile
(0.015) and non-tradable if the concentration index is below the 25𝑡ℎ percentile
(0.015).15 The complete classification of industries into tradable and non-tradable
sectors is presented in Appendix refapp:trad_nontrad_ind.
12
This excludes industry codes 41-43 (Construction), 84-88 (Education, Health, Defense and Pub-
lic Administration), 97-99 (Activities of households as employers and extraterritorial organs and
bodies). Additionally, due to missing data problems, we exclude industry codes 01 (Agriculture,
Fishing, and Forestry) 05-09 (Mining and Quarrying).
13
The industry classification schemes are similar to the ones used in Mian and Sufi (2014).
14
The aggregate data on exports, imports, gross output, and employment come from the ISTAT
data warehouse. The threshold values correspond to the median across 88 two-digit NACE indus-
tries for 2011.
15
The concentration index for each two-digit industry is based on 2011 municipal level employ-
ment.
 18

1.3.3 REAL ESTATE PRICES

The Real Estate Market Observatory (OMI) is the primary source for real estate
prices. OMI divides each municipality into areas with similarities in urban infras-
tructure, terrain, and socioeconomic characteristics. These areas are named homo-
geneous real estate markets. OMI estimates a maximum and minimum value16 for
real estate properties in each homogeneous market within the municipality. The
base for these estimates is restricted transaction data which can be supplemented
by surveys done to real estate agents on local market conditions. The estimates
differ by property type (residential, commercial, production, tertiary.) and main-
tenance estate (excellent, normal, and poor). The data is available for 2007-2014
with bi-annual frequency.
The property type classification in the data will define our categories for res-
idential and commercial real estate properties. For residential properties, we use
well-finished houses and economic dwelling-houses; both categories are the most
prevalent residential property types across the Italian territory. For commercial
real estate, we focus on property types used and owned by firms in the non-tradable
sector. Our approach is to employ retail stores and shopping malls. Both property
types are used by the retail industry, which given the industry classification de-
tailed before, is considered a non-tradable industry. Additionally, we keep data
for residential and commercial real estate properties classified as having a normal
maintenance state. Finally, to transform the frequency of the data to annual, we
keep the values for the second semester of each year.
𝑘 𝑘
Let 𝑃𝑚(𝑖),𝑡 and 𝑃𝑚(𝑖),𝑡 denote the upper and lower bound value of property type
𝑘 in homogeneous real estate market 𝑚(𝑖) within municipality 𝑚 for the second
16
The max and min estimates define the range of prices in which the average value of real estate
units fall with the highest probability
 19

semester of year 𝑡, respectively. Where 𝑘 can be either houses (ℎ) or commercial

real estate (𝑓 ). The measure of prices for real estate type 𝑘 = {ℎ , 𝑓 } will be an
𝑘 𝑘
average based on 𝑃𝑚(𝑖),𝑡 and 𝑃𝑚(𝑖),𝑡 across all homogeneous real estate markets in
𝑘 , is defined as in (1.1).
municipality 𝑚. This average, denoted by 𝑃𝑚

𝐻 𝐻
𝑘 1⎡ 1 𝑚 𝑘 1 𝑚 𝑘 ⎤
𝑃𝑚,𝑡 = ⎢ ∑ 𝑃𝑚(𝑖),𝑡 + ∑𝑃 ⎥, (1.1)
2 𝐻𝑚 𝑖=1 𝐻𝑚 𝑖=1 𝑚(𝑖),𝑡
⎣ ⎦

𝐻𝑚 represents the total number of homogeneous real estate markets within munic-
ipality 𝑚 in which an estimate for the minimum and maximum value of real estate
type 𝑘 is available in the data.

1.3.4 CONSUMPTION EXPENDITURE

There is no available data on consumption expenditure for Italy at the municipal

level. Therefore, we proxy household expenditures with a measure of new car pur-
chases following a similar procedure as in Mian et al. (2013).
cars = 𝑃Cars 𝑄Cars be the nominal household expenditure in car purchases
Let 𝑋𝑚,𝑡 𝑚,𝑡 𝑚,𝑡
Cars is the aggregate municipal price and 𝑄Cars
for municipality 𝑚 at year 𝑡, where 𝑃𝑚,𝑡 𝑚,𝑡

is the total number of new cars purchased by households in municipality 𝑚. Equa-

cars .
tion (1.2) provides an alternative way to define 𝑋𝑚,𝑡

Cars = 𝑝Cars 𝜔𝑄 𝑋 Cars

𝑋𝑚,𝑡 (1.2)
𝑚,𝑡 𝑚,𝑡 𝑡

where 𝑋𝑡Cars is the economy wide household expenditure in new vehicle purchases
𝑃Cars 𝑄 𝑄Cars
and 𝑝Cars
𝑚,𝑡 =
𝑚,𝑡
and 𝜔𝑚,𝑡 = 𝑚,𝑡
are the relative price of new cars and the share
𝑃Cars
𝑡 𝑄Cars
𝑡

of cars purchased in municipality 𝑚, respectively. Assuming that the relative price

Cars is captured by 𝜔𝑄 and 𝑋 Cars . If we
of cars is constant in time, any change in 𝑋𝑚,𝑡 𝑚,𝑡 𝑚,𝑡
 20

𝑄 Cars for 𝜔 𝑄 Cars which are observed in the data we can

replace 𝜔𝑚,𝑡 and 𝑋𝑚,𝑡 ̂𝑚,𝑡 and 𝐶𝑚,𝑡
obtain a proxy measure for the expenditure of households in new car purchases
defined as follows:
Cars = 𝜔 𝑄
𝐶𝑚,𝑡 ̂𝑚,𝑡 𝐶𝑡Cars , (1.3)

where 𝐶𝑡Cars is the aggregate household final expenditure on vehicle purchases in

nominal terms taken from Italian yearly national accounts.
𝑄
To obtain 𝜔
̂𝑚,𝑡 , we use data on the total number of vehicle registrations in each
municipality for 2007-2014. The data provided by the Italian Automobile Club
(ACI) distinguish between nine different types of automobiles: (1) cars, (2) buses,
(3) trucks for transportation of goods, (4) special use vehicles, (5) motorcycles, (6)
motorcycles/quadricycles for special use, (7) trailers used in goods transportation,
(8) trailers for special use, and (9) tractors.
We are interested in category (1) cars, defined as any vehicle intended for trans-
porting persons, with a maximum of nine seats, including that of the driver. Finally,
we restrict the data by using only registered vehicles purchased as new. Then, we
𝑄
can define 𝜔
̂𝑚,𝑡 as follows:

𝑄
New Cars Registered𝑚,𝑡
𝜔̂ 𝑚,𝑡 = . (1.4)
∑𝑚 New Cars Registered𝑚,𝑡

Notice that this procedure allocates a fraction of the aggregate expenditure on new
car purchases to a municipality based on the observed share of new cars registered
in that municipality. The main advantage of this approach is that the sum for ex-
penditure on new cars across municipalities is consistent with the aggregate value
observed for the entire Italian economy each year.
 21

1.3.5 SUMMARY STATISTICS: SELECTED SAMPLE

The balanced panel of municipal-level information contains data for 6,246 munici-
palities17 during 2008-2014. The selected sample is representative of the total uni-
verse of municipalities across the Italian territory. In particular, our sample repre-
sents 88% of the total population and captures 89.5% and 93% of total employment
and income across the Italian territory in 2012.

TABLE 1.1: Summary Statistics - 2012: Municipal Level Variables

Mean S.D 𝑃25 𝑃50 𝑃75

Population 8,278 44,961 1,209 2,819 6,919
Area (mi2 ) 58.38 108.65 8.63 21.79 54.39
pc
Income 11,376 2,961 8,854 11,740 13,469
𝐿tot 2,193 16,502 139 489 1,554
share 𝐿ntrad (%) 41 14 31 41 50
share 𝐿trad (%) 17 15 4 12 26
Δ𝜏 ℎ 0.43 0.07 0.40 0.40 0.50
Δ𝜏 𝑓 0.24 0.10 0.16 0.25 0.31
Δ𝐿tot -0.17 7.47 -3.52 -0.67 2.54
Δ𝐿ntrad 2.44 7.95 -2.20 1.28 5.67
Δ𝐿trad -2.08 19.35 -7.73 -1.02 3.36
Δ𝐶 -5.09 71.58 -57.17 -9.61 30.07
Δ𝑃House -1.81 4.03 -4.06 0.00 0.00
Δ𝑃CRE -1.88 3.43 -3.02 0.00 0.00
Notes: The table presents summary statistics for the municipality-level data in 2012. Employment
data comes from the ASIA yearly census. Income is based on income declarations of residents
in each municipality reported to the Finance Department. Tax rates are from the official acts
issued by municipal governments regarding the adopted tax rates. Real estate prices come from
the OMI dataset. 𝐿tot , 𝐿ntrad , and 𝐿trad refer to the total, non-tradable and tradable employment.
pc
Income and 𝐶 represent the real per-capita value of income and consumption expenditure in
car purchases; respectively, we use the CPI to deflate both variables (2010=100). 𝜏 ℎ and 𝜏 𝑓 refer
to the residential (prin) and commercial real estate (sec) tax rates, respectively. All variables with
Δ are expressed in percentages.

Table 1.1 present the summary statistics for the selected sample during 2012.
The median municipality has an area of 58 square miles,18 a population of 8,278
17
In 2012, the total number of municipalities equals 8,092 with 77% included in our sample.
18
A municipality is similar to a US county. As a helpful comparison for the length of a munici-
pality, the US New York County (i.e., Manhattan) has an area of 23 square miles. Then, the median
 22

residents, and 2,193 persons working within its territorial boundaries, out of which
41% are employed in the non-tradable sector.
For the average municipality in 2012, the growth rate of total employment, per-
capita expenditure in car purchases, residential prices, and commercial real estate
prices reduced by 0.67%, 5.09%, 1.81%, 1.88%, respectively. On the other hand,
for the mean municipality, the 2012 property tax reform produced an increase in
residential and commercial property taxes equal to 0.43% and 0.24%, respectively.
The increase in property taxes translated into an additional payment of 322 € for
households owning a property used as the principal residence. At the same time,
firms owning commercial real estate properties had to pay an additional 200 €.19

1.4 EMPIRICAL STRATEGY

This section presents the particulars regarding the diff-in-diff strategy used to es-
timate the effect of an increase in property tax rates during Italy’s 2012 property
tax reform. Next, we define and interpret our diff-in-diff estimator. Finally, we an-
alyze the conditions under which our diff-in-diff estimator identifies the average
treatment effect of higher property taxes on treated municipalities.

1.4.1 PARAMETRIC SPECIFICATION

ℎ represent the prin tax rate and 𝜏 𝑓
Let 𝜏𝑚,𝑡 𝑚,𝑡 the sec tax rate choose by municipality

𝑚 in year 𝑡. The yearly change of tax rate 𝑖 = {ℎ, 𝑓 } is defined as follows is presented
in (1.5).
𝑖
Δ𝜏𝑚,𝑡 𝑖
= 𝜏𝑚,𝑡 𝑖
𝜏𝑚,𝑡 (1.5)
1

municipality in our sample is approximately equivalent to the Borough of Manhattan.

19
Property taxes in 2011 for firms owning commercial real estate properties was equivalent to
1, 270 € for the average municipality.
 23

ℎ 𝑓
For our case, Δ𝜏𝑚,2012 and Δ𝜏𝑚,2012 represent the treatment intensity variables for
each municipality during the 2012 property tax reform in Italy. The main outcome
variables are employment (𝐿), residential prices (𝑃ℎ ), commercial real estate prices
(𝑃𝑓 ), and consumption expenditure (𝐶).
ℎ ,𝑃 𝑓
Let 𝑌𝑚,𝑡 = {𝐿𝑚,𝑡 , 𝐶𝑚,𝑡 , 𝑃𝑚,𝑡 𝑚,𝑡 } be the observed value of outcome 𝑌 for mu-
𝑓
nicipality 𝑚 at year 𝑡 and 𝑦𝑚,𝑡 = {𝑙𝑚,𝑡 , 𝑐𝑚,𝑡 , 𝑝ℎ𝑚,𝑡 , 𝑝𝑚,𝑡 } its symmetric growth rate20
defined as follows:
𝑌𝑚,𝑡 𝑌𝑚,𝑡 1
𝑦𝑚,𝑡 = (1.6)
(𝑌𝑚,𝑡 + 𝑌𝑚,𝑡 1 ) /2

Equation (1.7) presents the regression model. Notice that the functional form is
a generalization of the baseline diff-in-diff specification with two periods and a
single binary treatment. Unlike the baseline diff-in-diff case, we use a continuous
treatment intensity represented by the increase in property tax changes across mu-
nicipalities during the 2012 property tax reform.

ℎ × 1{𝑡 = 2012} + 𝛽 𝑓
𝑦𝑚,𝑡 = FE𝑚 + FE𝑡 + 𝛽𝑦,ℎ Δ𝜏𝑚,𝑡 𝑦,𝑓 Δ𝜏𝑚,𝑡 × 1{𝑡 = 2012} + 𝜖𝑚,𝑡 (1.7)

We include fixed effects at the municipality (FE𝑚 ) and year level (FE𝑡 ) to control for
unobserved growth trend components that are municipality specific but constant
across time and common across municipalities but changing in time, respectively.
To define
The variable 𝜖𝑚,𝑡 capture all the unobserved trend components of 𝑦. The covari-
ance matrix for 𝜖𝑚,𝑡 allows for arbitrary correlation across municipalities within the
same local labor market. 21

20
The main advantage of the symmetric growth rate is its intrinsic bounded range of [–2,2] which
helps limit the influence of outliers. On the other hand, the symmetric growth rate is a second-order
approximation of the log difference growth rate around 0.
21
ISTAT defines local Labour markets (LLM) as sub-regional geographical areas where the bulk of
the labor force lives and works and where establishments can find most of the labor force necessary
 24

The effect of the 2012 property tax reform is captured by 𝛽𝑦,𝑖 for 𝑖 = {ℎ, 𝑓 }. The
latter, is the coefficient associated to the interaction between the treatment intensity
𝑖 ) and a post-tax reform dummy (1{𝑡 = 2012}) that takes the value of one for the
(Δ𝜏𝑚,𝑡
year 2012. We only include the year 2012 in the post-tax reform dummy due to the
null variation for tax rate changes observed across municipalities after the property
tax reform (see figures 1.1 and 1.2 in section ??). The next subsection discusses the
exact interpretation for our diff-in-diff estimator 𝛽𝑦,𝑖 .

1.4.2 THE DIFF-IN-DIFF ESTIMATOR FOR THE 2012 TAX REFORM

To interpret the parameter 𝛽𝑦,𝑖 we will consider two types of municipalities, the
first group choose a high increase in 𝜏 𝑖 denoted as Δ𝜏 𝑖,1 , while the second group
choose a low increase in 𝜏 𝑖 denoted by Δ𝜏 𝑖,0 . We will assume that the difference
between high and low tax increase is just one percentage point (i.e Δ𝜏 𝑖,1 Δ𝜏 𝑖,0 = 1).
We can define the expected change in 𝑦 for municipalities with a high tax increase
in 2012 as follows:

E [𝑦𝑚,2012 |Δ𝜏 𝑖,1 ] − E [𝑦𝑚,2012 |Δ𝜏 𝑖,1 ] = FE2012 − FE2011 + 𝛽𝑦,𝑖 Δ𝜏 𝑖,1 (1.8)

The equivalent expected change in 𝑦 for municipalities with a lower tax increase is:

E [𝑦𝑚,2012 |Δ𝜏 𝑖,0 ] − E [𝑦𝑚,2012 |Δ𝜏 𝑖,0 ] = FE2012 − FE2011 + 𝛽𝑦,𝑖 Δ𝜏 𝑖,0 (1.9)

to occupy the offered jobs. To determine an LLM, ISTAT uses daily commuting flows recorded
during the General population and housing censuses.
 25

Our parameter of interest 𝛽𝑦,𝑖 results from subtracting (1.8) and (1.9).

⎧
{ ⎫
} ⎧
{ ⎫
}
𝛽𝑦,𝑖 = ⎨E [𝑦𝑚,2012 |Δ𝜏 ]−E [𝑦𝑚,2011 |Δ𝜏 ]⎬−⎨E [𝑦𝑚,2012 |Δ𝜏 ]−E [𝑦𝑚,2011 |Δ𝜏 ]⎬
𝑖,1 𝑖,1 𝑖,0 𝑖,0
{
⎩ }
⎭ {
⎩ }
⎭
(1.10)
Notice that (1.10) defines 𝛽𝑦,𝑖 as the difference of the expected change in 𝑦 between
municipalities with a high tax rate increase and a low tax rate increase after the tax
reform. From now on we will interpret 𝛽𝑦,𝑖 as the percentage point response of 𝑦
to a one percentage point higher increase in 𝜏 𝑖 .

1.4.3 IDENTIFICATION

The diff-in-diff strategy aims to identify the average treatment effect of the 2012
property tax reform across municipalities with different treatment intensities. Fol-
lowing Rubin (1978), we will use a potential outcome framework to describe the
conditions to achieve identification.
To simplify the analysis, assume a single property tax rate that changes due to
an exogenously imposed property tax reform at date 𝑡𝑅 . The tax reform forces each
̃ = {Δ1 , Δ2 , .., Δ𝐿 }
municipality to adjust the single property tax rate upwards. Let Δ
represents the set are possible tax rate increases, where Δ𝑖−1 < Δ𝑖 for all 𝑖 = 1, .., 𝐿.
Finally, assume that the difference between two consecutive tax increases is equal
to one percentage point (i.e. Δ𝑖 − Δ𝑖−1 = 1).
𝑖
For any outcome variable 𝑦, let 𝑦𝑚,𝑡 represent the potential outcome function
showing the value of outcome 𝑦 for municipality 𝑚 with treatment intensity Δ𝜏𝑚 ∈
̃ as if was treated with Δ𝑖 at the time of the tax reform. The potential outcome
Δ
function captures the counterfactual value of 𝑦 of having a tax increase Δ𝑖 for a
̃. The observed realized value for
municipality that choose a tax increase Δ𝜏𝑚 ∈ Δ
 26

each outcome variable 𝑦 is denoted by 𝑦𝑚,𝑡 . Equation (1.11) provides a relation

𝑖
between 𝑦𝑚,𝑡 and 𝑦𝑚,𝑡

𝐿−1
𝑦𝑚,𝑡 = ∑ 1 {Δ𝜏𝑚 = Δ𝑖 } × 𝑦𝑚,𝑡
𝑖 (1.11)
𝑖=0

𝑇 𝑀
Let {(𝑦𝑚,𝑡 )𝑡=1 } denote the balanced panel data of 𝑀 municipalities, 𝑇 periods
𝑚=1
for outcome variables 𝑦. We will assume 1 < 𝑡𝑅 < 𝑇.
We are interested in the mean effect on 𝑦 of tax increase Δ𝑖 relative to a lower
increase Δ𝑖−1 during a property tax reform in period 𝑡𝑅 . We call this moment the
average treatment effect of a tax increase on treated municipalities and denote it
𝑦
with 𝐴𝑇𝑇𝑡𝑅 . Equation (1.12) provides a formal definition for the latter.

𝑦
𝐴𝑇𝑇𝑡𝑅 = E [𝑦𝑚,𝑡
𝑖
𝑅
𝑖−1 ∣ Δ𝜏 = Δ𝑖 ] .
− 𝑦𝑚,𝑡 𝑅 𝑚 (1.12)

𝑦
On the other hand, let 𝐷𝑖𝐷𝑡𝑅 denote our diff-in diff estimator with its formal defi-
nition presented in (1.13)

𝑦
𝐷𝐼𝐷𝑡𝑅 = E [𝑦𝑚,𝑡𝑅 − 𝑦𝑚,𝑡𝑅 −1 ∣ Δ𝜏𝑚 = Δ𝑖 ] − E [𝑦𝑚,𝑡𝑅 − 𝑦𝑚,𝑡𝑅 −1 ∣ Δ𝜏𝑚 = Δ𝑖−1 ] .
(1.13)
𝑦 𝑖−1 making this mo-
Notice that 𝐴𝑇𝑇𝑡𝑅 depend on the potential outcome function 𝑦𝑚,𝑡
𝑖−1 is an unknown object to the econometrician. In contrast
ment not observable as 𝑦𝑚,𝑡
𝑦
to (1.12), the diff-in-diff estimator 𝐷𝑖𝐷𝑡𝑟 in (1.13) is defined in terms of observable
𝑇 𝑀
moments and therefore can be estimated with the our sample {(𝑦𝑚,𝑡 )𝑡=1 } .
𝑚=1
𝑦
Proposition 1 provides the necessary conditions to achieve identification of 𝐴𝑇𝑇𝑡𝑅
𝑦
in the case of the 2012 property tax reform using our diff-in-diff estimator 𝐷𝑖𝐷𝑡𝑅 .
 27

𝑦
Proposition 1 (Identification of 𝐴𝑇𝑇𝑡𝑅 ): If the potential outcome function sat-
isfies:

𝑗
E [𝑦𝑚,𝑡 ∣̃ Δ𝜏𝑚 = Δ𝑖 ] = E [𝑦𝑚,
𝑖 ∣ Δ𝜏 = Δ𝑖 ]
𝑡̃ 𝑚 (1.14a)

𝑗 𝑗 𝑗 𝑗
E [𝑦𝑚,𝑡 − 𝑦𝑚,𝑡−1 ∣ Δ𝜏𝑚 = Δ𝑖 ] = E [𝑦𝑚,𝑡 − 𝑦𝑚,𝑡−1 ∣ Δ𝜏𝑚 = Δ𝑗 ] , (1.14b)

where Δ𝑗 , Δ𝑖 ∈ Δ
̃ with 𝑗 < 𝑖, and for 1 < 𝑡 ̃ < 𝑡𝑅 , and 1 < 𝑡 < 𝑇. Then:

𝑦 𝑦
𝐴𝑇𝑇𝑡𝑅 = 𝐷𝑖𝐷𝑡𝑅 .

According to (1.14a), conditional on a municipality choosing a tax increase Δ𝑖 be-

fore the tax reform, the outcome 𝑦 is not different than the counterfactual value
under a lower tax increase Δ𝑗 . Intuitively, this condition states that municipalities
do not anticipate the tax reform before the time it happens and therefore do not self-
select into different tax treatment intensities. On the other hand, condition (1.14b)
implies that the trends in 𝑦 for municipalities choosing Δ𝑖 should not be different
from the trends in 𝑦 for municipalities choosing Δ𝑗 if the former chooses the lower
tax increase Δ𝑗 .
 28

1.5 ESTIMATION RESULTS

This section presents the empirical results and the potential threats to our identifi-
cation strategy. Initially, we show the baseline regression results of the diff-in-diff
estimator for Italy’s 2012 property tax reform. Then we discuss the potential issues
that could violate conditions (1.14a) and (1.14b) and the evidence and test provid-
ing credibility to our identification strategy and diff-in-diff estimator. Finally, we
examine other potential identification threats, how we address these concerns, and
the limitations of the empirical analysis in this paper.

1.5.1 BASELINE SPECIFICATION RESULTS

The baseline results are presented in Table 1.2. The first two rows shows the diff-
in-diff estimates for 𝛽𝑦,𝑖 for the the baseline specification (1.7). The first row shows
the point estimates of an increase in residential property tax rates. The second
row corresponds to the estimated effect of an increase in commercial real estate
taxes. The third and fourth column of Table 1.2 shows the interquartile range ratio
predicted by our diff-in-diff estimators relative to the empirical one observed with
our sample for 2012.
̂ ),
Notice that all the estimated coefficients in the first two rows of columns (𝛽𝑙,𝑖
̂ ), (𝛽 ̂ ℎ ) and (𝛽 ̂ 𝑓 ) have a negative sign. The latter implies that the increase
(𝛽𝑐,𝑖 𝑝 ,𝑖 𝑝 ,𝑖

in property tax rates during the 2012 tax reform in Italy is associated with a drop in
non-tradable employment, consumption expenditure, residential prices, and com-
mercial real estate prices.
̂ ), we can see that non-tradable employment reduction after an
For column (𝛽𝑙,𝑖
increase in residential taxes almost doubles the reduction due to higher commercial
 29

real estate taxes. In particular, a one pp higher increase in the tax rate for residential
properties implies a growth reduction of 0.087 pp for non-tradable employment. A
similar increase in taxes for commercial real estate properties reduces non-tradable
employment growth by 0.045 pp. Notice that these estimates capture at least eight
percent of the interquartile range of non-tradable employment growth observed in
our sample data.
On the other hand, columns (𝛽𝑝̂ ℎ ,𝑖 ) and (𝛽𝑝̂ 𝑓 ,𝑖 ) show that the price effect on the
real estate asset tax heavily is higher than its effect on the price for the other real
estate asset. That is to say, a one pp increase in the tax rate for residential properties
reduces its price growth by 0.02 pp with an almost zero effect on the price growth
for commercial real estate properties. At the same time, a one pp increase in the
tax rate for commercial real estate properties reduces its price growth by 0.032 pp,
while its effect on the residential price growth is approximately half (0.017 pp).
̂ ), we observe that the negative effect of property taxes
Finally, in column (𝛽𝑐,𝑖
on consumption expenditure is only statistically significant for residential taxes.
In particular, consumption expenditure growth drops over 0.50 pp after a one pp
higher tax rate increase for residential properties. This estimate captures approxi-
mately six percent of our sample’s empirical interquartile range for residential price
growth. The estimated coefficient for commercial real estate taxes is not only not
statistically significant, but its magnitude captures only less than one percent of the
interquartile range observed for consumption expenditure growth in our municipal
level data.
 30

TABLE 1.2: Baseline Estimation Results: 2012 Property Tax Reform

Non-Tradable Consumption Housing Commercial RE

Employment Expenditure Price Price
̂ )
(𝛽𝑙,𝑖 ̂ )
(𝛽𝑐,𝑖 (𝛽𝑝̂ h ,𝑖 ) (𝛽𝑝̂ f ,𝑖 )
ℎ × 1 {𝑡 = 2012}
Δ𝜏𝑚,𝑡 -0.087*** -0.517*** -0.022** -0.005
(0.015) (0.145) (0.009) (0.010)
𝑓
Δ𝜏𝑚,𝑡 × 1 {𝑡 = 2012} -0.045*** -0.177 -0.017*** -0.032***
(0.011) (0.120) (0.006) (0.008)
̂ 𝑦,ℎ (%)
𝐼𝑄𝑅𝑦 /𝐼𝑄𝑅 11.0 5.9 5.5 1.6
̂ 𝑦,𝑓 (%)
𝐼𝑄𝑅𝑦 /𝐼𝑄𝑅 8.5 0.87 6.3 15.7

𝑁obs 43,540 34,881 38,731 29,471

𝑁mun 6,220 6,104 5,534 3,687
𝑅2 0.13 0.12 0.33 0.31

Municipality FE ✓ ✓ ✓ ✓
Year FE ✓ ✓ ✓ ✓
Notes: Table present the baseline results. The dependent variables are expressed in growth rates, the sample used covers the period
2008-2014. Real state prices are defined as the average value per squared meter across homogeneous real state markets within each
municipality. Commercial real estate is composed by all properties used in the retail sector, while houses are properties used by
households for residential purposes only. Consumption expenditure is proxied by expenditures on new vehicles. Standard errors in
parentheses are clustered at the local labor market level, ∗ , ∗∗ , ∗∗∗ indicate significance at the 10% 5% and 1% respectively

In the next part, we will discuss how credible our estimation results are regard-
ing the necessary identification conditions presented in Proposition 1. Our partic-
ular focus will be on the appropriateness of the two conditions for the case of the
2012 tax reform in Italy.
 31

1.5.2 ANTICIPATION OF THE 2012 PROPERTY TAX REFORM

Proposition 1 shows that condition (1.14a) requires that municipalities do not an-
ticipate the 2012 tax reform. The characteristics of the tax reform and property tax
system in Italy make this condition an appropriate assumption for the empirical
strategy in this paper. We discuss next the reasons supporting this argument.
The Experimental-IMU was completely different from what was announced in
April 2011, both in terms of timing and intensity of the tax increase. On the one
hand, the initial announcement of the introduction of the IMU tax due on January
2014 cemented expectations about the future property tax increase. Therefore, the
Experimental-IMU scheduled for January 2012 surprised the population and local
authorities, given the initial expectations. Moreover, the tax increase under the
Experimental-IMU was significantly higher and affected all property types across
the Italian territory. In other words, the way expectations formed by the end of
2011 made the 2012 property tax reform utterly unexpected for local authorities and
the general population and, therefore, impossible to anticipate and foresee across
municipalities.
Even if anticipation is not an issue for the 2012 tax reform in Italy, there is still
the issue of past property tax changes and their lag effect on the outcome variables
of interest. First, if there is a correlation in property tax changes across time, mu-
nicipalities would self-select into treatment intensity groups during the 2012 tax
reform. Second, if past changes in property taxes have a long-lasting impact on
municipal-level outcomes, then a confounding factor would contaminate the re-
sponse of the outcome variables during the 2012 tax reform. In either case, the
diff-in-diff estimator would be biased.
We argue that the latter is not of great importance for our particular empirical
 32

strategy. The main reason for this is that property tax rates across Italian munici-
palities barely change from year to year. As shown in Figure 1.1, local authorities
are reluctant to change property tax rates unless forced by the central government
during tax reforms. In other words, property tax rates are persistent and uncorre-
lated to local economic fluctuations.
All the previous arguments and evidence allow us to conclude that anticipation
issues are not of significant concern for our diff-in-diff strategy.

