MASSACHUSETS INSTRWE
OF TECHNOLOGY
Essays in Financial Economics
MAY 15 2014
by
LIBRARIES
Felipe Severino
B.Sc., Pontificia Universidad Catolica de Chile, 2005
M.Sc., Pontificia Universidad Catolica de Chile, 2007
Submitted to the Alfred P. Sloan School of Management
in partial fulfillment of the requirements for the degree of
Doctor of Philosophy
at the
MASSACHUSETTS INSTITUTE OF TECHNOLOGY
June 2014
® Massachusetts Institute of Technology 2014. All rights reserved.
Signature redacted
...........................
Sloa
Alfred
Author................
chool of Management
May 2, 2014
Signature redacted
Antoinette Schoar
Certified by.......................
Michael Koerner '49 Professor of Entrepreneurial Finance
Thesis Supervisor
Signature redacted
Accepted by..........
.......
Ezra Zuckerman
Director, Sloan School of Management PhD Program
2
Essays in Financial Economics
by
Felipe Severino
Submitted to the Alfred P. Sloan School of Management
on May 2, 2014, in partial fulfillment of the
requirements for the degree of
Doctor of Philosophy
Abstract
This thesis consists of three empirical essays in financial economics, examining the
consequences of imperfect financial markets for households, small business and house
prices. In the first chapter (co-authored with Meta Brown and Brandi Coates) we explore the effect of personal bankruptcy laws on household debt. Personal bankruptcy
laws in the US, and many other countries, protect a fraction of an individual's assets from seizure by unsecured creditors in case of default. An increase in the level
of bankruptcy protection diminishes the collateral value of assets, and can therefore
reduce borrowers' access to credit. However, it might also increase the demand for
credit especially from risk averse borrowers by improving risk-sharing. Using changes
in the level of protection across US states and across time, we show that bankruptcy
protection laws increase borrowers' holdings of unsecured credit, but leave secured
debt -mortgage and auto loans- unchanged. At the same time we find an increase in
the interest rate for unsecured credit, but not for other types of credit. The effect is
predominantly driven by lower-income areas and regions with higher home ownership
concentration, for which an increase in the level of protection explains between 10%
and 30% of the growth in their credit card debt. Using detailed individual data,
we find no measurable increase in delinquency rates of households in the subsequent
three years. These results suggest that changes in bankruptcy protections did not
reduce the aggregate level of household debt, but they might have affected the composition of borrowing. In the second chapter (co-authored with Manuel Adelino and
Antoientte Schoar) we document the role of the collateral lending channel in small
business employment and self-employment in the period before the financial crisis of
2008. Small businesses in areas with a bigger run up in prices experienced a stronger
increase in employment than large firms in the same industries. This increase in small
business employment was more pronounced in industries that need little startup capital and can be financed more easily using housing as collateral. The increase is not
limited to the non-tradable sector and is also present in manufacturing industries,
in particular in those that ship goods over long distances. This indicates that this
channel is separate from the aggregate demand channel by which home equity based
borrowing leads to higher demand and employment creation. In aggregate, the collat3
eral lending channel explains 15-25 % of employment variation. In the third chapter
(co-authored with Manuel Adelino and Antoinette Schoar) we use exogenous changes
in the conforming loan limit as an instrument for lower cost of financing, and show
that cheaper credit significantly increases house prices. Houses that become eligible
for financing with a conforming loan show an increase in value of 1.16 dollars per
square foot (for an average price per square foot of 220 dollars). These coefficients
are consistent with a local elasticity of house prices to interest rates that is lower than
some previous studies proposed (below 10). In addition, loan to value ratios around
the conforming loan limit deviate significantly from the common 80 percent norm,
which confirms that it is an important factor in the financing choices of home buyers.
In line with our interpretation, the results are stronger in the first half of our sample
(1998-2001) when the conforming loan limit was more important, given that other
forms of financing were less common and substantially more expensive. Results are
also stronger in zip codes where personal income growth is low or declining, and in
regions with lower elasticity of housing supply.
Thesis Supervisor: Antoinette Schoar
Title: Michael Koerner '49 Professor of Entrepreneurial Finance
4
Acknowledgments
I always thought that writing the acknowledgments to my thesis was not going to be
easy, because I received encouragement and support from so many people along the
way. Even if they are not mentioned here, I am truly grateful to each of them.
I am deeply indebted to Antoinette Schoar: she has been an outstanding mentor.
Her advice, comments and support were always insightful; our many discussions and
conversations largely shaped the way I now think about research and finance. She has
always been there. Working with her and learning from her has been a true privilege.
I am extremely grateful to Nittai Bergman and Andrey Malenko, who provided
invaluable advice. They always pushed me to deepen my understanding and focus on
the important things. I also want to thank Xavier Giroud for his constant support
and willingness to help. I also benefited from discussion and guidance with Hui Chen,
John Cox, Sharon Cayley, Raj Iyer, Leonid Kogan, Gustavo Manso, Jun Pan, Stephen
Ross, Hillary Ross, Adrien Verdelhan and Jiang Wang. Thanks you all for your time
and dedication to make me a better researcher.
My research has benefited from working with many people; my conversations with
Manuel Adelino helped me understand the way research works. I will also want to
thank Meta Brown and the Federal Reserve Bank of New York for their generous
support. I cannot fail to mention my undergrad professors that encouraged me to
start this adventure, especially Jaime Casassus, Gonzalo Cortazar and Nicolas Majluf.
I am also grateful to Patricio Agusti, for his support during my first undergrad years.
I had the great pleasure of sharing my experience with an incredible group of
friends. I can still remember the first years, crammed into in the study room trying
to make sense of our problem sets. I am very grateful to Marco Di Maggio, Sebastian
Di Tella, Juan Passadore, Vicent Pons, Yang Sun, Tyler Williams, Luis Zermeno
and especially to Will Mullins thanks a lot for always being there. Their help and
friendship are something that I will always remember with affection, and I hope it
will continue in the future.
I have always felt the love and support of my family. I want to thank my parents, Fernando Severino and Fresia Diaz, for always believing in me, and for their
encouragement to always give the best of me: you taught me all that I know, and
are a true inspiration. To my brother and sister, Fernando and Francisca, for many
years of friendship, conversation and joy together. To my daughter, Ema, and my
son, Mateo, for bringing that special and unique happiness to my life: when you smile
nothing else matters, and I feel truly blessed to have you.
Last, but certainly not least, I would like to thank my wife Daniela Agusti. She
has been by my side every step of the way. Since the beginning you believed in me,
and left everything that was important to you to start this adventure with me. These
have been years of hard work, but also of wonderful experiences, but none of this
would have been the same without you. You make me want to be a better man.
Thank you for everything that you have done. For your unconditional support and
love, I will be forever grateful.
5
To Daniela, Ema and Mateo.
... en la calle codo a codo somos mucho mas que dos
... "
(Mario Benedetti)
6
Contents
1
13
13
19
19
22
24
24
26
27
29
32
Personal Bankruptcy and Household Debt
1.1
1.2
1.3
1.4
1.5
1.6
1.7
1.8
1.9
...................
Introduction ....
Bankruptcy Procedure and Related Literature
1.2.1 Institutional Framework . . . . . . . .
1.2.2 Related Literature . . . . . . . . . . .
Data and Summary Statistics . . . . . . . . .
1.3.1 Data Description . . . . . . . . . . . .
1.3.2 Summary Statistics . . . . . . . . . . .
Empirical Hypothesis . . . . . . . . . . . . . .
Empirical Strategy . . . . . . . . . . . . . . .
Results and discussion . . . . . . . . . . . . .
1.6.1 Bankruptcy Protection and Household Leverage and Interest
Rates . . . . . . . . . . . . . . . . . .
1.6.2 Robustness Test . . . . . . . . . . . . .
1.6.3 Magnitude of the effect .
1.6.4 Borrowers, Delinquency and Self-Employment .
Conclusion . . . . . . . . . . . . . . . . . . . . . . . .
.
Bibliography . . . . . . . . . . . . . . . . . . . . . . .
Appendix A. Model of Effect of Bankruptcy Protection on Household
B orrow ing . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
2 House Prices, Collateral and Self-Employment
2.1
2.2
2.3
2.4
2.5
Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Data and Empirical Methodology . . . . . . . . . . . . . . . .
2.2.1 Data Description . . . . . . . . . . . . . . . . . . . . .
2.2.2 Summary Statistics . . . . . . . . . . . . . . . . . . . .
2.2.3 Empirical Model . . . . . . . . . . . . . . . . . . . . .
Empirical Results . . . . . . . . . . . . . . . . . . . . . . . . .
2.3.1 House Prices and Employment at Small Establishments
2.3.2 Sole Proprietorships . . . . . . . . . . . . . . . . . . .
2.3.3 Crisis Period (2007-2009) . . . . . . . . . . . . . . . .
2.3.4 M igration . . . . . . . . . . . . . . . . . . . . . . . . .
2.3.5 Credit Conditions and Elasticity of Housing Supply . .
C onclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Bibliography . . . . . . . . . . . . . . . . . . . . . . . . . . . .
7
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
32
34
35
35
37
38
42
73
73
77
77
80
81
84
84
90
90
91
91
92
93
2.6
3
Appendix. Calculating the magnitude of the collateral effect
105
Credit Supply and House Prices: Evidence from Mortgage Market
Segmentation
115
3.1 Introduction . . . . . . . . . . . . . . . . . .
115
3.2
3.3
3.4
3.5
3.6
3.7
3.8
The User Cost Model . . . . . . . . . . . . .
Data and Methodology . . . . . . . . . . . .
3.3.1 Summary Statistics . . . . . . . . . .
3.3.2 Hedonic Regression . . . . . . . . . .
3.3.3 Empirical Approach . . . . . . . . .
Cost of Credit and House Prices . . . . . . .
3.4.1 Main Regression Results . . . . . . .
3.4.2 Credit Supply and Income . . . . . .
3.4.3 Robustness and Refinements . . . . .
3.4.4 Economic Magnitude of the Effect . .
Conclusion . . . . . . . . . . . . . . . . . . .
Bibliography . . . . . . . . . . . . . . . . . .
Appendix A. Robustness and Refinements - .dditional Tests
3.7.1 Restrict LTV Choices . . . . . . . . .
3.7.2 Different Bands . . . . . . . . . . . .
3.7.3 Timing of the Control Group . . . .
3.7.4 Pos-October Effect . . . . . . . . . .
3.7.5 Value per Square Foot by ZIP Code Income
Appendix B. Data Manipulation . . .
3.8.1
Data Cleaning . . . . . . . . .
3.8.2
Variable Construction
. . . .
8
119
120
120
121
122
128
128
129
130
133
135
137
153
153
153
154
154
154
155
155
157
List of Figures
1-1
1-2
1-3
1-4
Debt Growth and Bankruptcy Filings . . . . . . . . . . . . . . .
States that Changed their Level of Bankruptcy Protection . . .
Ilustration of Different Demand and Supply Responses . . . . .
Ilustration of a Solution of the Model . . . . . . . . . . . . . . .
.
.
.
.
.
.
.
.
.
.
.
.
44
45
46
47
3-1
3-2
3-3
3-4
3-5
3-6
3-7
Transaction-Loan Value Surface . . . . . . . . . . . . . . . . . .
Borrower Composition for the Regression Sample . . . . . . . .
Frequency of Transactions as Percentage of CLL Threshold . . .
Share of Unused Mortgage Applications . . . . . . . . . . . . . .
Fraction of Transactions with a Second Lien Loan by Year . . .
Value per Square Foot by House Value and by ZIP Code Income
Income as a Percentage of CLL Threshold . . . . . . . . . . . .
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
139
140
141
142
160
161
162
9
10
List of Tables
1.1
1.2
1.3
1.4
1.5
1.6
1.7
1.8
1.9
1.10
1.11
1.12
1.13
1.14
1.15
1.16
1.17
1.18
1.19
1.20
1.21
1.22
1.23
1.24
Summary Statistics Data. . . . . . . . . . . . . . . . . . . . . . . . .
Summary Statistics Protection Level . . . . . . . . . . . . . . . . . .
Effect of Bankruptcy Protection on Debt. Credit Card Debt . . . . .
Effect of Bankruptcy Protection on Debt. Mortgage Debt . . . . . . .
Effect of Bankruptcy Protection on Debt. Auto Debt . . . . . . . . .
Determinants of Bankruptcy Protection Levels and Changes . . . . .
Dynamics of the Change in Protection Levels on Credit Card Debt .
Local Business Conditions. Neighboring County-pairs across State
Borders. Credit Card Debt . . . . . . . . . . . . . . . . . . . . . . . .
Heterogeneous Treatment of Bankruptcy Protection on Credit Card
Debt: Income and Home ownership . . . . . . . . . . . . . . . . . . .
Effect of Bankruptcy Protection on Interest Rates: Personal Unsecured
Loans and Credit Cards . . . . . . . . . . . . . . . . . . . . . . . . .
Effect of Bankruptcy Protection on Interest Rates: Mortagage Credit
Effect of Bankruptcy Protection on Debt. Number of Credit Cards
and E ntry . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Effect of Bankruptcy Protection on Credit Card Delinquency . . . . .
Effect of Bankruptcy Protection on Self-Employment . . . . . . . . .
Effect of Bankruptcy Protection on Credit Card Debt. Alternative
Specifications . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Other Heterogeneous Treatment of Bankruptcy Protection. Credit
48
49
50
51
52
53
54
C ard D ebt . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
63
Determinants of Bankruptcy Protection Levels and Changes. Eventually Treated . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Dynamics of the Change in Protection. Mortgage Debt . . . . . . . .
Dynamics of the Change in Protection. Auto Debt . . . . . . . . . .
Local Business Conditions. Neighboring County-pairs across State
Borders. Mortgage Debt . . . . . . . . . . . . . . . . . . . . . . . . .
Local Business Conditions. Neighboring County-pairs across State
Borders. Auto Debt . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Heterogeneous Treatment of Bankruptcy Protection: Income and Homeownership. Mortgage Debt . . . . . . . . . . . . . . . . . . . . . . . .
Heterogeneous Treatment of Bankruptcy Protection: Income and Homeownership. Auto Debt . . . . . . . . . . . . . . . . . . . . . . . . . .
Effect of Bankruptcy Protection on County Delinquency Proportions
11
55
56
57
58
59
60
61
62
64
65
66
67
68
69
70
71
1.25 Effect of Bankruptcy Protection on Debt After Bankruptcy Reform 2005 72
2.1
Sum mary Statistics . . . . . . . . . . . . . . . . . . . . . . . . . . . .
96
2.2
Employment Growth, Firm Size, and House Price Appreciation
97
2.3
Employment Growth and House Prices: Excluding Construction, NonTradable, and Finance Industries and Considering Manufacturing Only
Breakdown of Manufacturing Industries by Distance Shipped . . . . .
Employment and House Price Appreciation across Industry Types . .
Proprietorships and House Price Appreciation . . . . . . . . . . . . .
Employment Growth, Firm Size, and House Price Appreciation, Crisis
Period (2007-2009) . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Total Employment, Unemployment, and Migration . . . . . . . . . .
Denial Rates . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Employment Growth, Firm Size, and House Price Appreciation: Individual Industries by Firm Size . . . . . . . . . . . . . . . . . . . . . .
Robustness Test: Difference between High and Low Start-up Capital
Effect of One Standard Deviation Change in the Independent Variable
Dollar-weighted Average Distance Shipped in Manufacturing (miles) .
Detail on Average Start-up Amount by 2-digit NAICS Sector . . . . .
Distance Shipped and Share of Employees at Large Establishments .
House Price Growth and Creation of Establishments . . . . . . . . . .
List of 3-digit NAICS Industries Excluding Non-tradables, Manufacturing, F.I.R.E., and Construction . . . . . . . . . . . . . . . . . . . .
2.4
2.5
2.6
2.7
2.8
2.9
2.10
2.11
2.12
2.13
2.14
2.15
2.16
2.17
3.1
3.2
3.3
3.4
3.5
3.6
3.7
3.8
3.9
3.10
3.11
3.12
3.13
3.14
3.15
3.16
. . .
98
99
100
101
102
103
104
107
108
109
110
111
112
113
114
Sum mary Statistics . . . . . . . . . . . . . . . . . . . . . . . . . . . . 143
Summary Statistics by Geography and Year . . . . . . . . . . . . . . 144
Verification of the Impact of the CLL on Financing Choices . . . . . . 145
Impact of CLL on Number of Transactions . . . . . . . . . . . . . . . 146
Effect of the CLL on House Valuation Measures . . . . . . . . . . . . 147
Effect of the CLL on House Valuation in Different Income Growth Areas148
Placebo Test for Coefficient of Interest . . . . . . . . . . . . . . . . . 149
Effect of the CLL on the Valuation of Different Groups of Transactions 150
Effect of the CLL on House Valuation in Low Supply Elasticity Areas
( Elasticity< 1) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 151
Elasticity Estim ates . . . . . . . . . . . . . . . . . . . . . . . . . . . . 152
Data Cleaning Description . . . . . . . . . . . . . . . . . . . . . . . . 155
Effect of the CLL on House Valuation Measures, Constrained Sample
(0.5<LTV < 0.8) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 163
Effect of CLL on Valuation Measures - Alternative Timing of the Control G roup . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 164
Effect of the CLL on Valuation - Alternative Bands . . . . . . . . . . 165
Effect of CLL on Valuation: Post October . . . . . . . . . . . . . . . 166
Effect of the CLL on House Valuation with In-Sample Controls . . . . 167
12
Chapter 1
Personal Bankruptcy and
Household Debt
1.1
Introduction
The last two decades in the US have seen a massive increase in household leverage,
from 320 billion dollars in 1994 to 1060 billion dollars in 2010, and at the same time
1
an increase in personal bankruptcies, which peaked in 2005 with 2.04 million filings.
These trends have brought renewed attention from academics and policy makers on
the role that bankruptcy rules play in helping people manage their debt load, but
also the incentives they provide to take on leverage in the first place.
Personal bankruptcy laws in the US protect a fraction of a household's assets from
seizure by unsecured creditors; under Chapter 7 bankruptcy, households are protected
from creditors up to a monetary limit set by each state - the personal bankruptcy
exemption. An increase in the level of this exemption (referred to as protection
henceforth) may strengthen the demand for credit but can also decrease the supply
of credit. In case of default, the lender cannot seize the borrower's assets if their
value does not exceed the protection level dictated by law, while if they do the lender
can only seize the excess value. Consider any simple model of a credit market with
financially constrained, risk-averse borrowers, and a risk-neutral lender. If borrowers
have a stochastic income, increased bankruptcy protection makes defaulting attractive
to borrowers in more states of the world. As a result it diminishes the collateral value
of assets, forcing lenders to charge a higher interest rate ex ante to break even (Hart
and Moore, 1994). Therefore, this is akin to reducing the supply of credit, increasing
prices, and/or reducing quantities. In addition, such a change in the supply of credit
could increase the riskiness of the pool of loan applicants; increases in lending rates
might foster borrowers' incentives to undertake riskier projects, or could intensify the
entry of riskier borrowers (Stiglitz and Weiss, 1981)2.
'Debt amounts converted to year 2000 constant dollars to reflect change adjusted by inflation,
see Figure .1-1
2
Furthermore, lenders' willingness to supply credit will vary depending on their ability to screen
borrowers.
13
Most of the existing empirical literature has focused on the effects described above
that tend to reduce the supply of credit (the seminal paper in the area is Gropp et al.,
1997). However, a higher protection level will also improve risk-sharing by increasing
the insurance function of bankruptcy: in bad states of the world the borrower declares
bankruptcy and, as a result of the higher protection level, is allowed to keep a larger
proportion of their assets - the protection amount (Dubey et al. 2005, Zame 1993)3.
This increases the demand for credit at a given interest rate. Changes in the level of
protection will also affect the composition of borrowers: more risk averse borrowers
might choose to use more debt since they weight the loss of their assets more severely.
Therefore, an increase in level of asset protection might also lead to a change in the
mix of borrowers, but in this case by drawing in new (more risk-averse borrowers),
or by encouraging existing borrowers to take on more debt. Interest rates must
therefore rise in equilibrium; but depending on which effect dominates (demand or
supply), there can be an increase or decrease in the amount of credit extended.4
We use the timing of state changes in the levels of bankruptcy protection in a
difference in difference design to identify their effect on the credit market equilibrium. We find that bankruptcy protection laws increase borrowers' unsecured credit
holdings, mainly credit cards, leaving their level of secured debt - mortgage and auto
loans - unchanged. At the same time we find an increase in the interest rate for
unsecured credit, but not for other types of credit. These results are predominantly
driven by low-income areas, and suggest that bankruptcy protection levels provide
important downside insurance, which has first order effects on the supply and also
on the demand for credit. Interestingly, using detailed individual data, we do not
find an increase in default rates, which suggests that households are not necessarily
over-borrowing or risk shifting as a response to the increase in protection.
Empirically identifying the true effect of bankruptcy protection levels on household
leverage is challenging, as these levels are correlated with unobservable borrower and
lender characteristics that might simultaneously affect credit availability, and the
level of protection. For example, states with higher protection levels may be states
in which households are less financially savvy, or they might be states with higher
house prices, and therefore more willing to take on more debt. This in turn will lead
to a positive correlation between debt and protection.
Therefore, we exploit changes in the dollar amounts of asset protection under
bankruptcy to identify the effect of this protection on household debt 5 . Our identification benefits from the fact that states increased bankruptcy protection at different
times and by different amounts over our sample period. We show that changes in
3
Non-state contingent contracts are a key friction here; in the absence of this friction, the effect
of personal bankruptcy protection on household borrowing disappears. One possible explanation for
why lenders do not offer more flexible contracts (more protection in "bad" states, or state contingent
repayment) is that these lenders could face a collective action problem: if only one lender offered
such a contract it would attract predominantly bad type borrowers, which is not an equilibrium.
Alternatively, customized state contingent contracts could be hard to enforce.
4
For a more developed model see Appendix A.
5
Asset protection in our empirical implementation is the sum of homestead exemption and
personal assets exemption levels for each state and year. Our results are invariant to the use of only
homestead exemption.
14
protection levels are uncorrelated with macroeconomic conditions and other determinants of credit equilibrium, most importantly changes in state level house prices
and unemployment rates. This allows us to disentangle the effect of bankruptcy protection levels on household leverage from other determinants of household debt that
may be changing as well.
We then estimate the effect of the changes in the levels of protection on changes
in household debt. In doing so, we compare the change in the level of household debt
between counties in a state that increases the level of protection between t and t+1,
with other counties in a state that did not change their level of protection during
the same period. The variation in bankruptcy protection changes over time and
across states, which helps us to deal with two crucial assumptions of any difference in
difference estimator. First, that the timing of the changes in the levels of protection
are uncorrelated with other determinants of household leverage, as discussed above.
And second, that after controlling for observed time-varying characteristics, linear
county trends, and time-invariant county characteristics, changes in protection at the
state level only affect the states which adopted the change, making the exogenous
change in the level of protection the only determinant of the difference in household
debt across states. Our empirical strategy is therefore similar to Cerqueiro and Penas
(2011) and Cerqueiro et al (2013) who examine the effect of bankruptcy protection
on start-up performance and innovation respectively.
Our results show that the exogenous variation in the levels of protection causally
increases the level of credit card debt held by households during our sample period
(1999-2005)6, leaving secure debt (mortgage and auto) unchanged. This is consistent
with the fact that personal bankruptcy allows households to discharge only unsecured
debt 7 . Using novel bank branch-level data on credit rates for different types of credit,
we explore the effect of bankruptcy protection changes on interest rates, and we
find that an increase in bankruptcy protection leads to an increase in the interest
rate on unsecured credit, which is consistent with a credit market equilibrium, where
supply decreases and demand increases but the net effect is dominated by the demand
response.
A possible concern may be that states which did not change the level of protection
within our sample period are not a good control group, as they could be systematically different from the group which did opt to change their level of bankruptcy
protection, and this would therefore invalidate our empirical inference. However, the
staggered nature of our empirical strategy, whereby each state which changed its level
of protection is a control for past and future periods for other changes, allows us to
6
We focus on the Pre-Bankruptcy Abuse Prevention and Consumer Protection Act of 2005
(BAPCPA), where the cost of filing for bankruptcy was low, and therefore the intensity of the
treatment was higher. The bankruptcy reform makes the process of filing for bankruptcy harder,
which ex ante diminished the incentives to take on more credit. The nature of the subprime crisis
of 2007 and financial shock of 2008 may have affected household willingness to take on credit, and
lenders' ability to supply credit, contributing to the lack of the effect during the post-reform period.
We empirically investigate this by extending our sample until 2009; we find that changes in the law
have no effect on unsecured debt held after the reform, see Appendix B8.
7The fact that the levels of protection only affect unsecured credit holdings helps to rule out
that protection levels do not endogenously increase when the credit market becomes looser.
15
replicate our findings focusing only on the states where changes in protection levels
were implemented in our sample period (i.e. "eventually" treated). In this case the
effects we estimate are unchanged.
We also look at the dynamics of the changes. By analyzing the timing, we can
rule out that the level of protection may be correlated with pre-existing state specific
trends that survive our controls, and thus that our results are a reflection of these
differential pre-trends rather than changes in the levels of protection. We show that
our estimates are not affected by the inclusion of lag changes in the levels of protection,
and that the coefficients on the lags are small in economic terms, and statistically
insignificant.'
We now explore the heterogeneity of the average treatment effect. Exploiting
within-state variation on the levels of debt held by counties, we find a stronger increase in the level of unsecured debt held by lower-income counties'. These results
are consistent with the fact that increases in personal bankruptcy protection levels
improve risk-sharing; this improvement should be stronger for lower-income regions,
as they have fewer resources to diversify their risk exposure than wealthier ones, for
which the differential impact of the increase should be smaller.
Personal bankruptcy levels of protection are heavily concentrated on home equity;
a big fraction of the protected nominal amount is exclusively linked to the home equity
of the borrower. In line with a demand driven channel, we find that the effect is almost
three times stronger in areas where homeownership is higher, after we condition on the
level of income. Also conditioning on the level of income10 , we find that the increase
in credit is stronger in areas where the banking industry is more concentrated (fewer
banks), which is consistent with the relationship lending model proposed Petersen
and Rajan (1995), where creditors are more likely to finance a credit constrained
borrower when credit markets are concentrated, because it is easier for these creditors
to internalize the benefits of assisting these borrowers; although this is only suggestive
evidence.
Overall we find that the average credit card balance in a county in our period is
290 million dollars in credit card debt, and the average increase in credit card debt is
7.6%. Our main estimate explains 10% of this balance growth". However, this value
more than triples for low-income homeowners and for our micro-level sample, for
8
Considering that our exogenous variation is at the state level, we cannot control for state-time
unobserved heterogeneity that is contemporaneous to the effects we observe.
9
Within each state, counties are divided into terciles based on total wages and salary levels in
1999.
' 0Homeownership and bank concentration are correlated with income at the county level. Therefore, looking at cross-sectional variation without controlling for income is not informative, as it
provides confounding information within all the correlated variables. In order to overcome this limitation of the data, we replicated the specification of interest for each income subgroup; this strategy
proved to be useful. For example, under this setup, unemployment heterogeneity within income
groups has no cross-sectional implications. However, homeownership and bank concentration still
provide meaningful variation within income groups.
" 1This percentage is estimated using the average change in protection in our sample period,
approximately 40k dollars, which represents a 54% change with respect to the average exemption
level of 70k dollars. This value is a more conservative measure than using one standard deviation of
level (70k dollars).
16
which our estimate explains 34% and 47% respectively of the increase in credit card
balance. This heterogeneity seems to suggest that this affects only a subset of people:
2
homeowners who are expecting to be close to distress level on their credit cards .
There is also the possibility that our estimates are biased downward (attenuation
bias), due to measurement errors of our treatment.
Finally, local economic conditions could produce spurious effects due to geographical heterogeneity that is uncorrelated to changes in the levels of protection. To
overcome this endogeneity we compare neighboring county-pairs across state borders,
within the same income bucket. The results of the estimation of changes in protection
within each county-pair are very similar to the main estimates, and stronger when
we concentrate on county-pairs in the lower end of the county income distribution.
The aggregate results raise important questions about how credit expands in response to bankruptcy protection, and by whom; and whether it affects the overall
composition and default probability of borrowers. We use detailed individual data
containing debt levels and specific account information to understand and empirically
test household behavior. We find that changes in protection levels increase the number of credit cards per household; this increase is stronger among households that had
ex ante credit card accounts and those that had a positive balance. Finally, changes
in protection are uncorrelated with entry into the credit card market, defined as the
time when a member of a household opens their first account, or as the time when
a credit card balance goes from zero to positive. All these results provide evidence
that in this sample, the effect is driven by existing debtors expanding their current
balance, or their number of accounts, rather than new households entering the credit
market.
Focusing on the same sample, we explore their delinquency behavior up to three
years after the increase in credit card usage induced by the change in protection.
Within this sample there is no measurable increase in the level of delinquency; if
anything, the probability of being delinquent in the future decreases. If the households which are increasing their level of debt are over-borrowing, or taking on more
risky projects, we would expect delinquency rates to increase. Although we cannot
completely rule out over-borrowing or risk shifting behavior, the results described are
more consistent with risk-averse borrowers increasing their debt as a result of the
13
increase in downside protection
Furthermore, using county self-employment information, we show that areas that
experienced an increase in the level of credit card debt also experienced an increase
in the level of self-employment creation, specifically in industries that use more credit
cards as start-up capital". It is important to point out that these outcome variables
are only suggestive evidence of the real effect of the increase in the level of unsecured
Appendix B2 shows that within low-income areas the effect is differentially stronger for areas
with a higher proportion of credit card delinquency (90+).
13
Also, at the county level, delinquency rates do not seem to increase, which implies that also
at the aggregate level, increases in the level of protection did not lead to an increase in the level of
delinquencies.
4
1 For example, construction, photography, and other low capital-intensity industries that can be
financed with credit card debt.
12
17
debt, as they represent county aggregates.
The results are also robust to restricting the sample to states which changed the
level of protection only once during the sample period, to considering only states with
large changes in protection as treated states, and to the use of an indicator instead of
the magnitude of the change. Given the nature of our empirical strategy, as we argue
before, time-varying changes at state levels may be omitted variables explaining our
results; one candidate is the level of unemployment insurance in each state (Hsu et
al., 2012). However, the inclusion of this variable has no impact on the estimated
coefficient. 5
Our results suggest that existing borrowers increase their leverage without increasing their ex post delinquency, consistent with risk-averse, constrained borrowers
reacting to the increase in insurance. We cannot say anything about the welfare effect of these changes. In a world with complete markets, increases in protection will
constrain the contract space and therefore may lead to inefficiencies. Furthermore, in
the presence of limited commitment, harsher penalties for defaulting could improve
welfare ex ante (Kehoe and Levine 1993, Alvarez and Jermann 2000). However, if
state contingent contracts are not available (i.e. incomplete markets), a pro-debtor
bankruptcy code could lead to welfare gains (Link 2004). Therefore, theoretically the
effect of increased bankruptcy protection on welfare is undetermined, and dependent
on modeling choices.
A number of earlier papers have looked at the cross sectional relationship between
the level of bankruptcy protection and consumer credit. See for example Gropp et
al. (1997), the first paper to examine this relationship. Using household data from
the 1983 Survey of Consumer Finances, they found that higher levels of protection
were associated with both reduced credit availability for low-asset households and increased debt balances among higher-asset households. Similarly Berger et al. (2010)
found that higher protection is associated with lower access to credit for unlimited
liability firms. Also, Lin and White (2001) found the same relationship for mortgage
credit. The recent legislative history of staggered introduction of bankruptcy exemptions in combination with household data allows us to identify the effects of changes
in bankruptcy protection on the change in the supply and demand of credit for different types of debt. Most importantly, we find that an increase in personal bankruptcy
protection leads to an increase in the amount of unsecured debt held by households,
leaving secured debt unchanged. Therefore, using an improved empirical strategy, we
see that the demand effect of bankruptcy protection, arguably driven by improved
risk-sharing, dominates its supply-deterring effects. Hence increased bankruptcy protection increases equilibrium debt reliance, particularly for low-income homeowners.
Increases in personal bankruptcy protection results in a weakening of creditor
rights. There is a vast literature in corporate finance that has examined the effect of
15
As a case study during our relevant sample period, 1999-2005, one state went from having
some level of protection to unlimited protection. When we include this time-varying dummy in
the regression, we find that the main effect is unchanged, but the unlimited protection dummy is
negative and significant for mortgage and credit card debt. This suggests that the effect of protection
is a non-linear function of the level of exemption, and therefore above a certain threshold lenders
increase prices to a magnitude which decreases quantities.
18
changes in creditor protection on debt (La Porta et al. 1998, Levine 1998, Djankov
et al. 2007). Most related to this paper is Vig (2013), which looks at increases
in the seizability of assets for large firms in India, and how this triggers a drop
in the demand for secured debt. Vig (2013) suggests that this demand response
is driven by an increase in the threat of early liquidation due to the increase in
creditor protection. Our paper focuses on a different channel, i.e. changes in the
self-selection of households with different risk aversion levels, or their willingness to
default strategically.
The rest of the paper proceeds as follows: Section 2 explains the institutional
framework of personal bankruptcy laws and related existing literature; Section 3
outlines the empirical hypothesis with a theoretical focus; Section 4 describes the
data; Section 5 develops the empirical strategy; and Section 6 shows the results
before the conclusion.
1.2
1.2.1
Bankruptcy Procedure and Related Literature
Institutional Framework
Personal bankruptcy procedures determine both the total amount that borrowers
must repay their creditors and how repayment is shared among individual creditors.
An increase in the amount repaid may benefit all individuals who borrow, because
higher repayment levels may cause creditors to lend more, and at lower interest rates.
However, a larger repayment amount implies that borrowers need to use more of
their existing assets and/or post-bankruptcy earnings to repay pre-bankruptcy debt,
therefore reducing their willingness to borrow and their incentive to work 6 .
US bankruptcy law has two separate personal bankruptcy procedures, which are
named as they appear in bankruptcy law, Chapter 7, and Chapter 13. Under both
procedures, creditors must immediately terminate all efforts to collect from the borrower (such as letters, wage garnishment, telephone calls, and lawsuits). Most consumer debt is discharged in bankruptcy, however most tax obligations, student loans,
allowance and child support obligations, debts acquired by fraud, and some credit
card debt used for luxury purchases or cash advances are not.
Mortgages, car loans, and other secured debts are not discharged in bankruptcy,
but filing for bankruptcy generally allows debtors to delay creditors from retrieving
assets or foreclosure. Prior to the Bankruptcy Abuse Prevention and Consumer Protection Act of 2005 (BAPCPA), debtors were allowed to freely choose between the
two.
Bankruptcy Law Before 2005
The most commonly used procedure before 2005 was Chapter 7. Under it, bankrupts
must list all their assets. Bankruptcy law makes some of these assets exempt, meaning
that they cannot be seized by creditors. Asset exemption amounts are determined by
16 See Dobbie and Song (2013) for a more detailed description of this issue.
19
the state in which the borrower lives. Most states will have personal asset protection,
which exempts debtors' clothing, furniture, "tools of the trade", and sometimes equity
in a vehicle. In addition, nearly all states have some level of homestead protection for
equity in owner-occupied homes, but the levels vary from a few thousand dollars, to
unlimited amounts in six states, including Texas, Florida, and DC". This exemption
level is what we refer to here as the protection level. Under Chapter 7, debtors must
use their non-protected assets to repay creditors, but they are not obliged to use any
of their future income to make repayments.
Under the alternative procedure in Chapter 13, bankrupts are not obliged to
repay from assets, but they must use part of their post-bankruptcy income to make
repayments. Before 2005, there was no predetermined income exemption; on the
contrary, borrowers who filed under Chapter 13 proposed their own repayment plans.
They often proposed to repay an amount equal to the value of their non-protected
assets under Chapter 7. Also, borrowers were not allowed to repay less than the
value of their non-protected assets and, since they had always the option to file under
Chapter 7, they had no incentive to offer any more. Judges did not need the approval
of creditors to approve repayment plans.18
The cost of filing for bankruptcy before 2005 was low: about 600 dollars under
Chapter 7, and 1,600 dollars under Chapter 13, as of 2001 (White 2007). The punishment for bankruptcy included making bankrupts' names public and the appearance
of the bankruptcy filing on their credit records for 10 years subsequently. In addition,
bankrupts were not allowed to file again under Chapter 7 for another six years, (but
they were allowed to file under Chapter 13 as often as every six months) 1 9 .
Overall, these features made US bankruptcy law very pro-debtor. Since debtors
could choose between the procedures under Chapters 7 and 13, they would select the
procedure which would maximize their gain from filing. Around three quarters of all
those filing for bankruptcy used Chapter 7 (Flynn and Bermant, 2002). Most debtors
who filed under Chapter 13 did so because their gains were even higher using this
17
See Table 1.2 for summary statistics of the level of protection.
Even when households file under Chapter 13, the amount that they are willing to repay is
affected by Chapter 7 bankruptcy protection. For example, suppose that a household that is considering filing for bankruptcy has 40,000 dollars in assets and is located in a state in which the
protection level is 20,000 dollars. Since the household would have 20,000 dollars of unprotected
assets if filing under Chapter 7, it would be willing to repay no more than 20,000 dollars (in present
value) from future income if it were to file under Chapter 13. As a result of this close relationship between Chapter 7 and Chapter 13 bankruptcy filings, we assume that changes in Chapter 7
protection levels will affect household willingness to file for bankruptcy (either under Chapter 7 or
13).
19
US bankruptcy law allowed additional debt to be discharged under Chapter 13. Debtors' car
loans could be discharged to the extent that the loan principal exceeded the market value of the
car (negative equity). Also, debts acquired by fraud and cash advances obtained shortly before
filing could be discharged under Chapter 13, but not under Chapter 7. These characteristics were
known as the Chapter 13 "super-discharge", and some households took advantage of the situation
by filing first under Chapter 7, where most of their debts were discharged, and then converting their
filings to Chapter 13, where they proposed a plan to repay part of the additional debt covered under
Chapter 13. This two-step procedure, known as "Chapter 20", increased borrowers' financial gains
from bankruptcy as opposed to filing under either procedure separately.
18
20
procedure than under Chapter 7.
The Bankruptcy Abuse Prevention and Consumer Protection Act
The Bankruptcy Abuse Prevention and Consumer Protection Act (BAPCPA) of 2005
made several major changes to bankruptcy law. First, it abolished the right of debtors
to choose between Chapters 7 and 13; now debtors must pass a new "means test" to
file under Chapter 7. Debtors qualify for Chapter 7 if their monthly family income
average over the six months prior to filing is less than the median monthly family
income level in the state in which they live, adjusted for family size. In some places
households could be allowed to file under Chapter 7, without satisfying the means test,
as long as their monthly "disposable income" was lower than 166 dollars per month.
Thus, the 2005 law prevents some wealthy debtors from taking advantage of the
unlimited income exemption in Chapter 7. The reform also imposed new restrictions
on strategies used to protect high value assets in bankruptcy. For example, state of
residence home-equity protection is only valid after two years of residency in that
state, and within 2.5 years the level is capped at 125,000 dollars. Finally if borrowers
convert non-exempt assets into home-equity by making a down payment on their
mortgage, they must do so at least 3 and one third years before filing (White, 2007).
The second major change under the BAPCA is a uniform procedure that determines repayment obligations under Chapter 13. Debtors must now use 100 percent
of their "disposable income" for five years following their bankruptcy filing to make
repayments 20 . Third, BAPCPA greatly raised bankruptcy costs, and households are
now required to take a financial management, and also a credit counseling course
before their debts are discharged. They must file detailed financial documents, including copies of their tax returns for the previous four years, which may force them
to prepare unfiled tax returns. Filing fees have also increased. These new requirements have increased debtors' out-of-pocket costs of filing to around 2,500 dollars to
file under Chapter 7 and 3,500 dollars under Chapter 13 (Elias, 2005), not forgetting
21
the cost of the two training courses, and the preparation of tax returns.
BAPCPA among other things also increased the minimum time that must pass
between bankruptcy filings from six to eight years for Chapter 7, and from six months
to two years for Chapter 13 filings 22 . Therefore, fewer debtors than before are eligible
for bankruptcy at any given period.
Overall, the adoption of BACPA increases the cost of bankruptcy, decreases the
possible amount of debt discharged in bankruptcy, while implicitly decreasing income
protection. Therefore, setting a maximum income level above which debtors can no
longer gain from filing, making the US bankruptcy law more pro-creditor.
20
BAPCPA defines disposable income as the difference between debtors' average monthly family
during the six months prior to filing, with a new income exemption.
income
2
'A large proportion of the cost is attributable to the fact that bankruptcy lawyers can be fined
information is not accurate.
if debtors'
22
BAPCPA also imposes a four-year minimum period, where no such minimum existed previously,
for filing first under Chapter 7 and then under Chapter 13; and it also eliminates the "superdischarge" effect.
21
1.2.2
Related Literature
Gropp et al. (1997) was the first paper to use household level debt data to look at
the difference on credit availability for different levels of protection. Using the Survey of Consumer Finance of 1983, they found that higher protection under personal
bankruptcy is associated with a lower probability of access to credit, and a lower level
of debt for low asset households, in states with more generous bankruptcy exemptions. Using detailed bank information, Berger et al. (2010) found that unlimited liability small businesses have lower access to credit in states with more debtor-friendly
bankruptcy laws. In addition, these businesses face harsher loan terms: they are
more likely to pledge business collateral, have shorter maturities, pay higher rates,
and borrow smaller amounts. Also, Lin and White (2001) looked at how the protection levels affect the availability of mortgage credit application granting, finding that
accepted applications are negatively correlated with the level of protection. However,
all these studies use cross-sectional variation on protection to look at how these levels
correlate with credit availability. Hynes et al. (2004) find that state levels of exemptions are correlated with bankruptcy filing rates and state redistributional policies to
help the poor, among other variables that can be correlated with the supply of credit,
suggesting that the examination of the impact of bankruptcy laws should not treat
protection levels as exogenous variables. This paper contributes to this literature
using state time variation in bankruptcy protection levels to overcome these endogeneity concerns when looking at relationship between bankruptcy protection and
credit markets. Using this empirical strategy we find that increases in bankruptcy
protection did not lead to a reduction in the amount of debt held by households.
Our empirical strategy is more closely related to the work of Cerqueiro and Penas (2011), who use state level variation in the level of bankruptcy protection to
look at start-up creation, finding that increases in protection decrease start-up performance; and to Cerqueiro et al. (2013), who uses a similar strategy to look at
the effect of personal bankruptcy laws on innovation, finding that there is an aggregate decrease in the level of innovative activity among small firms in places in which
protection increased. The effect of the use of credit cards in entrepreneurial activity has also been studied by Chatterji and Seamans (2012). Using states' removal of
credit card interest rate ceilings in 1978 they show that this deregulation increases the
probability of entrepreneurial entry, arguably through an access to finance channel.
Finally, Fan and White (2003) find that personal bankruptcy protection motivates
entrepreneurial activity using cross-sectional variation in the level of protection. In
this paper, we show that increases in bankruptcy protection are correlated with increases in self-employment. Although we cannot rule out a demand channel, it seems
that bankruptcy laws could have an expansive impact on self-employment through
an increase in the credit channel.
Bankruptcy laws directly affect unsecured debt, given that secured debt cannot be
discharged. Therefore this paper is related to the literature on credit card borrowing.
Agarwal et al. (2013), analyze the effectiveness of consumer financial regulation in the
credit card market, using the 2009 credit card reform. They find that regulatory limits
on credit card fees reduce the overall borrowing cost to consumers by 2.8% of average
22
daily balances. Gross and Souleles (2002a) use credit card account data to analyze
how people respond to increases in the supply of credit; they find that increases in
credit limits generate an immediate response to debt, which implies a big sensitivity
of households to credit market changes. Gross and Souleles (2002b) use credit card
accounts to analyze credit card delinquency to highlight the importance of timevarying household characteristics on their ex post behavior. Our paper contributes
to this literature, showing new evidence of how bankruptcy protection affects the
demand for credit card debt.
This paper also relates to the studies that focus on the effect of personal bankruptcy
on filings and delinquency rates. Gross et al. (2013) use tax rebates to find that
households have a significant sensitivity of income to probability of filing, which is
consistent with the high sensitivity of financially constrained agents to increase leverage as credit availability increases, found by Gross and Souleles (2002b). White
(2007) looks at the effect of the interaction between personal bankruptcy filings and
credit card growth before the adoption of the new Bankruptcy Abuse Prevention and
Consumer Protection Act (BAPCA), arguing that the increase is due to the debtor
friendly bankruptcy laws in the pre-2005 period. In a related article, Jagtiani and
Li (2013) focus on the ex post effect of filing, and find that after a consumer files
for bankruptcy, there are long-lasting effects on their availability of credit. This paper contributes to this literature providing suggestive evidence of how bankruptcy
protection affects the mix of borrowing with no impact on delinquency behavior.
Furthermore, the protection of assets under bankruptcy affects the amount of
household collateral, and thus, their access to credit. Since Bernanke and Gertler
(1989), or Kiyotaki and Moore (1997), a number of theories have suggested that
improvements in collateral values ease credit constraints for borrowers. The collateral
lending channel builds on the idea that information asymmetries between lenders and
borrowers can be alleviated when collateral values are high (Hart and Moore, 1994).
From an empirical point of view, the collateral channel has been explored in its effect
on firms, by Benmelech and Bergman (2011), and Chaney et al. (2012); and credit
availability for small businesses, by Hurst and Lusardi (2004), and Adelinot et al.
(2013). The effect of housing collateral on household leverage has also been analyzed,
by Mian and Sufi (2011).
Increases in bankruptcy protection can also be seen as decreases in creditor rights,
which connects this paper to a large literature tracing the link between creditor rights
and financial development, pioneered by La Porta et al. (1998), and including Levine
(1998); Djankov et al. (2007); and Haselmann et al. (2010). Overall, this literature
reports a positive correlation between increases in creditor rights and the amount of
credit. 2 4 Most relevant to the current paper is Vig (2013), which looks at the increase
in creditor protection for secured debtors in the context of large firms in India. The
main difference between Vig (2013) and this paper (besides the fact that this paper
looks at US households, as opposed to firms in India), is how demand responds to
Rampini and Viswanathan (2010) in the context of a firm's access to credit.
Most recently, there are other papers which have looked at the same relationship but using crosscountry settings: Gianetti (2003); Qian and Strahan (2007); Acharya et al. (2011); and Davydenko
and Franks (2008).
2 3
24
23
changes in creditor protection. In Vig (2013), the decrease in the amount of secured
debt is driven by an increase in the threat of early liquidation, which firms face due
to the increase in creditor protection.25 In the current paper, the demand response
(increases in the demand for credit card debt), is based on an insurance channel
which relies on household risk aversion, and/or an increase in the number of strategic
borrowers. 26
This paper is also related to previous studies that have looked at the effect of
bankruptcy laws design in the context of corporate bankruptcy (Baird and Rasmussen, 2002; Bolton and Scharfstein 1996). In this context there is a large literature that describes the tension between ex ante and ex post efficiency in any
bankruptcy design. For instance, Gertner and Scharfstein (1991), and Hart (2000),
show the incentives of the debtor and creditors under corporate resolution in a theoretical framework, and demonstrate how debt contracts can lead to inefficient liquidation and underinvestment. This framework is also relevant when thinking about
the incentives for households to file for bankruptcy. Empirically, Chang and Schoar
(2013) look at the judge-specific fixed effect, showing that pro-debtor judges have
worse firm outcomes after Chapter 11, suggesting that this is a result of managers
and shareholders' incentives misalignment, highlighting how bankruptcy codes can
have a significant impact on ex post outcomes. Furthermore, Iverson (2013) looks
at the effect of bankruptcy courts' reduction in court caseloads due to the consumer
bankruptcy reform in 2005, finding that firms in more pro-debtor courts allow more
firms to reorganize and liquidate fewer firms.
Finally, this paper is complementary to studies looking at the effect of personal
bankruptcy laws on labor markets. Dobbie and Song (2013) find that filing for
bankruptcy under Chapter 13 has a significant effect on increasing earnings and employment, and also decreases mortality, suggesting that consumer bankruptcy benefits
are an order of magnitude larger than previously estimated".
1.3
1.3.1
Data and Summary Statistics
Data Description
In order to address the impact of changes in bankruptcy protection on household
debt, we collect and combine different data sources. The three main data sources
include time series of state levels of protection under bankruptcy, and geographical
distribution of household debt and interest rates information. In this section we
describe this datasets in detail.
