Chapter 1 Personal Bankruptcy and Household Debt 1.1 Introduction
Transcript of Chapter 1 Personal Bankruptcy and Household Debt 1.1 Introduction
MASSACHUSETS INSTRWEOF TECHNOLOGY
Essays in Financial EconomicsMAY 1 5 2014
by
Felipe Severino LIBRARIES
B.Sc., Pontificia Universidad Catolica de Chile, 2005M.Sc., Pontificia Universidad Catolica de Chile, 2007
Submitted to the Alfred P. Sloan School of Managementin 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 redactedAuthor................ ...........................
Alfred Sloa chool of ManagementMay 2, 2014
Signature redactedCertified by....................... Antoinette Schoar
Michael Koerner '49 Professor of Entrepreneurial FinanceThesis Supervisor
Signature redactedAccepted by.......... .......
Ezra ZuckermanDirector, Sloan School of Management PhD Program
2
Essays in Financial Economics
by
Felipe Severino
Submitted to the Alfred P. Sloan School of Managementon May 2, 2014, in partial fulfillment of the
requirements for the degree ofDoctor 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 ex-
plore 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 as-
sets 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 com-
position 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 cap-
ital 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 collat-
3
eral lending channel explains 15-25 % of employment variation. In the third chapter(co-authored with Manuel Adelino and Antoinette Schoar) we use exogenous changesin the conforming loan limit as an instrument for lower cost of financing, and showthat cheaper credit significantly increases house prices. Houses that become eligiblefor financing with a conforming loan show an increase in value of 1.16 dollars persquare foot (for an average price per square foot of 220 dollars). These coefficientsare consistent with a local elasticity of house prices to interest rates that is lower thansome previous studies proposed (below 10). In addition, loan to value ratios aroundthe 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 otherforms of financing were less common and substantially more expensive. Results arealso stronger in zip codes where personal income growth is low or declining, and inregions with lower elasticity of housing supply.
Thesis Supervisor: Antoinette SchoarTitle: 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 par-
ents, 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 Personal Bankruptcy and Household Debt1.1 Introduction .... ...................1.2 Bankruptcy Procedure and Related Literature
1.2.1 Institutional Framework . . . . . . . .1.2.2 Related Literature . . . . . . . . . . .
1.3 Data and Summary Statistics . . . . . . . . .1.3.1 Data Description . . . . . . . . . . . .1.3.2 Summary Statistics . . . . . . . . . . .
1.4 Empirical Hypothesis . . . . . . . . . . . . . .1.5 Empirical Strategy . . . . . . . . . . . . . . .1.6 Results and discussion . . . . . . . . . . . . .
1.6.1 Bankruptcy Protection and HouseholdR ates . . . . . . . . . . . . . . . . . .
1.6.2 Robustness Test . . . . . . . . . . . . .
Leverage and Interest
1.6.3 Magnitude of the effect .
1.71.8
1.6.4 Borrowers, Delinquency and Self-EmploymentConclusion . . . . . . . . . . . . . . . . . . . . . . . .
Bibliography . . . . . . . . . . . . . . . . . . . . . . .1.9 Appendix A. Model of Effect of Bankruptcy Protection on Household
B orrow ing . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
2 House Prices, Collateral and Self-Employment2.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .2.2 Data and Empirical Methodology . . . . . . . . . . . . . . . . . . . .
2.2.1 Data Description . . . . . . . . . . . . . . . . . . . . . . . . .2.2.2 Summary Statistics . . . . . . . . . . . . . . . . . . . . . . . .2.2.3 Empirical Model . . . . . . . . . . . . . . . . . . . . . . . . .
2.3 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 . . . . . .
2.4 C onclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .2.5 Bibliography . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
7
1313191922242426272932
323435353738
42
7373777780818484909091919293
.
.
2.6 Appendix. Calculating the magnitude of the collateral effect
3 Credit Supply and House Prices: EvidenceSegmentation3.1 Introduction . . . . . . . . . . . . . . . . . .3.2 The User Cost Model . . . . . . . . . . . . .3.3 Data and Methodology . . . . . . . . . . . .
3.3.1 Summary Statistics . . . . . . . . . .3.3.2 Hedonic Regression . . . . . . . . . .3.3.3 Empirical Approach . . . . . . . . .
3.4 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 . .
3.5 Conclusion . . . . . . . . . . . . . . . . . . .3.6 Bibliography . . . . . . . . . . . . . . . . . .3.7 Appendix A. Robustness and Refinements -
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
3.8 Appendix B. Data Manipulation . . .3.8.1 Data Cleaning . . . . . . . . .3.8.2 Variable Construction . . . .
Code I
from Mortgage Market115
.dditional
ncome
Tests
115119120120121122128128129130133135137153153153154154154155155157
8
105
List of Figures
1-1 Debt Growth and Bankruptcy Filings . . . . . . . . . . . . . . . . . . 441-2 States that Changed their Level of Bankruptcy Protection . . . . . . 451-3 Ilustration of Different Demand and Supply Responses . . . . . . . . 461-4 Ilustration of a Solution of the Model . . . . . . . . . . . . . . . . . . 47
3-1 Transaction-Loan Value Surface . . . . . . . . . . . . . . . . . . . . . 1393-2 Borrower Composition for the Regression Sample . . . . . . . . . . . 1403-3 Frequency of Transactions as Percentage of CLL Threshold . . . . . . 1413-4 Share of Unused Mortgage Applications . . . . . . . . . . . . . . . . . 142
3-5 Fraction of Transactions with a Second Lien Loan by Year . . . . . . 1603-6 Value per Square Foot by House Value and by ZIP Code Income . . . 1613-7 Income as a Percentage of CLL Threshold . . . . . . . . . . . . . . . 162
9
10
List of Tables
1.1 Summary Statistics Data. . . . . . . . . . . . . . . . . . . . . . . . . 48
1.2 Summary Statistics Protection Level . . . . . . . . . . . . . . . . . . 49
1.3 Effect of Bankruptcy Protection on Debt. Credit Card Debt . . . . . 50
1.4 Effect of Bankruptcy Protection on Debt. Mortgage Debt . . . . . . . 51
1.5 Effect of Bankruptcy Protection on Debt. Auto Debt . . . . . . . . . 52
1.6 Determinants of Bankruptcy Protection Levels and Changes . . . . . 53
1.7 Dynamics of the Change in Protection Levels on Credit Card Debt . 54
1.8 Local Business Conditions. Neighboring County-pairs across State
Borders. Credit Card Debt . . . . . . . . . . . . . . . . . . . . . . . . 55
1.9 Heterogeneous Treatment of Bankruptcy Protection on Credit Card
Debt: Income and Home ownership . . . . . . . . . . . . . . . . . . . 56
1.10 Effect of Bankruptcy Protection on Interest Rates: Personal Unsecured
Loans and Credit Cards . . . . . . . . . . . . . . . . . . . . . . . . . 57
1.11 Effect of Bankruptcy Protection on Interest Rates: Mortagage Credit 58
1.12 Effect of Bankruptcy Protection on Debt. Number of Credit Cards
and E ntry . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 59
1.13 Effect of Bankruptcy Protection on Credit Card Delinquency . . . . . 60
1.14 Effect of Bankruptcy Protection on Self-Employment . . . . . . . . . 61
1.15 Effect of Bankruptcy Protection on Credit Card Debt. Alternative
Specifications . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 62
1.16 Other Heterogeneous Treatment of Bankruptcy Protection. Credit
C ard D ebt . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 63
1.17 Determinants of Bankruptcy Protection Levels and Changes. Eventu-
ally Treated . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 64
1.18 Dynamics of the Change in Protection. Mortgage Debt . . . . . . . . 65
1.19 Dynamics of the Change in Protection. Auto Debt . . . . . . . . . . 66
1.20 Local Business Conditions. Neighboring County-pairs across State
Borders. Mortgage Debt . . . . . . . . . . . . . . . . . . . . . . . . . 67
1.21 Local Business Conditions. Neighboring County-pairs across State
Borders. Auto Debt . . . . . . . . . . . . . . . . . . . . . . . . . . . . 68
1.22 Heterogeneous Treatment of Bankruptcy Protection: Income and Home-
ownership. Mortgage Debt . . . . . . . . . . . . . . . . . . . . . . . . 69
1.23 Heterogeneous Treatment of Bankruptcy Protection: Income and Home-
ownership. Auto Debt . . . . . . . . . . . . . . . . . . . . . . . . . . 70
1.24 Effect of Bankruptcy Protection on County Delinquency Proportions 71
11
1.25 Effect of Bankruptcy Protection on Debt After Bankruptcy Reform 2005 72
2.1 Sum m ary Statistics . . . . . . . . . . . . . . . . . . . . . . . . . . . . 962.2 Employment Growth, Firm Size, and House Price Appreciation . . . 972.3 Employment Growth and House Prices: Excluding Construction, Non-
Tradable, and Finance Industries and Considering Manufacturing Only 982.4 Breakdown of Manufacturing Industries by Distance Shipped . . . . . 992.5 Employment and House Price Appreciation across Industry Types . . 1002.6 Proprietorships and House Price Appreciation . . . . . . . . . . . . . 1012.7 Employment Growth, Firm Size, and House Price Appreciation, Crisis
Period (2007-2009) . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1022.8 Total Employment, Unemployment, and Migration . . . . . . . . . . 1032.9 D enial R ates . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1042.10 Employment Growth, Firm Size, and House Price Appreciation: Indi-
vidual Industries by Firm Size . . . . . . . . . . . . . . . . . . . . . . 1072.11 Robustness Test: Difference between High and Low Start-up Capital 1082.12 Effect of One Standard Deviation Change in the Independent Variable 1092.13 Dollar-weighted Average Distance Shipped in Manufacturing (miles) . 1102.14 Detail on Average Start-up Amount by 2-digit NAICS Sector . . . . . 1112.15 Distance Shipped and Share of Employees at Large Establishments . 1122.16 House Price Growth and Creation of Establishments . . . . . . . . . . 1132.17 List of 3-digit NAICS Industries Excluding Non-tradables, Manufac-
turing, F.I.R.E., and Construction . . . . . . . . . . . . . . . . . . . . 114
3.1 Sum m ary Statistics . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1433.2 Summary Statistics by Geography and Year . . . . . . . . . . . . . . 1443.3 Verification of the Impact of the CLL on Financing Choices . . . . . . 1453.4 Impact of CLL on Number of Transactions . . . . . . . . . . . . . . . 1463.5 Effect of the CLL on House Valuation Measures . . . . . . . . . . . . 1473.6 Effect of the CLL on House Valuation in Different Income Growth Areas1483.7 Placebo Test for Coefficient of Interest . . . . . . . . . . . . . . . . . 1493.8 Effect of the CLL on the Valuation of Different Groups of Transactions 1503.9 Effect of the CLL on House Valuation in Low Supply Elasticity Areas
( Elasticity< 1) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1513.10 Elasticity Estim ates . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1523.11 Data Cleaning Description . . . . . . . . . . . . . . . . . . . . . . . . 1553.12 Effect of the CLL on House Valuation Measures, Constrained Sample
(0.5<LTV < 0.8) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1633.13 Effect of CLL on Valuation Measures - Alternative Timing of the Con-
trol G roup . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1643.14 Effect of the CLL on Valuation - Alternative Bands . . . . . . . . . . 1653.15 Effect of CLL on Valuation: Post October . . . . . . . . . . . . . . . 1663.16 Effect of the CLL on House Valuation with In-Sample Controls . . . . 167
12
Chapter 1
Personal Bankruptcy andHousehold 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
an increase in personal bankruptcies, which peaked in 2005 with 2.04 million filings. 1
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 protectionhenceforth) 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 borrowershave 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
2Furthermore, 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 abovethat 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 increasingthe insurance function of bankruptcy: in bad states of the world the borrower declaresbankruptcy and, as a result of the higher protection level, is allowed to keep a largerproportion 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 ofprotection will also affect the composition of borrowers: more risk averse borrowersmight 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 themix 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 musttherefore rise in equilibrium; but depending on which effect dominates (demand orsupply), 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 adifference in difference design to identify their effect on the credit market equilib-rium. We find that bankruptcy protection laws increase borrowers' unsecured creditholdings, mainly credit cards, leaving their level of secured debt - mortgage and autoloans - unchanged. At the same time we find an increase in the interest rate forunsecured credit, but not for other types of credit. These results are predominantlydriven by low-income areas, and suggest that bankruptcy protection levels provideimportant downside insurance, which has first order effects on the supply and alsoon the demand for credit. Interestingly, using detailed individual data, we do notfind an increase in default rates, which suggests that households are not necessarilyover-borrowing or risk shifting as a response to the increase in protection.
Empirically identifying the true effect of bankruptcy protection levels on householdleverage is challenging, as these levels are correlated with unobservable borrower andlender characteristics that might simultaneously affect credit availability, and thelevel of protection. For example, states with higher protection levels may be statesin which households are less financially savvy, or they might be states with higherhouse prices, and therefore more willing to take on more debt. This in turn will leadto a positive correlation between debt and protection.
Therefore, we exploit changes in the dollar amounts of asset protection underbankruptcy to identify the effect of this protection on household debt5 . Our identifi-cation benefits from the fact that states increased bankruptcy protection at differenttimes and by different amounts over our sample period. We show that changes in
3Non-state contingent contracts are a key friction here; in the absence of this friction, the effectof personal bankruptcy protection on household borrowing disappears. One possible explanation forwhy lenders do not offer more flexible contracts (more protection in "bad" states, or state contingentrepayment) is that these lenders could face a collective action problem: if only one lender offeredsuch a contract it would attract predominantly bad type borrowers, which is not an equilibrium.Alternatively, customized state contingent contracts could be hard to enforce.
4For a more developed model see Appendix A.5Asset 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 onlyhomestead exemption.
14
protection levels are uncorrelated with macroeconomic conditions and other deter-
minants of credit equilibrium, most importantly changes in state level house prices
and unemployment rates. This allows us to disentangle the effect of bankruptcy pro-
tection 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 systemat-
ically 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
6We 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 levelswere implemented in our sample period (i.e. "eventually" treated). In this case theeffects we estimate are unchanged.
We also look at the dynamics of the changes. By analyzing the timing, we canrule out that the level of protection may be correlated with pre-existing state specifictrends that survive our controls, and thus that our results are a reflection of thesedifferential pre-trends rather than changes in the levels of protection. We show thatour 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 statisticallyinsignificant.'
We now explore the heterogeneity of the average treatment effect. Exploitingwithin-state variation on the levels of debt held by counties, we find a stronger in-crease in the level of unsecured debt held by lower-income counties'. These resultsare consistent with the fact that increases in personal bankruptcy protection levelsimprove risk-sharing; this improvement should be stronger for lower-income regions,as they have fewer resources to diversify their risk exposure than wealthier ones, forwhich 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 equityof the borrower. In line with a demand driven channel, we find that the effect is almostthree times stronger in areas where homeownership is higher, after we condition on thelevel of income. Also conditioning on the level of income10 , we find that the increasein credit is stronger in areas where the banking industry is more concentrated (fewerbanks), which is consistent with the relationship lending model proposed Petersenand Rajan (1995), where creditors are more likely to finance a credit constrainedborrower when credit markets are concentrated, because it is easier for these creditorsto internalize the benefits of assisting these borrowers; although this is only suggestiveevidence.
Overall we find that the average credit card balance in a county in our period is290 million dollars in credit card debt, and the average increase in credit card debt is7.6%. Our main estimate explains 10% of this balance growth". However, this valuemore 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-timeunobserved 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 in1999.
' 0Homeownership and bank concentration are correlated with income at the county level. There-fore, looking at cross-sectional variation without controlling for income is not informative, as itprovides confounding information within all the correlated variables. In order to overcome this limi-tation of the data, we replicated the specification of interest for each income subgroup; this strategyproved to be useful. For example, under this setup, unemployment heterogeneity within incomegroups has no cross-sectional implications. However, homeownership and bank concentration stillprovide 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 exemptionlevel of 70k dollars. This value is a more conservative measure than using one standard deviation oflevel (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:
homeowners who are expecting to be close to distress level on their credit cards 2 .
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 geograph-
ical 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 re-
sponse 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 num-
ber 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 house-
holds 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 the13
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
1 2Appendix B2 shows that within low-income areas the effect is differentially stronger for areas
with a higher proportion of credit card delinquency (90+).13Also, 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.14 For example, construction, photography, and other low capital-intensity industries that can be
financed with credit card debt.
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 withlarge changes in protection as treated states, and to the use of an indicator instead ofthe magnitude of the change. Given the nature of our empirical strategy, as we arguebefore, time-varying changes at state levels may be omitted variables explaining ourresults; one candidate is the level of unemployment insurance in each state (Hsu etal., 2012). However, the inclusion of this variable has no impact on the estimatedcoefficient. 5
Our results suggest that existing borrowers increase their leverage without in-creasing their ex post delinquency, consistent with risk-averse, constrained borrowersreacting to the increase in insurance. We cannot say anything about the welfare ef-fect of these changes. In a world with complete markets, increases in protection willconstrain the contract space and therefore may lead to inefficiencies. Furthermore, inthe presence of limited commitment, harsher penalties for defaulting could improvewelfare ex ante (Kehoe and Levine 1993, Alvarez and Jermann 2000). However, ifstate contingent contracts are not available (i.e. incomplete markets), a pro-debtorbankruptcy code could lead to welfare gains (Link 2004). Therefore, theoretically theeffect of increased bankruptcy protection on welfare is undetermined, and dependenton modeling choices.
A number of earlier papers have looked at the cross sectional relationship betweenthe level of bankruptcy protection and consumer credit. See for example Gropp etal. (1997), the first paper to examine this relationship. Using household data fromthe 1983 Survey of Consumer Finances, they found that higher levels of protectionwere associated with both reduced credit availability for low-asset households and in-creased debt balances among higher-asset households. Similarly Berger et al. (2010)found that higher protection is associated with lower access to credit for unlimitedliability firms. Also, Lin and White (2001) found the same relationship for mortgagecredit. The recent legislative history of staggered introduction of bankruptcy exemp-tions in combination with household data allows us to identify the effects of changesin bankruptcy protection on the change in the supply and demand of credit for differ-ent types of debt. Most importantly, we find that an increase in personal bankruptcyprotection leads to an increase in the amount of unsecured debt held by households,leaving secured debt unchanged. Therefore, using an improved empirical strategy, wesee that the demand effect of bankruptcy protection, arguably driven by improvedrisk-sharing, dominates its supply-deterring effects. Hence increased bankruptcy pro-tection increases equilibrium debt reliance, particularly for low-income homeowners.
Increases in personal bankruptcy protection results in a weakening of creditorrights. There is a vast literature in corporate finance that has examined the effect of
15As a case study during our relevant sample period, 1999-2005, one state went from havingsome level of protection to unlimited protection. When we include this time-varying dummy inthe regression, we find that the main effect is unchanged, but the unlimited protection dummy isnegative and significant for mortgage and credit card debt. This suggests that the effect of protectionis a non-linear function of the level of exemption, and therefore above a certain threshold lendersincrease prices to a magnitude which decreases quantities.
18
changes in creditor protection on debt (La Porta et al. 1998, Levine 1998, Djankovet al. 2007). Most related to this paper is Vig (2013), which looks at increasesin 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 responseis driven by an increase in the threat of early liquidation due to the increase increditor 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 3outlines 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 Bankruptcy Procedure and Related Literature
1.2.1 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 bor-
rower (such as letters, wage garnishment, telephone calls, and lawsuits). Most con-
sumer 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 Pro-
tection 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
16See 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 equityin a vehicle. In addition, nearly all states have some level of homestead protection forequity in owner-occupied homes, but the levels vary from a few thousand dollars, tounlimited amounts in six states, including Texas, Florida, and DC". This exemptionlevel is what we refer to here as the protection level. Under Chapter 7, debtors mustuse their non-protected assets to repay creditors, but they are not obliged to use anyof their future income to make repayments.
Under the alternative procedure in Chapter 13, bankrupts are not obliged torepay from assets, but they must use part of their post-bankruptcy income to makerepayments. Before 2005, there was no predetermined income exemption; on thecontrary, 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-protectedassets under Chapter 7. Also, borrowers were not allowed to repay less than thevalue of their non-protected assets and, since they had always the option to file underChapter 7, they had no incentive to offer any more. Judges did not need the approvalof creditors to approve repayment plans.18
The cost of filing for bankruptcy before 2005 was low: about 600 dollars underChapter 7, and 1,600 dollars under Chapter 13, as of 2001 (White 2007). The punish-ment for bankruptcy included making bankrupts' names public and the appearanceof 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, (butthey were allowed to file under Chapter 13 as often as every six months)19 .
Overall, these features made US bankruptcy law very pro-debtor. Since debtorscould choose between the procedures under Chapters 7 and 13, they would select theprocedure which would maximize their gain from filing. Around three quarters of allthose filing for bankruptcy used Chapter 7 (Flynn and Bermant, 2002). Most debtorswho filed under Chapter 13 did so because their gains were even higher using this
17See Table 1.2 for summary statistics of the level of protection.18Even 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 con-sidering filing for bankruptcy has 40,000 dollars in assets and is located in a state in which theprotection level is 20,000 dollars. Since the household would have 20,000 dollars of unprotectedassets if filing under Chapter 7, it would be willing to repay no more than 20,000 dollars (in presentvalue) from future income if it were to file under Chapter 13. As a result of this close relation-ship between Chapter 7 and Chapter 13 bankruptcy filings, we assume that changes in Chapter 7protection levels will affect household willingness to file for bankruptcy (either under Chapter 7 or13).
19 US bankruptcy law allowed additional debt to be discharged under Chapter 13. Debtors' carloans could be discharged to the extent that the loan principal exceeded the market value of thecar (negative equity). Also, debts acquired by fraud and cash advances obtained shortly beforefiling could be discharged under Chapter 13, but not under Chapter 7. These characteristics wereknown as the Chapter 13 "super-discharge", and some households took advantage of the situationby filing first under Chapter 7, where most of their debts were discharged, and then converting theirfilings to Chapter 13, where they proposed a plan to repay part of the additional debt covered underChapter 13. This two-step procedure, known as "Chapter 20", increased borrowers' financial gainsfrom bankruptcy as opposed to filing under either procedure separately.
20
procedure than under Chapter 7.
The Bankruptcy Abuse Prevention and Consumer Protection Act
The Bankruptcy Abuse Prevention and Consumer Protection Act (BAPCPA) of 2005made several major changes to bankruptcy law. First, it abolished the right of debtorsto 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 familyincome 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 deter-
mines 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, in-
cluding 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 require-
ments 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
the cost of the two training courses, and the preparation of tax returns.2 1
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 filings22 . 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.
2 0BAPCPA defines disposable income as the difference between debtors' average monthly family
income during the six months prior to filing, with a new income exemption.2 'A large proportion of the cost is attributable to the fact that bankruptcy lawyers can be fined
if debtors' information is not accurate.22BAPCPA 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 "super-
discharge" effect.
21
1.2.2 Related Literature
Gropp et al. (1997) was the first paper to use household level debt data to look atthe difference on credit availability for different levels of protection. Using the Sur-vey of Consumer Finance of 1983, they found that higher protection under personalbankruptcy is associated with a lower probability of access to credit, and a lower levelof debt for low asset households, in states with more generous bankruptcy exemp-tions. Using detailed bank information, Berger et al. (2010) found that unlimited lia-bility small businesses have lower access to credit in states with more debtor-friendlybankruptcy laws. In addition, these businesses face harsher loan terms: they aremore likely to pledge business collateral, have shorter maturities, pay higher rates,and borrow smaller amounts. Also, Lin and White (2001) looked at how the protec-tion levels affect the availability of mortgage credit application granting, finding thataccepted applications are negatively correlated with the level of protection. However,all these studies use cross-sectional variation on protection to look at how these levelscorrelate with credit availability. Hynes et al. (2004) find that state levels of exemp-tions are correlated with bankruptcy filing rates and state redistributional policies tohelp 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 treatprotection levels as exogenous variables. This paper contributes to this literatureusing state time variation in bankruptcy protection levels to overcome these endo-geneity concerns when looking at relationship between bankruptcy protection andcredit markets. Using this empirical strategy we find that increases in bankruptcyprotection 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 Pe-nas (2011), who use state level variation in the level of bankruptcy protection tolook at start-up creation, finding that increases in protection decrease start-up per-formance; and to Cerqueiro et al. (2013), who uses a similar strategy to look atthe effect of personal bankruptcy laws on innovation, finding that there is an aggre-gate decrease in the level of innovative activity among small firms in places in whichprotection increased. The effect of the use of credit cards in entrepreneurial activ-ity has also been studied by Chatterji and Seamans (2012). Using states' removal ofcredit card interest rate ceilings in 1978 they show that this deregulation increases theprobability of entrepreneurial entry, arguably through an access to finance channel.Finally, Fan and White (2003) find that personal bankruptcy protection motivatesentrepreneurial activity using cross-sectional variation in the level of protection. Inthis paper, we show that increases in bankruptcy protection are correlated with in-creases in self-employment. Although we cannot rule out a demand channel, it seemsthat bankruptcy laws could have an expansive impact on self-employment throughan increase in the credit channel.
Bankruptcy laws directly affect unsecured debt, given that secured debt cannot bedischarged. Therefore this paper is related to the literature on credit card borrowing.Agarwal et al. (2013), analyze the effectiveness of consumer financial regulation in thecredit card market, using the 2009 credit card reform. They find that regulatory limitson 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 time-
varying 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 lever-
age 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 pa-
per 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
2 3 Rampini and Viswanathan (2010) in the context of a firm's access to credit.2 4 Most recently, there are other papers which have looked at the same relationship but using cross-
country settings: Gianetti (2003); Qian and Strahan (2007); Acharya et al. (2011); and Davydenko
and Franks (2008).
23
changes in creditor protection. In Vig (2013), the decrease in the amount of secureddebt is driven by an increase in the threat of early liquidation, which firms face dueto 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 channelwhich relies on household risk aversion, and/or an increase in the number of strategicborrowers. 26
This paper is also related to previous studies that have looked at the effect ofbankruptcy laws design in the context of corporate bankruptcy (Baird and Ras-mussen, 2002; Bolton and Scharfstein 1996). In this context there is a large lit-erature that describes the tension between ex ante and ex post efficiency in anybankruptcy design. For instance, Gertner and Scharfstein (1991), and Hart (2000),show the incentives of the debtor and creditors under corporate resolution in a the-oretical framework, and demonstrate how debt contracts can lead to inefficient liq-uidation and underinvestment. This framework is also relevant when thinking aboutthe incentives for households to file for bankruptcy. Empirically, Chang and Schoar(2013) look at the judge-specific fixed effect, showing that pro-debtor judges haveworse firm outcomes after Chapter 11, suggesting that this is a result of managersand shareholders' incentives misalignment, highlighting how bankruptcy codes canhave a significant impact on ex post outcomes. Furthermore, Iverson (2013) looksat the effect of bankruptcy courts' reduction in court caseloads due to the consumerbankruptcy reform in 2005, finding that firms in more pro-debtor courts allow morefirms to reorganize and liquidate fewer firms.
Finally, this paper is complementary to studies looking at the effect of personalbankruptcy laws on labor markets. Dobbie and Song (2013) find that filing forbankruptcy under Chapter 13 has a significant effect on increasing earnings and em-ployment, and also decreases mortality, suggesting that consumer bankruptcy benefitsare an order of magnitude larger than previously estimated".
1.3 Data and Summary Statistics
1.3.1 Data Description
In order to address the impact of changes in bankruptcy protection on householddebt, we collect and combine different data sources. The three main data sourcesinclude time series of state levels of protection under bankruptcy, and geographicaldistribution of household debt and interest rates information. In this section wedescribe this datasets in detail.
The level of protection or exemptions represents the dollar amount of equity thatthe debtor is entitled to protect in the event of bankruptcy; it represents the amount
25This is consistent with the corporate literature on bankruptcy reorganization which suggestedthat 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).26Examples 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).
2 7 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 Con-
sumer 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 so-
cial 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 be-
havior of households, as we are able to track the delinquency behavior of consumers
before they are affected by the change in protection2 9
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 prod-
ucts, 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
28How to file for Chapter 7 Bankruptcy, Elias Renauer and Leonard Michon. Nolo editorial
(1999-2009)29See Lee and van der Klaauw (2010) for details on the sample design.
25
the Bureau of Labor Statistics Local Area estimates. Local Area UnemploymentStatistics (LAUS) are available between 1976 and 2012 for approximately 7,300 ar-eas that range from census regions and divisions to counties and county equivalent.We match the county equivalent data to the CCP data using Federal InformationProcessing Standard (FIPS) county unique identifiers.
To look at the determinants of change in exemptions, we use four additional datasources: changes in state total medical expenses extracted from the National HealthExpenditure Data, Centers for Medicare and Medicaid Services; state level changesin GDP and Personal Income from Bureau of Economic Analysis (BEA); bankruptcyfiling statistics at the state level from the Statistics Division of the AdministrativeOffice of the United States Courts30; and measures of political climates using theshare of votes for the Democratic Party in the last House of Representatives electionobtained from the Clerk of the House of Representatives (CHR).
The net creation of sole proprietorships at a county level is obtained from Censusnon-employer statistics; we obtain the number of establishments for the period of 1999to 2009 at the 2-digit NAICS level. In order to construct a measure of industries thatuse 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 responsesfrom the 2007 SBO and provides access to survey data at a more detailed level thanthat of the previously published SBO results. The SBO PUMS is designed to studyentrepreneurial activity by surveying a random sample of businesses selected from alist of all firms operating during 2007 with receipts of $1,000 or more provided bythe IRS. The survey provides business characteristics such as firm size, employer-paidbenefits, minority- and women-ownership, access to capital, and firm age. For thepurposes of this paper, we classified industries based on the "use of credit card as astart-up capital" for each firm and we group the answers to this question at the 2-digitNAICS industry level (the finest level available in the data) for firms established in2007, 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 to2005. The total debt balance in a county is 2.91 billion dollars. The level of creditcard balance is 0.29 billion dollars. When looking at states that "eventually" changetheir level of protection during our sample period and compare them to states thatnever 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 growthduring the same period experienced the same pattern, with a 7.6% average annualgrowth, with no significant difference between the "eventually" treated and the nevertreated group. The summary statistics seem to show that credit card balances are asmall proportion of the average household balance sheet, as mortgage debt accountsfor 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 aver-
age 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 in-
creased 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 protec-
tion. 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' equilib-
rium 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 pledg-
ing 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 collat-
eral 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 risk-
averse 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 strategicdefaulters with private information about their future income or propensity to defaultcould 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. Againthe equilibrium response will be driven by lenders' ability to screen new borrowers.
Therefore, the theoretical prediction is unclear, given that the net effect will de-pend on the relative magnitudes of the supply and demand response3 1 . Interest mustweakly rise in equilibrium, independent of the prevailing force. If the supply demanddominates, quantities should go down, but if the demand effect dominates, quantitiesshould go up. We attempt to distinguish between these channels empirically.
It is plausible to imagine that in the presence of agency problems, a demanddriven equilibrium takes place. In an extreme case, if the lender overestimates thequality of the pool of borrowers, the increase in protection would lead to an increasein quantities. However, in Appendix A we show that given very simple conditions,and without asymmetric information, we can observe a demand driven equilibriumwhere quantities and prices increase. This model of the credit market considers arisk-averse borrower who is financially constrained and a risk-neutral lender. Theborrower has a stochastic income, and exogenous home equity that is realized inperiod 2. Only debt contracts are available. In case of default, the lender can seizethe borrower's assets up to the exemption level dictated by law. The agents need toborrow in order to consume in period 1, while the interest rate is set such that thebank breaks even (zero profit). For a given interest rate, a risk-averse borrower willconsume until a point where the marginal utility of consumption today is equal tothe expected marginal utility in the future. Increased bankruptcy protection makesdefaulting attractive to the borrower in more states of the world, and forces lendersto charge a higher interest rate to break even.
The model shows that for a certain region with a given level of protection inbankruptcy, when the level of protection is increased, the agent will be willing to takeon more debt despite the increase in interest rates. This happens when the marginalbenefit from the increase in consumption at period 1 is greater than the loss of utilityin the good state in period 2, due to the repayment of their debt claim; as in the badstate they are indifferent due to the protection level. Furthermore, if the marginalbenefit is not enough to overcome the loss of consumption during the good state, weshould see a decrease in quantities and increase in prices. Using exogenous variationon the level of protection, we aim to identify the type of equilibrium that rises after anincrease in the level of consumer protection under bankruptcy. These results, whichare highlighted by the model, are relevant as they show that the insurance channelin itself could lead to a demand driven credit market equilibrium shift, without thepresence of moral hazard or adverse selection.