1.5.3 PARALLEL TRENDS AND THE 2012 TAX REFORM

𝑦
Condition (1.14b) showed the second condition necessary to identify 𝐴𝑇𝑇𝑡𝑅 with
our diff-in-diff estimator. As discussed earlier, this condition states that trends in 𝑦
need to be similar across municipalities with different treatment intensities if they
would have been treated with the lowest tax rate increase Δ0 .
Testing for parallel trends in a diff-in-diff strategy is impossible as it requires
knowledge about the outcome under alternative property tax rate changes, and we
know that counterfactual values are unknown by definition. However, the litera-
ture on the diff-in-diff estimator provides some alternatives to tests for violation in
the parallel trend condition. The only caveat is that this condition is not guaranteed
to be completely satisfied. In what follows, we will discuss the most common ap-
proach to test for violations in the parallel trend assumption, its implementation,
and the results we obtain for this test in our particular case.

EVENT-STUDY APPROACH: IMPLEMENTATION

To test for violation in the parallel trend condition using an event-study analysis,
we exploit pre-tax reform data. The idea is to check if, before the 2012 tax re-
 33

form, trends in outcome variables differ significantly across municipalities choos-

ing different property taxes increases. If the parallel trend condition is not satisfied,
we should observe significant differences in outcomes trends across municipalities
during the pre-tax reform period. However, the lack of evidence for differences in
pre-tax reform trends does not imply that the parallel trend assumption is satisfied
with certainty.
The parametric specification for this approach is a generalization of (1.7). First,
we will include leads for the interaction between treatment intensity and the post-
tax reform dummy. To estimate the lead coefficients, we can employ pre-tax reform
data for the outcome variables and then use the estimation results to test for dif-
ferences in pre-tax reform outcome dynamics across municipalities with different
property tax rate increases. We can also include lags for the treatment intensity
variable to test how persistent the effect of higher property taxes is on the outcome
variables of interest. Equation (1.15) shows the empirical implementation of the
event-study analysis for the 2012 property tax reform.

̃ ̃ 𝑓
𝑦𝑚,𝑡 = FE𝑚 + FE𝑡 + ∑ 𝛽𝑡𝑦,ℎ 1{𝑡 = 𝑡}̃ × Δ𝜏𝑚,2012
ℎ + ∑ 𝛽𝑡𝑦,𝑓 1{𝑡 = 𝑡}̃ × Δ𝜏𝑚,2012 + 𝜖𝑚,𝑡
̃
𝑡≠2011 ̃
𝑡≠2011
(1.15)
− −
We are mainly interested in the lead coefficients 𝛽𝑡𝑦,ℎ and 𝛽𝑡𝑦,𝑓 , for 𝑡− = {2008, 2009, 2010}.
Notice there is no lead coefficient for 2011, given that this is our base year. This
also means that 𝛽2012
𝑦,𝑖 for 𝑖 = {ℎ, 𝑓 } capture the immediate effect of the 2012 tax re-
+ +
form for the average Italian municipality. Finally, lag coefficients 𝛽𝑡𝑦,ℎ and 𝛽𝑡𝑦,𝑓 for
𝑡+ = {2013, 2014} capture the effects of the increase in property taxes 𝑇 + years after
the tax reform, for 𝑇 + = 𝑡+ − 2012. Similarly to the baseline specification in (1.7),
we also control for municipality and year fixed effects in (1.15).
 34

We use the lead coefficients to test for evidence against parallel trends. In par-
ticular, for 𝑖 = {ℎ, 𝑓 } we perform the following hypothesis test:

𝐻𝑜 ∶ 𝛽2008
𝑦,𝑖 = 𝛽2009
𝑦,𝑖 = 𝛽2010
𝑦,𝑖 =0
−
𝐻𝑎 ∶ 𝛽𝑡𝑦,𝑖 ≠ 0 , for some 𝑡− = {2008, 2009, 2010}

Rejecting the compound null 𝐻𝑜 is interpreted as evidence against parallel trends

because implies that outcome trends during the pre-tax reform period are statisti-
cally different across municipalities with different property tax increases.

EVENT-STUDY APPROACH: RESULTS

Figure 1.3 present the results for the increase in residential property taxes, while
Figure 1.4 show the estimated dynamic coefficients for the increase in property
taxes for commercial real estate. In each figure, the blue dotted lines and red dashed
̃
lines capture the points estimates of 𝛽𝑡𝑦,𝑖 and 95% confidence intervals, respectively.
For each plot, the point estimates to the left the vertical black dashed line for 2011
capture the lag coefficients. On the other hand, the contemporaneous and lag co-
efficients are to the right of the vertical black dashed line.
The first result we discuss is related to the evidence of pre-trends across Italian
municipalities. As we can see, the lag coefficients for all outcome variables of in-
terest are not statistically different from zero at the 5% confidence level. Therefore,
we do not reject the null hypothesis 𝐻𝑜 of no pre-trend differences across munici-
palities with a different tax rate increase for residential and commercial real estate
properties. In other words, we find no evidence of different growth dynamics in
non-tradable employment, consumption expenditure, and real estate prices across
Italian municipalities before the 2012 tax reform.
 35

The absence of systematic differences in trends for all the outcome variables
across municipalities with different changes in property taxes is consistent with the
findings of Alesina and Paradisi (2017). In this paper, the authors use Italy’s 2012
property tax reform to study the relationship between political budget cycles and
local governments’ tax policy choices. Their main finding is that changes in prop-
erty taxes across municipalities are primarily explained by the staggered timing of
local elections, with the latter defined in the 1940s. The authors argue that election
dates for Italian municipalities are as good as a random assignment. Then, it seems
logical that we do not find any evidence of pre-trends in outcomes across munic-
ipalities, given that property tax changes in Italian municipalities during 2012 are
mainly explained by political factors unrelated to business cycle fluctuations.
Another interesting result from our event-study analysis is related to the lack
of statistical significance at the 5% for the lead coefficients on all outcomes of inter-
est. The last result implies that we only observe a response on outcome variables
during the year of the property tax increase with no significant effect in the future.
To explain this result, we rely on two factors: (i) the expectations of municipali-
ties about additional property tax changes in the future and (ii) the shortly-lived
Experimental-IMU tax system.
First, as shown in section 1.5.2, Italian municipalities change property taxes only
after a tax reform imposed by the central government. If private agents internalize
this, only the probability of new tax reforms should be relevant for expectations of
new tax changes in the future. We employ the 2012 Survey on Household Income
and Wealth for Italy to provide evidence about this. In particular, households were
asked if they think the new property tax system will be changed within the next five
years. Approximately 67% of households expected the new property tax system to
last more than five years. Therefore, we can conclude that municipalities regarded
 36

the tax rate change as relatively permanent, which explains the significant response
in outcome variables observed during the year of the property tax reform.
Second, as we describe in section 1.2.2, the unpopularity of the 2012 property
tax reform forced the central government to cancel the Experimental-IMU tax after
just one year. The latter explains why the increase in property taxes during the 2012
tax reform had no statistically significant effect on the outcome variables after 2012.

1.5.4 COVARIATE BALANCE ACROSS ITALIAN MUNICIPALITIES

Quasi-experimental research designs often examine the similarities between treat-

ment and control groups in pre-treatment observable characteristics to reassure
that both groups were comparable prior to treatment exposure (Imbens and Ru-
bin, 2015). This issue is also relevant in any diff-in-diff design. For our partic-
ular case, we need to check for potential unbalances in observable characteristics
across municipalities with different property tax rate increases. Moreover, examin-
ing covariate balance across municipalities shows how property tax change choices
during the 2012 tax reform are correlated with other observable municipality-level
features.
However, a simple covariate balance table is not sufficiently informative for our
research design. What matters for our empirical strategy is that differences across
treatment intensity groups are constant over time and that changes in treatment
intensity exposure across municipalities are not associated with changes in the dis-
tribution of other observable characteristics at the municipality level. We follow the
covariate-balance regression approach described in Wing et al. (2018) to examine
this.
 FIGURE 1.3: Event Study: 2012 Change in the Tax rate for Residential Properties
(a) Non-Tradable Employment (c) Consumption

0.22 1.00

0.11 0.50

0.00 0.00

-0.11 -0.50

-0.22 -1.00

2008 2009 2010 2011 2012 2013 2014 2008 2009 2010 2011 2012 2013 2014
(e) Residential Prices (g) Commercial RE Prices

0.04 0.14

0.02 0.07

Growth rate relative to 2011 (%)

0.00 0.00

-0.02 -0.07

-0.04 -0.14

2008 2009 2010 2011 2012 2013 2014 2008 2009 2010 2011 2012 2013 2014
Note: The figure depicts the impact of the increase in residential property taxes on the main outcome variables. The solid blue lines plot the estimated
coefficients, and the red dashed line represents the 95 percent confidence intervals. The plots on the left are for the increase in the residential property
37

tax rate, while the figures on the left are for the increase in the commercial real estate tax rate. Each point, shows the average difference in growth rate
of 𝑦 in year 𝑡 for municipalities treated with Δ𝜏 ℎ relative to the 2011 growth rate, where 𝑡 = 2008, 2009, 2010, 2012, 2013, 2014, 2015,
𝑦 = {𝑙, 𝑐, 𝑝ℎ , 𝑝𝑓 }. The sample includes 6,246 municipalities over the 2008- 2014 period. Standard errors clustered at the Local Labor Market level.
 FIGURE 1.4: Event Study: 2012 Change in the Tax rate for Commercial Real Estate Properties
(b) Non-Tradable Employment (d) Consumption

0.16 1.00

0.08 0.50

0.00 0.00

-0.08 -0.50

-0.16 -1.00

2008 2009 2010 2011 2012 2013 2014 2008 2009 2010 2011 2012 2013 2014
(f) Residential Prices (h) Commercial RE Prices

0.04 0.08

0.02 0.04

Growth rate relative to 2011 (%)

0.00 0.00

-0.02 -0.04

-0.04 -0.08

2008 2009 2010 2011 2012 2013 2014 2008 2009 2010 2011 2012 2013 2014
Note: The figure depicts the impact of the increase in commercial real estate property taxes on the main outcome variables. The solid blue lines plot the
estimated coefficients, and the red dashed line represents the 95 percent confidence intervals. The plots on the left are for the increase in the residential
38

property tax rate, while the figures on the left are for the increase in the commercial real estate tax rate. Each point, shows the average difference in
growth rate of 𝑦 in year 𝑡 for municipalities treated with Δ𝜏 𝑓 relative to the 2011 growth rate, where 𝑡 = 2008, 2009, 2010, 2012, 2013, 2014, 2015,
𝑦 = {𝑙, 𝑐, 𝑝ℎ , 𝑝𝑓 }. The sample includes 6,246 municipalities over the 2008- 2014 period. Standard errors clustered at Local Labor Market-level.
 39

COVARIATE-BALANCE REGRESSIONS: IMPLEMENTATION

Let 𝑥 be a variable we want to use to test for covariate balance. Assuming that
data for that covariate is available at the municipality level, we can use 𝑥𝑚,𝑡 as a
left-hand side variable to estimate a specification similar to (1.7). In particular, to
𝑖 𝑓
check if covariate 𝑥 is balanced across treatment intensities Δ𝜏𝑚,2012 and Δ𝜏𝑚,2012
we estimate model (1.17).

ℎ 𝑓
𝑥𝑚,𝑡 = FE𝑚 + FE𝑡 + 𝜃𝑥,ℎ Δ𝜏𝑚,2012 + 𝜃𝑥,𝑓 Δ𝜏𝑚,2012 + 𝜇𝑚,𝑡 (1.17)

The coefficients 𝜃𝑥,ℎ and 𝜃𝑥,𝑓 capture compositional changes in 𝑥 due to choices in
property tax rate changes during the 2012 tax reform. In other words, these coeffi-
cients measure the variation in 𝑥 explained by differences in residential and com-
mercial tax rate changes across municipalities. Using these coefficients, we test for
the composite null 𝐻𝑜 ∶ 𝜃𝑥,ℎ = 𝜃𝑥,𝑓 = 0 of no compositional changes across munic-
ipalities with different treatment intensities. Rejecting the null will be interpreted
as evidence of imbalances in 𝑥 across municipalities during the 2012 tax reform.
We gather data on additional observable characteristics at the municipality level
for 2008-2014. We categorize the covariates used in this analysis into four groups.
First, we include variables related to local economic and financial conditions, specif-
𝑝𝑐 𝑝𝑐
ically: log of income per capita (Income ), growth rate of income per capita (ΔIncome ),
and log of loans (Loan) and deposits (Depos). The second group includes vari-
In
ables related to migration patterns, specifically in-migration (Mig ) and out-migration
Out
(Mig ) rates. The third group contains employment shares for manufactures
(𝑠ℎ𝐿man ), retail (𝑠ℎ𝐿ret ) and construction (𝑠ℎ𝐿cons ) computed using two-digit NACE
industry codes. Finally, we include observable characteristics related to financial
 40

conditions of local governments, in particular: per capita real growth rate for cur-
rent revenues (Δ𝑇C ) and current expenditures (Δ𝐺C ), capital investment rate (𝐺K /𝐺C ),
deficit-to-revenues ratio (Deficit/𝑇C ) and the debt-to-revenues ratio (𝐵/𝑇 C ).

COVARIATE BALANCE REGRESSIONS: RESULTS

Table 1.3 and Table 1.5 present the results for first and fourth groups of municipal
level covariates, respectively. On the other hand, Table 1.4 shows the estimation
results for covariates in the second and third groups. For each table, each column’s
first and second row presents the estimated coefficients for 𝜃𝑥,ℎ and 𝜃𝑥,𝑓 . While the
third row on each table reports the p-value for the compositional changes hypothesis
test described earlier.
First, notice that Tables 1.3 and 1.4 report no significant correlation between the
increase in property tax rates and the covariates in the first and second group. The
latter explains why we cannot reject the null of no compositional changes across
municipalities during the 2012 tax reform for these groups of covariates. In other
words, our results show that municipalities with different property tax rate in-
creases are similar in terms of economic and financial local conditions, migration
patterns, and industry employment shares.
We arrive at a different conclusion when testing for balance with variables re-
lated to the financial situation of local governments. Table 1.5 shows that munic-
ipalities with a higher tax rate increase for residential and commercial real estate
properties have higher growth of current revenues, lower primary deficit, and are
less indebted. This result translates into a rejection of the null for no compositional
changes. That is to say; we find evidence of imbalances in per capita revenues
growth, deficit-to-revenues ratio, and debt-to-revenues ratio across municipalities
during the 2012 tax reform.
 41

On the other hand, Table 1.5 also shows a significant positive correlation be-
tween the increase in the tax rate for residential properties and the per capita ex-
penditure growth. However, there are no significant differences in per capita ex-
penditure growth across municipalities with different commercial real estate tax
rate changes. As a result, we do not reject the null of no compositional effects across
municipalities for per capita expenditure growth. However, this evidence is not as
strong as for other covariates.
Lastly, we also find no evidence of imbalances in local public investment rates
across municipalities during the 2012 property tax reform.

TABLE 1.3: Covariate Balance: Local Economic and Financial Conditions

Income Growth Income Loans Deposits

(𝜃ΔInc𝑝𝑐 ,𝑖 ) (𝜃Inc𝑝𝑐 ,𝑖 ) (𝜃Loans,𝑖 ) (𝜃Depos,𝑖 )
ℎ
Δ𝜏𝑚,2012 -0.001 -0.002 -0.009 0.028
(0.007) (0.004) (0.025) (0.022)
𝑓
Δ𝜏𝑚,2012 -0.008 -0.002 -0.001 0.004
(0.005) (0.004) (0.016) (0.017)

𝐻𝑜 ∶ 𝜃𝑥,ℎ = 𝜃𝑥,𝑓 = 0 (𝑝-val) 0.25 0.77 0.93 0.42

𝑁obs 43,540 43,540 14,185 14,185

𝑁mun 6,220 6,220 2,089 2,089
𝑅̄ 2 0.10 0.99 0.99 0.99

Notes: The table shows the covariate balance test for observables related to local economic and financial con-
ditions. Income per capita is the municipality’s total taxable income divided by the end of the year population
expressed in 2010. Loans and deposits measure the stock of loans and deposits in bank branches within the
municipality. Standard errors in parentheses are clustered at the local labor market level, ∗ , ∗∗ , ∗∗∗ indicate sig-
nificance at the 10% 5% and 1% respectively
 42

TABLE 1.4: Covariate Balance: Migration Patterns and Supply Side Controls

Migration Rate Employment Share

In Out Manuf. Const. Retail
(𝜃Migin ,𝑖 ) (𝜃Migout ,𝑖 ) (𝜃sh 𝐿man ,𝑖 ) (𝜃sh 𝐿cons ,𝑖 ) (𝜃sh 𝐿ret ,𝑖 )
ℎ
Δ𝜏𝑚,2012 0.002 0.001 0.004 0.010** 0.000
(0.002) (0.002) (0.005) (0.005) (0.005)
𝑓
Δ𝜏𝑚,2012 0.001 -0.001 -0.004 0.002 -0.001
(0.001) (0.001) (0.004) (0.004) (0.003)

𝐻𝑜 ∶ 𝜃𝑥,ℎ = 𝜃𝑥,𝑓 = 0 (𝑝-val) 0.39 0.73 0.51 0.10 0.94

𝑁obs 43,540 43,540 43,540 43,540 43,540

𝑁mun 6,220 6,220 6,220 6,220 6,220
𝑅̄ 2 0.40 0.62 0.96 0.90 0.90

Notes: The table shows the covariate balance test for observables related to migration patterns and supply side conditions.
In(out) migration rates is the share of persons moving in(out) to the municipality at the end of the year. Employment
share is the sector’s employment ratio to total municipal employment. Manufacturing, Construction, and Wholesale &
Retail sector comprise total establishment employment classified in sections C, F, and G according to Nace 2. Rev industry
codes, respectively. Standard errors in parentheses are clustered at the local labor market level, ∗ , ∗∗ , ∗∗∗ indicate significance
at the 10% 5% and 1% respectively
 43

TABLE 1.5: Covariate Balance: Financial Situation of Local Governments

Rev. Expend. Investment Deficit-to-𝑇 𝑐 Debt-to-𝑇 𝑐

Growth Growth Rate Ratio Ratio
(𝜃Δ𝑇 𝑐 ,𝑖 ) (𝜃Δ𝐺𝑐 ,𝑖 ) (𝜃𝐺𝐾 /𝐺𝑐 ,𝑖 ) (𝜃Deficit/𝑇 𝑐 ,𝑖 ) (𝜃𝐵/𝑇 𝑐 ,𝑖 )
ℎ
Δ𝜏𝑚,2012 0.072** -0.065** -0.025 -0.137*** -0.122*
(0.032) (0.029) (0.057) (0.024) (0.069)
𝑓
Δ𝜏𝑚,2012 0.20*** -0.006 -0.052 -0.175*** -0.143***
(0.024) (0.025) (0.046) (0.015) (0.047)

𝐻𝑜 ∶ 𝜃𝑥,ℎ = 𝜃𝑥,𝑓 = 0 (𝑝-val) 0.00 0.10 0.46 0.00 0.01

𝑁obs 43,519 43,519 43,540 43,519 43,519

𝑁mun 10,158 10,158 6,220 10,158 10,158
𝑅̄ 2 0.92 0.93 0.53 0.27 0.59

Notes: The table shows the covariate balance test for variables related to migration patterns and supply side conditions. Current
revenues (𝑇 𝑐 ) include income from local taxes and tariffs collected by the municipal government. Current expenditures (𝐺𝑐
) represent all spending done by a municipality government for regular operations during the fiscal year. Per capita revenues
and expenditures are expressed in 2010 euros. Local government investment rate is the ratio of municipal government capital
expenditure to total current expenditure. Deficit is the difference between current expenditure and current revenues. Finally,
debt is the book value of the stock of local government debt at the end of the fiscal year. Standard errors in parentheses are
clustered at the local labor market level, ∗ , ∗∗ , ∗∗∗ indicate significance at the 10% 5% and 1% respectively

1.5.5 ROBUSTNESS AND OTHER POTENTIAL THREATS

CONTROLLING FOR MUNICIPAL LEVEL COVARIATES

The results on covariate-balance reported in 1.5.4 provide us with a guide test the
robustness of the baseline results presented in Table 1.2. In particular, we will
change our baseline specification (1.7) by controlling explicitly for 𝑥𝑚,𝑡 . Let 𝑋𝑚,𝑡
be the vector of all relevant covariates for municipality 𝑚 at year 𝑡, then (1.18) will
 44

be our new specification.

ℎ × 1{𝑡 = 2012} + 𝛽 𝑓
𝑦𝑚,𝑡 = FE𝑚 + FE𝑡 + 𝑋 𝑚,𝑡−1 𝛤 + 𝛽𝑦,ℎ Δ𝜏𝑚,𝑡 𝑦,𝑓 Δ𝜏𝑚,𝑡 × 1{𝑡 = 2012} + 𝜖𝑚,𝑡

(1.18)
To avoid the potential endogeneity between X𝑚,𝑡 and 𝑦𝑚,𝑡 we use the value of the
covariates lagged one year. The vector 𝑋𝑚,𝑡−1 includes the per capita real growth
rate for current revenues, the deficit-to-revenues ratio, and the debt-to-revenues
ratio. The reason to include only these covariates as additional controls is the evi-
dence we found for imbalances across municipalities during the 2012 tax reform in
Italy. We also include variables that should capture the effect of other local policy
changes happening during 2008-2014.22

TESTING ALTERNATIVE HYPOTHESIS

We next consider if the negative effect of the 2012 Italian property tax reform on
our outcomes of interest can be explained by some spurious correlation between the
change in property tax rates across municipalities and some unobserved municipality-
specific shock.
In particular, we examine how local credit supply, productivity, and uncer-
tainty shocks can explain the reduction observed in non-tradable employment, con-
sumption expenditure, and real estate prices across municipalities after the increase
in property taxes in 2012.
On the one hand, the idea that local credit supply is behind our baseline results
is based on the possibility that municipalities choosing a higher property tax in-
crease also faced a significant decline in credit supply. The latter could produce a
drop in consumption expenditure on non-housing and housing goods on the side
22
See appendix B for more details.
 45

of households and force firms to cut employment and reduce or cancel any invest-
ment in fixed assets altogether. Moreover, the evidence provided by Arellano et al.
(2019) shows that banks’ exposure to default risk during the 2012 Italian debt crisis
produced a heterogeneous drop in the supply of loans across regional credit mar-
kets and produced higher output declines for firms in regions highly exposed to
default risk.
On the other hand, the uncertainty hypothesis argues that the perception of
higher uncertainty induced by policy changes or severe economic contractions could
result in declines in economic activity. On the side of firms, Bloom (2009) shows
that uncertainty produces drops in hiring and temporal pauses in investment. While
on the households side Christelis et al. (2020) finds that changes in risk perception
can produce substantial declines in consumption expenditure due to a precaution-
ary saving motive. Then, a positive correlation between the 2012 increase in prop-
erty taxes and municipality-specific uncertainty changes generated by past adjust-
ments in local tax policy can explain the negative effect of the 2012 property tax
reform on our outcomes of interest.
The significant decline in aggregate TFP for Italy after the 2008 financial cri-
sis23 (Sgherri and Morsy, 2010) could impacted municipalities differently. For this
reason, we examine the potential confounding effect of local productivity shocks,
which could explain our baseline results if correlated with the 2012 property tax
increase across municipalities.
To test for the three hypotheses mentioned earlier, we will examine the stability
of our baseline regression results under an alternative specification which includes
a proxy measure for the unobserved credit supply, productivity, and uncertainty
23
Pellegrino and Zingales (2017) argues that stagnation problems with aggregate productivity in
Italy can be even as far back as the mid-1990s.
 46

shock and the interaction of this proxy with our treatment intensities.
𝑗
̂ be a proxy measure for the 𝑗-shock, where 𝑗={ credit supply,
In particular, let 𝜂𝑚,𝑡
uncertainty, productivity}. Then (1.19) will be our new parametric specification.

ℎ 𝑓 𝑗
𝑦𝑚,𝑡 = FE𝑚 + FE𝑡 + 𝛽𝑦,ℎ Δ𝜏𝑚,2012 + 𝛽𝑦,𝑓 Δ𝜏𝑚,2012 + 𝛿𝑗,0 𝜂𝑚,𝑡
̂ (1.19)
𝑗 ℎ 𝑗 𝑓
+ 𝛿𝑗,ℎ 𝜂𝑚,𝑡
̂ × Δ𝜏𝑚,2012 + 𝛿𝑗,𝑓 𝜂𝑚,𝑡
̂ × Δ𝜏𝑚,2012 + 𝜖𝑚,𝑡

Where 𝛿𝑗,0 is the coefficient for the 𝑗-shocks while 𝛿𝑗,ℎ and 𝛿𝑗,𝑓 are the coefficients
associated with the interaction between the proxy measure for the shock and the
change in residential and commercial real estate tax rates, respectively. The details
𝑗
̂ can be found in appendix B.
on the construction for 𝜂𝑚,𝑡

SPILLOVERS EFFECTS OF THE 2012 TAX REFORM

A final issue we will discuss is related to spillover effects problems. The choice of an
increase in property taxes during the 2012 tax reform not only affects the outcomes
related to that municipality but could potentially impact neighboring municipal-
ities through spillover effects. This is especially true for municipalities regarded
as metropolitan areas. In this case, the labor changes and migration patterns pro-
duced by a large municipality will spill over to its smaller neighbors. Moreover,
spillovers can also affect the choice of the tax rate change due to some leader-
follower strategic behavior of local governments.
We will include covariates that can partially capture spillover effects to examine
the importance of spillover effects on our baseline results. This approach does not
eliminate all the bias from spillover effects but at least provides some indicative
evidence of its importance for the initial results we obtained.
Our approach will exploit the classification of municipalities in Local Labor
 47

Markets (LLM) provided by the Italian National Institute of Statistics. An LLM

is a group of neighboring municipalities where the bulk of the labor force lives and
works and where establishments can find the main part of the labor force necessary
to occupy the offered jobs. Using the classification of municipalities in each LLM,
we can capture any unobserved LLM-time varying trend by controlling for LLM ×
year fixed effects. Equation (1.20) shows the latter.