The level of protection or exemptions represents the dollar amount of equity that
the debtor is entitled to protect in the event of bankruptcy; it represents the amount
25
This is consistent with the corporate literature on bankruptcy reorganization which suggested
that excessive creditor rights can lead to ex post inefficiencies in the form of a liquidation bias
(Aghion et al. (1992); Hart et al. (1997); Stromberg (2000); Pulvino (1998); and Povel (1999).
26
Examples of papers showing the costs of increases in creditor rights include: Acharya et al.
(2011); Acharya and Subramanian (2009); and Lilienfeld-Toal et al. (2012).
27
See White (2005) for a complete review of the literature.
24
of home equity and other personal assets that are protected. This information was
manually extracted and compiled from many sources, from state bankruptcy codes
to bankruptcy filing manual books2 8
We obtain level debt balances from the Federal Reserve Bank of New York Consumer Credit Panel/Equifax (CCP). This quarterly panel dataset is a 5% random
sample of individuals in the US who have a credit history with Equifax and a social security number associated with their credit file. Debt data reported includes
mortgage balances, home equity installment loans, and home equity lines of credit;
auto loans, including loans from banks, savings and loan associations, credit unions,
auto dealers and auto financing companies; and credit card debt: revolving accounts
from banks, national credit companies, credit unions, and bankcard companies. The
county level data is an aggregate of this information from 1999 to 2005 where, for
privacy reasons, reporting is done only for counties with an estimated population of
at least 10,000. This information is available for all debt types and the fraction of
household with delinquency status of 90 days late is provided as well. The micro
level data includes household level data of the debt variables described above, plus
detailed information on credit card accounts and individual level delinquency status:
current, 30 days late, 60 days late, 90 days late, 120 or more days late, and severely
derogatory. The individual level data permits a unique insight into the ex post behavior of households, as we are able to track the delinquency behavior of consumers
29
before they are affected by the change in protection
We obtain interest rates from Rate-Watch. It provides historical rate and fee data
from banks and credit unions across the country for a wide variety of banking products, such as CDs, checking, savings, money markets, promotional specials, auto loans,
unsecured loans, and credit cards. They collect information at the branch-setters level
by survey, and archive the information on a regular basis. For our purpose, interest
rates for unsecure loans, credit cards, and mortgage loans are aggregate at the county
level using branch-setter rate levels for the last quarter of each year to be consistent
with the aggregate debt balances measure. We then use this detailed geographically
dispersed measure of interest rates from 1999 to 2005 to analyze the supply response
of changes in personal bankruptcy protection.
County level income is measured as total wages and salary in a county according
to the IRS; this data is available from 1999 to 2005. The house prices used in the
regressions are obtained from the Federal Housing Finance Agency (FHFA) House
Price Index (HPI) data at a state level. The FHFA house price index is a weighted,
repeat-sales index and it measures average price changes in repeat sales or refinancing
on the same properties. This information is obtained by reviewing repeat mortgage
transactions on single-family properties whose mortgages have been purchased or
securitized by Fannie Mae or Freddie Mac since January 1975. We use data on the
state level index between 1999 and 2005.
County based unemployment levels and unemployment rates are obtained using
28
How to file for Chapter 7 Bankruptcy, Elias Renauer and Leonard Michon.
(1999-2009)
29
See Lee and van der Klaauw (2010) for details on the sample design.
25
Nolo editorial
the Bureau of Labor Statistics Local Area estimates. Local Area Unemployment
Statistics (LAUS) are available between 1976 and 2012 for approximately 7,300 areas that range from census regions and divisions to counties and county equivalent.
We match the county equivalent data to the CCP data using Federal Information
Processing Standard (FIPS) county unique identifiers.
To look at the determinants of change in exemptions, we use four additional data
sources: changes in state total medical expenses extracted from the National Health
Expenditure Data, Centers for Medicare and Medicaid Services; state level changes
in GDP and Personal Income from Bureau of Economic Analysis (BEA); bankruptcy
filing statistics at the state level from the Statistics Division of the Administrative
Office of the United States Courts3 0 ; and measures of political climates using the
share of votes for the Democratic Party in the last House of Representatives election
obtained from the Clerk of the House of Representatives (CHR).
The net creation of sole proprietorships at a county level is obtained from Census
non-employer statistics; we obtain the number of establishments for the period of 1999
to 2009 at the 2-digit NAICS level. In order to construct a measure of industries that
use credit card as a source of capital, we look at the Survey of Business Owners (SBO)
Public Use Microdata Sample (PUMS). The SBO PUMS was created using responses
from the 2007 SBO and provides access to survey data at a more detailed level than
that of the previously published SBO results. The SBO PUMS is designed to study
entrepreneurial activity by surveying a random sample of businesses selected from a
list of all firms operating during 2007 with receipts of $1,000 or more provided by
the IRS. The survey provides business characteristics such as firm size, employer-paid
benefits, minority- and women-ownership, access to capital, and firm age. For the
purposes of this paper, we classified industries based on the "use of credit card as a
start-up capital" for each firm and we group the answers to this question at the 2-digit
NAICS industry level (the finest level available in the data) for firms established in
2007, and then focus specifically in 1-4 employee firms only.
1.3.2
Summary Statistics
Table 1.1 shows a description of our main variables; the sample spans from 1999 to
2005. The total debt balance in a county is 2.91 billion dollars. The level of credit
card balance is 0.29 billion dollars. When looking at states that "eventually" change
their level of protection during our sample period and compare them to states that
never change their level of protection, the former holds 0.36 billion dollars on average,
and the latter 0.22; however the difference is not statistically significant.
The average debt growth in a county was 12.2%, and credit card debt growth
during the same period experienced the same pattern, with a 7.6% average annual
growth, with no significant difference between the "eventually" treated and the never
treated group. The summary statistics seem to show that credit card balances are a
small proportion of the average household balance sheet, as mortgage debt accounts
for most of consumers' debt claim. However, it is important to point out that when
30
See http://www.uscourts.gov/Statistics/BankruptcyStatistics.aspx
26
compared in terms of monthly payments, this difference is much smaller, and arguably
credit card debt is an important part of household budget and a relevant medium
to relax budget constraint, allowing households to shift inter-temporal consumption
(White 2007).
The only strong significant difference between the two groups is seen in average house price growth. States which were never treated experienced a house price
growth of 6.2% on average annually, and states which were eventually treated increased their house price growth by 8.8%. This difference is consistent with the fact
that house prices are argued to be determinants of the changes in bankruptcy protection. However, we find in Table 1.6 that they have no predictive power in the changes
in protection.
Table 1.2 shows the description of the exemption levels and changes from 1999 to
2005. First, it is important to notice that bankruptcy exemption changes are quite
common within our sample period; over the whole time there are 37 changes within
26 states. The average level of protection is around 73,000 dollars, and a median
of 55,800 dollars, with most of the value coming from the homestead exemption
(protection over homeowners' equity). The average change in protection is close to
40,000 dollars, with a median of 15,400 dollars, with some changes being very small
and associated to inflation adjustments, and others being very substantial. Figure
1-2 shows the geographical dispersion of these changes.
1.4
Empirical Hypothesis
Changes in the level of asset protection in bankruptcy affects credit markets' equilibrium through demand and supply. In order to guide our empirical analysis we review
the differences dimension through which increases in asset protection can affect the
supply and demand of credit, and review the implications for our empirical exercise.
Collateral channel. If markets are incomplete, the possibility of collateral pledging enhances agents' debt capacity, as it gives the lender the option to repossess assets
ex post, reducing the risk of borrowers, and easing borrowers' access to finance ex ante
(Hart and Moore, 1994). In our case, the increase in protection diminishes the collateral value of assets, as it decreases the availability of assets to be seized by lenders,
making the supply of credit less attractive; therefore reducing borrowers' access to
credit.
Insurance channel. In the presence of incomplete markets, increased protection
also makes borrowing more attractive for risk-averse agents by improving risk-sharing.
Effectively, the higher protection on the bad state of the world will incentivize riskaverse agents to take on leverage, increasing the demand for credit.
Moral hazard channel. An increase in the level of protection might also foster
borrowers' incentives to undertake riskier projects or over-borrowing, increasing the
demand for credit, and the ability of lenders to distinguish the type of borrower that
are they facing will define the supply response. Furthermore, according to Stiglitz
and Weiss (1981), lenders' profit functions could set an upper limit to the increase in
interest rates, leading to a decrease in the quantities due to the increase in borrower
27
risk. In summary, moral hazard increases the demand for credit, and in most cases,
will reduce the supply of credit.
Adverse selection channel. If the level of protection increases, more strategic
defaulters with private information about their future income or propensity to default
could participate in the markets, aiming to profit from the new borrowing conditions,
increasing the riskiness of the pool of borrowers and also the demand for credit. Again
the equilibrium response will be driven by lenders' ability to screen new borrowers.
Therefore, the theoretical prediction is unclear, given that the net effect will depend on the relative magnitudes of the supply and demand response3 1 . Interest must
weakly rise in equilibrium, independent of the prevailing force. If the supply demand
dominates, quantities should go down, but if the demand effect dominates, quantities
should go up. We attempt to distinguish between these channels empirically.
It is plausible to imagine that in the presence of agency problems, a demand
driven equilibrium takes place. In an extreme case, if the lender overestimates the
quality of the pool of borrowers, the increase in protection would lead to an increase
in quantities. However, in Appendix A we show that given very simple conditions,
and without asymmetric information, we can observe a demand driven equilibrium
where quantities and prices increase. This model of the credit market considers a
risk-averse borrower who is financially constrained and a risk-neutral lender. The
borrower has a stochastic income, and exogenous home equity that is realized in
period 2. Only debt contracts are available. In case of default, the lender can seize
the borrower's assets up to the exemption level dictated by law. The agents need to
borrow in order to consume in period 1, while the interest rate is set such that the
bank breaks even (zero profit). For a given interest rate, a risk-averse borrower will
consume until a point where the marginal utility of consumption today is equal to
the expected marginal utility in the future. Increased bankruptcy protection makes
defaulting attractive to the borrower in more states of the world, and forces lenders
to charge a higher interest rate to break even.
The model shows that for a certain region with a given level of protection in
bankruptcy, when the level of protection is increased, the agent will be willing to take
on more debt despite the increase in interest rates. This happens when the marginal
benefit from the increase in consumption at period 1 is greater than the loss of utility
in the good state in period 2, due to the repayment of their debt claim; as in the bad
state they are indifferent due to the protection level. Furthermore, if the marginal
benefit is not enough to overcome the loss of consumption during the good state, we
should see a decrease in quantities and increase in prices. Using exogenous variation
on the level of protection, we aim to identify the type of equilibrium that rises after an
increase in the level of consumer protection under bankruptcy. These results, which
are highlighted by the model, are relevant as they show that the insurance channel
in itself could lead to a demand driven credit market equilibrium shift, without the
presence of moral hazard or adverse selection.
Empirical Predictions
The exposed theoretical framework allows us to sharpen our empirical exploration.
31 Figure
1-3 shows the possible outcomes in a simple demand and supply graph.
28
Based on the arguments above we have the following predictions.
First, if the demand effect dominates, we should see an increase in quantities
and prices. Furthermore, the increase in prices should be stronger for low-income
borrowers, as the increase in risk-sharing (insurance channel) is more important for
these borrowers, and they are also more likely to be under financial constraints.
The effect should be stronger for homeowners, as the change in asset protection affects home-equity holding predominantly (see Table 1.2). The increase in bankruptcy
protection does not directly affect secured debt, as the bankruptcy code only discharges unsecured debt. Therefore, we should see weaker or no effect on secured
debt.
Finally, if agency problems are an important driver of the increase in demand,
we would expect to see a significant effect on ex post default, arguably driven by
individuals who over-borrowed ex ante or invested in riskier projects.
Second, if the supply effect dominates, we should see an increase in prices
and a decrease in quantities. The rise in prices should be higher in places where the
riskiness of the pool of borrowers, or the ex ante probability of defaults, increases
more. The effect should also be stronger where the fundamental value of the ability
to pledge assets is higher, and court enforcement of bankruptcy contracts is lower.
Further, the effect should be stronger in areas where lenders have less information
about their borrowers, as the dominance of the supply effect suggests that lenders are
reducing the supply of credit more intensively.
In the next section we show the empirical strategy we used to identify the equilibrium change: we find that the quantities and price effect is consistent with a stronger
demand effect, and we describe the set of tests that we used to assure this finding,
and the empirical test that attempts to distinguish between the different channels.
1.5
Empirical Strategy
Empirically identifying the actual effect of bankruptcy protection levels on household
leverage is challenging, as these levels are correlated with unobservable borrower and
lender characteristics, which might simultaneously affect credit availability and the
level of protection. For example, on the one hand, states with a higher protection
level may be states where households are less financially savvy and, as a result, are
more willing to take on more debt; this in turn will lead to a positive correlation
between debt and protection. On the other hand, if the level of protection correlates
with better local economic conditions, people will be less financially constrained,
potentially taking on less debt, and thus leading to a negative correlation between
debt and protection levels.
In this paper, we exploit exogenous variation in state level bankruptcy protection
dollar amounts to identify the effect of this protection on household debt. We use
different timing in the changes to exemption levels by state to identify how exemptions
affect household leverage (there were a total of 37 changes in exemptions between 1999
and 2005)
The proposed baseline specification is the following,
29
ADebtit
=
ai + at+
ppAProtectiont + FAXt + Eit (1)
Where ADebtit is the log change in either credit card debt, mortgage debt, auto
loan debt, in a county i and year t .AProtectiont represents the log change in the
level of Chapter 7 protection (homestead plus personal) in a state s and year t .aj is
a county fixed effect, and at are year fixed effect.AXit represents a vector of county
controls changes, such as county unemployment rate, log of house prices, and log of
income in a county.
We use the same specification in (1) to measure the effect of changes in protection
on interest rates. To do so we replace the log change in debt, by changes in interest
rates in percentage for mortgages, personal unsecure loans and credit cards.
Since changes in protection vary at the state level, but debt balances and interest rates are observed at the county or individual level, the error term in equation
(1) has a potentially time-varying state component. Following Bertrand, Duflo and
Mullainathan (2004), the residuals are clustered by state. This allows for maximum
flexibility in the variance-covariance matrix of residuals. It is also more general than
state-year clustering, which would leave intact the possibility of serial correlation in
the error term.
If the measure of debt and the controls all display heterogeneous trends across
counties, the most parsimonious treatment of these trends is to take first-differences,
as in the equation above3 2 , with variables in differences; the presence of county fixed
effects guarantees that differential county specific trends are controlled for in all variables. A first-differences specification is suitable in our case as it accommodates
the repeated treatment present in our sample (in our sample period some states did
change their level of protection more than once). The regressor 13p captures the
changes in debt within the year as the level of protection increases. Additionally,
the use of the amount of protection, i.e., intensity of treatment, guarantees that the
main estimate is driven by big changes in the level of protection. Furthermore, we
will conduct alternative specifications to show that our results are robust to the use
of level specification, and to the use of alternative measures of the treatment effect.
Effectively, we compare the change in the amount of debt between a county belonging to a state which increased the level of protection between t and t+1, with the
amount of debt of a county belonging to a state in which the level of protection did not
change during the same period. The two identifying assumptions are first, that the
timing of the changes in the levels of protection are uncorrelated with determinants of
household leverage; and second, that after controlling for observed time-varying characteristics, linear county trends, and time-invariant county characteristics, changes
in the state level of protection will only affect the state which adopted the change,
thus the only determinant of the difference in household debt across states is the
exogenous change in the level of protection.
We assess the first identifying assumption by looking at the correlation between
32
Paravisini (2008).
30
suspected determinants in the level of protection and changes in the levels of protection. Conventional wisdom attributes changes in the levels of bankruptcy protection
to the gap between house prices and homestead exemption levels, as well as the cost
of medical expenses. If our identification strategy is valid, changes in the measurable
variables should be uncorrelated with changes in the level of protection, suggesting
that the actual timing of the change is an exogenous shock to the credit demand and
supply of credit in the affected regions.
To assess the second identifying assumption, we need to rule out alternative hypotheses that could explain our results. First, changes in the level of protection could
be correlated with state specific pre-existing trends that survive our controls, and
thus our results are a reflection of this differential pre-trend rather than a result arising from changes in the levels of protection. For example, states which increase their
protection levels are states where economic conditions are booming in the period prior
to the increase. We should expect that looking at the dynamic of the change, the
inclusion of lags of the changes should have no effect on the coefficients and have no
significant correlation with the levels of debt.
A second alternative hypothesis is that there are state specific credit market trends
that are correlated with the changes in protection that would explain our findings. For
example, the areas where the level of protection increased were areas where all credit
availability for all types was expanded. To meaningfully differentiate the impact of
the change in the level of protection from these alternative hypotheses, we use the
fact that personal bankruptcy laws allow households to renege only on unsecured
debt, which implies that changes in personal bankruptcy laws will only directly affect
unsecured debt.
A third alternative hypothesis is that the observed increase in quantities is due to
a contemporaneous decrease in prices that is correlated with the timing of the changes
in bankruptcy protection. In other words, areas that increased the level of protection
were areas where credit became cheaper. Using novel bank branch level data on credit
rates for different types of credit, we can explore the effect of bankruptcy protection
changes on interest rates; if interest rates are positively affected by the increase in the
level of protection, it is less likely that our effect is driven by a relaxation of lending
standards in credit markets.
Local economic conditions could produce spurious effects due to geographical heterogeneity that is uncorrelated to changes in the levels of protection. To overcome this
33
endogeneity we compare neighboring county-pairs across state borders , but within
the same income categories, using the following empirical specification:
6
ADebtipt = oi + aYipt + /pAProtectionst + FAXit + Ept (2)
Where ADebtipt is the log change in either credit card debt, mortgage debt, auto
loan debt; in a county i, pair p and year t. AProtectionst represents the log change in
the level of Chapter 7 protection (homestead plus personal) in a state s in year t. ai
3
3
This methodology is similar to Heider and Ljungqvist (2013) and Dube et al. (2010)
31
is a county fixed effect, and aipt, is a dummy for each neighboring county pair for each
year. Note that variables for county i maybe repeated for all pairs of which they are
part. In this setup our estimate fp only uses debt variation within each neighboring
county-pair across state borders. Our additional identifying assumption implies that
the changes in protection are uncorrelated with the residual Eipt after controlling for
observable characteristics, county fixed effects and county-pairs year fixed effect. We
also assign counties to income buckets, and run the proposed specification only within
county-pairs that are in the same income category.
To attempt to identify the channel that is driving the demand effect we use individual level data to look at debt change, entry to the credit card market, and
delinquency. We use the same specification (1) as for the county aggregates, but
changing the dependent variable, and including in this case the zipcode level house
prices, income, and county unemployment rates.
The change in debt for each individual is estimated using log changes, and it
therefore represents the change in debt for existing debtors. When looking at the
number of accounts, our dependent variable is the difference between the number of
credit cards in t -1 and t. Entry is defined in two ways as follows: opening the first
credit card, which is a dummy equal to one if the household did not have a credit
card in t-1, and have one or more credit cards in t. Alternatively, entry is defined as a
dummy equal to one if the balance becomes positive between t and t -1. Both measures
attempt to capture the entry of new borrowers to the credit card market. Finally,
to measure delinquency, this is a dummy equal to one if household i is delinquent at
time t, t+1, t+2, and t+3 respectively, and the regressions are estimated separately.
Therefore, the estimated coefficient represents an intent-to-treat effect, as the same
individual may be affected by the change in the levels of protection more than once
during our sample period.
Finally, we look at changes in the levels of self-employment to explore the effect
on real outcomes. For this we use specification (1) but in this case, using the change
in total county self-employment as a left hand side variable, or the change in selfemployment in an industry and county between t and t-1.
1.6
Results and discussion
1.6.1
Bankruptcy Protection and Household Leverage and
Interest Rates
We find that growth in bankruptcy protection leads to an increase in the level of
credit card debt held by households (unsecured debt) between 1999 and 2005 (Table
1.3 ). Moreover, the increase in protection has no effect on other types of secured
debt (auto and mortgage, Table 1.4 and 1.5)3.
34
The average effect is only present in the pre-bankruptcy reform period, when filing for
bankruptcy was easier and cheaper (Table B8). If the cost of filing for bankruptcy increases enough,
the effective protection is smaller, decreasing the ex ante benefit of increasing the amount of debt
today. Considering that there is evidence that household bankruptcy filings are highly sensitive to
32
A possible concern may be that states which did not change the level of protection
within our sample period are not a good control group, as they could be systematically
different from the group which did opt to change their level of bankruptcy protection,
and this would therefore invalidate our empirical inference. To overcome this concern,
we replicated our main specification (Table 1.3 column 1), focusing only on the states
in which changes in protection levels were implemented in our sample period (i.e.
"eventually" treated, Table 1.3 column 6). In this case the main effects we estimate
are basically unchanged, mitigating the endogeneity concern about the changes.
Tables 1.10 and 1.11 replicates our main specification, but using interest rates
changes as a dependent variable for personal unsecured loans, credit cards, and mortgage rates. The results show that the increase in bankruptcy protection leads to an
increase in the level of interest rates for unsecured loans, but does not affect mortgage rates. These results suggest a demand driven credit market equilibrium, as we
observe increases in quantities, and prices.
Furthermore, in Table 1.6, columns 1 and 2, we look at the correlation between
the levels of protection and contemporaneous and lag levels of determinants, which
in a traditional view would be seen as driving the changes in the level of protection.
Empirically, levels seems to be correlated with housing price and bankruptcy filing
rates, which is consistent with evidence that cross-sectional variation in the level of
protection is a state specific characteristic. Furthermore, Table 1.6, columns 3 to
6, looks at how changes in the levels of exemptions correlates with change in the
determinants above, using an OLS estimation clustering standard errors at the state
level, or running a linear probability model of the likelihood of change. In both cases,
lag change in the candidates' determinants have no predictive power on changes in
the level of protection. This is consistent with our identification assumption, that
the timing of the changes is exogenous to characteristics which define the supply and
demand of credit.
While our results support the empirical strategy, there are alternative hypotheses
that we need to rule out as explaining our results. First, changes in the level of
protection could be correlated with pre-existing state specific trends that survive
our controls, and thus our results are a reflection of these differential pre-trends
rather than changes in the levels of protection. For example, states which increase
their protection levels are states in which employment conditions are booming in
the period prior to the change in protection levels. Table 1.7 looks at the effect of
changes in protection when lags and leads of the changes are incorporated into the
main specification; the first 4 columns show the specification without fixed effect, the
second sets out with state fixed effect, and the last one with county fixed effect. These
results show that our estimates are not affected by the inclusion of lag changes in the
levels of protection, and that the coefficient in the lags is economically small and
statistically insignificant 35 . Furthermore, the coefficients in the leads are increasing
and statistically significant, especially for two periods after the change, which suggests
liquidity constraint (Gross et al., 2013), we should expect the effect to be weaker or nonexistent
during the post period.
35
Considering that our exogenous variation is at the state level, we cannot control for state-time
unobserved heterogeneity that is contemporaneous to our effect.
33
that there may be an overreaction of households to the changes in the first year and
a long term effect that continues up to year two.
Table 1.3 shows that the effect is concentrated in credit card debt (unsecured).
This allows us to rule out the alternative explanation that our strategy is picking up
state specific credit market trends that are correlated with the changes in protection
and that can be confounded with our identified effect.
Table 1.9 shows the effect is stronger in counties that are in the lowest tercile of
the within state income distribution, monotonically decreasing as the level of income
increases. It is expected that lower-income areas may be more affected by increases
in protection, as the impact of the improvement in risk sharing should be more significant.
Homeowner households should be more affected by the changes in the level of
protection, as a big proportion of their protection comes from home equity protection.
However, county level homeownership is correlated with income, so in order to gain a
meaningful perspective on this variation, we look at the within income group variation
on county level homeownership. Table 1.9 column 3 shows that the differential effect
is aligned with the prediction, as the estimated coefficient for these particular areas
almost triples with respect to the baseline specification.
Following the same logic, we look at the within income group variation on bank
concentration - a measure based on share of deposit holding at the branch level. Table 1.16column 2 shows that the effect is stronger in areas where markets are more
concentrated, which is consistent with the Peterson and Rajan (1995) relationship
lending model, where creditors are more likely to finance a credit constrained borrower when credit markets are concentrated because it is easier for these creditors to
internalize the benefits of assisting these borrowers.
Another alternative explanation of our finding is that the increase in quantities is
due to a contemporaneous decrease in prices, which correlates with the timing of the
changes in bankruptcy protection. In other words, areas which increased the level of
protection were areas in which credit became cheaper. As mentioned above, Tables
1.10 and 1.11 show that the increase in bankruptcy protection leads to an increase
in the level of interest rates for unsecure loans, not affecting mortgage rates. These
results support our causal interpretation of the results, alleviating the concern that we
are picking up a relaxation in the price of credit leading to an increase in quantities.
Local economic conditions could produce spurious effects due to geographical heterogeneity that is uncorrelated with changes in the levels of protection. To overcome this endogeneity, we compare neighboring county-pairs within the same income
bucket. Table 1.8 shows that when focusing on a county-pair in the same income
bucket, the estimated results are very similar to the main specification. Moreover the
effect is stronger when we concentrate on county-pairs in the lower end of the county
income distribution.
1.6.2
Robustness Test
We choose a first difference specification with county fixed effect to parsimoniously
account for county level linear trends, and to account for multiples treatment for the
34
same state across time. However, in Table 1.1 Panel A, we show that our estimation
is the same if we exclude county fixed effect, and change them by state level fixed
effect or run debt levels on protection level with county fixed effect. In other words,
our effect is invariant to the specific difference in difference specification. Table 1.1
shows how the effect changes with different measures of the treatment. We choose to
use an intensity of treatment measure as our treatment; however, as Table 1.1 Panel
A shows, our results are invariant to the use of only large changes, use of exemption
dummies instead of the intensity of treatment, or if we restrict the analysis to only
states which change their level of protection only once.
Given the nature of our empirical strategy, as we argue before, time-varying
changes at state levels may be omitted variables explaining our results; one candidate is the level of unemployment insurance in each state (Hsu et al., 2012). Table
1.15 shows that the inclusion of this variable has no impact on the estimated coefficient. The results are also robust to change, the depend variable for changes in debt
to income, or percentage changes, or to replace the treatment only by the amount of
homestead protection. Finally, all the results exclude DC, because within our sample
period, this state changed the protection from a very low level to an unlimited level.
If we include a time-varying dummy to account for this extreme change in the level of
protection, Table 1.15 shows that it generates a decrease in the level of debt available
to households, consistent with the empirical prediction of our model.
1.6.3
Magnitude of the effect
In terms of magnitude, we find that the average county in our relevant period (19992005) has a credit card balance of 290 million dollars, and the average increase in
credit card debt is 7.6%. Our main estimate explains 10% of this balance growth.
This magnitude represents the average treatment effect over the entire population.
However, we believe that our effect is driven mostly by people close to financial
distress, for whom the possibility of filing for bankruptcy is a real one. When we
estimate the magnitude of the effect for the particular subgroup of areas, counties in
the low-income tercile with higher homeownership percentage, we find that the effect
now explains between 34% and 47% of the increases in their credit card balance.
This heterogeneity is consistent with our interpretation that there is only one subset
of people affected, e.g., homeowners within a county close to distress level on their
credit cards. However, there is also the possibility that our estimates are biased
downward (attenuation biased), due to measurement errors in our variables
1.6.4
Borrowers, Delinquency and Self-Employment
Important remaining questions to address, include which households are expanding
the amount of credit they hold, how they are doing so, and what their ex post conduct
may be. Using individual level data to look at the ex ante and ex post behavior of
households, first we replicate the county level results focusing on areas that are below
the median county income. Table 1.12 Panel A shows that the effect of changes in
protection is similar to those found when we focus on the lower end of the county
35
level distribution or county borders. When we focus on homeowners, defined as an
individual for whom we observe home-related debt at some point between 1999 and
2005, the effect is stronger, which again is consistent with the county estimates (Table
1.12 Panel B).
Furthermore, using detailed account information, we show in Table 1.12 columns
2-4, that changes in protection causally increase the number of credit cards per household; this increase is stronger among households that had ex ante credit card accounts.
Even more interestingly, the increase in number of credit cards is stronger for households that also had a positive balance. This finding suggests that the credit expansion
is due to existing borrowers acquiring more credit. Finally, Table 1.12 columns 5-6,
show how changes in protection are uncorrelated with entry into the credit card market, defined as the time when a member of a household opens their first account,
or as the time when their credit card balance goes from zero to positive. All these
results provide evidence that in this sample, the effect is being driving by existing
debtors expanding their current balance or their number of accounts, rather than new
households entering the credit market.
Focusing on the same sample, we explore their delinquency behavior up to three
years after the increase in credit card usage induced by the change in protection.
Three years is a long time frame when considering holdings on a credit card. Table
1.13 shows that within this sample there is no measurable increase in the level of
delinquency; if anything, the probability of individuals becoming delinquent in the
future decreases. If the households which are increasing their level of debt are overborrowing, or taking on more risky projects, we would expect delinquency rates to
increase. Although we cannot completely rule out an over-borrowing behavior, the
results described are more consistent with risk-averse borrowers increasing their debt
holding in response to the greater insurance received from the increase in protection.
We show that areas which experienced an increase in the level of credit card
debt also experienced an increase in the level of self-employment creation, specifically
within industries that make more use of credit cards as start-up capital. Table 2.6
shows that, on average, the increase in self-employment is only positively correlated
with the changes in the level of protection in low-income regions. Also, the estimated
effect is stronger when we focus on industries for which credit card debt is an important source of financing (for example, construction or photography). It is important
to point out that these outcome variables are only suggestive evidence of the real
effect of the increase on the level of unsecured debt.
Taking all this evidence together, the rise in credit card debt induced by the
increase in the level of protection could have led to an increase in small business
creation, and a decrease (or no increase) in the delinquency rates of unsecure creditors.
The individual results seems to suggest that the channel driving the demand effect is
consistent with a large impact from the insurance channel on existing borrowers, as we
do not observe increases in the entry rates of new borrowers and ex post delinquencies
within our micro level sample. Although this evidence is only suggestive, it highlights
the important potential benefits of increasing the level of bankruptcy protection,
especially for people in areas on the lower end of the wealth distribution, for which
the insurance effect is more significant.
36
1.7
Conclusion
Overall, the evidence we present in this paper identifies the causal effect of the increase
in the level of protection under personal bankruptcy on household leverage. We show
that increases in the level of bankruptcy protection within our sample period, leads
to an expansion in the levels of credit card debt that is stronger in counties that
are in the lowest tercile of the within state income distribution, and monotonically
decreasing as the level of income increases. Using micro level data we find that the
expansion is concentrated among existing borrowers. This expansion is also correlated
with an increase in small business creation, and seems to have no effect on counties'
overall delinquency rates.
These findings highlight the importance role that personal bankruptcy laws play
as an insurance mechanism, providing down side protection especially for low-income
regions. Therefore, the documented credit increase has important implications for our
understanding of personal bankruptcy protection as a risk-sharing improving policy.
37
1.8
Bibliography
Acharya, Viral V., and Krishnamurthy V. Subramanian. Bankruptcy codes and
innovation. Review of Financial Studies 22.12 (2009): 4949-4988.
Acharya, Viral V., Rangarajan K. Sundaram, and Kose John. Cross-country
variations in capital structures: The role of bankruptcy codes. Journal of Financial
Intermediation 20.1 (2011): 25-54.
Acharya, Viral V., Yakov Amihud, and Lubomir Litov. Creditor rights and corporate risk-taking. Journal of Financial Economics 102.1 (2011): 150-166.
Adelino, Manuel, Antoinette Schoar, and Felipe Severino. House Prices, Collateral
and Self-Employment. No. w18868. National Bureau of Economic Research (2013).
Agarwal, Sumit, et al. Regulating Consumer Financial Products: Evidence from
Credit Cards. No. w19484. National Bureau of Economic Research (2013).
Aghion, Philippe, Oliver Hart, and John Moore. The economics of bankruptcy
reform. Journal of Law Economics and Organization. 8 (1992): 523-546.
Alvarez, Fernando, and Urban J. Jermann. Efficiency, equilibrium, and asset
pricing with risk of default. Econometrica 68.4 (2000): 775-797.
Baird, Douglas, and Robert Rasmussen. The End of Bankruptcy. Stanford Law
Review 55 (2002).
Benmelech, Efraim, and Nittai K. Bergman. Bankruptcy and the collateral channel. The Journal of Finance 66.2 (2011): 337-378.
Berger, Allen N., Geraldo Cerqueiro, and Mara F. Penas. Does debtor protection
really protect debtors? Evidence from the small business credit market. Journal of
Banking & Finance 35.7 (2011): 1843-1857.
Bernanke, Ben, and Mark Gertler. Agency costs, net worth, and business fluctuations. The American Economic Review (1989): 14-31.
Bertrand, Marianne, Esther Duflo, and Sendhil Mullainathan. How much should
we trust differences-in-differences estimates? The Quarterly Journal of Economics
119.1 (2004): 249-275.
Bolton, Patrick, and David S. Scharfstein. Optimal debt structure and the number
of creditors. Journal of Political Economy (1996): 1-25.
Cerqueiro, Geraldo, and Maria Penas. How does personal bankruptcy law affect
start-ups? Working Paper (2011).
Cerqueiro, Geraldo, Deepak Hegde, Maria Penas, and Robert Seamans. Personal
Bankruptcy Law and Innovation. Working Paper (2013).
Chaney, Thomas, David Sraer, and David Thesmar. The Collateral Channel: How
did Real Estate Shocks Affect Corporate Investment. American Economic Review102
(6) (2012): 2381-2409.
Chang, Tom, and Antoinette Schoar. Judge Specific Differences in Chapter 11
and Firm Outcomes. National Bureau of Economic Research (2013).
Chatterji, Aaron K., and Robert C. Seamans. Entrepreneurial finance, credit
cards, and race. Journal of Financial Economics 106.1 (2012): 182-195.
Davydenko, Sergei A., and Julian R. Franks. Do bankruptcy codes matter? A
study of defaults in France, Germany, and the UK. The Journal of Finance 63.2
38
(2008): 565-608.
Djankov, Simeon, Caralee McLiesh, and Andrei Shleifer. Private credit in 129
countries. Journal of financial Economics 84.2 (2007): 299-329.
Dobbie, Will, and Jae Song. Debt Relief and Debtor Outcomes:. Measuring the
Effects of Consumer Bankruptcy Protection. Harvard University mimeo, (2013).
Dube, Arindrajit, T. William Lester, and Michael Reich. Minimum wage effects
across state borders: Estimates using contiguous counties. The Review of Economics
and Statistics 92.4 (2010): 945-964.
Elias, Stephen R. The New Bankruptcy: Will it Work for You?. Nolo, (2005).
Fan, Wei, and Michelle J. White. Personal Bankruptcy and the Level of Entrepreneurial Activity Journal of Law and Economics 46.2 (2003): 543-567.
Flynn, Ed, Gordon Bermant, and Karen Bakewell. A closer look at elderly chapter
7 debtors. American Bankruptcy Institute Journal. 21 (2002): 22-49.
Gertner, Robert, and David Scharfstein. A Theory of Workouts and the Effects
of Reorganization Law. The Journal of Finance 46.4 (1991): 1189-1222.
Giannetti, Mariassunta. Do better institutions mitigate agency problems? Evidence from corporate finance choices. Journal of Financial and Quantitative Analysis
38.1 (2003): 185-212.
Gropp, Reint, John Karl Scholz, and Michelle J. White. Personal bankruptcy
and credit supply and demand. The Quarterly Journal of Economics 112.1 (1997):
217-251.
Gross, David B., and Nicholas S. Souleles. Do liquidity constraints and interest
rates matter for consumer behavior? Evidence from credit card data. The Quarterly
Journal of Economics 117.1 (2002a): 149-185.
Gross, David B., and Nicholas S. Souleles. An Empirical Analysis of Personal
Bankruptcy and Delinquency. Review of Financial Studies, 15.1 (2002b) 319-347.
Gross, Tal, Matthew J. Notowidigdo, and Jialan Wang. Liquidity constraints
and consumer bankruptcy: Evidence from tax rebates. Review of Economics and
Statistics (2012).
Hart, Oliver and John Moore. A Theory of Debt Based on the Inalienability of
Human Capital, The Quarterly Journal of Economics, 109.4 (1994) 841-79
Hart, Oliver, and John Moore. Foundations of incomplete contracts. The Review
of Economic Studies 66.1 (1999): 115-138.
Hart, Oliver, Rafael La Porta, Florencio Lopez-de-Silanes, and John Moore. A
new bankruptcy procedure that uses multiple auctions. European Economic Review
41.3 (1997): 461-473.
Haselmann, Rainer, Katharina Pistor, and Vikrant Vig. How law affects lending.
Review of Financial Studies 23.2 (2010): 549-580.
Heider, Florian, and Alexander Ljungqvist. As Certain As Debt And Taxes:
Estimating The Tax Sensitivity Of Leverage From Exogenous State Tax Changes.
No. w18263. National Bureau of Economic Research (2012).
Hsu, Joanne, David Matsa, and Brian Melzer. Unemployment Insurance and
Consumer Credit. Working Paper (2012).
Hynes, Richard M., Anup Malani, and Eric A. Posner. The Political Economy of
Property Exemption Laws. Journal of Law and Economics 47.1 (2004): 19-43.
39
Jagtiani, Julapa, and Wenli Li. Credit access and credit performance after consumer bankruptcy filing: new evidence. No. 13-24. (2013).
Kehoe, Timothy J., and David K. Levine. Debt-constrained asset markets. The
Review of Economic Studies 60.4 (1993): 865-888.
Kiyotaki, Nobuhiro, and John Moore. Credit Chains. Journal of Political Economy 105.21 (1997): 211-248.
La Porta, Rafael, Florencio Lopez-de-Silanes, Andrej Shleifer, and Robert W.
Vishny. Law and Finance. Journal of Political Economy 106 (1998): 1113-1155.
Lee, Donghoon and Wilbert van der Klaauw. An Introduction to the FRBNY
Consumer Credit Panel. Federal Reserve Bank of New York Staff Report no. 479.
(2010)
Lefgren, Lars, and Frank McIntyre. Explaining the Puzzle of Cross-State Differences in Bankruptcy Rates. Journal of Law and Economics 52.2 (2009): 367-393.
Levine, Ross. The legal environment, banks, and long-run economic growth. Journal of Money, Credit and Banking (1998): 596-613.
Li, Wenli, Michelle J. White, and Ning Zhu. Did Bankruptcy Reform Cause
Mortgage Default to Rise?. No. w15968. National Bureau of Economic Research,
2010.
Lilienfeld-Toal, Ulf von, Dilip Mookherjee, and Sujata Visaria. The Distributive
Impact of Reforms in Credit Enforcement: Evidence from Indian Debt Recovery
Tribunals. Econometrica 80.2 (2012): 497-558.
Lin, Emily Y., and Michelle J. White. Bankruptcy and the market for mortgage
and home improvement loans. Journal of Urban Economics 50.1 (2001): 138-162.
Link, Frederick. The Economics of Personal Bankruptcy. MIT Thesis (2004)
Martin, Nathalie, and Stewart Paley. JK Lasser's The New Bankruptcy Law and
You. John Wiley & Sons, 2006.'
Mian, Atif and Amir Sufi. 2011. House Prices, Home Equity-Based Borrowing,
and the US Household Leverage Crisis. American Economic Review, 101 (2011):
2132-2156.
Paravisini, Daniel. Local bank financial constraints and firm access to external
finance. The Journal of Finance 63.5 (2008): 2161-2193.
Petersen, Mitchell A., and Raghuram G. Rajan. The benefits of lending relationships: Evidence from small business data. The Journal of Finance 49.1 (1994):
3-37.
Povel, Paul. Optimal soft or tough bankruptcy procedures. Journal of Law,
Economics, and Organization 15.3 (1999): 659-684.
Pulvino, Todd C. Do asset fire sales exist? An empirical investigation of commercial aircraft transactions. The Journal of Finance 53.3 (1998): 939-978.
Qian, Jun, and Philip E. Strahan. How laws and institutions shape financial
contracts: The case of bank loans. The Journal of Finance 62.6 (2007): 2803-2834.
Rampini, Adriano A., and S. Viswanathan. Collateral, risk management, and the
distribution of debt capacity. The Journal of Finance 65.6 (2010): 2293-2322.
Stango, Victor, and Jonathan Zinman. Limited and varying consumer attention:
Evidence from shocks to the salience of bank overdraft fees. No. w17028. National
Bureau of Economic Research, 2011.
40
Stiglitz, Joseph E., and Andrew Weiss. Credit rationing in markets with imperfect
information. The American Economic Review 71.3 (1981): 393-410.
Stromberg, Per. Conflicts of interest and market illiquidity in bankruptcy auctions: Theory and tests. The Journal of Finance 55.6 (2000): 2641-2692.
Vig, Vikrant. Access to collateral and corporate debt structure: Evidence from a
natural experiment. The Journal of Finance 68.3 (2013): 881-928.
41
1.9
Appendix A. Model of Effect of Bankruptcy
Protection on Household Borrowing
To explore the previous explanation, gain further insights into the effects of changes
in the bankruptcy reforms on the supply of credit, and to guide the empirical analysis,
we provide a simple model of the credit market where we abstract from considering the
moral hazard and adverse selection behavior of borrowers. In our model, we highlight
the effect of the increase of partial insurance provided by bankruptcy protection in
the credit market equilibrium outcome, and how even in the absence of asymmetric
information we could observe a demand effect.
We do this using a two period model, where the agent needs to borrow in order to
consume at period 1. Formally, the agent will consume c, at t=O and ci(s) at t=1,
where s C {B, G} (good and bad states in t=1), with the correspondent probability
{p, 1 - p}
The agent is endowed with a wealth only at t=1, his wealth is a combination
of home equity H (exogenous), and income y. For simplicity, assume that income
follows a binomial distribution given by y(G) = W > 0 and y(B) = 0 . Exists a level
of protection P (exogenously determinate)
The agent's consumption will be given by
co = b
ci = y + H - Min{(1 + R)b, y + Max(H - P,0)}
where R is endogenously determined
Agent's Maximization Problem
Given this setup, the agent will solve the following problem
V(b)
=
Max u(co) +
E[u(c)]
Subject to the consumption above. Therefore, the agent's consumption in period
2 will be given by:
* No default, total repayment: ci = y + H - (1
+ R)b
* Default and home-equity is not fully protected (H - P) > 0: c1
* Default and home-equity is fully protected (H - P) < 0: ci = H
42
=
P
Bank's break even condition
It is given by
(1 +r)b = E[Min{(1 + R)b,y + Max(H - P,0)}]
where r is the risk free rate (exogenous). The payoff for the bank are given by:
* No default, total repayment: b(1
+ R)
" Default and home-equity is not fully protected: y
+ H- P
" Default and home-equity is fully protected: y
Consider a risk-averse agent, u(x) = ln(x), the solution of the problem above defines
three regions as a function of the level of protection. Figure 1-4 illustrate the shape
of the numerical solution using the following set of parameters r = 0.05, /3 = 0.925,
p = 0.5, W = 5k.
Fixed borrowing (between 0, P): There is no default; banks lend at a risk-free
rate and the borrower demands a fixed quantity not related to the level of protection.
Increase in borrowing (between P, P*): There is a probability of default
greater than zero, interest rates go up, but quantities go up too. The agent's marginal
utility of consumption at t = 0 is greater than the marginal cost in the good state,
conditional on the level of protection on the bad state, that ensure a given level of
consumption.
Decrease in borrowing (between P*, P): The probability of default increases,
and interest rates go up even more. Agents will decrease the equilibrium amount of
debt with respect to the previous region, and the marginal cost in the good state
overcomes the benefit of consumption today, given the level of protection in the bad
state.
43
Figure 1-1: Debt Growth and Bankruptcy Filings
This figure plots the yearly number of non-business filings in the US from 1994 until 2012 extracted
from the Statistics Division of the Administrative Office of the United States Courts, and the adjusted
total revolving debt in the US extracted from the Federal Reserve Board of Governors Consumer
Credit Report.
8
N4
OS
1994
IM9
1998
20DO
Yearly non-businms
2004
2002
fifins US
-
44
2006
2008
2010
Consumer Revo"vn Debt US
2012
Figure 1-2: States that Changed their Level of Bankruptcy Protection
This figure shows in dark the counties that were at some point treated between 1999 and 2005;
"eventually" treated, in other words the level of bankruptcy protection changed at some point during
that period. Lightly colored counties are the counties in which the level never changed, "never"'
treated. Counties in gray represent counties for which FRBNY Consumer Credit Panel/Equifax did
not provide information because their population was below 10,000 households during our sample
period.
4.-
45
Figure 1-3: Iustration of Different Demand and Supply Responses
This figure uses supply and demand curves to illustrate possible net effects. Baseline Equilibrium is
the initial equilibrium before the change. Increase in Price, No Increase in Q, show the effect when
the supply response totally and perfectly upsets the demand increase. Increase in Price, Decrease
in Q, show the effect when the supply response is stronger than the demand increase. Increase in
Price, Increase in Q, show the effect when the demand effect dominates.
P
Do
Da
s
Do- - - -- -- - -=
So
So
so
Pa -
Po
!
Qo
P
Db
Qo=Qa
Q
Baseline Equilibrium
Q
Increase in Price, No Increase in Q
Dc
Sb
Sc
Do
Pb
Do
so
so
Pc
....
.-
-......--- - ..-.- ..PO
Po
s c, QC
Increase in Price, Increase in Q
Q
Increase in Price, Decrease in Q
46
Q
Figure 1-4: Ilustration of a Solution of the Model
This figure shows a stylized, schematic solution of the path obtained by solving numerically the model
in Appendix A; the top figure shows the relationship between the debt amount and protection levels.
The bottom figure shows the relationship between price and protection levels.
0
E
ee
Crdi ExasoardtCnrcio
fpoetoP
V
Credit Expansion
Credit Contation
47
Level of protection (P)
Table 1.1: Summary Statistics Data
All S ample
N=1 5,519
Levels
Eventually Treated
N=7,091
Mean
Std. Dev.
Never Treated
N=8,428
Mean
Std. Dev.
Mean
Std. Dev.
Debt to Income (DTI)
Mortgage Debt to Income (MTI)
Credit Card Debt to Income (CCTI)
Auto Loan Debt to Income (ATI)
1.23
0.90
0.16
0.17
0.48
0.45
0.04
0.06
1.29
0.97
0.16
0.16
0.52
0.49
0.04
0.06
County Total
County Mortgage
County Credit Card
County Auto
USD)
USD)
USD)
USD)
2.89
2.33
0.29
0.26
10.51
9.01
0.83
0.76
3.93
3.25
0.36
0.32
13.95
12.08
1.03
0.92
2.01
Pers. Unsec. Int. Rate (bp)
Credit Card hit. Rate (bp)
30 yr Fix. Mtg. Int. Rate (bp)
12.8
13.1
6.6
2.2
2.7
0.7
12.8
13.4
6.6
2.2
2.7
0.7
12.9
12.8
6.6
2.2
2.7
0.7
Mortgage Delinquency ('/ of pop)
Credit Card Deliquency (W of pop)
Auto Delinquency (W of pop)
1.5
8.2
2.4
1.3
3.5
1.5
1.5
7.8
2.3
1.2
3.1
1.4
1.6
8.5
2.4
1.3
3.8
1.5
100.306
269,477
5.56
1.90
2051
7.84
123,735
2.46
5.35
331.573
80.594
1.43
5.30
200.934
4.11
1.93
Debt
Debt
Debt
Debt
(bil.
(bil.
(bil.
(bil.
County Household Population
IRS County Income (bil. USD)
Unemployment Rate
No. of Bankruptcy Filing (1998)
W of Owner Occupancy (2000 )
Changes
1.90
5.32
604
73.35
N=13,302
Mean
Std. Dev.
6.85
1.87
N=6,078
Mean
Std. Dev.
1.18
0.84
0.17
0.17
1.57
0.22
0.22
0.45
0.41
0.05
0.07
6.18
5.06
0.61
0.60
*
N=7,224
Mean
Std. Dev.