Empirical PredictionsThe exposed theoretical framework allows us to sharpen our empirical exploration.
31Figure 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 af-
fects home-equity holding predominantly (see Table 1.2). The increase in bankruptcy
protection does not directly affect secured debt, as the bankruptcy code only dis-
charges 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 equilib-
rium 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, autoloan debt, in a county i and year t .AProtectiont represents the log change in thelevel of Chapter 7 protection (homestead plus personal) in a state s and year t .aj isa county fixed effect, and at are year fixed effect.AXit represents a vector of countycontrols changes, such as county unemployment rate, log of house prices, and log ofincome in a county.
We use the same specification in (1) to measure the effect of changes in protectionon interest rates. To do so we replace the log change in debt, by changes in interestrates in percentage for mortgages, personal unsecure loans and credit cards.
Since changes in protection vary at the state level, but debt balances and inter-est 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 andMullainathan (2004), the residuals are clustered by state. This allows for maximumflexibility in the variance-covariance matrix of residuals. It is also more general thanstate-year clustering, which would leave intact the possibility of serial correlation inthe error term.
If the measure of debt and the controls all display heterogeneous trends acrosscounties, 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 fixedeffects guarantees that differential county specific trends are controlled for in all vari-ables. A first-differences specification is suitable in our case as it accommodatesthe repeated treatment present in our sample (in our sample period some states didchange their level of protection more than once). The regressor 13p captures thechanges 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 themain estimate is driven by big changes in the level of protection. Furthermore, wewill conduct alternative specifications to show that our results are robust to the useof 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 be-longing to a state which increased the level of protection between t and t+1, with theamount of debt of a county belonging to a state in which the level of protection did notchange during the same period. The two identifying assumptions are first, that thetiming of the changes in the levels of protection are uncorrelated with determinants ofhousehold leverage; and second, that after controlling for observed time-varying char-acteristics, linear county trends, and time-invariant county characteristics, changesin 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 theexogenous change in the level of protection.
We assess the first identifying assumption by looking at the correlation between
3 2Paravisini (2008).
30
suspected determinants in the level of protection and changes in the levels of protec-
tion. 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 hy-
potheses 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 aris-
ing 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 het-
erogeneity that is uncorrelated to changes in the levels of protection. To overcome this
endogeneity we compare neighboring county-pairs across state borders33 , but within
the same income categories, using the following empirical specification:
ADebtipt = oi + aYipt + /pAProtectionst + FAXit + 6Ept (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 3This 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 eachyear. Note that variables for county i maybe repeated for all pairs of which they arepart. In this setup our estimate fp only uses debt variation within each neighboringcounty-pair across state borders. Our additional identifying assumption implies thatthe changes in protection are uncorrelated with the residual Eipt after controlling forobservable characteristics, county fixed effects and county-pairs year fixed effect. Wealso assign counties to income buckets, and run the proposed specification only withincounty-pairs that are in the same income category.
To attempt to identify the channel that is driving the demand effect we use in-dividual level data to look at debt change, entry to the credit card market, anddelinquency. We use the same specification (1) as for the county aggregates, butchanging the dependent variable, and including in this case the zipcode level houseprices, income, and county unemployment rates.
The change in debt for each individual is estimated using log changes, and ittherefore represents the change in debt for existing debtors. When looking at thenumber of accounts, our dependent variable is the difference between the number ofcredit cards in t -1 and t. Entry is defined in two ways as follows: opening the firstcredit card, which is a dummy equal to one if the household did not have a creditcard in t-1, and have one or more credit cards in t. Alternatively, entry is defined as adummy equal to one if the balance becomes positive between t and t -1. Both measuresattempt 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 attime 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 sameindividual may be affected by the change in the levels of protection more than onceduring our sample period.
Finally, we look at changes in the levels of self-employment to explore the effecton real outcomes. For this we use specification (1) but in this case, using the changein total county self-employment as a left hand side variable, or the change in self-employment in an industry and county between t and t-1.
1.6 Results and discussion
1.6.1 Bankruptcy Protection and Household Leverage andInterest Rates
We find that growth in bankruptcy protection leads to an increase in the level ofcredit card debt held by households (unsecured debt) between 1999 and 2005 (Table1.3 ). Moreover, the increase in protection has no effect on other types of secureddebt (auto and mortgage, Table 1.4 and 1.5)3.
3 4The average effect is only present in the pre-bankruptcy reform period, when filing forbankruptcy 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 debttoday. 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 mort-
gage 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 mort-
gage 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.3 5 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 anda 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 upstate specific credit market trends that are correlated with the changes in protectionand that can be confounded with our identified effect.
Table 1.9 shows the effect is stronger in counties that are in the lowest tercile ofthe within state income distribution, monotonically decreasing as the level of incomeincreases. It is expected that lower-income areas may be more affected by increasesin protection, as the impact of the improvement in risk sharing should be more sig-nificant.
Homeowner households should be more affected by the changes in the level ofprotection, 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 ameaningful perspective on this variation, we look at the within income group variationon county level homeownership. Table 1.9 column 3 shows that the differential effectis aligned with the prediction, as the estimated coefficient for these particular areasalmost triples with respect to the baseline specification.
Following the same logic, we look at the within income group variation on bankconcentration - a measure based on share of deposit holding at the branch level. Ta-ble 1.16column 2 shows that the effect is stronger in areas where markets are moreconcentrated, which is consistent with the Peterson and Rajan (1995) relationshiplending model, where creditors are more likely to finance a credit constrained bor-rower when credit markets are concentrated because it is easier for these creditors tointernalize the benefits of assisting these borrowers.
Another alternative explanation of our finding is that the increase in quantities isdue to a contemporaneous decrease in prices, which correlates with the timing of thechanges in bankruptcy protection. In other words, areas which increased the level ofprotection were areas in which credit became cheaper. As mentioned above, Tables1.10 and 1.11 show that the increase in bankruptcy protection leads to an increasein the level of interest rates for unsecure loans, not affecting mortgage rates. Theseresults support our causal interpretation of the results, alleviating the concern that weare 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 het-erogeneity that is uncorrelated with changes in the levels of protection. To over-come this endogeneity, we compare neighboring county-pairs within the same incomebucket. Table 1.8 shows that when focusing on a county-pair in the same incomebucket, the estimated results are very similar to the main specification. Moreover theeffect is stronger when we concentrate on county-pairs in the lower end of the countyincome distribution.
1.6.2 Robustness Test
We choose a first difference specification with county fixed effect to parsimoniouslyaccount 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 can-
didate 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 coeffi-
cient. 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 (1999-
2005) 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 anindividual for whom we observe home-related debt at some point between 1999 and2005, the effect is stronger, which again is consistent with the county estimates (Table1.12 Panel B).
Furthermore, using detailed account information, we show in Table 1.12 columns2-4, that changes in protection causally increase the number of credit cards per house-hold; 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 house-holds that also had a positive balance. This finding suggests that the credit expansionis 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 mar-ket, 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 theseresults provide evidence that in this sample, the effect is being driving by existingdebtors expanding their current balance or their number of accounts, rather than newhouseholds entering the credit market.
Focusing on the same sample, we explore their delinquency behavior up to threeyears 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. Table1.13 shows that within this sample there is no measurable increase in the level ofdelinquency; if anything, the probability of individuals becoming delinquent in thefuture 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 toincrease. Although we cannot completely rule out an over-borrowing behavior, theresults described are more consistent with risk-averse borrowers increasing their debtholding 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 carddebt also experienced an increase in the level of self-employment creation, specificallywithin industries that make more use of credit cards as start-up capital. Table 2.6shows that, on average, the increase in self-employment is only positively correlatedwith the changes in the level of protection in low-income regions. Also, the estimatedeffect is stronger when we focus on industries for which credit card debt is an impor-tant source of financing (for example, construction or photography). It is importantto point out that these outcome variables are only suggestive evidence of the realeffect of the increase on the level of unsecured debt.
Taking all this evidence together, the rise in credit card debt induced by theincrease in the level of protection could have led to an increase in small businesscreation, 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 isconsistent with a large impact from the insurance channel on existing borrowers, as wedo not observe increases in the entry rates of new borrowers and ex post delinquencieswithin our micro level sample. Although this evidence is only suggestive, it highlightsthe important potential benefits of increasing the level of bankruptcy protection,especially for people in areas on the lower end of the wealth distribution, for whichthe 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 andinnovation. Review of Financial Studies 22.12 (2009): 4949-4988.
Acharya, Viral V., Rangarajan K. Sundaram, and Kose John. Cross-countryvariations in capital structures: The role of bankruptcy codes. Journal of FinancialIntermediation 20.1 (2011): 25-54.
Acharya, Viral V., Yakov Amihud, and Lubomir Litov. Creditor rights and cor-porate risk-taking. Journal of Financial Economics 102.1 (2011): 150-166.
Adelino, Manuel, Antoinette Schoar, and Felipe Severino. House Prices, Collateraland Self-Employment. No. w18868. National Bureau of Economic Research (2013).
Agarwal, Sumit, et al. Regulating Consumer Financial Products: Evidence fromCredit Cards. No. w19484. National Bureau of Economic Research (2013).
Aghion, Philippe, Oliver Hart, and John Moore. The economics of bankruptcyreform. Journal of Law Economics and Organization. 8 (1992): 523-546.
Alvarez, Fernando, and Urban J. Jermann. Efficiency, equilibrium, and assetpricing with risk of default. Econometrica 68.4 (2000): 775-797.
Baird, Douglas, and Robert Rasmussen. The End of Bankruptcy. Stanford LawReview 55 (2002).
Benmelech, Efraim, and Nittai K. Bergman. Bankruptcy and the collateral chan-nel. The Journal of Finance 66.2 (2011): 337-378.
Berger, Allen N., Geraldo Cerqueiro, and Mara F. Penas. Does debtor protectionreally protect debtors? Evidence from the small business credit market. Journal ofBanking & Finance 35.7 (2011): 1843-1857.
Bernanke, Ben, and Mark Gertler. Agency costs, net worth, and business fluctu-ations. The American Economic Review (1989): 14-31.
Bertrand, Marianne, Esther Duflo, and Sendhil Mullainathan. How much shouldwe trust differences-in-differences estimates? The Quarterly Journal of Economics119.1 (2004): 249-275.
Bolton, Patrick, and David S. Scharfstein. Optimal debt structure and the numberof creditors. Journal of Political Economy (1996): 1-25.
Cerqueiro, Geraldo, and Maria Penas. How does personal bankruptcy law affectstart-ups? Working Paper (2011).
Cerqueiro, Geraldo, Deepak Hegde, Maria Penas, and Robert Seamans. PersonalBankruptcy Law and Innovation. Working Paper (2013).
Chaney, Thomas, David Sraer, and David Thesmar. The Collateral Channel: Howdid Real Estate Shocks Affect Corporate Investment. American Economic Review102(6) (2012): 2381-2409.
Chang, Tom, and Antoinette Schoar. Judge Specific Differences in Chapter 11and Firm Outcomes. National Bureau of Economic Research (2013).
Chatterji, Aaron K., and Robert C. Seamans. Entrepreneurial finance, creditcards, and race. Journal of Financial Economics 106.1 (2012): 182-195.
Davydenko, Sergei A., and Julian R. Franks. Do bankruptcy codes matter? Astudy 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 Economicsand 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 En-
trepreneurial 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? Evi-
dence 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) 841-79Hart, 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 con-sumer bankruptcy filing: new evidence. No. 13-24. (2013).
Kehoe, Timothy J., and David K. Levine. Debt-constrained asset markets. TheReview of Economic Studies 60.4 (1993): 865-888.
Kiyotaki, Nobuhiro, and John Moore. Credit Chains. Journal of Political Econ-omy 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 FRBNYConsumer Credit Panel. Federal Reserve Bank of New York Staff Report no. 479.(2010)
Lefgren, Lars, and Frank McIntyre. Explaining the Puzzle of Cross-State Differ-ences in Bankruptcy Rates. Journal of Law and Economics 52.2 (2009): 367-393.
Levine, Ross. The legal environment, banks, and long-run economic growth. Jour-nal of Money, Credit and Banking (1998): 596-613.
Li, Wenli, Michelle J. White, and Ning Zhu. Did Bankruptcy Reform CauseMortgage Default to Rise?. No. w15968. National Bureau of Economic Research,2010.
Lilienfeld-Toal, Ulf von, Dilip Mookherjee, and Sujata Visaria. The DistributiveImpact of Reforms in Credit Enforcement: Evidence from Indian Debt RecoveryTribunals. Econometrica 80.2 (2012): 497-558.
Lin, Emily Y., and Michelle J. White. Bankruptcy and the market for mortgageand 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 externalfinance. The Journal of Finance 63.5 (2008): 2161-2193.
Petersen, Mitchell A., and Raghuram G. Rajan. The benefits of lending rela-tionships: 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 commer-cial aircraft transactions. The Journal of Finance 53.3 (1998): 939-978.
Qian, Jun, and Philip E. Strahan. How laws and institutions shape financialcontracts: The case of bank loans. The Journal of Finance 62.6 (2007): 2803-2834.
Rampini, Adriano A., and S. Viswanathan. Collateral, risk management, and thedistribution 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. NationalBureau 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 auc-
tions: 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 BankruptcyProtection on Household Borrowing
To explore the previous explanation, gain further insights into the effects of changesin 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 themoral hazard and adverse selection behavior of borrowers. In our model, we highlightthe effect of the increase of partial insurance provided by bankruptcy protection inthe credit market equilibrium outcome, and how even in the absence of asymmetricinformation we could observe a demand effect.
We do this using a two period model, where the agent needs to borrow in order toconsume 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 incomefollows a binomial distribution given by y(G) = W > 0 and y(B) = 0 . Exists a levelof 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 period2 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 = P
* Default and home-equity is fully protected (H - P) < 0: ci = H
42
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.
OS
N4 8
1994 IM9 1998 20DO 2002 2004 2006 2008 2010 2012
Yearly non-businms fifins US - Consumer Revo"vn Debt US
44
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 isthe initial equilibrium before the change. Increase in Price, No Increase in Q, show the effect whenthe supply response totally and perfectly upsets the demand increase. Increase in Price, Decreasein Q, show the effect when the supply response is stronger than the demand increase. Increase inPrice, Increase in Q, show the effect when the demand effect dominates.
P
Do So
Pa -
Po
QoBaseline Equilibrium Q
Db Sb
Do so
.... - .- Pc
PO
Q
Da
Do so- - - - -- - - -=
Qo=QaIncrease in Price, No Increase in Q
DcSc
Do so
....- ..-.- --- - ..- ..- -
s c, QCIncrease in Price, Increase in Q
46
P
Pb
Po
Q
Increase in Price, Decrease in QQ
s
So
!
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 modelin 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
Crdi ExasoardtCnrcio ee fpoetoP
V
Credit Expansion Credit Contation Level of protection (P)
47
Table 1.1: Summary Statistics Data
All SN=1
Levels Mean
ample Eventually Treated Never Treated5,519 N=7,091 N=8,428Std. Dev. 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)
County Total Debt (bil. USD)County Mortgage Debt (bil. USD)
County Credit Card Debt (bil. USD)County Auto Debt (bil. USD)
Pers. Unsec. Int. Rate (bp)Credit Card hit. Rate (bp)
30 yr Fix. Mtg. Int. Rate (bp)
Mortgage Delinquency ('/ of pop)Credit Card Deliquency (W of pop)
Auto Delinquency (W of pop)
County Household PopulationIRS County Income (bil. USD)
Unemployment RateNo. of Bankruptcy Filing (1998)
W of Owner Occupancy (2000 )
1.230.900.160.17
2.892.330.290.26
12.813.16.6
0.480.450.040.06
10.519.010.830.76
2.22.70.7
1.290.970.160.16
3.933.250.360.32
12.813.46.6
0.520.490.040.06
13.9512.081.030.92
2.22.70.7
1.180.840.170.17
2.011.570.220.22
12.912.86.6
0.450.410.050.07
6.185.060.610.60
2.22.70.7
*
1.5 1.3 1.5 1.2 1.6 1.38.2 3.5 7.8 3.1 8.5 3.82.4 1.5 2.3 1.4 2.4 1.5
100.3061.905.32604
73.35
269,4775.561.90
20517.84
N=13,302Changes Mean Std. Dev.
123,7352.465.35
331.5736.851.87
N=6,078Mean Std. Dev.
80.594 200.9341.43 4.115.30 1.93
N=7,224Mean Std. Dev.
*
DTI ChangeMTI Change
CCTI ChangeATI Change
Total Debt GrowthMortgage Debt Growth
Credit Card Debt GrowthAuto Debt Growth
Pers. Unsec. it. Rate Change (bp)Credit Card JIt. Rate Change (bp)
30-yr Fix. Mtg. Int. Rate Change (bp)
Income GrowthUnemployment Rate Change
House Price Growth
0.0990.1150.0510.098
0.1220.1330.0760.117
-0.09-0.75-0.34
0.0330.1110.075
0.1130.1490.1180.156
0.0910.1200.0990.125
0.941.880.50
0.0530.9630.046
0.1010.1150.0530.096
0.1230.1330.0780.115
-0.12-0.65-0.34
0.0320.1150.088
0.1090.1450.1120.146
0.0890.1190.0930.118
0.931.840.49
0.098 0.1160.115 0.1510.049 0.1240.101 0.165
0.122 0.0920.133 0.1200.075 0.1040.119 0.130
-0.06-0.84-0.33
0.951.910.51
0.054 0.033 0.0520.931 0.108 0.9890.050 0.062 0.037 ** *
Note. "All Sample" refers to all counties in the sample period. "Eventually Treated" refers to counties treated duringthe 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 autoloans, is obtained from the FRBNY Consumer Credit Panel/Equifax. IRS County Income (in bil. USD) is measuredas total wages and salary in that county. Debt to Income is constructed using the two county measures describedabove. Personal unsecured, credit card, and 30-year fixed mortgage rates are constructed from branch-setter level ratesfrom Rate-Watch. Delinquency rates for mortgage, credit card, and auto loans are from the FRBNY Consumer CreditPanel/Equifax, and represent the fraction of households that are 90+ days delinquent. County household populationis the number of household per county and year in the FRBNY Consumer Credit Panel/Equifax. No. of Filings isthe number of non-business filings in a county in 1998 from the American Court System. % of Owner Occupancy isthe percentage of home ownership in a county in 2000 from the Census Bureau. For a complete description of thedata sources see section 3.1. Data Description. House price growth is extracted from the Federal Housing FinanceAgency (FHFA) House Price Index (HPI) data at a state level. The number of observations refers to the number ofcounty-year observations. Almost all variables are available for every county (2,218), with the exception of interestrates, which are only available for (1232, 1323 and 1340 counties respectively). *, **, and *** denotes significanceat 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 p 2 5 p50 p 7 5 p9 5
Protection Level 73,627 75,125 13,000 23,200 55,800 166,200 unlimited
Homestead 63,932 73,356 7,500 20,000 40,000 150,000 unlimited
Personal Assets 9,695 5,965 2,900 5,000 8,400 11,000 25,000Unlimited States 7No. of States 50
Eventually Treated Mean Std. Dev. p5 p 2 5 p50 p 7 5 p 9 5
Protection Level 85,655 86,100 11,000 32,300 51,000 110,300 390,000Homestead 75,243 84,838 0,000 25,000 40,000 100,000 350,000Personal Assets 10,411 6,061 3,000 7,200 9,100 11,000 25,000No. of States 26
Protection Changes 38,841 52,992 2,000 3,250 15,400 50,000 200,000
No. of Changes 37
Never Treated Mean Std. Dev. p5 p 2 5 p50 p 7 5 p 9 5
Protection Level 56,922 52,366 14,400 20,700 57,700 586,000 unlimited
Homestead 48,222 49,678 10,000 13,750 45,000 575,000 unlimited
Personal Assets 8,700 5,705 2,900 4,800 6,300 12,300 42,000
No. of States 24Note. "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
Table 1.3:
Effect of B
ankruptcy Protection on D
ebt. C
redit Card D
ebt
Ch
anges
Level
Cou
nty
Lin
earT
rend
(1)
Protection
0.018G
rowth s,t
(0.008)
Sta
teL
inear
Tren
d
(2)
No
Lin
earT
rend
(3)
Level
Con
trols
(4)
0.019** 0.018**
0.017**(0.008)
(0.007) (0.007)
Con
trols +In
c-Year
Uep
-Year
(5)
Even
tually
Treated
(6)
Ch
anged
On
ce(7)
(7) ~(9)(1)
1)
Ch
ange
>
0.15(8)
Du
mm
yT
reatm
ent
0.017** 0.017**
0.022** 0.018**
0.012***(0.008)
(0.008) (0.009)
(0.008) (0.004)
0.023** 0.027**
(0.011) (0.013)
Unem
ployment
0.002 0.003
0.002 0.000
0.003R
ate Change
(0.002) (0.002)
(0.002) (0.002)
(0.003)0.002
0.002 0.002
0.002(0.003)
(0.002) (0.002)
(0.002)
House P
rice -0.102
-0.109 -0.139***
-0.203** -0.183*
-0.118 -0.049
-0.103 -0.105
Index Grow
th (0.086)
(0.085) (0.037)
(0.102) (0.099)
(0.086) (0.108)
(0.086) (0.083)
Incomue
0.079* 0.134***
0.142*** 0.073*
0.088** 0.138*
0.081 0.079*
0.079*G
rowth
(0.047) (0.041)
(0.040) (0.041)
(0.041) (0.077)
(0.051) (0.047)
(0.047)
0.005* 0.002
(0.003) (0.003)
0.083*** 0.070**
(0.031) (0.029)
0.023 0.010
(0.021) (0.020)
13,302 13.302
13,302 13,302
50 50
50 50
y
yyY
y
y
y
0.003 0.007*
(0.003) (0.004)
-0.166*** -0.263***
(0.042) (0.053)
0.251*** 0.951***
(0.047) (0.006)
6,078 11,478
13,302 13,302
26 39
50 50
y
y
y
y
y
y
y
y0.29
0.28 0.30
0.31 0.29
0.30 0.30
0.30
15,51950y
15,51950Y
y
y
Notes.
Th
is table
sho
ws th
e estimated
coefficien
t fo
llow
ing
specificatio
n (1)
of log ch
anges
to cred
it card
deb
t on
log chan
ges
in b
ank
rup
tcy
pro
tection
at th
e cou
nty
level. D
ebt co
unty
data
is fro
m th
eF
RB
NY
Consu
mer
Cred
it P
anel/E
quifa
x.
Pro
tection
Gro
wth
is the log ch
ange in th
e lev
el of p
rotectio
n
in state
s at time t.
Pro
tection
L
evel is the level o
f pro
tection
in sta
te s at tim
e t. U
nem
plo
ym
ent
rate
chan
ge
is the
chan
ge in u
nem
plo
ym
ent
rate
in co
un
ty i at tim
e t from
B
LS
. H
ouse
price g
row
th is th
e log chan
ge
in the
FH
FA
state
level ind
ex for sta
te
s at time t, an
d
Inco
me g
row
th is th
e in
com
elog ch
ange
in cou
nty
i at tim
e t from
IR
S.
Colu
mns 1
and
2 sho
w th
e resu
lt usin
g
cou
nty
and sta
te fix
ed effects
respectiv
ely in
the first
differen
ce specificatio
n.
Colu
mn
3 sho
ws th
e
results if w
e exclu
de
state
or co
un
ty
fixed
effect fro
m
specificatio
n
(1). C
olu
mn
4 sho
ws th
e estimates
inclu
din
g
level of th
e co
ntro
ls. C
olu
mn
5 show
s the
estimates
inclu
din
g
level contro
ls an
d
inco
me
and
un
emp
loym
ent
gro
ups
times y
ear fixed
effect, to
allo
w fo
r differen
tial tren
ds
across
state
s based
o
n th
ese observ
able ch
aracteristics. C
olu
mn 6 sh
ow
s the estim
ates for a reg
ression th
at o
nly
u
ses state
s treate
d
durin
gth
e sam
ple
perio
d,
that
is, state
s th
at ch
anged
th
eir lev
el o
f pro
tection
d
urin
g th
e sam
ple p
eriod
. C
olu
mn 7 sh
ow
s the
results
if we
only
consid
er as tre
ate
d sta
te th
at ch
anged
o
nce.
Colu
mn
8 sho
ws
the estim
ates if w
e replace b
y zero ch
anges b
elow
0.15. C
olu
mn
9 show
s results if w
e replace
the ch
ange
with
a du
mm
y in
dicato
r that
is one if th
e chan
ge
is greater th
an
zero.
Colu
mns
10 and
11 show
the resu
lts of reg
ression
log levels o
f credit
card d
ebt
on
log levels
of p
rotectio
n
and
in
clud
ing
cou
nty
an
d sta
te fix
ed effect
respectiv
ely.
Th
e sam
ple p
eriod
is
from
1999 to
2005. *,
* an
d
*** d
eno
tessig
nifican
ce at th
e
10%, 5%
, an
d 1%
clu
ster at th
e sta
te
level resp
ectively
.
ProtectionL
evel s,t
Levels
Level
onL
evelC
oun
ty FE
(10)
Level onL
evelS
tate F
E(11)
Un
emp
loy
men
t
Rate
House P
rice
Incomue
No.
of Obs.
No.
of Clusters
County F
ES
tate FE
Year F
ER
-Squared
13,30250yy0.30
Tab
le 1
.4:
Eff
ect
of B
ankr
uptc
y P
rote
ctio
n on
Deb
t. M
ortg
age
Deb
t
Ch
ange
s
Sta
teL
inea
rT
ren
d(2
)
No
Lin
ear
Tre
nd
(3)
Lev
elC
ontr
ols
(4)
Lev
elC
ontr
ols
+In
c-Y
ear
Uep
-Yea
r(5
)
Eve
ntu
ally
Tre
ated
(6)
Ch
ange
dO
nce
(7)
Ch
ange
>
0.15
(8)
Du
mm
yT
reat
men
t
(9)
Pro
tect
ion
0.01
1 0.
011
0.00
8 0.
005
0.00
7 0.
014
0.01
3 0.
012
0.00
6
Gro
wth
s.t
(0
.012
) (0
.012
) (0
.015
) (0
.010
) (0
.010
) (0
.014
) (0
.013
) (0
.012
) (0
.007
)
Pro
tect
ion
Lev
el s
t
Une
mpl
oym
ent
-0.0
04
-0.0
03
-0.0
03
-0.0
04
0.00
0 -0
.001
-0
.005
* -0
.004
-0
.004
Rat
e C
hang
e (0
.003
) (0
.003
) (0
.003
) (0
.003
) (0
.003
) (0
.002
) (0
.003
) (0
.003
) (0
.003
)
Hou
se P
rice
0.
086
0.07
8 0.
044
-0.3
78**
-0
.345
**
0.12
8 0.
046
0.08
6 0.
084
Inde
x G
row
th
(0.1
61)
(0.1
61)
(0.0
79)
(0.1
70)
(0.1
74)
(0.2
56)
(0.2
09)
(0.1
61)
(0.1
61)
Inco
me
0.11
4 0.
185*
* 0.
191*
*G
row
th
(0.1
07)
(0.0
91)
(0.0
91)
0.03
9(0
.079
)
0.00
7 0.
006
(0.0
31)
(0.0
26)
0.06
0 0.
208
0.12
5 0.
114
0.11
4(0
.081
) (0
.181
) (0
.114
) (0
.107
) (0
.107
)
0.00
0 -0
.004
(0.0
04)
(0.0
04)
0.27
8***
0.
265*
**(0
.041
) (0
.041
)
0.13
3***
0.
105*
**(0
.039
) (0
.036
)
0.00
1 -0
.055
***
(0.0
04)
(0.0
07)
0.01
3 -0
.223
**(0
.069
) (0
.089
)
0.319
***
1.12
3***
(0.0
67)
(0.0
12)
13,3
02
13.3
02
13,3
02
13.3
0250
50
50
50
Y
YY Y
Y
Y
Y
6,07
8 11
,478
13
.302
13
.302
26
39
50
50Y
Y
Y
Y
Y
Y
Y
Y().
1()
0.08
0.
11
0.13
0.1
1 0.
09
0.09
0.09
15.5
1950 y
15,5
1950 y
Note
s.
This
tab
le
show
s th
e e
stim
ated
coef
fici
ent
foll
ow
ing sp
ecif
icat
ion
(1)
of
log
chan
ges
to
m
ort
gag
e d
ebt
on
log
ch
anges
in
ban
kru
ptc
y
pro
tect
ion
at
th
e co
unty
le
vel.
D
ebt
cou
nty
data
is
fro
m t
he
FR
BN
Y
Co
nsu
mer
C
redit
Pan
el/E
quif
ax.
Pro
tect
ion G
row
th i
s th
e lo
g ch
ange
in t
he l
evel
of
pro
tect
ion
in
sta
te s
at
tim
e t.
Pro
tect
ion
L
evel
is
the
leve
l o
f p
rote
ctio
n
in s
tate
s a
t ti
me
t.
Unem
plo
ym
ent
rate
chan
ge
is t
he c
han
ge
in u
nem
plo
ym
ent
rate
in
cou
nty
i a
t ti
me
t fr
om
BL
S.
House
pri
ce g
row
th i
s th
e l
og c
han
ge
in t
he
FH
FA
st
ate
lev
el i
ndex
for
sta
te s
at
tim
e t,
and
Inco
me
gro
wth
, is
the i
nco
me
log
chan
ge
in c
ou
nty
i
at t
ime
t fr
om
IR
S.
Co
lum
ns
1 an
d 2
show
the
resu
lt
usi
ng c
ou
nty
an
d st
ate
fix
ed e
ffec
ts r
esp
ecti
vel
y
in t
he
firs
t d
iffe
ren
ce
spec
ific
atio
n.
Colu
mn 3
sh
ow
s th
e r
esu
lts
if w
e ex
clude
state
or
cou
nty
fi
xed
eff
ect
from
sp
ecif
icat
ion
1.
Co
lum
n 4
show
s th
e es
tim
ates
in
clu
din
g
lev
el o
f th
e co
ntr
ols
. C
olu
mn
5 sh
ow
s th
e es
tim
ates
in
cludin
g
leve
l co
ntr
ols
and
inco
me
and
unem
plo
ym
ent
gro
ups
tim
es y
ear
fixed
ef
fect
, to
all
ow
fo
r d
iffe
ren
tial
tre
nds
acro
ss st
ate
s b
ased
o
n t
hes
e obse
rvab
le c
har
acte
rist
ics.
C
olu
mn
6 sh
ows
the e
stim
ates
fo
r a
regre
ssio
n
that
only
u
ses
state
s tr
eate
d d
uri
ng
the s
amp
le
per
iod,
that
is,
state
s th
at
chan
ged
thei
r le
vel
of
pro
tect
ion
d
uri
ng
the
sam
ple
per
iod.
Colu
mn 7
show
s th
e
resu
lts
if w
e only
consi
der
as
tre
ate
d s
tate
th
at
chan
ged
on
ce.
Colu
mn
8 sh
ow
s
the e
sti
mate
s
if w
e re
pla
ce
by zero
changes
belo
w
0.1
5.
Colu
mn
9 sh
ow
s re
su
lts
if w
e re
pla
ce th
e c
hange w
ith
a d
um
my
in
dic
ato
r th
at
is o
ne if
th
e change
is
gre
ate
r th
an
zero
. C
olu
mn
10 and
11,
sho
w
the r
esult
s
of
reg
ress
ion
lo
g le
vel
s of
mo
rtg
ag
e
debt
on
lo
g le
vel
s of
pro
tecti
on
an
d in
clu
din
g co
un
ty an
d sta
te
fixed eff
ect
resp
ecti
vely
. T
he sam
ple
p
eri
od
is
fr
om
1
99
9 to
2
00
5.
*,
*,
an
d
***
den
ote
s
signif
ican
ce a
t th
e
10%
, 5%
, an
d
1%
clust
er at
th
e st
ate
lev
el r
esp
ecti
vel
y
Co
un
tyL
inea
rT
ren
d
(1)
Lev
els
Lev
el o
nL
evel
Cou
nty
FE
(10)
Lev
el o
nL
evel
Sta
te F
E(1
1)
Unen
iplo
yie
nt
Rat
e
Hou
se P
rice
Inco
me
No.
of
Obs
.N
o.
of C
lust
ers
Cou
nty
FE
Sta
te F
EY
ear
FE
R-Sq
uare
d
13.3
0250 Y Y 0.09
Y
Y0.
