Clase Notes Fin

You might also like

Download as pdf or txt
Download as pdf or txt
You are on page 1of 38

Introduction

Brief literature review


Model specification and econometric approach
Data
Results

Financial Information in Colombia


Mistery Shoppers

J.E. Gomez-Gonzalez1 X. Giné2 N. García3


1 Research Department
Banco de la República
2 Research Department
World Bank
3 FinancialEducation Department
Banco de la República

Banco de España, 2018


Gomez-Gonzalez Mistery shoppers
Introduction
Brief literature review
Model specification and econometric approach
Data
Results

Outline

1 Introduction

2 Brief literature review

3 Model specification and econometric approach

4 Data

5 Results

Gomez-Gonzalez Mistery shoppers


Introduction
Brief literature review
Model specification and econometric approach
Data
Results

Some recent facts

Analysts have reported that Canadian house price growth


during the last decade has been excessive.
Aggregate price-rent ratio at the end of 2017 was 89 per
cent higher than its historical average.
Warnings about a possible bubble have been raised.
Prices relative to fundamentals (household income, rent)
are on record high levels, comparable only to those in
Australia, New Zealand and Belgium (OECD, 2017).
Papers studying house price dynamics in Canada
emphasize in heterogeneous behavior among Canadian
provinces.

Gomez-Gonzalez Mistery shoppers


Introduction
Brief literature review
Model specification and econometric approach
Data
Results

What do we do in this paper?

We test for the existence of bubbles in Canadian housing


markets.
Instead of focusing in aggregate house prices at the
country level, we use province-level data.
We use two mispricing indicators: Price-to-rent ratios and
the non-fundamental component of house prices.
We test for bubble transmission within provinces.

Gomez-Gonzalez Mistery shoppers


Introduction
Brief literature review
Model specification and econometric approach
Data
Results

Preview of results

When using the price-to-rent ratio we identify bubbles in all


provinces.
Using the non-fundamental price component we find
bubbles in 6 (out of 9) provinces.
Bubble transmissions are identified from BC to other
provinces, such as QC, MB, ON, NL and AB.
Migration intensities are time-varying and exhibit inverted
U-shapes.

Gomez-Gonzalez Mistery shoppers


Introduction
Brief literature review
Model specification and econometric approach
Data
Results

Contributions

Regional transmission of bubbles within a country has


received little attention in the literature. Our study is the
first in studying the Canadian case.
The few papers studying bubble migration have used only
the price-to-rent ratio. We use also a measure of the
non-fundamental component of prices that reduces the
probability of false positive results (Shi, 2017).
We show that bubble migrations do not follow a
geographical pattern. Instead, they happen within regions
in which investment opportunities appear in a given period
of time.

Gomez-Gonzalez Mistery shoppers


Introduction
Brief literature review
Model specification and econometric approach
Data
Results

Policy implications

Bubble formation occurs in a period in which Canadian


interest rates remained in persistently low levels.
This finding coincides with predictions of several studies
showing that prolongued periods of monetary policy easing
increase financial vulnerability (Maddaloni and Peydró,
2011; Dell’Ariccia et al., 2013; Jiménez et al., 2014;
Cecchetti et al., 2017; Amador-Torres et al., 2018).
Hence, our results provide further evidence of important
collateral effects that monetary policy may have credit,
asset prices and overall financial stability.
Central banks should account for these collateral effects
when designing interest rate policies.

Gomez-Gonzalez Mistery shoppers


Introduction
Brief literature review
Model specification and econometric approach
Data
Results

House price bubble detection


Early bubble detection methods include Shiller’s variance
bound test (1981) and West’s two-step test (1987).
Campbell and Shiller (1987) and Diba and Grossman
(1988) introduced the most commonly used methods in the
literature, namely the right-tailed unit root test and the
cointegration test.
Standard right-tailed unit root tests suffer from a serios
limitation pointed-out by Evans (1991): They lose power
significantly to detect bubbles when the sample data
includes multiple bubbles that emerge and collapse.
PSY (2015) propose a recursive ADF test that improves
power significantly with respect to conventional unit root
tests, while allowing the date-stamping of bubbles.
Gomez-Gonzalez Mistery shoppers
Introduction
Brief literature review
Model specification and econometric approach
Data
Results