DTI
MTI
CCTI
ATI
Change
Change
Change
Change
0.099
0.115
0.051
0.098
0.113
0.101
0.109
0.149
0.118
0.156
0.115
0.053
0.096
0.145
0.112
0.146
0.098
0.115
0.049
0.101
0.116
0.151
0.124
0.165
Debt
Debt
Debt
Debt
Growth
Growth
Growth
Growth
0.122
0.133
0.076
0.117
0.091
0.120
0.099
0.125
0.123
0.133
0.078
0.115
0.089
0.119
0.093
0.118
0.122
0.133
0.075
0.119
0.092
0.120
0.104
0.130
Pers. Unsec. it. Rate Change (bp)
Credit Card JIt. Rate Change (bp)
30-yr Fix. Mtg. Int. Rate Change (bp)
-0.09
-0.75
-0.34
0.94
1.88
0.50
-0.12
-0.65
-0.34
0.93
1.84
0.49
-0.06
0.95
-0.84
-0.33
0.51
Income Growth
Unemployment Rate Change
House Price Growth
0.033
0.111
0.075
0.053
0.963
0.046
0.032
0.115
0.088
0.054
0.931
0.050
0.033
0.108
0.062
0.052
0.989
0.037
Total
Mortgage
Credit Card
Auto
*
1.91
** *
Note. "All Sample" refers to all counties in the sample period. "Eventually Treated" refers to counties treated during
the sample period, that is, states that changed their level of protection during the sample period. "Never Treated"
refers to counties not treated during the sample period. County Debt (in bil. USD) for mortgage, credit card and auto
loans, is obtained from the FRBNY Consumer Credit Panel/Equifax. IRS County Income (in bil. USD) is measured
as total wages and salary in that county. Debt to Income is constructed using the two county measures described
above. Personal unsecured, credit card, and 30-year fixed mortgage rates are constructed from branch-setter level rates
from Rate-Watch. Delinquency rates for mortgage, credit card, and auto loans are from the FRBNY Consumer Credit
Panel/Equifax, and represent the fraction of households that are 90+ days delinquent. County household population
is the number of household per county and year in the FRBNY Consumer Credit Panel/Equifax. No. of Filings is
the number of non-business filings in a county in 1998 from the American Court System. % of Owner Occupancy is
the percentage of home ownership in a county in 2000 from the Census Bureau. For a complete description of the
data sources see section 3.1. Data Description. House price growth is extracted from the Federal Housing Finance
Agency (FHFA) House Price Index (HPI) data at a state level. The number of observations refers to the number of
county-year observations. Almost all variables are available for every county (2,218), with the exception of interest
rates, which are only available for (1232, 1323 and 1340 counties respectively). *, **, and *** denotes significance
at the 10%, 5%, and 1% level cluster at the state level for the mean differences between "Eventually Treated" and
"Never Treated" sample. The sample period is from 1999 to 2005.
48
Table 1.2: Summary Statistics Protection Level
All Sample
Mean
Std. Dev.
p5
p2 5
p50
p7 5
p9 5
Protection Level
Homestead
Personal Assets
73,627
63,932
9,695
75,125
73,356
5,965
13,000
7,500
2,900
23,200
20,000
5,000
55,800
40,000
8,400
166,200
150,000
11,000
unlimited
unlimited
25,000
Unlimited States
No. of States
7
50
p50
p75
p95
p2
5
Eventually Treated
Mean
Std. Dev.
p5
Protection Level
Homestead
Personal Assets
85,655
75,243
10,411
86,100
84,838
6,061
11,000
0,000
3,000
32,300
25,000
7,200
51,000
40,000
9,100
110,300
100,000
11,000
390,000
350,000
25,000
No. of States
26
Protection Changes
No. of Changes
38,841
37
52,992
2,000
3,250
15,400
50,000
200,000
Never Treated
Mean
Std. Dev.
p5
p2 5
p50
p7 5
p9 5
Protection Level
Homestead
Personal Assets
56,922
48,222
8,700
52,366
49,678
5,705
14,400
10,000
2,900
20,700
13,750
4,800
57,700
45,000
6,300
586,000
575,000
12,300
unlimited
unlimited
42,000
No. of States
24
Note. "All Sample" refers to all counties in the sample period. "Eventually Treated" refers to counties treated during
the sample period, that is states that changed their level of protection during the sample period. "Never Treated"
refers to counties not treated during the sample period. Protection Level is the nominal value of household protection
under Chapter 7. Homestead is the amount of home-equity protected under Chapter 7. Personal Assets, is the amount
of assets protected under Chapter 7, such as, books, furniture, jewelry, etc. The exact description depends on the
state. Unlimited States is the number of states with unlimited home-equity protection during our sample period.
Protection Changes is constructed based on the yearly changes in the level of protection. Levels of protection and
homestead are different at 10% between "Eventually Treated" and "Never Treated". The sample period is from 1999
to 2005.
49
Protection
Growth s,t
State
Linear
Trend
No
Linear
Trend
Level
Controls
(3)
Changes
Level
Eventually
Treated
(7)
Changed
Once
Change
> 0.15
(8)
0.018**
(0.008)
~(9)(1)
0.012***
(0.004)
Dummy
Treatment
Levels
(0.013)
0.027**
Level on
Level
State FE
(11)
1)
(0.011)
0.023**
(10)
County FE
Level
Level on
Table 1.3: Effect of Bankruptcy Protection on Debt. Credit Card Debt
County
Linear
Trend
(2)
0.022**
(0.009)
(7)
(6)
0.017**
(0.008)
0.002
(0.002)
Controls +
Inc-Year
Uep-Year
(5)
0.017**
(0.008)
0.002
(0.002)
-0.105
(0.083)
(0.007)
0.002
(0.002)
-0.103
(0.086)
(4)
0.017**
0.002
(0.003)
-0.049
(0.108)
(0.007)
0.003
(0.003)
-0.118
(0.086)
0.018**
0.000
(0.002)
-0.183*
(0.099)
(0.008)
0.002
(0.002)
-0.203**
(0.102)
0.019**
0.003
(0.002)
-0.139***
(0.037)
(1)
0.018
(0.008)
0.002
(0.002)
-0.109
(0.085)
Protection
Level s,t
Unemployment
Rate Change
-0.102
(0.086)
0.007*
(0.004)
House Price
Index Growth
0.003
(0.003)
-0.263***
(0.053)
0.079*
(0.047)
-0.166***
(0.042)
0.951***
(0.006)
0.079*
(0.047)
0.251***
(0.047)
15,519
50
0.081
(0.051)
0.002
15,519
50
y
0.138*
(0.077)
(0.003)
13,302
50
y
0.088**
(0.041)
0.005*
13,302
50
y
0.073*
(0.041)
(0.003)
0.070**
(0.029)
11,478
39
y
y
0.142***
(0.040)
0.083***
(0.031)
0.010
(0.020)
6,078
26
y
y
0.30
0.134***
(0.041)
0.023
(0.021)
13,302
50
y
y
0.30
0.079*
(0.047)
Rate
13,302
50
y
y
0.30
Incomue
Growth
House Price
13.302
50
y
0.29
Unemployment
Incomue
13,302
50
y
0.31
13,302
50
y
y
0.30
Y
y
y
0.28
y
0.30
y
Y
0.29
No. of Obs.
No. of Clusters
County FE
State FE
Year FE
R-Squared
Notes. This table shows the estimated coefficient following specification (1) of log changes to credit card debt on log changes in bankruptcy protection at the county level.
Debt county data is from the
FRBNY Consumer Credit Panel/Equifax. Protection Growth is the log change in the level of protection in state s at time t. Protection Level is the level of protection
in state s at time t. Unemployment
rate change is the change in unemployment rate in county i at time t from BLS. House price growth is the log change in the FHFA state level index for state
s at time t, and Income growth is the income
log change in county i at time t from IRS. Columns 1 and 2 show the result using county and state fixed effects respectively in the first difference specification.
Column 3 shows the results if we exclude
state or county fixed effect from specification (1). Column 4 shows the estimates including level of the controls. Column 5 shows the estimates
including level controls and income and unemployment
groups times year fixed effect, to allow for differential trends across states based on these observable characteristics. Column 6 shows the estimates
for a regression that only uses states treated during
the sample period, that is, states that changed their level of protection during the sample period. Column 7 shows the results if we only consider as
treated state that changed once. Column 8 shows
the estimates if we replace by zero changes below 0.15. Column 9 shows results if we replace the change with a dummy indicator that is one
if the change is greater than zero. Columns 10 and 11 show
the results of regression log levels of credit card debt on log levels of protection and including county and state fixed effect respectively. The
sample period is from 1999 to 2005. *, * and *** denotes
significance at the 10%, 5%, and 1% cluster at the state level respectively.
Y
0.08
Y
Y
().1()
Y
0.11
Y
0.13
0.11
Y
Y
0.09
0.09
Y
Y
0.09
Y
0.86
Y
0.97
y
significance at the 10%, 5%, and 1% cluster at the state level respectively
the estimates if we replace by zero changes below 0.15. Column 9 shows results if we replace the change with a dummy indicator that is one if the change is greater than zero. Column 10 and 11, show
and *** denotes
*, *,
the results of regression log levels of mortgage debt on log levels of protection and including county and state fixed effect respectively. The sample period is from 1999 to 2005.
Notes. This table shows the estimated coefficient following specification (1) of log changes to mortgage debt on log changes in bankruptcy protection at the county level. Debt county data is from the
FRBNY Consumer Credit Panel/Equifax. Protection Growth is the log change in the level of protection in state s at time t. Protection Level is the level of protection in state s at time t. Unemployment
rate change is the change in unemployment rate in county i at time t from BLS. House price growth is the log change in the FHFA state level index for state s at time t, and Income growth, is the income
log change in county i at time t from IRS. Columns 1 and 2 show the result using county and state fixed effects respectively in the first difference specification. Column 3 shows the results if we exclude
state or county fixed effect from specification 1. Column 4 shows the estimates including level of the controls. Column 5 shows the estimates including level controls and income and unemployment
groups times year fixed effect, to allow for differential trends across states based on these observable characteristics. Column 6 shows the estimates for a regression that only uses states treated during
the sample period, that is, states that changed their level of protection during the sample period. Column 7 shows the results if we only consider as treated state that changed once. Column 8 shows
R-Squared
Y
0.09
13.302
50
Y
15,519
50
13.302
50
Y
15.519
50
y
11,478
39
Y
13.302
50
Y
13,302
50
Y
6,078
26
Y
0.105***
(0.036)
0.133***
(0.039)
Income
13.302
50
0.265***
(0.041)
0.278***
(0.041)
House Price
13,302
50
-0.004
(0.004)
0.000
(0.004)
Uneniployient
Rate
13.302
50
Y
1.123***
(0.012)
0.319***
0.114
(0.107)
0.114
(0.107)
0.125
(0.114)
0.208
(0.181)
0.060
(0.081)
0.039
(0.079)
0.191**
(0.091)
0.185**
(0.091)
0.114
(0.107)
Income
Growth
No. of Obs.
No. of Clusters
County FE
State FE
Year FE
-0.223**
(0.089)
0.013
(0.069)
0.084
(0.161)
0.086
(0.161)
0.046
(0.209)
0.128
(0.256)
-0.345**
(0.174)
-0.378**
(0.170)
0.044
(0.079)
0.078
(0.161)
0.086
(0.161)
House Price
Index Growth
(0.067)
-0.055***
(0.007)
0.001
(0.004)
-0.004
(0.003)
-0.004
(0.003)
-0.005*
(0.003)
0.006
(0.026)
-0.001
(0.002)
0.007
(0.031)
Level on
Level
State FE
(11)
0.000
(0.003)
(9)
0.006
(0.007)
Level on
Level
County FE
(10)
-0.004
(0.003)
0.012
(0.012)
(7)
0.013
(0.013)
0.014
(0.014)
Dummy
Treatment
-0.003
(0.003)
(5)
0.007
(0.010)
0.005
(0.010)
Change
> 0.15
(8)
Changed
Once
Eventually
Treated
(6)
-0.003
(0.003)
(3)
0.008
(0.015)
0.011
(0.012)
Inc-Year
Uep-Year
Level
Controls
(4)
-0.004
(0.003)
(1)
0.011
(0.012)
No
Linear
Trend
State
Linear
Trend
(2)
Levels
Unemployment
Rate Change
Protection
Level st
Protection
Growth s.t
County
Linear
Trend
Level
Controls +
Changes
Table 1.4: Effect of Bankruptcy Protection on Debt. Mortgage Debt
Protection
Growth s,t
State
Linear
Trend
No
Linear
Trend
Changes
Eventually
Treated
(6)
Changed
Once
Change
Dummy
Treatment
Table 1.5: Effect of Bankruptcy Protection on Debt. Auto Debt
County
Linear
Trend
-0.005*
(0.003)
(9)
0.002
(0.008)
-0.005*
(0.003)
-0.007
(0.112)
> 0.15
(8)
0.010
(0.013)
-0.004
(0.003)
-0.005
(0.113)
(7)
0.009
(0.015)
-0.011***
(0.003)
0.049
(0.150)
0.059
(0.038)
0.009
(0.012)
-0.005
(0.004)
-0.230*
(0.118)
0.059
(0.038)
Level
Controls +
Inc-Year
Uep-Year
(5)
0.013
(0.013)
-0.002
(0.003)
-0.134
(0.125)
0.054
(0.041)
Level
Controls
(4)
0.009
(0.012)
-0.005*
(0.003)
-0.104
(0.124)
0.121***
(0.043)
(3)
0.009
(0.014)
-0.004
(0.003)
0.107**
(0.054)
0.020
(0.032)
(2)
0.009
(0.013)
-0.005*
(0.003)
-0.013
(0.113)
0.031
(0.032)
(1)
0.009
(0.013)
Unemployment
Rate Change
-0.005
(0.113)
0.127***
(0.030)
Protection
Level s,t
House Price
Index Growth
0.124***
(0.032)
-().009*
(0.005)
0.059
(0.038)
-0.011**
(0.005)
(0.045)
Income
Growth
Unemployment
Rate
0.009
(0.043)
Levels
15,519
50
Y
0.249***
(0.038)
0.107*
(0.055)
15,519
50
0.928***
(0.1)08)
0.061
(0.069)
Level on
Level
State FE
(11)
13,302
50
Y
Y
0.85
Level on
Level
County FE
(10)
13,302
50
Y
Y
0.18
0.007
(0.027)
11,478
39
Y
Y
0.18
0.000
(0.024)
0.029
(0.030)
6,078
26
Y
Y
0.18
(0.005)
(0.029)
13,302
50
Y
Y
0.20
0.024***
13,302
50
Y
Y
0.19
-0.005
13,302
50
Y
0.19
(0.004)
13,302
50
Y
0.17
Y
Y
0.97
Y
Y
0.19
0.033
House Price
0.026
13,302
50
Y
Y
0.18
Income
No. of Obs.
No. of Clusters
County FE
State FE
Year FE
R-Squared
Notes. This table shows the estimated coefficient following specification (1) of log changes to auto debt on log changes in bankruptcy protection at the county level. Debt county data is from the FRBNY
Consumer Credit Panel/Equifax. Protection Growth is the log change in the level of protection in state s at time t. Protection Level is the level of protection in state s at time t. Unemployment rate
change is the change in unemployment rate in county i at time t from BLS. House price growth is the log change in the FHFA state level index for state s at time t, and Income growth is the income log
change in county i at time t from IRS. Columns 1 and 2 show the result using county and state fixed effects respectively in the first difference specification. Column 3 shows the results if we exclude state
or county fixed effect from specification 1. Column 4 shows the estimates including level of the controls. Column 5 shows the estimates including level controls and income and unemployment groups
times year fixed effect, to allow for differential trends across states based on these observable characteristics. Column 6 shows the estimates for a regression that only uses states treated during the sample
period, that is, states that changed their level of protection during the sample period. Column 7 shows the results if we only consider as treated state that changed once. Column 8 shows the estimates
if we replace by zero changes below 0.15. Column 9 shows results if we replace the change with a dummy indicator that is one if the change is greater than zero. Column 10 and 11 show the results of
regression log levels of auto debt on log levels of protection and including county and state fixed effect respectively. The sample period is from 1999 to 2005. *, *
and *** denotes significance at the
10%, 5%, and 1% cluster at the state level respectively.
Table 1.6: Determinants of Bankruptcy Protection Levels and Changes
Protection Level s,t
Protection Growth s,t
Protection Dummy s,t
-1.837***
(0.671)
2.983***
(0.770)
(3)
-0.809**
(0.354)
1.691***
(0.619)
(4)
-0.537
(0.572)
0.970
(0.762)
(5)
-0.697
(0.701)
2.700***
(0.776)
-0.858
(0.789)
1.806*
(0.994)
-3.332
(5.359)
4.635
(5.238)
0.836
(1.001)
0.348
(1.106)
-0.316
(0.644)
-0.537
(0.763)
-1.150
(0.821)
-1.805*
(1.001)
-1.101
(1.270)
-1.020
(1.115)
-2.380
(1.834)
-2.274*
(1.287)
-0.023
(0.190)
0.033
(0.148)
0.028
(0.036)
-0.081*
(0.042)
0.005
(0.027)
-0.016
(0.028)
0.002
(0.033)
-0.008
(0.032)
0.027
(0.042)
-0.056
(0.050)
0.026
(0.048)
-0.058
(0.065)
State Real GDP/Growth s,t
3.703
(4.464)
State Real GDP/Growth st-1 -6.950
(3.916)
0.504
(0.871)
-1.448
(0.742)
0.474
(0.668)
-0.277
(0.282)
1.028
(1.018)
0.425
(0.457)
-1.665
(1.034)
-1.429
(0.789)
-0.911
(1.343)
-0.547
(0.802)
No. Filings/Growth st
-0.299*
(0.250)
No. Filings/Growth st-1 -0.482
(0.245)
0.125*
(0.039)
0.194***
(0.072)
0.030
(0.045)
0.053
(0.047)
-0.123
(0.098)
-0.045
(0.071)
0.060*
(0.069)
0.026
(0.064)
-0.114
(0.098)
-0.080
(0.090)
Political Climate st-1 0.045**
(1.509)
-0.289***
(0.171)
0.010
(0.161)
0.400
(0.234)
0.151
(0.151)
0.608
(0.458)
15.885*
(8.597)
-13.235*
(9.202)
1.077
(1.257)
-0.219*
(1.206)
1.554
(1.299)
-0.720
(0.929)
0.996
(1.928)
-1.159
(1.477)
3.264
(2.009)
-0.525*
(1.849)
3.190
(2.399)
-0.893
(2.200)
350
350
Y
Y
0.12
300
300
Y
Y
0.22
300
300
Y
Y
0.25
(1)
House Price/Growth st
-3.900
(4.616)
House Price/Growth st-1 5.287
(4.503)
Medical Exp./Growth st
Medical Exp./Growth st-i
Unemp. Rate/Change s,t
Unermp. Rate/Change st-1
Personal Income/Growth s,t
Personal Income/Growth s,t-1
No. of Obs.
State FE
Year FE
R2
Y
0.13
(2)
Y
0.07
Y
0.13
(6)
Note. This table shows the estimated coefficient of regression of bankruptcy protection on contemporaneous and lag
values of variables that could determinate the changes in protection levels. House Price s,t is the level or growth of
house prices in state s at time t, from FHFA. Medical expenses is the level of growth in state's annual total medical
expenses from the National Health Statistic. No. of Filings, is the number or change in the number of filings for
non-business bankruptcies in a state. Political Climate s,t is defined as the share of democratic votes in the closer
House of Representative election. State GDP and Personal Income are from BEA, and Unemployment Rate from
BLS. Columns 1 and 2 show the coefficient of regressions of the protection level on levels of the explanatory variables
using only year, and year and state fixed effect. Columns 3 and 4 show the coefficient of regressions of the growth
in protection on growth of the explanatory variables using only year, and year and state fixed effect. Columns 5 and
6 show the coefficient of regressions of a dummy that is one if the growth in protection is greater than zero on the
explanatory variables growth using only year, and year and state fixed effect. The sample period is from 1999 to 2005.
*, **, and *** denotes significance at the 10%, 5%, and 1% cluster at the state level.
53
Table 1.7: Dynamics of the Change in Protection Levels on Credit Card Debt
No
Linear Trend
(1)
1 Period
County
Linear Trend
(2)
County
Linear Trend
(3)
No
Linear Trend
(4)
0.001
(0.019)
County
Linear Trend
(5)
-0.004
(0.026)
County
Linear Trend
(6)
-0.005
(0.025)
Protection
Growth st-1
-0.008
(0.008)
-0.010
(0.010)
-0.012
(0.009)
-0.007
(0.009)
-0.007
(0.015)
-0.010
(0.015)
Protection
Growth st
0.018**
(0.007)
0.019**
(0.008)
0.016**
(0.007)
0.018**
(0.007)
0.022**
(0.009)
0.020**
(0.008)
0.002
(0.006)
0.006
(0.008)
0.006
(0.009)
0.003
(0.006)
0.010
(0.010)
0.010
(0.011)
0.010**
(0.005)
0.016***
(0.005)
0.016***
(0.005)
2 Periods
Protection
Growth st-2
Protection
Growth st+
Protection
Growth st+2
Unemployment
Rate Change
0.002
(0.002)
0.002
(0.002)
0.000
(0.002)
0.002
(0.002)
0.002
(0.002)
0.001
(0.002)
House Price
Index Growth
-0.139***
(0.037)
-0.108
(0.085)
-0.212**
(0.101)
-0.142***
(0.037)
-0.120
(0.085)
-0.229**
(0.100)
Income
Growth
0.143***
(0.040)
0.080*
(0.047)
0.073*
(0.042)
0.143***
(0.040)
0.080*
(0.047)
0.072*
(0.041)
0.005*
Unemployment
Rate
(0.003)
0.004
(0.003)
House Price
0.085
(0.030)
0.086
(0.029)
Income
0.024
(0.021)
0.025
(0.021)
No. of Obs
No. of Clusters
County FE
Year FE
R-Squared
13,302
50
Y
0.28
13,302
50
Y
Y
0.30
13,302
50
Y
Y
0.30
13,302
50
Y
0.28
13,302
50
Y
Y
0.30
13,302
50
Y
Y
0.31
Note. This table shows the estimated coefficient following specification (1) of log changes to credit card debt on
log changes in bankruptcy protection at the county level. Debt county data is from the FRBNY Consumer Credit
Panel/Equifax. Protection Growth is the log change in the level of protection in state s at time t. Unemployment
rate change is the change in unemployment rate in county i at time t from BLS. House price growth is the log change
in the FHFA state level index for state s at time t, and Income growth is the log change in income in county i at time
t from IRS. Columns 1 and 4 show the without the inclusion of county fixed effects, including one lag and lead, and
two lags and two leads. Columns 2 and 5 show the results with the inclusion of county fixed effect for including one
lag and lead, and two lags and two leads, Columns 3 and 6 are the same than before but including level controls. The
sample period is from 1999 to 2005. *, **, and *** denotes significance at the 10%, 5%, and 1% cluster at the state
level respectively.
54
Table 1.8: Local Business Conditions. Neighboring County-pairs across State Borders. Credit Card Debt
All
Equal Income
County-Pairs
County-Pairs
Low Income
County-Pairs
Liner
Trend
(2)
State
Linear
Trend
(3)
County
Liner
Trend
(4)
State
Linear
Trend
(5)
County
Liner
Trend
(6)
-0.005
(0.011)
0.015
(0.010)
0.015*
(0.009)
0.099**
0.098**
(0.046)
(0.044)
0.003**
(0.002)
0.003**
(0.002)
0.002
(0.003)
0.001
(0.003)
0.002*
(0.005)
0.001**
(0.005)
House Price
Index Growth
-0.322**
(0.157)
-0.317**
(0.154)
-0.266
(0.178)
-0.261
(0.171)
-1.040*
(0.550)
-1.037**
(0.526)
Income
Growth
0.095***
(0.024)
0.043
(0.027)
0.122*
(0.071)
0.066
(0.075)
0.121
(0.125)
0.102
(0.122)
9,168
48
9,168
48
Y
3,984
46
3,984
46
Y
1,188
33
1,188
33
Y
County
Protection
Growth s,t
State
Linear
Trend
(1)
-0.006
(0.011)
Unemployment
Rate Change
No. of Obs
No. of Clusters
County FE
State FE
County-Pair-Year FE
R-Squared
Y
Y
0.70
Y
Y
0.67
Y
0.70
Y
0.67
Y
Y
0.63
Y
0.62
Note. This table shows the estimated coefficient following specification (2) of log changes in credit card debt on
log changes in bankruptcy protection at the county level. Debt county data is from the FRBNY Consumer Credit
Panel/Equifax. Protection Growth is the log change in the level of protection in state s at time t. Unemployment
rate change is the change in unemployment rate in county i at time t from BLS. House price growth is the log change
in the FHFA state level index for state s at time t, and Income growth is the log change in income in county i at time
t from IRS. Columns 1 and 2, show the estimates for state and county fixed effect for all neighboring county-pairs
sample. Columns 3 and 4 show the results including state and county fixed effect for the sub-sample of neighboring
county-pairs for which both counties are in the same income bucket. Columns 5 and 6 show estimates with state and
county fixed effect for only the neighboring county-pairs in the same income bucket and in the lowest tercile of the
income distribution. The sample period is from 1999 to 2005. *, *, and *** denotes significance at the 10%, 5%,
and 1% cluster at the state level respectively.
55
Table 1.9: Heterogeneous Treatment of Bankruptcy Protection on Credit Card Debt:
Income and Home ownership
Low Income
Income
Protection Growth s,t
(1)
0.007
(0.007)
Protection Growth st
x Low Income
0.022***
(0.007)
(2)
0.028**
(0.011)
Protection Growth s,t
x Low Home Ownership
Protection Growth st
x Med Income
Med Income
Home
Ownership
(3)
0.063***
(0.018)
Home
Ownership
(4)
0.020**
(0.010)
(5)
0.029
(0.019)
High Income
Home
Ownership
(6)
0.006
(0.006)
(7)
0.014
(0.009)
-0.050***
(0.018)
-0.012
(0.025)
(().009)
-0.049***
(0.016)
-0.014
(0.019)
-0.013
(0.012)
-0.011
0.013**
(0.006)
Protection Growth st
x Med Home Ownership
0.005*
Unemployment
Rate Change
0.003
(0.002)
0.005*
(0.003)
(0.003)
0.002
(0.002)
0.002
(0.002)
0.002
(0.003)
0.002
(0.003)
House Price
Index Growth
-0.109
(0.086)
-0.015
(0.094)
-0.012
(0.095)
-0.099
(0.098)
-0.099
(0.098)
-0.208**
(0.092)
-0.206**
(0.093)
0.137***
(0.040)
0.059**
(0.030)
0.057*
(0.031)
0.090***
(0.032)
0.088***
(0.028)
0.240***
(0.062)
0.227***
(0.064)
13,302
50
Y
0.29
4,536
50
Y
0.24
4,536
50
Y
0.24
4,422
50
Y
0.29
4,422
50
Y
0.30
4,344
50
Y
0.46
4,344
50
Y
0.48
Income
Growth
No. of Obs
No. of Clusters
State and Year FE
R-Squared
Note. This table shows the estimated coefficient following a variation of specification (1) that incorporates interactions.
Low/Med Income represents counties in the lowest/middle tercile of the within state income distribution. Low/Med
Ownership represents counties in the lowest/middle tercile of the within income bucket distribution. Column 1 shows
the result for the whole sample when interacted with income heterogeneity. Column 2 shows the result of specification
(1) restricted to the low income counties. Column 3 shows the within low income heterogeneity in homeownership.
Columns 4 to 7 replicates columns 2 and 3 for medium and high income levels. The sample period is from 1999 to
2005. *, **, and *** denotes significance at the 10%, 5%, and 1% cluster at the state level respectively.
56
0.820***
(0.157)
0.007
(0.217)
0.147
(0.183)
-0.004
(0.232)
0.317
(0.229)
Y
0.79
0.80
Y
0.29
0.21
0.23
0.21
Y
0.82
0.82
Note. This table shows the estimated coefficient following a variation of specification (1) of changes in interest rates (%) on changes in the level of protection. Personal Unsecured Loan and Credit Card
Debt are county averages of the interest rates in a county for each type of credit. Columns 1 and 7 show the result using state fixed effect. Columns 2 and 8 show the estimates using county fixed effect.
Columns 3 and 9 show the result restricting the sample to only the "eventually" treated sample. Columns 4 and 10 show the estimates looking at the dynamic effect of changes in protection on interest
rates. Colums 5, 6, 11, and 12 show the results including state and county fixed effect for the sub-sample of neighboring county-pairs for which both counties are in the same income bucket. The sample
period is from 1999 to 2005. *, **, and *** denotes significance at the 10%, 5%, and 1% cluster at the state level respectively.
0.14
1,621
45
Y
1.621
45
5,371
50
Y
2.430
26
Y
5.371
50
Y
5.371
50
1,621
44
Y
1,621
44
4.693
49
Y
2,338
25
Y
4.693
49
Y
4.693
49
No. of Obs
No. of Clusters
Cty and Year FE
State and Year FE
R-Squared
0.15
-0.868
(4.905)
-0.224
(4.195)
1.864***
(0.605)
1.440
(0.973)
1.886***
(0.600)
1.734***
(0.558)
2.904
(1.936)
2.299*
(1.255)
0.203
(0.383)
0.551
(0.622)
0.182
(0.385)
0.198
(0.268)
Income
Growth
0.13
-5.049
(8.608)
-5.857
(7.780)
3.625
(4.014)
2.606
(4.532)
3.691
(3.895)
5.179
(3.984)
1.072
(3.315)
-0.112
(3.153)
5.154***
(1.607)
4.363**
(2.159)
4.812***
(1.623)
4.938***
(1.629)
House Price
Index Growth
Y
0.17
-0.059
(0.160)
-0.038
(0.151)
-0.086
(0.095)
-0.100
(0.096)
-0.103
(0.090)
-0.118
(0.089)
0.084
(0.107)
0.106
(0.103)
-0.009
(0.048)
-0.020
(0.073)
0.001
(0.050)
0.775
(0.573)
0.003
(0.046)
0.875*
(0.515)
County-Pairs
St Linear Cty Linear
Trend
Trend
(12)
(11)
Unemploymient
Rate Change
0.256
(0.273)
0.755***
(0.177)
-0.286
(0.205)
0.296*
(0.170)
Protection
Growth s.t+2
0.373**
(0.147)
0.308*
(0.166)
0.415***
(0.144)
-0.132
(0.106)
0.389***
(0.147)
Protection
Growth s,t+1
Protection
Growth st
0.083
(0.677)
-0.022
(0.274)
Cty Linear
Trend
(10)
Protection
Growth st-1
St Linear
Trend
(7)
0.584
(0.464)
Cty Linear
Trend
(4)
Credit Card Debt
Eventually
Cty Linear Cty Linear
Trend
Trend
(8)
(9)
-0.260
(0.395)
Cty Linear
Trend
(2)
County-Pairs
St Linear Cty Linear
Trend
Trend
(6)
(5)
Protection
Growth s.t-2
(1)
St Linear
Trend
Eventually
Cty Linear
Trend
(3)
Personal Unsecured Loan
Table 1.10: Effect of Bankruptcy Protection on Interest Rates: Personal Unsecured Loans and Credit Cards
Protection
Growth s,t
15 Yr-Fixed
Eventually
Cty Linear
Trend
(6)
0.005
(0.035)
St Linear
Trend
(7)
(7)
0.026
(0.029)
(8)
0.029
(0.030)
-0.040
(0.017)
Eventually
Cty Linear
Trend
(9)
0.027
(0.034)
30 Yr-Fixed
Table 1.11: Effect of Bankruptcy Protection on Interest Rates: Mortagage Credit
3 Yr-ARM
Cty Linear
Trend
(5)
0.019
(0.042)
0.004
(0.022)
Trend
Cty Linear
St Linear
Trend
(4)
0.014
(0.041)
0.001
(0.019)
Eventually
Cty Linear
Trend
(3)
0.041
(0.057)
-0.022
(0.017)
Cty Linear
Trend
(2)
0.053
(0.062)
-0.002
(0.011)
0.234
(0.261)
St Linear
Trend
(1)
0.037
(0.051)
-0.001
(0.009)
0.017
(0.252)
-0.048**
(0.026)
-0.039
(0.246)
-0.100***
(0.041)
0.637
(0.403)
-0.066***
(0.031)
Unemployment
Rate Change
0.045
(0.332)
-0.317***
(0.115)
0.009
(0.319)
-0.034
(0.139)
2.690**
(1.094)
2.244***
(0.648)
-0.029
(0.107)
2.332***
(0.677)
House Price
Index Growth
-0.136
(0.111)
-0.005
(0.118)
-0.191
(0.290)
2;732
25
y
-0.003
(0.085)
-0.093
(0.228)
5,533
49
y
-0.485
(0.374)
Income
Growth
5,533
49
2,802
26
y
1,945
24
Y
5,723
50
y
3,919
47
Y
5,723
50
3,919
47
y
0.86
0.87
0.87
y
0.87
0.85
0.86
0.85
Y
0.85
0.85
No. of Obs
No. of Clusters
Cty and Year FE
State and Year FE
R-Squared
Note. This table shows the estimated coefficient following a variation of specification (1) of changes in interest rates (%) in the level of protection.
3 Yr-ARM, 15 Yr-Fixed, 30 Yr-Fixed, are county
averages of the interest rates in a county for each type of credit. Columns 1, 4, and 7 show the result using state fixed effect. Columns 2, 5 and 8,
show the estimates using county fixed effect. Columns
3, 6 and 9, show the result restricting the sample to only the "eventually" treated sample. The sample period is from 1999 to 2005. *, *, and *** denotes significance
at the 10%, 5%, and 1% cluster at
the state level respectively.
00
Table 1.12: Effect of Bankruptcy Protection on Debt. Number of Credit Cards and
Entry
Panel A. All individuals
Entry
Number of Credit Cards
A in
Debt Balance
Protection
Growth s.t
(1)
0.076***
(().009)
Unenploynient
Rate Change
A in
N Credit Cards
A in
N Credit Cards Conditional on n>0
(3)
(2)
A in
N Credit Cards
Conditional on n>0
& Balance >0
(4)
Open First
Credit Card
(5)
Credit Card
Balance
Becomes >0
(6)
0.054***
(0.019)
0.082***
(0.026)
0.093***
(0.029)
0.001
(0.003)
-0.002
(0.006)
0.002
(0.003)
0.008**
(0.004)
0.008
(0.005)
0.009*
(0.005)
0.001
(0.001)
0.001
(0.002)
House Price
Index Growth
-0.070*
(0.041)
-0.050
(0.037)
-0.043
(0.049)
-0.039
(0.044)
-0.005
(0.008)
-0.031
(0.020)
Incone
Growth
0.012
(0.016)
-0.048***
(0.016)
-0.017
(0.018)
0.001
(0.017)
-0.011***
(0.004)
-0.063***
(0.014)
366,362
40
Y
619,726
40
Y
0.02
454,688
40
Y
0.02
359,235
40
Y
0.02
555,007
40
Y
221,849
:39
Y
0.01
N of Ohs
N of Clusters
R-Squaredl
0.00
0.01
Panel B. Home owners
Entry
Number of Credit Cards
A in
N Credit Cards
A in
N Credit Cards Conditional on n>0
(3)
(2)
A in
N Credit Cards
Conditional on n>O
& Balance >0
(4)
Open First
Credit Card
(5)
Credit Card
0.081***
(0.020)
0.103***
(0.024)
0.115***
(0.032)
-0.002
(0.003)
Balance
Becomes >0
(6)
-0.006
(0.006)
0.000
(0.004)
0.009*
(0.005)
0.009
(0.006)
0.009
(0.007)
0.000
(0.001)
0.001
(0.002)
House Price
Index Growth
-0.088*
(0.052)
-0.052
(0.057)
-0.045
(0.067)
-0.032
(0.071)
-0.003
(0.007)
-0.044
(0.029)
Incone
Growth
0.014
(0.017)
-0.036*
(0.021)
-0.006
(0.024)
0.006
(0.021)
-0.005*
(0.003)
-0.060***
(0.016)
N of Obs
N of Clusters
Cty and Year FE
R-Squared
210.863
39
Y
304,005
39
Y
0.02
248,955
39
Y
205,458
:39
Y
10.02
291,353
39
Y
10:3,854
37
Y
0.01
0.01
A in
Debt Balance
Protection
Growth s.)
(1)
0.102***
(0.014)
Uneniployient
Rate Change
0()0
0.02
Note. This table shows the estimated coefficient following a variation of specification (1). Panel A uses all individuals
in counties below the median income. Panel B restricts the sample to homeowners, defined as individuals for whom
some home debt is observed during the sample period. Column 1 shows the estimated of log changes in individuals'
credit card balance on log changes in the levels of bankruptcy protection. Column 2 shows the estimates of the effect of
personal bankruptcy protection on the number of credit cards changes. Column 3 restricted the previous specification
to borrowers with more than 0 credit card. Column 4 shows the estimates for individual with more than 0 credit
cards and a positive balance. Column 5 shows the estimates for a linear probability model on the timing of opening
the first card, in this case the dependent variable is one if the individual did not have a credit card at t-1, but has one
at t. Column 6 shows the same linear probability model estimates, but defining entry based on the timing of going
to a positive balance, in other words the variable is one if the individual did not have a positive balance at t-1 but
has one at t. The sample period is from 1999 to 2005. *, **, and *** denotes significance at the 10%, 5%, and 1%
cluster at the state level respectively.
59
t+1
(3)
t+2
0.004
(0.003)
t+3
(4)
-0.002
(0.005)
t
(5)
-0.009**
(0.004)
t+1
(6)
361,444
40
Y
Y
0.02
-0.001
(0.003)
t+2
(7)
359,783
40
Y
Y
0.01
0.003
(0.004)
t+3
(8)
(5)
t+1
(6)
-0.014***
0.004)
t+2
(7)
-0.007
(0.006)
(9)
-0.003
(0.004)
t+1
(10)
-0.008**
(0.002)
t+2
(11)
0.001
359,783
40
Y
Y
0.01
(0.002)
t+3
(12)
0.002
Severe
(0.003)
361,444
40
Y
Y
0.02
366,362
40
Y
Y
0.02
t
(9)
-0.001
(0.003)
t+1
(10)
-0.013***
(0.005)
t+2
(11)
-0.005
(0.003)
0.000
t+3
(12)
(0.004)
Severe
363,498
40
Y
Y
0.02
t
Table 1.13: Effect of Bankruptcy Protection on Credit Card Delinquency
(2)
0.000
(0.004)
363,498
40
Y
Y
0.02
Panel A. All individuals
t
-0.008**
(0.004)
366,362
40
Y
Y
0.02
90+ days
(1)
-0.001
(0.004)
359,783
40
Y
Y
0.01
120+ days
Protection
Growth s,t
361,444
40
Y
Y
0.02
t+3
(4)
(0.003)
120+ days
363,498
40
Y
Y
0.02
t+2
(3)
0.003
(0.002)
90+ days
t+1
(2)
-0.007
(0.007)
-0.003
-0.014***
(0.005)
t
366,362
40
Y
Y
0.02
t
(1)
-0.004
(0.004)
t+3
(8)
0.001
(0.003)
N of Obs
N of Clusters
Cty and Year FE
Uep/Income/HP Controls
R-Squared
Panel B. Home owners
Protection
Growth s,t
N of Obs 210,863
209,878
209,173
208,616
210,863
209,878
209,173
208,616
210,863
209,878
209,173
208,616
N of Clusters
39
39
39
39
39
39
39
39
39
39
39
39
Cty and Year
FE
Y
Y
Y
Y
Y
Y
Y
Y
Y
Y
Y
Y
Uep/Income/HP Controls
Y
Y
Y
Y
Y
Y
Y
Y
Y
Y
Y
Y
R-Squared
0.02
0.02
0.02
0.01
0.02
0.02
0.02
0.01
0.02
0.02
0.02
0.01
Note. This table shows the estimated coefficient
following
a
variation
of
specification
(1),
where
we
replace
the
dependent
variable
for
a
dummy
indicator
that is equal to 1
if the person is delinquent at the specified time. Panel A uses all individuals in counties below the median income with a positive balance. Panel
B restricts the sample to
homeowners, defined as individuals for whom some home debt is observed during the sample period. Columns 1 to 4 show the estimates where
delinquency is defined as being
delinquent 90 days or more. Column 5 to 8 show the estimates
where
delinquency
is
defined
as
being
delinquent
120
days
or
more.
Columns
9 to 12 show the estimates where
delinquency is defined as being severely delinquent. All regressions include controls. The sample period is from 1999 to 2005. *, **, and
*** denotes significance at the 10%,
5%, and 1% cluster at the state level respectively
Table 1.14: Effect of Bankruptcy Protection on Self-Employment
Self Employment
Credit Card
Startup > p50
Credit Card
Startup < p50
(1)
(2)
0.000
(0.002)
-0.003
(0.003)
(3)
-0.0l0(**
(0.004)
Protection Gowth s,t
x Low Income
0.006**
(0.003)
0.012***
(0.004)
0.024***
(0.007)
0.005
(0.004)
Protection Gowth st
x Med Incoie
0.003
(0.002)
0.008***
(0.003)
0.012**
(0.005)
0.006
(0.003)
Protection Gowth s,t
(4)
-0.002
(0.007)
(5)
-0.014
(0.009)
(6)
(7)
-0.003
(0.002)
-0.007
(0.003)
0.001***
0 .0 0 1 ***
0.01***
0.001
0.001
0.001
0.001
Rate Change
(0.000)
(0.000)
(0.001)
(0.001)
(0.001)
(0.001)
(0.001)
Honse Price
0.096***
(0.023)
0.097***
(0.022)
0.058**
(0.028)
0.057
(0.035)
0.056
(0.035)
0.059*
(0.033)
0.059
(0.033)
0.063***
0.063***
(0.010)
(0.009)
0.101***
(0.028)
0.126***
(0.037)
0.127***
(0.037)
0.085***
(0.025)
0.085
(0.025)
12,738
50
Y
12,738
50
Y
194,011
50
73,081
50
73,081
50
120,930
50
120,930
50
Y
Y
0.23
Y
Y
0.01
Y
Y
0.02
Y
Y
0.02
Y
Y
0.02
Y
Y
0.02
Unemployment
Index Growth
Income
Growth
Number of Observations
Number of Clusters
State FE
State x 2-digit industry
Year FE
R-Squared
0.21
Note. This table shows the estimated coefficient following a variation of specification (1) of log changes in selfemployment measures on log changes in the levels of protection. Column 1 shows the estimates for county selfemployment aggregates. Column 2 shows the results for the effect interacted with income heterogeneity for aggregate
self-employment. Column 3 shows the estimates interacted with low income using self-employment changes by industry
and county. Column 4 and 5 show the estimates for industries that used the level of credit card debt as a start-up
capital and Column 6 and 7 for industries that do not. The sample period is from 1999 to 2005. *, **, and
denotes significance at the 10%, 5%, and 1% cluster at the state level respectively.
61
Event
Baseline
Cty Linear
Table 1.15: Effect of Bankruptcy Protection on Credit Card Debt. Alternative Specifications
(1)
Debt to
Income
Event
Debt to
Income
% Change
in Debt
(10)
0.020**
(0.010)
Event
% Change
in Debt
Homestead
Only
(11)
0.017***
(0.006)
(0.006)
0.015**
Event
Homestead
Only
(12)
Event
Unemp.
Insurance
(9)
0.022**
(0.009)
Unemp.
Insurance
Baseline
Cty Linear
(8)
0.020*
(0.011)
(5)
0.018**
(0.008)
(7)
0.023**
(0.011)
Event
Unlimited
Change
(4)
0.017**
(0.008)
(6)
0.017**
(0.008)
Unlimited
Change
(3)
0.018**
(0.008)
(0.008)
0.018**
(2)
0.017**
(0.008)
-0.139***
(0.027)
Protection
Gowth st
-0.156***
(0.027)
0.002
Unlimited
Protection s,t
(0.003)
0.003
(0.003)
0.002
(0.002)
0.002
(0.002)
0.002
(0.003)
0.002
(0.003)
0.002
(0.002)
0.002
(0.002)
0.002
(0.003)
-0.210**
(0.094)
0.008***
(0.003)
0.002
(0.002)
-0.134
(0.091)
0.008***
(0.002)
Unemployment
Rate Change
-0.099
(0.088)
-0.130
(0.091)
-0.119
(0.086)
-0.111
(0.087)
-0.103
(0.086)
-0.109
(0.097)
-0.118
(0.086)
0.138*
(0.076)
-0.106
(0.094)
-0.102
(0.086)
0.080*
(0.047)
-0.155
(0.111)
House Price
Index Growth
0.139*
(0.077)
0.141*
(0.077)
0.080*
(0.047)
0.081*
(0.047)
0.138*
(0.077)
6,078
26
Y
0.18
0.111*
(0.059)
0.079*
(0.047)
13,302
50
Y
0.22
0.065*
(0.037)
Income
Growth
6,078
26
Y
0.29
5,916
24
13,302
50
Y
0.30
13.140
48
6,084
27
Y
0.29
Y
0.29
13,308
51
Y
0.30
Y
0.30
6,078
26
Y
0,29
6,078
26
Y
0.27
13,302
50
Y
0.30
13,302
50
Y
0.29
N of Obs
N of Clusters
Cty and year FE
R-Squared
Note. This table shows the estimated coefficient following a variation of the specification (1). Columns 1 and 2 replicated the main results. Columns 3 and
4 show the result when unlimited change of DC
is included as a dummy. Columns 5 and 6 show the results when controlling for level of unemployment insurance. Columns 7 and 8 replace the dependent variable
for debt to income change. Columns 9
and 10 replace the dependent variable for percentage changes in level of debt, and Columns 11 and 12 show the result if changes in the level of
protection are measured only as a home-equity protection.
The sample period is from 1999 to 2005. *, **, and *** denotes significance at the 10%, 5%, and 1% cluster at the state level respectively.
Table 1.16: Other Heterogeneous Treatment of Bankruptcy Protection. Credit Card
Debt
Protection Gowth st
Number
of Filing
Credit Card
90+ Delinq
Low Inc
Baseline
Bank
Concentration
Total
Debt/Income
Credit Card
Debt/Income
(1)
0.028**
(0.011)
(2)
(4)
(5)
(6)
0.085***
(0.021)
(3)
0.041**
(0.018)
0.043*
(0.024)
0.026**
(0.012)
0.048***
(0.017)
-0.086***
(0.022)
-0.028
(0.021)
-0.005
(0.039)
-0.004
(0.026)
-0.042***
(0.014)
-0.076***
(0.016)
-0.009
(0.031)
-0.040
(0.037)
0.010
(0.019)
-0.014
(0.018)
Protection Gowth s,t
x Low Income
Protection Gowth s,t
x Med Income
Unemployment
Rate Change
0.005*
(0.003)
0.005*
(0.003)
0.005*
(0.003)
0.005*
(0.003)
0.005*
(0.003)
0.005**
(0.002)
House Price
Index Growth
-0.015
(0.094)
-0.012
(0.094)
-0.013
(0.095)
-0.018
(0.095)
-0.018
(0.094)
-0.013
(0.094)
Income
0.059**
Growth
(0.030)
0.060**
(0.030)
0.062**
(0.031)
0.101***
(0.031)
0.064**
(0.030)
0.058**
(0.029)
4,536
4,536
50
50
Y
Y
0.24
0.24
Note. This table shows the estimated coefficient following a variation of specification (1) that incorporate interactions,
N of Obs
N of Clusters
State and year FE
R-Squared
4,536
50
Y
0.24
4,536
50
Y
0.24
4,536
50
Y
0.24
4,536
50
Y
0.25
within low income counties. Low/Med represents counties in the lowest/middle tercile of the within state described
variable distribution. Column 2 shows the result for bank concentration. Column 3 for the total debt to income
heterogeneity. Column 4 for credit card debt to income. Column 5 for heterogeneity on the county level number of
filing in 1998. Column 6, using credit card delinquency heterogeneity defined as delinquency in 1999. The sample
period is from 1999 to 2005. *, **, and *** denotes significance at the 10%, 5%, and 1% cluster at the state level
respectively.