86
0.97
Table 1.5:
Effect of B
ankruptcy Protection on D
ebt. A
uto Debt
Ch
anges
Level
Cou
nty
Lin
earT
rend
(1)P
rotection 0.009
Grow
th s,t (0.013)
Sta
teL
inear
Tren
d(2)
0.009(0.013)
No
Lin
earT
rend
(3)0.009
(0.014)
Level
Con
trols
(4)
0.009(0.012)
Con
trols +In
c-Year
Uep
-Year
(5)
0.013(0.013)
Even
tually
Trea
ted(6)
0.009(0.012)
Ch
anged
On
ce
(7)0.009
(0.015)
Ch
ange
> 0.15(8)
0.010(0.013)
ProtectionL
evel s,t
Unem
ployment
-0.005* -0.004
-0.005* -0.002
-0.005 -0.011***
-0.004 -0.005*
-0.005*R
ate Change
(0.003) (0.003)
(0.003) (0.003)
(0.004) (0.003)
(0.003) (0.003)
(0.003)
House P
rice -0.005
-0.013 0.107**
-0.104 -0.134
-0.230* 0.049
-0.005 -0.007
Index Grow
th (0.113)
(0.113) (0.054)
(0.124) (0.125)
(0.118) (0.150)
(0.113) (0.112)
0.000 0.007
(0.024) (0.027)
Income
0.059G
rowth
(0.038)0.124***
0.127***(0.032)
(0.030)0.031
(0.032)0.020
0.121*** 0.054
0.059 0.059
(0.032) (0.043)
(0.041) (0.038)
(0.038)
-0.011** -().009*
(0.005) (0.005)
0.009 0.033
(0.043) (0.045)
0.026 0.029
(0.029) (0.030)
13,302 13,302
13,302 13,302
50 50
50 50
Y
YY
-0.005 0.0
24
**
*
(0.004) (0.005)
0.107* 0.061
(0.055) (0.069)
0.249*** 0.928***
(0.038) (0.1)08)
6,078 11,478
13,302 13,302
26 39
50 50
Y
Y
Y
Y
Y
Y
Y
Y
Y
Y
Y
Y0.19
0.17 0.19
0.19 0.20
0.18 0.18
0.18
15,51950Y
15,51950Y
Y
Y0.85
0.97
Notes.
Th
is table sh
ow
s the estim
ated co
efficient
follo
win
g sp
ecification
(1)
of log ch
anges
to au
to d
ebt o
n log ch
anges
in b
ank
rup
tcy p
rotectio
n
at the co
unty
level. D
ebt co
unty
data
is from
th
e FR
BN
YC
onsu
mer
Cred
it P
anel/E
quifa
x.
Pro
tection
Gro
wth
is th
e log
chan
ge in
the level o
f pro
tection
in
state
s at tim
e t. P
rotectio
n L
evel is th
e level
of p
rotectio
n
in state
s at
time t.
Unem
plo
ym
ent
rate
chan
ge is th
e ch
ange in u
nem
plo
ym
ent
rate
in co
un
ty i
at time t fro
m
BL
S. H
ou
se price g
row
th
is the log ch
ange in th
e
FH
FA
state
level index
for sta
te s at
time t, an
d
Inco
me g
row
th
is the
inco
me log
chan
ge in co
unty
i at tim
e t from
IRS
. C
olu
mns
1 and
2 sho
w th
e result
usin
g co
un
ty an
d sta
te fix
ed effects resp
ectively
in the first d
ifference
specificatio
n.
Colu
mn 3 sh
ow
s the resu
lts if we ex
clude sta
teor co
unty
fix
ed
effect fro
m sp
ecification
1. C
olu
mn
4 sho
ws
the estim
ates in
clud
ing
level o
f the co
ntro
ls. C
olu
mn 5
show
s the
estimates
inclu
din
g
level co
ntro
ls an
d
inco
me
and
un
emp
loym
ent
gro
ups
times y
ear fixed
effect, to allo
w for d
ifferential
trends acro
ss state
s based
on
these o
bserv
able ch
aracteristics. C
olu
mn 6 sh
ow
s the estim
ates for a regressio
n th
at
only
uses sta
tes tre
ate
d d
urin
g th
e samp
leperio
d, th
at
is, state
s that ch
anged
th
eir lev
el of pro
tection durin
g th
e sam
ple p
eriod
. C
olu
mn 7 sh
ow
s the
results if w
e only
consid
er as tre
ate
d sta
te th
at ch
anged
once.
Colu
mn
8 sho
ws th
e estim
atesif w
e rep
lace b
y zero ch
anges
belo
w 0.15.
Colu
mn
9 sho
ws resu
lts if w
e replace
the ch
ange
with
a dum
my in
dicato
r th
at is o
ne if th
e chan
ge is
greater th
an
zero
. C
olu
mn
10 and
11 show
th
e results of
regressio
n
log lev
els o
f auto
d
ebt o
n log
levels o
f pro
tection an
d in
clud
ing
co
unty
and
state
fix
ed effect
respectiv
ely.
Th
e sam
ple p
eriod
is fro
m
1999 to 2005.
*, *
and
*** d
eno
tes significan
ce at th
e10%
, 5%
, and
1%
cluster
at the sta
te
level resp
ectively
.
Levels
Level
onL
evelC
ou
nty
FE
(10)
Du
mm
yT
reatm
ent
(9)
0.002(0.008)
Level
onL
evelS
tate F
E(11)
Unem
ployment
Rate
House P
rice
Income
No.
of Obs.
No. of C
lustersC
ounty FE
State F
EY
ear FE
R-S
quared
13,30250YY0.18
Table 1.6: Determinants of Bankruptcy Protection Levels and Changes
Protection Level s,t
(1) (2)
House Price/Growth st
House Price/Growth st-1
Medical Exp./Growth st
Medical Exp./Growth st-i
Unemp. Rate/Change s,t
Unermp. Rate/Change st-1
State Real GDP/Growth s,t
State Real GDP/Growth st-1
No. Filings/Growth st
No. Filings/Growth st-1
-3.900(4.616)5.287
(4.503)
-3.332(5.359)4.635
(5.238)
-0.023(0.190)0.033
(0.148)
3.703(4.464)-6.950
(3.916)
-0.299*(0.250)-0.482
(0.245)
-1.837***(0.671)
2.983***(0.770)
0.836(1.001)0.348
(1.106)
0.028(0.036)-0.081*(0.042)
0.504(0.871)-1.448(0.742)
0.125*(0.039)
0.194***(0.072)
Protection Growth s,t
(3) (4)-0.809**
(0.354)1.691***(0.619)
-0.316(0.644)-0.537(0.763)
0.005(0.027)-0.016(0.028)
0.474(0.668)-0.277(0.282)
0.030(0.045)0.053
(0.047)
-0.537(0.572)0.970
(0.762)
-1.150(0.821)-1.805*(1.001)
0.002(0.033)-0.008(0.032)
1.028(1.018)0.425
(0.457)
-0.123(0.098)-0.045
(0.071)
Protection Dummy s,t
(5)-0.697
(0.701)2.700***(0.776)
-1.101(1.270)-1.020
(1.115)
0.027(0.042)-0.056
(0.050)
-1.665(1.034)-1.429
(0.789)
0.060*(0.069)0.026
(0.064)
(6)
-0.858(0.789)1.806*(0.994)
-2.380(1.834)-2.274*(1.287)
0.026(0.048)-0.058(0.065)
-0.911(1.343)-0.547(0.802)
-0.114(0.098)-0.080(0.090)
Political Climate st-1
Personal Income/Growth s,t
Personal Income/Growth s,t-1
No. of Obs.State FEYear FE
R2
0.045** -0.289*** 0.010 0.400(1.509) (0.171) (0.161) (0.234)
15.885*(8.597)
-13.235*(9.202)
350
Y0.13
1.077(1.257)-0.219*(1.206)
350YY
0.12
1.554(1.299)-0.720(0.929)
0.996(1.928)-1.159(1.477)
300 300Y
Y Y0.07 0.22
0.151 0.608(0.151) (0.458)
3.264(2.009)-0.525*(1.849)
3.190(2.399)-0.893
(2.200)
300 300Y
Y Y0.13 0.25
Note. This table shows the estimated coefficient of regression of bankruptcy protection on contemporaneous and lagvalues of variables that could determinate the changes in protection levels. House Price s,t is the level or growth ofhouse prices in state s at time t, from FHFA. Medical expenses is the level of growth in state's annual total medicalexpenses from the National Health Statistic. No. of Filings, is the number or change in the number of filings fornon-business bankruptcies in a state. Political Climate s,t is defined as the share of democratic votes in the closerHouse of Representative election. State GDP and Personal Income are from BEA, and Unemployment Rate fromBLS. Columns 1 and 2 show the coefficient of regressions of the protection level on levels of the explanatory variablesusing only year, and year and state fixed effect. Columns 3 and 4 show the coefficient of regressions of the growthin protection on growth of the explanatory variables using only year, and year and state fixed effect. Columns 5 and6 show the coefficient of regressions of a dummy that is one if the growth in protection is greater than zero on theexplanatory 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
1 Period
No County CountyLinear Trend Linear Trend Linear Trend
(1) (2) (3)Protection
Growth st-2
Protection -0.008Growth st-1 (0.008)
Protection 0.018**Growth st (0.007)
Protection 0.002Growth st+ (0.006)
ProtectionGrowth st+2
Unemployment 0.002Rate Change (0.002)
House Price -0.139***Index Growth (0.037)
Income 0.143***Growth (0.040)
Unemployment
Rate
House Price
Income
No. of ObsNo. of ClustersCounty FEYear FER-Squared
13,30250
Y0.28
Note. This table shows the estimated coefficient following specification (1) of log changes to credit card debt onlog changes in bankruptcy protection at the county level. Debt county data is from the FRBNY Consumer CreditPanel/Equifax. Protection Growth is the log change in the level of protection in state s at time t. Unemploymentrate change is the change in unemployment rate in county i at time t from BLS. House price growth is the log changein the FHFA state level index for state s at time t, and Income growth is the log change in income in county i at timet from IRS. Columns 1 and 4 show the without the inclusion of county fixed effects, including one lag and lead, andtwo lags and two leads. Columns 2 and 5 show the results with the inclusion of county fixed effect for including onelag and lead, and two lags and two leads, Columns 3 and 6 are the same than before but including level controls. Thesample period is from 1999 to 2005. *, **, and *** denotes significance at the 10%, 5%, and 1% cluster at the statelevel respectively.
54
-0.010(0.010)
0.019**(0.008)
0.006(0.008)
0.002(0.002)
-0.108(0.085)
0.080*(0.047)
13,30250YY
0.30
NoLinear Trend
(4)0.001
(0.019)
-0.007(0.009)
0.018**(0.007)
0.003(0.006)
0.010**(0.005)
0.002(0.002)
-0.142***(0.037)
0.143***(0.040)
-0.012(0.009)
0.016**(0.007)
0.006(0.009)
0.000(0.002)
-0.212**(0.101)
0.073*(0.042)
0.005*(0.003)
0.085(0.030)
0.024(0.021)
13,30250YY
0.30
2 Periods
CountyLinear Trend
(5)-0.004
(0.026)
-0.007(0.015)
0.022**(0.009)
0.010(0.010)
0.016***(0.005)
0.002(0.002)
-0.120
(0.085)
0.080*(0.047)
13,30250YY
0.30
13,30250
Y0.28
CountyLinear Trend
(6)-0.005(0.025)
-0.010(0.015)
0.020**
(0.008)
0.010(0.011)
0.016***(0.005)
0.001(0.002)
-0.229**(0.100)
0.072*(0.041)
0.004(0.003)
0.086(0.029)
0.025(0.021)
13,30250YY
0.31
Table 1.8: Local Business Conditions.ders. Credit Card Debt
Neighboring County-pairs across State Bor-
AllCounty-Pairs
CountyLinerTrend
(2)
Equal IncomeCounty-Pairs
StateLinearTrend
(3)
CountyLinerTrend
(4)
Low IncomeCounty-Pairs
State CountyLinear LinerTrend Trend
(5) (6)
Protection -0.006 -0.005 0.015 0.015* 0.099** 0.098**Growth s,t (0.011) (0.011) (0.010) (0.009) (0.046) (0.044)
Unemployment 0.003** 0.003** 0.002 0.001 0.002* 0.001**Rate Change (0.002) (0.002) (0.003) (0.003) (0.005) (0.005)
House Price -0.322** -0.317** -0.266 -0.261 -1.040* -1.037**Index Growth (0.157) (0.154) (0.178) (0.171) (0.550) (0.526)
Income 0.095*** 0.043 0.122* 0.066 0.121 0.102
Growth (0.024) (0.027) (0.071) (0.075) (0.125) (0.122)
No. of Obs 9,168 9,168 3,984 3,984 1,188 1,188No. of Clusters 48 48 46 46 33 33
County FE Y Y YState FE Y Y Y
County-Pair-Year FE Y Y Y Y Y YR-Squared 0.70 0.70 0.67 0.67 0.63 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
StateLinearTrend
(1)
Table 1.9: Heterogeneous Treatment of Bankruptcy Protection on Credit Card Debt:Income and Home ownership
Low Income
HomeOwnership
Med Income
HomeOwnership
High Income
HomeOwnership
(1) (2) (3) (4) (5) (6) (7)Protection Growth s,t
Protection Growth stx Low Income
Protection Growth s,tx Low Home Ownership
Protection Growth stx Med Income
Protection Growth stx Med Home Ownership
0.007(0.007)
0.022***(0.007)
0.028** 0.063***(0.011) (0.018)
0.020** 0.029(0.010) (0.019)
-0.050***(0.018)
0.006 0.014
(0.006) (0.009)
-0.012(0.025)
-0.011
(().009)
0.013**(0.006)
-0.049***(0.016)
-0.014(0.019)
-0.013(0.012)
UnemploymentRate Change
0.003 0.005*(0.002) (0.003)
0.005*(0.003)
House Price -0.109 -0.015 -0.012Index Growth (0.086) (0.094) (0.095)
Income 0.137*** 0.059** 0.057*Growth (0.040) (0.030) (0.031)
0.002 0.002(0.002) (0.002)
-0.099 -0.099(0.098) (0.098)
0.002 0.002(0.003) (0.003)
-0.208** -0.206**(0.092) (0.093)
0.090*** 0.088*** 0.240*** 0.227***(0.032) (0.028) (0.062) (0.064)
No. of Obs 13,302 4,536 4,536 4,422 4,422 4,344 4,344No. of Clusters 50 50 50 50 50 50 50
State and Year FE Y Y Y Y Y Y YR-Squared 0.29 0.24 0.24 0.29 0.30 0.46 0.48
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/MedOwnership represents counties in the lowest/middle tercile of the within income bucket distribution. Column 1 showsthe 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 to2005. *, **, and *** denotes significance at the 10%, 5%, and 1% cluster at the state level respectively.
56
Income
Tab
le 1
.10:
E
ffec
t of
Ban
krup
tcy
Pro
tect
ion
on I
nter
est
Rat
es:
Per
sona
l U
nsec
ured
Loa
ns a
nd C
redi
t C
ards
Per
son
al U
nse
cure
d
Loa
nC
red
it C
ard
Deb
t
St
Lin
ear
Tre
nd
(1)
Cty
Lin
ear
Tre
nd
(2)
Eve
ntu
ally
Cty
Lin
ear
Tre
nd
(3)
Pro
tect
ion
Gro
wth
s.t
-2
Pro
tect
ion
Gro
wth
st-
1
Cty
Lin
ear
Tre
nd
(4)
-0.2
60(0
.395
)
Co
un
ty-P
air
sS
t L
inea
r C
ty L
inea
rT
ren
d
Tre
nd
(5)
(6)
-0.0
22(0
.274
)
Pro
tect
ion
0.38
9***
0.
415*
**
0.37
3**
0.29
6*
0.75
5***
0.
820*
**G
row
th s
t (0
.147
) (0
.144
) (0
.147
) (0
.170
) (0
.177
) (0
.157
)
Pro
tect
ion
-0.1
32G
row
th s
,t+
1
(0.1
06)
Pro
tect
ion
Gro
wth
s.t+
2-0
.286
(0.2
05)
Une
mpl
oym
ient
0.
003
0.00
1 -0
.020
-0
.009
0.
106
0.08
4
Rat
e C
hang
e (0
.046
) (0
.050
) (0
.073
) (0
.048
) (0
.103
) (0
.107
)
Hou
se P
rice
4.
938*
**
4.81
2***
4.
363*
* 5.
154*
**
-0.1
12
1.07
2In
dex
Gro
wth
(1
.629
) (1
.623
) (2
.159
) (1
.607
) (3
.153
) (3
.315
)
Inco
me
0.19
8 0.
182
0.55
1 0.
203
2.29
9*
2.90
4
Gro
wth
(0
.268
) (0
.385
) (0
.622
) (0
.383
) (1
.255
) (1
.936
)
St
Lin
ear
Tre
nd
(7)
Cty
Lin
ear
Tre
nd
(8)
Eve
ntu
ally
Cty
Lin
ear
Tre
nd
(9)
Cty
Lin
ear
Tre
nd
(10)
0.58
4(0
.464
)
Cou
nty
-Pair
sS
t L
inea
r C
ty L
inea
rT
ren
d
Tre
nd
(11)
(1
2)
0.08
3(0
.677
)
0.00
7 0.
147
-0.0
04
0.31
7 0.
875*
0.
775
(0.2
17)
(0.1
83)
(0.2
32)
(0.2
29)
(0.5
15)
(0.5
73)
0.30
8*(0
.166
)
0.25
6(0
.273
)
-0.1
18
-0.1
03
-0.1
00
-0.0
86
-0.0
38
-0.0
59(0
.089
) (0
.090
) (0
.096
) (0
.095
) (0
.151
) (0
.160
)
5.17
9 3.
691
2.60
6(3
.984
) (3
.895
) (4
.532
)3.
625
-5.8
57
-5.0
49(4
.014
) (7
.780
) (8
.608
)
1.73
4***
1.
886*
**
1.44
0 1.
864*
**
-0.2
24
-0.8
68(0
.558
) (0
.600
) (0
.973
) (0
.605
) (4
.195
) (4
.905
)
No.
of
Obs
No.
of
Clu
ster
sC
ty a
nd Y
ear
FE
Sta
te a
nd Y
ear
FE
R-S
quar
ed
4.69
349 Y 0.17
4.69
349 Y
2,33
825 Y
4.69
349 Y
1,62
144
1,62
144 Y
Y
5.37
150 Y
5.37
150 Y
2.43
026 Y
5,37
150 Y
1.62
145
1,62
145 Y
Y
0.13
0.
15
0.14
0.
79
0.80
0.
29
0.21
0.
23
0.21
0.
82
0.82
No
te.
This
tab
le s
how
s th
e es
tim
ated
coef
fici
ent
foll
ow
ing a
var
iati
on
of
sp
ecif
icat
ion
(1
) of
chan
ges
in
inte
rest
ra
tes
(%)
on c
han
ges
in
the l
evel
of
pro
tect
ion
. P
erso
nal
Unse
cure
d
Lo
an a
nd
C
red
it
Car
d
Deb
t ar
e co
un
ty a
ver
ages
of
the
inte
rest
ra
tes
in a
co
unty
fo
r ea
ch ty
pe
of
cred
it.
Colu
mns
1 an
d
7 sh
ow
the
resu
lt
usi
ng s
tate
fix
ed e
ffec
t.
Colu
mns
2 an
d
8 sh
ow
the
esti
mat
es u
sing
county
fix
ed e
ffec
t.
Co
lum
ns
3 an
d
9 sh
ow
th
e
resu
lt r
estr
icti
ng th
e sa
mp
le t
o o
nly
th
e "e
ven
tual
ly"
treate
d
sam
ple
. C
olu
mns
4 an
d
10 s
how
the
esti
mat
es lo
okin
g at
the
dynam
ic e
ffec
t of
chan
ges
in
pro
tect
ion
on i
nte
rest
rate
s.
Colu
ms
5,
6,
11,
and
12 s
how
the
resu
lts
incl
ud
ing
sta
te a
nd
co
un
ty
fixed
eff
ect
for
the
sub
-sam
ple
of
nei
gh
bo
rin
g c
ou
nty
-pai
rs
for
wh
ich
both
counti
es a
re
in t
he
sam
e in
com
e buck
et.
The
sam
ple
per
iod
is
fro
m
1999
to 2
005.
*,
**,
and
***
d
eno
tes
signif
ican
ce
at t
he
10%
, 5%
, an
d
1%
clust
er at
th
e st
ate
lev
el r
esp
ecti
vel
y.
Table 1.11:
Effect of B
ankruptcy Protection on Interest R
ates: M
ortagage Credit
3 Yr-A
RM
St L
inear
Cty L
inear
Tren
d
Tren
d
(1) (2)
Even
tually
Cty
Lin
earT
rend
(3)
15 Yr-F
ixed
Even
tually
St L
inear
Cty L
inear
Cty L
inear
Tren
d
Tren
d
Tren
d(4)
(5) (6)
Protection
Grow
th s,t
Unem
plo
ym
ent
Rate
Ch
ang
e
0.037(0.051)
0.053(0.062)
0.041(0.057)
-0.066*** -0.100***
-0.048**(0.031)
(0.041) (0.026)
House P
rice 2.244***
2.332*** 2.690**
Index Grow
th (0.648)
(0.677) (1.094)
Income
-0.093G
rowth
(0.228)-0.191(0.290)
-0.485(0.374)
0.014
(0.041)
-0.001(0.009)
0.009(0.319)
-0.003
(0.085)
0.019(0.042)
-0.002
(0.011)
0.045
(0.332)
-0.005
(0.118)
0.005
(0.035)
-0.022
(0.017)
0.637(0.403)
-0.136
(0.111)
0.026
(0.029)
0.001
(0.019)
-0.039
(0.246)
-0.029
(0.107)
0.029(0.030)
0.004
(0.022)
0.017(0.252)
-0.034
(0.139)
0.027
(0.034)
-0.040
(0.017)
0.234
(0.261)
00
-0.317***
(0.115)
No.
of Obs
3,919 3,919
1,945 5,723
5,723 2,802
5,533 5,533
2;732N
o. of C
lusters 47
47 24
50 50
26 49
49 25
Cty and Y
ear FE
Y
Y
y
y
y
yS
tate and Year F
E
Y
y
yR
-Squared
0.85 0.85
0.85 0.87
0.86 0.87
0.86 0.85
0.87N
ote.
Th
is table
sh
ow
s the estim
ated
coefficien
t fo
llow
ing
a v
ariation
of sp
ecification
(1)
of chan
ges
in in
terest rates (%
) in th
e level o
f pro
tection.
3 Yr-A
RM
, 15
Yr-F
ixed
, 30 Y
r-Fix
ed,
are county
averag
es o
f the in
terest rates in
a cou
nty
for each
ty
pe o
f credit.
Colu
mns 1,
4, an
d
7 sho
w
the resu
lt usin
g sta
te fix
ed effect.
Colu
mns
2, 5 an
d 8,
show
the estim
ates usin
g
coun
ty fix
ed effect.
Colu
mns
3, 6
and
9, show
the
result re
strictin
g th
e sam
ple
to o
nly
th
e
"even
tually
" treate
d
samp
le. T
he sam
ple p
eriod
is from
1999 to
2005. *, *,
and ***
den
otes sig
nifican
ce at th
e 10%
, 5%,
and
1%
cluster at
the state
level resp
ectively
.
30 Yr-F
ixed
St L
inear
Tren
d(7)(7)
Cty
Lin
earT
rend
(8)
Even
tually
Cty
Lin
earT
rend
(9)
Table 1.12: Effect of Bankruptcy Protection on Debt. Number of Credit Cards and
Entry
Panel A. All individualsNumber of Credit Cards
A inDebt Balance
(1)
Protection 0.076***Growth s.t (().009)
Unenploynient 0.002Rate Change (0.003)
House Price -0.070*Index Growth (0.041)
Incone 0.012Growth (0.016)
N of Ohs 366,362N of Clusters 40
YR-Squaredl 0.00
A inN Credit Cards
(2)
0.054***(0.019)
0.008**(0.004)
-0.050(0.037)
-0.048***(0.016)
619,72640Y
0.02
A inN Credit Cards
Conditional on n>0(3)
0.082***(0.026)
0.008(0.005)
-0.043(0.049)
-0.017(0.018)
454,68840Y
0.02
A inN Credit Cards
Conditional on n>0& Balance >0
(4)
0.093***(0.029)
0.009*(0.005)
-0.039(0.044)
0.001(0.017)
359,23540Y
0.02
Entry
Open FirstCredit Card
(5)
0.001(0.003)
0.001(0.001)
-0.005(0.008)
-0.011*** -0.063***(0.004) (0.014)
555,00740Y
0.01
221,849:39Y
0.01
Panel B. Home ownersNumber of Credit Cards
A inDebt Balance
(1)
Protection 0.102***Growth s.) (0.014)
Uneniployient 0.000Rate Change (0.004)
House Price -0.088*Index Growth (0.052)
Incone 0.014Growth (0.017)
N of ObsN of Clusters
Cty and Year FER-Squared
210.86339Y
0()0
A inN Credit Cards
(2)
0.081***(0.020)
0.009*(0.005)
-0.052(0.057)
-0.036*(0.021)
304,00539Y
0.02
A inN Credit Cards
Conditional on n>0(3)
0.103***(0.024)
0.009(0.006)
-0.045(0.067)
-0.006(0.024)
248,95539Y
0.02
A inN Credit Cards
Conditional on n>O& Balance >0
(4)
0.115***(0.032)
0.009(0.007)
-0.032(0.071)
0.006(0.021)
205,458:39Y
10.02
Entry
Open FirstCredit Card
(5)
-0.002(0.003)
0.000(0.001)
-0.003(0.007)
-0.005*(0.003)
291,35339Y
Credit CardBalance
Becomes >0(6)
-0.006(0.006)
0.001(0.002)
-0.044(0.029)
-0.060***(0.016)
10:3,85437Y
0.01 0.01
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
Credit CardBalance
Becomes >0(6)
-0.002(0.006)
0.001(0.002)
-0.031(0.020)
Table 1.13: E
ffect of Bankruptcy P
rotection on Credit C
ard Delinquency
Pan
el A. A
ll in
divid
uals
90+
day
s120+
d
ayst
t+1
t+
2
t+3
t
t+1
t+
2
t+3
(1) (2)
(3) (4)
(5) (6)
(7) (8)
Pro
tection
-0.001 -0.008**
0.000 0.004
-0.002 -0.009**
-0.001 0.003
Grow
th s,t (0.004)
(0.004) (0.004)
(0.003) (0.005)
(0.004) (0.003)
(0.004)
N of O
bs 366,362
N of C
lusters 40
Cty
and Y
ear FE
Y
Uep
/Inco
me/H
P
Controls
YR
-Squared
0.02
363,49840Y
361,44440Y
359,78340Y
Y
Y
Y0.02
0.02 0.01
366,36240YY
0.02
363,498 361,444
359,78340
40 40
Y
Y
YY
Y
Y
0.02 0.02
0.01
t(9)
-0.003
(0.003)
366,36240YY0.02
Severe
t+1
t+2
(10) (11)
-0.008** 0.001
(0.004) (0.002)
363,49840YY0.02
361,44440YY
0.02
Pan
el B. H
ome ow
ners
t(1)
Pro
tection
-0.004G
rowth s,t
(0.004)
90+
days
t+1
t+2
t+
3(2)
(3) (4)
-0.014*** -0.007
0.003(0.005)
(0.007) (0.002)
t(5)
-0.003 -0
(0.003)
120+
days
t+1
t+
2
t+3
(6) (7)
(8).014***
-0.007 0.001
0.004) (0.006)
(0.003)
t(9)
-0.001
(0.003)
Severe
t+1
t+2
t+
3(10)
(11) (12)
-0.013*** -0.005
0.000
(0.005) (0.004)
(0.003)
N of O
bs 210,863
N of C
lusters 39
Cty and Y
ear FE
Y
Uep
/Inco
me/H
P
Controls
YR
-Squared
0.02
209,878 209,173
208,61639
39 39
Y
Y
YY
Y
Y
0.02 0.02
0.01
210,86339YY0.02
209,87839YY0.02
209,17339YY0.02
208,61639YY0.01
210,86339YY0.02
209,878 209,173
208,61639
39 39
Y
Y
YY
Y
Y
0.02 0.02
0.01N
ote. T
his table shows the estim
ated coefficient follow
ing a variation of specification (1),
where w
e replace the dependent variable for a dum
my indicator th
at is equal to 1if the person is delinquent at
the specified tim
e. P
anel A
uses all individuals in counties below
the m
edian income w
ith a positive balance.
Panel B
restricts the sample to
homeow
ners, defined as individuals for w
hom som
e home debt is observed during the sam
ple period. C
olumns 1 to 4 show
the estimates w
here delinquency is defined as beingdelinquent 90 days or m
ore. C
olumn 5 to 8 show
the estimates w
here delinquency is defined as being delinquent 120 days or m
ore. C
olumns 9 to 12 show
the estimates w
heredelinquency is defined
as being severely delinquent. A
ll regressions include controls. T
he sample period
is from 1999 to 2005.
*, **,
and *** denotes significance at the 10%
,5%
, and 1%
cluster at the state level respectively
t+3
(12)
0.002
(0.002)
359,78340YY0.01
Table 1.14: Effect of Bankruptcy Protection on Self-Employment
Self Employment
Protection Gowth s,t
Protection Gowth s,tx Low Income
Protection Gowth stx Med Incoie
(1)0.000
(0.002)
Unemployment 0.001***Rate Change (0.000)
Honse Price 0.096***Index Growth (0.023)
(2)
-0.003(0.003)
(3)-0.0l0(**(0.004)
0.006** 0.012***(0.003) (0.004)
0.003 0.008***(0.002) (0.003)
0 .0 0 1 *** 0.01***(0.000) (0.001)
0.097*** 0.058**(0.022) (0.028)
Credit CardStartup > p50
(4)-0.002(0.007)
(5)-0.014(0.009)
Credit CardStartup < p50
(6) (7)
-0.003 -0.007(0.002) (0.003)
0.024***(0.007)
0.012**(0.005)
0.005(0.004)
0.006(0.003)
0.001 0.001 0.001 0.001(0.001) (0.001) (0.001) (0.001)
0.057 0.056 0.059* 0.059(0.035) (0.035) (0.033) (0.033)
Income 0.063*** 0.063*** 0.101*** 0.126*** 0.127*** 0.085*** 0.085Growth (0.010) (0.009) (0.028) (0.037) (0.037) (0.025) (0.025)
Number of ObservationsNumber of Clusters
State FEState x 2-digit industry
12,73850Y
12,73850Y
194,01150
YYear FE Y Y Y
R-Squared 0.21
73,081 73,081 120,930 120,93050 50 50 50
Y Y Y YY Y Y Y
0.23 0.01 0.02 0.02 0.02 0.02
Note. This table shows the estimated coefficient following a variation of specification (1) of log changes in self-employment measures on log changes in the levels of protection. Column 1 shows the estimates for county self-employment aggregates. Column 2 shows the results for the effect interacted with income heterogeneity for aggregateself-employment. Column 3 shows the estimates interacted with low income using self-employment changes by industryand county. Column 4 and 5 show the estimates for industries that used the level of credit card debt as a start-upcapital and Column 6 and 7 for industries that do not. The sample period is from 1999 to 2005. *, **, anddenotes significance at the 10%, 5%, and 1% cluster at the state level respectively.
61
Table 1.15:
Effect of B
ankruptcy Protection on C
redit Card D
ebt. A
lternative Specifications
Event
Baselin
e B
aseline
Cty
Lin
ear C
ty L
inear
(1) (2)
Protection
0.018** 0.017**
Gow
th st
(0.008) (0.008)
Unlim
itedP
rotection s,t
Unem
ployment
0.002 0.002
Rate C
hange (0.002)
(0.003)
House Price
-0.102 -0.118
Index Grow
th (0.086)
(0.086)
Income
0.079*G
rowth
(0.047)0.138*(0.077)
Event
Unlim
ited
Unlim
itedC
han
ge
Chan
ge
Event
Unem
p.
Unem
p.