House price bubble detection

This test has been extensively used recently for detecting


bubbles in different financial markets: Stock markets (Lee
and Phillips, 2016; Deang and Xie, 2017; Chuliá et al.,
2017); commodity markets (Etienne, 2016; Alexakis et al.,
2017); energy markets (Narayan and Narayan, 2017);
exchange rates (Maldonado et al., 2016); housing markets
(Anundsen et al., 2016; Pavlidis et al., 2016;
Gomez-Gonzalez et al., 2017).
These papers have found several episodes of housing
bubbles, many of them relating to the recent international
financial crisis.

Gomez-Gonzalez Mistery shoppers


Introduction
Brief literature review
Model specification and econometric approach
Data
Results

Non-fundamental component of prices

Shi (2017) argues that papers applying test to price-to-rent


ratio may be subject to false positives, as rents may not be
a good indicator of the fundamental price.
She proposes "cleaning-up" housing prices before
implementing the bubble detection test.
First, a fundamental price is computed first within a VAR
setup. Then, tests are performed on the non-fundamental
price component.

Gomez-Gonzalez Mistery shoppers


Introduction
Brief literature review
Model specification and econometric approach
Data
Results

Bubble migration

Growing literature suggesting the interdependence


between national and global house prices (Vansteenkiste,
2009; Cesa-Bianchi, 2011).
Regarding regional transmission of shocks, Vansteenkiste
(2007) shows that house price shocks in California are an
important factor driving prices in other US states.
Only three papers studying regional migration of house
price bubbles within a country: Two study China (Shih et
al., 2014; Chen et al., 2017); the other New Zealand
(Greenaway-McGreevy and Phillips, 2016).
All three papers report evidence of regional transmission of
housing bubbles.

Gomez-Gonzalez Mistery shoppers


Introduction
Brief literature review
Model specification and econometric approach
Data
Results

Baseline model

Let (pt ) be the log real price of housing and (rt ) the log real
housing rent.
The price-to-rent ratio consists of a market fundamental
(Ft ) and a bubble component (Bt ):

pt − rt = Ft + Bt (1)

where Bt satisfies the submartingale property (Diba and


Grossman, 1988),

1 1
Et (Bt+1 ) = Bt , with > 1
ρ ρ

Gomez-Gonzalez Mistery shoppers


Introduction
Brief literature review
Model specification and econometric approach
Data
Results

Fundamental value component

Rents may not be sufficient to "clean-up" housing prices


from their fundamentals.
An alternative is to compute the fundamental value in a
VAR setup and subtract it from observed prices before
applying PSY bubble detection tests.
We compute the fundamental value considering the
one-period return to housing (Vt+1 ) as,

Pt+1 + Rt+1
Vt+1 = .
Pt
.

Gomez-Gonzalez Mistery shoppers


Introduction
Brief literature review
Model specification and econometric approach
Data
Results

Fundamental value component

Using a first order Taylor expansion, the log housing price


is given by

pt = κ + ρpt+1 + (1 − ρ)rt+1 − vt+1 , (2)

where lowercase variables refer to the natural logarithm of



the real variables, ρ = ep̄e+er̄ , κ = −log(ρ) + (1 − ρ)(p̄ − r̄ ),
with p̄ and r̄ being the sample means of log real housing
prices and real rents, respectively.

Gomez-Gonzalez Mistery shoppers


Introduction
Brief literature review
Model specification and econometric approach
Data
Results

Fundamental value component

Solving Eq 2 by iteration, yields an equation for log prices


as the sum of a long-term component, the discounted
future cash-flows and the bubble component:
∞ ∞
κ X
j
X
pt = + (1 − ρ) ρ rt+1+j − ρj vt+1+j + Bt
1−ρ |{z}
j=0 j=0
} Bubble component
| {z }
Long-term | {z
Discounted Cash Flows
(3)
1
with Bt ≡ limj→∞ ρj pt+j = ρ Bt−1 .
The bubble component Bt satisfies the submartingale
property mentioned above.