63
Table 1.17: Determinants of Bankruptcy Protection Levels and Changes. Eventually
Treated
Protection Level s,t
Protection Growth s,t
Protection Dummy s,t
(2)
-2.224**
(0.970)
3.147***
(0.985)
(3)
-0.984**
(0.445)
1.453*
(0.778)
(4)
-0.699
(0.776)
0.648
(1.259)
(5)
-1.123
(0.924)
2.087*
(1.152)
(6)
-1.503
(1.135)
1.631
(1.569)
Medical Exp./Growth s,t
0.027
(1.611)
0.863
(2.067)
-1.039
(1.124)
-0.733
(1.443)
-1.533
(1.604)
-3.089*
(1.590)
-1.851
(2.219)
-2.245
(2.110)
-3.300
(3.542)
-4.823**
(2.277)
Unemp. Rate/Change s,t
0.059
(0.068)
-0.093
(0.071)
0.016
(0.052)
0.000
(0.060)
0.010
(0.065)
0.008
(0.069)
0.100
(0.080)
-0.042
(0.091)
0.100
(0.101)
-0.077
(0.129)
State Real GDP/Growth s,t
0.899
(1.774)
-1.494
(1.210)
0.814
(1.185)
-0.241
(0.592)
1.301
(1.814)
0.391
(0.807)
-2.145
(1.858)
-1.183
(1.411)
-1.589
(2.519)
-1.055
(1.419)
No. Filings/Growth s,t
0.073
(0.051)
0.158
(0.083)
0.004
(0.069)
0.035
(0.057)
-0.129
(0.134)
-0.074
(0.089)
0.023
(0.087)
-0.030
(0.083)
-0.073
(0.126)
-0.127
(0.112)
0.209*
(1.547)
-0.123*
(0.374)
0.060
(0.266)
0.924
(0.535)
0.375
(0.211)
1.536
(0.887)
13.996*
(7.586)
Personal Income/Growth st-1 -9.635
(7.373)
2.387
(2.940)
-0.875
(2.035)
2.838
(2.642)
-0.722
(1.740)
2.147
(3.967)
-0.613
(2.809)
7.406**
(3.512)
-0.869
(3.266)
7.292
(4.611)
-0.545
(3.806)
(1)
House Price/Growth s,t
House Price/Growth st-i
-1.563
(2.581)
3.301
(2.676)
-1.237
(4.206)
Medical Exp./Growth st-1 0.670
(4.642)
0.150
(0.177)
Unemp. Rate/Change st-1 0.029
(0.129)
-0.994
(5.869)
State Real GDP/Growth st-1 -2.495
(5.177)
-0.284
(0.190)
No. Filings/Growth st-1 -0.268
(0.159)
Political Climate st-1
Personal Income/Growth st
No. of Obs.
196
196
168
168
168
168
State FE
Y
Y
Y
Year FE
Y
Y
Y
Y
Y
Y
R2
0.27
0.12
0.08
0.21
0.18
0.24
Note. This table shows the estimated coefficient of regression of bankruptcy protection on contemporaneous and lag
values of variables that could determinate the changes in protection levels. House Price s,t is the level or growth of
house prices in state s at time t, from FHFA. Medical expenses is the level of growth in state's annual total medical
expenses from the National Health Statistic. No. of Filings, is the number or change in the number of filings for
non-business bankruptcies in a state. Political Climate s,t is defined as the share of democratic votes in the closer
House of Representative election. State GDP and Personal Income are from BEA, and Unemployment Rate from
BLS. Columns 1 and 2 show the coefficient of regressions of the level protection on level of the explanatory variables
using only year, and year and state fixed effect. Columns 3 and 4 show the coefficient of regressions of the growth
in protection on growth of the explanatory variables using only year, and year and state fixed effect. Columns 5 and
6 show the coefficient of regressions of a dummy that is one if the growth in protection is greater than zero on the
explanatory variables' growth using only year, and year and state fixed effect. The sample period is from 1999 to
2005. *, **, and *** denotes significance at the 10%, 5%, and 1% cluster at the state level respectively.
64
Table 1.18: Dynamics of the Change in Protection. Mortgage Debt
1 Period
2 Periods
No
County
County
Linear Trend Linear Trend Linear Trend
(1)
(2)
(3)
Protection
Growth st-2
No
County
Linear Trend Linear Trend
(4)
(5)
County
Linear Trend
(6)
-0.024
(0.017)
-0.043*
(0.025)
-0.054**
(0.026)
Protection
Growth st-1
0.019
(0.013)
0.013
(0.014)
0.005
(0.012)
0.018
(0.013)
0.002
(0.017)
-0.006
(0.015)
Protection
Growth st
0.007
(0.016)
0.011
(0.014)
0.005
(0.012)
0.006
(0.016)
0.005
(0.014)
-0.002
(0.013)
Protection
Growth st+1
-0.009
(0.008)
-0.006
(0.009)
-0.004
(0.009)
-0.010
(0.008)
-0.012
(0.009)
-0.010
(0.010)
-0.016*
(0.009)
-0.014
(0.011)
-0.011
(0.011)
Protection
Growth s,t+2
Unemployment
Rate Change
-0.003
(0.003)
-0.004
(0.003)
-0.004
(0.003)
-0.003
(0.003)
-0.004
(0.003)
-0.005*
(0.003)
House Price
Index Growth
0.046
(0.078)
0.092
(0.163)
-0.372**
(0.172)
0.049
(0.075)
0.092
(0.163)
-0.385**
(0.173)
Income
Growth
0.190**
0.113
(0.107)
0.039
(0.078)
0.189**
(0.091)
0.114
(0.107)
0.040
(0.078)
(().091)
0.001
0.000
Rate
(0.004)
(0.004)
House Price
0.277
(0.040)
0.281
(0.039)
Unemployment
Income
No. of Obs
No. of Clusters
County FE
Year FE
0.133
0.132
(0.039)
(0.040)
13,302
50
13,302
13,302
13,302
13,302
13,302
50
50
50
50
50
Y
Y
Y
Y
Y
Y
Y
Y
Y
Y
R-Squared
0.09
0.09
0.11
0.09
0.09
0.12
Note. This table shows the estimated coefficient following specification (1) of log changes to mortgage debt on log
changes in bankruptcy protection at the county level. Debt county data is from the FRBNY Consumer Credit
Panel/Equifax. Protection Growth is the log change in the level of protection in state s at time t. Unemployment
rate change is the change in unemployment rate in county i at time t from BLS. House price growth is the log change
in the FHFA state level index for state s at time t, and Income growth is the log change in income in county i at time
t from IRS. Columns 1 and 4, show the without the inclusion of county fixed effects, including one lag and lead, and
two lags and two leads. Columns 2 and 5 show the results with the inclusion of county fixed effect for including one
lag and lead, and two lags and two leads, Columns 3 and 6 are the same than before but including level controls. The
sample period is from 1999 to 2005. *, **, and *** denotes significance at the 10%, 5%, and 1% cluster at the state
level respectively.
65
Table 1.19: Dynamics of the Change in Protection. Auto Debt
2 Periods
1 Period
No
Linear Trend
(1)
County
Linear Trend
(2)
County
Linear Trend
(3)
Protection
Growth s,t-2
No
Linear Trend
(4)
-0.022
(0.019)
County
Linear Trend
(5)
-0.015
(0.026)
County
Linear Trend
(6)
-0.020
(0.028)
Protection
Growth st-1
-0.006
(0.013)
-0.004
(0.017)
-0.004
(0.017)
-0.005
(0.013)
-0.002
(0.017)
-0.003
(0.016)
Protection
Growth st
0.008
(0.014)
0.006
(0.011)
0.007
(0.010)
0.009
(0.014)
0.010
(0.014)
0.010
(0.013)
Protection
Growth st+
-0.012*
(0.007)
-0.011
(0.010)
-0.008
(0.011)
-0.011
(0.007)
-0.007
(0.009)
-0.004
(0.011)
0.015
(0.011)
0.020*
(0.011)
0.022*
(0.012)
Protection
Growth s,t+2
Unemployment
Rate Change
-0.005*
(0.003)
-0.005*
(0.003)
-0.002
(0.003)
-0.005*
(0.003)
-0.005*
(0.003)
-0.002
(0.003)
House Price
Index Growth
0.110**
(0.053)
0.002
(0.113)
-0.097
(0.124)
0.105*
(0.054)
-0.015
(0.114)
-0.127
(0.125)
Income
Growth
0.127***
(0.030)
0.059
(0.038)
0.032
(0.032)
0.128***
(0.030)
0.060
(0.037)
0.031
(0.032)
Unemployment
Rate
-0.011**
(0.005)
-0.012**
(0.005)
House Price
0.009
(0.043)
0.012
(0.042)
Income
0.025
(0.030)
0.026
(0.029)
13,302
13,302
13,302
13,302
13,302
No. of Ohs
13,302
50
50
50
50
No. of Clusters
50
50
Y
Y
Y
Y
County FE
Y
Y
Y
Y
Y
Y
Year FE
0.19
0.17
0.18
0.19
0.17
0.18
R-Squared
Note. This table shows the estimated coefficient following specification (1) of log changes to auto debt on log changes
in bankruptcy protection at the county level. Debt county data is from the FRBNY Consumer Credit Panel/Equifax.
Protection Growth is the log change in the level of protection in state s at time t. Unemployment rate change is the
change in unemployment rate in county i at time t from BLS. House price growth is the log change in the FHFA
state level index for state s at time t, and Income growth is the log change in income in county i at time t from IRS.
Columns 1 and 4, show the without the inclusion of county fixed effects, including one lag and lead, and two lags
and two leads. Columns 2 and 5 show the results with the inclusion of county fixed effect for including one lag and
lead, and two lags and two leads, Columns 3 and 6 are the same than before but including level controls. The sample
period is from 1999 to 2005. *, **, and *** denotes significance at the 10%, 5%, and 1% cluster at the state level
respectively.
66
Table 1.20: Local Business Conditions. Neighboring County-pairs across State Borders. Mortgage Debt
Equal Income
County-Pairs
All
County-Pairs
State
State
County
Linear
Trend
Liner
Trend
0.007
(0.011)
(3)
0.006
(0.010)
(4)
0.006
(0.010)
County
Liner
Trend
Low Income
County-Pairs
State
Linear
Trend
County
Liner
Trend
(5)
(6)
0.051
(0.060)
0.051
(0.058)
Protection
Growth s,t
Linear
Trend
(1)
0.006
(0.011)
Unemployment
Rate Change
-0.002
(0.005)
-0.002
(0.005)
0.001
(0.005)
0.000
(0.005)
-0.001
(0.008)
-0.001
(0.008)
House Price
Index Growth
-0.116
(0.153)
-0.109
(0.150)
-0.050
(0.203)
-0.046
(0.196)
0.077
(0.639)
0.074
(0.617)
Income
Growth
0.089*
(0.054)
0.015
(0.064)
0.197***
(0.074)
0.151*
(0.083)
0.160
(0.115)
0.177
(0.126)
9,168
48
9,168
48
Y
3,984
46
3,984
46
Y
1,188
33
1,188
33
Y
No. of Obs
No. of Clusters
County FE
State FE
County-Pair-Year FE
R-Squared
Y
Y
0.65
(2)
Y
Y
0.62
Y
0.64
Y
0.61
Y
Y
0.55
Y
0.53
Note. This table shows the estimated coefficient following specification (2) of log changes in mortgage debt on log
changes in bankruptcy protection at the county level. Debt county data is from the FRBNY Consumer Credit
Panel/Equifax. Protection Growth is the log change in the level of protection in state s at time t. Unemployment
rate change is the change in unemployment rate in county i at time t from BLS. House price growth is the log change
in the FHFA state level index for state s at time t, and Income growth is the log change in income in county i at time
t from IRS. Columns 1 and 2 show the estimates for state and county fixed effect for all neighboring county-pairs
sample. Columns 3 and 4 show the results including state and county fixed effect for the sub-sample of neighboring
county-pairs for which both counties are in the same income bucket. Columns 5 and 6 show estimates with state and
county fixed effect for only the neighboring county-pairs in the same income bucket and in the lowest tercile of the
income distribution. The sample period is from 1999 to 2005. *, * and *** denotes significance at the 10%, 5%,
and 1% cluster at the state level respectively.
67
Table 1.21: Local Business Conditions. Neighboring County-pairs across State Borders. Auto Debt
All
County-Pairs
Low Income
County-Pairs
County
Liner
Trend
(4)
0.008
(0.013)
State
Linear
Trend
0.006
(0.010)
State
Linear
Trend
(3)
0.008
(0.014)
0.000
(0.004)
0.000
(0.004)
-0.001
(0.005)
-0.079
(0.197)
-0.072
(0.193)
0.143***
(0.049)
0.062
(0.057)
State
Linear
Trend
County
Liner
Trend
(2)
Protection
Growth s,t
(1)
0.006
(0.010)
Unemployment
Rate Change
House Price
Index Growth
Income
Growth
Equal Income
County-Pairs
County
Liner
Trend
(5)
(6)
-0.018
(0.050)
-0.017
(0.048)
-0.001
(0.005)
-0.004
(0.006)
-0.003
(0.006)
-0.275
(0.213)
-0.269
(0.206)
-0.381
(0.406)
-0.379
(0.389)
0.295***
(0.102)
0.239**
(0.118)
0.285*
(0.160)
0.279*
(0.167)
No. of Obs
9,168
9,168
3,984
3,984
1,188
No. of Clusters
48
48
46
46
33
County FE
Y
Y
State FE
Y
Y
Y
County-Pair-Year FE
Y
Y
Y
Y
Y
R-Squared
0.70
0.70
0.67
0.67
0.60
Note. rhis table shows the estimated coefficient following specification (2) of log changes in auto debt
1,188
33
Y
Y
0.60
on log ch anges
in bankruptcy protection at the county level. Debt county data is from the FRBNY Consumer Credit Panel/Equifax.
Protection Growth is the log change in the level of protection in state s at time t. Unemployment rate change is the
change in unemployment rate in county i at time t from BLS. House price growth is the log change in the FHFA state
level index for state s at time t, and Income growth is the log change in income in county i at time t from IRS. Columns
1 and 2, show the estimates for state and county fixed effect for all neighboring county-pairs sample. Columns 3 and
4 show the results including state and county fixed effect for the sub-sample of neighboring county-pairs for which
both counties are in the same income bucket. Columns 5 and 6 show estimates with state and county fixed effect for
only the neighboring county-pairs in the same income bucket and in the lowest tercile of the income distribution. The
sample period is from 1999 to 2005. *, **, and *** denotes significance at the 10%, 5%, and 1% cluster at the state
level respectively.
68
Table 1.22: Heterogeneous Treatment of Bankruptcy Protection: Income and Homeownership. Mortgage Debt
Low Income
Protection Growth st
(1)
0.018
(0.011)
Protection Growth s,t
x Low Income
-0.005
(0.014)
(2)
0.011
(0.013)
Protection Growth st
x Low Home Ownership
Protection Growth s,t
x Med Income
Med Income
(3)
0.012
(0.016)
High Income
Home
Ownership
Home
Ownership
Home
Ownership
Income
(4)
0.006
(0.016)
(5)
0.006
(0.019)
(6)
(7)
0.012
(0.010)
0.019
(0.015)
0.007
(0.024)
0.003
(0.015)
-0.016
(0.014)
-0.010
(0.016)
-0.004
(0.015)
-0.001
(0.015)
-0.013
(0.011)
Protection Growth s,t
x Med Home Ownership
Unenployient
Bate Change
-0.003
(0.003)
-0.003
(0.004)
-0.003
(0.004)
-0.001
(0.003)
-0.001
(0.003)
-0.007
(0.008)
-0.008
(0.008)
House Price
Index Growth
0.078
(0.161)
0.070
(0.141)
0.070
(0.141)
0.137
(0.185)
0.137
(0.185)
0.042
(0.182)
0.041
(0.182)
Income
Growth
0.189**
(0.089)
0.096**
(0.046)
0.096**
(0.045)
0.016
(0.053)
0.012
(0.052)
0.415***
(0.138)
0.403***
(0.143)
4,344
4,344
4,422
4,422
4.536
4,536
13,302
50
50
50
50
50
50
50
Y
Y
Y
Y
Y
Y
Y
0.29
0.31
0.11
0.10
0.08
0.11
0.08
R-Squared
Note. This table shows estimated coefficient a variation of specification (1) that incorporates interactions. Low/Med
Income represents counties in the lowest/middle tercile of the within state income distribution. Low/Med Ownership
represents counties in the lowest/middle tercile of the within income bucket distribution. Column 1 shows the result for
the whole sample when interacted with income heterogeneity. Column 2 shows the result of specification (1) restricted
to the low income counties. Column 3 shows the within low income heterogeneity in homeownership. Columns 4 to
7 replicates columns 2 and 3 for medium and high income levels. The sample period is from 1999 to 2005. *, **, and
*** denotes significance at the 10%, 5%, and 1% cluster at the state level respectively.
No. of Obs
No. of Clusters
State and Year FE
69
Table 1.23: Heterogeneous Treatment of Bankruptcy Protection: Income and Homeownership. Auto Debt
Low Income
Income
(1)
(2)
Protection Growth s,t
0.000
(0.013)
0.032*
(0.019)
Protection Growth st
x Low Income
0.027
(0.017)
Protection Growth s,t
x Low Home Ownership
Protection Growth s,t
x Med Income
Med Income
Home
Ownership
(3)
0.038*
(0.021)
Home
Ownership
4)
-0. 002
(0. 016)
(5)
-0.003
(0.027)
High Income
Home
Ownership
(6)
-0.006
(0.012)
(7)
-0.023
(0.016)
-0.020
(0.030)
0.006
(0.023)
0.021
(0.017)
0.008
(0.017)
-0.004
(0.025)
0.028**
(0.012)
0.001
(0.007)
Protection Growth s,t
x Med Home Ownership
Unemployment
Rate Change
-0.005*
(0.003)
-0.002
(0.004)
-0.002
(0.004)
-0.007**
(0.003)
-0.007**
(0.003)
-0.008
(0.005)
-0.009*
(0.005)
House Price
Index Growth
-0.013
(0.113)
-0.114
(0.146)
-0.112
(0.147)
0.070
(0.116)
0.072
(0.117)
0.020
(0.105)
0.020
(0.105)
0.120***
(0.031)
0.066
(0.057)
0.065
(0.054)
0.056*
(0.033)
0.059*
(0.031)
0.209***
(0.030)
0.196***
(0.030)
Income
Growth
No. of Obs
13,302
4,536
4,536
4,422
4,422
4,344
4,344
No. of Clusters
50
50
50
50
50
50
50
State and Year FE
Y
Y
Y
Y
Y
Y
Y
R-Squared
0.19
0.12
0.13
0.20
0.20
0.34
0.36
Note. This table shows estimated coefficient following a variation of specification (1) that incorporates interactions.
Low/Med Income represents counties in the lowest/middle tercile of the within state income distribution. Low/Med
Ownership represents counties in the lowest/middle tercile of the within income bucket distribution. Column 1 shows
the result for the whole sample when interacted with income heterogeneity. Column 2 shows the result of specification
(1) restricted to the low income counties. Column 3 shows the within low income heterogeneity in homeownership.
Columns 4 to 7 replicates columns 2 and 3 for medium and high income levels. The sample period is from 1999 to
2005. *, **, and *** denotes significance at the 10%, 5%, and 1% cluster at the state level respectively.
70
I,
-1
0.024
(0.034)
-1.910**
(1.448)
-1.579**
(0.695)
13,302
50
Y
0.10
Unemployment
Rate Change
House Price
Index Growth
Income
Growth
N of Obs
N of Clusters
county and year FE
R-Squared
13,302
50
Y
0.17
-1.130
(0.700)
-2.388
(1.185)
-0.018
(0.016)
0.020
(0.058)
2year
13,302
50
Y
0.22
-0.630***
(0.180)
-3.404***
(0.876)
-0.024
(0.018)
0.072
(0.065)
3 years
13,302
50
Y
0.02
-0.581*
(0.335)
-1.245*
(0.576)
0.021
(0.020)
-0.009
(0.116)
13,302
50
Y
0.05
-0.650***
(0.232)
-0.653***
(0.475)
-0.008
(0.014)
-0.018
(0.043)
2year
13,302
50
Y
0.06
-0.401***
(0.115)
0.385***
(0.659)
0.001
(0.009)
-0.002
(0.045)
3 years
Mortgage Debt
1 year
13,302
50
Y
0.03
-0.387
(0.255)
0.181
(0.606)
0.077***
(0.021)
0.001
(0.165)
13,302
50
Y
0.02
-0.067
(0.139)
0.508
(0.455)
0.012
(0.012)
-0.021
(0.077)
2year
Auto Debt
1 year
13,302
50
Y
0.02
-0.132*
(0.076)
0.998*
(0.311)
0.010
(0.009)
-0.037
(0.049)
3 years
Note. This table shows the estimated coefficient following a variation of specification (1) that uses as a dependent variable the change in the fraction of delinquent households in each county, for each type
of credit, for different periods: 1, 2, and 3 year annual changes. The sample period is from 1999 to 2005. *, **, and *** denotes significance at the 10%, 5%, and 1% cluster at the state level respectively.
0.088
(0.204)
Protection
Growth s,t
1 year
Credit Card Debt
Table 1.24: Effect of Bankruptcy Protection on County Delinquency Proportions
Table 1.25: Effect of Bankruptcy Protection on Debt After Bankruptcy Reform 2005
(3)
Cty Linear
Trend
-0.002
(0.005)
(1)
No Linear
Trend
0.007
(0.008)
(2)
Cty Linear
Trend
0.011
(0.013)
(3)
Cty Linear
Trend
-0.007
(0.005)
(1)
-0.003
(0.005)
(2)
0.013
(0.013)
(3)
Auto Debt
(2)
0.017**
(0.008)
Mortgage Debt
(1)
-0.006
(0.006)
Credit Card Debt
-0.002
Cty Linear
Trend
(0.004)
-0.001
(0.001)
(0.036)
0.065*
-0.002
(0.003)
0.160**
(0.079)
0.146***
(0.046)
-0.006**
(0.003)
22,170
50
Y
0.25
(0.092)
0.172*
0.070**
(0.033)
-0.005**
(0.002)
8,868
50
Y
0.40
0.420***
(0.063)
0.166***
(0.041)
-0.007**
(0.003)
8,868
50
Y
0.43
0.323***
(0.053)
0.125
(0.082)
-0.006**
(0.003)
22,170
50
Y
0.42
0.123***
(0.031)
0.161***
(0.033)
-0.007***
(0.002)
No Linear Cty Linear
Trend
Trend
Protection
-0.007***
(0.002)
-0.197***
(0.025)
0.455***
(0.087)
8,868
50
Y
0.38
No Linear Cty Linear
Trend
Trend
Growth s,t
-0.004**
(0.002)
-0.139***
(0.038)
0.057*
(0.033)
8,868
50
Y
0.34
-0.022
(0.014)
Unemployment
Rate Change
-0.254***
(0.034)
-0.174**
(0.076)
22,170
50
Y
0.43
-0.011
(0.014)
House Price
Index Growth
0.054
(0.091)
8,868
50
Y
0.48
-0.021**
(0.009)
Income
Growth
8,868
50
Y
0.43
Protection
Growth s,t x Post
N of Obs
N of Clusters
ety and year FE
R-Squared
Note. This table shows the estimated following specification (1) but extending the sample, for each for each type of credit until 2009. Columns 1, in each type shows the estimates without county fixed
effect. Columns 2, shows the estimates with fixed effect and Columns 3 shows the interaction with a post dummy equal to one for years greater or equal than 2006. The sample period is from 1999 to
2009. *, **, and *** denotes significance at the 10%, 5%, and 1% cluster at the state level respectively.
CA
Chapter 2
House Prices, Collateral and
Self-Employment
2.1
Introduction
The boom-and-bust cycle of house prices over the past decade has featured prominently in explanations of the low unemployment during the surge in house prices and
the high unemployment that followed the real-estate bust. The debate has focused on
two primary explanations for the observed employment dynamics. One view is that
consumers' use of their houses as "ATMs" drove demand and created employment
during the surge in prices, so employment suffered when aggregate demand dropped
because of household deleveraging and falling house prices (see, e.g., Mian and Sufi,
2011a; and Romer, 2011). The other view is that the increase in house prices and the
rise in labor demand in the construction industry masked structural mismatches in
the workforce caused by job losses in the manufacturing sector (see Charles, Hurst,
and Notowidigdo, 2012; and Kocherlakota, 2010).
Our paper documents an alternative channel that has received much less attention but significantly affects the dynamics of employment creation over the business
cycle: the impact of the collateral lending channel, especially mortgage lending, on
employment in small businesses. Seminal papers by Bernanke and Gertler (1989) and
Kiyotaki and Moore (1997) and research since then suggest that improvements in
collateral values ease credit constraints for borrowers and can have multiplier effects
on economic growth. This collateral lending channel builds on the idea that information asymmetries between banks and firms can be alleviated more easily when
collateral values are high, and therefore firms can have higher leverage (Rampini and
Viswanathan, 2010), and that these problems are especially acute for small, more
opaque firms (Gertler and Gilchrist, 1994; Kashyap, Stein, and Wilcox, 1993). Yet
it has been difficult to cleanly identify the causal direction of the collateral effect
empirically. The challenge is that, on the one hand, increased collateral values facilitate lending but that, on the other hand, higher collateral values can be the result of
improvements in economic conditions (e.g., lacoviello, 2005).
This paper is the first to look directly at shocks to home values and consider the
73
impact these shocks have on employment in small firms relative to large firms. To
identify the causal effect of higher house prices we instrument for the growth in prices
between 2002 and 2007 using the elasticity measure developed by Saiz (2010). The
measure uses exogenous geographic and regulatory constraints to housing supply to
differentiate areas where an increase in housing demand translates into higher house
prices and more collateral value (areas where it is hard to build - that is, in which the
elasticity of the housing supply is low) or into higher volume of houses built (areas
with high elasticity). By relying on exogenous restrictions on the expansion of housing
volumes, we can identify the effect of high collateral values on employment in small
businesses. This identification strategy is similar to Chaney, Sraer, and Thesmar
(2012), who look at corporate investment decisions, and Mian and Sufi (2011b), who
examine increases in consumption from household leverage.
We show that during the housing price boom of 2002-2007, areas with rising
house prices (and increased leverage) experienced a significantly bigger increase in
small business starts and a rise in the number of people who were employed in establishments with fewer than ten employees compared to areas that did not see an
increase in house prices. The same increase in employment cannot be found for large
establishments in these same areas. In fact, the effect of home prices on job creation
decreases monotonically with firm size. This asymmetric effect on small versus large
holds only for instrumented house prices, which suggests that the non-instrumented
part of the variation (the one that captures endogenous demand) chiefly impacts
employment at larger firms. This asymmetry points to the interpretation of the collateral lending channel as an important driver of employment creation particularly
for small firms, since large firms have access to other forms of financing and should be
less affected by the collateral channel. To the extent that large firms are also affected
by the increase in real estate values, our estimates may understate the effect of the
collateral channel on total employment.
Although the result above supports the importance of the collateral channel for
small business creation, two alternative hypotheses must be ruled out as explaining our results. First, increases in housing prices can drive local demand for goods
(Campbell and Cocco, 2007) and, consequently, employment at non-tradable industries (Mian and Sufi, 2011a). To the extent that small firms may be more sensitive
to changes in demand (Kashyap and Stein, 1994), the asymmetry in the results could
reflect increased consumer demand rather than use of the collateral lending channel. The second alternative hypothesis results from our use of housing and zoning
restrictions for obtaining identification, because we rely on cross-sectional differences
between high- and low-elasticity areas. These areas could also vary in other characteristics, such as the level of economic vitality. For example, not only could areas
with low housing elasticity see higher home prices when demand for housing picks up
- and therefore increased available collateral - but they could also be the areas where
more investment opportunities become available.
We devise a number of tests to differentiate the impact of the collateral lending
channel from these alternative hypotheses. First, we verify that the results are not
driven by changing industry composition: even within industries, areas with increasing home prices saw stronger employment growth in smaller establishments than areas
74
with stagnant prices. 1
Second, narrowing in on the importance of collateral for business financing, we
look at variation across industries in the amount of start-up capital needed to set up
a new firm. The minimal feasible scale of businesses differs across industries, and the
availability of collateral matters more or less depending on that minimal scale. For
example, some businesses, like home health-care services, can be started with small
amounts of capital that could reasonably be financed through appreciation in home
values. In contrast, many sectors within manufacturing, for example, require large
amounts of capital and fixed investments; the capital needs in these areas are too
high to be financed via individual loans against property. This strategy is similar to
the approach used in Hurst and Lusardi (2004).
Our results follow exactly the predicted pattern: when we repeat our regressions
disaggregated by industries above and below median needs for start-up capital, we
find that the effect of house price increases on the creation of employment in small
establishments is especially strong among industries with lower capital needs. These
results confirm that the collateral lending channel plays an important role in shaping
employment dynamics. Borrowing against housing wealth allows people in areas with
more rapid home price appreciation to start small businesses and drives the increase
in employment at these small firms.
Third, we confirm that our results are not driven by the non-tradable or construction sectors. As noted above, if the relation between increasing housing price and job
creation in small firms were purely constrained to the non-tradable or construction
sectors, one would be concerned that the results are driven not by changes in the collateral lending channel but by differences in local demand. However, our results are
almost unchanged when we eliminate these sectors from the analysis, and they also
hold for the manufacturing sector where products are easily tradable. The difference
in employment creation between large and small firms is also particularly strong for
industries in which firms report shipping goods across long distances. Our results
are thus distinguished from the work of Mian and Sufi (2011a), which shows that
areas where house prices increased most also exhibited an increase in unemployment
in non-tradable industries due to deleveraging and lower demand in the aftermath of
2008. Any change in output in the low-elasticity areas must therefore be driven by
changes on the input (production) side. This is the collateral lending channel.
Last, we rule out that our results are driven by generally loosening credit standards
in areas with rapid house price growth. The growth of small businesses could be
caused not by better access to collateral but rather by easier access to other forms of
credit because of banks' improved balance sheet position. We show that this is not
the case. If anything, banks became increasingly more selective in credit approval in
low-elasticity areas leading up to 2007.
Using a calculation similar to that used in Mian and Sufi (2011a), we compute
the approximate contribution of the collateral lending channel to changes in overall
employment in the pre-crisis period, 2002-2007. Using this approach, we find that
'A similar relationship exists when we include proprietorships and unincorporated businesses in
the regressions.
75
the collateral channel accounts for 10-25% of the increase in employment in these
years (depending on the specific assumptions about the reference group that best
isolates the collateral effect), whereas the demand channel explains about 40% over
the same period and the two effects are mutually non-overlapping. Interestingly,
although the point estimate for the effect of the demand channel is large, the effect
is noisily estimated for 2002-2007, so we cannot reject that there is no effect on
employment of increased demand driven by higher house prices before the crisis. This
is in stark contrast to the post-crisis period (2007-2009), when the drop in demand
of over-leveraged areas shows up very strongly in the data (as documented in Mian
and Sufi, 2011a). It is important to point out that these numbers provide rough
approximations of the relative magnitudes of these two channels, but they ignore any
general equilibrium effects in aggregation.
When we consider the period after the financial crisis when house prices started
to decline (2007-2009), we find that small firms experienced weaker employment
declines than large firms in areas where the increase in house prices was stronger in
the period before the crisis. This suggests that small firms that were created in lowelasticity areas during the time of increasing collateral values were more resilient than
larger ones in those areas and did not immediately disappear when the crisis struck.
This shows an interesting asymmetry in the mechanism behind the collateral lending
channel - although it is a powerful channel in facilitating the creation of new small
establishments, a contraction in the amount of available collateral does not lead to a
disproportionate amount of destruction of employment in those small establishments.
We are, however, cautious in interpreting our results for the post-2007 period. First,
given the nature of our data, we cannot disentangle whether the relative persistence of
jobs in small businesses is due to the survival of existing small businesses or a change
in the entrance of newly started firms. Second, although the elasticity measure has a
natural interpretation for positive housing demand shocks, we lack a good instrument
for the house price drop. In fact, an increase in housing demand can translate into
either higher house prices (inelastic areas) or an expansion of housing volume (elastic
areas). However, on the downside, a drop in housing demand does not lead to the
destruction of housing stock, and thus prices simply drop in both inelastic and elastic
areas. So, instead of instrumenting for the price drop in the crisis period, we instead
compare areas with large appreciation in the pre-crisis period (low elasticity) with
those that had smaller house price increases - that is, the timing of the housing
price changes remains 2002-2007, as in the rest of the analysis. Once the crisis hit,
areas that experienced larger house price increases in the pre-crisis period were more
leveraged (Mian and Sufi, 2011a, 2011b), so it should be harder for households to
access collateral in these areas in the crisis.
Our study builds on literature that shows that credit constraints at the household level matter for the creation of new businesses (Evans and Jovanovic, 1989;
Holtz-Eakin, Joulfaian, and Rosen, 1994; Gentry and Hubbard, 2004; Cagetti and
De Nardi, 2006), although some authors have argued that this relation is present
only at the very top of wealth distribution (Hurst and Lusardi, 2004). At the same
time, housing wealth in particular has been shown to be an important factor in the
funding of business start-ups (see Fan and White, 2003; Fairlie and Krashinsky, 2012;
76
Fort, Haltiwanger, Jarmin, and Miranda, 2012; Kleiner, 2013; Corradin and Popov,
2013; and Schmalz, Sraer, and Thesmar, 2013, for France; and Black, De Meza, and
Jeffreys, 1996; and Kleiner, 2013, for the United Kingdom). Previous work has also
found that bank credit is an important source of financing for small businesses (Petersen and Rajan, 1994; Robb and Robinson, 2012; Fracassi, Garmaise, Kogan, and
Natividad, 2013) and that entrepreneurs often have to provide personal guarantees
when they obtain financing (Berger and Udell, 1998). More recently, Greenstone and
Mas (2012) use the sharp reduction in credit supply following the 2008 crisis, and the
heterogeneity of this effect among banks, to show that a decrease in the origination of
small business loans leads to a decrease in county employment and business formation
during the period 2007-2009.
The rest of the paper proceeds as follows: Section 2 describes our data and the
empirical methodology. Section 3 discusses the results, and Section 4 concludes.
2.2
2.2.1
Data and Empirical Methodology
Data Description
We obtain employment growth from the County Business Patterns (CBP) data set
published by the U.S. Census Bureau. The CBP data contain employment data by
county, industry, and establishment size (measured in number of employees) between
1998 and 2010 as of March of the reported year. We use the data at the four-digit
National American Industry Classification System (NAICS) level, broken down by
county and establishment size, to construct our main dependent variable of interest:
the employment growth by establishment size between 2002 and 2007. The breakdown
of establishments by employee number allows us to differentially estimate the effect
of housing price growth in the net creation of establishments of different sizes. 2
We use five establishment categories in our regressions that the Census Bureau
commonly uses: establishments of one to four employees, five to nine, ten to 19, 20 to
49, and 50 or more. The CPB provides all but the final category. For establishments
with 50 or more employees, the CBP has multiple categories, but if we were to use each
one individually, it would add noise to our estimation because such large businesses
become rare at the county level and even scarcer at the county and industry levels,
which we need for some of the specifications discussed below. In order to create
the category of establishments with more than 50 employees, we take the number
of establishments in each category above 50 and multiply those by the midpoint of
the category (for example, for the category of 100 to 249 employees, we multiply the
number of establishments by 174.5), and then we add them all up at the country and
industry levels.
2
The data include only the number of establishments in each county, industry, and year by
category of employment size (1-4 employees, 5-9, 10-19, etc.), not the total employment for each
establishment category. As such to construct the employment in each bin we multiply the number of
establishments by the middle point of each category. For example, to calculate the total employment
of 1-4-employee establishments in a given industry, county, and year, we multiply the number of
establishments by 2.5.
77
The housing prices used in the regressions come from the Federal Housing Finance
Agency (FHFA) House Price Index (HPI) data at the Metropolitan Statistical Area
(MSA) level. The FHFA HPI is a weighted, repeat-sales index, and it measures
average price changes in repeat sales or refinancings on the same properties. We
obtain this information by reviewing repeat mortgage transactions on single-family
properties whose mortgages have been purchased or securitized by Fannie Mae or
Freddie Mac since January 1975. We use data on the MSA-level index between 2002
and 2007.
The use of MSA-level house prices is consistent with our identification strategy.
To identify the causal effect of house prices on small business creation, we instrument house price growth between 2002 and 2007 with the measure of housing supply
elasticity of Saiz (2010), which varies at the MSA level. The measure of the supply
elasticity is constructed using geographical and local regulatory constraints to new
construction. Areas where it is difficult to add new housing (due to geographic or
regulatory restrictions) are classified as low elasticity and vice versa for areas where
land is easily available. Low-elasticity areas correlate strongly with steeper house
price growth in the years 2002-2007. This measure is available for 269 MSAs that we
match to 776 counties using the correspondence between MSAs and counties for the
year 1999 as provided by the Census Bureau.3 Although employment growth and our
other controls are available for a much larger sample of counties, our regressions focus
on the subset of counties for which we have the housing supply elasticity measure.
An important measure for our analysis is the amount of capital needed to start
a firm, since these investment requirements might affect how much a given industry
depends on the housing collateral channel. To construct this variable we use the
Survey of Business Owners (SBO) Public Use Microdata Sample (PUMS). The SBO
PUMS was created using responses from the 2007 SBO and provides access to survey
data at a more detailed level than that of previously published SBO results. The SBO
PUMS is designed to study entrepreneurial activity by surveying a random sample of
businesses selected from a list of all firms operating during 2007 with receipts of $1,000
or more provided by the IRS. The survey provides such business characteristics as
firm size, employer-paid benefits, minority- and women-ownership, access to capital,
and firm age. We focus here on the "Amount of start-up or acquisition capital"
for each firm, and we group the answers to this question at the two-digit NAICS
industry level (the finest level available in the data) for firms established in 2007.
The classification is virtually identical if we use all years in the data or if we focus on
firms with one tO four employees only. The median amount of capital needed to start
a business in the data is $215 thousand. We follow Hurst and Lusardi (2004) and
split industries above and below the median to measure the differential effect of the
collateral channel on business creation for industries in the two groups. The average
amount of capital needed by firms below the median is $132 thousand, whereas the
average amount needed for industries above the median is $260 thousand (detailed
amounts by two-digit NAICS sector are in Appendix Table2.14).
3
This correspondence is available at and for the New England Metropolitan Component Areas
used by Saiz (2010).
78
Our classification of "non-tradable," "tradable," and "construction" industries at
the four-digict NAICS level is obtained from Appendix Table 2 of Mian and Sufi
(2011a). 4 Non-tradable codes are included largely in the 44 and 45 sectors (Retail
Trade), as well as under 72 (Accommodation and Food Services). Construction industries include most codes under the Construction two-digit NAICS sector (23), as
well as some subsectors in manufacturing, retail trade, and services that are directly
connected to construction (e.g., 3273 - Cement and Concrete Products Manufacturing). Manufacturing industries include all 31-33 subsectors (Manufacturing), and in
some specifications we restrict the sample to manufacturing industries that are also
classified as "tradable" in Mian and Sufi (2011a) (i.e., those not in construction or in
"other industries").
To address further the concern that the results might be driven by local demand,
we construct a measure of the average distance that firms in an industry ship their
goods similar to that used in Duranton, Morrow, and Turner (2013). These data are
available from the 2007 Census Commodity Flow Survey, which reports the distance
traveled by shipments of a sample of establishments in each three-digit NAICS manufacturing industry. 5 The unit of observation in the census data is at the state and
industry levels, so we construct a dollar-weighted average distance of shipments also
for each state and industry individually. Summary statistics of the average distance
shipped, as well as how often each industry appears in each decile, are shown in
Appendix Table 2.13.
We also use data on county-level births and deaths of establishments for each
two-digit NAICS industry between 2002 and 2010 from the Census Statistics of U.S.
Businesses (SUSB). Data on births and deaths of establishments is provided under
the "Employment Change" section of SUSB, and it does not include a breakdown
by establishment size at the county and industry levels, so we cannot use it as our
main dataset. However, given that most establishment births are of a very small scale
(Haltiwanger, Jarmin, and Miranda, 2011), we view the regressions performed on this
data set as an important test of the mechanism in our main results. We compute
the cumulative number of births and deaths between 2002 and 2007 for each county
and industry as our dependent variable of interest and scale this number by the total
number of establishments as of 2002 in the same county-industry cell.
The net creation of sole proprietorships at a county level is obtained from two
sources. We use both the yearly local area personal income and employment data
from the Bureau of Economic Analysis (BEA and the census nonemployer statistics.
From the BEA we use Non-Farm Proprietorship employment at the county level
between 2002 and 2007 to estimate the growth of sole proprietorships in this period.
From the census we obtain the number of establishments for the period 2002-2007 at
the two-digit NAICS level. We use both sources of data in the regressions to ensure
the robustness of our results.
Unemployment and unemployment rate at the county level are obtained using
4
The
current
version
of
the
online
appendix
can
be
found
here:
http://faculty.chicagobooth.edu/amir.sufi/data-and-appendices/
'The year 2007 is the first year in which the data is reported at the three-digit NAICS level
(previous years included only commodity identifiers rather than industry data).
79
the Bureau of Labor Statistics Local Area estimates. Local Area Unemployment
Statistics (LAUS) are available for approximately 7,300 areas that range from census
regions and divisions to counties and county equivalents, and these data are available
between 1976 and 2012. We match the county equivalent data to the CBP data using
Federal Information Processing Standard (FIPS) county unique identifiers.
The migrations data are extracted from the IRS county-to-county migration data
series. The migration estimates are based on year-to-year address changes reported
on individual income tax returns filed with the IRS. The data set presents migration
patterns by county for the entire United States and is split by inflows - the number of
new residents who moved to a county and where they migrated from - and outflows the number of residents leaving a county and where they went.' We also compute net
flows as inflows minus outflows, and we scale all figures by the number of nonmovers
in the county. The data are available from 1991 through 2009 filling years.
To better identify the effect of house prices on self-employment, we include a set
of controls that capture some of the cross-sectional differences across counties. We
use county-level information from the Census Bureau Summary Files for 2000 on: the
number of households in a county; the natural logarithm of county-level population;
the percentage of college-educated individuals, defined as the number of people over
25 with a bachelor degree or higher as a proportion of the total population over 25
years old; the percentage of employed people, defined as the employed population over
the total population 16 years old or older; the share of the population in the workforce,
defined as the total population in the civilian labor force over 16 years old divided by
the total population 16 years old or older; the percentage of owner-occupied houses;
and a measure of exposure of each county to imports from China, 7 and, therefore,
better control for changes in investment opportunities in those counties.
2.2.2
Summary Statistics
Panel A of Table 2.1 provides descriptive statistics for our data set: the first row shows
total employment in 2002 for all counties in our sample, as well as the employment
growth between 2002 and 2007 estimated from the CBP data. Our data include a
total of 775 counties with nonmissing total employment data. We split the sample
into counties above and below the median of the housing supply elasticity measure
and show t-statistics (with standard errors clustered by MSA) for the difference in
means between the two groups. We see that counties with low supply elasticity
are larger but have similar unemployment rates in 2002 as those with high supply
elasticity. The characteristics in 2002 from the census are broadly similar for the
6
The data used to produce migration data products come from individual income tax returns
filed before late September of each calendar year and represent between 95% and 98% of total annual
filings.
7
We construct the measure of competition from imports from China by multiplying the fraction
of employment in each county and in each industry by the share of imported goods from China as a
fraction of total domestic shipments in the industry in the United States. The variation is virtually
the same if we instead use the growth in the weight of imports for each industry as a fraction of
U.S. domestic shipments between 1998 and 2005. The import data at the industry level is obtained
from Peter K. Schott's website: http://faculty.som.yale.edu/peterschott/subinternational.htm.
80
two groups, with the one exception being the percentage of college-educated people
(somewhat higher in low-elasticity areas). Average household income is also higher in
those counties, but the difference is economically small (about 10% of the mean). As
expected, counties with a low elasticity of housing supply experienced much stronger
growth in house prices than did counties with a high elasticity of supply (a "crude"
version of the first stage in our regressions) and similarly experienced a much larger
increase in average debt-to-income ratio (consistent with Mian and Sufi, 2011a).
Panel B of Table 2.1 shows how employment is distributed across the different
employment-size categories. The biggest firm category, 50 employees or more, accounts for 51.7% of employment in 2002, whereas the smallest category, 1-4 employees, accounts for 8.9%. Growth in employment is stronger among larger companies
in the 2002-2007 period, especially among the industries that we classify as having
low start-up capital needs.
2.2.3
Empirical Model
We test whether increases in real estate prices affect the growth in employment by
facilitating the creation of small businesses (collateral channel). To differentiate the
collateral channel from a pure (expansionary) demand shock, we look at the differential effect of home prices on the net creation of establishments in different size
categories.8 Our identification relies on the idea that improved availability of collateral in the form of higher house prices can positively affect the creation of small
businesses, whereas it is likely to have no effect on the creation of larger establishments since these firms cannot be started with capital that can be extracted from a
house.
We measure the availability of collateral to small business entrepreneurs by the
growth in house prices in the area where the establishment is located. However, it is
challenging to establish a causal link from the availability of collateral to the creation
of small businesses, since there are many omitted variables that could simultaneously
affect both the value of real estate collateral and the demand faced by small businesses,
including changes in household income in the area and improvements in investment
opportunities. To overcome this difficulty, we instrument for the changes in house
prices during our period of interest (2002-2007) using the elasticity of housing supply
by MSA (see Saiz, 2010). Our identification relies on the assumption that the elasticity of the housing supply only impacts employment creation at establishments of
different sizes through its effect on house prices. The exclusion restriction is violated
if housing supply elasticity is correlated with employment or business creation for reasons other than house price growth. Similar approaches have been used extensively
in the recent literature (see, e.g., Mian and Sufi, 2011a, 2011b; Charles, Hurst, and
Notowidigdo, 2012; and Robb and Robinson, 2012). Davidoff (2012) argues that the
8
As we discuss in the data section, our data do not include changes in employment within
establishments (i.e., along the intensive margin), so our measure of changes in employment relies on
multiplying the number of establishments in each size category by the midpoint of the number of
employees in each bin. It is thus equivalent to interpret our results in terms of number of employees
or number of establishments.
81
supply elasticity measure does not capture the severity of the boom-and-bust bust
cycle of the 2000s. In our setting we are concerned only with price increases between
2002 and 2007, and the supply elasticity measure developed by Saiz is a strong predictor of the increase in prices (i.e., there is no weak instruments problem). As we
describe below, we also include specifications that include county fixed effects that
should further mitigate concerns about the cross-sectional elasticity measure.
We rely on two basic regression specifications for our analysis. The first specification aggregates data up to the level at which our instrument varies - that is, at
the county-year establishment-size level. Each individual observation is the change
between 2002 and 2007 of employees in a given county, year, and establishment size.
We thus add up the number of employees in all industries in each establishment category and take the growth in total number of employees as the dependent variable.
We then run two-stage least squares regressions of the type:
A 0 2-0 7 Employmentij = o + 0 1 ,AHpF2-O7 + 0 2 1i +
3 1jAHP0
2-
07
+ -lXj + 6Ej
We index counties by j and establishment size categories by i. A 02- 0 7 Employmentij
is the change in employment for establishment size category i in county j between
2002 and 2007. Similarly, AHP02- 0 7 is the growth in housing prices at the county
level for the same time period where, as we discuss above, we instrument for the
growth in house prices using the housing supply elasticity of Saiz (2010). 1i is a set of
dummy variables for each of the four included establishment categories (we omit the
largest category of more than 50 employees). We then also include the product of the
establishment size dummies and the growth in house prices, and 33 is the coefficient of
interest in our regressions. In particular, the test we are interested in is whether the
coefficient for the smallest establishments is larger (and positive) than those of the
larger categories, which would confirm that house prices had a stronger impact on the
creation of small establishments. Xj is a set of county-level controls that include the
size of the county, the percentage of the population with a bachelor's degree or higher,
the percentage of the population that is employed, the percentage of the population
in the labor force, the percentage of owner-occupied houses, and the county share
of China imports. Standard errors in this specification are heteroskedasticity robust
and clustered at the MSA level (given that the variation in the instrument we use is
at this level as well), and all regressions are weighted by the number of households in
a county as of 2000, as in Mian and Sufi (2011a).
The second specification disaggregates observations to the county, year, establishment size, and four-digit NAICS level, yielding a much larger number of observations
than the specification above (since each county now appears multiple times for each
industry). When using these disaggregated data we can include industry fixed effects
in the regression, which allows us to control even further for common shocks (namely,
nationwide demand shocks) to each four-digit industry. The coefficients in this case
represent the differential impact that house prices have on establishments of different
sizes within each industry. The specification becomes:
82
A 02-
7
Employentijz =
a+
O1AHP
2 07
-
+ /21
+ 031iAHPj2- 0 7 + 'XY
+
Z+
ij
in which z indexes the industries and lz is a set of indicator variables for each
industry.