ℎ × 1{𝑡 = 2012} + 𝛽 𝑓
𝑦𝑚,𝑡 = FE𝑚 + FE𝑡 + 𝛿𝐿𝐿𝑀(𝑚),𝑡 + 𝛽𝑦,ℎ Δ𝜏𝑚,𝑡 𝑦,𝑓 Δ𝜏𝑚,𝑡 × 1{𝑡 = 2012} + 𝜖𝑚,𝑡

(1.20)
Where 𝛿𝐿𝐿𝑀(𝑚),𝑡 represents the LLM × year fixed effects for municipality 𝑚. Includ-
ing 𝛿𝐿𝐿𝑀(𝑚),𝑡 in our diff-in-diff strategy has the additional benefit of capturing any
municipality-level trend component correlated with unobserved LLM trends.

ALTERNATIVE SPECIFICATIONS AND ROBUSTNESS: RESULTS

The estimation results for specifications (1.18), (1.19) and (1.20) is depicted in Fig-
ure 1.5. Each plot in Figure 1.5 shows the estimation results for 𝛽𝑦,ℎ and 𝛽𝑦,𝑓 on all
outcome variable of interest. The dashed vertical lines represent the 95% confidence
intervals, while the hollowed diamond at the center of the interval corresponds to
the point estimates.
In each of the four plots in Figure 1.5, a light gray line divides the x-axis into
two halves. The left part of the x-axis reports the point estimates and confidence
intervals for an increase in residential property taxes (i.e., 𝛽𝑦,ℎ ). At the same time,
the right half shows the estimation results for an increase in the tax rate for com-
mercial real estate properties (i.e., 𝛽𝑦,𝑓 ). The baseline results presented in Table 1.2
are depicted in blue. The results for models (1.18) and (1.20) are depicted with
red and orange, respectively. Finally, the point estimates and confidence intervals
 48

in purple, green, and gray represent the estimates of (1.19) when controlling for
productivity, uncertainty, and credit supply shocks, respectively.
The main conclusion we draw from Figure 1.5 is that the baseline results for
the response of non-tradable employment, consumption growth, and real estate
prices to the 2012 tax reform are strongly robust in magnitude and significance. In
most cases, our point estimates remain negative and relatively close to the base-
line results. However, in some cases, we observe a loss in precision, especially for
the response of residential and commercial real estate prices when controlling for
credit supply, productivity, and uncertainty shocks. The loss in efficiency, in this
case, is because we include interaction between the proxy shocks and the treatment
intensity variables. Lastly, we need to mention that the stability of our estimated
coefficients is only violated in the response of consumption expenditure to change
in commercial real estate taxes when controlling for credit supply shocks.
 FIGURE 1.5: Robustness of Baseline Estimates

Non-Trad. Employment Consumption

0.04

0.02 0.90
0.71
0 0.60

-0.02
0.30
-0.04 -0.04 -0.04 -0.04
-0.05 0
-0.05 -0.06 -0.08
-0.06 -0.18 -0.18 -0.16
-0.07 -0.06
-0.30 -0.33
-0.08 -0.41
-0.08 -0.08
-0.09 -0.52 -0.52
-0.09 -0.09 -0.60 -0.65
-0.10
-0.78
-0.9
-0.12

-0.14 -1.2

-0.16 -1.5

Residential Prices CRE Price

0.04

0.01 0.02

0 -0.00
0 -0.00
-0.00
-0.01 -0.01
-0.01
-0.01 -0.01
-0.01 -0.02 -0.02
-0.01 -0.02
-0.02 -0.02 -0.02
-0.02 -0.02
-0.02
-0.02 -0.03 -0.03
-0.02 -0.02
-0.04
-0.03 -0.03
-0.04
-0.03

-0.06
-0.04

-0.05 -0.08
h f h f
∆τ ∆τ ∆τ ∆τ

Baseline Add. Covariates LLM FE × Year FE Prod. Shocks Uncert. Shocks Credit Sup. Shocks
49

Note: The figure depicts the estimation results for specifications (1.18), (1.19) and (1.20) on each outcome variable of interest. The holo The sample includes
6,220 municipalities over the 2008- 2014 period. The dashed vertical lines represent the 95% confidence intervals with the point estimates depicted with a
hollowed diamond at the center of the interval. On each graph, the y-axis is expressed in percentage points. The left half x-axis reports the estimated coefficients
for 𝛽𝑦,ℎ . The right half x-axis presents the results for 𝛽𝑦,𝑓 . The y-axis on each graph is expressed in percentage points. Standard errors clustered at the Local
Labor Market level.
 50

1.6 MODEL

This section presents the quantitative model used to quantify the housing wealth
and firm collateral channel. The economy only lasts one period and comprises
households, firms producing goods, and a construction sector supplying houses
and commercial real estate. Firms produce non-tradable differentiated goods de-
noted by 𝑗, and each variety is within the unit interval 𝑗 ∈ [0, 1] using labor and
commercial real estate. Households supply labor and consume an aggregate basket
of differentiated goods. Additionally, households have preferences over housing
services.
Since residential properties provide services to households and commercial real
estate is used as production input by firms, there is no substitution between both
types of real estate assets; each will have a different market with a different equi-
librium price. The construction sector provides both real estate assets to firms and
households.
To finance expenditures, households and firms can get loans from the foreign
financial market. Loans are paid within the period and charged with a zero interest
rate. Borrowers need collateral to get a loan in the form of real estate (i.e., houses
or commercial real estate). Therefore, the maximum loan size is defined by the
collateral value that households and firms own.
Lastly, the economy has a government collecting revenues from property taxes
paid by households and firms. Property taxes are different for house owners and
commercial real estate owners. In particular, let 𝜏 ℎ and 𝜏 𝑓 represent the tax rates
charged to residential and commercial real estate, respectively.
 51

1.6.1 HOUSEHOLDS PROBLEM

Households are identical in all aspects. In a first stage, households decide on their
supply of labor(𝐿) and expenditure on housing (𝐻 ℎ ) and non-housing goods (𝐶).
The first stage problem is presented in (1.21).

𝜒 1
maximize (𝐶)𝛽 (𝐻 ℎ )1−𝛽 − 1
𝐿1+ 𝜈
{𝐶,𝐿𝑠 ,𝐻 ℎ } 1+ 𝜈

subject to 𝐶 + 𝑃ℎ 𝐻 ℎ (1 + 𝜏 ℎ ) = 𝑊 𝐿 + Π (1.21a)

𝐶 ≤ 𝜙ℎ 𝑃ℎ 𝐻 ℎ (𝜇ℎ ) (1.21b)

Preferences for housing goods, non-housing goods, and labor are separable. In
particular, the labor supply will respond to household wealth changes. The param-
eter 𝜈 represents the Frisch elasticity. Preferences over housing and non-housing
goods are Cobb-Douglas; in a frictionless economy, the share of total expenditure
allocated to housing goods is 𝛽. Notice that 𝐻 ℎ is a continuous variable. Moreover,
households do not have an initial endowment of homes.
Household’s budget constraint is presented in (1.21a). Income sources for house-
holds are twofold: (i) labor income 𝑊 𝐿, (ii) profits from firms Π. Total income is
allocated on consumption 𝐶, housing 𝑃ℎ 𝐻 ℎ and property taxes 𝑃ℎ 𝐻 ℎ 𝜏 ℎ . The con-
sumption good 𝐶 will be the numeraire for the economy (i.e 𝑃𝑐 = 1).
Equation (1.21b) is the household’s borrowing constraint with 𝜇ℎ as its the
shadow value. Households can fund consumption using loans, with the latter con-
strained by a fraction of the value of residential assets held by the household. This
fraction is determined by the loan-to-value ratio 𝜙ℎ .
In a second stage, households decide the demand for each variety, taking the ex-
penditure on non-housing goods in the first stage as given. A CES aggregator com-
 52

bines all varieties into the aggregate consumption good. The second stage problem
is outlined in (1.22).

1
minimize
𝑐
∫ 𝑝𝑗 𝑐𝑗 𝑑𝑗 (1.22a)
𝑗
0
1
1 1− 1
𝜖
1− 1𝜖
subject to 𝐶 ≥ ⎛
⎜
⎜∫ 𝑐𝑗 𝑑𝑗⎞
⎟
⎟ (1.22b)
⎝0 ⎠

From the second stage problem, we can derive the inverse demand function for
variety 𝑗, which is presented in (1.23). The parameter 𝜖 captures the price elasticity
of demand for variety 𝑗.
1
𝐶 𝜖
𝑝 (𝑐𝑗 ) = ⎛
⎜ ⎞ ⎟ (1.23)
⎝ 𝑐𝑗 ⎠

1.6.2 FIRMS PROBLEM

All varieties 𝑗 ∈ [0, 1] are produced by monopolistically competitive firms. Each

firm chooses labor 𝐿 and commercial real estate 𝐻 𝑓 to maximize profits. Each firm
has market power to set prices above the marginal cost.
(1.24) outlines the profit maximization problem for a firm producing variety 𝑗.

𝑓
maximize 𝑝𝑗 𝑐𝑗 − 𝑊 𝐿𝑗 − 𝑃𝑓 𝐻𝑗 (1 + 𝜏 𝑓 ) (1.24a)
𝑓
{𝐿,𝐻 }
1
𝐶 𝜖
subject to 𝑝 (𝑐𝑗 ) = ⎛
⎜ ⎞ ⎟ (1.24b)
⎝ 𝑐𝑗 ⎠
𝛼 1−𝛼
𝑓 𝑓
𝑐𝑗 = 𝐹(𝐿𝑗 , 𝐻𝑗 ) = (𝐿𝑗 ) (𝐻𝑗 ) (1.24c)
𝑓
𝑊 𝐿𝑗 ≤ 𝜙𝑓 𝑃𝑓 𝐻𝑗 (𝜇𝑓 ) (1.24d)

Equation (1.23) represents is the inverse demand function variety 𝑗. Production

 53

technology in (1.24c) is Cobb-Douglass. We assume that firms have working cap-

ital requirements for labor financed with loans from the foreign financial market.
(1.24d) represent the borrowing constraints for firms, similar to borrowing con-
straints for households. The main difference is that firms need to pledge a fraction
of the value of commercial real estate assets they own, with this fraction defined by
a specific loan-to-value ratio 𝜙𝑓 . The shadow value for firms’ collateral constraint
is denoted by 𝜇𝑓 .
In addition to labor costs and investment in commercial real estate, firms need
to pay property taxes according to the tax rate 𝜏 𝑓 . The amount of property taxes
𝑓
paid by firm will be 𝑃𝑓 𝐻𝑗 𝜏 𝑓 .

1.6.3 RESIDENTIAL AND COMMERCIAL REAL ESTATE SUPPLY

The final private agent in the economy is the construction sector. To simplify our
analysis, we assume an implicit optimization problem for firms producing houses
and commercial real estate results in the price-increasing supply functions and ,
respectively.

𝐻 ℎ (𝑃ℎ ) = (𝑃ℎ )𝜎ℎ (1.25a)

𝜎𝑓
𝐻 𝑓 (𝑃𝑓 ) = (𝑃𝑓 ) (1.25b)

The parameters 𝜎ℎ and 𝜎𝑓 in and captures the price elasticity of supply for houses
and commercial real estate, respectively. The implicit assumption behind equa-
tions (1.25a) and (1.25b) is that the construction firms use a specific land as input
for production. Recall that 𝐻 𝑓 is an input production, while 𝐻 ℎ provides housing
services. Neither households nor firms can substitute one real estate asset with
the other, and therefore, 𝐻 𝑓 and 𝐻 ℎ are sold in different markets given their two
 54

distinct purposes.

1.6.4 EQUILIBRIUM, PROPERTY TAXES AND FINANCIAL FRICTIONS

Definition 1 (Competitive Equilibrium): A competitive equilibrium in this econ-

omy is defined by a set of prices {𝑊, 𝑃ℎ , 𝑃𝑓 , (𝑝𝑗 ) }, allocations {𝐿, 𝐻 ℎ , 𝐻 𝑓 , 𝐶, (𝑐𝑗 ) },
𝑗∈[0 ,1] 𝑗∈[0 ,1]
shadow values {𝜇ℎ , 𝜇𝑓 }, and property tax rates {𝜏 ℎ , 𝜏 𝑓 } such that:

1. Given equilibrium prices {𝑊, 𝑃ℎ , 𝑃𝑓 , (𝑝𝑗 ) } and property tax rates {𝜏 ℎ , 𝜏 𝑓 }

𝑗∈[0 ,1]

(a) Household choices for 𝐿, 𝐻 ℎ and 𝐶 solve (1.21) with 𝜇ℎ ≥ 0 and (𝑐𝑗 )
𝑗∈[0 ,1]
solve (1.22).

(b) Firm choices for 𝐿 and 𝐻 𝑓 solve (1.24) with 𝜇𝑓 ≥ 0.

(c) Supply of houses and commercial real estate is consistent with (1.25b)
and (1.25a), respectively.

2. Given the allocation {𝐿, 𝐻 ℎ , 𝐻 𝑓 } and property tax rates {𝜏 ℎ , 𝜏 𝑓 }

(a) {𝑊, 𝑃𝑓 , 𝑃ℎ } clear the markets for labor, houses and commercial real es-
tate, respectively.

Proposition 2 (Binding Borrowing Constraints): Let {𝑊, 𝑃ℎ , 𝑃𝑓 , } and {𝐿, 𝐻 ℎ , 𝐻 𝑓 , 𝐶, }

denote the equilibrium price and allocation vector. Then, the household’s bor-
rowing constraint binds (i.e., 𝜇ℎ >0) if and only if:

𝐶
< 𝛽, (1.26)
𝑊𝐿 + Π
 55

𝑓
Furthermore, the firm’s collateral constraint binds (i.e., 𝜇𝑗 > 0) if and only if:

𝑊𝐿𝑗
<𝛼 (1.27)
𝑊𝐿𝑗 + 𝑃𝑓 𝐻 𝑓 (1 + 𝜏 𝑓 )

Proposition 2 describes the necessary and sufficient conditions for a competitive

equilibrium with binding financial constraints. We explain the intuition behind
this result next.
In the case of households, 𝛽 captures the expenditure share on non-housing
goods in an economy with no borrowing constraints for households. If 𝛽 is high
enough, households prefer to allocate most of their expenditure towards non-housing
goods, which can be done only by reducing housing expenditure. However, spend-
ing on non-housing goods requires a loan surpassing the household’s collateral
value. Therefore, the only option is to get the maximum loan consistent with their
housing wealth., which implies a drop in the share of expenditure on non-housing
goods relative to a case with no financial frictions.
For firms, the intuition is similar. In this case, 𝛼 captures the ratio of labor costs
to total costs in an economy with no financial constraints for firms. If 𝛼 is high
enough, the high marginal productivity for labor incentivizes firms to hire more
workers. However, this decision is inconsistent with the value of collateral owned
by firms. In the end, firms’ maximum loan available reduces the ratio of labor costs
to total costs compared to the one in a frictionless economy.

Proposition 3 (Log-linear Equilibrium): Let Θ = [𝛼, 𝛽, 𝜈, 𝜖, 𝜎𝑓 , 𝜎ℎ , 𝜙ℎ , 𝜙𝑓 ] be

the model’s vector of structural parameters. Then, we can characterize the com-
petitive equilibrium with binding financial constraints by the log-linear system
 56

of equations (1.28a)-(1.28d).

𝐴ℎ log (𝑃ℎ ) = 𝜅𝑃ℎ (Θ) + (1 + 𝜈) [log (𝑊) − log (1 + 𝜏 ℎ + 𝜙ℎ )] + log (1 + 𝜏 𝑓 + 𝜖𝜙𝑓 )

(1.28a)

𝐴𝑓 log (𝑃𝑓 ) = 𝜅𝑃𝑓 (Θ) + (1 + 𝜎ℎ ) log (𝑃ℎ ) − 𝛼(𝜖 − 1) log (𝑊) − 𝜖 log (1 + 𝜏 𝑓 + 𝜙𝑓 )
(1.28b)

log (𝐿) = log(𝜙𝑓 ) + (1 + 𝜎𝑓 ) log (𝑃𝑓 ) − log (𝑊) (1.28c)

log (𝐶) = log(𝜙ℎ ) + (1 + 𝜎ℎ ) log (𝑃ℎ ) (1.28d)

where 𝐴ℎ = 1 + 𝜎ℎ + (1 − 𝛽)𝜈 , 𝐴𝑓 = 1 + 𝜎𝑓 + (1 − 𝛼)(𝜖 − 1) and 𝜅𝑃ℎ (Θ), 𝜅𝑃𝑓 (Θ), 𝜅𝑊 (Θ)

are a functions of Θ.

PROPERTY TAX REFORM AND EQUILIBRIUM RESPONSE

We can use the result from proposition 3 to quantify the effect of a property tax
reform on the economy’s equilibrium. The property tax reform is characterized by
𝑓 𝑓 𝑓
a rise in tax rates from (𝜏0ℎ , 𝜏0 ) to (𝜏1 , 𝜏1 ); where 𝜏1𝑖 > 𝜏0𝑖 , for 𝑖 = {ℎ, 𝑓 }. Let 𝑌𝑡 =
𝑓 𝑓
{𝑃𝑡ℎ , 𝑃𝑡 , 𝑊𝑡 , 𝐿𝑡 , 𝐶𝑡 } denote the equilibrium under property tax scheme (𝜏𝑡ℎ , 𝜏𝑡 ) , 𝑡 =
0, 1 and 𝑦 = log (𝑌1 )−log (𝑌0 ) the equilibrium response to the property tax reform,
expressed in percentage change.

𝜏ℎ
Proposition 4 (Model’s Reduced Form Coefficients): For a given Θ, if 1+𝜙ℎ
,
𝜏𝑓 𝜏𝑓
1+𝜙𝑓
and 1+𝜖𝜙𝑓
are small enough the equilibrium response of 𝑌 = {𝑃ℎ , 𝑃𝑓 , 𝐿, 𝐶}
to an increase in property taxes equal to (Δ𝜏 ℎ , Δ𝜏 𝑓 ) can be characterized as fol-
 57

lows:
𝑦 = 𝛽𝑦,ℎ (Θ) Δ𝜏 ℎ + 𝛽𝑦,𝑓 (Θ) Δ𝜏 𝑓 (1.29)

where Δ𝜏 𝑖 = 𝜏1𝑖 − 𝜏0𝑖 is the percentage point (pp.) change in tax rate 𝑖 = {ℎ, 𝑓 }
and 𝛽𝑦,𝑖 (Θ) is the reduced form effect of a change in Δ𝜏 𝑖 on 𝑦.

There are two points worth mentioning regarding the results from Proposition 4.
First, the parametric specification (1.7) presented in the empirical section is equiva-
lent to the model-implied linear mapping in defined in (1.29). Second, the model-
implied reduced form effect 𝛽𝑦,𝑖 (Θ) are invariant to policy changes as they only
depend on the structural parameters in the model. Therefore, if our empirical strat-
egy correctly identifies the reduced-form effect of an increase in property taxes, we
can use the 2012 Italian tax reform estimation results to discipline the model.
Finally, notice that the model-implied reduced form effect 𝛽𝑦,𝑖 in (1.29) can be
interpreted as an elasticity.
𝑑𝑦
= 𝛽𝑦,𝑖 (Θ)
𝑑Δ𝜏 𝑖

where, 𝑑𝑦 and 𝑑Δ𝜏 𝑖 are the marginally changes for 𝑦 and Δ𝜏 𝑖 , respectively both
measured in percentage points. From now on, we use the terms reduced-form effect
of 𝑌 to changes in tax rate 𝑖 and 𝑖-tax rate-elasticity of 𝑌 interchangeably.

PROPERTY TAX CHANGES, HOUSING WEALTH AND FIRM COLLATERAL CHANNEL

As we will discuss later, the model-implied 𝑖-tax rate-elasticity of employment will

be partly explained by the housing wealth and firm collateral channel. Therefore
our initial efforts will focus on formally defining the housing wealth and firm col-
lateral channel of employment in terms of property tax changes.

Definition 2 (Housing Wealth and Firm Collateral Channel): Assume that Δ𝜏 ℎ >
0, Δ𝜏 𝑓 = 0, and 𝑤 = 𝑝𝑓 = 0. Then the housing wealth channel on employment
 58

(𝛿wealth ) is the response of 𝑙 generated labor demand changes due to a marginal

increase in Δ𝜏 ℎ . The latter is formally defined in (??).

𝜕𝑙 𝜕𝑙𝑑 𝜕𝑐 𝜕𝑝ℎ
𝛿wealth = = (1.30)
𝜕Δ𝜏 ℎ 𝜕𝑐 𝜕𝑝ℎ 𝜕Δ𝜏 ℎ

Equivalently, if Δ𝜏 𝑓 > 0, Δ𝜏 ℎ = 0, and 𝑤 = 𝑝ℎ = 0. Then the firm collateral channel

on employment is the labor demand response of 𝑙 after an infinitesimally higher
increase in Δ𝜏 𝑓 . The formal definition of the latter can be found in (1.31).

𝜕𝑙 𝜕𝑙𝑑 𝜕𝑝𝑓
𝛿coll = = (1.31)
𝜕Δ𝜏 𝑓 𝜕𝑝𝑓 𝜕Δ𝜏 𝑓

where 𝑙𝑑 = log 𝐿𝑑1 − log 𝐿𝑑0 is the log-percentage response of labor demand 𝐿𝑑 to the
change in property tax schemes.

According to Definition , higher residential property taxes—keeping commer-

cial real estate taxes constant—should induce a response of employment through
the housing wealth channel. Likewise, an increase in commercial real estate taxes—
leaving residential property taxes unchanged—produces an employment adjust-
ment through the firm collateral channel.
Moreover, notice that the definition of each channel is done in partial equilib-
rium as only the price of the real estate asset being taxed more heavily is free to
adjust. The latter also implies that the labor supply remains constant. Therefore,
only labor demand adjustments are relevant for quantifying the housing wealth
and firm collateral channel on employment.
Equations (1.30) and (1.31) provide additional insights on the internal work-
ings of each channel affecting employment after an increase in property taxes. On
the one hand, the far-right side of (1.30) shows that the housing wealth channel
 59

captures an indirect response of employment caused by lower household demand

due to a drop in residential prices after the increase in property taxes. On the other
hand, the firm collateral channel capture the direct response of employment de-
mand due to the lower firm’s collateral value resulting from a decline in commercial
real estate prices after the increase in property taxes.
We can further characterize each channel in the model by obtaining a closed-
form expression that depends on the vector of structural parameters Θ. Proposi-
tion 5 presents the analytical expressions for the housing wealth and firm collateral
channel.

Proposition 5 (Housing Wealth and Firm Collateral Channel): Applying def-

inition 1.6.4 to the log-linear constrained competitive equilibrium described in
Proposition 3, we can obtain equations (1.32) and (1.33), which represent the
closed-form expressions for the housing wealth and firm collateral channel on
employment in our model, respectively.

1+𝜈 1 + 𝜎ℎ
𝛿wealth (Θ) = − ( )( ) (1.32)
1 + 𝜙ℎ 1 + 𝜎ℎ + (1 − 𝛽)𝜈

𝜖 ⎞⎛ 1 + 𝜎𝑓
𝛿coll (Θ) = − ⎛
⎜ ⎟⎜ ⎞
⎟ (1.33)
⎝ 1 + 𝜙𝑓 ⎠ ⎝ 1 + 𝜎𝑓 + (1 − 𝛼)(𝜖 − 1) ⎠

As we can see in (1.32) and (1.33), both channels affect employment demand neg-
atively. Moreover, their values are determined mainly by the real estate supply
elasticities (𝜎ℎ , 𝜎𝑓 ) and loan-to-value ratios (𝜙ℎ , 𝜙𝑓 ).
 60

INSPECTING THE MECHANISM: EMPLOYMENT AND PROPERTY TAX CHANGES

We move to describe one of the main results of the paper. The latter provides a link
between the 𝑖-tax rate elasticity of employment (i.e., 𝛽𝑙,𝑖 ) and the the property tax-
induced housing wealth and firm collateral channel on employment (i.e., 𝛿wealth ),
𝛿coll ).
In particular, Proposition 5 provide a liinear decomposition of 𝛽𝑙,ℎ (Θ) and 𝛽𝑙,𝑓 (Θ),
parameters capturing the overall effect of a change in property taxes on employ-
ment in the model.

Proposition 6 (Decomposing the Employment Response): Coefficients 𝛽𝑙,ℎ

and 𝛽𝑙,𝑓 can be decompose as follows:

(1 + 𝜎ℎ )(1 + 𝜈)
𝛽𝑙,ℎ = 𝛿wealth − (1 − 𝛼)(𝜖 − 1)𝛽𝑝𝑓 ,ℎ + [ − (1 + 𝛼(𝜖 − 1))] 𝛽𝑤,ℎ
𝐴ℎ
(1.34a)

(1 + 𝜎𝑓 )(1 + 𝜎ℎ ) 𝛼(𝜖 − 1)(1 + 𝜎𝑓 )

𝛽𝑙,𝑓 = 𝛿coll + 𝛽𝑝ℎ ,𝑓 − [ + 1]𝛽𝑤,𝑓 (1.34b)
𝐴𝑓 𝐴𝑓

where 𝛿wealth and 𝛿coll are defined by (1.32) and (1.33) respectively.

According to eq:decomp_lh Proposition 5, we can separate the employment re-

sponse after an increase in residential taxes into two parts. The first part captures
the employment response through the housing wealth channel 𝛿wealth ). The second
part represents the combined response of employment to a general equilibrium ad-
justment of commercial real estate prices and wages. We can use a similar interpre-
tation to describe the linear decomposition of the 𝑓 -tax elasticity of employment. In
this case, the employment response after an increase in commercial real estate is the
 61

sum of the collateral channel 𝛿coll ) and a term representing the general equilibrium
adjustment of residential prices and wages.

1.7 CALIBRATION STRATEGY

This section describes the specific steps to discipline the quantitative model in de-
tail. The general idea behind the calibration procedure is to use the estimates ob-
tained with municipal-level data for Italy as target moments to be replicated by the
model. However, we will not target some moments estimated with the data. In-
stead, we use these to perform a validation test for the model. The specifics of this
procedure are presented in the following section.