Gomez-Gonzalez Mistery shoppers


Introduction
Brief literature review
Model specification and econometric approach
Data
Results

Fundamental value component

Computing the log price-to-rent ratio and using Equation 3,


we obtain the following expression,

κ X
pt − rt = Ft + Bt with Ft ≡ + ρk (∆rt+1+k − vt+1+k ).
1−ρ
k =0
(4)

Gomez-Gonzalez Mistery shoppers


Introduction
Brief literature review
Model specification and econometric approach
Data
Results

Fundamental value component

Campbell et al. (2009), Sung and Tsang (2013) and Shi


(2017) assume that the log gross return to housing (vt+1 )
can be split into the real risk-free rate (it+1 ) and a
time-varying risk premium (φt+1 ).
This premium is assumed to be the sum of an expected
long-term risk premium ϕ and a zero-mean disturbance
(t+1 ).
Hence, the log-gross return is given by
vt+1 = ϕ + it+1 + t+1 .

Gomez-Gonzalez Mistery shoppers


Introduction
Brief literature review
Model specification and econometric approach
Data
Results

Fundamental value component

Using Equation 4, the housing market fundamental


component Ft is given by
"∞ ∞
#
κ̃ − ϕ̂ X
k
X
k
F̂t = + ρ̃ ∆r̂t+1+k − ρ̃ ît+1+k , (5)
1 − ρ̃
k =0 k =0

where κ̃ and ρ̃ are calibrated based on historical data, and


ϕ̂ is estimated by Recursive Least Squares as the intercept
of a regression of the form (vt+1 − it+1 ) = ϕ + t+1 .

Gomez-Gonzalez Mistery shoppers


Introduction
Brief literature review
Model specification and econometric approach
Data
Results

VAR approach for forecasting the fundamental value

For computing the fundamental value, forecasts of the


growth rate of real rent and the real interest rate are
needed.
We do these forecasts within a VAR framework.
Let Ztj be a vector containing the variables of region j at
time t. In this context j can be either Canada or one of the
nine provinces included in the study.

ZtCA = {∆rtCA , it , ∆YtCA , ∆LCA CA


t , ∆Nt }

Ztj = {∆rtj , it , ∆Ytj , ∆Ljt , ∆Ntj , ∆rtCA , ∆YtCA , ∆LCA CA


t , ∆Nt }

Gomez-Gonzalez Mistery shoppers


Introduction
Brief literature review
Model specification and econometric approach
Data
Results

VAR approach for forecasting the fundamental value

The forecasting equation for the two main-variables for the


j − th province are:
p p p p p
CA CA CA CA
X X X X X
it = α0 + α1,j ∆rt−j + α2,j it−j + η1,j ∆Yt−j + η2,j ∆Lt−j + η3,j ∆Nt−j + 1,t (6)
j=1 j=1 j=1 j=1 j=1

p p p p p
j j j j j
X X X X X
∆rt = β0 + β1,j ∆rt−j + β2,j it−j + µ1,j ∆Yt−j + η2,j ∆Lt−j + η3,j ∆Nt−j + 2,t (7)
j=1 j=1 j=1 j=1 j=1

Gomez-Gonzalez Mistery shoppers


Introduction
Brief literature review
Model specification and econometric approach
Data
Results

VAR approach for forecasting the fundamental value

VAR models are estimated by the method of maximum


likelihood. The companion form of the VAR system is given
by,
Zt = A0 + A1 Zt−1 + εt (8)
and the conditional forecast of Zt conditional on the
information set Ht , containing current and lagged values of
the VAR variables, is given by,

E(Zt+s |Ht ) = (I − As1 )(I − A0 )−1 A0 + As1 Zt . (9)

Gomez-Gonzalez Mistery shoppers


Introduction
Brief literature review
Model specification and econometric approach
Data
Results

VAR approach for forecasting the fundamental value

Using Equations 5 and 9, the optimal forecast of the


market fundamental component ft is given by
κ̃ − ϕ̂t 0 0 −1 −1 0 0 −1
E(ft |Ht ) = + (h1 − h2 )(1 − ρ) (I − ρA1 ) A0 + (h1 − h2 )A1 (I − ρA1 ) Zt , (10)
1 − ρ̃

where h1 and h2 are column vectors with their elements all


equal to zero, except for the element corresponding to rent
growth (∆rt ) and the real interest rate it .