The breakdown at the industry level allows us to address an important alternative
hypothesis to the mechanism we identify- namely, that higher home prices caused
increased demand, which then prompted the growth in new businesses. This type of
demand story (as opposed to the collateral lending channel) comes in two versions.
The first is that rising house prices lead to an increase in demand because households
feel richer or have access to home equity. This channel is proposed in Mian and Sufi
(2011a) to explain the drop in employment during the Great Recession of 2007-2009.
A second version of the demand hypothesis is that increasing house prices may benefit
certain industries more than others and that these industries happen to be composed
of smaller establishments on average (i.e., a "composition" effect).
We address these alternative demand hypotheses in several ways. First, by holding constant industry fixed effects we identify how employment in the smallest establishments reacts differently from that of large establishments within each four-digit
NAICS industry. This addresses the composition effect described above. Second, as
we have argued before, a pure local demand story should affect establishments of all
sizes similarly, whereas the credit collateral channel is relevant mainly for small businesses. There is, however, still the possibility that smaller firms are more sensitive to
local demand shocks than large firms. To see if this effect could explain our results,
we exclude the most obvious candidate industries that might directly benefit from
local demand shocks due to higher house prices- namely, those linked to construction
and firms in the non-tradable sector as classified in Mian and Sufi (2011a), and we
repeat our tests only for manufacturing firms, those that should be least affected by
local demand shocks.
As a robustness check to our results we also implement the approach in Chaney,
Sraer, and Thesmar (2012) by constructing the product of the nationwide conventional mortgage rate (obtained from the Federal Reserve data website) with the local
elasticity of housing supply measure. This provides time-varying shocks to the demand for housing - when mortgage rates drop more, the shock to demand for housing
should be larger, consistent with Adelino, Schoar, and Severino (2012). This shock
then translates into higher prices in areas with a low elasticity of housing supply than
in places where it is easy to build. This specification uses a panel of yearly observations at the county level and includes county fixed effects, unlike the previous two
specifications. As before, we run two-stage least squares regressions of the form:
AEmploymentijt = a + 01 IAHPJt + 02 lit + ( 3 1stAHPt + h'11j +
1
Y2
t
+ Eij
The instrument for house prices is the product of mortgage rates and housing elasticity, not just the elasticity measure as before. We include county fixed effects (I),
83
which absorbs all county-level controls included in the previous two specifications, as
well as year fixed effects.'
2.3
Empirical Results
2.3.1
House Prices and Employment at Small Establishments
Our central hypothesis is that the availability of more valuable collateral (in our case
through increased real estate prices) in the period before the financial crisis has an
effect on the creation of small firms or on self-employment, since it provides individuals
with easier access to start-up capital. As a result, we should see a sharper increase
in self-employment and employment in small businesses in areas that had steeper
housing price appreciation. We also expect this effect to be concentrated in firms in
the smaller size categories, since large firms cannot finance themselves using home
equity. This hypothesis is tested in Table 2.2, where we run two-stage least squares
regressions of the growth in employment between 2002 and 2007 on five establishment
size categories and their interaction with house price growth in the same period. The
instrument for house price growth, as we discuss above, is the Saiz (2010) measure
of housing supply elasticity. In the first column of Table 2.2 we show the first-stage
regression of house price growth on the Saiz measure of housing supply elasticity to
confirm the validity of the instrument. The coefficient of -0.09 means that a one
standard deviation increase in elasticity of housing supply is associated with an 11.7
percentage point lower growth in prices (for an average house price growth of 33.9%).
The F statistic on this regression is 14.5 (above the conventional threshold of 10 for
evaluating weak instruments). This reflects that MSAs with a higher elasticity of
supply experienced significantly lower housing price growth between 2002 and 2007,
in line with previous literature. In Column 2 we run a regression of employment
change between 2002 and 2007 on the change in house prices during the same period.
In this regression we do not instrument the change in house prices in order to show
the raw correlation between house prices and employment. The effect is positive and
economically large. A one standard deviation increase in house prices is associated
with an increase in total employment of 3.95% over this period, for an average growth
in employment of 10.6%. In the simple weighted least squares regression we see no
distinction between the effect of home prices on small and large establishments. This
result highlights the need for an instrument for our dependent variable of interest,
given the numerous factors that are likely to drive both employment creation and
house prices (income growth, investment opportunities, etc.).
In Column 3 of Table 2.2 we repeat the same regression but instrument the change
in house prices with the Saiz measure for the elasticity of housing supply. We see
that there is a positive but not significant causal relation between county-level employment change and house price growth on average, in contrast to the results in the
9
We do not rely on the panel specifications for most tests because mortgage rates did not
experience large drops in the period we analyze. We effectively have one large shock to demand for
housing in the period 2002-2007, and the first two specifications capture this fact more clearly.
84
previous column. However, when we look at the differential effect of instrumented
housing price changes, the increase in home prices has a significant and large positive
effect on the small establishments but no significant effect on employment growth for
big establishments (50 or more employees). The coefficient on the interaction term
between house price growth and the one-to-four-employee size category shows that a
1% increase in house prices translates into a 0.19% increase in employment at these
establishments relative to the largest ones. This translates into an increase in employment of 5.3% for a one standard deviation change in house prices, for an average
change in employment at the smallest establishments of 9.4% (the effects of a one
standard deviation change in house prices for each size category are shown in Appendix Table 2.12). Furthermore, the effect of collateral is decreasing monotonically
with firm size. For firms with more than ten employees, the effect is indistinguishable
from that of the very largest firms. This is consistent with the collateral channel of
house price appreciation being an important mechanism for small firm creation, since
the amount of collateral that is provided by real estate appreciation is not enough to
start a larger firm. Also, these results suggest that the causal impact of house prices
on employment growth in 2002-2007 did not work through increased demand, since
in that case firms of all sizes (including the very large) should have been affected.
One concern with the above specification could be that the change in house prices
in areas with low Saiz housing elasticity induces a local demand shock that especially
affects certain industries. If those industries are also, on average, disproportionately
made up of smaller establishments, the result above might reflect a composition effect
rather than the collateral channel, as we suggest. Although a number of factors would
need to line up in a very specific way, we cannot rule this concern out on face value with
the specifications in Table 2.2. In order to eliminate the alternative hypothesis about
industry composition, we use our more disaggregated data, which provides data at the
county, four-digit NAICS, and establishment size level. This allows us to hold industry
fixed effects constant and test whether, conditional on an industry, the growth of small
establishments is significantly stronger than that of large establishments in counties
with greater increases in home prices. Intuitively, this specification asks whether
within an industry the fraction of employment generated by small firms grows more
quickly than that of large firms. This way we can confirm that the results are not a
consequence of changing industry composition. The results for this specification are
shown in Column 4 of Table 2.2. As before, we find that impact of house price changes
(instrumented with the Saiz measure) is stronger for establishments with one to four
employees when compared to the bigger firm categories. We again find that the effect
is monotonically decreasing and not statistically significant beyond firms with ten or
more employees. To be more specific about which industries show the strongest effects
from the collateral channel, in Table 2.17 we show the three-digit NAICS industries
that are not construction, manufacturing, non-tradable, and finance, insurance, and
real estate, as well as the employment share in each size bin. The sample includes a
variety of services and wholesale activities, with significant cross-sectional variation
in the proportion of employees in the very small establishment size categories (from
26.3% of employment in one-to-four-employee establishments in the case of "NAICS
425 - Wholesale Electronic Markets and Agents and Brokers" to 0% in this category
85
for "NAICS 622 - Hospitals").
The third version of the instrumented regression is shown in Column 5 of Table 2.2,
in which we use yearly observations on county-level employment and construct a timevarying instrument by taking the product of the average conventional mortgage rate in
the United States and the Saiz elasticity measure. We then add county and year fixed
effects to the regressions and run the specification described in Section 2.3, above.
The results are very consistent with the two previous specifications, with the same
monotonically decreasing effect of house prices on employment at establishments of
increasing size. We run the robustness specifications with the time-varying instrument
and county fixed effects to account for time-invariant differences across regions that
could be correlated with elasticity and new business starts. The fact that the results
are consistent with our main specification alleviates these concerns.
To confirm that the effect we estimate runs through the collateral channel, we
test whether our estimated effect is stronger in industries that have lower start-up
capital needs. We expect this to be the case, given that the median amount of home
debt at its peak in 2006 for all U.S. households was approximately $117 thousand
(Mian and Sufi, 2011b) and that only a fraction of this amount would be available
for use in starting a business. Also, Adelino, Schoar, and Severino (2012) show that
the average value of a single family home during this period is approximately $309
thousand and that most families obtain an 80% loan-to-value (LTV) loan. Even
accounting for the fact that most entrepreneurs are over age 35 and that almost half
are over 45 (Robb and Robinson, 2012), and so we expect them to have built home
equity relative to the initial 80% LTV, it is not plausible to finance a large amount
of capital using home equity as collateral. Brown, Stein, and Zafar (2013) show that
the average amount of home equity lines of credit (HELOC) in the boom period is
$2,623, with a standard deviation of $13,672. This implies that even homeowners
who are two standard deviations above the mean have less than $30 thousand in
home equity loans. The paper also shows that the fourth quartile of homeowners
in high house-price appreciation areas has about $8,500 in HELOC. These numbers
suggest the range of funds that can be obtained from homes as collateral for starting
a business.
We split our sample of industries at the median amount of capital needed to start
a firm to explore this source of variation. As we describe in Section 2, above, we
obtain this information from the SBO PUMS by selecting the sample of new firms in
each industry and averaging the amount of capital needed to start those firms.
We show the results split by the amount of start-up capital needed in each industry
in Columns 6-11 of Table 2.2. The results show that the effect of collateral on
employment growth in small establishments is stronger for industries in which the
amount of capital needed to start a firm is lower (the average amount of start-up
capital for industries below the median is approximately $132 thousand). In fact, for
this subset of industries the effect is statistically significantly different from that of
the largest group even for establishments with up to 49 employees- that is, the causal
effect of house prices extends to establishments other than the very smallest. When
we include industry fixed effects, only the coefficient on the smallest establishments is
statistically different from zero. For the group of industries that require more start-up
86
capital, the effect of house prices on employment is smaller and statistically significant
only for the very smallest group both with and without fixed effects. These results
confirm that job creation at small businesses in response to house prices changes is
strongest in industries with low start-up capital needs that can reasonably be financed
through loans on home equity. Notice that the assumption underlying these tests is
that the contribution of housing as collateral is more likely to matter at the margin for
firms that require low amounts of capital than for firms that require a lot of capital. In
fact, for firms that require large amounts of capital, we expect entrepreneurs to seek
out additional sources of capital, and housing collateral is unlikely to be as important
for the decision to start a firm.
Effect After Removing Non-tradable Industries
In this subsection we document that our results are not driven by certain industries,
in particular construction and non-tradable firms. One possible concern is that the
increase in house prices led to a growth in demand for construction services or for local
services (e.g., local retail or restaurants), resulting in an increase in new firms in these
industries (e.g., more remodeling and new housing construction, more dry cleaners).
This would be a consequence of increased demand rather than an effect through the
collateral channel. We rerun our main specifications excluding all industries linked
to the construction and non-tradable sectors as classified by Mian and Sufi (2011a),
as well as Finance, Insurance, and Real Estate firms (NAICS 52 and 53). We report
these results in Table 2.3.
The first takeaway from Table 2.3 is that the direction and magnitude of the
effects are virtually unchanged when we remove these sectors from the regressions.
If the effect we measure were driven largely by a local demand shock (instead of
the collateral channel), we would expect the coefficient to be significantly affected
when we remove from the sample the sectors that are most sensitive to local demand
(Columns 1-3 of Table 2.3).
In the last two columns of Table 2.3 we limit the regressions to the manufacturing
sector. These industries are the least likely to be affected by local demand. At the
same time, however, they typically require significant start-up capital, which makes
it harder to find the effect of the collateral channel using our experiment. Still, we
find that small firms created more employment relative to large firms in period 20022007 in areas where housing prices rose more (Columns 4 and 5 of Table 2.3). The
effect is similar in magnitude for establishments of one to four employees, five to
nine, and ten to 19, but it is statistically significant only at conventional levels for the
smallest size category. We know that, on average, firms in the manufacturing sector
lost jobs during this period, and the coefficient on the largest firms suggests that they
lost more jobs in places where house prices rose more (coefficient is -0.16). When we
combine this effect with the coefficient on the small firms, this implies that access
to collateral allowed the smallest firms to preserve employment, whereas the largest
firms were losing jobs during this period. This confirms that a simple demand-side
story is not driving our results and confirms the importance of the collateral channel
for the creation of smaller establishments in the period 2002-2007.
87
In Table 2.4 we perform an additional test for manufacturing industries. In this
test, we split industries based on the average distance of shipments in each three-digit
NAICS industry and state. This addresses further the concern that local demand
shocks might be driving the results for manufacturing firms. Table 4 we show that
the result for manufacturing shown in Table 2.3 is driven by firms in industries and
states that ship goods across large distances. The median reported distance in the
sample is 600 miles, so firms that report shipping goods over more than 600 miles
are unlikely to make decisions as a function of local demand shocks (details on the
distances shipped by firms in each industry and state are in Appendix Table ??).
One possible concern with the test using distances is that small firms in a given
sector could be very different from large firms, so the small firms in those industries
could depend more on local demand. Although we do not have shipment data by
firm, in Table A7 we consider the relation between the reported distance shipped
in a given state and industry cell and the share of small businesses in that cell.
We use the same distance measure from before and separately compute the share
of employment in establishments that have 50 or more employees for each state and
three-digit NAICS manufacturing industry. Then, for each industry, we compute
the average (over all states) of the distance shipped, as well as the average share
of employees in firms that have 50 or more employees. Finally, for each state and
industry observation, we compute the deviation from the industry mean for both
measures and classify observations into deciles based on these deviations. 0 The
takeaway from this table is that there is no visible relation between the distance
shipped and the share of employees at large firms versus small firms. In particular,
there is a lot of heterogeneity across industries in the fraction of small firms and the
distance shipped. This should mitigate the concern that a strong positive relation
between firm size and distance shipped might explain the results in the last two
columns of Table 2.4.
Our measure of growth of establishments by size category does not allow us to
observe the creation and destruction of establishments directly, so in a separate set
of regressions shown in Table A8 we use the Statistics of U.S. Businesses from the
census to look at births and deaths of establishments at the two-digit NAICS industry
level. The disadvantage of this data set is that it does not include the breakdown
of establishments by employment size. Given that an overwhelming percentage of
new businesses are very small (Haltiwanger, Jarmin, and Miranda, 2011; Robb and
Robinson, 2012), this robustness test directly speaks to the validity of our main
results. We find that births of establishments are very strongly affected by increasing
house prices instrumented with the elasticity of housing supply and that the result
holds when we consider the net creation of establishments (i.e., births minus deaths),
and the coefficient is unchanged when we include two-digit NAICS fixed effects (the
finest industry category available in this data set at a county level).
10 So, state-industry
observations that are in the first decile of the distance are those that ship
goods at short distances relative to the industry average. Similarly, those in the first decile of the
share of employment at large firms, are state-industry observations that have few employees in large
firms relative to the industry average.
88
Magnitude of the Collateral Effect Relative to Previous Work
One way to give a rough estimate of the importance of the collateral lending channel
is to compare the magnitude of the employment gains that can be attributed to
this channel to those that can be assigned to the demand channel shown in Mian
and Sufi (2011a). To do so, we follow the same calculation used in that paper to
aggregate the effect across all counties. The authors compute the effect of debtto-income (DTI) ratios as of the beginning of the crisis on the employment change
between 2007 and 2009 in non-tradable industries." These are the industries that are
most likely to be affected by a drop in local demand due to overleveraged households.
They aggregate this effect by computing the predicted change in employment in nontradable industries and then extrapolating this effect to the rest of the economy.
We perform essentially the same calculations for the period 2002-2007 to establish
a benchmark employment effect that can be attributed to the demand channel. We
start by obtaining the effect of a change in house prices on employment in the nontradable industries at a county level for the 2002-2007 period. That regression is
shown in Table 2.5 in Column 3. If we aggregate in the same way as described
above (where the baseline employment is now as of 2002), we obtain an increase in
employment in the non-tradable sector of 451.8 thousand jobs, which, given a share
of employment in this sector of 18.4% in 2002, translates into a predicted total job
gain due to increased aggregate demand of 2.452 million jobs. This is about 40% of
the jobs created in the private sector in the 660 counties used for the calculation. The
confidence interval for this estimation is very large and includes zero, which opens
the possibility that the aggregate demand effect for the period before the crisis may
actually be quite small. This is in sharp contrast to the estimates obtained by Mian
and Sufi (2011a) for the years after the crisis, where the same regression yields very
strong effects for the drop in demand on non-tradable employment.
We now turn to the calculation of the magnitude of the collateral channel over
the same period. We rely on the differential impact of house prices on employment
creation at small firms relative to firms with 50 or more employees, and we focus on
the specifications in which we exclude non-tradable industries and construction (Table
2.3, Column 2). We again first compute predicted county-level employment gains for
these industries (relative to the 10th percentile county) and then we aggregate to all
counties. When we do that, we obtain an estimated total job gain in firms with fewer
than 50 employees relative to those with 50 or more employees of 1.698 million jobs
in all counties, or 27.8% of jobs created between 2002 and 2007 in this period. If
we restrict our attention to the specification where the demand explanation for our
results is the least plausible - that is, the manufacturing sector and, in particular,
firms in industries and states where the shipment distance is largest (Column 6 of
Table 2.4), the same computation would yield an estimate of 676 thousand jobs, or
about 11% of jobs created in this period and subset of counties. Section Al of the
"Using county-level debt-to-income ratio or the run-up in house prices between 2002 and 2007
as the independent variable (as we do in this paper) yields virtually the same results, as counties
with high debt-to-income by the end of this period are also the ones that experienced large increases
in home values.
89
appendix describes the calculation we perform in more detail.
The magnitude we estimate above is a lower bound for the total importance of
collateral for job creation for two reasons. First, our data do not allow us to track firms
over time, so if a firm grows to become very large, we do not attribute the employment
creation of that firm to our effect (it would be in the 50+ category that we use as our
baseline). Second, we are focusing on the importance of this channel for very small
firms. This ignores the role that collateral value plays for larger firms, as pointed
out in Chaney, Sraer, and Thesmar (2012), Cvijanovic (2013), and Chakraborty,
Goldstein, and MacKinlay (2013).
Last, this exercise is useful as a comparison to previous work and not as a proper
calibration of the importance of the collateral effect for the whole economy. In extending the effect that we observe for a subset of firms and industries in individual counties
to the whole economy, we ignore general equilibrium effects that could potentially be
important.
2.3.2
Sole Proprietorships
We now expand our analysis to include the creation of businesses without employees,
also called sole proprietorships or nonemployer businesses. Table 2.6 shows the effect
of housing price growth on net creation of proprietorships relative to all the establishment categories listed in the previous tables using the Saiz measure to instrument for
exogenous movements in housing price changes. The first column in this table uses
employment data on sole proprietorships from the BEA, while the last three columns
rely on census data on nonemployer establishments (which includes information on
the two-digit NAICS sector in which the establishment operates). The coefficient on
housing price growth in Column 1 interacted with the sole proprietorship category
is significantly different from that on the largest establishments and close in magnitude to that on the 1-4-employee category. In Column 2 we use census data and
find a smaller coefficient on the sole proprietorships, and we cannot distinguish that
coefficient from the others in the regression.
In the last two columns we again split the sample by the amount of capital needed
to start a business in a given industry, as discussed above. We find that the effect
of home prices on the net creation of sole proprietorships is stronger in industries
with low start-up capital needs, which is in line with our findings for the other size
categories. Note, however, that the difference between the coefficients in the two
specifications (below and above median capital needs) is not statistically significant.
2.3.3
Crisis Period (2007-2009)
One question that remains regarding the business establishments created as a consequence of the increasing value of collateral during the rise in house prices is whether
these establishments were then eliminated after the housing bubble burst. In this section we try to distinguish whether these newly created businesses were particularly
fragile and were disproportionately affected by the crisis or, alternatively, whether
they behaved like the rest of the firms in the economy.
90
Our data do not allow us to track individual establishments, so we cannot know
whether the specific firms created in the 2002-2007 period survived the crisis. We
can, however, test whether small establishments in general were more or less likely to
downsize or disappear in the crisis. That is, we can assess whether employment loss
was stronger at larger or smaller firms during the crisis in counties where the increase
in house prices had been stronger in the precrisis period (which are also the most
leveraged counties, as shown in Mian and Sufi, 2011a). We run those regressions in
Table 2.7.
The results show that employment loss was either similar across large and small
establishments or, if anything, was worse at large firms (in the specifications without
industry fixed effects) in counties where house prices rose more. This suggests that,
at least as a group, small firms were no more likely to destroy jobs as a consequence
of the increased leverage accumulated during the precrisis period. This is consistent
with the findings of Mian and Sufi (2011a) regarding non-tradable industries for this
period.
2.3.4
Migration
Our final consideration is the effect of house price changes on the net migration
of people in and out of each county. We measure net migration as the difference
between inflows and outflows of individuals at the county level. Table 2.8 shows
county-level regressions of county-to-county Net Migration, as well also Inflows and
Outflows separately, on house prices changes instrumented with the Saiz measure and
the same county-level controls as the previous tables. The results on migration show
no significant effect of the (instrumented) change in house prices on net migration.
This masks stronger results when we break down the results by inflows and outflows.
Indeed, counties that experience higher growth in house prices had larger outflows
that were offset in part by somewhat bigger inflows of people at the same time. This
alleviates the concern that low-elasticity counties experience high growth in demand
due to large in-migration. If anything, the results seem to suggest the opposite. Of
course, we cannot observe who is entering and who is migrating out of each county, so
we cannot address the more detailed question of whether entrepreneurs were moving
in as other individuals were moving out, but the aggregate trends suggest stronger
outflows than inflows in the high-appreciation areas.
2.3.5
Credit Conditions and Elasticity of Housing Supply
One possible concern with the instrument we use is that the behavior of lenders
in high- and low-elasticity areas during our time frame was different. Specifically,
if it became easier to obtain credit in low-elasticity areas relative to high-elasticity
areas during our sample period for reasons unrelated to collateral availability, and if
this drove the creation of new businesses, this would violate the exclusion restriction
for our instrument. One mechanism for such an effect would be that banks might
become laxer on all their credit decisions because of the improvement on the quality
of their mortgage portfolio due to higher house prices. Although the evidence points
91
to commercial lending having become more difficult in places where house prices
boomed (Chakraborty, Goldstein, and MacKinlay, 2013), making it unlikely that
small business credit provision became easier because of stronger mortgage portfolios,
we wish to address this concern directly.
To test whether such an effect is plausible, we use data on denial rates of mortgage applications from HMDA. The underlying assumption is that the cross-sectional
variation on the looseness of credit conditions should be positively correlated with
the same variation for mortgage credit, especially given that the reason why credit
might have become laxer is the fact that house prices increased.
We consider the number of applications that are denied by financial institutions
as a proportion of the total loan applications in a county and in a year." Using
the yearly estimates we compute the proportional change in denial rates between
2002 and 2007. We focus on loans used for purchasing homes because they are less
sensitive to the issue of relationship lending and/or private lender information about
the borrower and therefore should better reflect the loosening of credit conditions.
Panel A of Table 2.9 shows that credit conditions tightened rather than loosened
in low-elasticity areas (those below median elasticity in the sample) when we use this
measure of credit supply. Denial rates increased by about 2% in counties with low
elasticity of housing supply, whereas they go down in high-elasticity areas by 1% - that
is, credit loosened in those areas. The difference between the two types of counties
is statistically significant at the 1% level. In addition, total volume of applications
decreases by 1% in low-elasticity areas in comparison to the 10% increase in the
high-elasticity areas.
We formally test these differences in a regression framework using a continuous
elasticity measure as our independent variable. Panel B of Table 2.9 shows the results.
Consistent with the summary statistics of Panel A, we find that lower elasticity is
associated with higher denial rates of loan applications, and these results are robust
to different specification and controls. Although the regressions condition on the
applicant pool (and so the denial ratc could mask riskier borrowers applying for
loans), we control for the debt-to-income in these regressions to account for changes
in applicant types.
Overall, this result allows us to rule out the concern that our instrument is picking
up changes in the way that lenders granted credit instead of access to credit through
an increase in collateral values.
2.4
Conclusion
Overall, the evidence we present identifies the causal effect of rising house prices in
the creation of new small firms. Increased access collateral allowed individuals to
start small businesses or to become self-employed. We conjecture that without access
to this collateral in the form of real estate assets, many individuals would not have
12
Volume of applications is calculated as the sum of all loans that are originated plus applications
that are approved but not accepted, applications denied by the financial institution, and loans
purchased by the financial institution itself.
92
made the transition to starting a new business or self-employment. Our study is in
line with recent survey evidence from the NY Fed" that shows that: (i) access to
capital is the top growth challenge for small firms in 2013; (ii) the most cited reason
for not receiving credit is insufficient collateral; and (iii) that the most used form of
collateral for small businesses is personal real estate (in line also with the findings of
Kleiner, 2013). This implies that the effect we uncover is a collateral effect and not
the result of changing household risk-aversion due to increased wealth (as suggested
by Kihlstrom and Laffont, 1979).
We show that the effect of house prices is concentrated in small firms only and
has no causal effect on employment at large firms. Importantly, our results also hold
when we exclude industries that are most likely to be affected by local demand shocks
and when we restrict our attention to manufacturing industries. The effect of house
prices is also stronger in industries where the amount of capital needed to start a new
firm is lower, consistent with the hypothesis that housing serves as collateral but is
not sufficient to fund large capital needs.
Our results on the collateral effect on the upside (2002-2007) and after the crisis
hit, paired with the results on the effect of demand on job creation, suggest an interesting asymmetry of these effects. Collateral was particularly important in explaining
job creation when more collateral became available, but we observe no significant destruction when collateral became scarce. This is consistent with a "bright side" of
bubbles (as suggested in Caballero, Farhi, and Hammour, 2006, although the effect
we emphasize is quite different). On the other hand, a drop in demand is a strong
predictor of employment loss, but a similar shock on the upside (at least in the recent
experience) does not seem as powerful in predicting where jobs will be created.
2.5
Bibliography
Adelino, M., Schoar, A., Severino, F., 2012. Credit supply and house prices:
evidence from mortgage market segmentation. NBER Working Paper No. 17832.
Berger, A., Udell, G. F., 1998. The economics of small business finance: the roles
of private equity and debt markets in the financial growth cycle. Journal of Banking
and Finance 22, 613-673.
Bernanke, B. S., Gertler, M., 1989. Agency costs, net worth, and business fluctuations. American Economic Review 79, 14-31.
Black, J., De Meza, D., Jeffreys, D., 1996. House prices, the supply of collateral
and the enterprise economy. Economic Journal, 60-75.
Brown, M., Stein, S., Zafar, B., 2013. The impact of housing markets on consumer
debt: credit report evidence from 1999 to 2012. Federal Reserve Bank of New York
Working Paper.
Caballero, R. J., Farhi, E., Hammour, M. L., 2006. Speculative growth: hints
from the US economy. American Economic Review, 1159-1192.
13
Small Business Credit Survey, May 2013, Federal Reserve Bank of New York, available at
http://www.newyorkfed.org/smallbusiness/2013/.
93
Cagetti, M., De Nardi, M., 2006. Entrepreneurship, frictions, and wealth. Journal
of Political Economy 114,
Campbell, J. Y., Cocco, J. F., 2007. How do house prices affect consumption?
Evidence from micro data. Journal of Monetary Economics 54,
Chakraborty, I., Goldstein, I., MacKinlay, A., 2013. Do asset price bubbles have
negative real effects? Unpublished working paper.
Chaney, T., Sraer, D., Thesmar, D., 2012. The collateral channel: how real estate
shocks affect corporate investment. American Economic Review 102, 2381-2409.
Charles, K. K., Hurst, E., Notowidigdo, M. J., 2012. Manufacturing busts, housing
booms, and declining employment: a structural explanation. Unpublished working
paper.
Corradin, S., Popov, A., 2013. House prices, home equity and entrepreneurships.
ECB Working Paper.
Cvijanovic, D., 2013. Real estate prices and firm capital structure. Unpublished
working paper.
Davidoff, T., 2012. Supply elasticity and the housing cycle of the 2000s. Real
Estate Economics, forthcoming.
Duranton, G., Morrow, P., Turner, M. A., 2013. Roads and trade: evidence from
the US. CEPR Discussion Papers 9393.
Evans, D. S., Jovanovic, B., 1989. An estimated model of entrepreneurial choice
under liquidity constraints. Journal of Political Economy 97, 808-827.
Fairlie, R. W., Krashinsky, H. A., 2012. Liquidity constraints, household wealth,
and entrepreneurship revisited. Review of Income and Wealth 59, 279-306.
Fan, W., White, M. J., 2003. Personal bankruptcy and the level of entrepreneurial
activity. Journal of Law and Economics 46.
Fort, T., Haltiwanger, J. C., Jarmin, R. S., Miranda, J., 2012. How firms respond
to business cycles: the role of firm age and firm size. Unpublished working paper.
Fracassi, C., Garmaise, M. J., Kogan, S., Natividad, G., 2013. How much does
credit matter for entrepreneurial success in the United States? Unpublished working
paper.
Gentry, W. M., Hubbard, R. G., 2005. Success taxes, entrepreneurial entry, and
innovation. In: Innovation Policy and the Economy, Volume 5. MIT Press, Cambridge, pp. 87-108.
Gertler, M., Gilchrist, S., 1994. Monetary policy, business cycles, and the behavior
of small manufacturing firms. Quarterly Journal of Economics 109, 309-340.
Greenstone, M., Mas, A., 2012. Do credit market shocks affect the real economy?
Quasi-experimental evidence from the Great Recession and "normal" economic times.
Unpublished working paper.
Haltiwanger, J. C., Jarmin, R. S., Miranda, J., 2011. Who creates jobs? Small
vs. large vs. young. Unpublished working paper.
Holtz-Eakin, D., Joulfaian, D., Rosen, H. S., 1994. Entrepreneurial decisions and
liquidity constraints. NBER Working Paper No. 4526.
Hurst, E., Lusardi, A., 2004. Liquidity constraints, household wealth, and entrepreneurship. Journal of Political Economy 112, 201-248.
94
Iacoviello, M., 2005. House prices, borrowing constraints, and monetary policy in
the business cycle. American Economic Review 95, 739-761.
Kashyap, A. K., Stein, J. C., 1994. Monetary policy and bank lending. In:
Monetary Policy, University of Chicago Press, Chicago, pp. 221-261.
Kashyap, A. K., Stein, J. C., Wilcox, D. W., 1993. Monetary policy and credit
conditions: evidence from the composition of external finance. American Economic
Review 83, 78-98.
Kihlstrom, R. E., Laffont, J.-J., 1979. A general equilibrium entrepreneurial theory of firm formation based on risk aversion. Journal of Political Economy 719-748.
Kiyotaki, N., Moore, J., 1997. Credit cycles. Journal of Political Economy 105,
211-248.
Kleiner, K., 2013. How real estate drives the economy: an investigation of small
firm balance sheet shocks on employment. Unpublished working paper.
Kocherlakota, N., 2010. Inside the FOMC. President's speech, Marquette, MI,
August 17, 2010.
Mian, A., Sufi, A., 2011a. What explains high enemployment? The aggregate
demand channel. NBER Working Paper No. .
Mian, A., Sufi, A., 2011b. House prices, home equity based borrowing, and the
U.S. household leverage crisis. American Economic Review 101, 2132-2156.
Petersen, M. A., Rajan, R. G., 1994. The benefits of lending relationships: evidence from small business data. Journal of Finance 49, 3-37.
Rampini, A. A., Viswanathan, S., 2010. Collateral, risk management, and the
distribution of debt capacity. Journal of Finance 65, 2293-2322.
Robb, A. M., Robinson, D. T., 2012. The capital structure decisions of new firms.
Review of Financial Studies, forthcoming.
Romer, C. D., 2011. Dear Ben: It's time for your Volcker moment. New York
Times, October 29, 2011.
Saiz, A., 2010. The geographic determinants of housing supply. Quarterly Journal
of Economics 125, 1253-1296.
Schmalz, M., Sraer, D., Thesmar, D., 2013. Housing collateral and entrepreneurship. Unpublished working paper.
95
Table 2.1: Summary Statistics
Panel A
All Counties
High Elasticity
Low Elasticity
113,918
45,454
(238,831)
5.4
5.3
(1.5)
100.2
46.2
(188.1)
10.6
8.2
(15.8)
51.8
42.6
(36.4)
27.6
23.9
(21.1)
33.9
26.8
(21.1)
-0.9
-0.8
(1.0)
775
69,057
33,228
(129,569)
5.3
5.2
(1.5)
59.3
34.2
(92.6)
10.2
7.5
(16.9)
36.6
34.9
(23.0)
27.2
23.0
(24.2)
23.5
19.4
(14.3)
-0.7
-0.5
(0.9)
382
157,523
63,286
(304,041)
5.4
5.4
(1.4)
139.8
66.4
(241.4)
11.0
8.9
(14.5)
66.3
58.3
(40.7)
28.0
24.5
(17.6)
43.7
40.9
(21.9)
-1.0
-1.0
(1.0)
Total Employment (2002)
Unemployment Rate (2002, percent)
Number of Households (2000, thousands)
Growth in Total Employment (02-07, percent)
Growth in DTI (02-07, percent)
Growth in Income (02-07. percent)
Growth in House Prices (02-07, percent)
Change in Unemployment Rate (02-07, percent)
Number of Counties
393
Panel B
1-4 Emp
5-9 Emp
10-19 Emp
20-49 Emp
50+ Emp
9,101
9,122
8.0
9.0
12,819
12.5
12.1
21,466
10.6
18.3
72,939
13.3
51.7
5.580
7,365
11,033
39,964
12.1
11.0
10.8
13.4
12.8
14.0
16.6
24.6
47.7
2.866
6.9
5.8
3,542
4.4
7.4
5,454
13.1
11.7
10,433
9.6
20.5
32,975
9.3
54.6
Emp. in All Sectors
Total
Growth (02-07)
Percentage of Total
Emp. in Firms <P50 of Start-Up Capital
Total
Growth (02-07)
Percentage of Total
Emp. in Firms >P50 of Start-Up Capital
Total
Growth (02-07)
Percentage of Total
9.4
8.9
6,235
10.8
Note. Panel A reports summary statistics for all counties in the sample in Column 1, and Columns 2 and 3 show
the summary statistics for counties above and below the median elasticity of housing supply in the sample. For
each variable we show the pooled average, median (italicized) and standard deviation (in parenthesis). The last
column shows the t-statistic for the difference in means of the two groups, adjusted for clustering at the Metropolitan
Statistical Area level. Total Employment refers to the total number of employees in a county in thousands across all
establishment sizes and industries using the County Business Patterns data as of 2002. Unemployment Rate is shown
in percentage and comes from the Bureau of Labor Statistics Local Area statistics in 2002. Percent College Educated
is the percentage of the population with a college degree, Percent Employed is the percentage of the labor force that
is employed, Workforce as a Percentage of Population is the share of the population in the workforce, and Percent
of Homes Owner-occupied is the percentage of homes that are owner-occupied (i.e., not rental properties). Average
Household Income is the total income in a county divided by the number of households as of 2002 and Growth in
Income is the percentage change in income in a county between 2002 and 2007. Change in DTI is the percentage
change in debt to income ratio in the same period. The debt to income ratio is estimated using county level household
debt data from the New York Fed-Equifax and income is computed using IRS county-level information. Growth in
House Prices is the percentage change in house prices between 2002 and 2007 at the MSA level from the Federal
Housing Finance Agency. Panel B shows the Total Employment in 2002 in thousands, Employment Growth between
2002 and 2007 in percentage points, and the percentage of Total Employment for each establishment size for all firms,
as well as split by the start-up amount of capital needed to start a firm.
96
-- I
0.06
(0.09)
(0.00)
-0.86***
0.00
(0.00)
-1.11***
(0.19)
0.00
(0.00)
-1.09***
(0.19)
R2
3.653
0.22
0.12
373.576
0.30
Y
-0.08
(0.32)
0.00
((.1)
(0.22)
0.00
(1.1 )
21,962
0.02
Y
-0.07**
(0.03)
-0.05
(0.04)
(0.04)
0.00
0.16***
(0.05)
0.02
(0.03)
0.00
(1.1))
0.08
(0.38)
0.00**
0.33
3,653
0.21
(0.26)
196,027
0.39
Y
-1.00***
(0.25)
0.00
(0.00)
0.00
(0.00)
-0.05***
((1.01)
0.10
(0.10)
0.02
(0.12)
(0.15)
0.14
0.32**
(0.12)
-0.04
(0.13)
-1.16***
(0.20)
0.00
(0.00)
0.00
(0.)))
-0.03***
(0.01)
(0.05)
0.13***
0.14***
(0.05)
(0.05)
0.19***
0.33***
(0.07)
(0.0)
(9)
0.04
(0.06)
0.17***
21.954
0.00
Y
0.06
(0.05)
(.10*
(0.05)
(0.05)
3,651
0.14
-
-0.01
(0.22)
0.00
(11.111)
-1.08***
(0.20)
0.00**
(0.00)
177,549
0.10
Y
-0.19
(0.30)
(11.1)
0.00*
-0.72***
(0.21)
0.00
(0.00)
0.00
(0.00)
0.00**
(0.00)
-0.04***
(0.01)
0.02
(0.08)
(0.09)
0.09
0.19**
(0.08)
0.18**
(0.09)
-0.07
(0.10)
-0.02***
(0.01)
-0.07
(0.05)
-0.07
(0.06)
0.14**
(0.06)
0.31***
(0.06)
0.06
(0.07)
-0.11***
(0.04)
(11)
21,949
0.03
Y
-0.14***
(0.04)
-0.12**
(0.05)
(0.05)
-0.10*
0.10*
(0.06)
0.10**
(0.04)
> P50
(10)
(8)
< P50
(7)
-0.01
(0.07)
(6)
Start-up Capital
Start-up Capital
establishments with 50 or more employees. All regressions control for the natural logarithm of population, the percentage of the population with a college degree, the percentage of the labor force that
is employed, the share of the population in the workforce, and the percentage of homes that are owner occupied. Controls are at a county level for the year 2000 and are obtained using Census Bureau
Data Summary Files. Standard errors are in parenthesis and are clustered by MSA. *, **, *** indicate statistical significance at 10, 5, and 1% levels, respectively.
level and then the IV results using yearly observations and the interaction of the elasticity measure with the conventional mortgage rates as the instrument. Columns 6 through 11 show the coefficients
split by the start-up capital amount (above and below the median) also at the county, at the county and industry level, and at the county level with yearly observations. The omitted category refers to
The table shows two-stage least squares regressions of employment growth on house price growth instrumented with the elasticity of housing supply, indicator variables for each establishment size (not
shown in the table) and interactions of house price growth with the size of establishments. All regressions are weighted by the number of households in a county as of 2000. Employment growth is the
percentage change in employment between 2002 and 2007 estimated using County Business Patterns (CBP) data. Growth in House prices is the percentage change between 2002 and 2007, and each
interaction is with a dummy indicator for the size of the establishment. Column 1 shows the first stage regression of the change in house prices between 2002 and 2007 on the Saiz elasticity measure.
Columns 2 through 5 "All Industries" shows the results for the whole sample of firms, first the weighted least squares results, then the IV at a county level, the IV results at a county and industry
731
0.30
3,653
0.27
-
-
4-Digit Industry Fixed Effects
County Fixed Effects
Number of Observations
(0.23)
0.09
(0.23)
0.10
China Import Share in County (2005)
(0.91)
0.00
((1.00)
Percent of Homes Owner-occupied
0.00**
(1.1 ))
0.00**
(1.1)0)
-0.69
(0.63)
(0.00)
-o.01***
Workforce as a Percentage of Population
Percent Employed (2000 Census)
((1.)))
(0.00)
0.00
0.00**
0.00**
0.00
-0.04***
(0.01)
0.07
(0.07)
(0.00)
0.00
(0.04)
-0.02***
(0.01)
Percent College Educated
0.01
(0.02)
0.17
-0.02***
(0.01)
0.00
(0.03)
Log of the Population
Growth in House Prices * 20-49 Employees
(0.02)
-0.02
0.01
(0.04)
10-19 Employees
(0.04)
Growth in House Prices
(0.10)
0.08**
-0.02
(0.03)
0.26**
(0.09)
Growth in House Prices * 5-9 Employees
0.20***
(0.05)
0.03
(0.03)
-0.06
(0.10)
Growth in House Prices * 1-4 Employees
0.05
(0.06)
All Industries
(IV)
(5)
(3)
(4)
0.19***
(0.04)
All Industries
(WLS)
(2)
Growth in House Prices
Housing Supply Elasticity
(1)
-0.09***
(0.02)
First Stage
Table 2.2: Employment Growth, Firm Size, and House Price Appreciation
Percent College Educated
20-49 Employees
10-19 Employees
Growth in House Prices * 5-9 Employees
Growth in House Prices * 1-4 Employees
Growth in House Prices
0.00
(0.00)
0.00
(0.00)
-0.04***
(0.01)
0.08
(0.06)
0.08
(0.09)
0.19*
(0.10)
(0.09)
0.27***
-0.09
(0.10)
Construction
Drop
(0.23)
-0.84***
0.00
(0.00)
0.00
(0.00)
-0.04***
(0.01)
0.12*
(0.06)
0.12
(0.09)
0.21*
(0.11)
0.32***
(0.09)
-0.12
(0.10)
and Non-Trad.
Drop Const.
0.00*
(0.00)
(0.24)
-0.84***
(1.00)
0.00
0.00
(0.00)
-0.04***
(0.01)
0.11*
(0.06)
0.12
(0.09)
0.24**
(0.11)
(0.10)
0.35***
-0.14
(0.10)
Non-Trad. and F.I.R.E.
Drop Const.,
-0.88*
0.00*
(0.00)
(0.29)
-0.64**
0.00
(0.00)
0.00
(0.00)
-0.02**
(0.01)
0.01
(0.12)
0.11
(0.11)
0.12
(0.08)
0.13*
(0.07)
-0.17
(0.11)
Manufacturing
Y
(0.56)
-1.24**
0.00
(0.00)
(0.30)
-0.66**
0.00
(0.00)
0.00
(0.00)
-0.02*
(0.01)
-0.05
(0.09)
0.16
(0.11)
0.10
(0.09)
0.15*
(0.09)
-0.16
(0.12)
Manufacturing (Tradable)
Table 2.3: Employment Growth and House Prices: Excluding Construction, Non-Tradable, and Finance Industries and Considering Manufacturing Only
Percent Employed (2000 Census)
(0.22)
-0.88***
0.00
(0.00)
-0.28
Y
(0.50)
Growth in House Prices
Log of the Population
0.04**
Y
325,349
0.29
0.00***
).00***
Y
264,901
0.30
0.02**
0.00***
0.00***
0.02**
Y
242,510
0.31
0.33
0.85
0.95
Y
55,345
0.02
0.91
0.48
Y
44,649
0.02
(Tradable)
Workforce as a Percentage of Population
0.00
(0.00)
-0.23
Y
(0.36)
Growth in House Prices
Percent of Homes Owner-occupied
-0.11
Y
(0.36)
*
China Import Share in County (2005)
Y
(0.34)
4-Digit Industry Fixed Effects
Number of Observations
R2
0.00***
0.00***
Controls
Growth HP * 1-4 E. = Growth HP * 5-9 E.
Growth HP * 1-4 E. = Growth HP * 10-19 E.
0.10*
Growth HP * 1-4 E. = Growth HP * 20-49 E.
The table shows two stage least squares regressions of employment growth on house price growth instrumented with the elasticity of housing supply, indicator variables for each establishment size (not
shown in the table) and interactions of house price growth with the size of establishments. Each observation is at a county, 4-digit NAICS industry, and establishment size level. All regressions are
weighted by the number of households in a county as of 2000. House Price Growth is instrumented using the Saiz (2010) measure of elasticity of housing supply at an MSA level. Employment growth is
the percentage change in employment between 2002 and 2007 estimated using County Business Patterns (CBP) data. Growth in House prices is the percentage change between 2002 and 2007, and each
interaction is with a dummy indicator for the size of the establishment. All regressions include 4-digit industry fixed effects. Column 1 shows the results when we exclude construction industries, column
2 excludes both construction and non-tradable industries, column 3 also excludes finance, insurance and real estate-related industries (NAICS codes 52 and 53), column 4 includes only manufacturing
industries (NAICS 31 to 33) and column 5 has manufacturing industries that are classified as "tradable" in Mian and Sufi (2011a). All regressions control for the natural logarithm of population, the
percentage of the population with a college degree, the percentage of the labor force that is employed, the share of the population in the workforce, and the percentage of homes that are owner-occupied.
All controls are at a county level for the year 2000 and are obtained using Census Bureau Data Summary Files. Standard errors are in parenthesis and are clustered by MSA. *, *, *
denote statistical
significance at the 10, 5, and 1% levels, respectively.
Table 2.4: Breakdown of Manufacturing Industries by Distance Shipped
Manufacturing
Dist. Shipped <P50
Manufacturing
Dist. Shipped >P50
Growth in House Prices
-0.11
(0.17)
-0.29**
(0.14)
Growth in House Prices * 1-4 Employees
0.07
(0.14)
0.21**
(0.09)
Growth in House Prices * 5-9 Employees
0.11
(0.17)
0.20**
(0.09)
Growth in House Prices * 10-19 Employees
-0.03
(0.17)
0.24**
(0.11)
0.06
(0.30)
0.04
(0.12)
Log of the Population
-0.02
(0.02)
-0.02*
(0.01)
Percent College Educated
0.00
(0.00)
0.00
(0.00)
Percent Employed (2000 Census)
0.00
(0.00)
0.00
(0.00)
Workforce as a Percentage of Population
-0.42
(0.36)
-0.58*
(0.32)
Percent of Homes Owner-occupied
0.00
(0.00)
0.00*
(0.00)
China Import Share in County (2005)
-0.29
(0.45)
Y
Y
27,599
0.02
-1.21**
(0.58)
Y
Y
27,294
0.02
0.82
0.90
0.77
Growth in House Prices * 20-49 Employees
Controls
4-Digit Industry Fixed Effects
Number of Observations
R2
Growth HP * 1-4 E. = Growth HP * 5-9 E.
Growth HP * 1-4 E. = Growth HP * 10-19 E.
Growth HP * 1-4 E. = Growth HP * 20-49 E.
0.59
0.96
0.13
The table shows two-stage least squares regressions of employment growth on house price growth instrumented with
the elasticity of housing supply, indicator variables for each establishment size (not shown in the table) and interactions
of house price growth with the size of establishments. Each observation is at a county, 4 digit NAICS industry, and
establishment size level. All regressions are weighted by the number of households in a county as of 2000. House Price
Growth is instrumented using the Saiz (2010) measure of elasticity of housing supply at an MSA level. Employment
growth is the percentage change in employment between 2002 and 2007 estimated using County Business Patterns
(CBP) data for manufacturing industries (NAICS codes 31 to 33). Growth in House prices is the percentage change
between 2002 and 2007, and each interaction is with a dummy indicator for the size of the establishment. All
regressions include 4 digit NAICS fixed effects. The table splits industries and states based on the median of the
shipment distance distribution (about 600 miles). Data for distance shipped is from the Census Commodity Flow
Survey for 2007 and represents a dollar weighted average of shipment distance calculated at the 3 digit NAICS and
state of origin level. All regressions control for the natural logarithm of population, the percentage of the population
with a college degree, the percentage of the labor force that is employed, the share of the population in the workforce,
and the percentage of homes that are owner occupied. All controls are at a county level for the year 2000 and are
obtained using Census Bureau Data Summary Files. Standard errors are in parenthesis and are clustered by MSA. *,
**' *** denote statistical significance at the 10, 5, and 1% levels, respectively.
99
Percent College Educated
-0.69
(0.63)
-0.01***
(0.00)
0.00
(0.00)
(0.03)
-0.23
(0.28)
0.00**
(0.00)
(0.23)
-1.15***
0.00
(0.00)
0.00*
(0.00)
-0.02**
(0.01)
0.09
(0.06)
All Industries
731
0.18
0.42
(0.32)
0.00
(0.00)
(0.28)
-1.13***
0.00*
(0.00)
0.00**
(0.00)
-0.01
(0.01)
(0.07)
0.10
Non-Tradable
730
0.10
-1.94***
(0.47)
0.00**
(0.00)
-0.82
(0.51)
0.00
(0.00)
0.00
(0.00)
-0.02**
(0.01)
-0.01
(0.11)
Tradable
0.30
731
-0.52
(0.42)
0.00(**
(0.00)
-0.83**
(0.37)
0.00
(0.00)
0.00
(0.00)
(0.01)
-0.02*
0.32***
(0.08)
Construction
731
0.21
0.42
(0.32)
0.00
(0.00)
-1.35
(0.24)
0.00
(0.00)
0.00
(0.00)
-0.03
(0.01)
0.06
(0.06)
Others
Table 2.5: Employment and House Price Appreciation across Industry Types
Percent Employed (2000 Census)
0.00
(0.00)
731
Log of the Population
First Stage
Workforce as a Percentage of Population
0.10
(0.91)
0.24
0.00
(0.02)
Percent of Homes Owner-occupied
731
Housing Supply Elasticity
China Import Share in County (2005)
0.30
Growth in House Prices
R2
Number of Observations
The table shows two stage least squares regressions at a county level of employment growth on house price growth between 2002 and 2007. Each observation is at a county level. All regressions are
weighted by the number of households in a county as of 2000. House Price Growth is instrumented using the Saiz (2010) measure of elasticity of housing supply at an MSA level. Employment growth
is the percentage change in employment between 2002 and 2007 estimated using County Business Patterns (CBP) data. Industry type definitions follow Mian and Sufi (2011a). All regressions control
for the natural logarithm of population, the percentage of the population with a college degree, the percentage of the labor force that is employed, the share of the population in the workforce, and the
percentage of homes that are owner occupied. All controls are at a county level for the year 2000 and are obtained using Census Bureau Data Summary Files. Standard errors are in parenthesis and are
clustered by MSA. *, **, *** denote statistical significance at the 10%, 5%, and 1% levels, respectively.