1.7.1 INTERNAL CALIBRATION

Recall that the model has eight structural parameters captured by the vector Θ =
[𝛼 , 𝛽 , 𝜈 , 𝜖 , 𝜎ℎ , 𝜎𝑓 , 𝜙ℎ , 𝜙𝑓 ]. We will calibrate internally the parameters 𝜎ℎ , 𝜎𝑓 , 𝜙ℎ , 𝜙𝑓 .
The logic behind this choice is explained by the results found in proposition 7.
In particular, equation (1.32) shows that the housing wealth effect on employ-
ment is determined by the housing elasticity of supply 𝜎ℎ and the loan-to-value
ratio for households 𝜙ℎ , given a value for 𝛼 and 𝛽. In the same manner, equation
(1.33) shows that the firm collateral effect on employment is defined by the sup-
ply elasticity of commercial real estate 𝜎𝑓 and loan-to-value ratio for firms 𝜙𝑓 , for a
fixed choice of𝛼 and 𝜖.
Let Θin = [𝜎ℎ , 𝜎𝑓 , 𝜙ℎ , 𝜙𝑓 ] denote the subset of parameters calibrated in the model,
and Θout = [𝛼, 𝛽, 𝜈, 𝜖] the parameters set externally.
 62

1.7.2 TARGET MOMENTS

According to equation (1.29) in proposition 4, the main variables of the economy

respond linearly to changes in property taxes. The linear functional relation is simi-
lar to the parametric specification in (1.29) used to estimate the reduced form effect
of the 2012 property tax reform in Italy. Thus, we can use the estimates presented
in table 1.2 as target values for the reduced form coefficients in the model.
To differentiate between model and estimation results, we denote 𝛽𝑦,𝑖 as the
̂ represent its empirical estimate. The
model 𝑖-tax rate elasticity for 𝑦, while 𝛽𝑦,𝑖
proposition presented below shows that there is a one-to-one relationship between
the parameters in Θ𝑖𝑛 and 𝛽𝑦,𝑖 .

Proposition 7 (Reduced form Coefficients and Calibrated Parameters):

Using equation (1.29) in proposition 3 and fixing the value for Θout , the subset
of parameters in Θin are related to 𝛽𝑦,𝑖 in the following way:

𝛽𝑝ℎ ,ℎ = 𝛽𝑝ℎ ,ℎ (𝜎ℎ , Θ0 ) (1.35a)

𝛽𝑐,ℎ = 𝛽𝑐,ℎ (𝜙ℎ , Θ0 ) (1.35b)

𝛽𝑝𝑓 ,𝑓 = 𝛽𝑝𝑓 ,𝑓 (𝜎𝑓 , Θ0 ) (1.35c)

𝛽𝑝ℎ ,𝑓 = 𝛽𝑝ℎ ,𝑓 (𝜙𝑓 , Θ0 ) , (1.35d)

The results of proposition 7 provide with a specific procedure to discipline

the model with the empirical estimates. Using equations (1.35a)-(1.35d), we can
clearly see that to calibrate Θin = [𝜎ℎ , 𝜎𝑓 , 𝜙ℎ , 𝜙𝑓 ] in the model, we need four mo-
 63

ments estimated with Italian municipal level data, namely 𝛽𝑝̂ ℎ ,ℎ , 𝛽𝑐,ℎ
̂ , 𝛽 ̂ 𝑓 and 𝛽 ̂ ℎ .
𝑝 ,𝑓 𝑝 ,𝑓

In particular, we choose the value of: (i) housing elasticity of supply (𝜎ℎ ) to match
𝛽𝑝̂ ℎ ,ℎ , (ii) loan-to-value ratio for households (𝜙ℎ ) to target 𝛽𝑐,ℎ
̂ , (iii) commercial

real estate elasticity of supply (𝜎𝑓 ) to match 𝛽𝑝̂ 𝑓 ,𝑓 and (iv) the loan-to-value ratio
for firms (𝜙𝑓 ) to match 𝛽𝑝̂ ℎ ,𝑓 .

1.7.3 CALIBRATION RESULTS

Regarding the parameters defined externally, we choose to fix the labor share 𝛼 to
0.6, while the Frisch elasticity 𝜈 and elasticity of substitution across varieties 𝜖 are
set to 1 and 4 respectively, these values are consistent with what is commonly used
in the literature. Finally, we set the expenditure share on housing goods 𝛽 to 0.8
following Berger et. al (2018). This is summarized in table 1.6. The results for the
parameters calibrated internally can be found in table 1.7. As explained earlier, we
calibrate 𝜎𝑓 , 𝜎ℎ , 𝜙ℎ and 𝜙𝑓 to target 𝛽𝑝̂ ℎ ,ℎ , 𝛽𝑐,ℎ
̂ , 𝛽 ̂ 𝑓 and 𝛽 ̂ ℎ respectively.
𝑝 ,𝑓 𝑝 ,𝑓

The first two rows in table 1.7 show the calibrated values for the housing and
commercial real estate supply elasticities. In comparison, Saiz (2010) estimate a
supply elasticity of 16.6724 using data for housing price changes across US metropoli-
tan areas during the period 1970-2000. In our case, the calibration exercise is based
on empirical estimates using short-run variation in housing and commercial real
estate prices. Therefore, our calibration results for the short-run elasticity of sup-
ply are consistent with the long-run estimates in the literature.
The last two rows present the calibrated values for the loan-to-value ratio for
households and firms, respectively. As a benchmark comparison, we use the 2012
Survey of Households, Income and Wealth (SHIW) for Italy to obtain information
on households that asked and received a loan during the past three years using their
24
See Saiz (2010) TABLE III, column (4).
 64

real estate assets as collateral. We focus on owners of a single house or commercial

real estate property and compute the ratio of the loan obtained and the self-report
value for the property owned. The average loan-to-value ratio for single house
owners and owners of a single commercial real estate property is 0.42 and 0.50,
respectively. These values are not too far from the calibrated results for 𝜙ℎ and 𝜙𝑓 .

TABLE 1.6: Externally Defined Parameters

Parameter Value Target

Labor Share 𝛼 0.6 Common in literature
Frisch elasticity 𝜈 1 Common in literature
Elasticity of demand 𝜖 4 Common in literature
Exp. share goods 𝛽 0.8 Berger et al. (2018)

TABLE 1.7: Calibrated Parameters

Parameter Value Target

Supply elast. houses 𝜎ℎ 4.87 𝛽𝑃̂ ℎ ,ℎ
Supply elast. CRE 𝜎𝑓 2.40 𝛽𝑃̂ 𝑓 ,𝑓
LTV HH’s 𝜙ℎ 0.23 ̂
𝛽𝐶,ℎ
LTV firms 𝜙𝑓 0.35 𝛽𝑃̂ ℎ ,𝑓

1.7.4 VALIDATION EXERCISE: MODEL PREDICTIONS VS. ESTIMATION

RESULTS

Notice that the calibration procedure do not target the reduced form coefficients
related to employment. We will use the non-target moments to test the validity of
the model’s predictions. In particular, using the calibration results for Θin and the
values for Θ𝑜𝑢𝑡 defined externally we can compare the model-implied coefficients
̂ .
𝛽𝑙,𝑖 (Θin , Θout ) with the empirical estimates 𝛽𝑙,𝑖
 65

The results for the consistency exercise are represented in table 1.8. The first col-
umn shows the predictions of the model regarding 𝛽𝑙,𝑖 . The second column repro-
duces the values of table 1.2 for non-tradable employment, while the third column
computes the 95% confidence intervals for the point estimates. The model predicts

TABLE 1.8: Labor Response: Model vs Data

Model Data
𝛽𝑙,𝑖 ̂
𝛽𝑙,𝑖 95 % CI
Δ𝜏 ℎ 0.074 0.087 [0.06, 0.12]
Δ𝜏 𝑓 0.061 0.045 [0.02, 0.07]

that a 1 pp increase in the tax rate for houses reduce employment growth by 0.074
pp, which is slightly lower than the 0.087 point estimate but still within the 95%
confidence interval. On the other hand, the model predicts that a 1 pp. increase
in commercial real estate taxes reduce employment growth by 0.052 pp. which is
higher than the 0.041 obtained with the data, but still within the 95% confidence in-
terval. Based on the previous results, we can conclude that the model does a fairly
good job replicating non-target empirical estimates for employment.

1.8 MEASURING COLLATERAL CHANNELS: QUANTITATIVE

RESULTS

We can now move to the paper’s primary goal, estimating the relative importance
of the housing wealth and firm collateral channel. We can use the calibrated model
to estimate the magnitude of each channel. Recall that our quantitative model is
not only consistent with the targeted empirical estimates but also replicates fairly
good non-targeted estimates for employment.
 66

To quantify each channel we replace the values for the calibrated parameters in Θin
and the externally defined parameters in Θout on equations (1.32) and (1.33). The
results from this are presented in table 1.9. The results in column 𝛿𝑖 of table 1.9

TABLE 1.9: Household Wealth and Firm Collateral Chanel

𝛿𝑖 𝛽𝐿 𝛿𝑖 /𝛽𝐿
𝛿𝐿wealth -0.073 -0.074 98 %
𝛿𝐿coll -0.052 -0.061 84 %

shows the model predictions for the housing wealth and firm collateral channel.
According to the model, a 1 pp point increase in the tax rate for houses reduces
non-tradable employment growth by 0.073 pp through the housing wealth channel,
representing 98% of the overall 0.074 pp drop in employment growth to higher
housing taxes.
On the other hand, a 1 pp point increase in the tax rate for commercial real
estate reduces non-tradable employment growth by 0.052 pp via the firm collateral
channel, equivalent to 84% of the overall 0.061 pp decline in employment growth
caused by the increase in commercial real estate taxes.
The results of table 1.9 have two main implications. First, both channels are cru-
cial to understanding the employment consequences of a drop in real estate prices
induced by higher property taxes. Second, even if the housing wealth channel rela-
tively dominates the firm collateral channel, the latter retains its importance when
an employment decline is caused by a drop in real estate prices.

1.9 CONCLUSIONS AND FUTURE WORK

This paper focus on understanding the employment consequences of a drop in real

estate prices through the housing wealth effect and firm collateral effect. Previ-
 67

ous empirical work on this subject focus on estimating each channel separately.
However, not controlling for the part explain by one channel can bias the estima-
tion results mainly because housing prices and commercial real estate prices are
correlated. Even if the empirical approach controls for each channel, we still need
an explicit strategy to separate both given that the housing wealth and collateral
channel affect employment in a similar way. This paper provides a unifying ap-
proach to model and estimate both channels on employment. The approach taken
in this paper is based on combining empirical evidence with a quantitative model.
In particular, the paper estimate the reduced form effect of a property tax change
on real estate prices and consumption expenditure, these estimates are then used
to disciple the model, the latter provides an estimate for each separate channel.
The theoretical relation between property taxes and real estate prices provides
an argument to use the former to quantify each channel separately. Intuitively,
an increase in property taxes translate into higher costs of purchasing and hold-
ing real estate which in turn reduce demand and market prices. If we observe an
increase in property taxes only applied to houses then its effect on employment
should be partially captured by the drop in labor demand through the housing
wealth channel, while if we only observe an increase in property taxes applied to
commercial real estate then its total effect on employment should include the drop
in labor demand via the firm collateral channel. The 2012 property tax reform in
Italy represents a unique event that can be used to estimate the separate effect of
an increase in housing and commercial real estate. First, the Italian property tax
system is defined by two tax rates named prin and sec, the former applies only to
households main residence (i.e houses) while the latter applies to other proper-
ties including the ones used for production purposes (i.e commercial real estate).
The 2012 tax reform in Italy forced municipalities to set higher tax rates for houses
 68

and commercial real estate and produce a unique variation in property tax changes
across municipalities not observed before or after 2012. The empirical approach
use variation in property tax changes across municipalities during the 2012 prop-
erty tax reform to estimate the reduced form effect of a change in property taxes on
employment, housing and commercial estate prices and consumption expenditure
using a difference-in-difference approach with municipal level panel data period
2008-2014. The empirical results show that municipalities with a high increase in
property taxes had lower growth rate of non-tradable employment, consumption
expenditure, housing prices and commercial real estate prices.
We use the empirical estimates to discipline a quantitative model. The model in-
cludes two types of real estate assets one used by households and the other by firms
(i.e houses and commercial real estate), owners of real estate need to pay differen-
tiated property taxes to the government. The equilibrium solution for prices and
allocations provides a closed form expression for the employment response caused
by higher property tax rates. In particular, the reduced form effect of an increase in
property taxes for housing prices, commercial real estate prices and household con-
sumption are sufficient statistics for the model-implied response of employment
to property tax changes. Moreover, the model shows that the employment effect
of an increase housing taxes is explained partially by the household collateral ef-
fect, while the firm collateral effect is included in the response of employment to
an increase in the tax rate for commercial real estate. The calibration approach
use four moments estimated using Italian data as targets, namely the estimates for
the response of housing prices and consumption expenditure to an increase hous-
ing taxes and the estimates for the response of commercial real estate and housing
prices to an increase in commercial real estate taxes. In the model, these four mo-
ments are pinned down by price-elasticity for housing supply, loan-to-value ratio
 69

for households, price-elasticity of supply for commercial real estate and the loan-
to-value ratio for firms respectively. To test for consistency between data estimates
and model predictions we compare the model-implied response of employment to
higher property taxes with their empirical counterparts, the results shows that the
model provides a good approximation of this this non-target moments compared
to the ones estimated using Italian data during the 2012 tax reform.
With the calibrated model we quantify the relative importance of the two chan-
nels on employment. The model predictions show that the reduced form response
of employment to an increase in property taxes is mostly explained by the housing
wealth and firm collateral channel. In particular, the model predicts that a 1 pp
increase in the tax rate for commercial real estate reduce non-tradable employment
growth by 0.042 pp. with 83% (0.035 pp.) of this decline explained by the collat-
eral channel, while a 1 pp. increase in the tax rate for houses reduce non-tradable
employment growth by 0.054 pp. with 77% (0.045 pp.) of this decline explained by
the housing wealth channel. The direct impact of a drop in commercial real estate
prices on the demand for employment explains the marginal dominance of the firm
collateral channel over the housing wealth channel. Nevertheless, the results of the
model clearly show that both channels are crucial to understand the employment
overall effect of an increase in property taxes.
 70

CHAPTER 2

INVESTMENT, CAPITAL STRUCTURE AND DE-

FAULT RISK

Sovereign debt crises are events associated with high default risk and a decline in
aggregate output. Several explanations are used in the sovereign default literature
to explain this pattern. One in particular examines the role of investment in explain-
ing the decline in aggregate output during a sovereign debt crisis episode through
a bank-lending channel. In particular, Palmén (2020) show that after an increase
in default risk, domestic banks with a higher share of government bonds in their
asset portfolio cut loans to the private sector more strongly. While, ? finds that: (i)
firms linked to highly exposed banks reduce output more aggressively, and (ii) the
negative response of firms to default risk exposure is more pronounced for highly
indebted firms.
In this paper we will try to understand how the capital structure of firms shape
the response of investment to changes in default risk. The capital structure of a
firm is summarized by two dimensions: leverage and maturity. With leverage cap-
 71

turing the importance for firms of external funds to fund operations. While ma-
turity defines the length of time, the firm chooses to repay its past commitments.
In the presence of financial frictions, leverage affect investment due to debt over-
hang problems, while maturity can impact investment decisions through changes
in rollover risk.
The corporate finance literature define debt overhang as
We will focus on the Italian economy during the euro-zone debt crisis of 2010-
12. The particular evolution of investment and capital structure in Italy during this
period motivates our emphasis on the Italian sovereign debt crisis. Italy, along with
other euro countries (i.e., Greece, Ireland, Portugal, and Spain), experienced a his-
torical increase in the spread of government bonds and a sharp decline in aggregate
investment not observed for the rest of the euro-area (Figures 2.1 and 2.2). How-
ever, relative to the other euro countries affected by the debt crisis, Italy’s aggregate
investment declined more smoothly and recovered more slowly. Other differences
also stand out, particularly when comparing the aggregate Debt-to-Assets ratio and
share of Long Term Debt for non-financial corporations (Figure 2.3 and 2.4). Non-
financial corporations in Italy went through a more substantial deleveraging pro-
cess while relocating liabilities toward long-term debt faster than other euro coun-
tries also affected by the debt crisis of 2010-12.
The inclusion of maturity and leverage in the analysis is not only realistic but
also creates new channels to study the effect of the capital structure on the invest-
ment policy function of the firm. In particular, the heterogeneous response of firms’
investment to aggregate financial shocks will differ if the firm can adjust its capital
structure using one margin or both. When leverage is the only margin that the firm
can adjust when choosing its capital structure, the response of firms’ investment to
financial shocks will be heterogeneous. Firms with high leverage will need to adjust
 72

their investment by more after a negative financial shock. The latter continues to be
true if the firm can adjust maturity and leverage, but now the interactions between
both margins create new layers of heterogeneity. The inclusion of maturity in the
analysis comes with some drawbacks, it will complicate the setup of the model,
and it will make the quantitative solution of the model a more challenging task.
Nevertheless, including both dimensions in the analysis is an appropriate choice
to correctly quantify the heterogeneous response of a firm’s investment to finan-
cial shocks. The main goal of the paper is to understand the effect of maturity and
leverage decisions on the response of firms’ investment to financial shocks. In par-
ticular, the paper answers the following question: Does the choice of maturity and
leverage affect the response of investment to financial shocks originated by changes
in the sovereign spread? The EU sovereign spread crisis of 2010-2012 will serve as
a laboratory to answer this question; this event had a significant negative impact
on the performance of firms in several EU countries. We will focus on Italian firms
during the sovereign spread crisis of 2010-2012. The case of Italy is of particular
interest for two reasons. First, the 2010-2012 sovereign spread crisis was a unique
event in the history of Italy as a democratic country. Second, the increase in the
sovereign spread for Italy during 2010-2012 is one of the highest relative to other
EU countries that were also affected by this episode1 .
The main contribution of the paper is to show empirically that the response of
firms’ investment to changes in the sovereign spread will differ depending on the
choices of leverage and maturity made by the firm. For high leverage firms, an in-
crease in the sovereign spread by 100 basis points is associated with a decrease of
0.59% in the growth rate of capital for high maturity firms relative to low maturity
1
From 2010 to 2012 the spread for 10-years Italian government bonds increased by more than
517%.
 73

firms. For low leverage firms, the same increase in the sovereign spread is associ-
ated with an increment of 0.38% in the growth rate of capital for high maturity firms
relative to low maturity firms. Moreover, differences in maturity will change the
magnitude of the response of firms’ investment to changes in the sovereign spread
when comparing high leverage versus low leverage firm. For firms with low ma-
turity, an increase in the sovereign spread by 100 basis points is associated with
a drop of 1.41% in the growth rate of capital for firms with high leverage relative
to firms with low leverage. For firms with high maturity, the same increase in the
sovereign spread is associated with a drop of 2.41% for the growth rate of capital
for firms with high leverage relative to low leverage firms. The empirical results
are robust to the inclusion of additional firm-level and aggregate controls, changes
in the measurement of investment, and to changes in the sample.
The paper is divided into three sections. The first section addresses the ques-
tion formulated initially by estimating the relationship between maturity, leverage,
investment, and sovereign spread risk using aggregate and firm level-data for Italy
during 2007-2015. Firm-level data comes from Orbis, and aggregate data comes
from Eurostat. The second part of the paper presents a quantitative model and de-
scribes the main mechanisms that allow the model to explain the empirical results
found in the paper. The last part concludes and delineates the next steps to be taken
in the future.
 74

250
Italy
Spain
Portugal
200 Ireland
Greece
150
Spread (bp.)
100
50
0

2006q1 2007q1 2008q1 2009q1 2010q1 2011q1 2012q1 2013q1 2014q1 2015q1 2016q1

Note: The spread of sovereign bonds is measured as the excess yield of long-term
government bonds relative to bonds for Germany.Source OCDE.

FIGURE 2.1: Sovereign Spread: Euro Countries

1.2
1.1
Investment (2005Q2=1)
1

Italy

Others
.9

Rest EU
.8
.7
.6

2005q2 2006q2 2007q2 2008q2 2009q2 2010q2 2011q2 2012q2 2013q2 2014q2 2015q2 2016q2

Note: Non-Financial Corporations Real Investment. Index 2005Q2=1. Others comprise

the aggregate of Greece, Portugal, Ireland and Spain, while Rest EU represents the
aggregate for the rest of EU countries. Source Eurostat.

FIGURE 2.2: Non-financial Corporations Investment

 75

1.1
Italy

Others

Rest EU

1.05
Debt/Assets (2005Q2=1)
1
.95
.9

2005q2 2006q2 2007q2 2008q2 2009q2 2010q2 2011q2 2012q2 2013q2 2014q2 2015q2 2016q2

Note: Aggregate Debt-to-Assets ratio for Non-Financial Corporations. Index

2005Q2=1.Source Eurostat.

FIGURE 2.3: Non-financial Corporations Leverage

1.2
1.15
Long-Term Debt Share (2005Q1=1)
1.1

Italy

Others

Rest EU
1.05
1
.95

2005q2 2006q2 2007q2 2008q2 2009q2 2010q2 2011q2 2012q2 2013q2 2014q2 2015q2 2016q2

Note: Aggregate share of Long-Term liabilities for Non-Financial Corporations. Index

2005Q2=1. Source Eurostat.

FIGURE 2.4: Non-financial Corporations Maturity

 76

2.1 DATA

Before turning to the parametric specification and identification strategy we use in

this paper, this section describe the data sources and variable construction that will
be used. The empirical analysis is based on Italian yearly data during the period
2007-2015.

2.1.1 FIRM-LEVEL DATA

Firm-level data on Italian firms comes from Orbis global database from Bureau van
Dijk (BvD). The Orbis dataset is the largest cross-country firm-level database pro-
viding detailed information on balance sheets variables for public and private firms.
The main firm-level variables used are: sales, operating revenue, employment, total
assets, total fixed assets, tangible fixed assets, intangible fixed assets, loans, credi-
tors, long term liabilities, operating profits, net income (EBITDA), and total interest
payments. In the balance sheet items, loans and creditors are considered current
liabilities, then will have a maturity of fewer than 12 months. Long-term debt is a
non-current liability with a maturity of more than 12 months. Then, we define the
former as short term liabilities while the latter is considered long term liabilities.
In order to improve the quality of the Orbis data we follow the cleaning pro-
cedure suggested by Gopinath et al. (2017). To eliminate reporting mistakes, we
drop firm-year observations with missing information in any of the main variables.
Also, firm-year observations with negative values in operating revenues, total as-
sets, and total fixed assets are eliminated. To eliminate outliers, I winsorized the
main variables at the 1st and 99th percentile. Moreover, we also eliminate firms
 77

that are in industries with strong government presence 2 . Finally, we only keep
firms with complete information during the period 2007-2015 to construct the final
balanced panel dataset.
I will focus on investment as the main as the outcome variable in the regression
analysis. The capital stock will be the sum of tangible fixed assets and intangible
fixed assets. Investment will be proxied by the log annual difference in capital stock
(Δ log(𝑘𝑖𝑡 )). Firms financial constraints are proxied by Leverage (𝑙𝑒𝑣𝑖𝑡 ) defined as
the ratio of total liabilities over total assets, where total liabilities are the sum of
short-term debt (loans plus creditors) and long term debt. Maturity (𝑚𝑎𝑡𝑖𝑡 ) is the
ratio of long-term liabilities over total liabilities. Other firm-level variables used in
the regression analysis are size (𝑙𝑜𝑔(𝐴𝑖𝑡 )) proxied by the log of total assets, profit
ratio (𝜋𝑖𝑡 ) defined as the ratio of operating profits over operating revenues and
average interest rate (𝐼𝑛𝑡𝑖𝑡 ) measured as the total interests paid over net income.

2.1.2 AGGREGATE DATA

All aggregate yearly data for Italy comes from Eurostat. The main variable of inter-
est is sovereign spread (𝑠𝑝𝑟𝑡 ), which is define as the difference between the nominal
interest rate of 10-year Italian government bonds and 10-year Germany government
bonds. Other aggregate variables used are real GDP (annual chain-linked volume
at 2010 prices.) and Consumer Price Index (base=2010). The latter is used to de-
flate all the nominal firm-level nominal variables to be expressed in constant prices
of 2010.
2
These industries correspond to NACE codes 84 (Public Administrations and National Defense)
85 (Education) and 86-88 (Health Care)
 78

2.1.3 SUMMARY STATISTICS

The summary statistics for firm-level variables in 2007 is presented in Table 2.1.
The sample comprise a balance panel of 80, 718 privately held firms that operated
continuously during 2007 2015. For 2007, the median firm received 21, 000 euros
as revenues, have eight employees, 1.432 millions of euros as assets, and 1.110 mil-
lion euros as total liabilities. Finally, in terms of capital structure, the median firm
liabilities represents 83% of total assets with 16% of the liabilities as long term debt.

TABLE 2.1: Summary Statistics - 2007 : Firm Level Variables

Mean S.D 𝑃25 𝑃50 𝑃75

Employment 12 27 4 8 15
Operating Revenue 33.5 33.8 10.2 21.3 44.1
Assets 2,202.5 2,165.4 677.6 1,432.9 2,965.2
Total Liabilities 1,735.6 1,802.5 514.5 1,110.2 2,296.1
Profit ratio 1.64 4.30 0.01 0.92 3.02
Interest Service 27.61 49.67 4.29 19.41 46.53
Maturity 0.216 0.190 0.067 0.163 0.314
Leverage 0.789 0.187 0.690 0.834 0.92
Notes: The table presents summary statistics for 80, 718 firms during 2007. All monetary variables are
expressed in hundred of real euros of 2010.

2.2 CAPITAL STRUCTURE AND FIRM CHARACTERISTICS

This section summarizes the main characteristics of firms in terms the capital struc-
ture. The analysis that follows allow us to understand how maturity and leverage
is associated with other firm level variables which in turn will be useful when pre-
senting the empirical strategy. We start by documenting the main facts regarding
maturity and leverage of Italian firms during 2007-2015, then we move to classify-
ing the firms into capital structure groups, and finally we will show how balanced
are these groups of firms in terms of other firm observable characteristics. Matu-
 79

rity and Leverage. Figure 2.5 plots the average maturity and leverage for Italian
firms during 2007 2015. In this period the average Italian firm deleverage continu-
ously and move their debt portfolio towards long term debt, these patter is specially
observed during the 2007-08 financial crisis and the 2010-12 euro zone debt crisis.

FIGURE 2.5: Average Leverage and Maturity for Italian Firms

Capital Structure and Firm Characteristics. To understand how firms with

different capital structure differ across other observable characteristics we clas-
sify firms along maturity and leverage groups using the cross section distribution
for 2007. In particular, a firm is considered high leverage/maturity if it is above
the 75th percentile and low leverage/maturity otherwise. This creates four cap-
ital structure groups of firms: (i) high leverage-maturity (High-High), (i) high
leverage-low maturity (High-Low), (i) low leverage-maturity (Low-Low), and (i)
low leverage-high maturity (Low-High).
 80

We compare firms along five observable dimensions: revenues per worker, as-
sets per worker, profit rate, investment rates, total productivity, and interest ser-
vice. Table 2.2 report the average values across different leverage (lev) and matu-
rity (mat) groups for the year 2007. We can see that capital structure groups firms
are relatively balanced in terms of revenue and assets per worker and total produc-
tivity. However, there seem to be significant differences in profit rates, interest ser-
vices and investment rates. In particular, high leverage firms have negative profits
and have higher interest costs. While high leverage firms with most of its liabilities
as long term debt invest more in capital.

TABLE 2.2: Balance across Capital Structure Groups

Revenues Assets Profit Investment Interest

lev - mat Log(TFP)
per worker per worker Rate Rate Service
High-High 3.3 317.9 -1.14 1.34 3.62 43.04
High-Low 5.9 355.0 -0.40 0.87 4.00 39.86
Low-High 3.2 304.9 2.34 0.78 3.70 27.01
Low-Low 5.5 333.1 2.40 0.76 4.00 22.00
S.D (All groups) 7.02 435.7 4.30 4.92 0.75 49.7
Notes: The table shows the average of different firm level variables across four different capital structure groups computed for the year 2007.
All monetary variables are expressed in hundreds of real euros of 2010. Total productivity is proxied by TFP estimated following Wooldridge
(2009) extension of the 2-step TFP procedure estimation proposed by Levinsohn and Petrin (2003).