Gomez-Gonzalez Mistery shoppers


Introduction
Brief literature review
Model specification and econometric approach
Data
Results

VAR approach for forecasting the fundamental value

Using the fundamental value obtained from Equation 10


and using Equation 1, we can compute the
non-fundamental component as the difference between the
log price-to-rent ratio and the fundamental component.

Gomez-Gonzalez Mistery shoppers


Introduction
Brief literature review
Model specification and econometric approach
Data
Results

Testing for bubbles


PSY show that a sufficient condition for the existence of a
bubble consist in the detection of an explosive behavior in
the price-to-dividend ratio.
They propose the application of recursive right-tailed unit
root tests using the price-to-rent ratio Pt /Dt .
The GSADF test for detecting the existence of at least one
bubble establishes a null hypothesis consisting on the
presence of unit root (or a martingale) as:
H0 : yt = kT −η + yt−1 + t with constant k and η > 0.5
versus a middly-explosive alternative hypothesis namely,
Ha : yt = δT yt−1 +t with δT = 1 + cT −α , c > 0 and α ∈ [0, 1)

Gomez-Gonzalez Mistery shoppers


Introduction
Brief literature review
Model specification and econometric approach
Data
Results

Testing for bubbles

The fitted regression model is the well-known ADF


equation with K lags and an intercept:
K
X
∆yt = α + βyt−1 + γi ∆yt−i + t (11)
i=1

As we are interested in detecting explosive behaviors in


different subsamples we perform this test recursively
following a rolling window method.
To determine the origination and collapse of each bubble,
we follow the date stamping strategy of PSY (2013), using
BSADF statistics.

Gomez-Gonzalez Mistery shoppers


Introduction
Brief literature review
Model specification and econometric approach
Data
Results

Migration methodologies

Let βˆj,s be the slope coefficient of equation 11 for province


j at an ending date of the subsample s = S, ..., T , and
ˆ the same slope coefficient for the province in which
βcore,s
the initial bubble originated.
We fit the following regression for each province:
 
s
β̂j,s = δj +γj β̂core,s−d +errors , ∀ j 6= core, for s = S, ..., T .
T −S+1
(12)
where S is the initial date. In this paper core is the
province of BC and d is a delay parameter.

Gomez-Gonzalez Mistery shoppers


Introduction
Brief literature review
Model specification and econometric approach
Data
Results

Migration methodologies

The time-varying coefficient function γj (r ) may be


estimated by local level kernel regression according to the
formula,
PT T
s=S Kh,s (r )β̃j,s β̃core,s−d 1 X
γ̂j (r ; h, d) = PT , β̃j,s := β̂j,s − β̂j,s ,
2 T −S+1
s=S Kh,s (r )β̃core,s−d s=S
(13)
where Kh,s (r ) = h1 K ( s/Th−r ). K (·) is a smooth kernel
function, and h is a bandwidth parameter.

Gomez-Gonzalez Mistery shoppers


Introduction
Brief literature review
Model specification and econometric approach
Data
Results

Data

Variable Source Frequency


Nominal New Housing Price Index Statistics Canada Monthly
Nominal Rental component CPI Statistics Canada Monthly
CPI excluding shelter Statistics Canada Monthly
Average house price value The Canadian Real Estate Association Only one value.
(CREA)
Average rental cost Canada Mortgage and Housing Corpora- Only one value.
tion (CMHC)
Real GDP Statistics Canada Quarterly.
Total retail trade SA, Index OECD statistics Monthly.
Real GDP by province Statistics Canada Annual.
Employed population S.A. Statistics Canada Monthly.
Population over 15 Statistics Canada Monthly.
Population estimates Statistics Canada Quarterly.
Marketable bond yields CANSIM and Bank of Canada Monthly.
Long term real and nominal bond yield CANSIM and Bank of Canada Monthly.

Data was collected from 1980 until 2017.