Table 2.6: Proprietorships and House Price Appreciation
Start-up Capital
> P50 (Census)
BEA
Data
Census
Data
Start-up Capital
< P50 (Census)
0.02
(0.06)
0.03
(0.06)
-0.04
0.05
(0.07)
(0.07)
Growth in House Prices * Proprietorships
0.14*
(0.07)
0.06
(0.06)
0.12*
(0.06)
0.08
(0.08)
Growth in House Prices * 1-4 Employees
0.20***
(0.05)
0.20***
(0.05)
0.33***
(0.07)
0.14**
(0.06)
Growth in House Prices * 5-9 Employees
0.08**
(0.04)
0.08**
(0.04)
0.19***
(0.05)
0.04
(0.06)
Growth in House Prices * 10-19 Employees
0.01
(0.04)
0.01
(0.04)
0.14***
(0.05)
-0.07
(0.06)
Growth in House Prices * 20-49 Employees
0.00
(0.04)
0.00
(0.04)
0.13**
(0.05)
-0.07
(0.05)
Log of the Population
-0.02**
(0.01)
-0.02**
(0.01)
-0.02***
(0.01)
-0.02**
(0.01)
Percent College Educated
0.00**
(0.00)
0.00*
(0.00)
0.00
(0.00)
0.00**
(0.00)
Percent Employed (2000 Census)
0.00
(0.00)
0.00
(0.00)
0.00
(0.00)
0.00
(0.00)
-1.02***
(0.19)
-1.16***
(0.20)
-1.21***
(0.21)
-1.13***
Percent of Homes Owner-occupied
0.00**
(0.00)
0.00**
(0.00)
0.00**
(0.00)
0.00*
(0.00)
China Import Share in County (2005)
0.02
(0.22)
0.03
(0.23)
0.18
(0.24)
-0.02
(0.23)
Number of Observations
R2
4,381
0.48
4,384
0.38
4,384
0.31
4,382
Growth in House Prices
Workforce as a Percentage of Population
(0.21)
0.28
The table shows two-stage least squares regressions at a county level of employment growth on house price growth,
indicator variables for each establishment size (not shown in the table) and interactions of house price growth with the
size of establishments. Proprietorships are establishments with zero employees. Each observation is at a county and
establishment size level. All regressions are weighted by the number of households in a county as of 2000. House Price
Growth is instrumented using the Saiz (2010) measure of elasticity of housing supply at an MSA level. Employment
growth is the percentage change in employment between 2002 and 2007 estimated using County Business Patterns
(CBP) data except in the case of proprietorships. The data on growth in proprietorships is obtained from the Bureau
of Economic Analysis in the first column and from the Census in Columns 2 to 4. All regressions control for the natural
logarithm of population, the percentage of the population with a college degree, the percentage of the labor force that
is employed, the share of the population in the workforce, and the percentage of homes that are owner-occupied. All
controls are at a county level for the year 2000 and are obtained using Census Bureau Data Summary Files. Standard
errors are in parenthesis and are clustered by MSA. *, *, *** denote statistical significance at the 10%, 5%, and 1%
levels, respectively.
101
Table 2.7: Employment Growth, Firm Size, and House Price Appreciation, Crisis
Period (2007-2009)
All Industries
(WLS)
All Industries
(IV)
Start-up Capital
< P50 (IV)
Growth in House Prices
-0.04*
(0.02)
-0.12***
(0.03)
-0.13***
(0.04)
> P50 (IV)
-0.14***
(0.04)
Growth in House Prices * 1-4 Employees
0.04**
(0.02)
0.10***
(0.03)
0.11***
(0.04)
0.13***
(0.05)
Growth in House Prices * 5-9 Employees
0.01
(0.02)
0.05*
(0.03)
0.05*
(0.03)
0.09
(0.05)
Growth in House Prices * 10-19 Employees
0.00
(0.02)
0.06*
(0.03)
0.07**
(0.03)
0.09**
(0.04)
Growth in House Prices * 20-49 Employees
-0.02
(0.02)
0.02
(0.03)
0.00
(0.03)
0.07
(0.05)
Log of the Population
-0.01***
(0.00)
0.00**
(0.00)
0.00*
(0.00)
-0.01***
(0.00)
Percent College Educated
0.00***
(0.00)
0.00***
(0.00)
0.00***
(0.00)
0.00***
(0.00)
0.00*
(0.00)
0.00***
(0.00)
0.00***
(0.00)
0.00***
(0.00)
Workforce as a Percentage of Population
-0.25***
(0.07)
-0.26***
(0.06)
-0.26***
(0.07)
-0.25***
(0.07)
Percent of Homes Owner-occupied
0.00***
(0.00)
0.00***
(0.00)
0.00***
(0.00)
0.00***
(0.00)
China Import Share in County (2005)
0.12*
(0.07)
0.14*
(0.08)
0.25***
(0.09)
0.06
(0.08)
Number of Observations
R2
3,654
0.16
3,654
0.12
3.651
0.08
3,653
0.13
Percent Employed (2000 Census)
Start-up Capital
The table shows two-stage least squares regressions of employment growth between 2007 and 2009 on house price
growth for the previous 5 years (2002-2007), indicator variables for each establishment size (not shown in the table)
and interactions of house price growth with the size of establishments. All regressions are weighted by the number of
households in a county as of 2000. House Price Growth is instrumented using the Saiz (2010) measure of elasticity of
housing supply at an MSA level. Employment growth is the percentage change in employment between 2007 and 2009
estimated using County Business Patterns (CBP) data. Growth in House prices is the percentage change between
2002 and 2007, and each interaction is with a dummy indicator for the size of the establishment. Columns 1 and 2, All
Industries, shows the results for the whole sample of firms (first the weighted least squares results and then the IV),
Columns 3 to 6 show the coefficients split by the startup capital amount. The omitted category refers to firms with
50 or more employees. The first column for each sample of industries is aggregated at the county and establishment
size level, whereas the second column is at the county, establishment size and industry level, and includes industry
fixed effects. All regressions control for the natural logarithm of population, the percentage of the population with a
college degree, the percentage of the labor force that is employed, the share of the population in the workforce, and
the percentage of homes that are owner occupied. All controls are at a county level for the year 2000 and are obtained
using Census Bureau Data Summary Files. Standard errors are in parenthesis and are clustered by MSA. *, **, ***
denote statistical significance at the 10%, 5%, and 1% levels, respectively.
102
731
0.24
Number of Observations
R2
731
0.41
0.18
731
(0.44)
731
-1.27***
(0.28)
(0.00)
-1.08***
(0.00)
.0** -0.01***
-0.62**
(0.26)
Net Migration is calculated by county using inflows of taxpayers minus outflow of taxpayers in a year as a proportion of non migrants (i.e. people that filed in
the same county in t-1 and t). For each dependent variable the first column shows the results for the regressions without controls, and the second column shows the coefficients controlling for log of
population, the percentage of the population with a college degree, the percentage of the labor force that is employed, the share of the population in the workforce, and the percentage of homes that are
owner occupied. All controls are at a county level for the year 2000 and are obtained using Census Bureau Data Summary Files. Standard errors are in parenthesis and are clustered by MSA. *, *, *
denote statistical significance at the 10%, 5%, and 1% levels, respectively.
county to county migration data series.
The table shows two stage least squares regressions at a county level of the net migration on house price growth between 2002 and 2007. All regressions are weighted by the number of households in a
county as of 2000. House Price Growth is instrumented using the Saiz (2010) measure of elasticity of housing supply at an MSA level. Net Migration, Inflows and Outflows are obtained from the IRS
721
0.33
721
0.26
0.19
(0.29)
-0.60
-0.23
(0.28)
China Import Share in County (2005)
-4.76
(3.65)
(0.00)
(0.01)
(0.64)
0.00**
0.03***
0.00***
(0.00)
0.00**
(0.00)
Percent of Homes Owner-occupied
-0.01
(0.19)
3.94
(2.67)
-0.13
(0.52)
Workforce as a Percentage of Population
-0.63*
(0.34)
0.00
(0.00)
0.00
(0.00)
0.00
(0.00)
0.00***
(0.00)
0.01***
(0.00)
(0.00)
0.00
(0.01)
0.04**
(0.02)
(0.01)
-0.07***
(0.01)
(0.17)
0.34**
0.19
(0.12)
Outflows
Inflows
-0.07***
0.00
(0.00)
(0.00)
-0.01***-0.03***
0.03
(0.10)
-0.01
(0.02)
(0.23)
(0.00)
0.00
(0.00)
0.00**
(0.66)
0.00
(0.01)
-0.16
(0.12)
-1.29**
Net Migration
-0.20
Unemp. Rate
(0.14)
Unemp.
-1.15***
Percent Employed (2000 Census)
Percent College Educated
-0.02***
Log of the Population
(0.01)
0.09
(0.06)
Growth in House Prices
Total Employment
Table 2.8: Total Employment, Unemployment, and Migration
Table 2.9: Denial Rates
Panel A
Low Elasticity
High Elasticity
0.12
0.02
(0.06)
9,454
0.07
-0.01
(0.27)
394
0.14
-0.01
(0.05)
3,811
0.06
0.10
(0.22)
382
Denial Rate (2002)
Change in Denial Rate (02-07)
Volume (2002)
Volume per Household (2002)
Change in Volume (02-07)
Number of Counties
Difference
0.03***
Panel B
.-0.11***
Denial Rates
Elasticity
-0.03***
(0.00)
Volume
-0.01
(0.02)
0.02
(0.02)
-0.01
(0.04)
-0.57***
(0.11)
-0.13
(0.21)
0.02*
(0.01)
0.06***
(0.01)
-0.26***
(0.05)
-0.29**
(0.10)
0.02***
(0.00)
0.02***
(0.00)
-0.05**
-0.08**
(0.02)
(0.03)
0.00***
(0.00)
0.00***
(0.00)
0.01**
0.00
(0.00)
(0.00)
0.00
(0.00)
0.00***
(0.00)
-0.01**
0.00
(0.00)
(0.00)
Workforce as a Percentage of Population
-0.15*
(0.08)
-0.08
(0.10)
-1.05**
(0.44)
-1.10*
(0.61)
Percent of Homes Owner-occupied
0.00*
(0.00)
0.00
(0.00)
-0.01***
(0.00)
-0.01***
(0.00)
-0.39***
(0.11)
-0.49***
(0.11)
-0.12
(0.66)
0.47
(0.90)
NY Fed / IRS
763
0.58
HMDA
774
0.55
NY Fed / IRS
763
0.42
HMDA
774
026
Debt to Income (2002)
Changre in Debt to Income (02-07)
Log of the Population
Percent College Educated
Percent Employed (2000 Census)
China Import Share in County (2005)
DTI data
Number of Observations
P2
776
0130
-0.01***
(0.00)
-0.01***
(0.00)
0.11***
(0.02)
0.07**
(0.03)
776
0.09
The table shows the relation between mortgage denial rates and mortgage volume at a county level and the elasticity of
housing supply. Total application volume is calculated as the sum of all loans that are originated plus applications that
are approved but not accepted, applications denied by the financial institution and loans purchased by the financial
institution itself in each county and year, all scaled by the total number of households in a county as of 2000. Denial
rates are computed as the proportion of applications denied by the financial institution over total volume in each
county and year. All the data is extracted from HMDA LAR records. Panel A shows the average denial rates and
average volume in 2002 and 2007, as well as the change in these variables during this period for counties above and
below the median elasticity of housing supply in the sample. Panel B shows OLS regressions of the change in denial
rate the change in total volume of applications on housing supply elasticity as a continuous variable and controls
(debt to income level and changes, the natural logarithm of the population, the percentage of the population with a
college degree, the percentage of the labor force that is employed, the share of the population in the workforce, the
percentage of homes that are owner occupied). All regressions are weighted by the number of households as of 2000.
*, **, *** denote statistical significance at the 10%, 5%, and 1% levels, respectively.
104
2.6
Appendix. Calculating the magnitude of the
collateral effect
We follow the same calculation as Mian and Sufi (2011a) to aggregate the collateral
effect across all counties in the data. We start with the differential impact of house
prices on employment creation at small firms relative to firms with 50 or more employees, and we focus on the specifications where we exclude non-tradable industries
and construction (Table 2.3, Column 2). We first compute predicted county-level employment gains for each establishment size bins in this subset of industries (relative to
the 10th percentile county), and then we aggregate to all counties. Below we describe
each step in detail.
First, we compute the county-level predicted change in employment in each establishment size category by multiplying the regression coefficient by the change in house
prices between 2002 and 2007 in each county. We then subtract the predicted change
in the 10th percentile county in the change in house prices (to avoid being affected
by outliers at the bottom of the distribution). Second, we multiply the predicted
county-level change in employment in each establishment size bin by the employment
in that size bin in each county as of the beginning of the period (2002) to obtain
a predicted change in employment in terms of numbers of workers for each county
and establishment size. Third, we sum up the predicted changes across all counties
and establishment size bins to obtain an economy-wide predicted change due to the
collateral channel in the subset of industries in our preferred specification. Fourth,
and last, we divide the number of employees obtained in step 3 by the share of the
economy made up by the industries included in the specification (for example, 70.8%
of employment is in the industries included in Table 2.3, Column 2).
As an illustration of the calculations, we can take the regression coefficient of 0.315
for size bin 1-4 employees from Column 2 in Table 2.3. Given a change in house prices
of 0.12 in the 10th percentile county, this yields a predicted employment change in
this size bin in the subset of industries in this regression (all except non-tradable and
construction) for the county in the 10th percentile growth in house prices of 3.8%
more than for the size bin 50 and more employees. If we take another county that has
a change in house prices at the median (0.267) the predicted change in that county for
this subset of industries is 0.267*0.315=8.4%. Subtracting the predicted employment
change in the 10th percentile county yields 4.6% predicted change in employment in
the smallest establishment size bin in this county for this subset of industries. We
would then multiply this change by the number of employees in this establishment size
bin in this county and in this subset of industries. When we obtain a total number
of employees by county and bin category, we sum across the four smallest categories
and divide by the share of the economy that is made up by the industries included in
each specification.
We estimate a total job gain in firms with fewer than 50 employees relative to
those with 50 or more employees of 1.698 million jobs in all counties, or 27.8% of jobs
created between 2002 and 2007. This is composed of 600 thousand employees in 1-4
employee establishments, 488 thousand employees in the 5-9 category, 291 thousand
105
for the 10-19 employee bin, and 319 thousand for the bin with 20-49 employees. If
we restrict our attention to the specification where the demand explanation for our
results is the least plausible - that is, the manufacturing sector and, in particular,
firms in industries and states where the shipment distance is largest (Column 6 of
Table 2.4), the same computation would yield an estimate of 676 thousand jobs, or
about 11% of jobs created in this period and subset of counties.
106
Table 2.10: Employment Growth, Firm Size, and House Price Appreciation: Individual Industries by Firm Size
Growth in House Prices
Log of the Population
Percent College Educated
Percent Employed (2000 Census)
Percent of Potential Worker Population
Percent of Homes Owner-occupied
4-Digit Industry Fixed Effects
Number of Observations
R-Square
1-4 Emp
5-9 Emp
10-19 Emp
20-49 Emp
50+ Emp
0.13***
0.11**
(0.05)
(0.05)
0.05
(0.05)
-0.02
(0.08)
0.03
(0.12)
-0.03***
-0.06***
-0.06***
(0.01)
(0.01)
(0.01)
-0.04***
(0.02)
-0.06***
(0.02)
0.00
0.00
0.00
(0.00)
0.00
(0.00)
0.00
(0.00)
(0.00)
(0.00)
0.00
0.00
0.00
(0.00)
(0.00)
(0.00)
0.00
(0.00)
(0.00)
-0.75***
-1.16***
-0.83***
-0.58*
-0.99**
(0.20)
(0.18)
(0.21)
(0.31)
(0.44)
0.00
(0.00)
0.00
(0.00)
Y
61,427
Y
50,381
0.34
0.27
0.00
0.00
0.00
(0.00)
(0.00)
(0.00)
Y
Y
80,915
0.37
Y
71,947
0.37
110,069
0.34
0.00
The table shows two-stage least squares regressions at a county level of employment growth on house price growth
split by size of establishment. All regressions are weighted by the number of households in a county as of 2000.
House Price Growth is instrumented using the Saiz (2010) measure of elasticity of housing supply at an MSA level.
Employment growth is the percentage change in employment between 2002 and 2007 estimated using County Business
Patterns (CBP) data. Growth in House prices is the percentage change between 2002 and 2007, and each interaction
is a dummy indicator for the size of the establishment. All regressions include 4 digit industry fixed effect and control
for log of population, the percentage of the population with a college degree, the percentage of the labor force that
is employed, the share of the population in the workforce and the percentage of homes that are owner occupied. We
drop the top and bottom one percentile of the change in employment in each county, industry and establishment
denote statistical significance at
category. Standard errors are in parenthesis and are clustered by MSA. *, **, *
the 10%, 5%, and 1% levels, respectively.
107
Table 2.11: Robustness Test: Difference between High and Low Start-up Capital
1-4 Emp
5-9 Emp
10-19 Emp
20-49 Emp
50+ Emp
Growth in House Prices
0.23***
(0.06)
0.11*
(0.06)
0.03
(0.06)
0.03
(0.09)
0.01
(0.13)
Growth in HP * High Startup Capital
-0.21***
(0.05)
0.00
(0.06)
0.05
(0.06)
-0.11
(0.07)
0.03
(0.09)
Log of the Population
-0.03***
(0.01)
-0.06***
(0.01)
-0.06***
(0.01)
-0.04***
(0.02)
-0.06***
(0.02)
0.00
(0.00)
0.00
(0.00)
0.00
(0.00)
0.00
0.00
(0.00)
(0.00)
Percent College Educated
Percent Employed (2000 Census)
Percent of Potential Worker Population
Percent of Homes Owner-occupied
4-Digit Industry Fixed Effects
Number of Observations
R2
0.00
0.00
0.00
0.00
(0.00)
(0.00)
(0.00)
(0.00)
0.00
(0.00)
-0.75***
(0.20)
-1.16***
(0.18)
-0.82***
(0.21)
-0.59*
(0.31)
-0.99**
(0.44)
0.00
(0.00)
0.00
(0.00)
0.00
(0.00)
0.00
(0.00)
0.00
(0.00)
Y
110,069
0.34
Y
80,915
0.37
Y
71,947
0.37
Y
61,427
0.34
Y
50,381
0.27
The table shows two-stage least squares regressions at a county level of employment growth on house price growth
split by size of establishment and interacted with a High Startup Capital indicator (indicator itself not shown). High
Startup Capital is defined as 4 digit industries for which the amount of capital to start the firm is higher than the
median for all industries. All regressions are weighted by the number of households in a county as of 2000. House Price
Growth is instrumented using the Saiz (2010) measure of elasticity of housing supply at an MSA level. Employment
growth is the percentage change in employment between 2002 and 2007 estimated using County Business Patterns
(CBP) data. Growth in House prices is the percentage change between 2002 and 2007, and each interaction is a
dummy indicator for the size of the establishment. All regressions include 4 digit industry fixed effect and control
for log of population, the percentage of the population with a college degree, the percentage of the labor force that
is employed, the share of the population in the workforce, and the percentage of homes that are owner occupied.
We drop the top and bottom one percentile of the change in employment in each county, industry and establishment
category. Standard errors are in parenthesis and are clustered by MSA. *, *, *** denote statistical significance at
the 10%, 5%, and 1% levels, respectively.
108
Table 2.12: Effect of One Standard Deviation Change in the Independent Variable
Employment in All Sectors
Effect of 1 sigma change in HP
Growth (02-07)
Employment as of 2002
Employment in Firms <P50 of Start-Up Capital
Effect of 1 sigma change in HP
Growth (02-07)
Employment as of 2002
1-4 Emp
5-9 Emp
10-19 Emp
20-49 Eip
50+ Emp
5.2
9.4
9,101
2.7
8.0
9,122
1.3
12.5
12.819
1.1
10.6
21,466
1.1
13.3
72,939
6.8
10.9
6,213
3.9
11.1
5,566
2.9
13.4
7,350
2.7
14.2
11,012
-0.1
25.0
39,921
4.2
6.6
2,888
2.1
4.3
3,556
-0.1
13.0
5,468
-0.2
9.4
10,453
1.3
9.3
33,018
Employment in Firms >P50 of Start-Up Capital
Effect of 1 sigrma change in HP
Growth (02-07)
Employment as of 2002
The table show effect of one standard deviation change in house prices on employment for different establishment
sizes.
109
Table 2.13: Dollar-weighted Average Distance Shipped in Manufacturing (miles)
Panel A: Summary Statistics
Industry x State
Industry
630.2
368.4
651.7
218.3
25.0
378.1
600.8
168.9
559.3
620.4
Average
Std. Dev.
Percentiles:
1%
25%
50%
75%
817.7
831.7
99%
1,789.2
1,021.3
950
21
Number of Observations
Panel B: Deciles of NAICS and State Dollar-weighted Average Distance Measure
Industry-State Deciles
NAICS
311
312
313
314
315
316
321
322
323
324
325
326
327
331
332
333
334
335
336
337
339
Description
Food Manuf.
Beverage & Tobacco Product Manuf.
Textile Mills
Textile Product Mills
Apparel Manuf.
Leather & Allied Product Manuf.
Wood Product Manuf.
Paper Manuf.
Printing & Related Support Activities
Petroleum & Coal Products Manuf.
Chemical Manuf.
Plastics & Rubber Products Manuf.
Nonmetallic Mineral Product Manuf.
Primary Metal Manuf.
Fabricated Metal Product Manuf.
Machinery Manuf.
Computer & Electronic Product Manuf.
Electrical Eq., App., & Component Manuf.
Transportation Equipment Manuf.
Furniture & Related Product Manuf.
Miscellaneous Manuf.
1
2
3
4
5
6
7
8
9
10
1
15
2
3
1
1
8
2
5
27
2
16
1
2
1
7
8
4
8
2
2
13
7
5
4
1
3
12
4
3
10
3
4
2
1
13
2
4
1
6
4
4
2
8
1
3
3
3
6
2
5
4
4
2
4
6
7
5
4
4
9
13
4
6
5
2
11
8
6
6
2
2
3
1
3
3
3
3
1
8
8
4
8
6
2
7
2
7
7
5
5
2
7
9
5
2
12
10
6
10
3
13
5
6
10
3
15
4
2
9
4
1
16
3
1
2
2
5
12
3
11
10
1
1
20
2
2
1
3
1
4
2
1
1
8
2
2
7
3
9
11
1
2
3
11
8
10
1
5
6
6
1
3
3
8
2
1
9
12
7
7
7
5
5
6
3
5
3
11
1
1
1
1
9
15
10
9
1
10
The table shows the dollar weighted distance of shipments for 3 digit NAICS manufacturing industries. Data is
obtained from the 2007 Commodity Flow Survey. The first column of Panel A shows the weighted average distance
for each industry and state, and the second column aggregates the distances shipped at the 3 digit NAICS level. Panel
B shows the frequency with which each industry appears in each state x industry decile.
110
Table 2.14: Detail on Average Start-up Amount by 2-digit NAICS Sector
Industry
Agriculture, Forestry, Fishing and Hunting
Mining, Quarrying, and Oil and Gas Extraction
Utilities
Construction
Manufacturing
Wholesale Trade
Retail Trade
Transportation and Warehousing
Information
Finance and Insurance
Real Estate and Rental and Leasing
Professional, Scientific, and Technical Services
Management of Companies and Enterprises
Admin. and Supp. and Waste Mgnt and Remediation Svcs
Educational Services
Health Care and Social Assistance
Arts, Entertainment, and Recreation
Accommodation and Food Services
Other Services (except Public Administration)
NAICS2
Average Start-Up
Amount (USD)
11
21
22
23
31
42
44
48
51
52
53
54
55
56
61
62
71
72
81
146,033
673,609
601,149
78,372
363,166
188,085
216,302
131,893
236,126
203,799
220,691
87,879
488,681
91,278
156,893
214,889
218,061
273,186
161,995
Above/Below
Median
0
1
1
0
1
0
1
0
1
0
1
0
1
0
0
0
1
1
0
The table shows the average startup amount by 2 digit NAICS sector used in Tables 2 and 3 in the paper. Data is
from the Survey of Business Owners (SBO) Public Use Microdata Sample (PUMS) using responses to the question
about "Amount of startup or acquisition capital" for each firm with employees in the 2007 survey year.
111
Table 2.15: Distance Shipped and Share of Employees at Large Establishments
Industry-Demeaned Fraction of Employees in
> 50 Employee Establishments (2002), Deciles
1
2
3
4
5
6
7
8
9
10
Industry-Demeaned Distance Deciles
1
10
15
11
5
8
5
9
6
8
16
2
3
4
5
6
7
8
9
10
7
6
3
2
3
2
5
9
10
12
6
3
5 10
5 13 12 16
9
5 10 12 10
6
9 12 11
7 13 11 10 12 11 13
8
9
10 10 11 10 13 17
5
8
7
9
9
9 14
7 17 15
8
6
15 12 17
6
9 12
4
6
7
9 12 14 14
7
5 15
7 10
9 11 10 10 12 11
8
9
5
5
9
4
9 10
6
6 13 11
This table uses the distance measure at the state and 3 digit NAICS manufacturing industry from the 2007 Census
Commodity Flow Survey, and also the share of employment in establishments that have more than 50 employees for
each state and 3 digit NAICS manufacturing industry. For each industry, we compute the average distance shipped,
as well as the average share of employees in firms that have more than 50 employees. Finally, for each state and
industry observation, we compute the deviation from the industry mean for both measures and classify observations
into deciles based on these deviations.
112
-I
0.20
Y
13,482
-
731
0.29
731
0.21
-0.46
(0.35)
-0.45
(0.67)
-0.62
(0.57)
Y
13,482
0.16
0.31
0.22
0.16
(0.35)
731
-0.16
(0.29)
Y
13.482
-0.60
(0.40)
-0.69
Y
6,315
0.20
731
0.27
Y
-0.68
(0.85)
7,167
0.20
(0.49)
-
-0.24
(0.61)
0.00
(0.00)
731
0.29
-0.58
(0.64)
(0.00)
0.00
(0.00)
(0.00)
0.01**
(0.00)
0.00*
0.00**
(0.77)
(0.63)
(0.88)
(0.00)
-1.35*
-2.17***
-2.17**
(0.71)
0.00**
0.00
(0.00)
0.00
(0.00)
-2.43***
0.00
(0.00)
0.00
(0.00)
(0.00)
0.01**
(0.33)
(0.00)
0.00
(0.00)
0.00*
(0.00)
0.00*
-0.01
(0.02)
-1.13***
-1.28***
(0.29)
-0.65
(0.49)
(0.40)
-1.06**
(0.79)
(0.67)
-1.78**
0.00
(0.00)
0.00
(0.00)
0.00
(0.00)
0.00
(0.00)
0.00
(0.00)
0.00
(0.00)
0.00
(0.00)
0.00
(0.00)
-2.34***
0.00
(0.00)
0.00
(0.00)
0.00
(0.00)
0.00*
(0.00)
0.00
(0.00)
0.00*
(0.00)
0.00
(0.01)
Births, Capital > P50
(10)
(9)
0.50***
0.32***
(0.11)
(0.13)
The table shows two stage least squares regressions of establishment births and deaths on house price growth instrumented with the elasticity of housing supply. Each observation is at a county level
for the regressions without sector fixed effects (odd numbered columns) and at a county and 2 digit NAICS industry level whenever we include fixed effects (even numbered columns). All regressions are
weighted by the number of households in a county as of 2000. House Price Growth is instrumented using the Saiz (2010) measure of elasticity of housing supply at an MSA level. Births and deaths of
establishments come from the Census Statistics of U.S. Businesses and are summed between 2002 and 2007 and scaled by the number of establishments in a county as of 2002. Growth in House prices is
the percentage change between 2002 and 2007, and each interaction is with a dummy indicator for the size of the establishment. Columns 1 and 2 shows the results for births of establishments, Columns
3 and 4 show results for disappearance of establishments and Columns 5 and 6 use the net creation of establishments as the dependent variable. The final four columns split the sample by the amount of
capital necessary for starting a business and show results for establishment births. All regressions control for the natural logarithm of population, the percentage of the population with a college degree,
the percentage of the labor force that is employed, the share of the population in the workforce, and the percentage of homes that are owner occupied. All controls are at a county level for the year
denote statistical significance at the 10%, 5%, and 1% levels,
2000 and are obtained using Census Bureau Data Summary Files. Standard errors are in parenthesis and are clustered by MSA. *, *, *
respectively.
2-Digit NAICS Fixed Effects
Number of Observations
R2
China Import Share in County (2005)
Percent of Homes Owner-occupied
Workforce as a Percentage of Population
Percent Employed (2000 Census)
-0.01
-0.01
(0.02)
(0.02)
0.43***
(0.14)
0.57***
(0.13)
0.00
-0.02***
(0.01)
-0.01*
(0.01)
(8)
(7)
Births, Capital < P50
(0.00)
0.01
(0.01)
0.00
(0.01)
0.18***
(0.06)
(5)
0.16**
(0.06)
(6)
Net Creation of Est.
0.01*
Percent College Educated
0.28***
(0.08)
(3)
0.31***
(0.07)
(4)
Deaths of Est.
(0.00)
-0.01
(0.02)
-0.01
(0.01)
(2)
0.46***
(0.12)
Log of the Population
Growth in House Prices
(1)
0.46***
(0.12)
Births of Est.
Table 2.16: House Price Growth and Creation of Establishments
Description
Forestry and Logging
Fishing, Hunting and Trapping
Support Activities for Agriculture and Forestry
Support Activities for Mining
Utilities
Heavy and Civil Engineering Construction
Merchant Wholesalers, Durable Goods
Merchant Wholesalers, Nondurable Goods
Wholesale Electronic Markets and Agents and Brokers
Nonstore Retailers
Air Transp.
Water Transp.
Truck Transp.
Transit and Ground Passenger Transp.
Pipeline Transp.
Scenic and Sightseeing Transp.
Support Activities for Transp.
Couriers and Messengers
Warehousing and Storage
Publishing Ind. (except Internet)
Motion Picture and Sound Recording Ind.
Broadcasting (except Internet)
Internet Publishing and Broadcasting
Telecomnmunications
ISPs, Web Search, and Data Processing
Other Information Serv.
Professional, Scientific, and Technical Serv.
Management of Companies and Enterprises
Administrative and Support Serv.
Waste Management and Remediation Serv.
Educational Serv.
Ambulatory Health Care Serv.
Hospitals
Nursing and Residential Care Facilities
Social Assistance
Performing Arts, Spectator Sports, and Related Ind.
Museums, Historical Sites, and Similar Institutions
Amusement, Gambling, and Recreation Ind.
Accommodation
Repair and Maintenance
Personal and Laundry Serv.
Religious, Grantmaking, Civic Org.
20-49 Emp.
50+ Emp.
5-9 Emp.
10-19 Emp.
1-4 Emp.
0.9%
19.8%
12.7%
17.9%
5.3%
1.5%
5.4%
7.8%
6.8%
26.3%
12.3%
7.3%
6.9%
8.2%
5.7%
2.4%
8.1%
14.6%
9.5%
11.9%
12.6%
1.5%
3.8%,
9.3%
4.6%
11.9%
8.1%
20.5%
5.5%
15.9%
5.9%
6.5%
8.3%
7.5%
0.0%
5.2%
16.5%
10.6%
5.3%
13.3%
11.1%
2.8%
5.9%
6.4%
4.8%
6.4%
6.9%
2.8%
6.9%
5.2%
10.2%
65.9%
64.9%
52.2%
75.1%
86.6%
64.2%
45.6%
61.3%
37.1%
49.2%
92.6%
82.7%
56.7%
74.3%
64.5%
49.6%
57.8%
86.0%
69.3%
78.4%
57.7%
77.1%
68.1%
74.9%
74.7%
49.5%
48.2%
87.2%
73.0%
55.3%
76.7%
39.3%
100.0%
80.9%
43.2%
61.3%
71.9%
61.6%
71.6%
15.4%
20.6%
37.3%
21.4%
15.2%
3.0%
9.7%
4.3%
3.7%
19.3%
7.6%
2.6%
2.8%
3.2%
12.1%
2.4%
7.8%
4.1%
5.9%
7.6%
16.8%
1.3%
5.9%
5.8%
3.2%
10.6%
0.0%
1.2%
5.3%
18.2%
4.7%
4.8%
2.3%
23.1%
19.7%
11.8%
4.3%
8.2%
13.7%
10.8%
8.1%
17.1%
22.7%
16.1%
12.6%
16.6%
4.3%
8.4%
18.8%
13.9%
16.0%
16.3%
17.0%
7.0%
17.8%
10.2%
18.7%
13.4%
12.7%
11.0%
10.9%
21.6%
14.9%
7.2%
11.2%
22.2%
12.0%
19.2%
0.0%
9.4%
28.8%
9.3%
12.4%
20.5%
16.5%
18.9%
18.8%
22.4%
2.6%
7.4%
8.0%
3.2%
1.5%
5.2%
9.3%
6.3%
12.1%
9.4%
0.6%
2.1%
5.5%
2.8%
3.9%
6.8%
7.4%
1.6%
3.4%
3.0%
4.5%
2.3%
5.0%
3.6%
3.2%
8.0%
9.0%
1.5%
4.1%
6.1%
2.9%
14.5%
0.0%
3.0%
6.8%
5.3%
4.4%
4.7%
2.1%,
22.1%
19.5%
13.3%
Table 2.17: List of 3-digit NAICS Industries Excluding Non-tradables, Manufacturing, F.I.R.E., and Construction
NAICS
113
114
115
213
221
237
423
424
425
454
481
483
484
485
486
487
488
492
493
511
512
515
516
517
518
519
541
551
561
562
611
621
622
623
624
711
712
713
721
811
812
813
The table shows the 3 digit NAICS codes, as well as the proportion of employees in each establishment size category and the total number of employees in each industry in our sample of counties.
Chapter 3
Credit Supply and House Prices:
Evidence from Mortgage Market
Segmentation
3.1
Introduction
One of the central debates in finance focuses on the impact of the cost of funding on
the level of asset prices (see, e.g., Brunnermeier, Eisenbach and Sannikov, 2012). A
salient recent example is the US housing market: many observers of the 2008 financial crisis have proposed that reduced cost of credit was the central factor fueling the
increase in housing prices as well as the subsequent reversal (Hubbard and Mayer,
2008; Mayer, 2011). Others have argued that cheaper credit alone cannot explain the
bubble (Glaeser, Gottlieb, and Gyourko, 2010) and that other factors must have also
been at play, including a reduction in collateral constraints (Favilukis, Ludvigson,
and Van Nieuwerburgh, 2010; Khandani, Lo, and Merton, 2009), financial innovation (Mian and Sufi, 2009; Calomiris, 2009; Pavlov and Wachter, 2011), or market
sentiment and expectations about future appreciation (Shiller, 2008).
The key difficulty in measuring the effect of the cost of credit on the price of
housing is establishing the direction of causality between cost of funding and house
price growth: On the one hand, cheaper credit is likely to reduce borrower financing
constraints and increase total demand for housing, which in turn would lead to higher
prices. On the other hand, however, credit conditions in general might be responding
to expectations of stronger housing demand and, as a consequence, higher house
prices. In this latter scenario, cheaper credit is not the driver of house price increases,
but a byproduct of increased demand for housing, since housing as collateral becomes
more valuable. As we see in the existing literature, it has been very difficult to
separate these two effects. 1
In this paper, we develop a new instrument that uses annual changes in the con'A recent paper by Favara and Imbs (2012) uses branching deregulation in the 1990s to identify the causal link between credit supply and house prices and finds that states where there is
deregulation subsequently experience larger house price increases.
115
forming loan limit (CLL) as exogenous variation in the cost of credit, which allows
us to provide clean estimates of the effect of lower cost of credit on house prices. The
CLL determines the maximum size of a mortgage that can be purchased or securitized by Fannie Mae or Freddie Mac. Mortgages below the CLL therefore have lower
interest rates compared to jumbo loans (loans that are above the CLL), since the
former benefit from implicit (and since 2008, explicit) government support for Fannie
Mae and Freddie Mac. The difference in interest rates between conforming loans and
jumbo loans has been estimated to be up to 24 basis points.2 . In addition, Loutskina
and Strahan (2009, 2011) show that more borrowers are able to access mortgages
below the conforming loan limit than above, which suggests that not only the cost of
credit it lower below the CLL, but also access to credit itself might be easier.
The underlying idea of our identification strategy is that changes in the conforming
loan limit (CLL) from one year to the next are exogenous to local housing markets and
the local economy, since this change is based on the national average appreciation in
house prices. That means that, in a given year, a house just above the CLL threshold
has to be financed by an expensive jumbo loan, while the next year the equivalent
house can be financed via cheaper conforming loan. Our empirical approach involves
comparing transactions that can be financed more easily using a conforming loan, and
houses that are more expensive so that buyers need to obtain larger (jumbo) loans to
maintain the same loan-to-value ratio. We track transactions in the price range just
above and below the CLL in the year that the limit is in effect and compare them to
the subsequent year, once the limit is raised and houses just above the CLL become
eligible for conforming loans. This setup enables us to cleanly identify the effect of
the cost of credit and control for any overall trends in house prices.
The threshold that we use to define houses that are "cheap" to finance with a
conforming loan in a given year is obtained by dividing the conforming loan limit by
0.8.' By construction, buyers of houses with a price below this threshold can get a
conforming loan that makes up 80 percent of the price of the house, whereas if the
price of the house is above 125 percent of the CLL, it can no longer be financed at
80 percent with a conforming loan. Loans with a loan-to-value (LTV) ratio below
80 are associated with more attractive terms, and conforming loans above 80 percent
require private mortgage insurance in order to qualify for purchase by Fannie Mae
or Freddie Mac (Green and Wachter, 2005). Above this price threshold, borrowers
either finance their home with an 80 percent first mortgage using a jumbo loan (i.e.
a loan above the CLL) at a higher interest rate, or, if they want to take advantage
of the lower interest rate below the CLL, they have to use savings or alternative
forms of financing to make a larger down payment. Importantly, our sample includes
all transactions in this price range independent of financing choice of each borrower.
This allows us to eliminate any bias due to the endogenous choice of financing of a
2
See for example McKenzie (2002), Ambrose, LaCour-Little, and Sanders (2004), Sherlund
(2008), Kaufman (2012), or DeFusco and Paciorek (2013)
3
Kaufman (2012) uses this threshold for appraisal values to study the effect of the conforming
status of a loan on its cost and contract structure. Loutskina and Strahan (2013) follow our approach
and use changes in the CLL interacted with regional constraints to look at financial integration and
the propagation of shocks.
116
specific transaction. An example of such a bias would be that richer people who can
afford to put more money down might also purchase houses that are more expensive
based on (unobservable) quality dimensions. Our instrument eliminates this type of
concern.
We first document that the conforming loan limit (CLL) impacts borrowers' choice
of financing. The data shows that the norm in the mortgage market during this period
was to borrow at an LTV of exactly 0.8 (on average 60 percent of transactions are
at an LTV of 0.8). However, for houses that transact just above 125 percent of
the CLL, a much larger fraction of purchases are at an LTV below 0.8, since many
borrowers choose to take out a mortgage to exactly max out the conforming loan
limit. Borrowers that buy houses with a price above the threshold have a higher
funding cost than borrowers who buy houses at a price below 125 percent of the CLL,
since they either have to take a jumbo loan or use a conforming loan and finance the
rest of the house price with other forms of financing.
In our main analysis, we measure the causal effect of cheaper credit on house
prices instrumented via the change in the conforming loan limit from one year to the
next. We run differences-in-differences regressions in which we compare transactions
just above and just below the threshold of 125 percent of the CLL in the year that
the limit is in effect, and in the subsequent year when all of the transactions can
obtain an 80 percent conforming loan. 4 We use three different dependent variables
to capture the value of a property: (1) the value per square foot; (2) the residuals
of log of house prices from a hedonic regression using a large set of controls for the
underlying characteristics of the house, and (3) the residuals of the value per square
foot from similar hedonic regressions. 5
We find that transactions just above 125 percent of the CLL, i.e. in the "high
cost" group of borrowers, are made at lower values per square foot than those for
the unconstrained group. We see a 1.16 dollar discount per square foot for a mean
value per square foot of 220 dollars (i.e., about 53 basis points of the average house
value). This difference is reduced to 0.65 dollars per square foot (30 basis points)
after we control for house characteristics, suggesting that part of the effect we find
can be accounted for by differences in the observable quality of houses above and
below the threshold. These effects are significantly different from those we obtain
when we use "placebo" loan limits elsewhere in the distribution, which confirms that
we are picking up a cost of credit effect of the CLL. The effect is smaller (and often
insignificant) in the second half of our sample (2002-2005), which is the period when
4
This is the case for all years between 1998 and 2005. For example, the CLL in 1999 is USD
240, 000, which gives a threshold of USD 240, 000/0.8 = 300, 000 for this year. This means that in
the regression for 1999, we include houses priced at between 290, 000 and 310, 000 in the years of
1999 (the year the CLL is in effect) and 2000. The CLL in the year 2000 was raised to 252, 700, so
the new threshold for that year is 315,875. Clearly, all the houses we included in the analysis for
1999 can be financed at 80 percent with a conforming loan in the year 2000.
'We run the hedonic regressions by year and by metropolitan statistical area (of which we have
10) and we use the set of controls available from deeds registries data, which includes common
variables such as number of rooms and number of bedrooms, but also detail on the type of heating,
architectural type, building type, among many others (we discuss these controls in more detail in
Section 3.3.2).
117
jumbo loans became cheaper and easier to obtain (partly due to the increased ease
with which they could be securitized) and also when second lien mortgages became
widely available (see Figure 3-5 ). Both these effects reinforce the idea that when the
CLL was more important in the earlier part of the sample, its impact is also more
significant in our estimates.
Given our estimate for the change in house prices due to changes in credit conditions, we can compute the semi-elasticity of prices to differences in interest rates in
the region close to the threshold. We use the differences in interest rates estimated
in the prior literature of 10 to 24 basis points between conforming and jumbo loans
as our measure of the cost differential for buyers above and below the threshold. We
obtain local elasticity estimates that range from a low end of 1.2 to an upper range of
9.1 depending on the period and the exact estimate for the interest rate differential
between jumbo and conforming loans that we use for our calculations. These elasticity estimates are at the lower end of what has been previously found in the literature,
and they imply that the 55
We next investigate the cross-sectional heterogeneity of our elasticity estimates
by focusing on whether the effect of cheaper credit is stronger when buyers face other
types of constraints at the same time, as proxied for by lower income. Specifically,
we interact the changes in the CLL with whether a zip code and year is below the
10th percentile of the income growth distribution for each individual regression. The
point estimate for these areas shows that value per square foot is 2.50 dollars higher
in the year that a house becomes eligible to be financed with a conforming loan. This
is more than double the size of the average elasticity that we found in the overall
sample, suggesting that cheaper credit may have had a disproportional impact on
economically more depressed households and regions.
We show that our results are not driven by a subsidy effect that provides a focal
point to draw in more bidders. First, there is no visible bunching in the number of
transactions just around the threshold of CLL divided by 0.8, suggesting that the
supply of housing does not react strongly to the CLL. We also do not find that there
is bunching in the number of unobserved bidders for homes around the CLL, which we
measure as the share of borrowers that apply for loans but ultimately either withdraw
or do not use the loans they are approved for. If the CLL served simply as a focal
point for home sales, we should expect more bidders for homes that are eligible for
conforming loans. Instead, we find that our measure of the share of failed bids is
lower, not higher, for borrowers that borrow up to the CLL. The fact that there is
neither a significant jump in the quantity of transactions nor in our proxy of failed
offers for homes suggests that the effect we find on prices is more consistent with a
cost of credit interpretation.
The rest of the paper is structured as follows: Section 3.2 discusses related literature and the user cost model. Section 3.3 describes our data and the identification
strategy. In Section 3.4, we lay out the regressions results and robustness checks of
our main analysis. Section 3.5 discusses the findings and concludes.
118
3.2
The User Cost Model
In this paper, we are interested in estimating the impact of changes in the cost
of credit on the price of housing. The existing literature has focused on different
versions of the user-cost model of Poterba (1984) to draw conclusions about the role
of interest rates and other costs of owning for house prices. In this model, agents
are indifferent between owning and renting if the housing market is in equilibrium,
where the mortgage interest rate is the main determinant of the cost of owning. The
existing literature shows that different assumptions yield very different conclusions
about the role of interest rates in driving the cost of housing and highlight why our
estimate of the impact of the cost of credit on prices is an important contribution to
this debate.
We follow the notation in Glaeser, Gottlieb, and Gyourko (2010) to describe the
basic elements of the user cost model. Renting a property involves paying rent equal
to Rt in each period. Owning a property, on the other hand, includes making a downpayment 0 that is a proportion of the price of the house Pt and obtaining a mortgage
that is rolled over each period, such that principal is never paid down completely. The
borrower pays interest on the mortgage at a rate rt that is deflated by the relevant
tax rate #, as well as property taxes and maintenance costs equal to T that both grow
at a rate g. The model assumes that individuals have a private discount factor of pt.
If we assume that market interest rates and private discount rates are constant and
equal to each other, we can write the indifference condition for users as:
= (1 -#)r - g+
(3.1)
Pt
This is shown in Glaeser et al (2010) and is similar to what is presented in Hubbard
and Mayer (2008) as well as a simplified version of the user cost in Himmelberg, Mayer,
and Sinai (2005). If the assumptions of this model hold, changes in the user cost (the
right-hand side of the equation) should lead to changes in the price to rent ratio. For
example, if the user cost is 5 percent, then the price of a house should be about 20
times its market rent. In such a world, a drop of 1 percentage point in mortgage rates
would lead to a decline of (1 - r) in the user cost, or 0.75 if we assume a marginal tax
rate of 25 percent. The price to rent ratio would then be 23.5, an increase in the price
of 17.5 percent. This is the magnitude of the elasticities proposed in Himmelberg et
al (2005), and in Hubbard and Mayer (2008).
Glaeser, Gottlieb, and Gyourko (2010) dispute some of the simplifying assumptions in the model above, and show that a more realistic model can produce much
lower elasticities of prices to interest rates. In particular, if private discount rates
are not the same as market rates, changes in interest rates wont alter the way users
discount future expected house price appreciation. Glaeser et al (2010) show that
this change alone can reduce the elasticity to just 8, instead of the initial 17.5. Other
mechanisms through which the elasticity could be substantially reduced include mean
reverting interest rates, which means borrowers anticipate having to sell a home at a
time when rates are higher, or the possibility of prepaying a mortgage. Our econo-
119
metric approach allows us to more carefully identify the magnitude of the change in
house prices due to changes in the average cost of financing, since we look at exogenous movements in the cost of capital for home buyers. Our empirical results provide
local estimates for the numerator of the elasticity calculation. In Section 3.4.4, we
discuss the range of elasticities that are consistent with our results.