2.3 EMPIRICAL STRATEGY

In this section we will all the relevant details on the estimation strategy. First, we set
out parametric specification based on the capital structure groups defined earlier,
the parameters of interest and the implicit identification requirements. Then, we
discuss the potential threats to identification, how we deal with the latter and the
limitations of the estimation strategy.
 81

2.3.1 PARAMETRIC SPECIFICATION

The baseline regression is presented in equation 2.1. The main outcome variable
of interest is investment of firm 𝑖 at year 𝑡 represented by Δ log 𝑘𝑖𝑡 . The main left
hand side variables are spread of Italian government bonds 𝑠𝑝𝑟𝑡 and the capital
structure group dummies 𝐷lev = 1{lev𝑖 > 𝑝75} and 𝐷lev = 1{mat𝑖 > 𝑝75} taking the
value of one if firm 𝑖 has a leverage or maturity above the 75𝑡ℎ percentile of the
2007 distribution, respectively. We include firm level fixed effects 𝐹𝐸𝑖 to control for
constant unobserved firm characteristics that explain investment differences across
firms. The baseline specification won’t include year fixed effects because these will
be perfectly collinear with aggregate spread 𝑠𝑝𝑟𝑡, this will be changed later when
3
testing the robustness of our baseline results. The coefficients of interest are {𝛽𝑘 }𝑘=0 .

Δ log 𝑘𝑖𝑡 = FE𝑖 + 𝛽0 𝑠𝑝𝑟𝑡 + 𝛽1 𝐷lev ×𝑠𝑝𝑟𝑡 + 𝛽2 𝐷mat ×𝑠𝑝𝑟𝑡 + 𝛽3 𝐷lev ×𝐷mat ×𝑠𝑝𝑟𝑡 + 𝜖𝑖,𝑡 (2.1)

Coefficient 𝛽0 capture the response of investment to changes in default risk, while

𝛽1 and 𝛽2 capture the differential response of investment due to high leverage and
maturity. The interaction 𝐷lev ×𝐷mat is necessary given that maturity and leverage
are not defined independently but simultaneously defined the capital structure of
firms. Then by estimating 𝛽3 we should be able to correctly capture the heteroge-
neous response of firm investment to changes in default risk generated by capital
structure differences. In particular if spread 𝑠𝑝𝑟𝑡 increase by 100 basis points, the
average response of investment is: (i) 𝛽0 for low leverage-low maturity firms, (ii)
𝛽0 + 𝛽2 for low leverage-high maturity firms, (iii) 𝛽0 + 𝛽1 for high leverage-low
maturity firms, and (iv) 𝛽0 + 𝛽1 + 𝛽2 + 𝛽3 for high leverage-high maturity firms.
Using (i)-(iv) we can measure the relative response investment to default risk
 82

due to one dimension of capital structure keeping the other dimension constant,
this is summarized in Table 2.3.

TABLE 2.3: Response of Investment and Capital Structure

lev-mat High mat Low mat Δ mat | lev

High lev 𝛽0 + 𝛽1 + 𝛽2 + 𝛽3 𝛽0 + 𝛽 1 𝛽2 + 𝛽3
Low lev 𝛽0 + 𝛽 2 𝛽0 𝛽2
Δlev | mat 𝛽1 + 𝛽 3 𝛽1

2.4 ESTIMATION RESULTS

Table 2.4 shows the estimates obtained for all the coefficients of interest. We can
see that leverage generates an asymmetric and heterogeneous response of a firm’s
investment to changes in the sovereign spread across maturity categories. For high-
leverage firms, an increase in the sovereign spread by 100 basis points is associ-
ated with an additional drop of 0.59% in the capital growth rate if the firm is high-
maturity relative to low-maturity firms. For low leverage firms, the same increase
in the sovereign spread is associated with an additional increase of 0.38% in the
capital growth rate if the firm is high-maturity relative to low-maturity firms. Both
estimations are statistically significant at 5% confidence level. On the other hand,
table 2.4 shows that maturity can intensify the negative response of a firm’s invest-
ment to changes in the sovereign spread across leverage categories. For high ma-
turity firms, an increase in the sovereign spread by 100 basis points is associated
with an additional reduction of 2.41% in the growth rate of capital if the firm is
high-leverage relative to low-leverage. For low maturity firms, the same increase in
the sovereign spread is associated with an additional drop of 1.41% in the growth
rate of capital if the firm is high-leverage relative to low-leverage firms. Therefore
 83

we can conclude that across maturity and leverage categories, an increase in the
sovereign spread by 100 basis points is correlated with a reduction of the growth
rate of capital by 0.97%.

TABLE 2.4: Baseline Estimation Results

lev-mat High mat Low mat Δ mat | lev

High lev -4.61** -4.02** -0.59**
Low lev - 2.20** - 2.58** 0.38**
Δlev | mat -2.41*** -1.44***

2.4.1 TESTING ROBUSTNESS OF BASELINE RESULTS

Table 2.5 present the coefficients of interest under different specification. Column
(1) controls only for firm fixed effects changes in the sovereign spread are nega-
tively correlated with the growth rate of capital. Column(2) includes an interaction
term between the spread and the categorical leverage variable. As we can see, the
growth rate of capital stock reduced more for high leverage firms after an increase
in the sovereign spread. The results are in line with previous empirical evidence
for Italy and other EU countries during the 2010-2012 sovereign spread episode.
Columns (3)-(6) include the interaction terms between the categorical variables
for maturity and leverage. The difference among columns (3)-(6) is the additional
controls included in the regression specification. Column (3) add firm fixed effects.
Column (4) includes firm-level and aggregate level controls. Column(5) is the
preferred regression specification, including all the controls described in equation
2.1. Across specifications, the estimation results are robust in terms of sign and
significance of the coefficients of interest. Only in column (4) the significance of 𝛽2
changes. The high collinearity between the sovereign spread and other aggregate
controls that this specification incorporates creates this robustness issue. Finally,
 84

column (6) includes an industry-year fixed effect to control for industry time trends
that could bias the estimation results in column (5). The estimations coefficients
are robust to the inclusion of this additional control.

TABLE 2.5: Alternative Specifications

(1) (2) (3) (4) (5) (6)

𝑠𝑝𝑟𝑡 -2.80*** -2.70*** -2.73*** -2.05*** -2.58***

(-72.40) (-58.46) (-57.25) (-42.22) (-37.47)
𝐷lev × 𝑠𝑝𝑟𝑡 -0.40*** -0.45** -0.95*** -1.44*** -0.93***
(-4.83) (-2.34) (-5.02) (-7.37) (-4.88)
𝐷mat × 𝑠𝑝𝑟𝑡 0.57*** 0.17 0.38** 0.48***
(3.04) (0.93) (2.07) (2.63)
𝐷lev × 𝐷lev × 𝑠𝑝𝑟𝑡 -0.48* -0.65** -0.97*** -0.71***
(-1.73) (-2.43) (-3.55) (-2.65)

Firm FE ✓ ✓ ✓ ✓ ✓ ✓
Firm/Aggregate Controls ✓ ✓ ✓
Interactions ✓ ✓
Industry-Year FE ✓
𝑁 726,462 726,462 726,462 726,462 726,462 726,462
𝑅2 0.110 0.110 0.110 0.151 0.151 0.180
 85

2.5 MODEL

The model presented in this section borrows heavily from Poeschl(2018). The main
concern of the model is to characterize the decisions of firms regarding maturity,
leverage, and investment. The model includes short term and long term debt,
which is risky due to limited commitment problems. The firm is affected by id-
iosyncratic shocks to productivity and fixed costs. Lenders are perfectly competi-
tive and price new debt issuances using all relevant information about firms’ de-
cisions. Information on firms’ characteristics is common knowledge to all lenders.
For simplicity, the model focus on partial equilibrium and includes sovereign spread
risk as an aggregate shock that changes the discount rate of lenders.

2.5.1 FIRMS

There is a continuum of firms that produce a homogeneous good 𝑦𝑡 sold in a per-

fectly competitive market at a price 𝑃𝑡 . The price of the homogeneous good is nor-
malized to one. Firms produce using only capital 𝑘𝑡 with decreasing returns to
scale. Capital depreciates each period at a rate of 𝛿. Investment 𝐼𝑡 is done by firms
and is subject to convex adjustment costs 𝐴𝐶𝑘 (𝐼𝑡 , 𝑘𝑡 ). The functional form of the
capital adjustment cost is as follows

2
𝐼
𝐴𝐶𝑘 (𝐼𝑡+1 , 𝑘𝑡 ) = 𝜃 ( 𝑡 ) 𝑘𝑡 , (2.2)
𝑘𝑡

where
𝐼𝑡 = 𝑘𝑡+1 − (1 − 𝛿)𝑘𝑡 , (2.3)
 86

will be the law of motion for capital stock. Each period, the firm is subject to a
productivity shock and a fixed cost shock. The productivity shock 𝑧𝑡 follows an
AR(1) process

log(𝑧𝑡+1 ) = (1 − 𝜇) + 𝜌𝑧 log(𝑧𝑡 ) + 𝜖𝑡 𝜖 ∼ 𝑁(0, 𝜎𝑧2 ) (2.4)

The fixed cost shock 𝑓𝑡 will be 𝑖𝑖𝑑 across time with 𝑃(𝑓𝑡 ≤ 𝑓 ) = 𝐺(𝑓 ) as cumulative
distribution function and support 𝐹 = ℝ++ . In order to continue operations in
the current and next period, the firm needs to have sufficient resources to pay the
fixed cost shock, investment expenditures for the next period, and past liabilities.
The firms can use internal resources to fund operations, which are represented by
operating profits (Π (𝑧𝑡 , 𝑓𝑡 , 𝑘𝑡 )) defined as follows

Π (𝑧𝑡 , 𝑓𝑡 , 𝑘𝑡 ) = 𝐴𝑧𝑡 𝑘𝑡𝛼 + (1 − 𝛿)𝑘𝑡 − 𝑓𝑡 . (2.5)

The firm can also issue short term debt 𝑏𝑆,𝑡 and long term debt 𝐼𝐿,𝑡 to finance oper-
ations at prices 𝑞𝑆,𝑡 , 𝑞𝐿,𝑡 respectively. Short term debt is in the form of one-period
non-contingent bonds. Each unit of long term debt is infinitesimally small matur-
ing each period with exogenous probability 𝜆, and with a coupon equal to 𝑐 < 1.
If 𝑏𝐿,𝑡 is the stock of long term debt of the firm, the principal payment is 𝜆 𝑏𝐿,𝑡 and
the coupon payment is (1 − 𝜆) 𝑐 𝑏𝐿,𝑡 . The stock of long term debt next period will
be
𝑏𝐿,𝑡+1 = 𝐼𝐿,𝑡 + (1 − 𝜆) 𝑏𝐿,𝑡 . (2.6)

By modeling long term debt in a probabilistic way, the state space of the model
will be significantly reduced. With random maturity, there is no need to follow
the entire path of debt issuances anymore. All relevant information about past
 87

decisions of long term debt is captured entirely by 𝑏𝐿,𝑡 .

In addition to the marginal cost to issue debt captured by the prices of each
bond, the firm will have to pay exogenous financing costs to issue new debt 𝐴𝐶𝑏 (𝑏𝑆,𝑡+1 , 𝐼𝐿,𝑡 ).
The exogenous adjustment function for debt captures all the additional costs in-
curred by the firm to issue more debt, for example, flotation costs, legal fees, or
bank fees for new loans. The issuance costs have a linear functional form3

𝐴𝐶𝑏 (𝑏𝑆,𝑡+1 , 𝐼𝐿,𝑡 ) = 𝜉 (∣𝑏𝑆,𝑡+1 ∣ + ∣𝐼𝐿,𝑡 ∣) (2.7)

Under this functional form, short term and long term debt have the marginal is-
suance cost. The model does not exclude repurchases of long term debt, but this
option will be costly to firms. There is a need to include exogenous issuance costs
with risky long term debt. Otherwise, the equilibrium maturity levels will not be
consistent with the ones observed using firm-level data.
Each period the firm can default on its outstanding liabilities, short term debt
default will always lead to long term debt default and vice-versa. A limited liability
clause protects shareholders from receiving a negative dividend payment. There-
fore, the firm can not issue external equity to finance operations. A non-negative
dividend constraint captures the limited liability clause4 .

𝐷𝑡 ≥ 0 (2.8)
3
The choice of the functional form is consistent with corporate finance literature (Titman and
Tsyplakov(2007), Poeschl(2018)), other alternative functional forms are a combination between a
purely linear and fixed issuance costs (Kuhen and Schmid (2014))
4
The corporate finance literature documented two empirical facts consistent with the non-
negative dividend condition: (i) Firms face high costs of equity issuance (Asquith and Mullins
(1986)), (ii) External equity issuance is rare even for listed companies (DeAngelo, DeAngelo, and
Stulz (2010)).
 88

where 𝐷𝑡 are the dividends of the firm. This additional constraint has to be satisfied
in each period that the firm operates in the market. If at any given period this
constraint can not be satisfied, the shareholders will force the firm to default. Let
𝑏𝑆,𝑡
𝑏𝑡 = 𝑏𝑆,𝑡 + 𝑏𝐿,𝑡 be the total stock of debt of the firm and maturity 𝑚𝑡 = 𝑏𝑡
the
maturity, it is convenient to express the model in terms of 𝑏𝑡 and 𝑚𝑡 . Therefore, the
current period dividends of the firm will be

2
𝜃 𝑘𝑡+1 −(1−𝛿)𝑘𝑡
𝐷𝑡 = Π (𝑧𝑡 , 𝑓𝑡 , 𝑘𝑡 ) − (1 − 𝑚𝑡 )𝑏𝑡 − (𝜆 + (1 − 𝜆) 𝑐) 𝑚𝑡 𝑏𝑡 − 𝑘𝑡+1 − 2
( 𝑘𝑡
) 𝑘𝑡

+𝑞𝑆,𝑡 (1 − 𝑚𝑡+1 ) 𝑏𝑡+1 + 𝑞𝐿,𝑡 (𝑚𝑡+1 𝑏𝑡+1 − (1 − 𝜆) 𝑚𝑡 𝑏𝑡 )

−𝜉 (∣(1 − 𝑚𝑡+1 ) 𝑏𝑡+1 ∣ + ∣𝑚𝑡+1 𝑏𝑡+1 − (1 − 𝜆) 𝑚𝑡 𝑏𝑡 ∣)

(2.9)
equation (2.9) captures the income sources and expenditures items of the firm
in the current period. The first row of equation (2.9) represents the firm’s inter-
nal funds net of financial repayment costs and investment expenditures, which in-
cludes the costs associated with capital adjustment. The second row includes the
firm’s revenue from debt issuances. The last row captures the exogenous cost of
new issuances of debt.
If a firm decides to default, then it will exit the market next period. A new firm
replaces the defaulting firm starting operations next period with zero liabilities,
low capital, and the lowest possible realizations for the productivity and fixed cost
shock.
Equity holders will not receive any residual value. Creditors with existing claims
will liquidate the defaulting firm, the total recovery value 𝑅(𝑧𝑡 , 𝑓𝑡 , 𝑘𝑡 ) after liquida-
tion is distributed among lenders using a pari-passu clause5 , this means that the
5
This clause establish that each creditor receive a fraction of 𝑅(𝑧𝑡 , 𝑓𝑡 , 𝑘𝑡 ) equal to the share of
claims that creditor had on the firm relative to the amount of debt that was defaulted.
 89

1
lender will receive 𝑏𝑡
𝑅(𝑧𝑡 , 𝑓𝑡 , 𝑘𝑡 ) for each unit of defaulted debt.
The liquidation value is proportional to the firm’s operative profits net of ad-
justment costs after sales of installed capital. Creditors will try to minimize the
adjustment costs from uninstalling capital when choosing the amount of capital to
liquidate. The creditors will only recover a fraction of 𝜒 of the liquidation value as
the liquidation process is costly. The total recovery value will be

̄ 𝑡 ) − 𝐴𝐶̄ 𝑘 (𝑘𝑡 ), 0} ,
𝑅(𝑧𝑡 , 𝑓𝑡 , 𝑘𝑡 ) = 𝜒 max{Π (𝑧𝑡 , 𝑓𝑡 , 𝑘𝑡 ) − 𝑘(𝑘 (2.10)

̄ 𝑡 ) = 𝑎𝑟𝑔 𝑚𝑖𝑛 𝐴𝐶𝑘 (𝑘𝑡+1 , 𝑘𝑡 )6 . The firm’s

where 𝐴𝐶̄ 𝑘 (𝑘𝑡 ) = 𝑚𝑖𝑛𝐴𝐶𝑘 (𝑘𝑡+1 , 𝑘𝑡 ) and 𝑘(𝑘
𝑘𝑡+1 𝑘𝑡+1
objective is to maximize expected discounted value of dividends conditional on the
information available at the current period subject to the non-negative dividend
constraint. If the firm discount rate is 𝛽, then its objective function will be

∞
E0 [ ∑ 𝛽𝑠 𝐷𝑡+𝑘 ] . (2.11)
𝑠=0

2.5.2 SOVEREIGN GOVERNMENT

Following Bocola (2016), the model includes a sovereign government in a reduced

form way. The government is an agent working in the background of the partial
equilibrium model. It will decide how much to spend on final goods that are fi-
nanced using lump-sum taxes and one-period bonds issued to lenders. Govern-
ment bonds are risky as each period the government can default on existing debt
1
levels. Let 𝑄𝑔 (𝑢𝑡 ) be the price of government bonds where 𝑄(𝑢)𝑔 = 1+𝑟𝑔 (𝑢𝑡 )
and
𝑟𝑔 (𝑢𝑡 ) is the interest rate of government bonds. The interest rate on government
6
Replacing equation (2.2) for the adjustment, the solution of the optimal minimization cost of
̄ 𝑡 ) = (1 − 𝛿) 𝑘𝑡 − 𝑘𝑡 and 𝐴𝐶
uninstalling capital will be 𝑘(𝑘 ̄ 𝑘 (𝑘𝑡 ) = 𝑘𝑡 .
𝜃 2𝜃
 90

bonds is an increasing function of a sovereign spread shock 𝑢𝑡 . The aggregate

sovereign spread shock follows an AR(1) process

𝑢∗
𝑢𝑡 = + (1 − 𝜌𝑔 ) 𝑢𝑡−1 + 𝜎𝑢 𝜖𝑡 , 𝜖𝑡 ∼ 𝑁(0, 1) , (2.12)
1 − 𝜌𝑔

where 𝑢∗ and 𝜎𝑢2 represents the mean and the variance of the aggregate sovereign
spread shock respectively.

2.5.3 INTERMEDIATE CREDITORS

The domestic financial sector is comprised of a large pool of perfectly competitive

intermediaries that lend resources to firms and the government. Financial interme-
diaries discount future profits using the interest rate for risky government bonds
1
1+𝑟𝑔 (𝑢𝑡 )
. To price firms’ risky debt, lenders use all relevant information on their de-
cisions, idiosyncratic shocks and aggregate shock. Let 𝑠𝑡 = (𝑚𝑡 , 𝑏𝑡 , 𝑘𝑡 ) be the vector
endogenous states and 𝑥𝑡 = (𝑧𝑡 , 𝑑𝑡 ) the vector of exogenous states that includes the
idiosyncratic and aggregate shock. Both 𝑥𝑡 and 𝑠𝑡 are sufficient for lenders to fore-
cast the future probability of default for the firm. As lenders are risk neutral and
perfectly competitive, the optimal pricing rules for short term and long term debt
satisfy the zero expected profit condition. Therefore 𝑞𝑆,𝑡 and 𝑞𝐿,𝑡 will be

⎡ ∣
𝑅(𝑧𝑡+1 ,𝑓𝑡+1 ,𝑘𝑡+1 ) ∣ ⎤
1
𝑞𝑆 (𝑥𝑡 , 𝑠𝑡+1 ) = E (1
1+𝑟(𝑢𝑡 ) 𝑥′ ⎢
− 𝑑𝑓 (𝑥𝑡+1 , 𝑓𝑡+1 , 𝑠𝑡+1 )) +𝑑𝑓 (𝑥𝑡 , 𝑓𝑡+1 , 𝑠𝑡+1 ) 𝑏𝑡+1 ∣𝑥 𝑡 ⎥ ,
∣
⎣ ∣ ⎦
(2.13)

1 ⎡
𝑞𝐿 (𝑥𝑡 , 𝑠𝑡+1 ) = E𝑥′ ⎢ 𝑓
⎢(1 − 𝑑 (𝑥𝑡+1 , 𝑓𝑡+1 , 𝑠𝑡+1 )) (𝜆+(1−𝜆)𝑐+(1−𝜆) 𝑞𝐿 (𝑥𝑡+1 , 𝑠𝑡+2
̂ ))
1 + 𝑟(𝑢𝑡 ) ⎢
⎣
 91

𝑅(𝑧𝑡+1 , 𝑓𝑡+1 , 𝑘𝑡+1 ) ∣∣ ⎤

+𝑑𝑓 (𝑥𝑡 , 𝑓𝑡+1 , 𝑠𝑡+1 ) ∣𝑥𝑡 ⎥ , (2.14)
𝑏𝑡+1 ∣
∣ ⎦

where 𝑑𝑓 (𝑥𝑡+1 , 𝑓𝑡+1 , 𝑠𝑡+1 ) is the optimal default policy of the firm for next period
taking the value of 1 if the firm default and 0 if the firm repay its debt commitments.
At equation (2.5.3) is clear that for long term debt the lender needs to forecast the
price of long term debt for two periods ahead 𝑞𝐿̂ (𝑥𝑡+1 , 𝑠𝑡+2
̂ ). In order to forecast long
term debt, the lender uses the firm’s policy functions on maturity, debt and capital
(𝑠𝑡+2
̂ ), which, by assumption, are perfectly observed by the lender. Notice that,
firms also can not commit to following a particular debt and investment plan that
will reduce the probability of default in the future. This additional commitment
problem is important for long term debt as the lender needs to consider the future
decisions of the firm to price debt correctly. As a result, long term debt will be
riskier and will have a higher premium relative to short term debt.

2.5.4 TIMING OF THE MODEL

It is time to describe the timing of the model. To simplify the notation the time sub-
script is omitted, to represent a value of a variable in the future prime superscripts
are used. The state of the firm in the current period is defined by its endogenous
state vector, and exogenous state vector (𝑥, 𝑓 , 𝑠) where 𝑥 = (𝑘, 𝑚, 𝑏) and 𝑠 = (𝑢, 𝑧).
The model divides the timing structure into five stages:

1. Beginning of the current period: Conditional on not defaulting in the pre-

vious period, the firm observes the realization of the idiosyncratic produc-
tivity and fixed cost shock (𝑧, 𝑓 ), and the aggregate sovereign spread shock
𝑢. The firm also knows the capital stock, debt level, and maturity level from
the decisions of the previous period 𝑠 = (𝑚, 𝑏, 𝑘). The complete state for the
 92

firm at the beginning of the current period is (𝑥, 𝑓 , 𝑠) = (𝑢, 𝑧, 𝑓 , 𝑚, 𝑘, 𝑏), where:
𝑥 = (𝑢, 𝑧) and 𝑠 = (𝑚, 𝑘, 𝑏).

2. Production stage: The firm pays the fixed cost, produces and receives Π(𝑧, 𝑓 , 𝑘)
as operating profits.

3. Repayment stage: The firm decides whether or not to default on its existing
debt levels. If the firm chooses to default, then lenders liquidate its remaining
assets and take the amount generated as operating profits. Next period, a new
firm will replace the defaulting one.

4. Continuation stage: If the firm decide not to default then it needs to pay (1 −
𝑚)𝑏+(𝜆+(1−𝜆) 𝑐) 𝑚 𝑏 as financial costs from past liabilities. After this the firm
is permitted to decide the optimal debt issuance, maturity, and investment
levels (𝑏′ , 𝑚′ , 𝑘 ′ ) subject to the non-negative dividend constraint and taking
as given the optimal prices of long and short term debt 𝑞𝑆 (𝑥, 𝑠′ ), 𝑞𝐿 (𝑥, 𝑠′ ).

5. End of the current period: The firm’s endogenous state for next period is
𝑠′ = (𝑘 ′ , 𝑏′ , 𝑚′ ).

2.5.5 FIRM’S PROBLEM

Following the timing of the model described in the last section, it will be convenient
to divide the problem of the firm into two sub-problems. The first sub-problem
is at the repayment stage, where the firm decides to default or not. The second
sub-problem will be at the continuation stage, where the firm chooses the optimal
levels of maturity, debt, and investment. Both problems are stated in the following
sections.
 93

REPAYMENT STAGE

Let V be the value of the firm, V𝑐 the value associated with repayment and V𝑑 the
value of defaulting on outstanding liabilities. At the repayment stage, the firm’s
problem will be

V(𝑥, 𝑓 , 𝑠) = max 1{𝐷<0} × V𝑑 + 1{𝐷≥0} × max {V𝑐 (𝑥, 𝑓 , 𝑠), V𝑑 } , (2.15)

𝑓
{𝑑 ={1,0}}

where 1{𝐷<0} is an indicator variable that takes the value of 1 if the firm satisfies the
non-negative dividend constraint and 0 otherwise. The problem at the repayment
stage determines the default policy of the firm 𝑑𝑓 (𝑥, 𝑓 , 𝑠). According to equation
(2.15), two reasons motivate firms’ default in this model. On the one hand, the
firm can default due to strategic reasons when the continuation value is lower than
the value of default. On the other hand, the firm will decide to default if there is
no available choice for maturity, debt, and capital that can satisfy the non-negative
dividend constrain. In the last case, the default is forced upon the firm by its share-
holders as the limited liability clause protects them from receiving negative divi-
dends. We know that after default equity holders walk away and do not receive
anything from the liquidation process then V𝑑 = 0, replacing this on the firm’s
problem at the repayment stage

V(𝑥, 𝑓 , 𝑠) = 1{𝐷≥0} × max {V𝑐 (𝑥, 𝑓 , 𝑠), 0} . (2.16)

CONTINUATION STAGE

If the firm decide not to default will move to the continuation stage where it can
choose the new levels for maturity, debt, and capital stock. The continuation value
 94

of the firm V𝑐 is defined by the bellman equation at the continuation stage problem,
the later is defined as follows

V𝑐 (𝑥, 𝑓 , 𝑠) = max {𝐷 + 𝛽 E[V(𝑥′ , 𝑓 ′ , 𝑠′ )∣𝑥]}

{𝑠′ =(𝑚′ ,𝑏′ ,𝑘′ )}

s.t 𝐷≥0,

𝑚′ ≥ 0 ,

𝑚′ ≤ 1 ,

𝑞𝐿 = 𝑞𝐿 (𝑥, 𝑠′ ), 𝑞𝑆 = 𝑞𝑆 (𝑥, 𝑠′ ),

where 𝐷 represent the dividends of the firms and V(𝑥′ , 𝑓 ′ , 𝑠′ ) is the value of the firm
defined at equation (2.16). The non-negative dividend condition will restrict the
firm’s choice set regarding the levels of maturity, debt, and capital. Additionally,
the upper and lower bounds on maturity prevent the firm from choosing maturity
levels below zero and above one. For a given price schedule, the firm will inter-
nalize that the choice regarding maturity, debt, and capital will adjust the price of
short term and long term debt.