Insu
rance
Insu
rance
Event
Deb
t to
Deb
t toIncom
e Incom
e
Event
% C
han
ge
% C
han
ge
in Deb
t in D
ebt
Event
Hom
estead
Hom
esteadO
nly O
nly(3)
(4) (5)
(6) (7)
(8) (9)
(10) (11)
(12)0.018**
0.017** 0.018**
0.017** 0.023**
0.020* 0.022**
0.020**(0.008)
(0.008) (0.008)
(0.008) (0.011)
(0.011) (0.009)
(0.010)
-0.156*** -0.139***
(0.027) (0.027)
0.002 0.002
0.002 0.002
0.008*** 0.008***
(0.002) (0.003)
(0.002) (0.003)
(0.002) (0.003)
0.002 0.002
(0.002) (0.003)
-0.103 -0.119
-0.099 -0.134
-0.210** -0.155
-0.106 -0.109
(0.086) (0.086)
(0.088) (0.091)
(0.094) (0.111)
(0.094) (0.097)
0.080* 0.139*
(0.047) (0.077)
0.080* 0.138*
(0.047) (0.076)
0.065* 0.111*
(0.037) (0.059)
0.017*** 0.015**
(0.006) (0.006)
0.002 0.003
(0.002) (0.003)
-0.111 -0.130
(0.087) (0.091)
0.081* 0.141*
(0.047) (0.077)
N of Obs
13,302N of C
lusters 50
Cty and year F
E
YR
-Squared 0.30
6,078 13,308
6,084 13,302
6,078 13,302
6,078 13,302
6,07826
51 27
50 26
50 26
50 26
Y
Y
Y
Y
Y
Y
Y
Y
Y0,29
0.30 0.29
0.30 0.29
0.22 0.18
0.29 0.27
Note.
Th
is table
sho
ws th
e estim
ated co
efficient
follo
win
g a v
ariation o
f the sp
ecification
(1). C
olu
mn
s 1 an
d 2 rep
licated th
e m
ain resu
lts. C
olu
mns 3
and
4 sho
w th
e result w
hen
un
limited
chan
ge o
f DC
is inclu
ded
as a d
um
my. C
olu
mns 5
and
6 sho
w th
e results
wh
en co
ntro
lling
for level o
f un
emp
loy
men
t in
suran
ce. C
olu
mns 7 an
d 8 rep
lace the
dep
enden
t variab
le fo
r deb
t to in
com
e ch
ange.
Colu
mns 9
and
10 replace th
e
dep
end
ent
variab
le for percen
tage
chan
ges
in level o
f deb
t, and C
olu
mns 11
and
12 sh
ow
the resu
lt if ch
anges
in the level o
f pro
tection
are measu
red on
ly as a h
om
e-equity
p
rotectio
n.
The sam
ple
perio
d
is from
1999 to
2005. *,
**, an
d ***
den
otes sig
nifican
ce at
the
10%,
5%, an
d
1%
cluster
at the
state
level resp
ectively
.
13.14048Y0.30
5,91624Y0.29
Table 1.16: Other Heterogeneous Treatment of Bankruptcy Protection. Credit Card
Debt
Protection Gowth st
Protection Gowth s,t
x Low Income
Protection Gowth s,tx Med Income
UnemploymentRate Change
Low IncBaseline
(1)0.028**(0.011)
0.005*(0.003)
House Price -0.015Index Growth (0.094)
Income 0.059**Growth (0.030)
BankConcentration
(2)
0.085***(0.021)
-0.086***(0.022)
-0.076***(0.016)
0.005*(0.003)
-0.012(0.094)
0.060**(0.030)
Total Credit CardDebt/Income Debt/Income
(3)
0.041**(0.018)
-0.028(0.021)
-0.009(0.031)
0.005*(0.003)
-0.013(0.095)
0.062**(0.031)
(4)
0.043*(0.024)
-0.005(0.039)
-0.040(0.037)
0.005*(0.003)
-0.018(0.095)
0.101***(0.031)
Number Credit Cardof Filing 90+ Delinq
(5) (6)
0.026** 0.048***(0.012) (0.017)
-0.004(0.026)
0.010(0.019)
0.005*(0.003)
-0.018(0.094)
0.064**(0.030)
-0.042***(0.014)
-0.014(0.018)
0.005**(0.002)
-0.013(0.094)
0.058**(0.029)
N of ObsN of Clusters
State and year FER-Squared
4,53650Y
0.24
4,53650Y
0.24
4,53650Y
0.24
4,53650Y
0.25
Note. This table shows the estimated coefficient following a variation of specification (1) that
4,536 4,53650 50Y Y
0.24 0.24
incorporate interactions,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 incomeheterogeneity. 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. EventuallyTreated
Protection Level s,t
House Price/Growth s,t
House Price/Growth st-i
Medical Exp./Growth s,t
Medical Exp./Growth st-1
Unemp. Rate/Change s,t
Unemp. Rate/Change st-1
State Real GDP/Growth s,t
State Real GDP/Growth st-1
No. Filings/Growth s,t
No. Filings/Growth st-1
(1)-1.563
(2.581)3.301
(2.676)
-1.237(4.206)0.670
(4.642)
0.150(0.177)0.029
(0.129)
-0.994(5.869)-2.495
(5.177)
-0.284(0.190)-0.268(0.159)
(2)
-2.224**(0.970)
3.147***(0.985)
0.027(1.611)0.863
(2.067)
0.059(0.068)-0.093
(0.071)
0.899(1.774)-1.494(1.210)
0.073(0.051)0.158
(0.083)
Protection
(3)-0.984**(0.445)1.453*
(0.778)
-1.039(1.124)-0.733(1.443)
0.016(0.052)0.000
(0.060)
0.814(1.185)-0.241
(0.592)
0.004(0.069)0.035
(0.057)
Growth s,t
(4)-0.699(0.776)0.648
(1.259)
-1.533(1.604)-3.089*(1.590)
0.010(0.065)0.008
(0.069)
1.301(1.814)0.391
(0.807)
-0.129(0.134)-0.074(0.089)
Protection
(5)-1.123(0.924)2.087*(1.152)
-1.851(2.219)-2.245
(2.110)
0.100(0.080)-0.042(0.091)
-2.145(1.858)-1.183(1.411)
0.023(0.087)-0.030
(0.083)
Dummy s,t
(6)-1.503(1.135)1.631
(1.569)
-3.300(3.542)
-4.823**(2.277)
0.100(0.101)-0.077(0.129)
-1.589(2.519)-1.055
(1.419)
-0.073(0.126)-0.127
(0.112)
Political Climate st-1 0.209* -0.123*(1.547) (0.374)
0.060 0.924(0.266) (0.535)
0.375 1.536(0.211) (0.887)
Personal Income/Growth st
Personal Income/Growth st-1
No. of Obs.State FEYear FE
R2
196
Y0.27
196YY
0.12
168 168Y
Y Y0.08 0.21
Note. This table shows the estimated coefficient of regression of bankruptcy protection
168 168Y
Y Y0.18 0.24
on contemporaneous and lagvalues of variables that could determinate the changes in protection levels. House Price s,t is the level or growth ofhouse prices in state s at time t, from FHFA. Medical expenses is the level of growth in state's annual total medicalexpenses from the National Health Statistic. No. of Filings, is the number or change in the number of filings fornon-business bankruptcies in a state. Political Climate s,t is defined as the share of democratic votes in the closerHouse of Representative election. State GDP and Personal Income are from BEA, and Unemployment Rate fromBLS. Columns 1 and 2 show the coefficient of regressions of the level protection on level of the explanatory variablesusing only year, and year and state fixed effect. Columns 3 and 4 show the coefficient of regressions of the growthin protection on growth of the explanatory variables using only year, and year and state fixed effect. Columns 5 and6 show the coefficient of regressions of a dummy that is one if the growth in protection is greater than zero on theexplanatory variables' growth using only year, and year and state fixed effect. The sample period is from 1999 to2005. *, **, and *** denotes significance at the 10%, 5%, and 1% cluster at the state level respectively.
64
13.996*(7.586)-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)
Table 1.18: Dynamics of the Change in Protection. Mortgage Debt
1 Period 2 Periods
No County County No County CountyLinear Trend Linear Trend Linear Trend Linear Trend Linear Trend Linear Trend
(1) (2) (3) (4) (5) (6)
Protection -0.024 -0.043* -0.054**Growth st-2 (0.017) (0.025) (0.026)
Protection 0.019 0.013 0.005 0.018 0.002 -0.006Growth st-1 (0.013) (0.014) (0.012) (0.013) (0.017) (0.015)
Protection 0.007 0.011 0.005 0.006 0.005 -0.002Growth st (0.016) (0.014) (0.012) (0.016) (0.014) (0.013)
Protection -0.009 -0.006 -0.004 -0.010 -0.012 -0.010Growth st+1 (0.008) (0.009) (0.009) (0.008) (0.009) (0.010)
Protection -0.016* -0.014 -0.011Growth s,t+2 (0.009) (0.011) (0.011)
Unemployment -0.003 -0.004 -0.004 -0.003 -0.004 -0.005*Rate Change (0.003) (0.003) (0.003) (0.003) (0.003) (0.003)
House Price 0.046 0.092 -0.372** 0.049 0.092 -0.385**Index Growth (0.078) (0.163) (0.172) (0.075) (0.163) (0.173)
Income 0.190** 0.113 0.039 0.189** 0.114 0.040Growth (().091) (0.107) (0.078) (0.091) (0.107) (0.078)
Unemployment 0.001 0.000Rate (0.004) (0.004)
House Price 0.277 0.281(0.040) (0.039)
Income 0.133 0.132(0.039) (0.040)
No. of Obs 13,302 13,302 13,302 13,302 13,302 13,302No. of Clusters 50 50 50 50 50 50County FE Y Y Y YYear FE Y Y Y Y Y YR-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 logchanges in bankruptcy protection at the county level. Debt county data is from the FRBNY Consumer CreditPanel/Equifax. Protection Growth is the log change in the level of protection in state s at time t. Unemploymentrate change is the change in unemployment rate in county i at time t from BLS. House price growth is the log changein the FHFA state level index for state s at time t, and Income growth is the log change in income in county i at timet from IRS. Columns 1 and 4, show the without the inclusion of county fixed effects, including one lag and lead, andtwo lags and two leads. Columns 2 and 5 show the results with the inclusion of county fixed effect for including onelag and lead, and two lags and two leads, Columns 3 and 6 are the same than before but including level controls. Thesample period is from 1999 to 2005. *, **, and *** denotes significance at the 10%, 5%, and 1% cluster at the statelevel respectively.
65
Table 1.19: Dynamics of the Change in Protection. Auto Debt
NoLinear Trend
(1)
1 Period
CountyLinear Trend
(2)
CountyLinear Trend
(3)Protection
Growth s,t-2
NoLinear Trend
(4)-0.022(0.019)
2 Periods
CountyLinear Trend
(5)-0.015(0.026)
Protection -0.006Growth st-1 (0.013)
Protection 0.008Growth st (0.014)
Protection -0.012*Growth st+ (0.007)
ProtectionGrowth s,t+2
UnemploymentRate Change
-0.005*(0.003)
House Price 0.110**Index Growth (0.053)
Income 0.127***Growth (0.030)
UnemploymentRate
House Price
Income
No. of Ohs 13,302 13,302 13,302 13,302 13,302 13,302No. of Clusters 50 50 50 50 50 50County FE Y Y Y YYear FE Y Y Y Y Y YR-Squared 0.17 0.18 0.19 0.17 0.18 0.19
Note. This table shows the estimated coefficient following specification (1) of log changes to auto debt on log changesin 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 thechange in unemployment rate in county i at time t from BLS. House price growth is the log change in the FHFAstate 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 lagsand two leads. Columns 2 and 5 show the results with the inclusion of county fixed effect for including one lag andlead, and two lags and two leads, Columns 3 and 6 are the same than before but including level controls. The sampleperiod is from 1999 to 2005. *, **, and *** denotes significance at the 10%, 5%, and 1% cluster at the state levelrespectively.
66
CountyLinear Trend
(6)
-0.020(0.028)
-0.004(0.017)
0.006(0.011)
-0.011(0.010)
-0.005*(0.003)
0.002(0.113)
0.059(0.038)
-0.005(0.013)
0.009(0.014)
-0.011(0.007)
0.015(0.011)
-0.005*(0.003)
0.105*(0.054)
0.128***(0.030)
-0.002(0.017)
0.010(0.014)
-0.007(0.009)
0.020*(0.011)
-0.005*(0.003)
-0.015(0.114)
0.060(0.037)
-0.004(0.017)
0.007(0.010)
-0.008(0.011)
-0.002(0.003)
-0.097(0.124)
0.032(0.032)
-0.011**(0.005)
0.009(0.043)
0.025(0.030)
-0.003(0.016)
0.010(0.013)
-0.004(0.011)
0.022*(0.012)
-0.002(0.003)
-0.127(0.125)
0.031(0.032)
-0.012**(0.005)
0.012(0.042)
0.026(0.029)
Table 1.20: Local Business Conditions. Neighboring County-pairs across State Bor-
ders. Mortgage Debt
AllCounty-Pairs
State CountyLinearTrend
(1)
LinerTrend
(2)
Equal IncomeCounty-Pairs
State CountyLinearTrend
(3)
LinerTrend
(4)
Low IncomeCounty-Pairs
State CountyLinear LinerTrend Trend
(5) (6)
Note.
Protection 0.006 0.007 0.006 0.006 0.051 0.051Growth s,t (0.011) (0.011) (0.010) (0.010) (0.060) (0.058)
Unemployment -0.002 -0.002 0.001 0.000 -0.001 -0.001Rate Change (0.005) (0.005) (0.005) (0.005) (0.008) (0.008)
House Price -0.116 -0.109 -0.050 -0.046 0.077 0.074Index Growth (0.153) (0.150) (0.203) (0.196) (0.639) (0.617)
Income 0.089* 0.015 0.197*** 0.151* 0.160 0.177Growth (0.054) (0.064) (0.074) (0.083) (0.115) (0.126)
No. of Obs 9,168 9,168 3,984 3,984 1,188 1,188No. of Clusters 48 48 46 46 33 33
County FE Y Y YState FE Y Y Y
County-Pair-Year FE Y Y Y Y Y YR-Squared 0.65 0.64 0.62 0.61 0.55 0.53
This table shows the estimated coefficient following specification (2) of log changes in mortgage debt on logchanges in bankruptcy protection at the county level. Debt county data is from the FRBNY Consumer CreditPanel/Equifax. Protection Growth is the log change in the level of protection in state s at time t. Unemploymentrate 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-pairssample. 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 Bor-ders. Auto Debt
AllCounty-Pairs
State CountyLinear LinerTrend Trend
(1) (2)
Equal IncomeCounty-Pairs
State CountyLinear LinerTrend Trend
(3) (4)
Low IncomeCounty-Pairs
State CountyLinear LinerTrend Trend
(5) (6)
Protection 0.006 0.006 0.008 0.008 -0.018 -0.017Growth s,t (0.010) (0.010) (0.014) (0.013) (0.050) (0.048)
Unemployment 0.000 0.000 -0.001 -0.001 -0.004 -0.003Rate Change (0.004) (0.004) (0.005) (0.005) (0.006) (0.006)
House Price -0.079 -0.072 -0.275 -0.269 -0.381 -0.379Index Growth (0.197) (0.193) (0.213) (0.206) (0.406) (0.389)
Income 0.143*** 0.062 0.295*** 0.239** 0.285* 0.279*Growth (0.049) (0.057) (0.102) (0.118) (0.160) (0.167)
No. of Obs 9,168 9,168 3,984 3,984 1,188 1,188No. of Clusters 48 48 46 46 33 33
County FE Y Y YState FE Y Y Y
County-Pair-Year FE Y Y Y Y Y YR-Squared 0.70 0.70 0.67 0.67 0.60 0.60
rhis table shows the estimated coefficient following specification (2) of log changes in auto debt on log ch angesin 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 thechange in unemployment rate in county i at time t from BLS. House price growth is the log change in the FHFA statelevel index for state s at time t, and Income growth is the log change in income in county i at time t from IRS. Columns1 and 2, show the estimates for state and county fixed effect for all neighboring county-pairs sample. Columns 3 and4 show the results including state and county fixed effect for the sub-sample of neighboring county-pairs for whichboth counties are in the same income bucket. Columns 5 and 6 show estimates with state and county fixed effect foronly the neighboring county-pairs in the same income bucket and in the lowest tercile of the income distribution. Thesample period is from 1999 to 2005. *, **, and *** denotes significance at the 10%, 5%, and 1% cluster at the statelevel respectively.
68
Note.
Table 1.22: Heterogeneous Treatment of Bankruptcy Protection: Income and Home-
ownership. Mortgage Debt
Protection Growth st
Protection Growth s,tx Low Income
Protection Growth stx Low Home Ownership
Protection Growth s,tx Med Income
Protection Growth s,tx Med Home Ownership
Income
(1)0.018
(0.011)
-0.005(0.014)
Low Income
HomeOwnership
(2)
0.011(0.013)
(3)
0.012(0.016)
0.007(0.024)
Med Income
HomeOwnership
(4)
0.006(0.016)
(5)
0.006(0.019)
0.003(0.015)
High Income
HomeOwnership
(6)
0.012(0.010)
(7)
0.019(0.015)
-0.016(0.014)
-0.013(0.011)
-0.010(0.016)
-0.004(0.015)
-0.001(0.015)
UnenployientBate Change
-0.003 -0.003(0.003) (0.004)
House Price 0.078 0.070Index Growth (0.161) (0.141)
Income 0.189** 0.096**Growth (0.089) (0.046)
No. of ObsNo. of Clusters
State and Year FER-Squared
13,30250Y
0.11
-0.003 -0.001 -0.001(0.004) (0.003) (0.003)
0.070 0.137 0.137(0.141) (0.185) (0.185)
0.096**(0.045)
4,536 4.53650 50Y Y
0.08 0.08
0.016 0.012(0.053) (0.052)
4,422 4,42250 50Y Y
0.10 0.11
-0.007 -0.008(0.008) (0.008)
0.042 0.041(0.182) (0.182)
0.415*** 0.403***(0.138) (0.143)
4,34450Y
0.29
4,34450Y
0.31
Note. This table shows estimated coefficient a variation of specification (1) that incorporates interactions. Low/MedIncome represents counties in the lowest/middle tercile of the within state income distribution. Low/Med Ownershiprepresents counties in the lowest/middle tercile of the within income bucket distribution. Column 1 shows the result forthe whole sample when interacted with income heterogeneity. Column 2 shows the result of specification (1) restrictedto the low income counties. Column 3 shows the within low income heterogeneity in homeownership. Columns 4 to7 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.
69
Table 1.23: Heterogeneous Treatment of Bankruptcy Protection: Income and Home-ownership. Auto Debt
Income
Protection Growth s,t
Protection Growth stx Low Income
Protection Growth s,tx Low Home Ownership
Protection Growth s,tx Med Income
Protection Growth s,tx Med Home Ownership
UnemploymentRate Change
(1)0.000
(0.013)
0.027(0.017)
Low Income
HomeOwnership
(2)
0.032*(0.019)
(3)0.038*(0.021)
-0.(0.
-0.020(0.030)
Med Income
HomeOwnership
4) (5)002 -0.003016) (0.027)
High Income
HomeOwnership
(6) (7)-0.006 -0.023(0.012) (0.016)
0.006(0.023)
0.021(0.017)
0.001(0.007)
-0.005* -0.002(0.003) (0.004)
0.008(0.017)
-0.002(0.004)
House Price -0.013 -0.114 -0.112Index Growth (0.113) (0.146) (0.147)
Income 0.120*** 0.066 0.065Growth (0.031) (0.057) (0.054)
No. of ObsNo. of Clusters
State and Year FER-Squared
13,30250Y
0.19
4,53650Y
0.12
4,53650Y
0.13
-0.004(0.025)
-0.007** -0.007**(0.003) (0.003)
0.070 0.072(0.116) (0.117)
0.056* 0.059*(0.033) (0.031)
4,422 4,42250 50Y Y
0.20 0.20Note. This table shows estimated coefficient following a variation of specification (1) that
0.028**(0.012)
-0.008 -0.009*(0.005) (0.005)
0.020 0.020(0.105) (0.105)
0.209*** 0.196***(0.030) (0.030)
4,34450Y
0.34incorporates
4,34450Y
0.36interactions.
Low/Med Income represents counties in the lowest/middle tercile of the within state income distribution. Low/MedOwnership represents counties in the lowest/middle tercile of the within income bucket distribution. Column 1 showsthe 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 to2005. *, **, and *** denotes significance at the 10%, 5%, and 1% cluster at the state level respectively.
70
Tab
le
1.24
: E
ffec
t of
Ban
kru
ptc
y P
rote
ctio
n o
n C
ou
nty
Del
inq
uen
cy P
roport
ions
Cre
dit
Car
d D
ebt
Mor
tgag
e D
ebt
1 year
2year
3 years
1
year
2y
ear
3 y
ears
1
year
2y
ear
3 years
Pro
tect
ion
0.08
8 0.
020
0.07
2 -0
.009
-0
.018
-0
.002
0.
001
-0.0
21
-0.0
37G
row
th s
,t (0
.204
) (0
.058
) (0
.065
) (0
.116
) (0
.043
) (0
.045
) (0
.165
) (0
.077
) (0
.049
)
Une
mpl
oym
ent
0.02
4 -0
.018
-0
.024
0.
021
-0.0
08
0.00
1 0.
077*
**
0.01
2 0.
010
Rat
e C
hang
e (0
.034
) (0
.016
) (0
.018
) (0
.020
) (0
.014
) (0
.009
) (0
.021
) (0
.012
) (0
.009
)
Hou
se P
rice
-1
.910
**
-2.3
88
-3.4
04**
* -1
.245
* -0
.653
***
0.38
5***
0.
181
0.50
8 0.
998*
Inde
x G
row
th
(1.4
48)
(1.1
85)
(0.8
76)
(0.5
76)
(0.4
75)
(0.6
59)
(0.6
06)
(0.4
55)
(0.3
11)
Inco
me
-1.5
79**
-1
.130
-0
.630
***
-0.5
81*
-0.6
50**
* -0
.401
***
-0.3
87
-0.0
67
-0.1
32*
Gro
wth
(0
.695
) (0
.700
) (0
.180
) (0
.335
) (0
.232
) (0
.115
) (0
.255
) (0
.139
) (0
.076
)
N o
f O
bs
13,3
02
13,3
02
13,3
02
13,3
02
13,3
02
13,3
02
13,3
02
13,3
02
13,3
02N
of
Clu
ster
s 50
50
50
50
50
50
50
50
50
coun
ty a
nd y
ear
FE
Y
Y
Y
Y
Y
Y
Y
Y
Y
R-S
quar
ed
0.10
0.
17
0.22
0.
02
0.05
0.
06
0.03
0.
02
0.02
Note
. T
his
tab
le s
how
s th
e es
tim
ated
coef
fici
ent
foll
ow
ing a
var
iati
on o
f sp
ecif
icat
ion
(1)
th
at
use
s as
a d
epen
den
t var
iable
the c
han
ge
in t
he
frac
tion o
f d
elin
qu
ent
house
hold
s in
eac
h co
un
ty,
for
each
type
of
cred
it,
for
dif
fere
nt
per
iod
s:
1, 2,
an
d
3 y
ear
annual
ch
anges
. T
he
sam
ple
per
iod
is f
rom
19
99 t
o 2
005.
*,
**
, an
d
***
den
ote
s si
gn
ific
ance
at
the
10%
, 5%
, an
d
1%
clu
ster
at
the
state
lev
el r
espec
tivel
y.
-1 I,
Au
to
Deb
t
Table 1.25:
Effect of B
ankruptcy Protection on D
ebt After B
ankruptcy Reform
2005
Cred
it Card
Deb
t
No
Lin
ear C
ty Lin
ear C
ty Lin
earT
rend
T
rend
T
rend
Mortgage D
ebt
No
Lin
ear C
ty L
inear
Cty L
inear
Tren
d
Tren
d
Tren
d
Au
to Deb
t
No
Lin
ear C
ty L
inear
Cty
Lin
earT
rend
T
rend
T
rend
(1)P
rotectio
n
-0.002G
rowth s,t
(0.004)
Pro
tection
Gro
wth
s,t x Post
Unem
plo
ym
ent
Rate
Ch
ang
e
House P
rice
Ind
ex G
row
th
-0.004**(0.002)
-0.254***(0.034)
(2)
-0.006(0.006)
-0.007***(0.002)
(3)0.017**(0.008)
-0.021**(0.009)
-0.001(0.001)
-0.139*** -0.197***
(0.038) (0.025)
(1)-0.002(0.005)
-0.002(0.003)
0.065*(0.036)
(2)
0.007(0.008)
-0.006**(0.003)
(3)0.011
(0.013)
-0.011(0.014)
-0.005**(0.002)
0.146*** 0.070**
(0.046) (0.033)
(1)-0.007(0.005)
-0.007**(0.003)
0.166***(0.041)
(2)
-0.003(0.005)
(3)0.013
(0.013)
-0.022(0.014)
-0.006** -0.007***
(0.003) (0.002)
0.125(0.082)
0.161***(0.033)
Income
0.054G
rowth
(0.091)-0.174**(0.076)
0.057*(0.033)
0.455*** 0.160**
(0.087) (0.079)
0.172*(0.092)
0.420*** 0.323***
0.123***(0.063)
(0.053) (0.031)
N of O
bs 8,868
8,868 22,170
8,868 8,868
22,170 8,868
8,868 22,170
N of C
lusters 50
50 50
50 50
50 50
50 50
ety and year FE
Y
Y
Y
Y
Y
Y
Y
Y
Y
R-S
quared 0.43
0.48 0.43
0.34 0.38
0.25 0.40
0.43 0.42
Note.
Th
is table
sho
ws th
e estimated
fo
llow
ing
specificatio
n (1)
but ex
tendin
g
the sam
ple,
for each
for each ty
pe o
f credit
until
2009. C
olu
mns 1,
in each ty
pe sh
ow
s the estim
ates w
ithou
t co
unty
fixed
effect. C
olu
mns
2, sho
ws
the estim
ates w
ith fix
ed effect
and
C
olu
mns
3 sho
ws
the in
teraction
w
ith a p
ost d
um
my
eq
ual to
one for
years g
reater or eq
ual
than
2006. T
he sam
ple
perio
d
is from
1999 to
2009. *, **,
and
*** d
eno
tes sig
nifican
ce at th
e
10%,
5%, an
d 1%
clu
ster at
the sta
te lev
el resp
ectively
.
CA
Chapter 2
House Prices, Collateral andSelf-Employment
2.1 Introduction
The boom-and-bust cycle of house prices over the past decade has featured promi-
nently 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 ontwo primary explanations for the observed employment dynamics. One view is thatconsumers' use of their houses as "ATMs" drove demand and created employmentduring 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 atten-
tion 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 infor-
mation 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 facili-
tate 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. Toidentify the causal effect of higher house prices we instrument for the growth in pricesbetween 2002 and 2007 using the elasticity measure developed by Saiz (2010). Themeasure uses exogenous geographic and regulatory constraints to housing supply todifferentiate areas where an increase in housing demand translates into higher houseprices and more collateral value (areas where it is hard to build - that is, in which theelasticity of the housing supply is low) or into higher volume of houses built (areaswith high elasticity). By relying on exogenous restrictions on the expansion of housingvolumes, we can identify the effect of high collateral values on employment in smallbusinesses. This identification strategy is similar to Chaney, Sraer, and Thesmar(2012), who look at corporate investment decisions, and Mian and Sufi (2011b), whoexamine increases in consumption from household leverage.
We show that during the housing price boom of 2002-2007, areas with risinghouse prices (and increased leverage) experienced a significantly bigger increase insmall business starts and a rise in the number of people who were employed in es-tablishments with fewer than ten employees compared to areas that did not see anincrease in house prices. The same increase in employment cannot be found for largeestablishments in these same areas. In fact, the effect of home prices on job creationdecreases monotonically with firm size. This asymmetric effect on small versus largeholds only for instrumented house prices, which suggests that the non-instrumentedpart of the variation (the one that captures endogenous demand) chiefly impactsemployment at larger firms. This asymmetry points to the interpretation of the col-lateral lending channel as an important driver of employment creation particularlyfor small firms, since large firms have access to other forms of financing and should beless affected by the collateral channel. To the extent that large firms are also affectedby the increase in real estate values, our estimates may understate the effect of thecollateral channel on total employment.
Although the result above supports the importance of the collateral channel forsmall business creation, two alternative hypotheses must be ruled out as explain-ing our results. First, increases in housing prices can drive local demand for goods(Campbell and Cocco, 2007) and, consequently, employment at non-tradable indus-tries (Mian and Sufi, 2011a). To the extent that small firms may be more sensitiveto changes in demand (Kashyap and Stein, 1994), the asymmetry in the results couldreflect increased consumer demand rather than use of the collateral lending chan-nel. The second alternative hypothesis results from our use of housing and zoningrestrictions for obtaining identification, because we rely on cross-sectional differencesbetween high- and low-elasticity areas. These areas could also vary in other charac-teristics, such as the level of economic vitality. For example, not only could areaswith 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 wheremore investment opportunities become available.
We devise a number of tests to differentiate the impact of the collateral lendingchannel from these alternative hypotheses. First, we verify that the results are notdriven by changing industry composition: even within industries, areas with increas-ing 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 construc-
tion 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 col-
lateral 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 theseyears (depending on the specific assumptions about the reference group that bestisolates the collateral effect), whereas the demand channel explains about 40% overthe 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 effectis noisily estimated for 2002-2007, so we cannot reject that there is no effect onemployment of increased demand driven by higher house prices before the crisis. Thisis in stark contrast to the post-crisis period (2007-2009), when the drop in demandof over-leveraged areas shows up very strongly in the data (as documented in Mianand Sufi, 2011a). It is important to point out that these numbers provide roughapproximations of the relative magnitudes of these two channels, but they ignore anygeneral equilibrium effects in aggregation.
When we consider the period after the financial crisis when house prices startedto decline (2007-2009), we find that small firms experienced weaker employmentdeclines than large firms in areas where the increase in house prices was stronger inthe period before the crisis. This suggests that small firms that were created in low-elasticity areas during the time of increasing collateral values were more resilient thanlarger ones in those areas and did not immediately disappear when the crisis struck.This shows an interesting asymmetry in the mechanism behind the collateral lendingchannel - although it is a powerful channel in facilitating the creation of new smallestablishments, a contraction in the amount of available collateral does not lead to adisproportionate 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 ofjobs in small businesses is due to the survival of existing small businesses or a changein the entrance of newly started firms. Second, although the elasticity measure has anatural interpretation for positive housing demand shocks, we lack a good instrumentfor the house price drop. In fact, an increase in housing demand can translate intoeither higher house prices (inelastic areas) or an expansion of housing volume (elasticareas). However, on the downside, a drop in housing demand does not lead to thedestruction of housing stock, and thus prices simply drop in both inelastic and elasticareas. So, instead of instrumenting for the price drop in the crisis period, we insteadcompare areas with large appreciation in the pre-crisis period (low elasticity) withthose that had smaller house price increases - that is, the timing of the housingprice 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 moreleveraged (Mian and Sufi, 2011a, 2011b), so it should be harder for households toaccess collateral in these areas in the crisis.
Our study builds on literature that shows that credit constraints at the house-hold level matter for the creation of new businesses (Evans and Jovanovic, 1989;Holtz-Eakin, Joulfaian, and Rosen, 1994; Gentry and Hubbard, 2004; Cagetti andDe Nardi, 2006), although some authors have argued that this relation is presentonly at the very top of wealth distribution (Hurst and Lusardi, 2004). At the sametime, housing wealth in particular has been shown to be an important factor in thefunding 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, andJeffreys, 1996; and Kleiner, 2013, for the United Kingdom). Previous work has alsofound that bank credit is an important source of financing for small businesses (Pe-tersen and Rajan, 1994; Robb and Robinson, 2012; Fracassi, Garmaise, Kogan, andNatividad, 2013) and that entrepreneurs often have to provide personal guaranteeswhen they obtain financing (Berger and Udell, 1998). More recently, Greenstone andMas (2012) use the sharp reduction in credit supply following the 2008 crisis, and theheterogeneity 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 formationduring 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 Data and Empirical Methodology
2.2.1 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 bycounty, 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 bycounty 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.
2The data include only the number of establishments in each county, industry, and year bycategory of employment size (1-4 employees, 5-9, 10-19, etc.), not the total employment for eachestablishment category. As such to construct the employment in each bin we multiply the number ofestablishments by the middle point of each category. For example, to calculate the total employmentof 1-4-employee establishments in a given industry, county, and year, we multiply the number ofestablishments by 2.5.