3.3
Data and Methodology
The dataset we use in this paper contains all the ownership transfers of residential properties available in deeds and assessors records over 11 years, from 1998 to
2008, and seventy-four counties in ten metropolitan statistical areas (MSAs) - Boston,
Chicago, DC, Denver, Las Vegas, Los Angeles, Miami, New York, San Diego, and San
Francisco. We limit our attention to transactions of single-family houses, which account for the large majority (approximately 78 percent) of all observations.
Each observation in the data contains the date of the transaction, the amount for
which a house was sold, the size of the first mortgage, and an extensive set of variables
about the property itself. These characteristics include the property address, interior
square footage, lot size, number of bedrooms, number of bathrooms, total rooms,
house age, type of house (single-family house or condo), renovation status, and date of
renovation. Additional characteristics include the availability of a fireplace, parking,
the architectural and structural style of the building, the type of construction, exterior
material, availability of heating or cooling, heating and cooling mechanism, type of
roof, view, attic, basement, and garage. We describe the procedure for cleaning the
raw data received from Dataquick in the Appendix to the paper.
3.3.1
Summary Statistics
The dataset that we use for this paper contains 3.98 million transactions of singlefamily houses that are summarized in Tables 3.1 and 3.2.6 We can see in Panel A of
Table 3.1 that the average transaction value is 309 thousand dollars with a standard
deviation of 124 thousand dollars. The average size of the houses is 1,735 sqft, and
the houses have, on average, 3 bedrooms and 2 bathrooms. The average loan to
value is 0.81 (including only the first mortgage for each transaction), and the median
LTV is 80 percent. The average value per square foot is 194 dollars with a standard
deviation of 92 dollars per square foot (first row of Panel B).
Table 3.1 also shows the summary statistics for the sample we use in the regressions
in the final three columns. For the regression sample, the average price for each
house is higher than in the whole dataset (at 371 thousand dollars, compared to 309
thousand in the first column). This is consistent with the fact that the conforming
loan limit was set to cover substantially more than 50 percent of the mortgages made
every year (Acharya, Richardson, Nieuwerburgh, White, 2011). These houses are also,
on average, larger and have more bedrooms and bathrooms than the whole dataset.
6
Please see the Appendix for a detailed description of the procedure for cleaning the data initially
obtained from Dataquick and how we arrive at the 3.98 million observations.
120
Panel A of Table 3.2 shows marked differences in the summary statistics for each
of the ten MSAs included in our data. The table shows that San Francisco is the
metropolitan area with the highest valuation, with an average house price of 384
thousand dollars. Denver and Las Vegas represent the areas with the lowest valuation,
with an average of approximately 250 and 262 thousand dollars respectively. When
we compare values per square foot, we get a similar picture, namely San Francisco is
the area with the highest valuation with an average of 266 dollars per square foot,
and Las Vegas is the area with the lowest valuation with an average of 137 dollars
per square foot.
Table 3.2 Panel B shows the evolution of prices through time. Here we see the
increase in house prices from an average of 240 thousand dollars in 1998 to a peak
of 366 thousand dollars in 2006, as well as the increase in the volume of transactions
over the same period. The increase in prices and volume is linked to an increase in
volatility. The standard deviation of the transactions increased from 102 thousand
dollars in 1998 to 122 thousand dollars in 2006. A similar pattern can be observed for
the value per square foot measure, where standard deviation is 51 dollars per square
foot in 1998, and increases to 106 dollars per square foot in 2006. Finally, the loan
to value average (including only the first mortgage) is stable both across MSAs and
through time at around 0.8.
3.3.2
Hedonic Regression
One of the advantages of using deeds registry data is the richness of the information
provided on the property characteristics, which allows us to account for price differences between houses that can be attributed to observable features. Specifically, we
will be able to assess whether the price impact we observe due to the changes in the
conforming loan limit can be attributed to differences in the quality of the houses, or
whether these differences are there even after accounting for quality.
In order to distinguish between these two explanations, we estimate hedonic regressions of value per square foot and log of house price on a number of house characteristics, and estimate the residuals for each of these two left-hand side variables
(which we denote by LHSi). Specifically, we estimate the following regressions by
MSA and by year:
LHS, = -yo + PX + monthi + zipcodei + Ej
We use both the logarithm of the price of a transaction as well as the value per
square foot as our dependent variables. By estimating these regressions by year and
by MSA, we allow the coefficients on the characteristics to vary along these two
dimensions. We also use monthly indicator variables to account for seasonality in the
housing market, as well as zip code fixed effects. The set of controls Xi is a similar set
of controls to that used in Campbell, Giglio, and Pathak (2010) with some additional
characteristics. The controls include square footage, high and low square footage
dummies, the size of the lot, number of bedrooms and bathrooms, and a number of
indicators for interior and exterior house characteristics (eg. fireplace, style of the
121
building, etc.). We describe which variables are included, as well as the detail of the
construction of each variable, in the Appendix to the paper.
The estimated R2 of each of these regressions (80 in total for each of the two
left-hand side variable-10 MSAs in 8 years) is between 40 and 60 percent for the
price of the transaction, and 50 to 70 percent when we use value per square foot as a
dependent variable.
Summary statistics for the residuals from the hedonic regressions for the whole
sample are shown in Panel B of Table 3.1. The average residuals are, by construction,
zero. The standard deviation of the errors is about 42 dollars per square foot, and
0.17 thousand dollars for the log of the price of the house. The hedonic regressions
are estimated on the whole dataset of transactions (the 3.98 million observations
mentioned above), so when we restrict our attention to the regression sample, the
average error no longer has to be zero. Indeed, for the regression sample, the average
residual from the hedonic regressions for the value per square foot is positive at 5.3
dollars, and the average error for the log of transaction value of the house is 0.05
dollars (last three columns of Panel B of Table 3.1). The standard deviation of the
residuals for the regression sample is similar in magnitude to what we obtain for all
the transactions.
3.3.3
Empirical Approach
Identification Strategy
To identify the effect of changes in credit conditions on house prices, we restrict our
analysis to two groups of buyers who all buy houses in a tight price range, but differ
in the financing available to them. The sample for our regressions is made up of
houses that transact in a band around 125 percent of each year's conforming loan
limit, as well as houses in the subsequent year in the same price range. Specifically,
we divide houses into two groups: houses below the threshold of 125 percent of the
year's CLL (i.e. transactions that fall between 125 percent of CLL and 125 percent
of CLL minus USD 10,000) and houses above that threshold that transact between
125 percent of CLL and 125 percent of CLL+10, 000. By construction, in the year
that the conforming loan limit is in effect, houses above the threshold of 125 percent
of the CLL cannot be financed at 80 percent using a conforming loan, whereas the
houses below the threshold can be financed. Thus, home buyers that bid for houses
priced above 125 percent of CLL cannot finance a full 80 percent of the transaction
with the cheaper and more easily available conforming loans. In the subsequent year,
the CLL is raised and both groups of transactions can be financed at 80 percent
with a conforming loan. 7 Our sample includes all transactions in this price range,
independent of the mortgage choice made by each buyer. This way, our estimates are
not biased by the endogeneity of the choice of financing of each specific transaction.
The identification strategy is best understood through an example. Consider the
7
While this was no longer true for the years after 2006, in all cases between 1998 and 2005, the
limit increases enough from year to year to make up 80 percent of the price of the transactions we
have in the sample.
122
year 1999: In that year, the conforming loan limit (CLL) for single-family houses
was USD 240, 000. The corresponding threshold for house prices that we use for
this year is 300, 000 (240, 000/0.8 or, equivalently, 1.25 * 240, 000). In this year, the
group of houses "above the threshold" have prices between USD 300, 000 and USD
(300, 000 + 10, 000) = 310, 000 and houses "below the threshold" have a transaction
price between USD (300, 000 - 10, 000) = 290, 000 and USD 300, 000 (those that
transact at exactly USD 300, 000 are included in this second group). For the purposes
of our main regressions, we track these two groups of houses from 1999 to 2000, where
1999 is the year in which the CLL is in effect and 2000 is the year in which all these
transactions could be bought using a conforming loan at a full 80 percent LTV. In
fact, the CLL changed in 2000 to USD 252, 700, so the threshold of 125 percent of
CLL was now USD 315,875 and even our "above the threshold" group for 1999 is
now eligible to get an 80 percent LTV conforming loan.
One important assumption in our analysis is that borrowers in the group "above
the threshold" of 125 percent are constrained in their choice of financing. In order
to stay at an LTV of 0.8, they have to take a jumbo loan and these have been
found to be more expensive by between 10 and 24 basis points relative to conforming
loans (McKenzie, 2002; Ambrose, LaCour-Little, and Sanders, 2004; Sherlund, 2008;
Kaufman, 2012; DeFusco and Paciorek, 2013). Alternatively, they can also borrow
up to the CLL and then cover the rest of the house price with savings or other
funding, which means having a first mortgage LTV of less than 80 percent. This
additional source of funding is likely substantially more expensive relative to the
conforming mortgage rate. For some borrowers, this may, in fact, be the only option,
as they may be excluded from the jumbo market altogether because of more careful
screening of jumbo loans done by originating banks (Loutskina and Strahan, 2009,
2011). Whether they choose a jumbo loan or they make up the difference using other
sources of financing, these borrowers have a higher average cost of capital than the
buyers below the threshold.
As Figure 3-1 shows, the most frequent choice on the part of borrowers is to have
an LTV of exactly 80 percent (that is, the large mass along the diagonal of the figure).
The main exception to this rule occurs exactly at the conforming loan limit, where a
significant mass of borrowers chooses an LTV below 0.8 by sticking to a conforming
loan (in 2000 the limit was USD 252,700, and in 2004 it was 333,7000).
The data
shows that in the year in which the CLL is in effect, about 45 percent of the houses
below the threshold in our sample are bought with an LTV of exactly 80 percent,
whereas for houses above this boundary just 19 percent of borrowers pick 80 percent
LTVs (which for these transactions means using a jumbo loan). Additionally, on
average 55 percent of the transactions just above the threshold are financed using a
conforming loan, which means having an LTV lower than 80 percent. These borrowers
end up with an LTV of 77-79.5 percent, which is a very infrequent choice anywhere
else in the distribution. Again, these borrowers might have a lower LTV because they
choose to stay below the CLL due to the cost of the loan, or because they are excluded
from the jumbo market altogether. Whatever the reason, this group of borrowers is
"constrained" in the set of options available for financing their house.
123
Empirical Specification
Our main regressions estimate the size of the effect of the constraint imposed by the
conforming loan limit on the valuation of transactions made just above the threshold
of 125 percent of the CLL. We run differences-in-differences regressions year-by-year
with one indicator variable for houses priced above the conforming loan limit divided
by 0.8, another indicator for the year in which the CLL is in effect, and an interaction
of these two indicator variables. We also include ZIP code fixed effects in all regressions, so our estimates do not reflect differences between neighborhoods, but rather
variation within zip codes.
The sample for each year-by-year regression includes houses within a USD 10,000
band around the conforming loan limit in the year in which the limit is in force, as
well as the subsequent year. This implies that the "Above the Threshold" indicator
variable takes a value of 1 if the price at which a house transacts is greater than 125
percent of the conforming loan limit of a certain year, and less than that amount plus
10,000 dollars. This same variable is a 0 for transactions between 125 percent of the
CLL and 125 percent of the CLL minus 10,000 dollars. The "Year CLL" indicator
variable is a 1 in the year in which the CLL is in effect for each regression, and
a 0 in the subsequent year. We use a tight band around the threshold so that all
transactions in the year after the limit is in effect are eligible for an 80 percent LTV
conforming loan. We thus have a group of transactions that is "easy to finance and
another one that is "hard to finance in the year that the limit is in effect, but all
transactions in the sample are "easy to finance once the limit is raised.'
We run regressions of the following form:
Valuation measurei =
/31Above
N
+
/11AboveThreshold
ThresholdxYear-CLL
+
+ YZIP +
/ 3 2lYearCLL±
Ei
We estimate this regression for each year between 1998 and 2005. We cannot
include 2006 and 2007 in our estimates because the conforming loan limit did not
change after 2006 in our data (house prices dropped and the administration left the
limit unchanged). 9 After we obtain 01, /2, and 03 for all 8 years (1998-2005), we
estimate Fama-MacBeth averages (Fama and MacBeth, 1973) of these coefficients
and obtain the standard errors of this average by using the standard deviation of
the estimated coefficients and dividing it by the square root of the number of coeffi8
An alternative way to run our test would be to compare the year in which each limit is in effect
with the previous year, when all transactions in this range would be above the threshold for that
year. The results for this alternative specification are reported in the Appendix.
9
We do not run our analysis on the changes that were made to conforming loan limits in 2008 in
high-cost areas as part of the Housing and Economic Recovery Act of 2008 for two reasons: First, the
limit was chosen by the government, as opposed to being mechanically related to previous limits, so
this introduces the possibility that the "jumbo-conforming" program was designed to assist specific
areas and thus would be endogenous to expected future appreciation. Second, to the best of our
knowledge, there is no empirical evidence that the program had any discernible impact on the cost
of funding of mortgages that were made between the old limit of USD 417,000 and the new, higher
limits.
124
cients. We test the robustness of our results to serial correlation in the error term by
constructing Newey-West standard errors, and all the results are unchanged.
We should point out that our approach is not a regression discontinuity design,
but rather differences-in-differences for each pair of years. There are a couple of
reasons for this: First, the threshold that we use does not imply a sharp discontinuity
in the ease of financing a home. For a house just one dollar above the threshold,
a homebuyer only has to come up with one additional dollar of equity (and still
obtain a conforming mortgage), which means the total cost of financing the house is
almost unchanged. As we move progressively away from the threshold, transactions
become harder to finance. For our differences-in-differences estimator to be valid, all
we need is that houses above the threshold are somewhat harder to finance, though
not necessarily discontinuously so.
The second reason for not using a regression discontinuity design is that in the
year that the limit is in effect, homebuyers choose to buy houses above or below
the threshold, i.e. the position with respect to the limit is not exogenous. On the
contrary, our differences-in-differences specification uses the exogenous change in the
conforming loan limit to compare a group of transactions that are above the limit
in a year, but below in the next with a group of transaction that are always below
the limit, achieving a clean identification of the effect of credit availability on house
prices.
Our estimation strategy allows us to estimate the causal effect of changes in the
cost of credit on the valuation of houses. Since house price levels differ across the
various states of the United States, the change in the CLL affects different parts of
the housing stock across areas depending on the price level of the area. Using this
instrument we can account for the possibility that there are differential growth rates
within the distribution of house types across the country. For example, one concern
would be that middle class families might buy a certain type of house and, at the
same time, have a different income growth from other parts of the population. Our
instrument allows us to rule this out, because the same "type" of house will have
different prices depending on where it is located in the country.
Finally, we can rule out that selection effects are driving our results: one could
worry that buyers of houses "above the threshold" in the year that the conforming
loan limit is in effect are different along some unobservable dimensions from the other
buyers. Several features of our analysis make selection an unlikely explanation of the
results. First, for a selection hypothesis to be a true alternative to our explanation,
it would have to involve arguments other than cost of credit to explain why buyers
were different above and below the threshold. Second, these "special" buyers would
both have to be better able to deal with the higher cost of credit (potentially because
they are wealthier or have higher income) and bargain harder for houses. It is unclear
why wealthier borrowers should pay less for a similar house than poorer borrowers.
If wealthier people bought higher quality houses and we did not observe these differences, these unobservable characteristics would create bias in the opposite direction.
Third, our identification strategy would require that the selection effect change each
year parallel to the change in the size of the conforming loan limit, which is very
unlikely. Lastly, to further alleviate any concerns about selection, we run our main
125
regressions excluding borrowers that choose LTVs below 80 percent in the year that
the CLL is in effect. If selection was the explanation of the results, these transactions
should be by "wealthy" borrowers driving the results. We find that the results do not
change materially when we exclude this subset of transactions.
Differences in Financing Choices
As we pointed out above, the equivalent to a first stage in our empirical strategy
is to show that the changes in the conforming loan limit have a significant effect
on the financing choices of borrowers. In Figure 3-1 we can see the importance of
both the 80 percent LTV rule, as well as the conforming loan limit, in determining
financing choices for the whole distribution of transactions. In Figure 3-2 we focus on
the groups of transactions that we include in the regressions. The first panel tracks
transactions up to USD 10,000 below 125 percent of the conforming loan limit in
each year, whereas the second panel includes transactions up to USD 10,000 above
the threshold. We show the total number of transactions (for all years between 1998
and 2006) in each month during the year prior to the limit being in effect, in the
year that the limit is valid, and in the subsequent year. We also break down the
transactions by the choice of LTV - the transactions at the bottom of each panel
have an LTV below 75 percent, the second group includes transactions with an LTV
between 75 percent and 79.5 percent, the third has transactions with LTV=80 percent,
and the top group has all the transactions with an LTV above 80.1 percent. The main
message from Figure 3-2 is that in the year that the CLL is in effect, the composition
of financing choices by borrowers differs very significantly, with the 80 percent group
becoming very prominent for the transactions below 125 percent of the CLL, whereas
it is small for the transactions above the threshold. At the same time, the borrowers
who stick with a conforming loan and buy houses above 125 percent of the CLL
become an important fraction of all borrowers (they have an LTV between 75 and
79.5 percent).1 0 In the year after the limit is in effect, the choice of LTV across the
two groups becomes indistinguishable.
In Table 3.3, we present the effect of the changes in the conforming loan limit
on the financing choices made by the borrowers included in the sample of our main
regressions. In this table, we are verifying what we see in the pictures, namely that
borrowers on average end up with lower LTVs when they buy houses above the threshold of 125 percent of CLL. We find that LTVs are, on average, 0.3 to 0.7 percentage
points lower for the group of transactions that happen above the threshold of 125
percent of the CLL in the year that the limit is in effect. This effect is statistically
and economically significant given how little variation there is in the modal choice of
LTV of borrowers. The second panel on Table 3.3 shows that borrowers also obtain,
'OThe first picture for the group below 125 percent of the CLL also shows a noticeable fraction
of borrowers with an LTV between 75 and 79.5 percent in the year before the CLL is in effect. This
is because these transactions were not eligible for a conforming loan at an 80 percent LTV in the
year before the new limit was in effect and were, in general, just slightly above that threshold. This
is thus a reflection of the same phenomenon we see for the group above 125 percent of the CLL in
the year that the new limit is in place.
126
on average, smaller loans in the year that the limit is in effect and when the price of
the house is above the threshold. The difference in log loan amount is, on average,
0.0056 to 0.0088 dollars, and based upon the findings in our main results, we conjecture that it is the fact that borrowers obtain smaller first mortgages that leads to
the difference of approximately 1.16 dollars per square foot (for an average value per
square foot of 220 dollars).
Differences in the Number of Transactions
There are several reasons to expect quantities to change due to differential cost of
credit, including different levels of down-payment (Stein, 1995) or sellers waiting for
buyers to obtain better credit conditions (Genesove and Mayer, 1997). In fact, unless
the supply elasticity of houses is very low (or zero), we expect the price effect due to
a change in the demand for housing to be accompanied by a change in the number of
transactions.
As discussed in Section 3.3.3, we do not use a regression discontinuity approach
to address the question of the change in the quantity of transactions. Figure 3-3
confirms that this would produce no significant result. This figure shows the number
of transactions relative to the threshold in each year. The figure is centered at 0, i.e.
the transactions at exactly 125 percent of the CLL. The figure shows that there is no
discontinuity in the number of transactions above and below the threshold.
Given that a regression discontinuity would not be appropriate in our setting,
we use a setup similar to our main regressions to look for changes in the number of
transactions above and below the threshold. We consider the difference in the share
of transactions in our sample that fall above and below the threshold in the year that
the limit is in effect and in the subsequent year in a differences-in-differences setup.
This test is equivalent to a T-test for the mean of the variable "Above Threshold"
that compares the average of this variable in the year that the limit is in effect and
in the subsequent year. If our instrument affects the quantity of transactions, we
should see an increase in the share of observations above the threshold when the limit
is raised, as credit becomes cheaper for those transactions. We show in Table 3.4
that this test reveals no changes in the share of transactions above and below the
threshold for the first part of our sample (1998-2001), and that there is a statistically
significant effect for the second part of the sample. This translates into a share of
transactions above the threshold approximately 60 basis points lower in the year that
the conforming loan limit is in effect during the period 2002-2005. This regression
shows that cheaper credit provided by conforming loans is reflected only on house
prices in the first part of our sample, and that in the second part of the sample, it
impacts both quantities and prices, i.e. local supply elasticity of houses seems to
have been higher in the second part of the sample. This, along with the reasons we
give in Section 3.4.1 on the availability of second liens and jumbo loans, may help
explain why the effect we find on prices is smaller relative to the earlier years (when
the quantity response is not there).
127
3.4
Cost of Credit and House Prices
3.4.1
Main Regression Results
We present the results for our canonical specification in Table 3.5. This table presents
Fama-MacBeth coefficients from year-by-year regressions, as described before in Section 3.3.3. The coefficient of interest in Panel A of Table 3.5 is that on the interaction
variable, and it shows that houses above the threshold of CLL/0.8 transacted at a
value per square foot that was lower by about 1.16 dollars in the year that the CLL
was in effect. The results are stronger for the first half of the sample, where the point
estimate is -1.55 dollars per square foot for this set of transactions.
The other coefficients on the regressions for value per square foot are consistent
with what we know about house prices over this period. First, houses that are above
the threshold of 125 percent of CLL (i.e. the more expensive houses in the regression
sample) are associated with a higher average value per square foot. In unreported
analyses, we find that more expensive houses are generally associated with a higher
value per square foot (i.e. price rises quicker than house size in the whole distribution
of transactions), and here we find that this is also the case for the regression sample.
Also, the "Year CLL" dummy variable is associated with a strong negative effect,
reflecting the strong increase in house valuations that we saw in this period in the
US. Given that the year in which the CLL is in effect is always the "pre" year in the
regressions, we expect those transactions, on average, to be associated with a lower
value per square foot.
In Panels B and C we use the residuals from the regressions we described in Section
3.3.2 as the dependent variable to account for differences in quality between houses.
The results are qualitatively and quantitatively very similar to the ones we present
in Panel A. In Panel B we are using the residuals of a regression of log of house price
on a set of characteristics, and we find a point estimate of -0.0017 that translates to
residual being lower by 620 dollars for houses above the threshold of 125 percent of
the CLL when the CLL binds, considering an average transaction value of 371,340
dollars. This suggests that transactions that cannot be financed at 80 percent with
conforming loans are made at lower prices even after we control for a rich set of house
characteristics. 11
Similarly in Panel C of Table 3.5, we confirm that even when we use the value per
square foot as a dependent variable but control for house quality, the interaction term
is significant and economically large even though the point estimate of 0.65 dollars
for houses above the threshold is slightly lower than the results in Panel A where we
do not adjust for house quality. The difference between the point estimate of 1.16
dollars of Panel A and 0.65 dollars in this specification indicates that houses above
the limit are of somewhat worse quality than those below the limit in the year that
the limit is in effect.
We also show that the estimated effect of the conforming loan limit on house prices
is stronger in the first half of the sample than in the second half. This result holds
"In the Appendix we show that the results are unchanged if we include the characteristics as
controls in the regressions, as opposed to running the regressions with the hedonic residuals.
128
for all three left-hand side variables. This is in line with our expectations, given that
borrowers had easier access to second lien loans after 2002 (we show the evolution of
the use of second liens in Figure 1 of the Appendix). Additionally, more borrowers
use jumbo loans, which may reflect a reduction of the cost differential of this type of
loan relative to conforming loans, and an increase in the ease of access to this type of
loan, possibly driven by an increased ease of securitization of these loans. Finally, in
the Appendix we show the robustness of our results to serial correlation in the error
term by constructing Newey-West standard errors, and all the results are unchanged
3.4.2
Credit Supply and Income
We now turn to how the effect of credit supply on house prices changes with the
growth in income in a zip code. To do this, we obtain data on zip code level average
2
household income each year from 2000 to 2007 from Melissa Data. We create a new
variable that is a "1" if a zip code has negative nominal average income growth from
one year to the next, and "0" otherwise. We then run similar regressions to what we
did before (year-by-year), adding an interaction between our previous variables and
this new zip code level "Negative Income Growth" variable. Looking at the coefficient
on the triple interaction term (negative income growth, the year that the CLL is in
effect, and being above 125 percent of the CLL) allows us to identify how the effect of
credit supply differs in times of positive and negative income growth. Our hypothesis
is that the effect of credit supply is stronger in times of negative income growth, as
households in a certain zip code are more likely to be constrained and there is likely
to be less competition for housing, which increases the probability that a seller sells
to a constrained buyer.
We show the results for these regressions in Table 3.6. In the first column of Table
3.6 we repeat our main regressions for the period 2001-2005 only, as this is the period
for which we were able to construct the income growth indicator variable. The results
are consistent with those in Table 3.5. In the second column of Table 3.6 we show
Fama-MacBeth coefficients from the regressions with the income growth interaction
term. The triple interaction terms show that the effect of credit supply on value per
square foot is significantly stronger in zip codes and years that are below the 10th
percentile of income growth for the individual regression. The point estimate shows
that value per square foot is 1.55 dollars lower in the year that the conforming loan
limit is in effect for houses above 125 percent of the limit when income growth is low
in a zip code. We also find that the main effect from our regressions in Table 3.5 is
quantitatively similar to before, implying that the simple inclusion of ZIP code level
income does not change any of our main results.
In the Appendix we plot the distribution of value per square foot for ZIP codes
of different income levels. Those pictures also suggest that the distribution of value
per square foot is affected by the conforming loan limit in ZIP codes in the lowest
quartile of the income distribution. In particular, the average value per square foot
is monotonically increasing for up to conforming loan limit threshold, and from this
12
Melissa Data obtains this data from the IRS and provides it in an easy-to-read format.
129
point onwards the distribution becomes flat. This pattern is not visible for zip codes
with higher median incomes.
3.4.3
Robustness and Refinements
Differential House Price Trends
We want to rule out that our results are driven by differences in secular trends between
houses above and below the threshold of CLL/0.8. Specifically, if more expensive
houses have, on average, lower house price growth from one year to the next relative
to less expensive houses, we might obtain the results reported in Table 3.5, but we
might also obtain similar results for samples with transactions above and below other
arbitrary thresholds.
In order to address whether the effect that we find is indeed the product of the
true conforming loan limits and not due to different trends along the distribution
of houses, we run the same regressions described in Section 3.3.3 for "placebo" loan
limits. We do this by shifting the true conforming loan limit in USD 10,000 steps
from the true value each year. We start at CLL-100,000 and move 20 steps until we
reach CLL+100,000. For each of these 21 tests, we first define the "shift" relative to
the true conforming loan limits, and then we change the limits for all years by that
amount. For example, when we are changing all the limits by -20,000, this means
that the "placebo" limit for 1999 is 220,000 dollars instead of the true 240,000 dollars,
the "placebo" limit for 2000 is 232,700 instead of 252,700, and so on. We then run
the same year-by-year regressions and produce Fama-MacBeth coefficients for each
of the 20 alternative "placebo" values for the CLL. The results from this exercise are
shown in Table 3.7.
The table shows that the coefficients of interest we obtain for all three dependent
variables (values per square foot, residuals from the transaction amounts, and residuals of values per square foot) are systematically among the lowest of all obtained with
the 20 "placebo" trials (the ranking is given in the last two rows of the table). The
coefficient on the value per square foot measure is the lowest of the 21 trials whether
we use the whole sample, or whether we limit our attention a sample of transactions
that all have an LTV between 0.5 and 0.8.13 When we use the whole sample and
the two residual measures from the hedonic regressions as the left-hand side variables
in the regressions, the coefficients for the true conforming loan limits are the second
and third lowest. In the restricted sample with LTVs between 0.5 and 0.8, these two
measures produce the second lowest and the lowest coefficient out of the 21 trials. If
we limit our attention to placebo limits that are below the true limits (i.e. the top
half of Table 3.7), all our measures produce the lowest coefficients out of those trials.
We consider these to be true "placebos", because all the transactions used for those
regressions are, by construction, below the "eligibility" criteria of 125 percent of the
true conforming loan limit both in the year that the limit is in effect, and in the
13
We discuss this subsample in more detail and show the equivalent to our Table 3.5 for this
sample in the Appendix
130
subsequent year. As such, these transactions should not have any changes in credit
availability from one year to the next.
When we compute the standard deviation of those coefficients, we find that the
coefficients using value per square foot as the dependent variable are statistically
significantly different from the average of the other coefficients at a 5 percent level in
both the whole sample and in the restricted sample with LTV between 0.5 and 0.8.
T-statistics for these tests are shown in the fourth row of Table 3.7. When we use
the value per square foot residual measure as a left-hand side variable, the coefficient
has a t-statistic of 1.77 in the whole sample, and above 2.37 in the restricted sample.
Finally, the coefficient from the regression that uses the residual from the log of house
price hedonic regression as a left-hand side variable is not significantly different from
the average of the other coefficients, as the t-statistics are between 1.0 and 1.2 in
both the whole sample and in the restricted sample. The fact that the results are
directionally the same when using all three left-hand side variables, and that there is
no "placebo" limit that consistently produces results that are as strong as the ones
from the true limit, further confirms that our coefficients are not obtained by pure
chance.
Selection Into Treatment
As discussed in the introduction, there can be at least two alternative mechanisms
for the effect of the conforming loan limits on house valuation. The first mechanism
is that cheaper credit around the threshold leads to an increase in the demand for
houses of a certain type, which then leads to higher valuation of these houses (or,
conversely, higher cost of credit reduces the demand for houses above the threshold in
the year that the limit is in effect). The alternative mechanism is that different credit
conditions above and below the threshold attract a type of buyer in the year that
the limit is in effect that is both better able to deal with the higher cost of funding
(possibly because of higher wealth or income), and is a more effective negotiator than
other "typical" buyers. This would still mean that our results are driven by credit
conditions being different above and below the threshold, but it would be a different
mechanism for our results. This selection effect results from the fact that borrowers
can choose the level of their LTV. If all borrowers mechanically had to use an LTV
of 80 percent, there would not be any possibility for selection.
To understand whether the aforementioned form of selection is important, we
divide transactions that are just above the cut off for being eligible for a CLL at
80 percent in a given year into two groups: (1) transactions that nevertheless use a
conforming loan and therefore choose to have an LTV below 80 percent (making up
the difference with other forms of financing), and (2) transactions that use a jumbo
loan with an 80 percent LTV, which means they do not get a conforming loan. The
first group isolates the set of borrowers where selection could be an issue. These
borrowers might be optimizing around the CLL threshold and could therefore have
other unobservable differences from the rest of the borrowers. For example, these
"special" buyers could have more wealth or higher income and thus might also differ
in other unobservables such as their ability to bargain. By excluding the group of
131
home buyers who choose this type of financing, we can test if these are driving our
results, i.e. whether they alone buy cheaper houses. As an aside, it is ex ante not clear
why those borrowers would buy cheaper houses (based on value per square foot). The
fact that they are wealthier would usually lead us to believe that the omitted variable
bias goes in the other direction, i.e. they buy houses with higher unobservable quality.
The following regressions show that this group of borrowers does not drive our results.
To test the importance of the selection effect, we run differences-in-differences
regressions excluding each of the two groups described above at a time (in the year
that the limit is in effect) and construct Fama-MacBeth coefficients, as we did in
Table 3.5. The results are shown in Table 3.8. We find that results do not change
much when we exclude the jumbo loans or when we exclude the conforming loans,
which implies that our main results are not being driven solely by either one of these
groups of transactions. The statistical significance of the results is similar, and the
magnitude of the coefficients sometimes is larger for one group and other times for
the other, depending on the left-hand side measure we use. Overall, the results point
in the same direction for both sets of regressions.
This robustness test shows that the effect of credit conditions on house prices in
our setting is not likely to be driven solely by selection of different buyers in our
"treated" group. If this were the case, we would expect the borrowers that pick a
conforming loan and end up with an LTV below 80 percent to be the ones driving our
main result. The fact that we also see similar results when we exclude this subgroup
increases the likelihood of our alternative explanation, namely that differential cost of
credit changes demand for housing, and that this shift in demand for housing drives
the change in house valuation.
In the Appendix we show that our results are stable if we use a 5,000 dollar
band around the threshold of CLL/0.8 instead of the 10,000, which suggests that the
difference in the cost of credit is likely to be similar for these two sets of buyers relative
to buyers below the threshold. This is further evidence that the result is not driven
solely by buyers who choose to obtain a conforming mortgage and put up additional
equity from other sources. Finally, we also show in the Appendix that the effect of
the CLL is similar for the first 9 months of the year and for the last three months,
indicating that borrowers do not behave differently after the limit for the subsequent
year has been defined by the administration.
Constraints to Housing Supply
To understand whether the effect of credit supply is amplified by the inability of
housing supply to adjust quickly to demand, we divide zip codes into high and low
house supply elasticity according to the measure in Saiz (2010). If the supply of
housing were perfectly elastic and able to adjust quickly to an increase in demand
for houses, the effect on prices should not be there. In this test, we find that the
constraint imposed by the conforming loan limit is stronger in zip codes located in
more inelastic, metropolitan, statistical areas (MSAs) according to the Saiz measure
(Table 3.9). This result is in line with what we expect and with previous literature
(e.g. Mian and Sufi, 2009), namely that cheaper credit will feed through to house
132
prices more frequently in regions where the supply of houses cannot adjust as easily.
We are cautious to interpret this result, however, because we have limited crosssectional variation in the elasticity measure in our data. In fact, all of the MSAs in
our sample are above the median elasticity found in Saiz (2010) for the whole country,
and 7 of the 10 MSAs are in the top 20 percent of MSAs with the least elasticity in
the nation.
3.4.4
Economic Magnitude of the Effect
As we discuss in Section 3.2, there is significant disagreement as to what the magnitude of the elasticity of house prices to interest rates is, as changes to the way
a standard user cost model is specified can produce vastly different estimates. To
understand the magnitude of our estimated effect, we compute the semi-elasticity of
house prices to interest rates, calculated as the percentage change in prices divided
by the change in interest rates. The change in the CLL gives us an unbiased local
estimate of the numerator of this semi-elasticity. To obtain an estimate of the denominator, we use the differential in interest rates between jumbo and conforming
loans estimated in the prior literature.
Table 3.10 shows that the change in house prices around the CLL ranges from 30
to 91 basis points. We obtain the low of 30 basis points when we use the residuals
from the hedonic regressions of value per square foot as the dependent variable and
include the whole time period (1998 to 2006).4 The high end of the estimate (91 basis
points) comes from the specification where we constrain the period to 1998-2001 and
use the raw value per square foot as the dependent variable. We exclude our estimates
for the period 2002 to 2005 since we know that the CLL was less important during
that time.
There is an extensive literature that provides estimates of the jumbo-conforming
spread, see McKenzie (2002), Ambrose, LaCour-Little, and Sanders (2004), Sherlund
(2008), Kaufman (2012) and DeFusco and Paciorek (2013). The most common estimates that have been found across all the papers range from a low of 10 basis points
to a high of 24 basis points. 15 If we divide our estimated range of house price changes
by the range in the jumbo-conforming spread, we obtain estimates for the elasticity
of house prices to interest rates that vary between 1.2 and 9.1 (Table 3.10). While
these estimates are local in nature, i.e. they do not use the full distribution of housing
transactions in the data nor do they take into account general equilibrium effects, this
is the first unbiased estimate of this semi-elasticity in the literature and the results
are at the lower end of the estimates that have been proposed previously (see, for
example, Glaeser, Gottlieb, and Gyourko, 2010). In fact, given our data, it is hard
The point estimate in the regressions is 0.65 dollars from Panel C in Table 3.5, and we scale
that by the average value per square foot for the sample to obtain 30 basis point changes in value
per square foot.
15
The paper by Kaufman (2012) obtains an estimate of 10 basis points by using a regression discontinuity approach on the access to conforming loans around the threshold of CLL/0.8 in appraisal
values. This estimate is particularly relevant for our purposes given that it explores the part of the
distribution of homes that we also consider.
14
133
to justify estimates above 10 without making very aggressive assumptions about the
cost differential above and below the threshold.
The prior calculation is our preferred method of obtaining an estimate of the elasticity. However, we can obtain an alternative estimate of the elasticity by considering
borrowers who choose to obtain a conforming loan of less than 80 percent LTV above
the threshold. This means they put up additional equity which either has to be financed through a third party loan or through savings. On average, given the range
of transactions used in the regressions, these borrowers put up an additional USD
5,000. If we assume that the cost of the additional equity is 5 percentage points or
more above the conforming mortgage rate, this is equivalent to a spread of 6-8 basis
points in the total cost of financing for these borrowers relative to those who buy a
house below the threshold. This translates into an elasticity of between 4.4 and 11.4,
depending on the house price effect we use from our regressions. The assumption for
the spread of 5 percentage points over the conforming mortgage rate is not high if we
consider that many people use a jumbo loan even very close to the threshold of the
CLL, indicating that the cost of additional equity is, at least for some borrowers, very
substantial. The fact that we see borrowers stick with a conforming loan and put up
additional equity above the threshold may, in fact, be an indication that they are excluded from the jumbo market altogether, rather than evidence that this is a cheaper
option. As Loutskina and Strahan (2009, 2011) show, jumbo loans are associated
with more careful screening of borrowers, which may mean that many households
simply could not use an 80 percent LTV above the threshold of 125 percent of the
CLL even if they were looking to do so.
Another way of assessing the economic importance of the effect we find is by comparing the dollar amount of savings through lower interest rates and the house price
differential. Assume a loan of USD 300,000, which is approximately the conforming
loan limit midway through our sample (2002). If we use the upper end of the jumboconforming spread of 24 basis points, we calculate a cost difference of USD 720 in
the first year of the life of the loan. The present value of the cost difference over 30
years is USD 8,557 assuming a 6 percent discount rate. If we use the lower end of the
jumbo-conforming spread that has been estimated (10 basis points), this cost difference is USD 3,604. Our estimated effect of the conforming loan is a price difference
of USD 0.65-1.16 per square foot for an average size of a house of 1,935 square feet.
This translates into a USD 1,256-2,244 difference in the price of the house. Thus, for
each dollar of savings in the present value of interest costs, home values increase by
about 25-60 cents (always less than 1 dollar).
One possible concern with our estimation is that home buyers might expect the
conforming loan limit to rise in the subsequent year and would thus refinance their
loan shortly after obtaining it. If refinancing were frictionless, buying a house above
the threshold would cost 10-24 basis points more than the conforming loan rate for
only one year, because borrowers who took a jumbo loan would immediately refinance
into a conforming loan in the following year (once the limit was raised). This would
imply a very high elasticity of house prices to interest rates, as the difference in
the effective interest rate over the life of the loan paid by a borrower who took a
conforming loan and one who took a jumbo loan would be very small. However,
134
this analysis misses the transaction costs of refinancing, and the estimates of these
transaction costs that have been found in the literature are very large. A paper by
Stanton (1995) finds that transaction costs for mortgage prepayment are around 30
to 50 percent of the remaining principal balance of a mortgage. These transaction
costs include both explicit monetary costs (about one-sixth of the total costs) and
non-monetary prepayment costs (the remaining five-sixths). A more recent paper by
Downing, Stanton and Wallace (2005) produced a lower, but still substantial, average
transaction cost of refinancing of 11.5 percent of face value. The bottom line from
both these studies is clear - transaction costs are too high for the jumbo conforming
spread alone to significantly change the prepayment behavior of borrowers. In other
words, the benefit from obtaining lower interest rates by refinancing to a conforming
loan in a year or two are too small to overcome the transaction costs of refinancing.
3.5
Conclusion
In this paper we use the exogenous changes in the annual level of the conforming loan
limit as an instrument for lower cost of funding. We find that a home that becomes
eligible for cheaper mortgages due to an increase in the CLL has, on average, a
1.16 dollar higher value per square foot compared to a house that is just above the
threshold that allows it to be financed with a conforming loan at 80 percent loan
to value. The magnitude of the difference that we find is economically important
given the average value per square foot of houses that transact around the CLL
of 220 dollars, which means that a 1.16 dollar increase constitutes almost a 0.45
percent increase in prices. Under our assumptions for the interest rate differential for
transactions above and below the threshold, this corresponds to a semi-elasticity of
prices to interest rates of less than 10.
Another way of stating our results is to say that the interest rate subsidy granted
by the GSEs and, ultimately, the taxpayer, does not fully benefit the buyers of homes
and, instead, partially accrues to the sellers of homes in the form of higher house
prices. Also, the results suggest that mortgages are being supplied in a competitive
fashion, and that originating banks are not appropriating the mortgage subsidy provided by the GSEs. In addition, we see that the CLL constitutes a first order factor
in how houses are financed: there is a significant fraction of borrowers who choose
an LTV below 80 percent, between 77 and 79.5 percent, in order to stay below the
conforming loan limit. These borrowers either were unable to get a jumbo loan, or
are trying to take advantage of the lower interest rate of a conforming loan. But, as a
result, many borrowers end up holding a larger fraction of equity in their house than
most other borrowers.
These results are stronger in the earlier part of our sample when borrowers were
less likely to have access to other forms of financing, such as second liens, and when
the interest rate differential between jumbo loans and conforming loans was larger.
After 2004 in particular, we see that the vast majority of borrowers even above the
threshold of 125 percent of the CLL choose an LTV of 80 percent, which supports the
idea that access to jumbo loans and other forms of financing became much easier in the
135
second half of the sample. At the same time, the house price impact of the conforming
loan limit is also smaller in this time period. This suggests that those houses which
were previously just out of reach of being financed by a conforming loan at 80 percent
could now be bid up in price since people had easier access to jumbo loans and other
forms of finance. The results are also stronger in ZIP codes with the lowest income
growth, usually negative, and also in areas with lower elasticity of housing supply.
While we can only estimate a local treatment effect around the CLL, this presents
a first test of the exogenous effect of cheaper mortgage loans on house prices. We
estimate an elasticity of house prices to interest rates that is below 10, implying that
the drop in mortgage rates cannot account for the increase in house prices between
2000 and 2006. However, we do show that those credit conditions matter for the
formation of prices. Our results do not support a view that credit market conditions
purely respond to housing demand, but point instead to a directional effect that easier
credit supply leads to an increase in house prices.
136
3.6
Bibliography
Acharya, V., Richardson, M., Nieuwerburgh, S. V., and White, L. J. (2010) Guaranteed to Fail: Fannie Mae, Freddie Mac, and the Debacle of Mortgage Finance.
Princeton University Press, March 2011.
Ambrose, B. W., LaCour-Little, M., and Sanders A.B. (2004) The Effect of Conforming Loan Status on Mortgage Yield Spreads: A Loan Level Analysis. Real Estate
Economics, Vol. 32, No. 4, 541-569.
Brunnermeier, Markus K., Eisenbach, T., and Sannikov, Y. (2012) Macroeconomics with Financial Frictions: A Survey. Working Paper.
Calomiris, C. W.. (2009). Financial Innovation, Regulation, and Reform. Cato
Journal (29) p. 6 5 .
Campbell, J.Y., Giglio, S., and Pathak, P. (2010) Forced Sales and House Prices.
American Economic Review, Forthcoming.
DeFusco, A. A., and Paciorek, A. (2013) The Interest Rate Elasticity of Mortgage
Demand: Evidence From Bunching at the Conforming Loan Limit. Working Paper.
Downing, C., Stanton, R., and Wallace, N. (2005) An Empirical Test of a TwoFactor Mortgage Valuation Model: How Much Do House Prices Matter? Real Estate
Economics, Vol. 33, Issue 4, 681-710.
Fama, E. F., and MacBeth, J. D. (1973) Risk, Return, and Equilibrium: Empirical
Tests. Journal of PoliticalEconomy, Vol. 81, No. 3, 607-636.
Favara, G., and Imbs, J. (2011) Credit Supply and the Price of Housing. CEPR
Discussion Paper, No. 8129.
FavilukisJ., Ludvigson, S.C., and Nieuwerburgh, S. V. (2010) The Macroeconomic
Effects of Housing Wealth, Housing Finance, and Limited Risk-Sharing in General
Equilibrium. NBER Working Paper, No. 15988.
Genesove, D. and Mayer, C. J. (1997) Equity and Time to Sale in the Real Estate
Market. American Economic Review, Vol. 87, No. 3. (Jun, 1997), 255-269.
Glaeser, E. L, Gottlieb, J., and Gyourko, J. (2010) Can Cheap Credit Explain the
Housing Boom. NBER Working Paper,No. 16230.
Green, R. K., and Wachter, S. M. (2005) The American Mortgage in Historical
and International Context. Journal of Economic Perspectives,Vol. 19, No. 4, 93-114.
Himmelberg, C., Mayer, C., and Sinai, T. (2005) Assessing High House Prices:
Bubbles, Fundamentals and Misperceptions. Journal of Economic Perspectives, Vol.
19(4), 67-92.
Hubbard, G., and Mayer, C. (2008) House Prices, Interest Rates, and the Mortgage Market Meltdown. Columbia Business School Working Paper.
Kaufman, A. (2012) What do Fannie and Freddie do? Unpublished Manuscript.
Khandani, A. E., Lo, A.W., and Merton, R.C. (2009) Systemic Risk and the
Refinancing Ratchet Effect. NBER Working Paper,No. 15362.
Loutskina, E., and Strahan, P. (2009) Securitization and the Declining Impact of
Bank Financial Condition on Loan Supply: Evidence from Mortgage Originations.
Journal of Finance,64(2), 861-922.
137
Loutskina, E., and Strahan, P. (2011) Informed and Uninformed Investment in
Housing: The Downside of Diversification. Review of FinancialStudies, 24(5), 144780.
Mayer, C. (2011) Housing Bubbles: A Survey. Annual Review of Economics,
3:55977.
McKenzie, J.A. (2002) A Reconsideration of the Jumbo/Non-jumbo Mortgage
Rate Differential. Journal of Real Estate Finance and Economics, Vol. 25, No. 2-3,
197-213.
Mian, A., and Sufi, A. (2009) The Consequences of Mortgage Credit Expansion:
Evidence from the U.S. Mortgage Default Crisis. Quarterly Journal of Economics,
Vol. 124, No. 4, 1449-1496.
Pavlov, A., and Wachter, S. (2011) Subprime Lending and Real Estate Prices.
Real Estate Economics, 39: 117
Poterba, J. (1984) Tax Subsidies to Owner-occupied Housing: An Asset-Market
Approach. Quarterly Journal of Economics, Vol. 99(4), 729-52.
Saiz, A. (2010) The Geographic Determinants of Housing Supply. Quarterly Journal of Economics, 125(3): 1253-1296.
Sherlund, S.M. (2008) The Jumbo-Conforming Spread: A Semiparametric Approach. Finance and Economics Discussion Series Working Paper,2008-01.
Stanton, R. (1995) Rational Prepayment and the Valuation of Mortgage-Backed
Securities Review of FinancialStudies, Vol. 8, No. 3, 677-708.
Stein, J.C. (1995) Prices and Trading Volume in the Housing Market: A Model
with Down-Payment Effects. Quarterly Journal of Economics, Vol. 100, No. 2,
379-406.
138
Figure 3-1: Transaction-Loan Value Surface
Note: This figure shows the frequency of transactions at each house price-loan value combination
for the year 2000 and 2004, and the 10 MSAs covered in our data, where both house prices and
loan values were binned at USD 10,000 intervals. The mass of transactions on the diagonal have a
loan to value of approximately 0.8.
(a) 2000
4500
4000
3500
2500
I'm
3500
1000
SM
Transaction
Loan Value
Value
(b) 2004
IiLIL
SODO
40M
Transaction
LOWn Value
Value
139
Figure 3-2: Borrower Composition for the Regression Sample
Note: This figure shows the number of transactions by month for transactions within USD 10,000
of the threshold of 125 percent of CLL. Transactions below and above this threshold are tracked
from the year prior to the CLL being in effect to the year after the CLL is lifted to its new value.
We break down transactions by LTV range to show the differences that emerge between houses
above and below 125 percent of the CLL.