DEFAULT AND REPAYMENT SETS

The firm’s problem at the repayment and continuation stage can be conveniently
expressed in terms of a default and continuation set. The repayment set ℭ(𝑥, 𝑠)
includes all the possible realizations of the fixed cost shock that makes repayment
optimal given (𝑥, 𝑠), this is presented as follows

⎧
{ ⎫
}
ℭ(𝑥, 𝑠)⎨𝑓 ∈ 𝐹 ∶ V𝑐 (𝑥, 𝑓 , 𝑠) ≥ 0 and 𝐷 ≥ 0⎬ . (2.17)
{
⎩ }
⎭
 95

The default set 𝔇(𝑥, 𝑠) includes all the different fixed cost shock realizations that
make default optimal in the current period given (𝑥, 𝑠)

𝔇(𝑥, 𝑠)𝐹 ℭ(𝑥, 𝑠) . (2.18)

We know that the firm will always default in a given period if there is no combina-
tion of next period maturity, debt, and capital levels that satisfy the non-negative
dividend constraint. The latter is true because the value of the firm is bounded
below by V𝑑 = 0. If the firm has at least one feasible alternative that satisfies the
non-negative dividend constraint, then the continuation value of the firm will be
weakly positive. The latter implies that only the non-negative constraint is needed
to define the continuation set.
𝐸𝐹
Let 𝐼 (𝑥, 𝑠) be the firm’s minimal investment expenditure that can be financed
with external funds, which is defined as follows

⎧
{ ′ 𝜃 𝑘 ′ − (1 − 𝛿)𝑘
2
𝐸𝐹
𝐼 (𝑥, 𝑠) = min ⎨𝑘 + ( ) 𝑘 − 𝑞𝑆 (𝑥, 𝑠′ )(1 − 𝑚′ )𝑏′
′ ′ ′ {
{𝑘 ,𝑏 ,𝑚 }⎩
2 𝑘
(2.19)
⎫
}
′ ′ ′ ′ ′ ′ ′
−𝑞𝐿 (𝑥, 𝑠 )(𝑚 𝑏 − (1 − 𝜆) 𝑚𝑏) + 𝜉 (∣(1 − 𝑚 )𝑏 ∣ + ∣𝑚 𝑏 − (1 − 𝜆)𝑚 𝑏∣)⎬ ,
}
⎭

notice that the last definition includes the adjustment costs of capital in addition
to the cost to purchase new units of capital. Also notice that equation (2.19) the
revenues generated from external funds are are net of exogenous issuing debt costs.
For a given state (𝑥, 𝑓 , 𝑠), the choice set of the firm is defined by all the combina-
tions of next period maturity, debt and capital levels combinations 𝑠′ = (𝑘 ′ , 𝑏′ , 𝑚′ )
that satisfy the non-negative dividend constraint. If the choice set of the firm is
empty then the firm will default with certainty in that period. The firm’s choice
 96

set will be non-empty if the following condition is satisfied weakly by at least one
𝑠′ = (𝑘 ′ , 𝑏′ , 𝑚′ )

𝐸𝐹
𝐴𝑧𝑘 𝛼 + (1 − 𝛿)𝑘 − 𝑓 − (1 − 𝑚)𝑏 − (𝜆 + (1 − 𝜆) 𝑐) 𝑚𝑏 − 𝐼 (𝑥, 𝑠) ≥ 0 , (2.20)

using equation (2.20) and the non-negative dividend condition we can define the
fixed cost threshold 𝑓 (𝑆, 𝑠) as the value of the fixed cost shock that satisfy equation
(2.20) with equality

𝐸𝐹
𝑓 (𝑥, 𝑠) = 𝐴𝑧𝑘 𝛼 + (1 − 𝛿)𝑘 − (1 − 𝑚)𝑏 − (𝜆 + (1 − 𝜆) 𝑐) 𝑚𝑏 − 𝐼 (𝑥, 𝑠) , (2.21)

For a given state (𝑥, 𝑓 , 𝑠), the firm will default if an only if the realization of the
fixed cost shock is strictly greater than the threshold 𝑓 > 𝑓 (𝑥, 𝑠) and will continue
otherwise 𝑓 ≤ 𝑓 (𝑥, 𝑓 , 𝑠). Therefore, using equation (2.21) we can redefine the con-
tinuation set in terms of the default threshold as follows

⎧
{ ⎫
}
ℭ(𝑥, 𝑠)⎨𝑓 ∈ 𝐹 ∶ 𝑓 ≤ 𝑓 (𝑥, 𝑠)⎬ , (2.22)
{
⎩ }
⎭

with the continuation set defined in terms of the fixed cost threshold, it is simple
to solve the problem of the firm at the repayment stage, its solution will define the
default policy of the firm as follows

⎧
{
𝑓 { 1 if 𝑓 ≤ 𝑓 (𝑥, 𝑠)
𝑑 (𝑥, 𝑓 , 𝑠) = ⎨ (2.23)
{
{
⎩0 if 𝑓 > 𝑓 (𝑥, 𝑠) .
 97

COMPLETE FORMULATION OF THE FIRM PROBLEM

Using the default policy at (2.23) we can redefine the problem of the firm at the
continuation stage as follows

⎧
{ 𝑓 (𝑥′ ,𝑠′ ) ⎫
{ ⎡ ∣ ⎤}
}
V𝑐 (𝑥, 𝑓 , 𝑠) = max ⎨𝐷 + 𝛽 E𝑆′ ⎢ ′ ′ ′ ′ ∣ ⎥
⎢ ∫ V𝑐 (𝑥 , 𝑓 , 𝑠 ) 𝑑𝐺(𝑓 ) ∣𝑆⎥⎬
{𝑚′ ,𝑏′ ,𝑘′ }{
⎢ ∣ ⎥}
{ 0 ∣ ⎦}
⎩ ⎣ ⎭
s.t 𝐷≥0, (2.24)

𝑚′ ≥ 0 ,

𝑚′ ≤ 1 ,

𝑞𝐿 = 𝑞𝐿 (𝑥, 𝑠′ ), 𝑞𝑆 = 𝑞𝑆 (𝑥, 𝑠′ ) .

Replacing the default policy on equations (2.25) and (2.26) we will get the prices
of short and long term as a function of the default threshold

+∞ ∣
1 ⎡ 𝑅(𝑧′ , 𝑓 ′ , 𝑘 ′ ) ′ )∣𝑥⎤ ,
𝑞𝑆 (𝑥, 𝑠′ ) = ′ ′
E𝑥′ ⎢𝐺(𝑓 (𝑥 , 𝑠 )) + ∫ 𝑑𝐺(𝑓 ∣ ⎥ (2.25)
1 + 𝑟𝑔 (𝑢) 𝑏′ ∣
⎣ ′ ′
𝑓 (𝑥 ,𝑠 ) ∣ ⎦

1 ⎡
𝑞𝐿 (𝑥, 𝑠′ ) = E ′ ⎢(𝜆 + (1 − 𝜆)𝑐)𝐺(𝑓 (𝑥′ , 𝑠′ ))
1 + 𝑟𝑔 (𝑢) 𝑥
⎣
𝑓 (𝑥′ ,𝑠′ ) ∣
+∞
𝑅(𝑧′ , 𝑓 ′ , 𝑘 ′ ) ∣
′ )∣𝑥⎤ ,
+ (1 − 𝜆) ∫ 𝑞𝐿̂ (𝑥′ , 𝑠″̂ ) 𝑑𝐺(𝑓 ′ ) + ∫ 𝑑𝐺(𝑓 ∣ ⎥
𝑏′ ∣ ⎦
0 𝑓 (𝑥′ ,𝑠′ )
∣
(2.26)

where 𝐺(𝑓 (𝑥′ , 𝑠′ )) is the probability of repayment and 𝑠′̂ = (𝑚′ (𝑥, 𝑓 , 𝑠), 𝑏′ (𝑥, 𝑓 , 𝑠), 𝑘 ′ (𝑥, 𝑓 , 𝑠))
is the vector formed with the optimal policy functions for maturity, debt and capi-
+∞
𝑅(𝑧′ ,𝑓 ′ ,𝑘′ )
tal. Notice that recovery value is integrated over the default set ∫ 𝑏′
𝑑𝐺(𝑓 ′ ),
𝑓 (𝑥′ ,𝑠′ )
 98

and the continuation value for long term debt is integrated over the continuation
𝑓 (𝑥′ ,𝑠′ )
set (1 − 𝜆) ∫ 𝑞𝐿̂ (𝑥′ , 𝑠″̂ ) 𝑑𝐺(𝑓 ′ ).
0

2.5.6 RECURSIVE PARTIAL COMPETITIVE EQUILIBRIUM

A recursive partial equilibrium for this economy will be defined by a set of value
functions {V(𝑥, 𝑓 , 𝑠), V𝑐 (𝑥, 𝑓 , 𝑠)}, policy functions {𝑘(𝑥, 𝑓 , 𝑠), 𝑏(𝑥, 𝑓 , 𝑠), 𝑚(𝑥, 𝑓 , 𝑠), 𝑑𝑓 (𝑥, 𝑓 , 𝑠)}
and price functions {𝑞𝑆 (𝑥, 𝑠′ ), 𝑞𝐿 (𝑥, 𝑠′ )} such that:

• Given {𝑞𝑆 (𝑥, 𝑠′ ), 𝑞𝐿 (𝑥, 𝑠′ )}

1. The value functions {V, V𝑐 } satisfies (2.16) and (2.24) respectively.

2. The default policy 𝑑𝑓 (𝑥, 𝑓 , 𝑠) is the solution of the repayment problem of

the firm (i.e satisfies (2.23)).

3. The optimal capital structure {𝑏(𝑥, 𝑓 , 𝑠), 𝑚(𝑥, 𝑓 , 𝑠)} and the investment
policy function {𝑘(𝑥, 𝑓 , 𝑠)} are the solution of the continuation problem
of the firms defined by (2.24).

• Given {V(𝑥, 𝑓 , 𝑠), V𝑐 (𝑥, 𝑓 , 𝑠)} and {𝑘(𝑥, 𝑓 , 𝑠), 𝑏(𝑥, 𝑓 , 𝑠), 𝑚(𝑥, 𝑓 , 𝑠), 𝑑𝑓 (𝑥, 𝑓 , 𝑠)}, the
pricing functions {𝑞𝑆 (𝑥, 𝑠′ ), 𝑞𝐿 (𝑥, 𝑠′ )} satisfies the zero expected profit condi-
tion for financial creditors (i.e satisfies (2.25), and (2.26) respectively).

2.5.7 CHARACTERIZING THE OPTIMAL DECISIONS OF THE FIRM

To characterize the optimal decisions of the firm, we will assume that the continua-
tion value of the firm and the debt pricing functions are continuous, differentiable.
Moreover, we will assume that the first-order conditions are sufficient to charac-
terize the optimal decisions of firms. To simplify the problem we will consider
 99

the case with no capital adjustment costs (i.e 𝜃 = 0), and no aggregate risk (i.e
𝑟
𝑟(𝑢) = 𝑟 > 0). The coupon 𝑐 will be equal to 1+𝑟
which implies that in the ab-
1
sence of risk default for the firm 𝑞𝑆 = 𝑞𝐿 = 1+𝑟
. There is no recovery for lenders
after firms’ default (i.e 𝜒 = 0). With these assumptions in mind, the continuation
problem of the firm will be

⎧
{ 𝑓 (𝑥′ ,𝑠′ ) ⎫
{ ⎡ ∣ ⎤}
}
V𝑐 (𝑥, 𝑓 , 𝑠) = max ⎨ 𝐷 + 𝛽 E𝑥′ ⎢ ′ ′ ′ ′ ∣ ⎥
⎢ ∫ V𝑐 (𝑥 , 𝑓 , 𝑠 ) 𝑑𝐺(𝑓 ) ∣𝑥⎥⎬
{𝑠′ =(𝑚′ ,𝑏′ ,𝑘′ )}{
⎢ ∣ ⎥}
{ 0 ∣ ⎦}
⎩ ⎣ ⎭
s.t 𝐷 ≥ 0 (𝜆𝐷 ) ,

𝑚′ ≥ 0 (𝜆𝑚0 ) ,

𝑚′ ≤ 1 (𝜆𝑚1 ) ,

where 𝜆𝐷 is the shadow value of internal funds, 𝜆𝑚0 is the multiplier for the lower
bound of maturity and 𝜆𝑚1 is the multiplier for the upper bound of maturity. The
first order condition for capital is as follows

⎡ 𝜕𝑞𝑆 (𝑥, 𝑠′ ) ′ ⎤
(1 + 𝜆𝐷 )⎢1 − (1 − 𝑚 ′ )𝑏′ − 𝜕𝑞𝐿 (𝑥, 𝑠 ) (𝑚′ 𝑏′ − (1 − 𝜆)𝑚𝑏)⎥ =
⎢ ⎥
⎢ 𝜕𝑘 ′ 𝜕𝑘 ′ ⎥
⎣ ⎦ (2.27)
𝑓 (𝑆′ ,𝑠′ )
⎡ 𝜕𝑓 (𝑥′ , 𝑠′ ) ∣ ⎤
𝛽E𝑥′ ⎢ ′ ′ ′𝛼−1 + 1 − 𝛿)𝑑𝐺(𝑓 ′ ) + V (𝑓 )𝑔(𝑓 )
⎢ ∫ (1 + 𝜆𝐷 )(𝛼𝐴𝑧 𝑘 𝑐
∣𝑥 ⎥
∣ ⎥
⎢ 𝜕𝑘 ′ ∣ ⎥
⎣
0 ∣ ⎦

𝜕𝐺(𝑓 (𝑥′ ,𝑠′ ))

where V𝑐 (𝑓 ) = V𝑐 (𝑥′ , 𝑓 (𝑥′ , 𝑠′ ), 𝑠′ ) and 𝑔(𝑓 ) = 𝜕𝑘′
. The left hand side of
equation (2.27) capture the marginal costs of an additional unit of capital. The
marginal cost must consider the effect of that additional unit capital the prices for
short term and long term debt prices. It is possible to show that an increase in the
capital stock will reduce the default risk of the firm. Differentiating the fixed cost
 100

threshold at (2.21) with respect to capital

𝜕𝑓 (𝑥′ , 𝑠′ )
= 𝛼 𝐴𝑧′ 𝑘 𝛼−1 + (1 − 𝛿) > 0, (2.28)
𝜕𝑘 ′

clearly, an additional unit of capital will increase the fixed cost threshold, then the
probability of default in the next period will reduce.
The right-hand side of equation (2.27) captures the expected marginal benefit
of an additional unit of capital. Additional investment increases the continuation
value of the firm in two ways. First, it will increase operating profits in the next
period by 𝛼 𝐴𝑧 𝑘 𝛼−1 + (1 − 𝛿). Second, it will reduce the probability of default in the
next period allowing the firm to receive V𝑐 in additional states in which the firm
chose to default.
For maturity the first order condition will be

⎡ 𝜕𝑞𝑆 (𝑆′ , 𝑠′ ) ′ ⎤
(1 + 𝜆𝐷 )⎢ ′ ′ )𝑏′ − 𝜕𝑞𝐿 (𝑥, 𝑠 ) (𝑚′ 𝑏′ − (1 − 𝜆)𝑚𝑏)⎥ + 𝜆
⎢(𝑞𝑆 − 𝑞𝐿 )𝑏 − (1 − 𝑚 ⎥ 𝑚1 − 𝜆𝑚0 =
⎢ 𝜕𝑚′ 𝜕𝑚′ ⎥
⎣ ⎦
𝑓 (𝑥′ ,𝑠′ )
⎡ 𝜕𝑓 (𝑥′ , 𝑠′ ) ∣
⎢ ′ ′ ′ ′ ″ ′
𝛽E𝑥′ ⎢ ∫ (1 + 𝜆𝐷 )𝑏 (1 − 𝜆)(1 − 𝑐 − 𝑞𝐿 (𝑥 , 𝑠 ̂ ) + 𝜉 )𝑑𝐺(𝑓 ) + V𝑐 (𝑓 )𝑔(𝑓 ) ∣𝑥 ,
∣
⎢ 𝜕𝑚′ ∣
⎣
0 ∣
(2.29)

The right-hand side of equation (2.29) captures the expected marginal benefit of
increasing maturity. For a fixed level of 𝑏′ , if the firm issues an additional unit of
long-term debt, the coupon repayment costs will increase by 𝑐 and will reduce the
rollover costs by −𝑞′𝐿 + 𝜒. Overall, each unit of long-term debt will reduce next-
period dividends by 1 − 𝑐 − 𝑞𝐿 + 𝜒 in the states that the firm chooses not to default.
Additionally, increasing maturity will also change the fixed cost threshold of the
 101

firm. This is easy to see by partially differentiating equation (2.21)

𝜕𝑓 (𝑥′ , 𝑠′ )
= 𝑏′ (1 𝜆)(1 − 𝑐 − 𝑞′𝐿 + 𝜉 ) > 0 , (2.30)
𝜕𝑚′

𝑟
at the beginning of this section we assume that 𝑐 = 1+𝑟
, it is also known that 𝑞𝐿 ≤
1
1+𝑟
; then 1 − 𝑐 − 𝑞′𝐿 + 𝜉 > 0. Consequently, higher levels of maturity reduce the
probability of default in the next period, and increase the expected marginal benefit
by having additional states in which the firm receives the continuation value V𝑐 .
The left-hand side of equation (2.29) captures the marginal costs from increas-
ing maturity marginally. Three terms affect the marginal costs of maturity. First,
the opportunity costs generated by substituting short term debt for long term debt,
this is captured by the term (𝑞𝑆 − 𝑞𝐿 ) 𝑏′ . Second, maturity will affect the price of
short term and long term debt by changing the fixed cost threshold of the firm and
its default risk. Using equation (2.30) we can find the partial effect of maturity on
the price of short term debt and long term debt

𝜕𝑞𝑆 (𝑥, 𝑠′ ) 1 ⎡ 𝜕𝑓 (𝑥′ , 𝑠′ ) ⎤

= E ′ ⎢𝑔(𝑓 ) ⎥, (2.31)
𝜕𝑚′ 1+𝑟 𝑥 𝜕𝑚′
⎣ ⎦

𝑓 (𝑥′ ,𝑠′ )
𝜕𝑞𝐿 (𝑥, 𝑠′ ) 1 ⎡ 𝜕𝑓 (𝑥′ , 𝑠′ ) 𝜕𝑞′𝐿̂ 𝜕𝑘 ″ 𝜕𝑞′𝐿̂ 𝜕𝑏″ 𝜕𝑞′𝐿̂ 𝜕𝑚″ ⎤
= E ′ ⎢𝜅𝑔(𝑓 ) +(1−𝜆) ∫ ( + ″ + ″ )𝑑𝐺(𝑓 ′ )⎥,
𝜕𝑚′ 1+𝑟 𝑥 𝜕𝑚′ ″ ′
𝜕𝑘 𝜕𝑚 𝜕𝑏 𝜕𝑚 𝜕𝑚 𝜕𝑚 ′ ′
⎣ 0 ⎦
(2.32)
where 𝜅 = 𝜆 + (1 − 𝜆) 𝑐 < 1. It is clear from equation (2.31) that higher maturity by
reducing the probability of default in the next period will increase the price of short
term debt; for long-term debt, this is not so obvious. By inspecting equation (2.32)
it is clear that maturity affects the price of long-term debt in the same way that the
price of short term debt by reducing the probability of default next period, but this
 102

effect is proportionally lower (𝜅 < 1), this is because default in the short run and
long run are relevant when pricing long-term debt as only a fraction 𝜆 (1 − 𝑚) of
the total outstanding liabilities mature in the current period.
Notice that equation (2.32) includes an additional term. This term captures the
effect of maturity on the long-run default risk; its sign is unknown as it depends
on the shape of the policy functions for maturity, debt, and capital. For this reason,
the effect of maturity on long term debt is ambiguous.
Finally, the first order condition for debt will be

⎡ 𝜕𝑞𝑆 (𝑆′ , 𝑠′ ) ′ ′ ⎤
(1 + 𝜆𝐷 )⎢ ′ ′ )𝑏′ + 𝑞 𝑚′ − 𝜕𝑞𝐿 (𝑆 , 𝑠 ) (𝑚′ 𝑏′ − (1 − 𝜆)𝑚𝑏) + 𝜉 ⎥ =
⎢𝑞𝑆 (1 − 𝑚 ) − (1 − 𝑚 𝐿 ⎥
⎢ 𝜕𝑏′ 𝜕𝑏′ ⎥
⎣ ⎦
𝑓 (𝑥′ ,𝑠′ )
⎡ 𝜕𝑓 (𝑥′ , 𝑠′ ) ∣ ⎤
⎢ ′ ′ ′ ′ ″ ′
𝛽E𝑥′ ⎢ ∫ (1 + 𝜆𝐷 )(1 − 𝑚 (1 − 𝜆) ((1 − 𝑐 − 𝑞𝐿 (𝑥 , 𝑠 ̂ ) + 𝜉 )))𝑑𝐺(𝑓 ) − V𝑐 (𝑓 )𝑔(𝑓 ) ∣𝑥 ⎥ ,
′ ∣ ⎥
⎢ 𝜕𝑏 ∣ ⎥
⎣
0 ∣ ⎦
(2.33)

The right-hand side of equation (2.33) captures the expected marginal costs from
increasing debt. The additional unit of debt will reduce future dividends because
of repayment costs in states where the firms do not default. If the additional unit
of debt is in the form of long term debt, only a fraction of 𝜆 must be repaid next
period. The fraction (1 − 𝜆) will be subject to the usual benefits of rollover costs
changes (𝑚′ (1 − 𝜆)(1 − 𝑐 − 𝑞′𝐿 + 𝜉 )). Overall, the marginal cost will change by 1 −
𝑚′ (1−𝜆)(1−𝑐−𝑞′𝐿 −𝜉 ) at each state where the firm does not default. The additional
unit of debt also changes the fixed cost threshold of the firm, using equation (2.21)
and differentiating with respect to debt

𝜕𝑓 (𝑥′ , 𝑠′ )
= −1 + 𝑚′ (1 − 𝜆)(1 − 𝑐 − 𝑞′𝐿 + 𝜉 ), (2.34)
𝜕𝑏′
 103

𝑟
as long as 𝜉 ≤ 1+𝑟
, the fixed cost threshold will reduce, and the probability of
default will increase with an additional unit of debt. In this case, the marginal
costs of the firm increase because there are additional states where the firm will
default. There will be an opportunity cost in terms of the continuation value loses
in states where previously the firm decided to repay.
The left-hand side of equation (2.33) captures the marginal revenue from an
additional unit of debt issued in the current period. By increasing debt in one ad-
ditional unit the firm receives 𝑞𝑆 (1 − 𝑚′ ) + 𝑞𝐿 𝑚′ . However, higher debt levels will
also increase the default risk of the firm and threfore reduce the revenues from
𝜕𝑞𝑆 (𝑆′ ,𝑠′ )
the one additional unit of debt, this is captured by the term − 𝜕𝑏′
(1 − 𝑚′ )𝑏′ −
𝜕𝑞𝐿 (𝑆′ ,𝑠′ )
𝜕𝑏′
(𝑚′ 𝑏′ − (1 − 𝜆)𝑚𝑏). Using equation (2.34), we can derive the effect of debt
on the prices of short-term and long-term debt

𝜕𝑞𝑆 (𝑥, 𝑠′ ) 1 ⎡ 𝜕𝑓 (𝑥′ , 𝑠′ ) ⎤

= E ′ ⎢𝑔(𝑓 ) ⎥, (2.35)
𝜕𝑏′ 1+𝑟 𝑥 𝜕𝑏′
⎣ ⎦

𝑓 (𝑥′ ,𝑠′ )
𝜕𝑞𝐿 (𝑥, 𝑠′ ) 1 ⎡ 𝜕𝑓 (𝑥′ , 𝑠′ ) 𝜕𝑞′𝐿̂ 𝜕𝑘 ″ 𝜕𝑞′𝐿̂ 𝜕𝑏″ 𝜕𝑞′𝐿̂ 𝜕𝑚″ ⎤
= E ′ ⎢𝜅𝑔(𝑓 ) +(1−𝜆) ∫ ( + ″ ′+ ″ )𝑑𝐺(𝑓 ′ )⎥,
𝜕𝑏′ 1+𝑟 𝑥 𝜕𝑏′ ″ ′
𝜕𝑘 𝜕𝑏 𝜕𝑏 𝜕𝑏 𝜕𝑚 𝜕𝑏 ′
⎣ 0 ⎦
(2.36)
equations (2.35) and (2.36) are similar to the ones derived for maturity. The price
of short-term and long-term debt will reduce due to the increase in the probability
of default next-period; for long term debt, this effect is lower as 𝜅 < 1. Moreover,
default is more likely in the long term if the firm’s additional debt is in the form
of long-term debt, this effect is captured by the second term of equation (2.36), its
sign is undetermined as it depends on the shape of the policy functions of the firm.
 104

2.5.8 INSPECTING THE MECHANISM OF THE MODEL

The model presented earlier can capture the empirical results presented in previous
sections. To understand why this is true, we need to differentiate a firm that faces
default risk in the short-run, and the long run. There is no short-run and long-
run default risk for the firm if it will not default with certainty in the next period
and every other period in the future. A firm has only long-run default risk if, with
certainty, it will not default in the next period, but there is a positive probability
that will default in other periods in the future. Finally, a firm has short-run and
long-run risks if there is a positive probability that the firm will default in future
periods.
When a firm is not subject to any default risk, then capital should be equal to
the frictionless level. The firm, in this case, will hold all its liabilities as long term
debt, given that prices of both types of debt are equal but long term debt provides
gains in terms of lower rollover costs (i.e., 𝜉 > 0). An increase in the sovereign
spread should only affect the amount of debt issued that period, which in turn will
depend on the needs of external funds to finance the frictionless capital level for
the next period. The firm will not adjust maturity either.
If the firm is only subject to long-run risk, we know that it will not default with
certainty in the next period. Therefore, the price of short term debt will be constant
1
and equal to 1+𝑟
. This is not the case for long-term debt. If the firm reduces matu-
rity, the total amount of debt will move to an option that bears no risk of default.
Maturity will be lower than one, but higher than zero as issuing long term debt
has positive gains in terms of lower exogenous and endogenous rollover costs. The
capital will be lower than the frictionless one. As the firm increases in size, and its
capital is closer to the first-best, the level of maturity will rise to one. An increase in
 105

the sovereign spread will create incentives for firms with high maturity to substi-
tute long-term debt with short-term debt. The long-run default risk of the firm will
be lower, and the price of long term debt will be lower. For low maturity firms, this
is the opposite, they face more liquidity risk relative to high maturity firms, so the
increase in the sovereign spread will create incentives to increase maturity. Long-
term default risk will be higher, and the price of long term debt will be lower. Low
maturity firms, in this case, will need more resources to roll over their debt. Will
move its liability portfolio towards an option that has default risk. This will mean
that firms with high maturity levels will cut their investment in a lower magnitude
relative to low maturity firms.
Finally, for firms that face both short term and long term default risk, the sit-
uation will be opposite to the previous case. Firms with high maturity will have
a higher probability of default in the long term relative to low maturity firms. An
increase in the sovereign spread will create incentives for low maturity firms to is-
sue more long term debt as a way to reduce rollover risk. If maturity is sufficiently
low for these firms, the probability of default in the future should not increase dra-
matically. Firms with high maturity will not have the same option. The issuances
of long term debt will drastically increase the cost of external funds. In turn, is-
sue more short term debt will increase default risk short-run, which is already a
problem that is faced by firms in this group. For high maturity firms, the choice is
to deleverage and cut the size of its operations, which means that investment will
reduce by more compare to the case of firms with high maturity.
It is clear from the above arguments that empirical evidence is consistent with
this model. If high leverage firms are the ones with both long term and short term
default risk and low leverage firms are the ones with only short-run default risk or
no default risk at all, the increase in the sovereign spread will make firms with at
 106

most short-run default risk to cut investment in a lower amount if the firm has high
maturity levels relative to a firm with low maturity levels. Moreover, if the firm has
both short term and long term default risk, then an increase in the sovereign spread
will invest in reducing more strongly for firms with high maturity levels.