Table: Data sources by variable

Gomez-Gonzalez Mistery shoppers


Introduction
Brief literature review
Model specification and econometric approach
Data
Results

Data
Province House price Rental cost
CA $480, 253 $11, 544
NL $252, 646 $10, 380
NS $222, 808 $12, 096
NB $163, 630 $9, 156
QC $284, 652 $9, 012
ON $517, 245 $13, 848
MB $275, 914 $12, 396
SK $305, 008 $12, 624
AB $404, 107 $14, 316
BC $663, 469 $14, 580
We use the December 2016 value in current Canadian dollars.
(Source: CREA and CMHC).

Table: House price and rental cost

Gomez-Gonzalez Mistery shoppers


Introduction
Brief literature review
Model specification and econometric approach
Data
Results

Data

Figure: House log price-to-rent ratio by province


Gomez-Gonzalez Mistery shoppers
Introduction
Brief literature review
Model specification and econometric approach
Data
Results

Price-to-rent ratio decomposition

Figure: House Non Fundamental component by province

Gomez-Gonzalez Mistery shoppers


Introduction
Brief literature review
Model specification and econometric approach
Data
Results

Bubble detection

Figure: Price-to-Rent: GSADF Test by Province

Gomez-Gonzalez Mistery shoppers


Introduction
Brief literature review
Model specification and econometric approach
Data
Results

Bubble detection

Figure: Non-Fundamental value: GSADF Test by Province

Gomez-Gonzalez Mistery shoppers


Introduction
Brief literature review
Model specification and econometric approach
Data
Results

Date stamping
Price-to-Rent ratio Non-Fundamental component
Province
N. Dates N. Dates
1991M2-1991M11 (N) 1990M12-1991M6 (N)
CA 2 2002M3-2017M6 3 2008M9-2009M5
2016M6-2017M6
1997M4-1997M11 (N) 2008M2-2009M1
NL 3 1999M12-2000M5 1
2008M2-2011M6
1991M2-1991M7 (N)
NS 2 0
1994M4-1995M3
1991M2-1992M1 (N)
NB 2 0
1996M1-2001M9
2001M7-2017M6 1991M4-1991M9 (N)
QC 1 2
2014M3-2014M9
1991M2-1992M3 (N) 1990M12-1991M6 (N)
ON 3 2002M4-2008M11 3 1997M8-1998M2
2009M10-2017M6 2016M6-2017M6
MB 1 2002M7-2017M6 1 2003M6-2004M4

(N) refers to negative bubbles.

Table: Date-stamping for Price-to-Rent ratio and Non-fundamental


component

Gomez-Gonzalez Mistery shoppers


Introduction
Brief literature review
Model specification and econometric approach
Data
Results

Date stamping
Price-to-Rent ratio Non-Fundamental component
Province
N. Dates N. Dates
1991M2-1991M7 (N) 2007M8-2008M9
2004M9-2005M5
SK 4 1
2006M4-2012M4
2014M10-2017M6
2005M12-2008M1 1991M2-1991M7 (N)
1997M10-1998M3
AB 1 4
2005M11-2007M2
2008M5-2009M3
2003M7-2017M6 1990M12-1991M6 (N)
1995M2-1995M10 (N)
2003M10-2004M8
BC 1 6
2006M6-2006M11
2008M9-2009M7 (N)
2016M3-2016M10
(N) refers to negative bubbles.

Table: Date-stamping for Price-to-Rent ratio and Non-fundamental


component

Gomez-Gonzalez Mistery shoppers


Introduction
Brief literature review
Model specification and econometric approach
Data
Results

Migrations

Province PY Stat Period


SK 8.22∗∗∗ 2006M8 - 2007M6
AB 4.75∗∗∗ 2006M8 - 2007M6
NL 3.31∗∗∗ 2006M8 - 2008M12
ON 2.18∗∗ 2006M8 - 2017M6
Significance at 1%(***), 5% (**) and 10% (*)
Table: Phillips and Yu Migration Test: BC is the
origin

Gomez-Gonzalez Mistery shoppers


Introduction
Brief literature review
Model specification and econometric approach
Data
Results

Migrations

Figure: Time-varying migration coefficients of BSADF statistics

Gomez-Gonzalez Mistery shoppers


Introduction
Brief literature review
Model specification and econometric approach
Data
Results

Migrations

Figure: Time-varying migration coefficients of non-fundamental


component
Gomez-Gonzalez Mistery shoppers

You might also like