(a) Transactions below 125 percent of CLL
3500
(A)
(B)
CLI year-i
3000
(C)
CLLyear+ 1
CLLyear
12500
2000
1500
Li
-02
0
1 2 3 4 5 6 7 8 9
101112131415161718192021222324252627282930313233343536
months
E LTV<75
m75<LTV<80
OLWV=80
(ILTV>80
(b) Transactions above 125 percent of CLL
3500
300 0
CLL year -1
i
CLLyear
CLL year +1
2500
2000
1500
500
1
2 3 4
5
6 7
8
9 1011121314Im1617181920212223242526272829303132333435
Months
EILTV<75
*75<LTV<80
140
IJLTV=80
EJLTV>80
Figure 3-3: Frequency of Transactions as Percentage of CLL Threshold
Note: This figure shows the frequency of transactions by their distance to the threshold of 125
percent of the conforming loan limit. The vertical red line is the threshold and the transactions for
all years are centered around that value. The x-axis is represented as one minus the transaction
value as a percentage of each year's threshold of 125 percent of the conforming loan limit (e.g. if
the threshold is 200,000, a transaction of 150,000 will appear as -25 percent).
0
03
C 0D
C
I0 03
0
C
$1
*@@
O
vfI %
E
z
0%
.met0406.
R
-100
0
-50
50
Transaction Value as Percentage of 1.25CLL
141
100
Figure 3-4: Share of Unused Mortgage Applications
Note: The horizontal axis indicates the difference between loan amounts and the conforming
loan
limit as a percentage of the conforming loan limit. The share of unused mortgages is
constructed
from HMDA as the number of "withdrawn" or "unused" mortgage applications as a percentage
of
total applications. We aggregate these proportions into 1% bins and each dot in the figure
represents the share of unused mortgages for each bin. We also plot third degree polynomials
( to
the left and right of the conforming loan limit) as well as 95% confidence intervals (dashed
lines).
Data extracted from HMDA, 1998-2006.
*~
-A
a
a
~-a
o
Soa
-a-'.
S
10
U
S
0~
U
r
0~
-50
-45
40
-35
-30
-25
-20
-15
-10
-5
0
5
10
15
20
Distance from the conforming loan limit (%)
142
25
30
35
40
45
50
Table 3.1: Summary Statistics
Panel A. House Characteristics.
Transaction Value (USD 1000)
Loan to value
House Size (sqft)
Lot Size (sqft)
Number of rooms
Number of bedrooms
Number of bathrooms
House age (years)
All Transactions N=3,983,575
Median
Mean Std. Dev.
286.00
123.93
308.52
Regression Sample N=262,671
Median
Mean Std. Dev.
380.00
54.92
371.34
0.81
0.15
0.80
0.76
0.13
0.80
1,735
10,197
6.84
3.20
1.93
35.40
672
15,495
1.60
0.78
1.03
27.70
1,592
6,700
7.00
3.00
2.00
34.00
1,935
11,734
7.23
3.33
2.11
34.74
701
17,923
1.61
0.78
1.07
27.40
1,816
7,203
7.00
3.00
2.00
34.00
Panel B. House Valuation.
Value per sqft (USD/sqft)
Value per sqft residual (USD/sqft
Log of transaction value residual (USD)
All Transactions N=3,983,575
Median
Mean Std. Dev.
172.03
91.60
193.59
-0.95
42.30
0.00
0.00
0.17
0.01
Regression Sample N=262,671
Median
Mean Std. Dev.
200.20
93.37
219.63
3.43
44.26
5.29
0.05
0.14
0.04
Note: Panel A shows the descriptive statistics for all transactions in our data from 1998 to 2008.
The data was extracted from deeds records by Dataquick. Panel B shows the different valuation
measures we use in the regression analysis. Value per sqft is the transaction amount divided by the
size of the house measured in square feet. Both the residual measures are obtained from hedonic
regressions run by year and by metropolitan area of value per sqft and transaction value on a set
of detailed house characteristics. We give more information on the construction of the residuals in
Section 2, Data and Methodology.
143
Table 3.2: Summary Statistics by Geography and Year
Panel A. Geographic Distribution
MSA
Boston
Chicago
DC
Denver
Las Vegas
Los Angeles
Miarni
New York
San Diego
San Francisco
Total
N Obs
279,261
377,031
396,211
397,293
345,219
725,897
483,541
487,104
219,489
272,529
3,983.575
Transaction Value
Mean
Std. Dev
320.29
112.40
262.41
108.15
329.95
126.16
250.22
94.93
262.24
102.87
332.28
129.71
270.10
111.74
341.00
121.13
353.14
124.63
383.59
123.74
308.52
123.93
Value per sqft
Mean Std. Dev
197.67
73.81
174.37
68.63
186.97
85.93
155.84
49.28
136.62
45.38
231.29
108.35
144.80
57.04
221.25
92.55
222.18
94.86
266.47
109.26
193.59
91.60
Loan to Value
Mean Std. Dev
0.78
0.16
0.81
0.15
0.82
0.14
0.83
0.15
0.82
0.14
0.81
0.13
0.81
0.14
0.78
0.17
0.79
0.14
0.79
0.13
0.81
0.15
Transaction Value
Mean Std. Dev
239.78
102.07
246.38
104.88
257.67
109.21
265.16
108.82
283.79
114.34
303.37
118.32
331.81
121.20
357.51
121.71
366.27
121.89
359.24
122.53
325.11
119.84
308.52
123.93
Value per sqft
Mean Std. Dev
133.84
50.59
139.33
54.03
149.65
61.64
156.74
63.81
171.06
71.85
187.40
80.05
212.65
90.51
237.24
100.72
247.02
105.50
237.79
101.57
206.92
91.62
193.59
91.60
Panel B. Distribution By Year and Thresholds
Year
1998
1999
2000
2001
2002
2003
2004
2005
2006
2007
2008
Total
N Obs
134,200
350,827
354,071
365,814
397,527
423,939
525,407
475,723
376,182
293,329
286,556
3,983,575
Thresholds
House Price Conf. Loan
283,938
227,150
300,000
240,000
315,875
252,700
343,750
275,000
375,875
300,700
403,375
322,700
417,125
333,700
449,563
359,650
521,250
417,000
521,250
417,000
521,250
417,000
Loan to Value
Mean Std. Dev
0.81
0.15
0.81
0.15
0.81
0.16
0.82
0.15
0.81
0.15
0.81
0.15
0.79
0.14
0.78
0.13
0.79
0.13
0.82
0.14
0.84
0.15
0.81
0.15
Note: This table uses all the deed registry data on house transactions for 10 MSAs. Panel A shows
the mean and standard deviation by city of (i) house price, (ii) value per sqft and (iii) loan to value.
Panel B the mean and standard deviation by year for the same three variables.
144
Table 3.3: Verification of the Impact of the CLL on Financing Choices
Panel A: Loan to Value
Above Threshold
Year CLL
Above Threshold x
Year CLL
No. Obs.
All years
-0.004***
(0.001)
-0.008***
(0.002)
-0.004***
(0.001)
242,753
1998-2001
-0.006***
(0.002)
-0.005**
(0.002)
-0.004*
(0.002)
100,870
2002-2005
-0.002***
(0.001)
-0.011***
(0.001)
-0.003*
(0.002)
141,883
1998-2001
0.024***
(0.003)
-0.009***
(0.003)
-0.007*
(0.004)
100,870
2002-2005
0.021***
(0.001)
-0.017***
(0.003)
-0.005
(0.003)
141,883
Panel B: Log Loan Amount
Above Threshold
Year CLL
Above Threshold x
Year CLL
No. Obs.
All years
0.023***
(0.002)
-0.013***
(0.002)
-0.006**
(0.002)
242,753
Note: This table shows Fama MacBeth coefficients computed from year by year regressions that
use two measures of financing choice as the dependent variable in each of the two panels. The
sample includes all transactions within USD 10,000 of each year's conforming loan limit, as well as
transactions of the same amount in the subsequent year. Above the Threshold refers to transactions
up to USD 10,000 above the conforming loan limit divided by 0.8 (i.e. the transactions that were
"ineligible" to be bought with a conforming loan at a full 80 percent LTV) and Year CLL is the year
in which the conforming loan limit is in effect.
145
Table 3.4: Impact of CLL on Number of Transactions
Year CLL
No. Obs.
All years
-0.003***
(0.000)
262,671
1998-2001
2002-2005
0.000
(0.001)
109,496
-0.006***
(0.001)
153,175
Note: This table shows Fama MacBeth coefficients computed from year by year regressions that use
a dummy variable for whether a transaction happens above the threshold of 125 percent of the CLL
as the dependent variable. The sample includes all transactions within USD 10,000 of each year's
conforming loan limit, as well as transactions of the same amount in the subsequent year. Year CLL
is the year in which the conforming loan limit is in effect. Zip Codes fixed effects are included on
each regression
146
Table 3.5: Effect of the CLL on House Valuation Measures
Panel A: Value Per Square Foot
Above Threshold
Year CLL
Above Threshold x
Year CLL
No. Obs.
All years
1.261**
(0.494)
-22.869***
(4.047)
-1.162***
(0.264)
262,671
1998-2001
1.669***
(0.573)
-14.851***
(2.314)
-1.553***
(0.297)
109,496
2002-2005
0.852
(0.836)
-30.886***
(5.314)
-0.771**
(0.369)
153,175
Panel B: Log of Transaction Value Residual from Hedonic Regressions
Above Threshold
Year CLL
Above Threshold x
Year CLL
No. Obs.
All years
0.0129***
(0.0013)
0.0387***
(0.0041)
-0.0017**
(0.0008)
251,431
1998-2001
0.0154***
(0.0015)
0.0356***
(0.0047)
-0.0020
(0.0015)
103,535
2002-2005
0.0104***
(0.0009)
0.0417***
(0.0072)
-0.0013***
(0.0004)
147,896
Panel C: Value Per Square Foot Residual from Hedonic Regressions
Above Threshold
Year CLL
Above Threshold x
Year CLL
No. Obs.
All years
1.733***
(0.360)
4.103***
(0.644)
-0.651***
(0.238)
251,764
1998-2001
2.060***
(0.425)
3.935***
(0.495)
-0.940***
(0.351)
103,709
2002-2005
1.407**
(0.595)
4.270***
(1.293)
-0.362
(0.291)
148,055
Note: This table shows Fama MacBeth coefficients computed from year by year regressions that
use three alternative measures of valuation as the dependent variable in each of the three panels.
The hedonic regressions that produce the residuals for panels B and C are described in Section
3.3.2. The sample for each year's regression includes all transactions within +/- USD 10,000 of that
year's conforming loan limit, as well as transactions in the same band in the subsequent year. All
year by year regressions include ZIP code fixed effects. Above the Threshold refers to transactions
up to USD 10,000 above the conforming loan limit divided by 0.8 (i.e. the transactions that were
"ineligible" to be bought with a conforming loan at a full 80 percent LTV) and Year CLL is the year
in which the conforming loan limit is in effect.
147
Table 3.6: Effect of the CLL on House Valuation in Different Income Growth Areas
Panel A: Value Per Square Foot
Above Threshold
Year CLL
Above Threshold x
Year CLL
Above Threshold x
Year CLL x Low Inc. Growth
No. Obs.
2001-2005
0.731
(0.667)
-28.869***
(4.706)
-0.846***
(0.257)
179,828
2001-2005
0.601
(0.638)
-29.364***
(4.510)
-0.953***
(0.210)
-1.548**
(0.652)
179,828
Panel B: Log of Transaction Value Residual from Hedonic Regressions
Above Threshold
Year CLL
Above Threshold x
Year CLL
Above Threshold x
Year CLL x Low Inc. Growth
No. Obs.
2001-2005
0.0109***
(0.0008)
0.0418***
(0.0056)
-0.0016***
(0.0003)
173,347
2001-2005
0.0108***
(0.0009)
0.0439***
(0.0057)
-0.0022***
(0.0006)
-0.0018
(0.0051)
173,347
Panel C: Value Per Square Foot Residual from Hedonic Regressions
Above Threshold
Year CLL
Above Threshold x
Year CLL
Above Threshold x
Year CLL x Low Inc. Growth
No. Obs.
2001-2005
1.396***
(0.453)
4.314***
(1.017)
-0.504**
(0.250)
173,550
2001-2005
1.347***
(0.412)
4.806***
(1.072)
-0.750***
(0.158)
-0.319
(0.651)
173,550
Note: This table shows Fama MacBeth coefficients computed from year by year regressions that
use three alternative measures of valuation as the dependent variable in each of the three panels.
The sample for each year's regression includes all transactions within +/- USD 10,000 of that year's
conforming loan limit, as well as transactions in the same band in the subsequent year. Above the
Threshold refers to transactions up to USD 10,000 above the conforming loan limit divided by 0.8
(i.e. the transactions that were "ineligible" to be bought with a conforming loan at a full 80 percent
LTV) and Year CLL is the year in which the conforming loan limit is in effect. This specification
interacts the diff-in-diff specification with a dummy variable that uses changes in income at a zipcode
level as proxy for good and bad times. Specifically, the dummy is 1 if the changes in the average
zipcode income are below the 10th percentile of each particular diff-in-diff regression and 0 otherwise.
We use tax income data at zipcode level available from 2000-2006, which restricted our sample to
2001-2005
148
Table 3.7: Placebo Test for Coefficient of Interest
Value Per
Square Foot
All Transactions
Value Per
Log of
Transaction Square Foot
Residual
Value
Residual
0.5<LTV<0.8 Transactions
Value Per
Log of
Value Per
Transaction Square Foot
Square Foot
Residual
Value
Residual
True CLL
-1.162
-0.002
-0.651
-1.257
-0.002
-0.931
Placebo
0.045
(0.467)
0.001
(0.002)
0.222
(0.494)
-0.107
(-0.107)
0.000
(0.002)
0.110
(0.440)
T-Statistic
2.586
1.206
1.770
2.626
1.009
CLL Rank
CLL Rank
below only
1
1
4
2
2
1
1
1
3
1
2.365
1
1
Note: This table shows the average and standard deviation (in parenthesis) of a series of 20 placebo
tests we perform by shifting the conforming loan limit in USD 10,000 intervals from CLL-100,000
until CLL+100,000 (i.e. the limits of all years are first changed by -100,000, then by -90,000, etc.).
The first row shows the coefficients when we use the true conforming loan limit. We use the placebo
loan limits to run year-by-year regressions and form Fama-MacBeth coefficients like those in Table
3.5 for each set of "false" loan limits. The t-statistic is for the difference between the coefficients
when we use the true conforming loan limit and the average of all the other coefficients, using the
standard deviation given by the 20 trials. The three dependent variables are the same we use in
Table 3.5. The coefficient of interest is on the interaction between our "above threshold" variable
and the year in which the conforming loan limit is in effect. As in the previous tables, the sample
for each year's regression includes transactions within +/- USD 10,000 of that year's CLL, as well
as transactions in the same band in the subsequent year. The first three columns include all such
transactions, whereas in the last three columns the sample is constrained to transactions with an
LTV between 0.5 and 0.8. All year by year regressions include ZIP code fixed effects. The last two
rows show the ranking of the coefficient when we use the true CLL, first for all placebo limits and
then when we only consider the placebo tests below the true CLL.
149
Table 3.8: Effect of the CLL on the Valuation of Different Groups of Transactions
Panel A: Value Per Square Foot
Keeping Conforming
Above Threshold
Year CLL
Above Threshold x
Year CLL
No. Obs.
All years
0.939**
(0.472)
-24.539***
(4.351)
-0.967**
(0.416)
177,227
1998-2001
1.580***
(0.568)
-15.953***
(2.564)
-1.314**
(0.572)
72,048
2002-2005
0.297
(0.666)
-33.126***
(5.712)
-0.621
(0.634)
105,179
Keeping Jumbo
All years
0.868*
(0.481)
-24.874***
(4.454)
-2.177***
(0.639)
160,342
1998-2001
1.530***
(0.545)
-16.040***
(2.596)
-2.618**
(1.119)
62,905
2002-2005
0.207
(0.701)
-33.708***
(5.813)
-1.736**
(0.724)
97,437
Panel B: Log of Transaction Value Residual from Hedonic Regressions
Keeping Conforming
Above Threshold
Year CLL
Above Threshold x
Year CLL
No. Obs.
All years
0.0117***
(0.0014)
0.0367***
(0.0038)
-0.0027**
(0.0011)
170,808
1998-2001
0.0145***
(0.0018)
0.0335***
(0.0041)
-0.0019
(0.0022)
68,719
2002-2005
0.0090***
(0.0007)
0.0398***
(0.0067)
-0.0034***
(0.0009)
102,089
Keeping Jumbo
All years
0.0119***
(0.0013)
0.0370***
(0.0039)
0.0004
(0.0015)
154,848
1998-2001
0.0146***
(0.0016)
0.0337***
(0.0042)
-0.0020
(0.0025)
60,114
2002-2005
0.0091***
(0.0008)
0.0402***
(0.0068)
0.0028**
(0.0012)
94,734
Panel C: Value Per Square Foot Residual from Hedonic Regressions
Keeping Conforming
Keeping Jumbo
All years
1998-2001
2002-2005
All years
1998-2001
2002-2005
Above Threshold
1.573***
1.947***
1.199***
1.583***
1.991***
1.175**
Year CLL
(0.290)
3.514***
(0.579)
(0.357)
3.485***
(0.431)
(0.414)
3.543***
(1.175)
(0.308)
3.529***
(0.573)
(0.333)
3.552***
(0.409)
(0.470)
3.507***
(1.168)
Above Threshold x
-1.399***
-1.216**
-1.583***
0.225
-0.462
0.911**
(0.344)
170,946
(0.535)
68,790
(0.493)
102,156
(0.418)
154,949
(0.536)
60,165
(0.464)
94,784
Year CLL
No. Obs.
Note: This table shows Fama Macbeth coefficients computed from year by year regressions that use
three alternative measures of valuation as the dependent variable in each of the three panels. The
hedonic regressions that produce the residuals for panels B and C are described in Section 3.3.2.
The sample for each year's regression includes transactions within +/- USD 10,000 of that year's
conforming loan limit. All year by year regressions include ZIP code fixed effects. We divide the
transactions that happen at a price above 125 percent of a year's CLL in the year that the limit
is in effect into two groups: those with a conforming loan and those with a jumbo loan. We then
run the same regressions including just one of these two groups at a time. The first three columns
include the transactions with a conforming loan and the last three columns include transactions with
a jumbo loan. Above the Threshold refers to transactions up to USD 10,000 above the conforming
loan limit divided by 0.8 (i.e. the transactions that were "ineligible" to be bought with a conforming
loan at a full 80 percent LTV) and Year CLL is the year in which the conforming loan limit is in
effect.
150
Table 3.9: Effect of the CLL on House Valuation in Low Supply Elasticity Areas
Elasticity<1)
(
Panel A: Value Per Square Foot
Above Threshold
Year CLL
Above Threshold x
Year CLL
All
1.261**
(0.494)
-22.869***
(4.047)
-1.162***
(0.264)
Above Threshold x
Year CLL x Low Elasticity
No. Obs.
262,671
All
1.221
(0.799)
-15.282***
(3.920)
-0.430
(0.831)
-0.870
(0.977)
262,671
1998-2001
1.669***
(0.573)
-14.851***
(2.314)
-1.553***
(0.297)
109,496
1998-2001
3.069***
(0.374)
-8.015***
(0.843)
-2.100**
(0.817)
0.726
(1.332)
109,496
2002-2005
0.852
(0.836)
-30.886***
(5.314)
-0.771**
(0.369)
153,175
2002-2005
-0.628
(0.749)
-22.550***
(5.981)
1.239
(0.832)
-2.466**
(0.992)
153,175
Panel B: Log of Transaction Value Residual from Hedonic Regressions
Above Threshold
Year CLL
Above Threshold x
Year CLL
Above Threshold x
All
0 .0129***
(0.0013)
(I).0387***
(0.0041)
0.0017**
(0.0008)
Year CLL x Low Elasticity
No. Obs.
251,431
All
0.0106***
(0.0030)
0.0263***
(0.0037)
0.0008
(0.0022)
-0.0032
(0.002)
251,431
1 998-2001
0 .0154***
(0.0015)
0 .0356***
(0.0047)
-0.0020
(0.0015)
103,535
1998-2001
0.0182***
(0.0012)
0.0306***
(0.0044)
-0.0018
(0.0037)
-0.0002
(0.004)
103,535
2002-2005
0.0104***
(0.0009)
0.0417***
(0.0072)
-0.0013***
(0.0004)
1998-2001
2.623***
(0.278)
3.316***
(0.270)
-1.620***
(0.306)
0.843
(0.744)
103,709
2002-2005
1.407**
(0.595)
4.270***
(1.293)
-0.362
(0.291)
147,896
2002-2005
0.0030*
(0.0016)
0.0219***
(0.0055)
0.0033*
(0.0018)
-0.0063***
(0.0020)
147,896
Panel C: Value Per Square Foot Residual from Hedonic Regressions
Above Threshold
Year CLL
Above Threshold x
Year CLL
Above Threshold x
Year CLL x Low Elasticity
No. Obs.
All
1.733***
(0.360)
4.103***
(0.644)
-0.651***
(0.238)
251,764
All
1. 338**
(( .524)
1. 811**
(C.716)
- ).503
(C.546)
).241
(2).740)
2C1,764
1998-2001
2.060***
(0.425)
3.935***
(0.495)
-0.940***
(0.351)
103,709
148,055
2002-2005
0.054
(0.319)
0.305
(0.898)
0.615
(0.684)
-1.325
(1.104)
148,055
Note: In this case the dummy is 1 for low elasticity places. For this specification that corresponde to
the lowest MSA ( Miami, San Francisco, San Diego, Los Angeles, New York, Chicago and Boston).
The areas with elasticity higher than 1 are Las Vegas, Denver and DC
151
Table 3.10: Elasticity Estimates
Jumbo-Conforming Spread
A House Prices in bp
Max: 91.2
Min: 29.7
Min (10 bp)
9.1
3.0
Max (24 bp)
3.8
1.2
Note: This table shows elasticity calculations for different scenarios of both the house price increase
estimated in the regressions and the interest rate differential implied for transactions above and
below the threshold of 125 percent of the conforming loan limit. We use the jumbo-conforming
spread in interest rates as the denominator in the elasticity calculation.
152
3.7
3.7.1
Appendix A. Robustness and Refinements Additional Tests
Restrict LTV Choices
We want to test that our estimates are not driven by borrowers with very unusual LTV
levels, namely those with LTV below 50 percent and above 80 percent. Borrowers
with those choices of LTV are likely to either have access to abundant equity to put
up when buying a home, or to be very constrained and need a very high LTV. By
limiting our sample to include only borrowers who choose a first lien LTV between
50 and 80 percent, we capture the transactions that should be most affected by the
conforming loan limit. In particular, this subsample includes the group of borrowers
that end up with an LTV between 77 percent and 79.5 percent in the year that
the CLL is in effect because they stick with a conforming loan, even though their
house costs more than 125 percent of the CLL. This choice of LTV is very common
for the "Above the Threshold" group of borrowers in the year that the limit is in
effect, but very infrequent everywhere else in the distribution of transactions. Also,
this subsample includes all the borrowers that choose an 80 percent LTV, the most
frequent choice in the data. This means getting a jumbo loan for transactions "Above
the Threshold" and a conforming loan for transactions below that threshold. Finally,
the transactions that are excluded from this sample should be least affected by the
conforming loan limit, either because their LTVs are very low, in which case they
are never affected by the limit anyway, or alternatively, because they have high LTVs
and thus obtain jumbo loans in the year in which the limit is in effect whether the
price of the transactions is above or below the 125 percent of the CLL threshold.
Table 3.12 shows the results for Fama-MacBeth coefficients from year-by-year regressions, much like we described in the Main Results section of the paper, except
using only transactions with an LTV between 0.5 and 0.8. The results are quantitatively similar to those we obtain for the whole sample, which means that our main
results are not being driven by very low or very high LTVs. This reinforces our interpretation that our main results are caused by the CLL and not some other spurious
factor. The magnitude of the coefficients is very similar to the ones in the previous
table, but we lose statistical significance for the coefficient of interest when we use
the "Value Residual" measure as the left-hand side measure.
3.7.2
Different Bands
Table 3.14 shows that the result is very stable as we move away from the threshold of
CLL/0.8. In fact, the point estimates are indistinguishable from each other whether
we use a band of USD 5,000 or USD 10,000, which suggests that the difference in
the cost of credit is likely to be similar for these two sets of buyers relative to buyers
below the threshold. This is further evidence that the result is not driven solely by
buyers who choose to obtain a conforming mortgage and put up additional equity
from other sources.
153
3.7.3
Timing of the Control Group
We run an additional robustness test in which, instead of comparing the year in
which the limit is in effect with the subsequent year, we compare it to the previous
year. In this way, we are comparing houses that are never eligible for an 80 percent
conforming loan (those above the threshold) to transactions that initially are not
eligible, but become eligible once the limit changes. The research design is the same
as before, but we shift the window of analysis back one year. Table 3.13 shows the
Fama-MacBeth coefficients for this specification. The point estimates are smaller than
the ones in Table 3.13, but they are in the same direction and remain statistically
significant for the first years in the sample.
3.7.4
Pos-October Effect
One concern with our tests is that the conforming loan limit is announced in or
around October of each year, which might mean that the anticipation of a raise of
the conforming loan limit would confound our results. In order to address this issue,
we interact our main effect with the last three months of the year, to see if the
coefficients are being driven by this time period. Table 3.15 shows the results for this
specification, and we see that the estimates for the effect are the same for the last
three months of the year as they are for the first nine. The main effect is almost
unchanged.
3.7.5
Value per Square Foot by ZIP Code Income
In Figure 3-6, we split ZIP codes by their median income in order to consider the effect
of the conforming loan limit on the distribution of value per square foot on the whole
sample of transactions. We plot the average value per square foot as a function of the
distance of each transaction to the threshold of 125 percent of the CLL. We can see
that for the ZIP codes in the lowest quartile of the income distribution, the average
value per square foot is monotonically increasing for up to conforming loan limit
threshold, and from this point onwards the distribution becomes flat. This pattern
is not visible for zip codes with higher median incomes, where the distribution seems
monotonically increasing both below and above the threshold.
154
3.8
3.8.1
Appendix B. Data Manipulation
Data Cleaning
In order to clean the raw data received from Dataquick, we perform the following
modifications to the data:
Table 3.11: Data Cleaning Description
Criterion
Initial data
Transaction value equal to zero
Missing zipcode
Missing square feet
Mislabeled year
First loan greater than transaction value
House of less than 500 square feet
Transaction greater than 1,2 MM and smaller than 30 M
Company owned observation based on Dataquick flag
Company owned obs based on owner/seller/buyer information
Simple duplicated transactions
Value per square feet yearly outliers
Same property, date and buyer/seller information
Same property, and (late and no seller information
Same property, (late and transaction value
Same property, date and A sell to B and B sell to C
Special Transaction, based on Dataquick flag
Same property and (late, multiple sales in a day
Clean data
Ifemove single-family houses
Transaction greater than 600 M and smaller than 130 M
Whole sample for hedonic regressions
Transactions outside the 10k band for each year
Transactions used twice ( treatment in year t and control in
year t+1
Regression sample
Deleted Observations Remaining Observations
11,884,730
10,519,301
1,365,429
10,500,535
18,766
8,990,803
1,509,732
8,990,798
5
8,637,246
353,552
8,590,187
47,059
8,208,401
381,786
7,757,106
451,295
7,010,352
746,754
7,010,352
0
6,868,273
142,079
6,856,696
11,577
6,856,332
364
6,814,477
41,855
6,792,219
22,258
6,791,610
609
6,791,362
248
6,791,362
5,039,692
1,751,670
3,983,575
1,056,117
3,983,575
240,735
3,742,840
262,671
+21,936
262,671
Note: This table enumerates the steps taken in the data cleaning process and gives the number of
observations that are dropped in each step, as well as the remaining observations after each step.
Table 3.11 shows the number of observations deleted in each step of the data
preparation and a basic description of the criterion used to drop those observations
from the sample. In the following paragraphs, we categorize each step and describe
the criteria we used in detail, providing additional information about the data construction. We start with 11,884,730 observations.
Missing observations and outliers
We drop records with missing transaction value, house size, zip code, property
unique identifier, or mislabeled year.
155
- We drop a record if the house size is smaller than 500 square feet, as well as
records with transaction values smaller than three thousand and greater than
one million and two hundred thousand dollars.
- Value per square foot outliers per year: We drop observations that are above
the ninety-ninth percentile for the value per square foot variable or below the
first percentile each year.
Company owned observations
- We drop observations that Dataquick identifies as being bought by a corporation.
- Company owned observations based on owner/seller/buyer information: If the
owner, seller, or buyer names contain LLC, CORP, or LTD, the observation is
removed from the sample.
Duplicate transactions
Simple duplicated transactions: Remove records for which all the property information is the same.
Same property, date, and buyer/seller information: Drop observations that are
duplicated based on transaction value, date, and buyer/seller information.
Same property and date, no seller information: Drop observations for which the
property unique identifier and date are the same and have no seller information.
Same property, date, and transaction value: Drop observations for which property unique identifier, date, and transaction value are the same.
Same property and date and A sells to B and B sells to C: If person A sells to
B and B sells to C in the same date, we keep the most recent transaction.
Special transaction, based on Dataquick flag: This flag allows us to identify
records that are not actual transactions. For example, if a transaction was only
an ownership transfer without a cash transfer, this field is populated, allowing
us to delete this transaction.
Same property and date, multiple sales in a day: If a property is sold more than
twice during the same day, we keep only one transaction.
Additional information
We merge the Metropolitan Statistical Area (MSA) classification obtained from
the Census Bureau definition, using FIPS unique code identifier by county1 6 .
16 FIPS county code is a five-digit Federal Information Processing Standard (FIPS) code which
uniquely identifies counties and county equivalents in the United States, certain U.S. possessions,
and certain freely associated states. The first two digits are the FIPS state code and the last three
are the county code within the state or possession.
156
Change the second lien amount to missing if the first loan amount is equal
to the second loan amount, or if the second loan amount is greater than the
transaction value.
- Change the second lien amount to missing if combined loan to value (CLT) is
greater than two and loan to value (LTV) is equal to one.
- Change house age to missing if house age, calculated using transaction year
minus year built, is smaller than zero.
This procedure gives us our clean sample with 6,791,362.
Whole Sample for Hedonic Regression Sample
- We further restricted the sample for the hedonic regressions to transactions that
are between one hundred and thirty thousand and six hundred thousand dollars.
This selection aims to avoid that the estimates from the hedonic regression be
driven by transactions that are far from the region of interest.
This gives us our whole sample with 3,983,575 observations that are summarized in
the Summary Statistics section of the paper.
Regression Sample
Non-single-family houses: Our identification strategy relies on the change in
the conforming loan limit for single-family houses, therefore, we restrict our
attention to this type of house.
Transactions outside the USD 10,000 band for each year: Based on the threshold
value for each year that we describe in the Identification Strategy subsection,
we define a relevant transaction band around that threshold. For example, in
1999 the house threshold (1.25 of the conforming loan limit) is 300,000 dollars.
Therefore, we keep records with transaction values between 290,000 and 310,000
dollars that happened between 1999 and 2000. This subsample will be the
sample used to run the differences-in-differences specification using the 1999
threshold. For years when transaction bands overlapped, transaction will be
treatment in year t and controls in year t+1, and therefore used twice in the
empirical strategy
This gives us our regression sample with 262,671 observations
3.8.2
Variable Construction
In this appendix, we describe in more detail the variables used in the hedonic regressions. The hedonic regressions use two left-hand side variables: value per square
foot and price of each transaction. As we pointed out when we describe the hedonic
157
regression in the paper (Section 3.2), we use a similar set of controls as those used in
Campbell, Giglio, and Pathak (2010), and we add a few more characteristics.
The variables we use are interior square feet (linearly, high and low square feet dummies), lot size, bedrooms, bathrooms, total rooms, house age (linearly and squared),
type of house, an indicator for whether the house was renovated, an indicator for fireplace and parking, indicators for style of building (architectural style and structural
style), and additional indicators for type of construction, exterior material, heating
and cooling, heating and cooling mechanism, type of roof, view, attic, basement, and
garage.
While interior square feet, lot size, and age are included as continuous variables, all
the other controls are included as indicator variables.
Type of house: This variable is 1 if the house is a single-family house and 0 if
it is a condo or a multifamily property.
Bedrooms: This characteristic is divided into four categories (dummies): one
bedroom, two bedrooms, three bedrooms, and more than three bedrooms.
Bathrooms: This characteristic is divided into four categories: one bathroom,
one and a half bathrooms, two bathrooms, and more than two bathrooms.
Rooms: This characteristic is divided into five categories (dummies): one room,
two rooms, three rooms, four rooms, and more than four rooms.
Building Shape, Architectural Code, Structural Code, Exterior Material, Construction Code, Roof Code, View Code: These characteristics were divided based
on the numeric categorization of the original field. For example, construction
code was divided into 10 different categories that indicated the material used
on the framework of the building. In this case, we created 10 dummies based
on this categorization.
Heating and cooling: This information was divided into four categories: only
heating, only cooling, both heating and cooling, and heating-cooling information missing. The last variable was created to avoid dropping transactions for
which the information was not available.
Heating and cooling type: These characteristics were divided based on the numeric categorization of the original field. In this case, they discriminate the
type of cooling or heating system that is being used in the house.
158
- Garage and Garage Carport: A dummy is created to account for houses that
have garage surface greater than 50 square feet. For those transactions without
the information, a missing dummy is created for this category. Finally, we used
additional information to create a dummy that indicates if the houses have a
garage carport or not.
- Renovation: This variable accounts for the number of years since the last renovation. Based on this continuous variable, five categories (dummies) are defined:
missing renovation if the renovation date is missing or renovation period is negative, last renovation in less than 10 years, renovated between 10 and 20 years,
renovated between 20 and 30 years , and last renovation in more than or equal
to 30 years.
Attic: This characteristic is accounted for using a dummy for houses with an
attic greater than 50 square feet, and another dummy to account for missing
information about the attic in the houses.
Basement Finished and Unfinished: For the finished basement information, we
created a dummy for houses with basement size greater than 100 square feet, and
another dummy to account for missing information about the finished basement.
The same procedure is used to incorporate the information about unfinished
basement.
We use both the price of a transaction as well as the value per square foot as our
dependent variables. By estimating these regressions by year and by Metropolitan
Statistical Areas (MSA), we allow the coefficients on the characteristics to vary along
these two dimensions. We included monthly indicator variables to account for seasonality in the housing market, as well as zip code fixed effects. The set of controls
Xi is composed of all the variables described above, but in the case of the value per
square foot regression, we exclude the interior square feet continuous variables.
LHS, = -yo + IX, + monthi + zipcodej + Ei
When a record is missing the interior square feet, the lot size, the number of bedrooms
or bathrooms, or information on a houses age, we do not include this observation in
the hedonic regressions. This explains the difference between the number of observations for the value per square foot hedonic regressions (where we exclude interior
square footage) and the transaction value residual in our main regression results.
159
Figure 3-5: Fraction of Transactions with a Second Lien Loan by Year
Note: This figure shows the average fraction of transactions with a second lien loan by year for the
whole sample and the restricted sample used in the regression. Years 2007 and 2008 are excluded
from the regression sample because there was no change on the conforming loan limits on those
years.
0
0-
CD
0
XtD -
0
U--
0
-'-
1998
1999
2000
2002
2001
whole sample
160
2003
2004
2005
restricted sample
2006
Figure 3-6: Value per Square Foot by House Value and by ZIP Code Income
Note: This figure shows the average value per square foot plotted against the value of the house.
We split ZIP codes into quartiles according to their median income, where 1 includes the ZIP codes
in the lowest income quartile and 4 includes the ZIP codes with the highest median income. We use
the average of the median yearly income over the whole sample to place ZIP codes into the quartiles.
The x-axis is represented as one minus the transaction value as a percentage of each year's threshold
of 125 percent of the conforming loan limit (e.g. if the threshold is 200,000, a transaction of 150,000
will appear as -25 percent). The vertical red line is the threshold and the transactions for all years
are centered around that value.
2
1
A- OR
W
V -
A"-
-'04
-- W4PW *W
700-l
0
.0.
LL.
4
3
S
-04
0
0
7
-1100
-
woprl"- --
70/
0
0
-50
16o -ibo
so
Transaction Value as Percentage of 1.25CLL
161
!b
10
Figure 3-7: Income as a Percentage of CLL Threshold
Note: The horizontal axis indicates the difference between loan amounts and the conforming loan
limit as a percentage of the conforming loan limit. The figure plots average mortgage applicant
income computed from HMDA mortgage applications. We aggregate these proportions into 1%
bins and each dot in the figure represents the share of unused mortgages for each bin. We also plot
third degree polynomials (to the left and right of the conforming loan limit) as well as 95%
confidence intervals (dashed lines). Data extracted from HMDA, 1998-2006.
~0
C
0
6
;O
E
'C
Ok
I I
-50 -45
-40 -35
-30
-25
-20
-15
-10
1
i
-5
0
5
10
15
20
Distance from the conforming loan limit (%)
162
I
I
25
30
r -
35
40
45
50
Table 3.12: Effect of the CLL on House Valuation Measures, Constrained Sample
(0.5<LTV<0.8)
Panel A: Value Per Square Foot
Above Threshold
Year CLL
Above Threshold x
Year CLL
No. Obs.
All years
0.956**
(0.462)
-24.627***
(4.386)
-1.257***
(0.422)
190,450
1998-2001
1.584***
(0.556)
-15.935***
(2.576)
-1.610**
(0.646)
75,304
2002-2005
0.328
(0.650)
-33.319***
(5.726)
-0.904
(0.576)
115,146
Panel B: Log of Transaction Value Residual from Hedonic Regressions
Above Threshold
Year CLL
Above Threshold x
Year CLL
No. Obs.
All years
0.0118***
(0.0014)
0.0367***
(0.0038)
-0.0017
(0.0011)
183,643
1998-2001
0.0145***
(0.0017)
0.0335***
(0.0040)
-0.0019
(0.0022)
71,843
2002-2005
0.0090***
(0.0007)
0.0398***
(0.0066)
-0.0015*
(0.0008)
111,800
Panel C: Value Per Square Foot Residual from Hedonic Regressions
Above Threshold
Year CLL
Above Threshold x
Year CLL
No. Obs.
All years
1.565***
(0.298)
3.431***
(0.550)
-0.931***
(0.260)
183,789
1998-2001
1.958***
(0.356)
3.470***
(0.417)
-1.085***
(0.413)
71,917
2002-2005
1.172***
(0.431)
3.392***
(1.113)
-0.777**
(0.360)
111,872
Note: This table shows Fama Macbeth coefficients computed from year by year regressions that
use three alternative measures of valuation as the dependent variable in each of the three panels.
The hedonic regressions that produce the residuals for panels B and C are described in Section 3.2.
The sample for each year's regression includes transactions within +/- USD 10,000 of that year's
conforming loan limit, as well as transactions in the same band in the subsequent year. Unlike the
main regression table in the paper, the sample for these regressions is constrained to transactions
with an LTV between 0.5 and 0.8. All year by year regressions include ZIP code fixed effects. Above
the Threshold refers to transactions up to USD 10,000 above the conforming loan limit divided by
0.8 (i.e. the transactions that were "ineligible" to be bought with a conforming loan at a full 80
percent LTV) and Year CLL is the year in which the conforming loan limit is in effect.
163
Table 3.13: Effect of CLL on Valuation Measures - Alternative Timing of the Control
Group
Panel A: Value Per Square Foot
Below Threshold
Pre-Year CLL
Below Threshold X
Pre-Year CLL
No. Obs.
All years
0.012
(0.236)
-23.739***
(4.391)
-0.375
(0.473)
227,325
All Transactions
1999-2002 2003-2006
-0.005
0.029
(0.282)
(0.423)
-15.890***
-31.588***
(2.489)
(6.534)
-0.817
0.068
(0.549)
(0.783)
93,612
133,713
0.5<LTV<0.8 Transactions
All years
1999-2002 2003-2006
0.522*
0.628
0.417
(0.270)
(0.412)
(0.404)
-25.061***
-16.995***
-33.127***
(4.636)
(2.666)
(7.057)
-0.555
-0.812***
-0.298
(0.434)
(0.233)
(0.884)
168,865
66,072
102,793
Panel B: Transaction Value Residual from Hedonic Regressions
All years
Below Threshold . -0.0099***
(0.0010)
Pre-Year CLL
0.0346***
(0.0045)
Below Threshold X
0.0000
Pre-Year CLL
(0.0016)
No. Obs.
217,410
All Transactions
1999-2002
2003-2006
-0.0106***
-0.0092***
(0.0010)
(0.0017)
0.0342***
0.0350***
(0.0037)
(0.0089)
-0.0019
0.0019
(0.0021)
(0.0023)
88,416
128,994
0.5<LTV<0.8 Transactions
All years
1999-2002 2003-2006
-0.0086***
-0.0087***
-0.0085***
(0.0011)
(0.0018)
(0.0015)
0.0342***
0.0334***
0.0350***
(0.0045)
(0.0042)
(0.0088)
-0.0011
-0.0031
0.0009
(0.0016)
(0.0023)
(0.0020)
162,584
62,897
99,687
Panel C: Value Per Square Foot Residual from Hedonic Regressions
Below Threshold
Pre-Year CLL
Below Threshold X
Pre-Year CLL
No. Obs.
All years
-0.903***
(0.289)
3.215***
(0.712)
-0.175
(0.351)
217.804
All Transactions
1999-2002
2003-2006
-0.881***
-0.925
(0.197)
(0.593)
3.019***
3.411**
(0.529)
(1.436)
-0.605**
0.256
(0.245)
(0.625)
88,613
129,191
0.5<LTV<0.8 Transactions
All years
1999-2002
2003-2006
-0.524**
-0.446**
-0.603
(0.208)
(0.206)
(0.395)
2.852***
2.591***
3.112**
(0.699)
(0.547)
(1.392)
-0.467
-0.915***
-0.020
(0.315)
(0.130)
(0.560)
162,788
62,997
99,791
Note: Table shows Fama McBeth coefficients computed from year by year regressions that use
three alternative measures of valuation as the dependent variable in each of the three panels. The
sample includes all transactions within USD 10,000 of each year's conforming loan limit, as well
as transactions of the same amount in the previous year (unlike the previous tables where we use
the subsequent year). In this table we include the results for all transactions, as well as those for
the sample that is restricted to having an LTV between 0.5 and 0.8. Below the Threshold refers to
transactions up to USD 10,000 below the conforming loan limit at year t divided by 0.8 (i.e. the
transactions that were "eligible" to be bought with a conforming loan at a full 80 percent LTV in year
t , but were "ineligible" in year t-1) and Pre-Year CLL is the previous year in which the conforming
loan limit is in effect. This specification makes the interaction coefficient directly comparable to the
main regression on signs and magnitudes.
164
Table 3.14: Effect of the CLL on Valuation - Alternative Bands
Panel A: Value Per Square Foot
Above Threshold
Year CLL
Above Threshold x
Year CLL
No. Obs.
10K
1.261**
(0.494)
-22.869***
(4.047)
-1.162***
(0.264)
262,671
Ok to 5K
0.969
(0.722)
-23.008***
(3.988)
-1.064*
(0.556)
134,117
5K to 1OK
1.406***
(0.544)
-23.194***
(4.177)
-1.181**
(0.581)
128,554
Panel B: Log of Transaction Value Residual from Hedonic Regressions
Above Threshold
Year CLL
Above Threshold x
Year CLL
No. Obs.
10K
0.0129**
(0.0013)
0.0387***
(0.0041)
-0.0017***
(0.0008)
251,431
Ok to 5K
0.0071
(0.0019)
0.0384***
(0.0045)
-0.0015*
(0.0011)
128,429
5K to 10K
0.0180***
(0.0013)
0.0389***
(0.0038)
-0.0023**
(0.0016)
123,002
Panel C: Value Per Square Foot Residual from Hedonic Regressions
Above Threshold
Year CLL
Above Threshold x
Year CLL
No. Obs.
10K
1.733***
(0.360)
4.103***
(0.644)
-0.651***
(0.238)
251,764
Ok to 5K
1.255*
(0.700)
4.052***
(0.678)
-0.712
(0.508)
128,601
5K to 1OK
2.110***
(0.387)
3.946***
(0.763)
-0.623***
(0.238)
123,163
Note: This table shows Fama MacBeth coefficients computed from year by year regressions that
use three alternative measures of valuation as the dependent variable in each of the three panels.
The hedonic regressions that produce the residuals for panels B and C are described in Section
3.3.2. The sample for each year's regression includes all transactions within +/- USD 10,000 of that
year's conforming loan limit, as well as transactions in the same band in the subsequent year. All
year by year regressions include ZIP code fixed effects. Above the Threshold refers to transactions
up to USD 10,000 above the conforming loan limit divided by 0.8 (i.e. the transactions that were
"ineligible" to be bought with a conforming loan at a full 80 percent LTV) and Year CLL is the year
in which the conforming loan limit is in effect.
165
Table 3.15: Effect of CLL on Valuation: Post October
Panel A: Value Per Square Foot
Above Threshold
0.000
Year CLL
0.000
Above Threshold x
Year CLL
Above Threshold x
Year CLL x Post October
No. Obs.
1998-2005
1.261**
(0.625)
-22.869***
(5.119)
-1.162***
(0.334)
262,671
1998-2005
1.039*
(0.531)
-23.460***
(5.079)
-1.086***
(0.393)
-0.213
(1.031)
262,671
Panel B: Log of Transaction Value Residual from Hedonic Regressions
Above Threshold
Year CLL
Above Threshold x
Year CLL
Above Threshold x
Year CLL x Post October
No. Obs.
1998-2005
0.0129***
(0.0016)
0.0387***
(0.0052)
-0.0017*
(0.0010)
251,431
1998-2005
0.0132***
(0.0014)
0.0398***
(0.0056)
-0.0027**
(0.0013)
0.0033
(0.0027)
251,431
Panel C: Value Per Square Foot Residual from Hedonic Regressions
Above Threshold
Year CLL
Above Threshold x
Year CLL
Above Threshold x
Year CLL x Post October
No. Obs.
1998-2005
1.733***
(0.456)
4.103***
(0.815)
-0.651**
(0.301)
251,764
1998-2005
1.751***
(0.407)
4.176***
(0.813)
-0.696**
(0.277)
0.031
(0.805)
251,764
Note: This table shows Fama MacBeth coefficients computed from year by year regressions that
use three alternative measures of valuation as the dependent variable in each of the three panels.
The sample for each year's regression includes all transactions within +/- USD 10,000 of that year's
conforming loan limit, as well as transactions in the same band in the subsequent year. Above the
Threshold refers to transactions up to USD 10,000 above the conforming loan limit divided by 0.8
(i.e. the transactions that were "ineligible" to be bought with a conforming loan at a full 80 percent
LTV) and Year CLL is the year in which the conforming loan limit is in effect. This specification
interacts the diff-in-diff specification with a dummy variable that is 1 in October, November and
December of each year.
166
Table 3.16: Effect of the CLL on House Valuation with In-Sample Controls
Panel A: Value Per Square Foot
Above Threshold
Year CLL
Above Threshold x
Year CLL
No. Obs.
All years
2.926***
(0.366)
-15.158***
(2.706)
-0.771**
(0.299)
251,764
1998-2001
3.272***
(0.416)
-9.681***
(1.206)
-1.211***
(0.428)
103,709
2002-2005
2.581***
(0.612)
-20.634***
(3.567)
-0.332
(0.327)
148,055
1998-2001
0.0323***
(0.0011)
-0.0005***
(0.0001)
-0.0001
(0.0001)
103,535
2002-2005
0.0239***
(0.0011)
-0.0004***
(0.0001)
0.0001
(0.0001)
147,896
Panel B: Log of Transaction Value
Above Threshold
Year CLL
Above Threshold x
Year CLL
No. Obs.
All years
0.0281***
(0.0018)
-0.0004***
(0.0001)
0.0000
(0.0000)
251,431
Note: This table shows Fama MacBeth coefficients computed from year by year regressions that use
two alternative measures of valuation as the dependent variable in each of the two panels. Instead
of using residuals from a hedonic regression, all characteristics of the houses are included as controls
within the estimation sample. The sample for each year's regression includes all transactions within
+/- USD 10,000 of that year's conforming loan limit, as well as transactions in the same band in the
subsequent year. All year by year regressions include ZIP code fixed effects. Above the Threshold
refers to transactions up to USD 10,000 above the conforming loan limit divided by 0.8 (i.e. the
transactions that were "ineligible" to be bought with a conforming loan at a full 80 percent LTV)
and Year CLL is the year in which the conforming loan limit is in effect.
167