2.6 CONCLUSIONS AND FURTHER WORK

The maturity and leverage decisions of the firm define its capital structure. If fi-
nancial frictions are present, the capital structure will have a non-trivial role for
the future investment plans of the firm. The paper studied the way that the cap-
ital structure can affect the response of investment to aggregate financial shocks
focusing in particular in shocks to the sovereign spread. Using firm-level data for
Italy during the period 2007-2015, the paper showed that the response of firms’ in-
vestment to changes in the sovereign spread differs depending on the choices of
leverage and maturity made by the firm. For high leverage firms, an increase of the
sovereign spread by 100 basis points is associated with a decrease of 0.59% in the
growth rate of capital for high maturity firms relative to low maturity firms. For
low leverage firms, the same increase in the sovereign spread is associated with an
increment of 0.38% in the growth rate of capital for high maturity firms relative to
low maturity firms.
Moreover, differences in maturity will change the magnitude of the response of
firms’ investment to changes in the sovereign spread when comparing high lever-
age versus low leverage firms. For firms with low maturity, an increase in the
sovereign spread by 100 basis points is associated with a drop of 1.41% in the
growth rate of capital for firms with high leverage relative to firms with low lever-
age. For firms with high maturity, the same increase in the sovereign spread is
 107

associated with a drop of 2.41% for the growth rate of capital for firms with high
leverage relative to low leverage firms. The empirical results are robust to the inclu-
sion of additional firm-level and aggregate controls, changes in the measurement
definition for investment, and changes in the sample.
The paper also built a partial equilibrium model with exogenous sovereign spread
shocks where firms can issue risky long term and short term debt. The characteriza-
tion capital structure and investment decisions of the firm in this setup showed that
it is possible to understand the empirical results for Italian firms using the model. If
a firm is subject to long-run and short-run default risk and has low maturity levels,
an aggregate sovereign spread shock will make the firm to increase maturity. This
will reduce default risk in the short run without affecting too much the default risk
in the long-run. The firm, in this case, will not adjust its investment levels so much
compared to a firm that holds high levels of maturity. If the firm faces only long-
run default risk, then a sovereign spread shock will make firms with high maturity
to change their liabilities portfolio from risky long term debt towards risk-free short
term debt. The latter means that maturity will reduce and the problems associated
with long-run default risk will decline. In this case, the firm will reduce investment
by less compared to a firm with low maturity levels.
In terms of the work to be done, many areas to be explored. In the empirical
part, the robustness of the regression will be tested again in multiple ways. First,
by changing the main dummy variables for maturity and leverage with continuous
variables. Second, by changing the sample period. Third, by changing the defi-
nitions used for long-term and short-term debt, the new definitions will include
other items of the current and non-current liabilities for the firm. Lastly, it will be
interesting to check how this result holds for other EU countries that were affected
by the sovereign spread crisis as well. Spain and Portugal are good candidates for
 108

this robustness check. Besides improving the robustness of the model, the future
work will add dynamics to the estimation strategy. This can be done with the use
of local projections methods (Jorda (2005)) to find the impulse response function
for investment after a sovereign spread shock.
The model is where more work needs to be done. First, a calibration strategy
needs to be specified; in particular, a way to estimate the process for the sovereign
spread shock must be proposed. Second, the model needs to fix some moments
in the firm-level data for Italy that the model wants to match. Third, a solution
algorithm must be defined to find a quantitative solution for the model. Once the
model is calibrated, it will be used to simulate artificial firm-level data. The data
generated by the model will then be used to check how good the model reproduces
the empirical regressions of the paper.
 109

BIBLIOGRAPHY

Adelino, Manuel, Antoinette Schoar, and Felipe Severino (2015) “House prices, collateral,

and self-employment,” Journal of Financial Economics, 117 (2), 288–306.

Alesina, Alberto and Matteo Paradisi (2017) “Political budget cycles: Evidence from Italian

cities,” Economics & Politics, 29 (2), 157–177.

Arellano, Cristina, Yan Bai, and Luigi Bocola (2019) “Sovereign default risk and firm het-

erogeneity,” NBER Working Paper (23314).

Asquith, Paul and David W Mullins Jr (1986) “Signalling with dividends, stock repur-

chases, and equity issues,” Financial management, 27–44.

Bahaj, Saleem, Angus Foulis, and Gabor Pinter (2020) “Home values and firm behavior,”

American Economic Review, 110 (7), 2225–70.

Bahaj, Saleem, Angus Foulis, Gabor Pinter, and Paolo Surico (2022) “Employment and the

residential collateral channel of monetary policy,” Journal of Monetary Economics.

Berger, David, Veronica Guerrieri, Guido Lorenzoni, and Joseph Vavra (2018) “House

prices and consumer spending,” The Review of Economic Studies, 85 (3), 1502–1542.

Bloom, Nicholas (2009) “The impact of uncertainty shocks,” econometrica, 77 (3), 623–685.

Chaney, Thomas, David Sraer, and David Thesmar (2012) “The collateral channel: How
 110

real estate shocks affect corporate investment,” American Economic Review, 102 (6), 2381–

2409.

Chodorow-Reich, Gabriel, Plamen T Nenov, and Alp Simsek (2021) “Stock market wealth

and the real economy: A local labor market approach,” American Economic Review, 111

(5), 1613–57.

Christelis, Dimitris, Dimitris Georgarakos, Tullio Jappelli, and Maarten van Rooij (2020)

“Consumption uncertainty and precautionary saving,” Review of Economics and Statistics,

102 (1), 148–161.

DeAngelo, Harry, Linda DeAngelo, and Rene M Stulz (2010) “Seasoned equity offerings,

market timing, and the corporate lifecycle,” Journal of financial economics, 95 (3), 275–295.

Giroud, Xavier and Holger M Mueller (2017) “Firm leverage, consumer demand, and em-

ployment losses during the great recession,” The Quarterly Journal of Economics, 132 (1),

271–316.

Gopinath, Gita, Şebnem Kalemli-Özcan, Loukas Karabarbounis, and Carolina Villegas-

Sanchez (2017) “Capital allocation and productivity in South Europe,” The Quarterly

Journal of Economics, 132 (4), 1915–1967.

Guren, Adam M, Alisdair McKay, Emi Nakamura, and Jón Steinsson (2021) “Housing

wealth effects: The long view,” The Review of Economic Studies, 88 (2), 669–707.

Hansen, Benjamin, Joseph J Sabia, and Daniel I Rees (2017) “Have cigarette taxes lost their

bite? New estimates of the relationship between cigarette taxes and youth smoking,”

American Journal of Health Economics, 3 (1), 60–75.

Hubbard, R Glenn and Christopher J Mayer (2009) “The mortgage market meltdown and

house prices,” The BE Journal of Economic Analysis & Policy, 9 (3).

 111

Imbens, Guido W and Donald B Rubin (2015) Causal inference in statistics, social, and biomed-

ical sciences: Cambridge University Press.

Jordà, Òscar (2005) “Estimation and inference of impulse responses by local projections,”

American economic review, 95 (1), 161–182.

Kalemli-Özcan, Ṣebnem, Luc Laeven, and David Moreno (2018) “Debt overhang, rollover

risk, and corporate investment: Evidence from the european crisis,”Technical report, Na-

tional Bureau of Economic Research.

Kuehn, Lars-Alexander and Lukas Schmid (2014) “Investment-based corporate bond pric-

ing,” The Journal of Finance, 69 (6), 2741–2776.

Lechner, Michael et al. (2011) The estimation of causal effects by difference-in-difference methods,

4: Now Hanover, MA.

Levinsohn, James and Amil Petrin (2003) “Estimating production functions using inputs

to control for unobservables,” The review of economic studies, 70 (2), 317–341.

Levitin, Adam J (2009) “Worsening Foreclosure Crisis: Is It Time to Reconsider Bankruptcy

Reform?: Hearing Before the Subcomm. on Administrative Oversight and the Courts of

the S. Comm. on the Judiciary, 111th Cong., July 23, 2009 (Statement of Adam J. Levitin,

Associate Prof. of Law, Geo. UL Center).”

Liu, Zheng, Jianjun Miao, and Tao Zha (2016) “Land prices and unemployment,” Journal

of Monetary Economics, 80, 86–105.

Liu, Zheng, Pengfei Wang, and Tao Zha (2013) “Land-price dynamics and macroeconomic

fluctuations,” Econometrica, 81 (3), 1147–1184.

Marsh, Tanya D (2011) “Too big to fail vs. too small to notice: Addressing the commercial

real estate debt crisis,” Ala. L. Rev., 63, 321.

 112

Mian, Atif, Kamalesh Rao, and Amir Sufi (2013) “Household balance sheets, consumption,

and the economic slump,” The Quarterly Journal of Economics, 128 (4), 1687–1726.

Mian, Atif and Amir Sufi (2014) “What explains the 2007–2009 drop in employment?”

Econometrica, 82 (6), 2197–2223.

Oliviero, Tommaso and Annalisa Scognamiglio (2019) “Property tax and property values:

Evidence from the 2012 Italian tax reform,” European Economic Review, 118, 227–251.

Palmén, Olli (2020) “Sovereign default risk and credit supply: Evidence from the euro

area,” Journal of International Money and Finance, 109, 102257.

Panel, Congressional Oversight (2010) “Commercial real estate losses and the risk to finan-

cial stability,” February Oversight Report.

Pellegrino, Bruno and Luigi Zingales (2017) “Diagnosing the Italian disease,”Technical re-

port, National Bureau of Economic Research.

Piazzesi, Monika and Martin Schneider (2016) “Housing and macroeconomics,” Handbook

of macroeconomics, 2, 1547–1640.

Poeschl, Johannes (2020) “Corporate debt maturity and investment over the business cy-

cle,” Available at SSRN 3796066.

Rubin, Donald B (1978) “Bayesian inference for causal effects: The role of randomization,”

The Annals of statistics, 34–58.

Saiz, Albert (2010) “The geographic determinants of housing supply,” The Quarterly Journal

of Economics, 125 (3), 1253–1296.

Sgherri, Ms Silvia and Mrs Hanan Morsy (2010) After the Crisis: Assessing the Damage in

Italy: International Monetary Fund.

 113

Surico, Paolo and Riccardo Trezzi (2019) “Consumer spending and property taxes,” Journal

of the European Economic Association, 17 (2), 606–649.

Titman, Sheridan and Sergey Tsyplakov (2007) “A dynamic model of optimal capital struc-

ture,” Review of Finance, 11 (3), 401–451.

Wing, Coady, Kosali Simon, and Ricardo A Bello-Gomez (2018) “Designing difference in

difference studies: best practices for public health policy research,” Annual review of public

health, 39.

Wooldridge, Jeffrey M (2009) “On estimating firm-level production functions using proxy

variables to control for unobservables,” Economics letters, 104 (3), 112–114.

114
 115

APPENDIX

A TRADABLE AND NON-TRADABLES INDUSTRIES

TABLE A.1: Non-Tradable NACE Industries

Division Name Section TradeE TradeY HHI

19 Manuf. coke & petroleum C 595,208 0.31 0.03
20 Manuf. chemicals C 487,905 0.79 0.013
29 Manuf. vehicles C 336,130 0.79 0.03
24 Manuf. basic metals C 285,574 0.6 0.017
26 Manuf. computer/elect/opt C 239,425 0.44 0.027
21 Manuf. Pharma C 218,005 0.9 0.013
30 Manuf. other transport equip. C 156,098 0.17 0.013
10 Manuf. food products C 138,202 0.2 0.002
28 Manuf. machinery and equip. C 135,429 0.27 0.003
17 Manuf. paper/products C 131,726 0.29 0.004
27 Manuf. electrical equip. C 116,954 0.24 0.003
15 Manuf. leather/products C 108,611 0.67 0.009
32 Other manuf. C 89,349 0.13 0.008
22 Manuf. rubber/plastic C 82,638 0.23 0.002
13 Manuf. textiles C 75,699 0.44 0.009
14 Manuf. wearing apparel C 73500 0.59 0.003
23 Manuf. other non-metalic C 49033 0.25 0.003
31 Manuf. furniture C 28915 0.22 0.005
61 Telecom. H 0.03
53 Postal/courier serv. J 0.03
63 Information serv. J 0.035
62 Computer programming serv. J 0.036
93 Sport/Recreation activ. R 0.06
50 Water transport H 0.115
65 Insurance/pension funding K 0.132
60 Broadcast. activ. J 0.17
51 Air transport H 0.305
12 Manuf. tobacco C 0.338

Notes: The table shows industries classified as tradables using the geographic concentration and global trade
criteria as in Mian and Sufi (2014). First and second column shows the 2-digit NACE code and the name of
the industry, respectively. The fourth and fifth column shows ratio of trade (exports+imports) relative to total
employment and the ratio of trade to gross-output ratio, respectively computed using aggregate industry data
for the year 2007. The last column reports the Herfindahl-index computed with municipal level employment
shares for the year 2007.
 116

TABLE A.2: Tradable NACE Industries

Division Division Name Section HHI

49 Land transport and transport via pipelines H 0.0092
55 Accommodation I 0.0075
46 Wholesale trade G 0.0078
56 Food and beverage service activities I 0.0074
47 Retail trade G 0.0056
33 Repair & inst. of machinery & equip. C 0.0051
45 Wholesale and retail trade vehicles & motorcycles G 0.0043
43 Specialised construction activities F 0.0032
42 Civil Engineering F 0.0034
41 Construction of buildings F 0.0035

Notes: The table shows industries classified as non-tradables using the geographic concentration criteria as in
Mian and Sufi (2014). First and second column shows the 2-digit NACE code and the name of the industry.
The fourth column reports the Herfindahl-index computed with municipal level employment shares for the year
2007. The last three rows marked with red are exclude industries from the non-tradable classification due to
being related to construction.

B EMPIRICAL ANALYSIS: ADDITIONAL DETAILS

ABOUT MUNICIPAL LEVEL COVARIATES

This appendix provide additional detail on the data sources for the variables in-
cluded in the covariate-balance analysis in 1.5.4.
Income. The Italian Finance Department dependent of the Ministry of Economy
and Finance provides data gross income (reddito complessivo) for each municipality.
The income measure we use is an aggregate of income declarations from resident
population in each municipality.
Local Credit Market. The Aggregated Data - Statistical Database of The Italian
Central Bank (Banca D’Italia) is the main source for loans and deposits at the mu-
nicipal level. Loans and deposits present the stock at the end of the year reported
by all bank branches in each municipality.
 117

Population and Migration.

Local government Financial Statements. The XXX dependent of the Ministry of
Interior is the source foe the yearly financial statements of municipal governments.
For each municipalty, financial statements of local governments contains detailed
information on financial accounts for revenues, expenditures and debt flows and
stocks.

ABOUT OTHER LOCAL POLICY TAX CHANGES

The variables controlling for other policy changes included in (1.18) are described
as follows:

1. 2008 Exemption for Main Residence: Interaction between a dummy for 2008 and
the implied reduction in the residential property tax rate relevant for the main
residence of households.

2. 2011 Income Tax Reform: Interaction between a dummy for 2011 and the log
value for per-capita revenues from income surcharge tax (IRPEF).

3. 2014 Property Tax Changes: Interaction between a dummy for 2014 and the log
value for per-capita revenues from new the new property service tax (TASI)

ABOUT PRODUCTIVITY SHOCKS

ABOUT CREDIT SUPPLY SHOCKS

ABOUT UNCERTAINTY SHOCKS

 118

C ROBUSTNESS CHECKS: ALTERNATIVE SPECIFICATIONS

TABLE B.3: Non-Tradable Employment: Robustness Checks

(1) (2) (3) (4) (5) (6)

ℎ
Δ𝜏𝑚,𝑡 × 1 {𝑡 = 2012} -0.087*** -0.077*** -0.083*** -0.089*** -0.091*** -0.051***
(0.015) (0.015) (0.016) (0.015) (0.022) (0.018)
𝑓
Δ𝜏𝑚,𝑡 × 1 {𝑡 = 2012} -0.045*** -0.045*** -0.047*** -0.044*** -0.067** -0.064***
(0.011) (0.010) (0.012) (0.011) (0.028) (0.023)
𝑁obs 43,540 43,539 43,519 43,540 43,540 31,486
𝑁mun 6,220 6,220 6,220 6,220 6,220 6,220
2
𝑅 0.13 0.21 0.22 0.13 0.13 0.13
Municipality FE ✓ ✓ ✓ ✓ ✓ ✓
Year FE ✓ ✓ ✓ ✓ ✓ ✓
Region × Year FE ✓ ✓
LLM × Year FE ✓
Additional Controls ✓
𝑖
Productivity × Δ𝜏𝑚,2012 ✓
𝑖
Uncertainty × Δ𝜏𝑚,2012 ✓
𝑖
Credit Supply × Δ𝜏𝑚,2012 ✓

Notes: Main dependent variable is the non-tradable employment growth rate. The sample used covers the period 2008-2014.
Column (1) reports the baseline results, column (2) add municipal level covariates, column (3) controls for local labor market
time trends. Columns (4), (5), and (6) include the interaction between the 2012 residential/CRE tax rate change and proxies for
productivity, uncertainty, and credit supply shocks, respectively. To proxy for productivity shocks we use the annual growth of
income per worker. The proxy for uncertainty shocks is based on the yearly standard deviation for the growth rate of income
per worker across municipalities within the same province. Finally, to proxy for credit supply shocks we use the number of local
banks branches in each municipality per 1000 residents. Standard errors in parentheses are clustered at the local labor market
level, *, **, *** indicate significance at the 10% 5% and 1% respectively
 119

TABLE B.4: Consumption Expenditure: Robustness Checks

(1) (2) (3) (4) (5) (6)

ℎ
Δ𝜏𝑚,𝑡 × 1 {𝑡 = 2012} -0.517*** -0.334** -0.407** -0.525*** -0.652*** -0.776***
(0.145) (0.144) (0.162) (0.144) (0.221) (0.184)
𝑓
Δ𝜏𝑚,𝑡 × 1 {𝑡 = 2012} -0.177 -0.058 -0.181 -0.163 -0.080 0.710***
(0.120) (0.119) (0.126) (0.121) (0.285) (0.230)
𝑁obs 34,517 34,517 34,350 34,517 34,517 28,599
𝑁mun 5,740 5,740 5,740 5,740 5,740 5,740
2
𝑅 0.10 0.10 0.21 0.10 0.10 0.10
Municipality FE ✓ ✓ ✓ ✓ ✓ ✓
Year FE ✓ ✓ ✓ ✓ ✓ ✓
Population deciles × Year ✓ ✓
LLM × Year FE ✓
Additional Controls ✓
𝑖
Productivity × Δ𝜏𝑚,2012 ✓
𝑖
Uncertainty × Δ𝜏𝑚,2012 ✓
𝑖
Credit Supply × Δ𝜏𝑚,2012 ✓

Notes: Main dependent variable is the growth rate of consumption expenditure, the latter is proxied by expenditures on new
vehicles. The sample used covers the period 2008-2014. Column (1) reports the baseline results, column (2) add municipal level
covariates, column (3) controls for local labor market time trends. Columns (4), (5), and (6) include the interaction between
the 2012 residential/CRE tax rate change and proxies for productivity, uncertainty, and credit supply shocks, respectively. To
proxy for productivity shocks we use the annual growth of income per worker. The proxy for uncertainty shocks is based
on the yearly standard deviation for the growth rate of income per worker across municipalities within the same province.
Finally, to proxy for credit supply shocks we use the number of local banks branches in each municipality per 1000 residents.
Standard errors in parentheses are clustered at the local labor market level, *, **, *** indicate significance at the 10% 5% and
1% respectively
 120

TABLE B.5: Residential Prices: Robustness Checks

(1) (2) (3) (4) (5) (6)

ℎ
Δ𝜏𝑚,𝑡 × 1 {𝑡 = 2012} -0.022** -0.027*** -0.014** -0.022** -0.019 -0.008
(0.009) (0.009) (0.006) (0.009) (0.016) (0.011)
𝑓
Δ𝜏𝑚,𝑡 × 1 {𝑡 = 2012} -0.017*** -0.018*** -0.013*** -0.017*** -0.002 -0.028***
(0.006) (0.006) (0.005) (0.006) (0.012) (0.010)
𝑁obs 38,731 38,729 38,494 38,731 38,731 27,445
𝑁mun 5,534 5,534 5,534 5,534 5,534 5,534
2
𝑅 0.33 0.34 0.66 0.33 0.63 0.34
Municipality FE ✓ ✓ ✓ ✓ ✓ ✓
Year FE ✓ ✓ ✓ ✓ ✓ ✓
Province × Year FE ✓
LLM × Year FE ✓
Population deciles × Year ✓ ✓ ✓ ✓ ✓ ✓
𝑖
Productivity × Δ𝜏𝑚,2012 ✓
𝑖
Uncertainty × Δ𝜏𝑚,2012 ✓
𝑖
Credit Supply × Δ𝜏𝑚,2012 ✓

Notes: Main dependent variable is the growth rate of residential properties prices. Residential properties price is defined as the
average value per meter2 across homogeneous real state markets within each municipality. The sample used covers the period
2008-2014. Column (1) reports the baseline results, column (2) add municipal level covariates, column (3) controls for local
labor market time trends. Columns (4), (5), and (6) include the interaction between the 2012 residential/CRE tax rate change
and proxies for productivity, uncertainty, and credit supply shocks, respectively. To proxy for productivity shocks we use the
annual growth of income per worker. The proxy for uncertainty shocks is based on the yearly standard deviation for the growth
rate of income per worker across municipalities within the same province. Finally, to proxy for credit supply shocks we use the
number of local banks branches in each municipality per 1000 residents. Standard errors in parentheses are clustered at the
local labor market level, *, **, *** indicate significance at the 10% 5% and 1% respectively
 121

TABLE B.6: Commercial Real Estate Prices: Robustness Checks

(1) (2) (3) (4) (5) (6)

ℎ
Δ𝜏𝑚,𝑡 × 1 {𝑡 = 2012} -0.005 -0.007 -0.002 -0.010 -0.023 -0.012
(0.010) (0.009) (0.008) (0.011) (0.018) (0.015)
𝑓
Δ𝜏𝑚,𝑡 × 1 {𝑡 = 2012} -0.032*** -0.025*** -0.016*** -0.038*** -0.018 -0.030**
(0.008) (0.007) (0.005) (0.009) (0.026) (0.015)
𝑁obs 29,471 25,805 29,000 25,805 25,805 19,332
𝑁mun 3,687 3,687 3,687 3,687 3,687 3,687
2
𝑅 0.31 0.46 0.65 0.21 0.21 0.21
Municipality FE ✓ ✓ ✓ ✓ ✓ ✓
Year FE ✓ ✓ ✓ ✓ ✓ ✓
Population deciles × Year ✓
Region × Year FE ✓ ✓ ✓
LLM × Year FE ✓
𝑖
Productivity × Δ𝜏𝑚,2012 ✓
𝑖
Uncertainty × Δ𝜏𝑚,2012 ✓
𝑖
Credit Supply × Δ𝜏𝑚,2012 ✓

Notes: Main dependent variable is the growth rate of Commercial Real Estate (CRE) prices. CRE price is defined as the average
value per meter2 across homogeneous real state markets within each municipality of all properties used in the retail sector. The
sample used covers the period 2008-2014. Column (1) reports the baseline results, column (2) add municipal level covariates,
column (3) controls for local labor market time trends. Columns (4), (5), and (6) include the interaction between the 2012
residential/CRE tax rate change and proxies for productivity, uncertainty, and credit supply shocks, respectively. To proxy for
productivity shocks we use the annual growth of income per worker. The proxy for uncertainty shocks is based on the yearly
standard deviation for the growth rate of income per worker across municipalities within the same province. Finally, to proxy
for credit supply shocks we use the number of local banks branches in each municipality per 1000 residents. Standard errors
in parentheses are clustered at the local labor market level, *, **, *** indicate significance at the 10% 5% and 1% respectively
 122

D MODEL SOLUTION CHAPTER 1

We begin with the the profit maximization problem (1.24) of firm producing vari-
ety 𝑐𝑗 . The first order conditions are as follows:

𝜖−1 1 𝛼( 𝜖−1 )−1 (1−𝛼)( 𝜖−1

𝜖 ) 𝑓
{𝐿𝑗 } ∶ 𝛼( ) 𝐶𝜖 𝐿 𝜖 (𝐻 ℎ ) = 𝑊(1 + 𝜇𝑗 ) (C.1a)
𝜖

𝑓 𝜖−1 1 𝛼( 𝜖−1 ) (1−𝛼)( 𝜖−1

𝜖 )−1 𝑓
{𝐻𝑗 } ∶ (1 − 𝛼) ( ) 𝐶 𝜖 𝐿𝑗 𝜖 (𝐻𝑗ℎ ) = 𝑃𝑓 (1 + 𝜏 𝑓 − 𝜙𝑓 𝜇𝑗 )
𝜖
(C.1b)

𝑓 𝑓
𝜇𝑗 [𝑊𝐿𝑗 − 𝜙𝑓 𝑃𝑓 𝐻𝑗 ] = 0 (C.1c)

𝑓
where 𝜇𝑗 represent the multiplier of the firm collateral constraint (1.24d). Assum-
ing the collateral constraint is binding, we can use (C.1a) and (C.1b) to solve for 𝐿𝑗
𝑓
and 𝐻𝑗 :
𝜖
𝜖−1 𝐶
𝐿𝑗 = [𝛼 ] (1−𝛼)(𝜖−1)
(C.2)
𝜖 𝑓
𝑊 1+𝛼(𝜖−1) (𝜙𝑓 𝑃𝑓 ) (1 + 𝜇𝑗 )𝜖

𝑓 𝜖−1
𝜖 𝜙𝛼(𝜖−1)
𝑓
𝐶
𝐻𝑗 = [(1 − 𝛼) ] 1+(1−𝛼)(𝜖−1)
(C.3)
𝜖 𝑓
𝑊 𝛼(𝜖−1) (𝑃𝑓 ) (1 + 𝜏 𝑓 − 𝜙𝑓 𝜇𝑗 )𝜖

using the ratio (C.2) to (C.3) and solving for 𝜇𝑓 :

𝑓 𝛼 (1 + 𝜏 𝑓 + 𝜙𝑓 )
𝜇𝑗 = −1 (C.4)
𝜙𝑓
 123

replacing (C.4) into (C.3) and (C.2) we obtain the demand functions for employ-
ment and commercial real estate properties:

𝜖−1
𝜖 𝜙1+𝛼(𝜖−1)
𝑓
𝐶
𝐿𝑑𝑗 =( ) (1−𝛼)(𝜖−1)
(C.5)
𝜖
𝑊 1+𝛼(𝜖−1) (𝑃𝑓 ) (1 𝑓
+ 𝜏 + 𝜙𝑓 )𝜖

𝑓 ,𝑑 𝜖−1
𝜖 𝜙𝛼(𝜖−1)
𝑓
𝐶
𝐻𝑗 =( ) 1+(1−𝛼)(𝜖−1)
(C.6)
𝜖
𝑊 𝛼(𝜖−1) (𝑃𝑓 ) (1 + 𝜏 𝑓 + 𝜙𝑓 )𝜖

replacing (C.5) and (C.6) into the firm’s profit function (1.24):

𝑊 𝐿𝑑𝑗 (1 + 𝜏 𝑓 + 𝜙𝑓 )
Π𝑗 = (C.7)
𝜙𝑓 (𝜖 − 1)