77
The housing prices used in the regressions come from the Federal Housing FinanceAgency (FHFA) House Price Index (HPI) data at the Metropolitan Statistical Area(MSA) level. The FHFA HPI is a weighted, repeat-sales index, and it measuresaverage price changes in repeat sales or refinancings on the same properties. Weobtain this information by reviewing repeat mortgage transactions on single-familyproperties whose mortgages have been purchased or securitized by Fannie Mae orFreddie Mac since January 1975. We use data on the MSA-level index between 2002and 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 instru-ment house price growth between 2002 and 2007 with the measure of housing supplyelasticity of Saiz (2010), which varies at the MSA level. The measure of the supplyelasticity is constructed using geographical and local regulatory constraints to newconstruction. Areas where it is difficult to add new housing (due to geographic orregulatory restrictions) are classified as low elasticity and vice versa for areas whereland is easily available. Low-elasticity areas correlate strongly with steeper houseprice growth in the years 2002-2007. This measure is available for 269 MSAs that wematch to 776 counties using the correspondence between MSAs and counties for theyear 1999 as provided by the Census Bureau.3 Although employment growth and ourother controls are available for a much larger sample of counties, our regressions focuson 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 starta firm, since these investment requirements might affect how much a given industrydepends on the housing collateral channel. To construct this variable we use theSurvey of Business Owners (SBO) Public Use Microdata Sample (PUMS). The SBOPUMS was created using responses from the 2007 SBO and provides access to surveydata at a more detailed level than that of previously published SBO results. The SBOPUMS is designed to study entrepreneurial activity by surveying a random sample ofbusinesses selected from a list of all firms operating during 2007 with receipts of $1,000or more provided by the IRS. The survey provides such business characteristics asfirm 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 NAICSindustry 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 onfirms with one tO four employees only. The median amount of capital needed to starta business in the data is $215 thousand. We follow Hurst and Lusardi (2004) andsplit industries above and below the median to measure the differential effect of thecollateral channel on business creation for industries in the two groups. The averageamount of capital needed by firms below the median is $132 thousand, whereas theaverage amount needed for industries above the median is $260 thousand (detailedamounts by two-digit NAICS sector are in Appendix Table2.14).
3This correspondence is available at and for the New England Metropolitan Component Areasused by Saiz (2010).
78
Our classification of "non-tradable," "tradable," and "construction" industries atthe 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 (RetailTrade), as well as under 72 (Accommodation and Food Services). Construction in-dustries include most codes under the Construction two-digit NAICS sector (23), aswell as some subsectors in manufacturing, retail trade, and services that are directlyconnected to construction (e.g., 3273 - Cement and Concrete Products Manufactur-ing). Manufacturing industries include all 31-33 subsectors (Manufacturing), and insome specifications we restrict the sample to manufacturing industries that are alsoclassified 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 theirgoods similar to that used in Duranton, Morrow, and Turner (2013). These data areavailable from the 2007 Census Commodity Flow Survey, which reports the distancetraveled by shipments of a sample of establishments in each three-digit NAICS man-ufacturing industry.5 The unit of observation in the census data is at the state andindustry levels, so we construct a dollar-weighted average distance of shipments alsofor each state and industry individually. Summary statistics of the average distanceshipped, as well as how often each industry appears in each decile, are shown inAppendix Table 2.13.
We also use data on county-level births and deaths of establishments for eachtwo-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 underthe "Employment Change" section of SUSB, and it does not include a breakdownby establishment size at the county and industry levels, so we cannot use it as ourmain dataset. However, given that most establishment births are of a very small scale(Haltiwanger, Jarmin, and Miranda, 2011), we view the regressions performed on thisdata set as an important test of the mechanism in our main results. We computethe cumulative number of births and deaths between 2002 and 2007 for each countyand industry as our dependent variable of interest and scale this number by the totalnumber of establishments as of 2002 in the same county-industry cell.
The net creation of sole proprietorships at a county level is obtained from twosources. We use both the yearly local area personal income and employment datafrom the Bureau of Economic Analysis (BEA and the census nonemployer statistics.From the BEA we use Non-Farm Proprietorship employment at the county levelbetween 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 atthe two-digit NAICS level. We use both sources of data in the regressions to ensurethe robustness of our results.
Unemployment and unemployment rate at the county level are obtained using
4The 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 UnemploymentStatistics (LAUS) are available for approximately 7,300 areas that range from censusregions and divisions to counties and county equivalents, and these data are availablebetween 1976 and 2012. We match the county equivalent data to the CBP data usingFederal Information Processing Standard (FIPS) county unique identifiers.
The migrations data are extracted from the IRS county-to-county migration dataseries. The migration estimates are based on year-to-year address changes reportedon individual income tax returns filed with the IRS. The data set presents migrationpatterns by county for the entire United States and is split by inflows - the number ofnew 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 netflows as inflows minus outflows, and we scale all figures by the number of nonmoversin 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 setof controls that capture some of the cross-sectional differences across counties. Weuse county-level information from the Census Bureau Summary Files for 2000 on: thenumber of households in a county; the natural logarithm of county-level population;the percentage of college-educated individuals, defined as the number of people over25 with a bachelor degree or higher as a proportion of the total population over 25years old; the percentage of employed people, defined as the employed population overthe 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 bythe 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 showstotal employment in 2002 for all counties in our sample, as well as the employmentgrowth between 2002 and 2007 estimated from the CBP data. Our data include atotal of 775 counties with nonmissing total employment data. We split the sampleinto counties above and below the median of the housing supply elasticity measureand show t-statistics (with standard errors clustered by MSA) for the difference inmeans between the two groups. We see that counties with low supply elasticityare larger but have similar unemployment rates in 2002 as those with high supplyelasticity. The characteristics in 2002 from the census are broadly similar for the
6The data used to produce migration data products come from individual income tax returnsfiled before late September of each calendar year and represent between 95% and 98% of total annualfilings.
7We construct the measure of competition from imports from China by multiplying the fractionof employment in each county and in each industry by the share of imported goods from China as afraction of total domestic shipments in the industry in the United States. The variation is virtuallythe same if we instead use the growth in the weight of imports for each industry as a fraction ofU.S. domestic shipments between 1998 and 2005. The import data at the industry level is obtainedfrom 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, ac-
counts for 51.7% of employment in 2002, whereas the smallest category, 1-4 employ-
ees, 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 differ-
ential 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 col-
lateral 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 establish-
ments 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 elas-
ticity 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 rea-
sons 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
8As 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 bustcycle of the 2000s. In our setting we are concerned only with price increases between2002 and 2007, and the supply elasticity measure developed by Saiz is a strong pre-dictor of the increase in prices (i.e., there is no weak instruments problem). As wedescribe below, we also include specifications that include county fixed effects thatshould further mitigate concerns about the cross-sectional elasticity measure.
We rely on two basic regression specifications for our analysis. The first specifi-cation aggregates data up to the level at which our instrument varies - that is, atthe county-year establishment-size level. Each individual observation is the changebetween 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 cat-egory and take the growth in total number of employees as the dependent variable.We then run two-stage least squares regressions of the type:
A02-0 7 Employmentij = o + 0 1 ,AHpF2-O7 + 02 1i + 3 1jAHP0 2- 07 + -lXj + 6Ej
We index counties by j and establishment size categories by i. A 02-0 7 Employmentijis the change in employment for establishment size category i in county j between2002 and 2007. Similarly, AHP02- 0 7 is the growth in housing prices at the countylevel for the same time period where, as we discuss above, we instrument for thegrowth in house prices using the housing supply elasticity of Saiz (2010). 1i is a set ofdummy variables for each of the four included establishment categories (we omit thelargest category of more than 50 employees). We then also include the product of theestablishment size dummies and the growth in house prices, and 33 is the coefficient ofinterest in our regressions. In particular, the test we are interested in is whether thecoefficient for the smallest establishments is larger (and positive) than those of thelarger categories, which would confirm that house prices had a stronger impact on thecreation of small establishments. Xj is a set of county-level controls that include thesize 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 populationin the labor force, the percentage of owner-occupied houses, and the county shareof China imports. Standard errors in this specification are heteroskedasticity robustand clustered at the MSA level (given that the variation in the instrument we use isat this level as well), and all regressions are weighted by the number of households ina county as of 2000, as in Mian and Sufi (2011a).
The second specification disaggregates observations to the county, year, establish-ment size, and four-digit NAICS level, yielding a much larger number of observationsthan the specification above (since each county now appears multiple times for eachindustry). When using these disaggregated data we can include industry fixed effectsin 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 caserepresent the differential impact that house prices have on establishments of differentsizes within each industry. The specification becomes:
82
A 02- 7 Employentijz = a + O1AHP 2-0 7 + /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 hold-
ing constant industry fixed effects we identify how employment in the smallest estab-
lishments 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 busi-
nesses. 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 conven-
tional mortgage rate (obtained from the Federal Reserve data website) with the local
elasticity of housing supply measure. This provides time-varying shocks to the de-
mand 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 obser-
vations 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 + Y21 t + Eij
The instrument for house prices is the product of mortgage rates and housing elas-
ticity, 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, aswell 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 casethrough increased real estate prices) in the period before the financial crisis has aneffect on the creation of small firms or on self-employment, since it provides individualswith easier access to start-up capital. As a result, we should see a sharper increasein self-employment and employment in small businesses in areas that had steeperhousing price appreciation. We also expect this effect to be concentrated in firms inthe smaller size categories, since large firms cannot finance themselves using homeequity. This hypothesis is tested in Table 2.2, where we run two-stage least squaresregressions of the growth in employment between 2002 and 2007 on five establishmentsize categories and their interaction with house price growth in the same period. Theinstrument for house price growth, as we discuss above, is the Saiz (2010) measureof housing supply elasticity. In the first column of Table 2.2 we show the first-stageregression of house price growth on the Saiz measure of housing supply elasticity toconfirm the validity of the instrument. The coefficient of -0.09 means that a onestandard deviation increase in elasticity of housing supply is associated with an 11.7percentage 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 forevaluating weak instruments). This reflects that MSAs with a higher elasticity ofsupply experienced significantly lower housing price growth between 2002 and 2007,in line with previous literature. In Column 2 we run a regression of employmentchange 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 showthe raw correlation between house prices and employment. The effect is positive andeconomically large. A one standard deviation increase in house prices is associatedwith an increase in total employment of 3.95% over this period, for an average growthin employment of 10.6%. In the simple weighted least squares regression we see nodistinction between the effect of home prices on small and large establishments. Thisresult highlights the need for an instrument for our dependent variable of interest,given the numerous factors that are likely to drive both employment creation andhouse prices (income growth, investment opportunities, etc.).
In Column 3 of Table 2.2 we repeat the same regression but instrument the changein house prices with the Saiz measure for the elasticity of housing supply. We seethat there is a positive but not significant causal relation between county-level em-ployment change and house price growth on average, in contrast to the results in the
9We do not rely on the panel specifications for most tests because mortgage rates did notexperience large drops in the period we analyze. We effectively have one large shock to demand forhousing 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 instrumentedhousing price changes, the increase in home prices has a significant and large positiveeffect on the small establishments but no significant effect on employment growth forbig establishments (50 or more employees). The coefficient on the interaction termbetween house price growth and the one-to-four-employee size category shows that a1% increase in house prices translates into a 0.19% increase in employment at theseestablishments relative to the largest ones. This translates into an increase in em-ployment of 5.3% for a one standard deviation change in house prices, for an averagechange in employment at the smallest establishments of 9.4% (the effects of a onestandard deviation change in house prices for each size category are shown in Ap-pendix Table 2.12). Furthermore, the effect of collateral is decreasing monotonicallywith firm size. For firms with more than ten employees, the effect is indistinguishablefrom that of the very largest firms. This is consistent with the collateral channel ofhouse price appreciation being an important mechanism for small firm creation, sincethe amount of collateral that is provided by real estate appreciation is not enough tostart a larger firm. Also, these results suggest that the causal impact of house priceson employment growth in 2002-2007 did not work through increased demand, sincein 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 pricesin areas with low Saiz housing elasticity induces a local demand shock that especiallyaffects certain industries. If those industries are also, on average, disproportionatelymade up of smaller establishments, the result above might reflect a composition effectrather than the collateral channel, as we suggest. Although a number of factors wouldneed to line up in a very specific way, we cannot rule this concern out on face value withthe specifications in Table 2.2. In order to eliminate the alternative hypothesis aboutindustry composition, we use our more disaggregated data, which provides data at thecounty, four-digit NAICS, and establishment size level. This allows us to hold industryfixed effects constant and test whether, conditional on an industry, the growth of smallestablishments is significantly stronger than that of large establishments in countieswith greater increases in home prices. Intuitively, this specification asks whetherwithin an industry the fraction of employment generated by small firms grows morequickly than that of large firms. This way we can confirm that the results are not aconsequence of changing industry composition. The results for this specification areshown 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 fouremployees when compared to the bigger firm categories. We again find that the effectis monotonically decreasing and not statistically significant beyond firms with ten ormore employees. To be more specific about which industries show the strongest effectsfrom the collateral channel, in Table 2.17 we show the three-digit NAICS industriesthat are not construction, manufacturing, non-tradable, and finance, insurance, andreal estate, as well as the employment share in each size bin. The sample includes avariety of services and wholesale activities, with significant cross-sectional variationin the proportion of employees in the very small establishment size categories (from26.3% of employment in one-to-four-employee establishments in the case of "NAICS425 - 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 time-
varying 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 $309thousand 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. Whenwe include industry fixed effects, only the coefficient on the smallest establishments isstatistically 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 2002-
2007 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 thistest, we split industries based on the average distance of shipments in each three-digitNAICS industry and state. This addresses further the concern that local demandshocks might be driving the results for manufacturing firms. Table 4 we show thatthe result for manufacturing shown in Table 2.3 is driven by firms in industries andstates that ship goods across large distances. The median reported distance in thesample is 600 miles, so firms that report shipping goods over more than 600 milesare unlikely to make decisions as a function of local demand shocks (details on thedistances 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 givensector could be very different from large firms, so the small firms in those industriescould depend more on local demand. Although we do not have shipment data byfirm, in Table A7 we consider the relation between the reported distance shippedin 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 shareof employment in establishments that have 50 or more employees for each state andthree-digit NAICS manufacturing industry. Then, for each industry, we computethe average (over all states) of the distance shipped, as well as the average shareof employees in firms that have 50 or more employees. Finally, for each state andindustry observation, we compute the deviation from the industry mean for bothmeasures and classify observations into deciles based on these deviations. 0 Thetakeaway from this table is that there is no visible relation between the distanceshipped 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 thedistance shipped. This should mitigate the concern that a strong positive relationbetween firm size and distance shipped might explain the results in the last twocolumns of Table 2.4.
Our measure of growth of establishments by size category does not allow us toobserve the creation and destruction of establishments directly, so in a separate setof regressions shown in Table A8 we use the Statistics of U.S. Businesses from thecensus to look at births and deaths of establishments at the two-digit NAICS industrylevel. The disadvantage of this data set is that it does not include the breakdownof establishments by employment size. Given that an overwhelming percentage ofnew businesses are very small (Haltiwanger, Jarmin, and Miranda, 2011; Robb andRobinson, 2012), this robustness test directly speaks to the validity of our mainresults. We find that births of establishments are very strongly affected by increasinghouse prices instrumented with the elasticity of housing supply and that the resultholds 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 (thefinest industry category available in this data set at a county level).
10So, state-industry observations that are in the first decile of the distance are those that shipgoods at short distances relative to the industry average. Similarly, those in the first decile of theshare of employment at large firms, are state-industry observations that have few employees in largefirms 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 channelis to compare the magnitude of the employment gains that can be attributed tothis channel to those that can be assigned to the demand channel shown in Mianand Sufi (2011a). To do so, we follow the same calculation used in that paper toaggregate the effect across all counties. The authors compute the effect of debt-to-income (DTI) ratios as of the beginning of the crisis on the employment changebetween 2007 and 2009 in non-tradable industries." These are the industries that aremost 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 non-tradable industries and then extrapolating this effect to the rest of the economy.
We perform essentially the same calculations for the period 2002-2007 to establisha benchmark employment effect that can be attributed to the demand channel. Westart by obtaining the effect of a change in house prices on employment in the non-tradable industries at a county level for the 2002-2007 period. That regression isshown in Table 2.5 in Column 3. If we aggregate in the same way as describedabove (where the baseline employment is now as of 2002), we obtain an increase inemployment in the non-tradable sector of 451.8 thousand jobs, which, given a shareof employment in this sector of 18.4% in 2002, translates into a predicted total jobgain due to increased aggregate demand of 2.452 million jobs. This is about 40% ofthe jobs created in the private sector in the 660 counties used for the calculation. Theconfidence interval for this estimation is very large and includes zero, which opensthe possibility that the aggregate demand effect for the period before the crisis mayactually be quite small. This is in sharp contrast to the estimates obtained by Mianand Sufi (2011a) for the years after the crisis, where the same regression yields verystrong effects for the drop in demand on non-tradable employment.
We now turn to the calculation of the magnitude of the collateral channel overthe same period. We rely on the differential impact of house prices on employmentcreation at small firms relative to firms with 50 or more employees, and we focus onthe specifications in which we exclude non-tradable industries and construction (Table2.3, Column 2). We again first compute predicted county-level employment gains forthese industries (relative to the 10th percentile county) and then we aggregate to allcounties. When we do that, we obtain an estimated total job gain in firms with fewerthan 50 employees relative to those with 50 or more employees of 1.698 million jobsin all counties, or 27.8% of jobs created between 2002 and 2007 in this period. Ifwe restrict our attention to the specification where the demand explanation for ourresults 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 ofTable 2.4), the same computation would yield an estimate of 676 thousand jobs, orabout 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 2007as the independent variable (as we do in this paper) yields virtually the same results, as countieswith high debt-to-income by the end of this period are also the ones that experienced large increasesin 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 firmsover time, so if a firm grows to become very large, we do not attribute the employmentcreation of that firm to our effect (it would be in the 50+ category that we use as ourbaseline). Second, we are focusing on the importance of this channel for very smallfirms. This ignores the role that collateral value plays for larger firms, as pointedout 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 propercalibration of the importance of the collateral effect for the whole economy. In extend-ing the effect that we observe for a subset of firms and industries in individual countiesto the whole economy, we ignore general equilibrium effects that could potentially beimportant.
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 effectof housing price growth on net creation of proprietorships relative to all the establish-ment categories listed in the previous tables using the Saiz measure to instrument forexogenous movements in housing price changes. The first column in this table usesemployment data on sole proprietorships from the BEA, while the last three columnsrely on census data on nonemployer establishments (which includes information onthe two-digit NAICS sector in which the establishment operates). The coefficient onhousing price growth in Column 1 interacted with the sole proprietorship categoryis significantly different from that on the largest establishments and close in mag-nitude to that on the 1-4-employee category. In Column 2 we use census data andfind a smaller coefficient on the sole proprietorships, and we cannot distinguish thatcoefficient from the others in the regression.
In the last two columns we again split the sample by the amount of capital neededto start a business in a given industry, as discussed above. We find that the effectof home prices on the net creation of sole proprietorships is stronger in industrieswith low start-up capital needs, which is in line with our findings for the other sizecategories. Note, however, that the difference between the coefficients in the twospecifications (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 conse-quence of the increasing value of collateral during the rise in house prices is whetherthese establishments were then eliminated after the housing bubble burst. In this sec-tion we try to distinguish whether these newly created businesses were particularlyfragile and were disproportionately affected by the crisis or, alternatively, whetherthey behaved like the rest of the firms in the economy.
90
Our data do not allow us to track individual establishments, so we cannot knowwhether 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 todownsize or disappear in the crisis. That is, we can assess whether employment losswas stronger at larger or smaller firms during the crisis in counties where the increasein house prices had been stronger in the precrisis period (which are also the mostleveraged counties, as shown in Mian and Sufi, 2011a). We run those regressions inTable 2.7.
The results show that employment loss was either similar across large and smallestablishments or, if anything, was worse at large firms (in the specifications withoutindustry 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 consistentwith 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 migrationof 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 andOutflows separately, on house prices changes instrumented with the Saiz measure andthe 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 outflowsthat were offset in part by somewhat bigger inflows of people at the same time. Thisalleviates 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. Ofcourse, we cannot observe who is entering and who is migrating out of each county, sowe cannot address the more detailed question of whether entrepreneurs were movingin as other individuals were moving out, but the aggregate trends suggest strongeroutflows 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-elasticityareas during our sample period for reasons unrelated to collateral availability, and if
this drove the creation of new businesses, this would violate the exclusion restrictionfor 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 pricesboomed (Chakraborty, Goldstein, and MacKinlay, 2013), making it unlikely thatsmall 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 mort-gage applications from HMDA. The underlying assumption is that the cross-sectionalvariation on the looseness of credit conditions should be positively correlated withthe same variation for mortgage credit, especially given that the reason why creditmight have become laxer is the fact that house prices increased.
We consider the number of applications that are denied by financial institutionsas a proportion of the total loan applications in a county and in a year." Usingthe yearly estimates we compute the proportional change in denial rates between2002 and 2007. We focus on loans used for purchasing homes because they are lesssensitive to the issue of relationship lending and/or private lender information aboutthe borrower and therefore should better reflect the loosening of credit conditions.
Panel A of Table 2.9 shows that credit conditions tightened rather than loosenedin low-elasticity areas (those below median elasticity in the sample) when we use thismeasure of credit supply. Denial rates increased by about 2% in counties with lowelasticity of housing supply, whereas they go down in high-elasticity areas by 1% - thatis, credit loosened in those areas. The difference between the two types of countiesis statistically significant at the 1% level. In addition, total volume of applicationsdecreases by 1% in low-elasticity areas in comparison to the 10% increase in thehigh-elasticity areas.
We formally test these differences in a regression framework using a continuouselasticity 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 isassociated with higher denial rates of loan applications, and these results are robustto different specification and controls. Although the regressions condition on theapplicant pool (and so the denial ratc could mask riskier borrowers applying forloans), we control for the debt-to-income in these regressions to account for changesin applicant types.
Overall, this result allows us to rule out the concern that our instrument is pickingup changes in the way that lenders granted credit instead of access to credit throughan increase in collateral values.
2.4 Conclusion
Overall, the evidence we present identifies the causal effect of rising house prices inthe creation of new small firms. Increased access collateral allowed individuals tostart small businesses or to become self-employed. We conjecture that without accessto this collateral in the form of real estate assets, many individuals would not have
1 2 Volume of applications is calculated as the sum of all loans that are originated plus applicationsthat are approved but not accepted, applications denied by the financial institution, and loanspurchased 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 inter-
esting asymmetry of these effects. Collateral was particularly important in explaining
job creation when more collateral became available, but we observe no significant de-
struction 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 fluctu-
ations. 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.
1 3 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. Journalof 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 havenegative real effects? Unpublished working paper.
Chaney, T., Sraer, D., Thesmar, D., 2012. The collateral channel: how real estateshocks affect corporate investment. American Economic Review 102, 2381-2409.
Charles, K. K., Hurst, E., Notowidigdo, M. J., 2012. Manufacturing busts, housingbooms, and declining employment: a structural explanation. Unpublished workingpaper.
Corradin, S., Popov, A., 2013. House prices, home equity and entrepreneurships.ECB Working Paper.
Cvijanovic, D., 2013. Real estate prices and firm capital structure. Unpublishedworking paper.
Davidoff, T., 2012. Supply elasticity and the housing cycle of the 2000s. RealEstate Economics, forthcoming.
Duranton, G., Morrow, P., Turner, M. A., 2013. Roads and trade: evidence fromthe US. CEPR Discussion Papers 9393.
Evans, D. S., Jovanovic, B., 1989. An estimated model of entrepreneurial choiceunder 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 entrepreneurialactivity. Journal of Law and Economics 46.
Fort, T., Haltiwanger, J. C., Jarmin, R. S., Miranda, J., 2012. How firms respondto 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 doescredit matter for entrepreneurial success in the United States? Unpublished workingpaper.
Gentry, W. M., Hubbard, R. G., 2005. Success taxes, entrepreneurial entry, andinnovation. In: Innovation Policy and the Economy, Volume 5. MIT Press, Cam-bridge, pp. 87-108.
Gertler, M., Gilchrist, S., 1994. Monetary policy, business cycles, and the behaviorof 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? Smallvs. large vs. young. Unpublished working paper.
Holtz-Eakin, D., Joulfaian, D., Rosen, H. S., 1994. Entrepreneurial decisions andliquidity constraints. NBER Working Paper No. 4526.
Hurst, E., Lusardi, A., 2004. Liquidity constraints, household wealth, and en-trepreneurship. Journal of Political Economy 112, 201-248.
94
Iacoviello, M., 2005. House prices, borrowing constraints, and monetary policy inthe 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 the-
ory 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: evi-
dence 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 entrepreneur-
ship. Unpublished working paper.
95
Table 2.1: Summary Statistics
Panel A
All Counties High Elasticity Low Elasticity
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
1-4 Emp 5-9 Emp
Emp. in All SectorsTotal
Growth (02-07)Percentage of Total
Emp. in Firms <P50 of Start-Up CapitalTotal
Growth (02-07)Percentage of Total
Emp. in Firms >P50 of Start-Up CapitalTotal
Growth (02-07)Percentage of Total
9,1019.48.9
6,23510.812.1
2.8666.95.8
10-19 Emp 20-49 Emp 50+ Emp
9,122 12,819 21,466 72,9398.0 12.5 10.6 13.39.0 12.1 18.3 51.7
5.580 7,365 11,033 39,96411.0 13.4 14.0 24.610.8 12.8 16.6 47.7
3,542 5,454 10,433 32,9754.4 13.1 9.6 9.37.4 11.7 20.5 54.6
Note. Panel A reports summary statistics for all counties in the sample in Column 1, and Columns 2 and 3 showthe summary statistics for counties above and below the median elasticity of housing supply in the sample. Foreach variable we show the pooled average, median (italicized) and standard deviation (in parenthesis). The lastcolumn shows the t-statistic for the difference in means of the two groups, adjusted for clustering at the MetropolitanStatistical Area level. Total Employment refers to the total number of employees in a county in thousands across allestablishment sizes and industries using the County Business Patterns data as of 2002. Unemployment Rate is shownin percentage and comes from the Bureau of Labor Statistics Local Area statistics in 2002. Percent College Educatedis the percentage of the population with a college degree, Percent Employed is the percentage of the labor force thatis employed, Workforce as a Percentage of Population is the share of the population in the workforce, and Percentof Homes Owner-occupied is the percentage of homes that are owner-occupied (i.e., not rental properties). AverageHousehold Income is the total income in a county divided by the number of households as of 2002 and Growth inIncome is the percentage change in income in a county between 2002 and 2007. Change in DTI is the percentagechange in debt to income ratio in the same period. The debt to income ratio is estimated using county level householddebt data from the New York Fed-Equifax and income is computed using IRS county-level information. Growth inHouse Prices is the percentage change in house prices between 2002 and 2007 at the MSA level from the FederalHousing Finance Agency. Panel B shows the Total Employment in 2002 in thousands, Employment Growth between2002 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
113,91845,454
(238,831)5.45.3
(1.5)100.246.2
(188.1)10.68.2
(15.8)51.842.6
(36.4)27.623.9
(21.1)33.926.8
(21.1)-0.9-0.8(1.0)
775
69,05733,228
(129,569)5.35.2
(1.5)59.334.2
(92.6)10.27.5
(16.9)36.634.9
(23.0)27.223.0
(24.2)23.519.4
(14.3)-0.7-0.5(0.9)
382
157,52363,286
(304,041)5.45.4
(1.4)139.866.4
(241.4)11.08.9
(14.5)66.358.3
(40.7)28.024.5
(17.6)43.740.9
(21.9)-1.0-1.0(1.0)
393
Panel B
Tab
le 2
.2:
Em
ploy
men
t G
row
th,
Firm
Siz
e, a
nd H
ouse
Pri
ce A
ppre
ciat
ion
Hou
sing
Su
pp
ly E
last
icit
y
Fir
st S
tage
A
ll In
dust
ries
(WL
S)
(1)
(2)
-0.0
9***
(0.0
2)
All
Indu
stri
es(I
V)
(3)
(4)
Sta
rt-u
p C
apit
al<
P50
(5)
(6)
(7)
Sta
rt-u
p C
apit
al
>
P50
(8)
(9)
(10)
Gro
wth
in
Hou
se P
rice
s
Gro
wth
in
Hou
se
Pri
ces
* 1-
4 E
mpl
oyee
s
Gro
wth
in
Hou
se
Pri
ces
* 5-
9 E
mpl
oyee
s
Gro
wth
in
Hou
se P
rice
s 10
-19
Em
ploy
ees
Gro
wth
in
Hou
se P
rice
s *
20-4
9 E
mpl
oyee
s
Log
of
the
Po
pu
lati
on
Per
cen
t C
olle
ge E
du
cate
d
Perc
ent
Em
ploy
ed (
2000
Cen
sus)
Wor
kfor
ce a
s a
Per
centa
ge
of P
op
ula
tio
n
Perc
ent
of H
omes
Ow
ner-
occu
pied
Chin
a Im
port
Sh
are
in C
ounty
(20
05)
4-D
igit
Indust
ry
Fix
ed E
ffec
tsC
ounty
F
ixed
Eff
ects
Nu
mb
er o
f O
bse
rvat
ions R2
--I
0.19
***
0.05
-0
.06
0.02
-0
.01
-0.0
4 -0
.11*
**
0.06
-0
.07
0.10
**(0
.04)
(0
.06)
(0
.10)
(0
.03)
(0
.07)
(0
.13)
(0
.04)
(0
.07)
(0
.10)
(0
.04)
0.03
0.
20**
* 0.
26**
0.
16**
* 0.
33**
* 0.
32**
0.3
1***
0.
14**
0.
18**
0.
10*
(0.0
3)
(0.0
5)
(0.0
9)
(0.0
5)
(0.0
7)
(0.1
2)
(0.0
6)
(0.0
6)
(0.0
9)
(0.0
6)
-0.0
2 0.
08**
0.
17
0.00
0.
19**
* 0.
14
0.17*
**
0.04
0.
19**
-0
.10*
(0.0
3)
(0.0
4)
(0.1
0)
(0.0
4)
(0.0
5)
(0.1
5)
(0.0
5)
(0.0
6)
(0.0
8)
(0.0
5)
-0.0
2 0.
01
0.06
-0
.05
0.14
***
0.02
(.1
0*
-0.0
7 0.0
9 -0
.12*
*(0
.02)
(0
.04)
(0
.09)
(0
.04)
(0
.05)
(0
.12)
(0
.05)
(0
.06)
(0
.09)
(0
.05)
0.01
0.00
0.
07
-0.0
7**
0.13*
**
0.10
0.
06
-0.0
7 0.
02
-0.1
4***
(0.0
2)
(0.0
4)
(0.0
7)
(0.0
3)
(0.0
5)
(0.1
0)
(0.0
5)
(0.0
5)
(0.0
8)
(0.0
4)
0.00
-0
.02*
**
-0.0
2***
-0
.04*
**(0
.03)
(0
.01)
(0
.01)
(0
.01)
0.00
(0.0
0)0.0
0**
0.00
**
0.00
(0.00
) ((1
.)))
(1.1
)
-o.0
1***
0.
00
0.00
0.
00(0
.00)
(0
.00)
(0
.00)
(0
.00)
-0.6
9(0
.63)
0.00
((1.
00)
0.10
(0.9
1)
-1.0
9***
-1
.11*
**
-0.8
6***
(0.1
9)
(0.1
9)
(0.2
2)
0.00
**
0.00
**
0.00
(1.1
)0)
(1.1
))
((.1
)
0.09
0.
12
-0.0
8(0
.23)
(0
.23)
(0
.32)
--
YY
-0.0
3***
-0
.05*
**(0
.01)
((1
.01)
0.00
0.
00(0
.)))
(0
.00)
0.00
0.
00(0
.00)
(0
.00)
-1.1
6***
-1
.00*
**(0
.20)
(0
.25)
0.00*
* 0.
00(0
.0)
(1.1))
0.33
0.08
(0.2
6)
(0.3
8) YY
-0.0
2***
-0
.04*
**(0
.01)
(0
.01)
0.00*
* 0.
00(0
.00)
(0
.00)
0.00
**
0.00
(0.0
0)
(0.0
0)
-1.0
8***
-0
.72*
**(0
.20)
(0
.21)
0.00
0.
00*
(11.11
1) (11
.1)
-0.0
1 -0
.19
(0.2
2)
(0.3
0)
-Y
Y73
1 3,
653
3.65
3 37
3.57
6 21
,962
3,
653
196,
027
21.9
54
3,65
1 17
7,54
9 21
,949
0.30
0.
27
0.22
0.
30
0.02
0.
21
0.39
0.
00
0.14
0.