𝑓 ,𝑑
Finally, aggregating Π𝑗 , 𝐿𝑑𝑗 , and 𝐻𝑗 across all varierties 𝑗 ∈ [0, 1]:

𝜖−1
𝜖 𝜙1+𝛼(𝜖−1)
𝑓
𝐶
𝐿𝑑 =( ) (1−𝛼)(𝜖−1)
(C.8)
𝜖
𝑊 1+𝛼(𝜖−1) (𝑃𝑓 ) (1 𝑓
+ 𝜏 + 𝜙𝑓 )𝜖

𝜖−1
𝜖 𝜙𝛼(𝜖−1)
𝑓
𝐶
𝑓 ,𝑑
𝐻 =( ) 1+(1−𝛼)(𝜖−1)
(C.9)
𝜖
𝑊 𝛼(𝜖−1) (𝑃𝑓 ) (1 𝑓
+ 𝜏 + 𝜙𝑓 )𝜖

𝑊 𝐿𝑑 (1 + 𝜏 𝑓 + 𝜙𝑓 )
Π= (C.10)
𝜙𝑓 (𝜖 − 1)
1 1 𝑓 ,𝑑 1
where 𝐿𝑑 = ∫0 𝐿𝑑𝑗 𝑑𝑗, 𝐻 𝑓 ,𝑑 = ∫0 𝐻𝑗 𝑑𝑗, and Π = ∫0 Π𝑗 𝑑𝑗.
Using the first stage problem for households (1.21) we get the following first order
conditions:

1−𝛽
{𝐶} ∶ 𝛽 𝐶𝛽−1 (𝐻 ℎ ) = 𝜆 + 𝜇ℎ (C.11a)
 124

1
{𝐿} ∶ 𝜒 𝐿𝜈 = 𝜆 𝑊 (C.11b)

−𝛽
{𝐻 ℎ } ∶ (1 − 𝛽) 𝐶𝛽 (𝐻 ℎ ) = 𝜆 𝑃ℎ (1 + 𝜏 ℎ ) − 𝜇ℎ 𝜙ℎ 𝑃ℎ (C.11c)

𝜆 [𝑊𝐿 + Π − 𝐶 − 𝑃ℎ 𝐻 ℎ (1 + 𝜏 ℎ )] (C.11d)

𝜇ℎ [𝐶 − 𝜙ℎ 𝑃ℎ 𝐻 ℎ ] = 0 (C.11e)

where 𝜇ℎ and 𝜆 are the multipliers for the borrowing constraint (1.21b) and budget
constraint (1.21a) of households, respectively. Assuming the household borrowing
constraint is binding, we can use (1.21b) and (1.21a) to find a solution for 𝐶 and
𝐻ℎ:
𝜙ℎ
𝐶= (𝑊𝐿 + Π) (C.12)
1 + 𝜏 ℎ + 𝜙ℎ
1
𝐻ℎ = (𝑊𝐿 + Π) (C.13)
𝑃ℎ (1 + 𝜏 ℎ + 𝜙ℎ )

Using (C.11a), (C.11c), and (1.21b), we can solve for 𝜇ℎ and 𝜆:

1 𝜙ℎ
𝜇ℎ = 1−𝛽
[𝛽 − ] (C.14)
(𝜙ℎ 𝑃ℎ ) 1 + 𝜏 ℎ + 𝜙ℎ

𝛽
𝜙ℎ
𝜆= 1−𝛽
(C.15)
(𝑃ℎ ) (1 + 𝜏ℎ + 𝜙ℎ )

finding the labor supply function by replacing (C.15) into (C.11b):

𝛽 𝜈
⎡ 𝑊𝜙ℎ ⎤
𝐿𝑠 = ⎢ 1−𝛽 ⎥ (C.16)
𝜒 (𝑃 ℎ) (1 + 𝜏 ℎ+𝜙 )
⎣ ℎ ⎦

we can use the aggregate profit function for firms in (C.10), the labor market equi-
 125

librium 𝐿𝑑 − 𝐿𝑠 , and the labor supply function to replace for Π and 𝑊𝐿𝑠 in (C.12)
and (C.13):
1+𝛽𝜈
𝜙ℎ 𝑊 1+𝜈 (1 + 𝜏 𝑓 + 𝜖𝜙𝑓 )
𝐶= (1−𝛽)𝜈
(C.17)
𝜒 𝜈 𝜙𝑓 (𝜖 − 1) (𝑃ℎ ) (1 + 𝜏ℎ + 𝜙ℎ )1+𝜈
𝛽𝜈
𝜙ℎ 𝑊 1+𝜈 (1 + 𝜏 𝑓 + 𝜖𝜙𝑓 )
𝐻 ℎ,𝑑 = 1+(1−𝛽)𝜈
(C.18)
𝜒 𝜈 𝜙𝑓 (𝜖 − 1) (𝑃ℎ ) (1 + 𝜏 ℎ + 𝜙ℎ )1+𝜈

E PROOF OF PROPOSITIONS CHAPTER 1

𝑦
Proof Proposition 1. We can expressed 𝐴𝑇𝑇𝑡𝑅 in (1.12) as follows:

𝑦
𝐴𝑇𝑇𝑡𝑅 = E[𝑦𝑚,𝑡
𝑖
𝑅
𝑖
− 𝑦𝑚,𝑡 𝑅 −1
𝑖
+ 𝑦𝑚,𝑡 𝑅 −1
𝑖−1 ∣Δ𝜏 = Δ𝑖 ]
− 𝑦𝑚,𝑡𝑅 𝑚

+ E[𝑦𝑚,𝑡
𝑖−1 − 𝑦𝑖−1
𝑅 𝑚,𝑡𝑅 −1 ∣Δ𝜏𝑚 = Δ
𝑖−1 ] − E[𝑦𝑖−1 − 𝑦𝑖−1
𝑚,𝑡𝑅 𝑚,𝑡𝑅 −1 ∣Δ𝜏𝑚 = Δ
𝑖−1 ]

(D.19a)

Using condition (1.14a) we know that E[𝑦𝑚,𝑡

𝑖
𝑅 −1
∣Δ𝜏𝑚 = Δ𝑖 ] = E[𝑦𝑚,𝑡
𝑖−1
𝑅 −1
∣Δ𝜏𝑚 =

Δ𝑖 ]

𝑦
𝐴𝑇𝑇𝑡𝑅 = E[𝑦𝑚,𝑡
𝑖
𝑅
𝑖
− 𝑦𝑚,𝑡 𝑅 −1
∣Δ𝜏𝑚 = Δ𝑖 ] − E[𝑦𝑚,𝑡
𝑖−1 − 𝑦𝑖−1
𝑅 𝑚,𝑡𝑅 −1 ∣Δ𝜏𝑚 = Δ
𝑖−1 ]

+ E[𝑦𝑚,𝑡
𝑖−1 − 𝑦𝑖−1
𝑅 𝑚,𝑡𝑅 −1 ∣Δ𝜏𝑚 = Δ
𝑖−1 ] − E[𝑦𝑖−1 − 𝑦𝑖−1
𝑚,𝑡𝑅
𝑖
𝑚,𝑡𝑅 −1 ∣Δ𝜏𝑚 = Δ ] (D.20a)

Finally using condition (1.14b):

E[𝑦𝑚,𝑡
𝑖−1 − 𝑦𝑖−1
𝑅 𝑚,𝑡𝑅 −1 ∣Δ𝜏𝑚 = Δ
𝑖−1 ] − E[𝑦𝑖−1 − 𝑦𝑖−1
𝑚,𝑡𝑅
𝑖
𝑚,𝑡𝑅 −1 ∣Δ𝜏𝑚 = Δ ] = 0
 126

therefore, equation (D.20a) transforms into:

𝑦
𝐴𝑇𝑇𝑡𝑅 = E[𝑦𝑚,𝑡
𝑖
𝑅
𝑖
− 𝑦𝑚,𝑡𝑅 −1
∣Δ𝜏𝑚 = Δ𝑖 ] − E[𝑦𝑚,𝑡
𝑖−1 − 𝑦𝑖−1
𝑅 𝑚,𝑡𝑅 −1 ∣Δ𝜏𝑚 = Δ
𝑖−1 ] (D.21)

notice that the left hand side of equation (D.21) is the diff-in-diff estimator defined
in (1.13) QED.
Proof Proposition 2. Using equation (C.12), we can see that the expenditure share
in consumption goods is equal to:

𝐶 𝜙ℎ
=
𝑊𝐿 + Π 1 + 𝜏 ℎ + 𝜙ℎ

while 𝛽 is the expenditure share in consumption goods in a frictionless economy.

Therefore, using equation (C.14) we can see that 𝜇ℎ > 0 if

𝜙ℎ 𝐶
𝛽− >0⟺ <𝛽
ℎ
1 + 𝜏 + 𝜙ℎ 𝑊𝐿 + Π

On the other hand, using equations (C.5) and (C.6) the expression for the cost
function is as follows:

𝑓 𝜖−1
𝜖 𝜙𝛼(𝜖−1)
𝑓 1 + 𝜏 𝑓 + 𝜙𝑓
𝑓 𝑓
𝑊𝐿𝑗 +(1+𝜏 )𝑃 𝐻𝑗 =( ) (1−𝛼)(𝜖−1) 𝜖−1
= 𝑊𝐿
𝜖 𝜙𝑓
𝑊 𝛼(𝜖−1) (𝑃𝑓 ) (1 𝑓
+ 𝜏 + 𝜙𝑓 )

then the labor cost-to-total costs is equal to:

𝑊𝐿𝑗 𝜙𝑓
𝑓
=
𝑊𝐿𝑗 + (1 + 𝜏 𝑓 )𝑃𝑓 𝐻𝑗 1 + 𝜏 𝑓 + 𝜙𝑓

while the same ratio in the case of a frictionless economy is 𝛼. Using (C.4), 𝜇𝑓 > 0
 127

if and only if:

𝛼 (1 + 𝜏 𝑓 + 𝜙𝑓 ) 𝑊𝐿𝑗
−1>0⟺ <𝛼
𝜙𝑓 𝑊𝐿𝑗 + (1 + 𝜏 𝑓 )𝑃𝑓 𝐻𝑗
𝑓

QED.
Proof Proposition 3. We can find the equilibrium price for the housing market using
demand function (C.9) and the supply function (1.25b):

𝛽𝜈
1+𝜎ℎ +(1−𝛽)𝜈 𝜙ℎ 𝑊 1+𝜈 (1 + 𝜏 𝑓 + 𝜙𝑓 𝜖)
(𝑃ℎ ) = (D.22)
𝜒 𝜈 𝜙𝑓 (𝜖 − 1)(1 + 𝜏 ℎ + 𝜙ℎ )1+𝜈

using (D.22) we can reduce (C.17) to the following expression

1+𝜎ℎ
𝐶 = 𝜙ℎ (𝑃ℎ ) (D.23)

In the same manner, the equilibrium price for the commercial real estate market can
be found using (C.9) and (1.25b) and replacing the value for 𝐶 in equation (D.23):

1+𝜎ℎ
1+𝜎𝑓 +(1−𝛼)(𝜖−1) (𝜖 − 1)𝜖 𝜙ℎ 𝜙𝛼(𝜖−1)
𝑓
(𝑃ℎ )
(𝑃𝑓 ) = (D.24)
𝜖 𝜖 𝑊 𝛼(𝜖−1) (1 + 𝜏 𝑓 + 𝜙𝑓 )𝜖

using (D.24), (D.23), and (C.8) we can reduce equilibrium employment to the fol-
lowing expression:
1+𝜎𝑓
𝜙𝑓 (𝑃𝑓 )
𝐿= (D.25)
𝑊

Finally, applying log’s to equations (D.22), (D.24), (D.23), and (D.25) and defining:
 128

𝛽𝜈
⎡ 𝜙 ⎤
𝜅𝑃ℎ (Θ) = log ⎢ 𝜈 ℎ ⎥
𝜒 𝜙𝑓 (𝜖 − 1)
⎣ ⎦
𝜖 𝛼(𝜖−1)
⎡ (𝜖 − 1) 𝜙ℎ 𝜙𝑓 ⎤
𝜅𝑃𝑓 (Θ) = log ⎢ 𝜖 ⎥
𝜖
⎣ ⎦
𝐴ℎ = 1 + 𝜎ℎ + (1 − 𝛽)𝜈

𝐴𝑓 = 1 + 𝜎𝑓 + (1 − 𝛼)(𝜖 − 1)

we obtain equations (1.28a) - (1.28d) QED.

𝜏ℎ 𝜏𝑓 𝜏𝑓
Proof Proposition 4. Given that ,
1+𝜙ℎ 1+𝜙𝑓
and 1+𝜖𝜙𝑓
are close to zero, we use the
following three approximations

1 + 𝜏ℎ
log(1 + 𝜏ℎ + 𝜙ℎ ) ≈ log(1 + 𝜙ℎ ) +
1 + 𝜙ℎ
1 + 𝜏𝑓
log(1 + 𝜏 𝑓 + 𝜙𝑓 ) ≈ log(1 + 𝜙𝑓 ) +
1 + 𝜙𝑓

𝑓 1 + 𝜏𝑓
log(1 + 𝜏 + 𝜖𝜙𝑓 ) ≈ log(1 + 𝜖𝜙𝑓 ) +
1 + 𝜖𝜙𝑓

then, using equations (1.28b) and (1.28a):

1 1
𝐴ℎ 𝑝ℎ = (1 + 𝜈) [𝑤 − Δ𝜏 ℎ ] + Δ𝜏 𝑓 (D.28a)
1 + 𝜙ℎ 1 + 𝜖𝜙𝑓
𝜖
𝐴𝑓 𝑝𝑓 = (1 + 𝜎ℎ )𝑝ℎ − 𝛼(𝜖 − 1)𝑤 − Δ𝜏 𝑓 (D.28b)
1 + 𝜙𝑓

𝑓 𝑓
where: 𝑝ℎ = log(𝑃1ℎ ) − log(𝑃0ℎ ), 𝑝𝑓 = log(𝑃1 ) − log(𝑃0 ), 𝑤 = log(𝑊1 ) − log(𝑊0 ),
𝑓 𝑓
Δ𝜏 ℎ = 𝜏1ℎ − 𝜏0ℎ , and Δ𝜏 𝑓 = 𝜏1 − 𝜏0 . Using the labor equilibrium condition with
 129

equations (C.8), (C.16), and (D.23) we can solve for the equilibrium wage log(𝑊):

𝜈
⎡ 𝜒 𝜙𝑓 ⎤
(1 + 𝜈) log (𝑊) = log ⎢ 𝛽𝜈 ⎥ + (1 + 𝜎𝑓 ) log 𝑃𝑓 + (1 − 𝛽)𝜈 log 𝑃ℎ + 𝜈 log (1 + 𝜏 ℎ + 𝜙ℎ )
⎣ 𝜙ℎ ⎦
(D.29)
which can be in terms of 𝑤, 𝑝ℎ , 𝑝𝑓 , and Δ𝜏 ℎ as follows:

𝜈
(1 + 𝜈)𝑤 = (1 + 𝜎𝑓 )𝑝𝑓 + (1 − 𝛽)𝜈𝑝ℎ + Δ𝜏 ℎ (D.30)
1 + 𝜙ℎ

Using (D.28a), (D.28b), and (D.30) solving for 𝑝ℎ , 𝑝𝑓 and 𝑤 as a function of Δ𝜏 ℎ

and Δ𝜏 𝑓 :
𝑝ℎ = 𝛽𝑝𝑓 ,ℎ (Θ)Δ𝜏 ℎ + 𝛽𝑝𝑓 ,𝑓 (Θ)Δ𝜏 𝑓 (D.31)

𝑝ℎ = 𝛽𝑝ℎ ,ℎ (Θ)Δ𝜏 ℎ + 𝛽𝑝ℎ ,𝑓 (Θ)Δ𝜏 𝑓 (D.32)

𝑤 = 𝛽𝑤,ℎ (Θ)Δ𝜏 ℎ + 𝛽𝑤,𝑓 (Θ)Δ𝜏 𝑓 (D.33)

where:

(1 + 𝜈) [𝛼(𝜖 − 1)(1 + 𝜎𝑓 ) + (1 + 𝜈)𝐴𝑓 ]

𝛽𝑝ℎ ,ℎ (Θ) =− (D.34a)
𝐴ℎ𝑓 (𝜖 − 1)(1 + 𝜙ℎ )

(1 + 𝜙𝑓 ) [(1 − 𝛼)(1 + 𝜈)𝜎𝑓 + 𝛼𝜈(1 + 𝜎𝑓 )] + 𝜖𝜙𝑓 (1 + 𝜈)(1 + 𝜎𝑓 )

𝛽𝑝ℎ ,𝑓 (Θ) =−
𝐴ℎ𝑓 (𝜖 − 1)(1 + 𝜙𝑓 )(1 + 𝜖𝜙𝑓 )
(D.34b)

(1 + 𝜈) [(1 + 𝜈)(1 + 𝜎ℎ ) − 𝛼(𝜖 − 1)(1 − 𝛽)𝜈]

𝛽𝑝𝑓 ,ℎ (Θ) = (D.34c)
𝐴ℎ𝑓 (𝜖 − 1)(1 + 𝜙ℎ )

(1 + 𝜈)(1 + 𝜎ℎ )(1 + (𝜖 + 1)𝜙𝑓 ) + 𝛼(1 − 𝛽)𝜈(1 + 𝜙𝑓 )

𝛽𝑝𝑓 ,𝑓 (Θ) =− (D.34d)
𝐴ℎ𝑓 (𝜖 − 1)(1 + 𝜙𝑓 )(1 + 𝜖𝜙𝑓 )
 130

(1 + 𝜈) [𝜎𝑓 (1 − 𝛼)(1 + 𝜈) + 𝛼𝜈(1 + 𝜎𝑓 )] + 𝜖𝜙𝑓 (1 + 𝜈)(1 + 𝜎𝑓 )

𝛽𝑤,ℎ (Θ) =−
𝐴ℎ𝑓 (𝜖 − 1)(1 + 𝜙ℎ )
(D.34e)

[1 + 𝜙𝑓 (𝜖 + 1)] [(1 + 𝜎ℎ ) + 𝜎𝑓 𝐴ℎ ] + (1 − 𝛽)𝜈 [𝛼 + (𝜖 + 1)𝜙𝑓 ]

𝛽𝑤,𝑓 (Θ) =−
𝐴ℎ𝑓 (𝜖 − 1)(1 + 𝜙𝑓 )(1 + 𝜖𝜙𝑓 )
(D.34f)

𝐴𝑓 = 1 + 𝜎𝑓 + (1 − 𝛼)(𝜖 − 1)

𝐴ℎ = 1 + 𝜎ℎ + (1 − 𝛽)𝜈

𝐴ℎ𝑓 = 𝛼(1 + 𝜎𝑓 )𝐴ℎ + (1 − 𝛼)(1 + 𝜈)(1 + 𝜎ℎ )

The equilibrium response for consumption and employment can be found using
equations (D.25), and (D.23):

𝑐 = (1 + 𝜎ℎ )𝑝ℎ (D.35a)

𝑙 = (1 + 𝜎𝑓 )𝑝𝑓 − 𝑤 (D.35b)

replacing (D.32), (D.31), and (D.33) into (D.35a) and (D.35b) we get:

𝑙 = 𝛽𝑙,ℎ (Θ)Δ𝜏 ℎ + 𝛽𝑙,𝑓 (Θ)Δ𝜏 𝑓 (D.36)

𝑐 = 𝛽𝑐,ℎ (Θ)Δ𝜏 ℎ + 𝛽𝑐,𝑓 (Θ)Δ𝜏 𝑓 (D.37)

where:

𝛽𝑙,ℎ (Θ) = (1 + 𝜎𝑓 )𝛽𝑝𝑓 ,ℎ (Θ) − 𝛽𝑤,ℎ (Θ) (D.38a)

 131

𝛽𝑙,𝑓 (Θ) = (1 + 𝜎𝑓 )𝛽𝑝𝑓 ,𝑓 (Θ) − 𝛽𝑤,𝑓 (Θ) (D.38b)

𝛽𝑐,ℎ (Θ) = (1 + 𝜎ℎ )𝛽𝑝ℎ ,ℎ (Θ) (D.38c)

𝛽𝑐,𝑓 (Θ) = (1 + 𝜎ℎ )𝛽𝑝ℎ ,𝑓 (Θ) (D.38d)

Equations (D.32), (D.31), (D.36), and (D.37) can be generalized as done in (1.29)
QED.
Proof Proposition 5. The housing wealth channel is the partial equilibrium re-
sponse of employment demand after an increase in housing prices induced by an
higher residential. Using equations (D.23) and (D.22) to find the percentage equi-
librium response of consumption and housing prices after an increase in residential
taxes equivalent to Δ𝜏 ℎ , all else constant:

𝑐 = (1 + 𝜎ℎ )𝑝ℎ
1+𝜈
𝑝ℎ =− Δ𝜏 ℎ
(1 + 𝜙ℎ )(1 + 𝜎ℎ + (1 − 𝛽)𝜈)

then 𝜕𝑐/𝜕𝑝ℎ and 𝜕𝑝ℎ /𝜕Δ𝜏 ℎ are equal to:

𝜕𝑐
= 1 + 𝜎ℎ (D.40a)
𝜕𝑝ℎ

𝜕𝑝ℎ 1+𝜈
=− (D.40b)
𝜕Δ𝜏 ℎ (1 + 𝜙ℎ )(1 + 𝜎ℎ + (1 − 𝛽)𝜈)

defining the percentage response of labor demand to changes in consumption with

equation (C.8), all else constant we find the following expression:

𝑙𝑑 = 𝑐
 132

therefore 𝜕𝑙𝑑 /𝜕𝑐 will be:

𝜕𝑙𝑑
=1 (D.41)
𝜕𝑐

The firm collateral channel is the partial equilibrium response of labor demand due
to lower CRE prices induced by an increase in CRE taxes. Keeping everything else
constant, the percentage equilibrium response of CRE prices after an increase in
CRE taxes equal to Δ𝜏 𝑓 can be defined with equation (D.24) as follows:

𝜖
𝑝𝑓 = − Δ𝜏 𝑓
(1 + 𝜙𝑓 )(1 + 𝜎𝑓 + (1 − 𝛼)(𝜖 − 1))

then 𝜕𝑝𝑓 /𝜕Δ𝜏 𝑓 will be:

𝜕𝑝𝑓 𝜖
=− (D.42)
𝜕Δ𝜏 𝑓 1 + 𝜎𝑓 + (1 − 𝛼)(𝜖 − 1)

using (D.25) to define the percentage change of labor demand after a decrease in
CRE price, all else constant, we will get the following expression:

𝑙𝑑 = (1 + 𝜎𝑓 )𝑝𝑓

then 𝜕𝑙𝑑 /𝜕𝑝𝑓 will be:

𝜕𝑙𝑑
= 1 + 𝜎𝑓 (D.43)
𝜕𝑝𝑓

Finally, replacing equations (D.40a), (D.40b), (D.41), (D.42), and (D.43) in defini-
tion 1.6.4 we obtain 𝛿wealth (Θ) and 𝛿coll (Θ) QED.
Proof Proposition 6. Let 𝜏 𝑓 be constant, obtaining the percentage employment, con-
sumption, and housing prices response after an increase in residential taxes equal
 133

to Δ𝜏 ℎ from equations (C.8), (D.23), and (D.22):

𝑙 = 𝑐 − (1 + 𝛼(𝜖 − 1))𝑤 − (1 − 𝛼)(𝜖 − 1)𝑝𝑓 (D.44a)

𝑐 = (1 − 𝜎ℎ )𝑝ℎ (D.44b)
1+𝜈 1
𝑝ℎ = [𝑤 − Δ𝜏 ℎ ] (D.44c)
1 + 𝜎ℎ + (1 − 𝛽)𝜈 1 + 𝜙ℎ

According to proposition 4, the reduced form effect of Δ𝜏 ℎ on 𝑝𝑓 and 𝑤 can be

represented as follows: 𝑝𝑓 = 𝛽𝑝𝑓 ,ℎ (Θ)Δ𝜏 ℎ , 𝑤 = 𝛽𝑤,ℎ (Θ)Δ𝜏 ℎ . Replacing (D.44b),
(D.44c) and the latter two expressions for 𝑝𝑓 and 𝑤 into (D.44c):

⎧
{ ⎫
}
{ (1 − 𝜈)(1 + 𝜎ℎ ) (1 + 𝜎ℎ )(1 + 𝜈) }
𝑙 = ⎨− −(1−𝛼)(𝜖−1)𝛽𝑝𝑓 ,ℎ +[ −(1−𝛼(𝜖−1))]𝛽𝑤,ℎ ⎬Δ𝜏 ℎ
{ (1 + 𝜙ℎ )𝐴ℎ 𝐴ℎ }
{ }
⎩ ⎭
(D.45)
where 𝐴ℎ = 1 + 𝜎ℎ + (1 − 𝛽)𝜈. Notice that the expression in curly brackets in equa-
tion (D.45) is 𝛽𝑙,ℎ (Θ).
Now, keeping 𝜏 ℎ constant the percentage change in employment and CRE prices
due to an increase in CRE taxes equivalent to Δ𝜏 𝑓 can be represented using equa-
tions (D.25) and (D.24) as follows:

𝑙 = (1 + 𝜎𝑓 )𝑝𝑓 − 𝑤 (D.46a)

1 ⎡ ⎤
𝑝𝑓 = ⎢(1 + 𝜎 )𝑝ℎ − 𝛼(𝜖 − 1)𝑤 − 𝜖 Δ𝜏 𝑓 ⎥ (D.46b)
⎢ 𝑓 ⎥
1 + 𝜎𝑓 + (1 − 𝛼)(𝜖 − 1) ⎢ 1 + 𝜙𝑓 ⎥
⎣ ⎦

Again using proposition 4, we know that 𝑝ℎ = 𝛽𝑝ℎ ,𝑓 (Θ)Δ𝜏 𝑓 and 𝑤 = 𝛽𝑤,𝑓 (Θ)Δ𝜏 𝑓 .
Replacing the previous expressions together with (D.46b) into (D.46a) we obtain
 134

the following expression:

⎧
{ ⎫
}
{ (1 + 𝜎𝑓 )𝜖 (1 + 𝜎𝑓 )(1 + 𝜎ℎ ) 𝛼(𝜖 − 1)(1 + 𝜎𝑓 ) }
𝑙=⎨− + 𝛽𝑝ℎ ,𝑓 − [ + 1]𝛽𝑤,𝑓 ⎬Δ𝜏 𝑓
{ (1 + 𝜎𝑓 )𝐴𝑓 𝐴𝑓 𝐴𝑓 }
{ }
⎩ ⎭
(D.47)
where 𝐴𝑓 = 1+𝜎𝑓 +(1−𝛼)(𝜖 −1). Notice again that the expression in curly brackets
in equation (D.47) is 𝛽𝑙,𝑓 (Θ). Finally, according to proposition 5: (i) 𝛿wealth (Θ) =
−(1 − 𝜈)(1 + 𝜎ℎ )/(1 + 𝜙ℎ )𝐴ℎ , (ii) 𝛿coll (Θ) = −𝜖(1 + 𝜎𝑓 )/(1 + 𝜙𝑓 )𝐴𝑓 .
Then, replacing the definitions for the housing wealth and firm collateral channel
into (D.45) and (D.47) we obtain the decomposition of the reduced form effects for
employment in (1.34a) and (1.34b) QED.
Proof Proposition 7. Using equations (D.34a), (D.38c), (D.34d) and (D.34b) we
arrive to the desired result in (1.35a), (1.35b), (1.35c), and (1.35d), respectively
QED.