10
0.03
The
table
sh
ow
s tw
o-s
tag
e le
ast
squ
ares
re
gre
ssio
ns
of
emplo
ym
ent
gro
wth
on
house
pri
ce g
row
th
inst
rum
ente
d
wit
h t
he
elas
tici
ty o
f housi
ng
sup
ply
, in
dic
ato
r var
iable
s fo
r ea
ch e
stab
lish
men
t si
ze
(not
show
n i
n th
e ta
ble
) an
d in
tera
ctio
ns
of
house
pri
ce g
row
th
wit
h t
he
size
of
esta
bli
shm
ents
. A
ll re
gre
ssio
ns
are
wei
ghte
d b
y th
e
num
ber
of
house
hold
s in
a c
ounty
as
of
2000
. E
mplo
ym
ent
gro
wth
is
the
per
cen
tag
e ch
ange
in
emp
loy
men
t b
etw
een
200
2 an
d
2007
es
tim
ated
usi
ng
Co
un
ty
Busi
nes
s P
att
ern
s (C
BP
) data
. G
row
th i
n H
ouse
pri
ces
is t
he
per
centa
ge
chan
ge
bet
wee
n
2002
an
d
2007
, an
d
each
inte
ract
ion i
s w
ith a
du
mm
y
indic
ator
for
the s
ize
of t
he
esta
bli
shm
ent.
C
olu
mn
1 sh
ow
s th
e fi
rst
stag
e re
gre
ssio
n
of
the
chan
ge
in h
ouse
pri
ces
bet
wee
n
2002
an
d 2
007
on t
he S
aiz
elas
tici
ty
mea
sure
.C
olu
mns
2 th
rough
5 "A
ll In
dust
ries
" sh
ow
s th
e re
sult
s fo
r th
e w
hole
sa
mp
le
of
firm
s,
firs
t th
e
wei
ghte
d l
east
sq
uar
es
resu
lts,
th
en t
he
IV
at
a co
unty
le
vel,
th
e
IV r
esu
lts
at
a co
unty
an
d in
dust
ryle
vel
and th
en
th
e IV
re
sult
s usi
ng y
earl
y o
bserv
ati
on
s
an
d th
e in
tera
cti
on
of
the ela
sti
cit
y m
easu
re w
ith th
e co
nv
en
tio
nal
mort
gage
rate
s as
the
instr
um
en
t.
Colu
mns
6 th
rough
11
sho
w th
e co
eff
icie
nts
spli
t b
y t
he sta
rt-u
p
capit
al
am
ou
nt
(above an
d
belo
w t
he
media
n)
als
o at
the co
un
ty,
at
the c
ou
nty
an
d in
du
str
y
lev
el,
and at
the c
ounty
le
vel
wit
h yearl
y o
bserv
ati
on
s.
The
om
itte
d cate
gory
re
fers
to
esta
bli
shm
ents
w
ith 5
0 or
more
em
plo
yee
s.
All
reg
ress
ions
con
tro
l fo
r th
e n
atu
ral
logar
ithm
of
po
pu
lati
on
, th
e per
centa
ge
of
the p
op
ula
tion
wit
h
a co
lleg
e deg
ree,
th
e per
centa
ge
of
the
lab
or
forc
e th
at
is e
mplo
yed
, th
e s
har
e of
the
popula
tion i
n th
e w
ork
forc
e,
and
th
e per
centa
ge
of
ho
mes
that
are
ow
ner
occ
upie
d.
Contr
ols
are
at
a
cou
nty
le
vel
for
th
e yea
r 20
00
and a
re o
bta
ined
usi
ng
Cen
sus
Bure
auD
ata
Sum
mar
y
Fil
es.
Sta
ndar
d
erro
rs a
re
in p
aren
thes
is
and
are
cl
ust
ered
by
MSA
. *,
**
, **
* in
dic
ate
stati
stic
al
signif
ican
ce a
t 10
, 5,
an
d
1%
level
s,
resp
ecti
vel
y.
(11)
Table 2.3:
Em
ployment
Grow
th and House P
rices: E
xcluding Construction,
Non-T
radable, and F
inance Industries and Con-
sidering Manufacturing
Only
Dro
p
Dro
p C
onst.
Dro
p C
onst.,
Man
ufa
ctu
ring
M
an
ufa
ctu
ring
(T
rad
able
)C
onstru
ction
and Non-T
rad. N
on-Trad.
and
F.I.R
.E.
(Trad
able)
Grow
th in House Prices
-0.09 -0.12
-0.14 -0.17
-0.16(0.10)
(0.10) (0.10)
(0.11) (0.12)
Gro
wth
in House P
rices * 1-4 E
mployees
0.27*** 0.32***
0.35*** 0.13*
0.15*(0.09)
(0.09) (0.10)
(0.07) (0.09)
Grow
th in House Prices * 5-9 E
mployees
0.19* 0.21*
0.24** 0.12
0.10(0.10)
(0.11) (0.11)
(0.08) (0.09)
Grow
th in House Prices
10-19 Em
ployees 0.08
0.12 0.12
0.11 0.16
(0.09) (0.09)
(0.09) (0.11)
(0.11)
Grow
th in House Prices *
20-49 Em
ployees 0.08
0.12* 0.11*
0.01 -0.05
(0.06) (0.06)
(0.06) (0.12)
(0.09)
Log of the Population -0.04***
-0.04*** -0.04***
-0.02** -0.02*
(0.01) (0.01)
(0.01) (0.01)
(0.01)
Percent College E
ducated 0.00
0.00 0.00
0.00 0.00
(0.00) (0.00)
(0.00) (0.00)
(0.00)
Percent Em
ployed (2000 Census)
0.00 0.00
0.00 0.00
0.00(0.00)
(0.00) (1.00)
(0.00) (0.00)
Workforce as
a Percen
tage of P
opulatio
n
-0.88*** -0.84***
-0.84*** -0.64**
-0.66**(0.22)
(0.23) (0.24)
(0.29) (0.30)
Percent of Hom
es Ow
ner-occupied 0.00
0.00 0.00*
0.00* 0.00
(0.00) (0.00)
(0.00) (0.00)
(0.00)
Chin
a Import S
hare in Co
un
ty (2005)
-0.11 -0.23
-0.28 -0.88*
-1.24**(0.34)
(0.36) (0.36)
(0.50) (0.56)
Contro
ls Y
Y
Y
Y
Y
4-Digit Industry Fixed E
ffects Y
Y
Y
Y
Y
Num
ber of O
bservations 325,349
264,901 242,510
55,345 44,649
R2
0.29 0.30
0.31 0.02
0.02G
rowth H
P * 1-4 E. = G
rowth H
P * 5-9 E. 0.04**
0.02** 0.02**
0.95 0.48
Gro
wth
H
P *
1-4 E. =
Gro
wth
H
P
* 10-19 E
. 0.00***
).00*** 0.00***
0.85 0.91
Grow
th H
P * 1-4 E. =
Grow
th HP * 20-49 E.
0.00*** 0.00***
0.00*** 0.33
0.10*
The tab
le sh
ow
s two
stage
least sq
uares reg
ressions
of em
plo
ym
ent g
row
th
on
house
price g
row
th
instru
men
ted
with
the elasticity
of h
ousin
g su
pply
, ind
icator
variab
les for each
estab
lishm
ent
size (n
ot
show
n
in th
e tab
le) an
d
interactio
ns
of h
ouse
price
gro
wth
w
ith th
e
size of estab
lishm
ents.
Each
o
bserv
ation
is
at a co
un
ty,
4-d
igit
NA
ICS
in
dustry
, and estab
lishm
ent
size level.
All
regressio
ns
arew
eighted
b
y the
num
ber o
f househ
old
s in
a co
un
ty as o
f 2000. H
ouse
Price G
row
th
is instru
men
ted
usin
g th
e S
aiz (2010) m
easure
of elasticity
of h
ousin
g su
pply
at
an M
SA level.
Em
plo
ym
ent
gro
wth
isth
e percen
tage
chan
ge in
emplo
ym
ent
betw
een
2002 and
2007 estimated
usin
g
County
B
usin
ess Patte
rns
(CB
P)
data
. G
row
th in H
ouse
prices is th
e percen
tage
chan
ge b
etween
2002 and
2007, and
eachin
teraction
is w
ith a d
um
my in
dicato
r for th
e size
of th
e estab
lishm
ent.
All
regressio
ns in
clude 4
-dig
it industry
fix
ed effects.
Colu
mn
1 show
s the resu
lts wh
en
we ex
clud
e con
structio
n in
du
stries, co
lum
n2 ex
cludes
both
co
nstru
ction
and n
on-trad
able
ind
ustries,
colu
mn
3 also ex
cludes
finan
ce, in
suran
ce an
d real
estate-related
industries
(NA
ICS
co
des 52
and
53), co
lum
n
4 inclu
des
only
m
anufactu
ring
ind
ustries (N
AIC
S
31 to
33) an
d co
lum
n
5 has
man
ufactu
ring
in
du
stries that
are classified
as
"tradable
" in M
ian an
d
Sufi (2
01
1a).
All
regressio
ns
con
trol for th
e natu
ral
log
arithm
of p
op
ulatio
n,
the
percen
tage
of th
e p
opulatio
n
with
a colleg
e deg
ree, th
e
percen
tage o
f the
labor force
that is em
plo
yed
, the
share o
f the p
op
ulatio
n in th
e wo
rkfo
rce, an
d th
e p
ercentag
e of h
om
es that
are ow
ner-o
ccupied
.A
ll contro
ls are at a co
un
ty lev
el for th
e y
ear 2000 an
d are o
btain
ed u
sing C
ensu
s B
ureau
Data
S
um
mary
Files.
Stan
dard
errors are in p
arenth
esis an
d are
clustered
by M
SA.
*, *,
* d
eno
te statistic
al
significan
ce at th
e
10, 5,
and
1%
levels,
respectiv
ely.
Table 2.4: Breakdown of Manufacturing Industries by Distance Shipped
Manufacturing ManufacturingDist. Shipped <P50 Dist. Shipped >P50
Growth in House Prices -0.11 -0.29**(0.17) (0.14)
Growth in House Prices * 1-4 Employees 0.07 0.21**(0.14) (0.09)
Growth in House Prices * 5-9 Employees 0.11 0.20**(0.17) (0.09)
Growth in House Prices * 10-19 Employees -0.03 0.24**(0.17) (0.11)
Growth in House Prices * 20-49 Employees 0.06 0.04(0.30) (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 -0.58*(0.36) (0.32)
Percent of Homes Owner-occupied 0.00 0.00*(0.00) (0.00)
China Import Share in County (2005) -0.29 -1.21**(0.45) (0.58)
Controls Y Y4-Digit Industry Fixed Effects Y Y
Number of Observations 27,599 27,294R2 0.02 0.02
Growth HP * 1-4 E. = Growth HP * 5-9 E. 0.82 0.90Growth HP * 1-4 E. = Growth HP * 10-19 E. 0.59 0.77Growth HP * 1-4 E. = Growth HP * 20-49 E. 0.96 0.13
The table shows two-stage least squares regressions of employment growth on house price growth instrumented withthe elasticity of housing supply, indicator variables for each establishment size (not shown in the table) and interactionsof house price growth with the size of establishments. Each observation is at a county, 4 digit NAICS industry, andestablishment size level. All regressions are weighted by the number of households in a county as of 2000. House PriceGrowth is instrumented using the Saiz (2010) measure of elasticity of housing supply at an MSA level. Employmentgrowth 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 changebetween 2002 and 2007, and each interaction is with a dummy indicator for the size of the establishment. Allregressions include 4 digit NAICS fixed effects. The table splits industries and states based on the median of theshipment distance distribution (about 600 miles). Data for distance shipped is from the Census Commodity FlowSurvey for 2007 and represents a dollar weighted average of shipment distance calculated at the 3 digit NAICS andstate of origin level. All regressions control for the natural logarithm of population, the percentage of the populationwith 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 areobtained 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
Table 2.5:
Em
ployment and H
ouse Price A
ppreciation across Industry T
ypes
First S
tage A
ll Industries
Non-T
radab
leT
radab
le C
on
structio
n
Housing S
upply Elasticity
Grow
th in House P
rices
Log of th
e Populatio
n
Percent C
ollege Educated
Percent E
mployed (2000 C
ensus)
Workforce
as a Percentage of P
opulatio
n
Percent of H
omes O
wner-occupied
Chin
a Import S
hare in County (2005)
Num
ber of Observations
(0.02)
0.00(0.03)
0.00(0.00)
-0.01***(0.00)
-0.69(0.63)
0.00(0.00)
0.10(0.91)
731R
2 0.30
0.09(0.06)
-0.02**(0.01)
0.00*(0.00)
0.00(0.00)
-1.15***(0.23)
0.00**(0.00)
-0.23(0.28)
7310.24
0.10(0.07)
-0.01(0.01)
0.00**(0.00)
0.00*(0.00)
-1.13***(0.28)
0.00(0.00)
0.42(0.32)
7310.18
-0.01(0.11)
-0.02**(0.01)
0.00(0.00)
0.00(0.00)
-0.82(0.51)
0.00**(0.00)
-1.94***(0.47)
7300.10
0.32*** 0.06
(0.08) (0.06)
-0.02*
(0.01)
0.00(0.00)
0.00(0.00)
-0.83**(0.37)
0.00(**(0.00)
-0.52(0.42)
731
-0.03(0.01)
0.00(0.00)
0.00(0.00)
-1.35(0.24)
0.00(0.00)
0.42(0.32)
7310.30
0.21T
he
table
sho
ws
two
stag
e least sq
uares
regressio
ns
at a co
un
ty lev
el of em
plo
ym
ent
gro
wth
o
n h
ouse
price
gro
wth
betw
een 2002
and
2007.
Each
ob
servatio
n
is at a
coun
ty
level. A
ll reg
ressions
arew
eighted
b
y the n
um
ber
of h
ouseh
old
s in
a co
unty
as o
f 2000. H
ou
se P
rice Gro
wth
is in
strum
ented
usin
g th
e
Saiz
(2010) m
easure
of elasticity
of h
ousin
g su
pply
at
an M
SA lev
el. E
mp
loym
ent
gro
wth
is th
e p
ercentag
e ch
ange
in em
plo
ym
ent b
etween
2002
and
2007 estimated
usin
g C
ou
nty
B
usin
ess P
atte
rns
(CB
P)
data
. In
dustry
ty
pe d
efinitio
ns follow
M
ian
and
S
ufi (20
11a).
All reg
ressions
contro
lfor th
e n
atu
ral
log
arithm
of p
opulatio
n, th
e p
ercentag
e o
f the p
op
ulatio
n
with
a colleg
e deg
ree, the
percen
tage
of th
e labor force
that
is emplo
yed
, th
e sh
are of th
e pop
ulatio
n in
th
e wo
rkfo
rce, an
d th
epercen
tage
of h
om
es that are o
wner
occu
pied
. A
ll con
trols
are at a cou
nty
level for th
e y
ear 2000 an
d are
ob
tained
usin
g C
ensu
s B
ureau
Data
Sum
mary
F
iles. S
tandard
erro
rs are in p
arenth
esis and
areclu
stered
by M
SA.
*, **,
*** d
eno
te sta
tistical sig
nifican
ce at th
e
10%,
5%, an
d
1%
levels,
respectiv
ely.
Oth
ers
Table 2.6: Proprietorships and House Price Appreciation
BEA Census Start-up CapitalData Data < P50 (Census)
Growth in House Prices
Growth in House Prices * Proprietorships
Growth in House Prices * 1-4 Employees
Growth in House Prices * 5-9 Employees
Growth in House Prices * 10-19 Employees
Growth in House Prices * 20-49 Employees
Log of the Population
Percent College Educated
Percent Employed (2000 Census)
Workforce as a Percentage of Population
Percent of Homes Owner-occupied
China Import Share in County (2005)
0.02 0.03(0.06) (0.06)
0.14* 0.06(0.07) (0.06)
0.20*** 0.20***(0.05) (0.05)
0.08** 0.08**(0.04) (0.04)
0.01 0.01(0.04) (0.04)
0.00 0.00(0.04) (0.04)
-0.02** -0.02**(0.01) (0.01)
0.00** 0.00*(0.00) (0.00)
0.00 0.00(0.00) (0.00)
-1.02*** -1.16***(0.19) (0.20)
0.00** 0.00**(0.00) (0.00)
0.02 0.03(0.22) (0.23)
Number of Observations 4,381R2 0.48
4,3840.38
-0.04(0.07)
0.12*(0.06)
0.33***(0.07)
0.19***
(0.05)
0.14***(0.05)
0.13**(0.05)
-0.02***(0.01)
0.00(0.00)
0.00(0.00)
-1.21***(0.21)
0.00**(0.00)
0.18(0.24)
4,3840.31
Start-up Capital> P50 (Census)
0.05
(0.07)
0.08(0.08)
0.14**(0.06)
0.04(0.06)
-0.07(0.06)
-0.07(0.05)
-0.02**(0.01)
0.00**(0.00)
0.00(0.00)
-1.13***(0.21)
0.00*(0.00)
-0.02(0.23)
4,3820.28
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,Period (2007-2009)
Firm Size, and House Price Appreciation, Crisis
All Industries All Industries Start-up Capital(WLS) (IV) < P50 (IV)
Growth in House Prices
Growth in House Prices * 1-4 Employees
Growth in House Prices * 5-9 Employees
Growth in House Prices * 10-19 Employees
Growth in House Prices * 20-49 Employees
Log of the Population
Percent College Educated
Percent Employed (2000 Census)
Workforce as a Percentage of Population
Percent of Homes Owner-occupied
China Import Share in County (2005)
Number of ObservationsR2
-0.04*(0.02)
0.04**(0.02)
0.01(0.02)
0.00(0.02)
-0.02(0.02)
-0.01***(0.00)
0.00***(0.00)
0.00*(0.00)
-0.25***(0.07)
0.00***(0.00)
0.12*(0.07)
3,6540.16
-0.12***(0.03)
0.10***(0.03)
0.05*(0.03)
0.06*(0.03)
0.02(0.03)
0.00**(0.00)
0.00***(0.00)
0.00***(0.00)
-0.26***(0.06)
0.00***(0.00)
0.14*(0.08)
3,6540.12
-0.13***(0.04)
0.11***(0.04)
0.05*(0.03)
0.07**(0.03)
0.00(0.03)
0.00*(0.00)
0.00***(0.00)
0.00***(0.00)
-0.26***(0.07)
0.00***(0.00)
0.25***(0.09)
3.6510.08
Start-up Capital> P50 (IV)
-0.14***(0.04)
0.13***(0.05)
0.09(0.05)
0.09**(0.04)
0.07(0.05)
-0.01***(0.00)
0.00***(0.00)
0.00***(0.00)
-0.25***(0.07)
0.00***(0.00)
0.06(0.08)
3,6530.13
The table shows two-stage least squares regressions of employment growth between 2007 and 2009 on house pricegrowth 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 ofhouseholds in a county as of 2000. House Price Growth is instrumented using the Saiz (2010) measure of elasticity ofhousing supply at an MSA level. Employment growth is the percentage change in employment between 2007 and 2009estimated using County Business Patterns (CBP) data. Growth in House prices is the percentage change between2002 and 2007, and each interaction is with a dummy indicator for the size of the establishment. Columns 1 and 2, AllIndustries, 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 with50 or more employees. The first column for each sample of industries is aggregated at the county and establishmentsize level, whereas the second column is at the county, establishment size and industry level, and includes industryfixed effects. All regressions control for the natural logarithm of population, the percentage of the population with acollege degree, the percentage of the labor force that is employed, the share of the population in the workforce, andthe percentage of homes that are owner occupied. All controls are at a county level for the year 2000 and are obtainedusing 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
Tab
le 2
.8:
Tot
al E
mpl
oym
ent,
Une
mpl
oym
ent,
and
Mig
rati
on
Gro
wth
in
Hou
se P
rice
s
Log
of
the
Popula
tion
Per
cent
Col
lege
E
duca
ted
Perc
ent
Em
ploy
ed (
2000
Cen
sus)
Wor
kfor
ce
as a
Per
cent
age
of P
opula
tion
Per
cent
of
Hom
es O
wne
r-oc
cupi
ed
Chi
na I
mport
S
hare
in
Cou
nty
(200
5)
Num
ber
of O
bser
vati
ons
R2
Tota
l E
mplo
ym
ent
0.09
(0.0
6)
-0.0
2***
(0.0
1)
0.00
**(0
.00)
0.00
(0.0
0)
-1.1
5***
(0.2
3)
0.00
**(0
.00)
-0.2
3(0
.28)
731
0.24
Unem
p.
Unem
p.
Rate
-0.2
0(0
.14)
-0.0
1(0
.02)
(0.0
0)
0.00
(0.0
0)
-0.1
3(0
.52)
0.00
***
(0.0
0)
-0.6
0(0
.64)
721
0.26
-1.2
9**
(0.6
6)
0.03
(0.1
0)
(0.0
1)
0.04
**(0
.02)
3.94
(2.6
7)
0.03
***
(0.0
1)
-4.7
6(3
.65)
721
0.33
Net
Mig
rati
on
-0.1
6(0
.12)
0.00
(0.0
1)
0.00
(0.0
0)
0.00
(0.0
0)
-0.0
1(0
.19)
0.00*
*(0
.00)
0.19
(0.2
9)
731
Infl
ow
s O
utf
low
s
0.19
0.
34**
(0.1
2)
(0.1
7)
-0.0
7***
-0
.07*
**(0
.01)
(0
.01)
0.01
***
0.00
***
(0.0
0)
(0.0
0)
0.00
0.
00(0
.00)
(0
.00)
-0.6
3*
-0.6
2**
(0.3
4)
(0.2
6)
.0**
-0
.01*
**(0
.00)
(0
.00)
-1.0
8***
-1
.27*
**(0
.28)
(0
.44)
731
731
0.41
0.
18T
he
table
sh
ow
s tw
o s
tage
leas
t sq
uar
es re
gre
ssio
ns
at a
co
un
ty
lev
el
of
the n
et
mig
rati
on o
n
house
pri
ce g
row
th
bet
wee
n
2002
an
d 2
007.
A
ll re
gre
ssio
ns
are
wei
ghte
d b
y th
e num
ber
of
house
hold
s in
aco
unty
as
of
2000
. H
ouse
P
rice
G
row
th i
s in
stru
men
ted
usi
ng t
he
Sai
z (2
010)
m
easu
re o
f el
asti
city
of
housi
ng
sup
ply
at
an
MSA
le
vel.
N
et
Mig
rati
on,
Infl
ow
s an
d O
utf
low
s ar
e o
bta
ined
fr
om
the
IRS
county
to
co
un
ty m
igra
tion data
se
ries.
N
et
Mig
rati
on
is
calc
ula
ted
by
co
un
ty usi
ng
infl
ow
s of
taxpayers
m
inus
ou
tflo
w
of
taxpayers
in
a year
as
a p
roport
ion
of
non m
igra
nts
(i
.e.
people
th
at
file
d
inth
e sa
me
cou
nty
in
t-1
an
d t)
. F
or
each
dep
enden
t v
aria
ble
th
e fi
rst
colu
mn
sho
ws
the
resu
lts
for
the
regre
ssio
ns
wit
hout
contr
ols
, an
d t
he
seco
nd
colu
mn
sho
ws
the
coef
fici
ents
co
ntr
oll
ing
for
log
ofpopula
tion,
the
per
centa
ge
of
the
popula
tion w
ith a
coll
ege
deg
ree,
the
per
cen
tag
e of
the
lab
or
forc
e th
at
is e
mplo
yed
, th
e sh
are
of
the
po
pu
lati
on
in
the
wo
rkfo
rce,
and th
e p
erce
nta
ge
of
ho
mes
that
are
ow
ner
o
ccu
pie
d.
All
con
tro
ls
are
at
a co
un
ty
leve
l fo
r th
e y
ear
2000
an
d
are
ob
tain
ed
usi
ng
Cen
sus
Bu
reau
Data
Su
mm
ary
F
iles
. S
tan
dar
d
erro
rs a
re
in p
aren
thes
is
and a
re c
lust
ered
by
MSA
. *,
*,
*
den
ote
st
ati
stic
al
signif
ican
ce
at t
he
10%
, 5%
, an
d
1%
lev
els,
res
pec
tiv
ely
.
-0.0
1***-0
.03***
Table 2.9: Denial Rates
Panel ALow Elasticity High Elasticity Difference
Denial Rate (2002) 0.12 0.14Change in Denial Rate (02-07) 0.02 -0.01 0.03***
(0.06) (0.05)Volume (2002) 9,454 3,811
Volume per Household (2002) 0.07 0.06Change in Volume (02-07) -0.01 0.10 .-0.11***
(0.27) (0.22)Number of Counties 394 382
Denial Rates
Elasticity -0.03*** -0.01*** -0.01*** 0.07**(0.00) (0.00) (0.00) (0.03)
Debt to Income (2002) 0.11*** -0.01(0.02) (0.04)
Changre in Debt to Income (02-07)
Log of the Population
Percent College Educated
Percent Employed (2000 Census)
Workforce as a Percentage of Population
Percent of Homes Owner-occupied
China Import Share in County (2005)
DTI dataNumber of Observations
P2
0.02*(0.01)
0.06***(0.01)
0.02*** 0.02***(0.00) (0.00)
0.00*** 0.00***(0.00) (0.00)
0.00 0.00***(0.00) (0.00)
-0.15*(0.08)
0.00*(0.00)
-0.08(0.10)
0.00(0.00)
-0.39*** -0.49***(0.11) (0.11)
NY Fed / IRS776 7630130 0.58
HMDA774
0.55
-0.57*** -0.13(0.11) (0.21)
-0.26*** -0.29**(0.05) (0.10)
-0.05** -0.08**(0.02) (0.03)
0.01**(0.00)
-0.01**(0.00)
0.00(0.00)
0.00(0.00)
-1.05** -1.10*(0.44) (0.61)
-0.01*** -0.01***(0.00) (0.00)
-0.12(0.66)
7760.09
NY Fed / IRS7630.42
0.47(0.90)
HMDA774026
The table shows the relation between mortgage denial rates and mortgage volume at a county level and the elasticity ofhousing supply. Total application volume is calculated as the sum of all loans that are originated plus applications thatare approved but not accepted, applications denied by the financial institution and loans purchased by the financialinstitution itself in each county and year, all scaled by the total number of households in a county as of 2000. Denialrates are computed as the proportion of applications denied by the financial institution over total volume in eachcounty and year. All the data is extracted from HMDA LAR records. Panel A shows the average denial rates andaverage volume in 2002 and 2007, as well as the change in these variables during this period for counties above andbelow the median elasticity of housing supply in the sample. Panel B shows OLS regressions of the change in denialrate 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 acollege degree, the percentage of the labor force that is employed, the share of the population in the workforce, thepercentage 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
Panel B
Volume
-0.01(0.02)
0.02(0.02)
2.6 Appendix. Calculating the magnitude of thecollateral effect
We follow the same calculation as Mian and Sufi (2011a) to aggregate the collateraleffect across all counties in the data. We start with the differential impact of houseprices on employment creation at small firms relative to firms with 50 or more em-ployees, and we focus on the specifications where we exclude non-tradable industriesand construction (Table 2.3, Column 2). We first compute predicted county-level em-ployment gains for each establishment size bins in this subset of industries (relative tothe 10th percentile county), and then we aggregate to all counties. Below we describeeach step in detail.
First, we compute the county-level predicted change in employment in each estab-lishment size category by multiplying the regression coefficient by the change in houseprices between 2002 and 2007 in each county. We then subtract the predicted changein the 10th percentile county in the change in house prices (to avoid being affectedby outliers at the bottom of the distribution). Second, we multiply the predictedcounty-level change in employment in each establishment size bin by the employmentin that size bin in each county as of the beginning of the period (2002) to obtaina predicted change in employment in terms of numbers of workers for each countyand establishment size. Third, we sum up the predicted changes across all countiesand establishment size bins to obtain an economy-wide predicted change due to thecollateral 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 theeconomy 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 0.315for size bin 1-4 employees from Column 2 in Table 2.3. Given a change in house pricesof 0.12 in the 10th percentile county, this yields a predicted employment change inthis size bin in the subset of industries in this regression (all except non-tradable andconstruction) 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 hasa change in house prices at the median (0.267) the predicted change in that county forthis 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 tothose with 50 or more employees of 1.698 million jobs in all counties, or 27.8% of jobscreated 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. Ifwe restrict our attention to the specification where the demand explanation for ourresults 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 ofTable 2.4), the same computation would yield an estimate of 676 thousand jobs, orabout 11% of jobs created in this period and subset of counties.
106
Table 2.10: Employment Growth,ual Industries by Firm Size
Firm Size, and House Price Appreciation: Individ-
Growth in House Prices
Log of the Population
1-4 Emp 5-9 Emp
0.13*** 0.11**(0.05) (0.05)
0.05(0.05)
-0.03*** -0.06*** -0.06***(0.01) (0.01) (0.01)
-0.02(0.08)
50+ Emp
0.03(0.12)
-0.04*** -0.06***(0.02) (0.02)
Percent College Educated
Percent Employed (2000 Census)
Percent of Potential Worker Population
Percent of Homes Owner-occupied
4-Digit Industry Fixed EffectsNumber of Observations
R-Square
0.00 0.00(0.00) (0.00)
0.00 0.00(0.00) (0.00)
-0.75*** -1.16*** -0.83***(0.20) (0.18) (0.21)
0.00 0.00(0.00) (0.00)
Y110,069
0.34
Y80,915
0.37
The table shows two-stage least squares regressions at a county level of employment growth on house price growthsplit 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 BusinessPatterns (CBP) data. Growth in House prices is the percentage change between 2002 and 2007, and each interactionis a dummy indicator for the size of the establishment. All regressions include 4 digit industry fixed effect and controlfor log of population, the percentage of the population with a college degree, the percentage of the labor force thatis 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.
107
10-19 Emp 20-49 Emp
0.00(0.00)
0.00(0.00)
0.00(0.00)
0.00(0.00)
-0.58*(0.31)
0.00(0.00)
Y61,4270.34
0.00(0.00)
0.00(0.00)
-0.99**(0.44)
0.00(0.00)
Y50,3810.27
0.00(0.00)
Y71,947
0.37
Table 2.11: Robustness Test: Difference between High and Low Start-up Capital
Growth in House Prices
Growth in HP * High Startup Capital
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 EffectsNumber of Observations
R2
1-4 Emp 5-9 Emp
0.23*** 0.11*(0.06) (0.06)
-0.21*** 0.00(0.05) (0.06)
10-19 Emp 20-49 Emp 50+ Emp
0.03(0.06)
0.05(0.06)
0.03(0.09)
-0.11(0.07)
0.01(0.13)
0.03(0.09)
-0.03*** -0.06*** -0.06*** -0.04*** -0.06***(0.01) (0.01) (0.01) (0.02) (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)
-0.75*** -1.16*** -0.82***(0.20) (0.18) (0.21)
0.00 0.00 0.00(0.00) (0.00) (0.00)
Y Y110,069 80,915
0.34 0.37
Y71,9470.37
0.00(0.00)
0.00(0.00)
-0.59*(0.31)
0.00(0.00)
Y61,427
0.34
0.00(0.00)
0.00(0.00)
-0.99**(0.44)
0.00(0.00)
Y50,381
0.27
The table shows two-stage least squares regressions at a county level of employment growth on house price growthsplit by size of establishment and interacted with a High Startup Capital indicator (indicator itself not shown). HighStartup Capital is defined as 4 digit industries for which the amount of capital to start the firm is higher than themedian for all industries. All regressions are weighted by the number of households in a county as of 2000. House PriceGrowth is instrumented using the Saiz (2010) measure of elasticity of housing supply at an MSA level. Employmentgrowth 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 adummy indicator for the size of the establishment. All regressions include 4 digit industry fixed effect and controlfor log of population, the percentage of the population with a college degree, the percentage of the labor force thatis 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 establishmentcategory. Standard errors are in parenthesis and are clustered by MSA. *, *, *** denote statistical significance atthe 10%, 5%, and 1% levels, respectively.
108
Table 2.12: Effect of One Standard Deviation Change in the Independent Variable
1-4 Emp 5-9 Emp 10-19 Emp 20-49 Eip 50+ Emp
Employment in All SectorsEffect of 1 sigma change in HP
Growth (02-07)Employment as of 2002
Employment in Firms <P50 of Start-Up CapitalEffect of 1 sigma change in HP
Growth (02-07)Employment as of 2002
Employment in Firms >P50 of Start-Up Capital
Effect of 1 sigrma change in HPGrowth (02-07)
Employment as of 2002
5.29.4
9,101
2.7 1.3 1.1 1.18.0 12.5 10.6 13.3
9,122 12.819 21,466 72,939
6.8 3.9 2.9 2.7 -0.110.9 11.1 13.4 14.2 25.0
6,213 5,566 7,350 11,012 39,921
4.2 2.1 -0.1 -0.2 1.36.6 4.3 13.0 9.4 9.3
2,888 3,556 5,468 10,453 33,018
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
AverageStd. Dev.
Percentiles:
630.2368.4
651.7218.3
1% 25.0 168.925% 378.1 559.350% 600.8 620.475% 817.7 831.799% 1,789.2 1,021.3
Number of Observations 950 21
Panel B: Deciles of NAICS and State Dollar-weighted Average Distance Measure
Industry-State Deciles
NAICS Description 1 2 3 4 5 6 7 8 9 10
Food Manuf. 1 2 7 10 13 2 6 4 4Beverage & Tobacco Product Manuf. 15 16 8 3
Textile Mills 2 1 4 4 3Textile Product Mills 3 2 8 2 3
Apparel Manuf. 1 1 2 1 3Leather & Allied Product Manuf. 1 2
Wood Product Manuf. 8 12 13 4 4Paper Manuf. 2 3 7 9 6
Printing & Related Support Activities 5 11 5 13 5Petroleum & Coal Products Manuf. 27 10 4 2
Chemical Manuf. 1 1 2 11Plastics & Rubber Products Manuf. 1 1 3 7 8
Nonmetallic Mineral Product Manuf. 16 20 12 3Primary Metal Manuf. 2 4 9 8
Fabricated Metal Product Manuf. 3 2 3 11 10Machinery Manuf. 1 1 1
Computer & Electronic Product Manuf. 3 1 1 5Electrical Eq., App., & Component Manuf. 2 1 2
Transportation Equipment Manuf. 2 4 1 3 6Furniture & Related Product Manuf. 5 2 8 11 6
Miscellaneous Manuf. 2 1
2 1 26 5 84 4 4 6
4 5 33 2 2 33 2 38 6 3 32 6 1 119 8 4 612 8 8 2
1
374
1111117
7 2 5 5 47 7 2 67 7 12 10 95 5 10 3 155 5 6 15 106 2 10 4 93 7 3 2 15 9 13 9 10
The table shows the dollar weighted distance of shipments for 3 digit NAICS manufacturing industries. Data isobtained from the 2007 Commodity Flow Survey. The first column of Panel A shows the weighted average distancefor each industry and state, and the second column aggregates the distances shipped at the 3 digit NAICS level. PanelB shows the frequency with which each industry appears in each state x industry decile.
110
311312313314315316321322323324325326327331332333334335336337339
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 ExtractionUtilities
ConstructionManufacturing
Wholesale TradeRetail Trade
Transportation and WarehousingInformation
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 SvcsEducational Services
Health Care and Social AssistanceArts, Entertainment, and Recreation
Accommodation and Food Services
Other Services (except Public Administration)
NAICS2
11
212223314244485152535455566162717281
Average Start-UpAmount (USD)
146,033673,609601,14978,372
363,166188,085216,302131,893236,126203,799220,69187,879488,68191,278156,893214,889218,061273,186161,995
Above/BelowMedian
0110101010101000110
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 Industry-Demeaned Distance Deciles
> 50 Employee Establishments (2002), Deciles 1 2 3 4 5 6 7 8 9 10
1 10 7 6 3 2 3 2 5 9 102 15 12 6 3 5 10 5 13 12 163 11 9 5 10 12 10 6 9 12 114 5 7 13 11 10 12 11 13 8 95 8 10 10 11 10 13 17 5 8 76 5 9 9 9 14 7 17 15 8 67 9 15 12 17 6 9 12 4 6 78 6 9 12 14 14 7 5 15 7 109 8 9 11 10 10 12 11 8 9 510 16 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 CensusCommodity Flow Survey, and also the share of employment in establishments that have more than 50 employees foreach 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 andindustry observation, we compute the deviation from the industry mean for both measures and classify observationsinto deciles based on these deviations.
112
Tab
le 2
.16:
H
ouse
Pri
ce G
row
th a
nd C
reat
ion
of E
stab
lish
men
ts
Gro
wth
in
Hou
se P
rice
s
Log
of
the
Popula
tion
Per
cent
Col
lege
E
duca
ted
Per
cent
E
mpl
oyed
(2
000
Cen
sus)
Wor
kfor
ce a
s a
Per
cent
age
of P
opula
tion
Per
cent
of
Hom
es O
wne
r-oc
cupi
ed
Chi
na I
mp
ort
Sha
re i
n C
ount
y (2
005)
Bir
ths
of
Est
. D
eath
s o
f E
st.
(1)
(2)
(3)
(4)
0.46
***
0.46
***
0.31
***
0.28
***
(0.1
2)
(0.1
2)
(0.0
7)
(0.0
8)
731
Net
Cre
ati
on
of
Est
. B
irth
s, C
ap
ital
<
P5
0(5
) (6
) (7
) (8
)
0.16
**(0
.06)
Bir
ths,
Cap
ital
> P
50
(9)
(10)
0.18
***
0.57
***
0.43
***
0.32
***
0.50
***
(0.0
6)
(0.1
3)
(0.1
4)
(0.1
1)
(0.1
3)
-0.0
1 -0
.01
0.00
0.
01
-0.0
1*
-0.0
2***
-0
.01
(0.0
1)
(0.0
2)
(0.0
1)
(0.0
1)
(0.0
1)
(0.0
1)
(0.0
2)
0.01
* 0.
00
0.00
* 0.
00
0.00*
(0.0
0)
(0.0
0)
(0.0
0)
(0.0
0)
(0.0
0)
0.00
0.
00
0.00
(0.0
0)
(0.0
0)
(0.0
0)
-2.3
4***
-1
.78*
* -1
.06*
*
0.00
0.
00(0
.00)
(0
.00)
-0.6
5 -1
.28*
**(0
.67)
(0
.79)
(0
.40)
(0
.49)
(0
.29)
0.00
* 0.
00*
(0.0
0)
(0.0
0)
-0.6
2
0.00
0.
00(0
.00)
(0
.00)
0.00
(0.0
0)0.
00(0
.00)
-0.0
1(0
.02)
0.00
(0.0
0)
0.00
(0.0
0)
0.00
(0.0
1)
0.01
**(0
.00)
0.00
(0.0
0)
-0.0
1(0
.02)
0.00
(0.0
0)
0.00
(0.0
0)
-1.1
3***
-2
.43*
**
-2.1
7**
-2.1
7***
-1
.35*
(0.3
3)
(0.7
1)
(0.8
8)
(0.6
3)
(0.7
7)
0.00
0.0
0 0.
00**
0.
00**
0.
00*
(0.0
0)
(0.0
0)
(0.0
0)
-0.4
5 -0
.46
-0.6
0 -0
.16
(0.5
7)
(0.6
7)
(0.3
5)
(0.4
0)
(0.2
9)
2-D
igit
NA
ICS
Fix
ed E
ffec
tsN
umbe
r of
Obs
erva
tion
sR
2
-Y
731
0.29
13,4
820.
20
Y
(0.0
0)
(0.0
0)
0.16
-0
.58
(0.3
5)
(0.6
4)
Y13
.482
73
1 13
,482
0.01
**(0
.00)
-0.2
4(0
.61)
-Y
731
7,16
70.
21
0.22
0.
31
0.16
0.
29
0.20
0.00
(0.0
0)
-0.6
9(0
.49)
731
0.27
0.00
(0.0
0)
-0.6
8(0
.85)
Y6,
315
0.20
The
table
sh
ow
s tw
o
stag
e le
ast
squ
ares
re
gre
ssio
ns
of
esta
bli
shm
ent
bir
ths
and
dea
ths
on
house
pri
ce g
row
th
inst
rum
ente
d
wit
h t
he
elas
tici
ty
of h
ousi
ng su
pply
. E
ach
obse
rvat
ion
is a
t a
county
le
vel
for
the
regre
ssio
ns
wit
hout
sect
or
fixed
eff
ects
(o
dd
nu
mb
ered
colu
mns)
an
d a
t a
county
and
2 dig
it
NA
ICS
in
du
stry
le
vel
wh
enev
er w
e in
clude
fix
ed e
ffec
ts
(even
num
ber
ed c
olu
mns)
. A
ll re
gre
ssio
ns
are
wei
gh
ted
b
y th
e n
um
ber
of
ho
use
ho
lds
in a
county
as
of
2000
. H
ou
se P
rice
Gro
wth
is
inst
rum
ente
d
usi
ng t
he
Sai
z (2
010)
m
easu
re o
f el
asti
city
of
housi
ng
sup
ply
at
an
M
SA l
evel
. B
irth
s an
d d
eath
s of
esta
bli
shm
ents
co
me
from
th
e C
ensu
s S
tati
stic
s of
U.S
. B
usi
nes
ses
and
ar
e su
mm
ed b
etw
een 2
002
and
20
07 a
nd
sc
aled
by
the n
um
ber
of
esta
bli
shm
ents
in
a
county
as
of
2002
. G
row
th
in H
ou
se
pri
ces
isth
e p
erce
nta
ge
chan
ge
bet
wee
n 2
002
and
200
7,
and e
ach
in
tera
ctio
n
is w
ith
a d
um
my
in
dic
ato
r fo
r th
e si
ze o
f th
e es
tab
lish
men
t.
Colu
mns
1 an
d 2
sh
ow
s th
e r
esu
lts
for
bir
ths
of
esta
bli
shm
ents
, C
olu
mns
3 an
d
4 sh
ow
res
ult
s fo
r dis
appea
rance
of
esta
bli
shm
ents
an
d
Colu
mns
5 an
d 6
use
the n
et c
reat
ion
of e
stab
lish
men
ts
as t
he
dep
enden
t var
iable
. T
he
fin
al
four
colu
mns
spli
t th
e s
ample
b
y th
e am
ount
ofca
pit
al n
eces
sary
fo
r st
art
ing
a
bu
sin
ess
and
sh
ow
res
ult
s fo
r es
tab
lish
men
t bir
ths.
A
ll re
gre
ssio
ns
con
tro
l fo
r th
e natu
ral
log
arit
hm
of
po
pu
lati
on
, th
e p
erce
nta
ge
of
the
po
pu
lati
on
wit
h
a co
lleg
e deg
ree,
the
per
centa
ge
of
the
lab
or
forc
e th
at
is e
mplo
yed
, th
e sh
are
of
the
po
pu
lati
on
in
the
wo
rkfo
rce,
an
d th
e p
erce
nta
ge
of
ho
mes
th
at
are
ow
ner
occ
upie
d.
All
contr
ols
ar
e at
a
county
le
vel
fo
r th
e
yea
r20
00
and
are
obta
ined
usi
ng
Cen
sus
Bure
au D
ata
S
um
mar
y F
iles
. S
tandar
d
erro
rs a
re
in p
aren
thes
is
and
ar
e cl
ust
ered
b
y M
SA.
*,
*,
* d
eno
te st
ati
stic
al
signif
ican
ce
at t
he
10%
, 5%
, an
d
1%
level
s,re
spec
tiv
ely
.
-I
Table 2.17: L
ist of 3-digit NA
ICS Industries E
xcluding Non-tradables, M
anufacturing, F.I.R.E
., and Construction
NA
ICS
113114115213221237423424425454481483484485486487488492493511512515516517518519541551561562611621622623624711712713721811812813
Th
e table sh
ows
the 3 d
igit N
AIC
S
codes,
as well
as the
pro
portio
n of em
ployees
in each
estab
lishm
ent size
category and
the
tota
l nu
mb
er of em
ployees
in each
in
du
stry in
our sam
ple
of coun
ties.
Descrip
tion
1-4 E
mp
.
Forestry and L
ogging 19.8%
Fishing, H
unting and Trapping
12.7%S
upport Activities for A
griculture and Forestry
17.9%S
upport A
ctivities for Mining
5.3%U
tilities 1.5%
Heavy and C
ivil Engineering C
onstruction 5.4%
Merchant
Wholesalers, D
urable G
oods 7.8%
Merchant
Wholesalers, N
ondurable Goods
6.8%W
holesale Electronic M
arkets and Agents and B
rokers 26.3%
Nonstore R
etailers 12.3%
Air T
ransp. 0.9%
Water
Transp.
3.0%T
ruck T
ransp. 9.7%
Transit and G
round Passenger T
ransp. 4.3%
Pipeline T
ransp. 3.7%
Scenic and Sightseeing T
ransp. 19.3%
Support
Activities for T
ransp. 7.6%
Couriers and M
essengers 2.6%
Warehousing and S
torage 2.8%
Publishing
Ind. (except Internet)
3.2%M
otion Picture
and Sound R
ecording Ind.
12.1%B
roadcasting (except Internet)
2.4%Internet P
ublishing and B
roadcasting 7.8%
Telecom
nmunications
4.1%IS
Ps, W
eb Search, and D
ata Processing
5.9%O
ther Information Serv.
7.6%P
rofessional, Scientific, and T
echnical Serv. 16.8%
Managem
ent of Com
panies and Enterprises
1.3%A
dministrative and S
upport Serv. 5.9%
Waste M
anagement and R
emediation Serv.
5.8%E
ducational Serv.
3.2%A
mbulatory H
ealth Care Serv.
10.6%H
ospitals 0.0%
Nursing and R
esidential Care F
acilities 1.2%
Social Assistance
5.3%P
erforming A
rts, Spectator S
ports, and Related Ind.
18.2%M
useums, H
istorical S
ites, and Sim
ilar Institutions 4.7%
Am
usement, G
ambling, and R
ecreation Ind. 4.8%
Accom
modation
2.3%R
epair and Maintenance
23.1%P
ersonal and Laundry Serv.
19.7%R
eligious, G
rantmaking, C
ivic Org.
11.8%
5-9 Em
p.
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%
10-19 Em
p.
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%10.2%2.8%6.9%5.2%6.9%4.8%6.4%6.4%5.3%13.3%11.1%2.8%5.9%10.6%5.2%16.5%0.0%5.5%
15.9%5.9%6.5%8.3%7.5%
20.5%21.4%15.2%
20-49 E
mp
.
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%
50+
Em
p.
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%
Chapter 3
Credit Supply and House Prices:Evidence from Mortgage MarketSegmentation
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 finan-
cial 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 innova-
tion (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 iden-
tify 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 allowsus to provide clean estimates of the effect of lower cost of credit on house prices. TheCLL determines the maximum size of a mortgage that can be purchased or securi-tized by Fannie Mae or Freddie Mac. Mortgages below the CLL therefore have lowerinterest rates compared to jumbo loans (loans that are above the CLL), since theformer benefit from implicit (and since 2008, explicit) government support for FannieMae and Freddie Mac. The difference in interest rates between conforming loans andjumbo loans has been estimated to be up to 24 basis points.2 . In addition, Loutskinaand Strahan (2009, 2011) show that more borrowers are able to access mortgagesbelow the conforming loan limit than above, which suggests that not only the cost ofcredit 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 conformingloan limit (CLL) from one year to the next are exogenous to local housing markets andthe local economy, since this change is based on the national average appreciation inhouse prices. That means that, in a given year, a house just above the CLL thresholdhas to be financed by an expensive jumbo loan, while the next year the equivalenthouse can be financed via cheaper conforming loan. Our empirical approach involvescomparing transactions that can be financed more easily using a conforming loan, andhouses that are more expensive so that buyers need to obtain larger (jumbo) loans tomaintain the same loan-to-value ratio. We track transactions in the price range justabove and below the CLL in the year that the limit is in effect and compare them tothe subsequent year, once the limit is raised and houses just above the CLL becomeeligible for conforming loans. This setup enables us to cleanly identify the effect ofthe 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 aconforming loan in a given year is obtained by dividing the conforming loan limit by0.8.' By construction, buyers of houses with a price below this threshold can get aconforming loan that makes up 80 percent of the price of the house, whereas if theprice of the house is above 125 percent of the CLL, it can no longer be financed at80 percent with a conforming loan. Loans with a loan-to-value (LTV) ratio below80 are associated with more attractive terms, and conforming loans above 80 percentrequire private mortgage insurance in order to qualify for purchase by Fannie Maeor Freddie Mac (Green and Wachter, 2005). Above this price threshold, borrowerseither 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 advantageof the lower interest rate below the CLL, they have to use savings or alternativeforms of financing to make a larger down payment. Importantly, our sample includesall 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
2See for example McKenzie (2002), Ambrose, LaCour-Little, and Sanders (2004), Sherlund(2008), Kaufman (2012), or DeFusco and Paciorek (2013)
3Kaufman (2012) uses this threshold for appraisal values to study the effect of the conformingstatus of a loan on its cost and contract structure. Loutskina and Strahan (2013) follow our approachand use changes in the CLL interacted with regional constraints to look at financial integration andthe 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
4This 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 easewith which they could be securitized) and also when second lien mortgages becamewidely available (see Figure 3-5 ). Both these effects reinforce the idea that when theCLL was more important in the earlier part of the sample, its impact is also moresignificant in our estimates.
Given our estimate for the change in house prices due to changes in credit condi-tions, we can compute the semi-elasticity of prices to differences in interest rates inthe region close to the threshold. We use the differences in interest rates estimatedin the prior literature of 10 to 24 basis points between conforming and jumbo loansas our measure of the cost differential for buyers above and below the threshold. Weobtain local elasticity estimates that range from a low end of 1.2 to an upper range of9.1 depending on the period and the exact estimate for the interest rate differentialbetween jumbo and conforming loans that we use for our calculations. These elastic-ity 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 estimatesby focusing on whether the effect of cheaper credit is stronger when buyers face othertypes 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 the10th percentile of the income growth distribution for each individual regression. Thepoint estimate for these areas shows that value per square foot is 2.50 dollars higherin the year that a house becomes eligible to be financed with a conforming loan. Thisis more than double the size of the average elasticity that we found in the overallsample, suggesting that cheaper credit may have had a disproportional impact oneconomically more depressed households and regions.
We show that our results are not driven by a subsidy effect that provides a focalpoint to draw in more bidders. First, there is no visible bunching in the number oftransactions just around the threshold of CLL divided by 0.8, suggesting that thesupply of housing does not react strongly to the CLL. We also do not find that thereis bunching in the number of unobserved bidders for homes around the CLL, which wemeasure as the share of borrowers that apply for loans but ultimately either withdrawor do not use the loans they are approved for. If the CLL served simply as a focalpoint for home sales, we should expect more bidders for homes that are eligible forconforming loans. Instead, we find that our measure of the share of failed bids islower, not higher, for borrowers that borrow up to the CLL. The fact that there isneither a significant jump in the quantity of transactions nor in our proxy of failedoffers for homes suggests that the effect we find on prices is more consistent with acost of credit interpretation.
The rest of the paper is structured as follows: Section 3.2 discusses related liter-ature and the user cost model. Section 3.3 describes our data and the identificationstrategy. In Section 3.4, we lay out the regressions results and robustness checks ofour 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 down-
payment 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 assump-
tions 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 inhouse prices due to changes in the average cost of financing, since we look at exoge-nous movements in the cost of capital for home buyers. Our empirical results providelocal estimates for the numerator of the elasticity calculation. In Section 3.4.4, wediscuss 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 residen-tial properties available in deeds and assessors records over 11 years, from 1998 to2008, and seventy-four counties in ten metropolitan statistical areas (MSAs) - Boston,Chicago, DC, Denver, Las Vegas, Los Angeles, Miami, New York, San Diego, and SanFrancisco. We limit our attention to transactions of single-family houses, which ac-count for the large majority (approximately 78 percent) of all observations.
Each observation in the data contains the date of the transaction, the amount forwhich a house was sold, the size of the first mortgage, and an extensive set of variablesabout the property itself. These characteristics include the property address, interiorsquare 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 ofrenovation. Additional characteristics include the availability of a fireplace, parking,the architectural and structural style of the building, the type of construction, exteriormaterial, availability of heating or cooling, heating and cooling mechanism, type ofroof, view, attic, basement, and garage. We describe the procedure for cleaning theraw 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 single-family houses that are summarized in Tables 3.1 and 3.2.6 We can see in Panel A ofTable 3.1 that the average transaction value is 309 thousand dollars with a standarddeviation of 124 thousand dollars. The average size of the houses is 1,735 sqft, andthe houses have, on average, 3 bedrooms and 2 bathrooms. The average loan tovalue is 0.81 (including only the first mortgage for each transaction), and the medianLTV is 80 percent. The average value per square foot is 194 dollars with a standarddeviation 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 regressionsin the final three columns. For the regression sample, the average price for eachhouse is higher than in the whole dataset (at 371 thousand dollars, compared to 309thousand in the first column). This is consistent with the fact that the conformingloan limit was set to cover substantially more than 50 percent of the mortgages madeevery year (Acharya, Richardson, Nieuwerburgh, White, 2011). These houses are also,on average, larger and have more bedrooms and bathrooms than the whole dataset.
6Please see the Appendix for a detailed description of the procedure for cleaning the data initiallyobtained 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 differ-
ences 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 re-
gressions of value per square foot and log of house price on a number of house char-
acteristics, 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 theconstruction of each variable, in the Appendix to the paper.
The estimated R2 of each of these regressions (80 in total for each of the twoleft-hand side variable-10 MSAs in 8 years) is between 40 and 60 percent for theprice of the transaction, and 50 to 70 percent when we use value per square foot as adependent variable.
Summary statistics for the residuals from the hedonic regressions for the wholesample 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, and0.17 thousand dollars for the log of the price of the house. The hedonic regressionsare estimated on the whole dataset of transactions (the 3.98 million observationsmentioned above), so when we restrict our attention to the regression sample, theaverage error no longer has to be zero. Indeed, for the regression sample, the averageresidual from the hedonic regressions for the value per square foot is positive at 5.3dollars, and the average error for the log of transaction value of the house is 0.05dollars (last three columns of Panel B of Table 3.1). The standard deviation of theresiduals for the regression sample is similar in magnitude to what we obtain for allthe transactions.
3.3.3 Empirical Approach
Identification Strategy
To identify the effect of changes in credit conditions on house prices, we restrict ouranalysis to two groups of buyers who all buy houses in a tight price range, but differin the financing available to them. The sample for our regressions is made up ofhouses that transact in a band around 125 percent of each year's conforming loanlimit, 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 theyear's CLL (i.e. transactions that fall between 125 percent of CLL and 125 percentof CLL minus USD 10,000) and houses above that threshold that transact between125 percent of CLL and 125 percent of CLL+10, 000. By construction, in the yearthat the conforming loan limit is in effect, houses above the threshold of 125 percentof the CLL cannot be financed at 80 percent using a conforming loan, whereas thehouses below the threshold can be financed. Thus, home buyers that bid for housespriced above 125 percent of CLL cannot finance a full 80 percent of the transactionwith 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 percentwith 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 arenot biased by the endogeneity of the choice of financing of each specific transaction.
The identification strategy is best understood through an example. Consider the7 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 wehave 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 theconforming loan limit on the valuation of transactions made just above the thresholdof 125 percent of the CLL. We run differences-in-differences regressions year-by-yearwith one indicator variable for houses priced above the conforming loan limit dividedby 0.8, another indicator for the year in which the CLL is in effect, and an interactionof these two indicator variables. We also include ZIP code fixed effects in all regres-sions, so our estimates do not reflect differences between neighborhoods, but rathervariation within zip codes.
The sample for each year-by-year regression includes houses within a USD 10,000band around the conforming loan limit in the year in which the limit is in force, aswell as the subsequent year. This implies that the "Above the Threshold" indicatorvariable takes a value of 1 if the price at which a house transacts is greater than 125percent of the conforming loan limit of a certain year, and less than that amount plus10,000 dollars. This same variable is a 0 for transactions between 125 percent of theCLL and 125 percent of the CLL minus 10,000 dollars. The "Year CLL" indicatorvariable is a 1 in the year in which the CLL is in effect for each regression, anda 0 in the subsequent year. We use a tight band around the threshold so that alltransactions in the year after the limit is in effect are eligible for an 80 percent LTVconforming loan. We thus have a group of transactions that is "easy to finance andanother one that is "hard to finance in the year that the limit is in effect, but alltransactions in the sample are "easy to finance once the limit is raised.'
We run regressions of the following form:
Valuation measurei = N + /11AboveThreshold + /32lYearCLL±
/31Above ThresholdxYear-CLL + YZIP + Ei
We estimate this regression for each year between 1998 and 2005. We cannotinclude 2006 and 2007 in our estimates because the conforming loan limit did notchange after 2006 in our data (house prices dropped and the administration left thelimit unchanged).9 After we obtain 01, /2, and 03 for all 8 years (1998-2005), weestimate Fama-MacBeth averages (Fama and MacBeth, 1973) of these coefficientsand obtain the standard errors of this average by using the standard deviation ofthe estimated coefficients and dividing it by the square root of the number of coeffi-
8An alternative way to run our test would be to compare the year in which each limit is in effectwith the previous year, when all transactions in this range would be above the threshold for thatyear. The results for this alternative specification are reported in the Appendix.
9We do not run our analysis on the changes that were made to conforming loan limits in 2008 inhigh-cost areas as part of the Housing and Economic Recovery Act of 2008 for two reasons: First, thelimit was chosen by the government, as opposed to being mechanically related to previous limits, sothis introduces the possibility that the "jumbo-conforming" program was designed to assist specificareas and thus would be endogenous to expected future appreciation. Second, to the best of ourknowledge, there is no empirical evidence that the program had any discernible impact on the costof funding of mortgages that were made between the old limit of USD 417,000 and the new, higherlimits.
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 differ-
ences, 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 thatthe CLL is in effect. If selection was the explanation of the results, these transactionsshould be by "wealthy" borrowers driving the results. We find that the results do notchange 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 strategyis to show that the changes in the conforming loan limit have a significant effecton the financing choices of borrowers. In Figure 3-1 we can see the importance ofboth the 80 percent LTV rule, as well as the conforming loan limit, in determiningfinancing choices for the whole distribution of transactions. In Figure 3-2 we focus onthe groups of transactions that we include in the regressions. The first panel trackstransactions up to USD 10,000 below 125 percent of the conforming loan limit ineach year, whereas the second panel includes transactions up to USD 10,000 abovethe threshold. We show the total number of transactions (for all years between 1998and 2006) in each month during the year prior to the limit being in effect, in theyear that the limit is valid, and in the subsequent year. We also break down thetransactions by the choice of LTV - the transactions at the bottom of each panelhave an LTV below 75 percent, the second group includes transactions with an LTVbetween 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 mainmessage from Figure 3-2 is that in the year that the CLL is in effect, the compositionof financing choices by borrowers differs very significantly, with the 80 percent groupbecoming very prominent for the transactions below 125 percent of the CLL, whereasit is small for the transactions above the threshold. At the same time, the borrowerswho stick with a conforming loan and buy houses above 125 percent of the CLLbecome an important fraction of all borrowers (they have an LTV between 75 and79.5 percent).10 In the year after the limit is in effect, the choice of LTV across thetwo groups becomes indistinguishable.
In Table 3.3, we present the effect of the changes in the conforming loan limiton the financing choices made by the borrowers included in the sample of our mainregressions. In this table, we are verifying what we see in the pictures, namely thatborrowers on average end up with lower LTVs when they buy houses above the thresh-old of 125 percent of CLL. We find that LTVs are, on average, 0.3 to 0.7 percentagepoints lower for the group of transactions that happen above the threshold of 125percent of the CLL in the year that the limit is in effect. This effect is statisticallyand economically significant given how little variation there is in the modal choice ofLTV 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 fractionof borrowers with an LTV between 75 and 79.5 percent in the year before the CLL is in effect. Thisis because these transactions were not eligible for a conforming loan at an 80 percent LTV in theyear before the new limit was in effect and were, in general, just slightly above that threshold. Thisis thus a reflection of the same phenomenon we see for the group above 125 percent of the CLL inthe 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 con-
jecture 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 presentsFama-MacBeth coefficients from year-by-year regressions, as described before in Sec-tion 3.3.3. The coefficient of interest in Panel A of Table 3.5 is that on the interactionvariable, and it shows that houses above the threshold of CLL/0.8 transacted at avalue per square foot that was lower by about 1.16 dollars in the year that the CLLwas in effect. The results are stronger for the first half of the sample, where the pointestimate is -1.55 dollars per square foot for this set of transactions.
The other coefficients on the regressions for value per square foot are consistentwith what we know about house prices over this period. First, houses that are abovethe threshold of 125 percent of CLL (i.e. the more expensive houses in the regressionsample) are associated with a higher average value per square foot. In unreportedanalyses, we find that more expensive houses are generally associated with a highervalue per square foot (i.e. price rises quicker than house size in the whole distributionof 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 theUS. Given that the year in which the CLL is in effect is always the "pre" year in theregressions, we expect those transactions, on average, to be associated with a lowervalue per square foot.
In Panels B and C we use the residuals from the regressions we described in Section3.3.2 as the dependent variable to account for differences in quality between houses.The results are qualitatively and quantitatively very similar to the ones we presentin Panel A. In Panel B we are using the residuals of a regression of log of house priceon a set of characteristics, and we find a point estimate of -0.0017 that translates toresidual being lower by 620 dollars for houses above the threshold of 125 percent ofthe CLL when the CLL binds, considering an average transaction value of 371,340dollars. This suggests that transactions that cannot be financed at 80 percent withconforming loans are made at lower prices even after we control for a rich set of housecharacteristics. 11
Similarly in Panel C of Table 3.5, we confirm that even when we use the value persquare foot as a dependent variable but control for house quality, the interaction termis significant and economically large even though the point estimate of 0.65 dollarsfor houses above the threshold is slightly lower than the results in Panel A where wedo not adjust for house quality. The difference between the point estimate of 1.16dollars of Panel A and 0.65 dollars in this specification indicates that houses abovethe limit are of somewhat worse quality than those below the limit in the year thatthe limit is in effect.
We also show that the estimated effect of the conforming loan limit on house pricesis 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 ascontrols 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
household income each year from 2000 to 2007 from Melissa Data. 2 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
1 2 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 codeswith 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 betweenhouses above and below the threshold of CLL/0.8. Specifically, if more expensivehouses have, on average, lower house price growth from one year to the next relativeto less expensive houses, we might obtain the results reported in Table 3.5, but wemight also obtain similar results for samples with transactions above and below otherarbitrary thresholds.
In order to address whether the effect that we find is indeed the product of thetrue conforming loan limits and not due to different trends along the distributionof houses, we run the same regressions described in Section 3.3.3 for "placebo" loanlimits. We do this by shifting the true conforming loan limit in USD 10,000 stepsfrom the true value each year. We start at CLL-100,000 and move 20 steps until wereach CLL+100,000. For each of these 21 tests, we first define the "shift" relative tothe true conforming loan limits, and then we change the limits for all years by thatamount. For example, when we are changing all the limits by -20,000, this meansthat 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 runthe same year-by-year regressions and produce Fama-MacBeth coefficients for eachof the 20 alternative "placebo" values for the CLL. The results from this exercise areshown in Table 3.7.
The table shows that the coefficients of interest we obtain for all three dependentvariables (values per square foot, residuals from the transaction amounts, and residu-als of values per square foot) are systematically among the lowest of all obtained withthe 20 "placebo" trials (the ranking is given in the last two rows of the table). Thecoefficient on the value per square foot measure is the lowest of the 21 trials whetherwe use the whole sample, or whether we limit our attention a sample of transactionsthat all have an LTV between 0.5 and 0.8.13 When we use the whole sample andthe two residual measures from the hedonic regressions as the left-hand side variablesin the regressions, the coefficients for the true conforming loan limits are the secondand third lowest. In the restricted sample with LTVs between 0.5 and 0.8, these twomeasures produce the second lowest and the lowest coefficient out of the 21 trials. Ifwe limit our attention to placebo limits that are below the true limits (i.e. the tophalf 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 thoseregressions are, by construction, below the "eligibility" criteria of 125 percent of thetrue 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 thissample 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 statisticallysignificantly different from the average of the other coefficients at a 5 percent level inboth 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 coefficienthas 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 onesfrom 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 ourresults, i.e. whether they alone buy cheaper houses. As an aside, it is ex ante not clearwhy those borrowers would buy cheaper houses (based on value per square foot). Thefact that they are wealthier would usually lead us to believe that the omitted variablebias 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-differencesregressions excluding each of the two groups described above at a time (in the yearthat the limit is in effect) and construct Fama-MacBeth coefficients, as we did inTable 3.5. The results are shown in Table 3.8. We find that results do not changemuch 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 thesegroups of transactions. The statistical significance of the results is similar, and themagnitude of the coefficients sometimes is larger for one group and other times forthe other, depending on the left-hand side measure we use. Overall, the results pointin the same direction for both sets of regressions.
This robustness test shows that the effect of credit conditions on house prices inour 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 aconforming loan and end up with an LTV below 80 percent to be the ones driving ourmain result. The fact that we also see similar results when we exclude this subgroupincreases the likelihood of our alternative explanation, namely that differential cost ofcredit changes demand for housing, and that this shift in demand for housing drivesthe change in house valuation.
In the Appendix we show that our results are stable if we use a 5,000 dollarband around the threshold of CLL/0.8 instead of the 10,000, which suggests that thedifference in the cost of credit is likely to be similar for these two sets of buyers relativeto buyers below the threshold. This is further evidence that the result is not drivensolely by buyers who choose to obtain a conforming mortgage and put up additionalequity from other sources. Finally, we also show in the Appendix that the effect ofthe 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 subsequentyear has been defined by the administration.
Constraints to Housing Supply
To understand whether the effect of credit supply is amplified by the inability ofhousing supply to adjust quickly to demand, we divide zip codes into high and lowhouse supply elasticity according to the measure in Saiz (2010). If the supply ofhousing were perfectly elastic and able to adjust quickly to an increase in demandfor houses, the effect on prices should not be there. In this test, we find that theconstraint imposed by the conforming loan limit is stronger in zip codes located inmore 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 cross-
sectional 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 mag-
nitude 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 de-
nominator, 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 esti-
mates 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
14 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.15The paper by Kaufman (2012) obtains an estimate of 10 basis points by using a regression dis-
continuity 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.
133
to justify estimates above 10 without making very aggressive assumptions about thecost differential above and below the threshold.
The prior calculation is our preferred method of obtaining an estimate of the elas-ticity. However, we can obtain an alternative estimate of the elasticity by consideringborrowers who choose to obtain a conforming loan of less than 80 percent LTV abovethe threshold. This means they put up additional equity which either has to be fi-nanced through a third party loan or through savings. On average, given the rangeof transactions used in the regressions, these borrowers put up an additional USD5,000. If we assume that the cost of the additional equity is 5 percentage points ormore above the conforming mortgage rate, this is equivalent to a spread of 6-8 basispoints in the total cost of financing for these borrowers relative to those who buy ahouse 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 forthe spread of 5 percentage points over the conforming mortgage rate is not high if weconsider that many people use a jumbo loan even very close to the threshold of theCLL, indicating that the cost of additional equity is, at least for some borrowers, verysubstantial. The fact that we see borrowers stick with a conforming loan and put upadditional equity above the threshold may, in fact, be an indication that they are ex-cluded from the jumbo market altogether, rather than evidence that this is a cheaperoption. As Loutskina and Strahan (2009, 2011) show, jumbo loans are associatedwith more careful screening of borrowers, which may mean that many householdssimply could not use an 80 percent LTV above the threshold of 125 percent of theCLL even if they were looking to do so.
Another way of assessing the economic importance of the effect we find is by com-paring the dollar amount of savings through lower interest rates and the house pricedifferential. Assume a loan of USD 300,000, which is approximately the conformingloan limit midway through our sample (2002). If we use the upper end of the jumbo-conforming spread of 24 basis points, we calculate a cost difference of USD 720 inthe first year of the life of the loan. The present value of the cost difference over 30years is USD 8,557 assuming a 6 percent discount rate. If we use the lower end of thejumbo-conforming spread that has been estimated (10 basis points), this cost differ-ence is USD 3,604. Our estimated effect of the conforming loan is a price differenceof 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, foreach dollar of savings in the present value of interest costs, home values increase byabout 25-60 cents (always less than 1 dollar).
One possible concern with our estimation is that home buyers might expect theconforming loan limit to rise in the subsequent year and would thus refinance theirloan shortly after obtaining it. If refinancing were frictionless, buying a house abovethe threshold would cost 10-24 basis points more than the conforming loan rate foronly one year, because borrowers who took a jumbo loan would immediately refinanceinto a conforming loan in the following year (once the limit was raised). This wouldimply a very high elasticity of house prices to interest rates, as the difference inthe effective interest rate over the life of the loan paid by a borrower who took aconforming 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 byStanton (1995) finds that transaction costs for mortgage prepayment are around 30to 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 byDowning, 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 pro-
vided 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 conformingloan limit is also smaller in this time period. This suggests that those houses whichwere previously just out of reach of being financed by a conforming loan at 80 percentcould now be bid up in price since people had easier access to jumbo loans and otherforms of finance. The results are also stronger in ZIP codes with the lowest incomegrowth, 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 presentsa first test of the exogenous effect of cheaper mortgage loans on house prices. Weestimate an elasticity of house prices to interest rates that is below 10, implying thatthe drop in mortgage rates cannot account for the increase in house prices between2000 and 2006. However, we do show that those credit conditions matter for theformation of prices. Our results do not support a view that credit market conditionspurely respond to housing demand, but point instead to a directional effect that easiercredit supply leads to an increase in house prices.
136
3.6 Bibliography
Acharya, V., Richardson, M., Nieuwerburgh, S. V., and White, L. J. (2010) Guar-
anteed 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 Con-
forming 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) Macroeco-
nomics 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 Two-
Factor 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 Political Economy, 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 Mort-
gage 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 inHousing: The Downside of Diversification. Review of Financial Studies, 24(5), 1447-80.
Mayer, C. (2011) Housing Bubbles: A Survey. Annual Review of Economics,3:55977.
McKenzie, J.A. (2002) A Reconsideration of the Jumbo/Non-jumbo MortgageRate 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-MarketApproach. Quarterly Journal of Economics, Vol. 99(4), 729-52.
Saiz, A. (2010) The Geographic Determinants of Housing Supply. Quarterly Jour-nal of Economics, 125(3): 1253-1296.
Sherlund, S.M. (2008) The Jumbo-Conforming Spread: A Semiparametric Ap-proach. Finance and Economics Discussion Series Working Paper, 2008-01.
Stanton, R. (1995) Rational Prepayment and the Valuation of Mortgage-BackedSecurities Review of Financial Studies, Vol. 8, No. 3, 677-708.
Stein, J.C. (1995) Prices and Trading Volume in the Housing Market: A Modelwith 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
(b) 2004
I iLIL
Transaction Value LOWn Value
139
4500
4000
3500
2500
I'm
3500
1000
SM
Transaction Value Loan Value
SODO
40M
(A)CLI year-i
(B)CLL year
Li -02
Figure 3-2: Borrower Composition for the Regression Sample
Note: This figure shows the number of transactions by month for transactions within USD 10,000of 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
3000
12500
2000
1500
0
(C)CLL year+ 1
1 2 3 4 5 6 7 8 9 101112131415161718192021222324252627282930313233343536months
E LTV<75 m75<LTV<80 OLWV=80 (ILTV>80
(b) Transactions above 125 percent of CLL
3500
CLL year -1 CLLyear CLL year +13 0 0 0i
2500
2000
1500
500
1 2 3 4 5 6 7 8 9 1011121314Im1617181920212223242526272829303132333435
Months
140
EILTV<75 *75<LTV<80 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).
003
0D
030 $1 *@@ O
vfI %
-100R
-50 0 50Transaction Value as Percentage of 1.25CLL
100
141
CC
I0
C
Ez
0 %.met0406.
Figure 3-4: Share of Unused Mortgage Applications
Note: The horizontal axis indicates the difference between loan amounts and the conforming loanlimit as a percentage of the conforming loan limit. The share of unused mortgages is constructedfrom 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 figurerepresents the share of unused mortgages for each bin. We also plot third degree polynomials ( tothe 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 -a-'.
-50 -45
142
Soa
S
10US
0~
Ur0~
40 -35 -30 -25 -20 -15 -10 -5 0 5 10 15 20 25 30 35 40 45 50Distance from the conforming loan limit (%)
Table 3.1: Summary Statistics
Panel A. House Characteristics.All Transactions N=3,983,575 Regression Sample N=262,671
Mean Std. Dev. Median Mean Std. Dev. Median
Transaction Value (USD 1000) 308.52 123.93 286.00 371.34 54.92 380.00
Loan to value 0.81 0.15 0.80 0.76 0.13 0.80House Size (sqft) 1,735 672 1,592 1,935 701 1,816
Lot Size (sqft) 10,197 15,495 6,700 11,734 17,923 7,203
Number of rooms 6.84 1.60 7.00 7.23 1.61 7.00
Number of bedrooms 3.20 0.78 3.00 3.33 0.78 3.00
Number of bathrooms 1.93 1.03 2.00 2.11 1.07 2.00
House age (years) 35.40 27.70 34.00 34.74 27.40 34.00
Panel B. House Valuation.All Transactions N=3,983,575 Regression Sample N=262,671
Mean Std. Dev. Median Mean Std. Dev. Median
Value per sqft (USD/sqft) 193.59 91.60 172.03 219.63 93.37 200.20
Value per sqft residual (USD/sqft 0.00 42.30 -0.95 5.29 44.26 3.43
Log of transaction value residual (USD) 0.00 0.17 0.01 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 Transaction Value Value per sqft Loan to ValueN Obs Mean Std. Dev Mean Std. Dev Mean Std. Dev
Boston 279,261 320.29 112.40 197.67 73.81 0.78 0.16Chicago 377,031 262.41 108.15 174.37 68.63 0.81 0.15DC 396,211 329.95 126.16 186.97 85.93 0.82 0.14Denver 397,293 250.22 94.93 155.84 49.28 0.83 0.15Las Vegas 345,219 262.24 102.87 136.62 45.38 0.82 0.14Los Angeles 725,897 332.28 129.71 231.29 108.35 0.81 0.13Miarni 483,541 270.10 111.74 144.80 57.04 0.81 0.14New York 487,104 341.00 121.13 221.25 92.55 0.78 0.17San Diego 219,489 353.14 124.63 222.18 94.86 0.79 0.14San Francisco 272,529 383.59 123.74 266.47 109.26 0.79 0.13Total 3,983.575 308.52 123.93 193.59 91.60 0.81 0.15
Panel B. Distribution By Year and Thresholds
Year Thresholds Transaction Value Value per sqft Loan to ValueN Obs House Price Conf. Loan Mean Std. Dev Mean Std. Dev Mean Std. Dev
1998 134,200 283,938 227,150 239.78 102.07 133.84 50.59 0.81 0.151999 350,827 300,000 240,000 246.38 104.88 139.33 54.03 0.81 0.152000 354,071 315,875 252,700 257.67 109.21 149.65 61.64 0.81 0.162001 365,814 343,750 275,000 265.16 108.82 156.74 63.81 0.82 0.152002 397,527 375,875 300,700 283.79 114.34 171.06 71.85 0.81 0.152003 423,939 403,375 322,700 303.37 118.32 187.40 80.05 0.81 0.152004 525,407 417,125 333,700 331.81 121.20 212.65 90.51 0.79 0.142005 475,723 449,563 359,650 357.51 121.71 237.24 100.72 0.78 0.132006 376,182 521,250 417,000 366.27 121.89 247.02 105.50 0.79 0.132007 293,329 521,250 417,000 359.24 122.53 237.79 101.57 0.82 0.142008 286,556 521,250 417,000 325.11 119.84 206.92 91.62 0.84 0.15Total 3,983,575 308.52 123.93 193.59 91.60 0.81 0.15
Note: This table uses all the deed registry data on house transactions for 10 MSAs. Panel A showsthe 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
All years 1998-2001 2002-2005Above Threshold -0.004*** -0.006*** -0.002***
(0.001) (0.002) (0.001)Year CLL -0.008*** -0.005** -0.011***
(0.002) (0.002) (0.001)Above Threshold x -0.004*** -0.004* -0.003*Year CLL (0.001) (0.002) (0.002)
No. Obs. 242,753 100,870 141,883
Panel B: Log Loan Amount
All years 1998-2001 2002-2005Above Threshold 0.023*** 0.024*** 0.021***
(0.002) (0.003) (0.001)Year CLL -0.013*** -0.009*** -0.017***
(0.002) (0.003) (0.003)Above Threshold x -0.006** -0.007* -0.005Year CLL (0.002) (0.004) (0.003)
No. Obs. 242,753 100,870 141,883
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
All years 1998-2001 2002-2005Year CLL -0.003*** 0.000 -0.006***
(0.000) (0.001) (0.001)No. Obs. 262,671 109,496 153,175
Note: This table shows Fama MacBeth coefficients computed from year by year regressions that usea dummy variable for whether a transaction happens above the threshold of 125 percent of the CLLas the dependent variable. The sample includes all transactions within USD 10,000 of each year'sconforming loan limit, as well as transactions of the same amount in the subsequent year. Year CLLis the year in which the conforming loan limit is in effect. Zip Codes fixed effects are included oneach regression
146
Table 3.5: Effect of the CLL on House Valuation Measures
Panel A: Value Per Square Foot
All years 1998-2001 2002-2005
Above Threshold 1.261** 1.669*** 0.852(0.494) (0.573) (0.836)
Year CLL -22.869*** -14.851*** -30.886***(4.047) (2.314) (5.314)
Above Threshold x -1.162*** -1.553*** -0.771**Year CLL (0.264) (0.297) (0.369)
No. Obs. 262,671 109,496 153,175
Panel B: Log of Transaction Value Residual from Hedonic Regressions
All years 1998-2001 2002-2005
Above Threshold 0.0129*** 0.0154*** 0.0104***(0.0013) (0.0015) (0.0009)
Year CLL 0.0387*** 0.0356*** 0.0417***(0.0041) (0.0047) (0.0072)
Above Threshold x -0.0017** -0.0020 -0.0013***Year CLL (0.0008) (0.0015) (0.0004)
No. Obs. 251,431 103,535 147,896
Panel C: Value Per Square Foot Residual from Hedonic Regressions
All years 1998-2001 2002-2005
Above Threshold 1.733*** 2.060*** 1.407**(0.360) (0.425) (0.595)
Year CLL 4.103*** 3.935*** 4.270***(0.644) (0.495) (1.293)
Above Threshold x -0.651*** -0.940*** -0.362
Year CLL (0.238) (0.351) (0.291)
No. Obs. 251,764 103,709 148,055
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
2001-2005 2001-2005Above Threshold 0.731 0.601
(0.667) (0.638)Year CLL -28.869*** -29.364***
(4.706) (4.510)Above Threshold x -0.846*** -0.953***Year CLL (0.257) (0.210)Above Threshold x -1.548**Year CLL x Low Inc. Growth (0.652)No. Obs. 179,828 179,828
Panel B: Log of Transaction Value Residual from Hedonic Regressions
2001-2005 2001-2005Above Threshold 0.0109*** 0.0108***
(0.0008) (0.0009)Year CLL 0.0418*** 0.0439***
(0.0056) (0.0057)Above Threshold x -0.0016*** -0.0022***Year CLL (0.0003) (0.0006)Above Threshold x -0.0018Year CLL x Low Inc. Growth (0.0051)No. Obs. 173,347 173,347
Panel C: Value Per Square Foot Residual from Hedonic Regressions
2001-2005 2001-2005Above Threshold 1.396*** 1.347***
(0.453) (0.412)Year CLL 4.314*** 4.806***
(1.017) (1.072)Above Threshold x -0.504** -0.750***Year CLL (0.250) (0.158)Above Threshold x -0.319Year CLL x Low Inc. Growth (0.651)No. Obs. 173,550 173,550
Note: This table shows Fama MacBeth coefficients computed from year by year regressions thatuse 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'sconforming loan limit, as well as transactions in the same band in the subsequent year. Above theThreshold 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 percentLTV) and Year CLL is the year in which the conforming loan limit is in effect. This specificationinteracts the diff-in-diff specification with a dummy variable that uses changes in income at a zipcodelevel as proxy for good and bad times. Specifically, the dummy is 1 if the changes in the averagezipcode 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 to2001-2005
148
Table 3.7: Placebo Test for Coefficient of Interest
All Transactions 0.5<LTV<0.8 Transactions
Value Per Log of Value Per Value Per Log of Value Per
Square Foot Transaction Square Foot Square Foot Transaction Square Foot
Value Residual Value Residual
Residual Residual
True CLL -1.162 -0.002 -0.651 -1.257 -0.002 -0.931
Placebo 0.045 0.001 0.222 -0.107 0.000 0.110(0.467) (0.002) (0.494) (-0.107) (0.002) (0.440)
T-Statistic 2.586 1.206 1.770 2.626 1.009 2.365
CLL Rank 1 4 2 1 3 1
CLL Rank 1 2 1 1 1 1
below only
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 Keeping JumboAll years 1998-2001 2002-2005 All years 1998-2001 2002-2005
Above Threshold 0.939** 1.580*** 0.297 0.868* 1.530*** 0.207(0.472) (0.568) (0.666) (0.481) (0.545) (0.701)
Year CLL -24.539*** -15.953*** -33.126*** -24.874*** -16.040*** -33.708***(4.351) (2.564) (5.712) (4.454) (2.596) (5.813)
Above Threshold x -0.967** -1.314** -0.621 -2.177*** -2.618** -1.736**Year CLL (0.416) (0.572) (0.634) (0.639) (1.119) (0.724)No. Obs. 177,227 72,048 105,179 160,342 62,905 97,437
Panel B: Log of Transaction Value Residual from Hedonic Regressions
Keeping Conforming Keeping JumboAll years 1998-2001 2002-2005 All years 1998-2001 2002-2005
Above Threshold 0.0117*** 0.0145*** 0.0090*** 0.0119*** 0.0146*** 0.0091***(0.0014) (0.0018) (0.0007) (0.0013) (0.0016) (0.0008)
Year CLL 0.0367*** 0.0335*** 0.0398*** 0.0370*** 0.0337*** 0.0402***(0.0038) (0.0041) (0.0067) (0.0039) (0.0042) (0.0068)
Above Threshold x -0.0027** -0.0019 -0.0034*** 0.0004 -0.0020 0.0028**Year CLL (0.0011) (0.0022) (0.0009) (0.0015) (0.0025) (0.0012)No. Obs. 170,808 68,719 102,089 154,848 60,114 94,734
Panel C: Value Per Square Foot Residual from Hedonic Regressions
Keeping Conforming Keeping JumboAll 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**(0.290) (0.357) (0.414) (0.308) (0.333) (0.470)
Year CLL 3.514*** 3.485*** 3.543*** 3.529*** 3.552*** 3.507***(0.579) (0.431) (1.175) (0.573) (0.409) (1.168)
Above Threshold x -1.399*** -1.216** -1.583*** 0.225 -0.462 0.911**Year CLL (0.344) (0.535) (0.493) (0.418) (0.536) (0.464)No. Obs. 170,946 68,790 102,156 154,949 60,165 94,784
Note: This table shows Fama Macbeth coefficients computed from year by year regressions that usethree alternative measures of valuation as the dependent variable in each of the three panels. Thehedonic 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'sconforming loan limit. All year by year regressions include ZIP code fixed effects. We divide thetransactions that happen at a price above 125 percent of a year's CLL in the year that the limitis in effect into two groups: those with a conforming loan and those with a jumbo loan. We thenrun the same regressions including just one of these two groups at a time. The first three columnsinclude the transactions with a conforming loan and the last three columns include transactions witha jumbo loan. Above the Threshold refers to transactions up to USD 10,000 above the conformingloan limit divided by 0.8 (i.e. the transactions that were "ineligible" to be bought with a conformingloan at a full 80 percent LTV) and Year CLL is the year in which the conforming loan limit is ineffect.
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 xYear CLLAbove Threshold xYear CLL x Low Elasticity
No. Obs.
All All1.261** 1.221(0.494) (0.799)
-22.869*** -15.282***(4.047) (3.920)
-1.162*** -0.430(0.264) (0.831)
-0.870(0.977)
262,671 262,671
1998-2001 1998-20011.669*** 3.069***(0.573) (0.374)
-14.851*** -8.015***(2.314) (0.843)
-1.553*** -2.100**(0.297) (0.817)
0.726(1.332)
109,496 109,496
2002-2005 2002-20050.852 -0.628
(0.836) (0.749)-30.886*** -22.550***
(5.314) (5.981)-0.771** 1.239(0.369) (0.832)
-2.466**(0.992)
153,175 153,175
Panel B: Log of Transaction Value Residual from Hedonic Regressions
All0
(I
Above Threshold
Year CLL
Above Threshold xYear CLLAbove Threshold xYear CLL x Low Elasticity
No. Obs.
All.0129*** 0.0106***(0.0013) (0.0030)).0387*** 0.0263***(0.0041) (0.0037)0.0017** 0.0008(0.0008) (0.0022)
-0.0032(0.002)
251,431 251,431
10
0
998-2001 1998-2001 2002-2005 2002-2005.0154*** 0.0182*** 0.0104*** 0.0030*(0.0015) (0.0012) (0.0009) (0.0016).0356*** 0.0306*** 0.0417*** 0.0219***(0.0047) (0.0044) (0.0072) (0.0055)-0.0020 -0.0018 -0.0013*** 0.0033*(0.0015) (0.0037) (0.0004) (0.0018)
-0.0002 -0.0063***(0.004) (0.0020)
103,535 103,535 147,896 147,896
Panel C: Value Per Square Foot Residual from Hedonic Regressions
Above Threshold
Year CLL
Above Threshold xYear CLLAbove Threshold xYear CLL x Low ElasticityNo. Obs.
All1.733*** 1.(0.360) ((
4.103*** 1.(0.644) (C0.651*** -(0.238) (C
(2251,764 2C
All 1998-2001 1998-2001 2002-2005338** 2.060*** 2.623*** 1.407**.524) (0.425) (0.278) (0.595)
811** 3.935*** 3.316*** 4.270***.716) (0.495) (0.270) (1.293)).503 -0.940*** -1.620*** -0.362.546) (0.351) (0.306) (0.291)).241).740)1,764
0.843(0.744)
103,709 103,709 148,055
2002-20050.054
(0.319)0.305(0.898)0.615
(0.684)-1.325(1.104)148,055
151
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
-
Table 3.10: Elasticity Estimates
A House Prices in bpMax: 91.2Min: 29.7
Jumbo-Conforming SpreadMin (10 bp) Max (24 bp)
9.1 3.83.0 1.2
Note: This table shows elasticity calculations for different scenarios of both the house price increaseestimated in the regressions and the interest rate differential implied for transactions above andbelow the threshold of 125 percent of the conforming loan limit. We use the jumbo-conformingspread in interest rates as the denominator in the elasticity calculation.
152
3.7 Appendix A. Robustness and Refinements -Additional Tests
3.7.1 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. Bylimiting 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 re-
gressions, 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 quantita-
tively 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 inter-
pretation 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 inwhich the limit is in effect with the subsequent year, we compare it to the previousyear. In this way, we are comparing houses that are never eligible for an 80 percentconforming loan (those above the threshold) to transactions that initially are noteligible, but become eligible once the limit changes. The research design is the sameas before, but we shift the window of analysis back one year. Table 3.13 shows theFama-MacBeth coefficients for this specification. The point estimates are smaller thanthe ones in Table 3.13, but they are in the same direction and remain statisticallysignificant 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 oraround October of each year, which might mean that the anticipation of a raise ofthe 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 thecoefficients are being driven by this time period. Table 3.15 shows the results for thisspecification, and we see that the estimates for the effect are the same for the lastthree months of the year as they are for the first nine. The main effect is almostunchanged.
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 effectof the conforming loan limit on the distribution of value per square foot on the wholesample of transactions. We plot the average value per square foot as a function of thedistance of each transaction to the threshold of 125 percent of the CLL. We can seethat for the ZIP codes in the lowest quartile of the income distribution, the averagevalue per square foot is monotonically increasing for up to conforming loan limitthreshold, and from this point onwards the distribution becomes flat. This patternis not visible for zip codes with higher median incomes, where the distribution seemsmonotonically increasing both below and above the threshold.
154
3.8 Appendix B. Data Manipulation
3.8.1 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
CriterionInitial dataTransaction value equal to zeroMissing zipcodeMissing square feetMislabeled yearFirst loan greater than transaction valueHouse of less than 500 square feetTransaction greater than 1,2 MM and smaller than 30 M
Company owned observation based on Dataquick flagCompany owned obs based on owner/seller/buyer informationSimple duplicated transactionsValue per square feet yearly outliersSame property, date and buyer/seller informationSame property, and (late and no seller informationSame property, (late and transaction valueSame property, date and A sell to B and B sell to CSpecial Transaction, based on Dataquick flagSame property and (late, multiple sales in a dayClean dataIfemove single-family housesTransaction greater than 600 M and smaller than 130 MWhole sample for hedonic regressionsTransactions outside the 10k band for each yearTransactions used twice ( treatment in year t and control in
year t+1Regression sample
Deleted Observations Remaining Observations11,884,730
1,365,429 10,519,30118,766 10,500,535
1,509,732 8,990,8035 8,990,798
353,552 8,637,24647,059 8,590,187381,786 8,208,401451,295 7,757,106746,754 7,010,352
0 7,010,352142,079 6,868,27311,577 6,856,696
364 6,856,33241,855 6,814,47722,258 6,792,219
609 6,791,610248 6,791,362
6,791,3621,751,670 5,039,6921,056,117 3,983,575
3,983,5753,742,840 240,735+21,936 262,671
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 con-
struction. 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 asrecords with transaction values smaller than three thousand and greater thanone million and two hundred thousand dollars.
- Value per square foot outliers per year: We drop observations that are abovethe ninety-ninth percentile for the value per square foot variable or below thefirst percentile each year.
Company owned observations
- We drop observations that Dataquick identifies as being bought by a corpora-tion.
- Company owned observations based on owner/seller/buyer information: If theowner, seller, or buyer names contain LLC, CORP, or LTD, the observation isremoved from the sample.
Duplicate transactions
Simple duplicated transactions: Remove records for which all the property in-formation is the same.
Same property, date, and buyer/seller information: Drop observations that areduplicated based on transaction value, date, and buyer/seller information.
Same property and date, no seller information: Drop observations for which theproperty unique identifier and date are the same and have no seller information.
Same property, date, and transaction value: Drop observations for which prop-erty 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 toB 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 identifyrecords that are not actual transactions. For example, if a transaction was onlyan ownership transfer without a cash transfer, this field is populated, allowingus to delete this transaction.
Same property and date, multiple sales in a day: If a property is sold more thantwice during the same day, we keep only one transaction.
Additional information
We merge the Metropolitan Statistical Area (MSA) classification obtained fromthe Census Bureau definition, using FIPS unique code identifier by county16 .
16FIPS county code is a five-digit Federal Information Processing Standard (FIPS) code whichuniquely 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 threeare 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 re-
gressions. 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 inCampbell, 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 dum-mies), 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 fire-place and parking, indicators for style of building (architectural style and structuralstyle), and additional indicators for type of construction, exterior material, heatingand cooling, heating and cooling mechanism, type of roof, view, attic, basement, andgarage.While interior square feet, lot size, and age are included as continuous variables, allthe other controls are included as indicator variables.
Type of house: This variable is 1 if the house is a single-family house and 0 ifit is a condo or a multifamily property.
Bedrooms: This characteristic is divided into four categories (dummies): onebedroom, 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, Con-struction Code, Roof Code, View Code: These characteristics were divided basedon the numeric categorization of the original field. For example, constructioncode was divided into 10 different categories that indicated the material usedon the framework of the building. In this case, we created 10 dummies basedon this categorization.
Heating and cooling: This information was divided into four categories: onlyheating, only cooling, both heating and cooling, and heating-cooling informa-tion missing. The last variable was created to avoid dropping transactions forwhich the information was not available.
Heating and cooling type: These characteristics were divided based on the nu-meric categorization of the original field. In this case, they discriminate thetype 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 thathave garage surface greater than 50 square feet. For those transactions withoutthe information, a missing dummy is created for this category. Finally, we usedadditional information to create a dummy that indicates if the houses have agarage carport or not.
- Renovation: This variable accounts for the number of years since the last reno-vation. Based on this continuous variable, five categories (dummies) are defined:missing renovation if the renovation date is missing or renovation period is neg-ative, 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 equalto 30 years.
Attic: This characteristic is accounted for using a dummy for houses with anattic greater than 50 square feet, and another dummy to account for missinginformation about the attic in the houses.
Basement Finished and Unfinished: For the finished basement information, wecreated a dummy for houses with basement size greater than 100 square feet, andanother dummy to account for missing information about the finished basement.The same procedure is used to incorporate the information about unfinishedbasement.
We use both the price of a transaction as well as the value per square foot as ourdependent variables. By estimating these regressions by year and by MetropolitanStatistical Areas (MSA), we allow the coefficients on the characteristics to vary alongthese two dimensions. We included monthly indicator variables to account for sea-sonality in the housing market, as well as zip code fixed effects. The set of controlsXi is composed of all the variables described above, but in the case of the value persquare 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 bedroomsor bathrooms, or information on a houses age, we do not include this observation inthe hedonic regressions. This explains the difference between the number of obser-vations 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.
00-
CD
0
XtD -
0
U--
0 -'-
1998 1999 2000 2001 2002 2003 2004 2005
whole sample restricted sample
2006
160
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 codesin the lowest income quartile and 4 includes the ZIP codes with the highest median income. We usethe 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 thresholdof 125 percent of the conforming loan limit (e.g. if the threshold is 200,000, a transaction of 150,000will appear as -25 percent). The vertical red line is the threshold and the transactions for all yearsare centered around that value.
1 2
700-l-'04
4
7100 - 0
70/so 16o -ibo -50
Transaction Value as Percentage of 1.25CLL
161
0
LL..0.3
S
0
0
-1 0 !b 10
-04
woprl"- --
A- OR A"- -W V - -- W4PW *W
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.
I I 1 i I I r -
-20 -15 -10 -5 0 5 10 15 20Distance from the conforming loan limit (%)
25 30 35 40 45 50
162
~0C
0
E
'C
Ok
-50 -45 -40 -35 -30 -25
6; O
Table 3.12: Effect of the CLL on House Valuation Measures, Constrained Sample
(0.5<LTV<0.8)
Panel A: Value Per Square Foot
All years 1998-2001 2002-2005
Above Threshold 0.956** 1.584*** 0.328(0.462) (0.556) (0.650)
Year CLL -24.627*** -15.935*** -33.319***(4.386) (2.576) (5.726)
Above Threshold x -1.257*** -1.610** -0.904
Year CLL (0.422) (0.646) (0.576)
No. Obs. 190,450 75,304 115,146
Panel B: Log of Transaction Value Residual from Hedonic Regressions
All years 1998-2001 2002-2005
Above Threshold 0.0118*** 0.0145*** 0.0090***(0.0014) (0.0017) (0.0007)
Year CLL 0.0367*** 0.0335*** 0.0398***(0.0038) (0.0040) (0.0066)
Above Threshold x -0.0017 -0.0019 -0.0015*Year CLL (0.0011) (0.0022) (0.0008)
No. Obs. 183,643 71,843 111,800
Panel C: Value Per Square Foot Residual from Hedonic Regressions
All years 1998-2001 2002-2005
Above Threshold 1.565*** 1.958*** 1.172***(0.298) (0.356) (0.431)
Year CLL 3.431*** 3.470*** 3.392***(0.550) (0.417) (1.113)
Above Threshold x -0.931*** -1.085*** -0.777**
Year CLL (0.260) (0.413) (0.360)
No. Obs. 183,789 71,917 111,872
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 by0.8 (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 ControlGroup
Panel A: Value Per Square Foot
All Transactions 0.5<LTV<0.8 TransactionsAll years 1999-2002 2003-2006 All years 1999-2002 2003-2006
Below Threshold 0.012 -0.005 0.029 0.522* 0.628 0.417(0.236) (0.282) (0.423) (0.270) (0.412) (0.404)
Pre-Year CLL -23.739*** -15.890*** -31.588*** -25.061*** -16.995*** -33.127***(4.391) (2.489) (6.534) (4.636) (2.666) (7.057)
Below Threshold X -0.375 -0.817 0.068 -0.555 -0.812*** -0.298Pre-Year CLL (0.473) (0.549) (0.783) (0.434) (0.233) (0.884)No. Obs. 227,325 93,612 133,713 168,865 66,072 102,793
Panel B: Transaction Value Residual from Hedonic Regressions
All Transactions 0.5<LTV<0.8 TransactionsAll years 1999-2002 2003-2006 All years 1999-2002 2003-2006
Below Threshold . -0.0099*** -0.0106*** -0.0092*** -0.0086*** -0.0087*** -0.0085***(0.0010) (0.0010) (0.0017) (0.0011) (0.0018) (0.0015)
Pre-Year CLL 0.0346*** 0.0342*** 0.0350*** 0.0342*** 0.0334*** 0.0350***(0.0045) (0.0037) (0.0089) (0.0045) (0.0042) (0.0088)
Below Threshold X 0.0000 -0.0019 0.0019 -0.0011 -0.0031 0.0009Pre-Year CLL (0.0016) (0.0021) (0.0023) (0.0016) (0.0023) (0.0020)No. Obs. 217,410 88,416 128,994 162,584 62,897 99,687
Panel C: Value Per Square Foot Residual from Hedonic Regressions
All Transactions 0.5<LTV<0.8 TransactionsAll years 1999-2002 2003-2006 All years 1999-2002 2003-2006
Below Threshold -0.903*** -0.881*** -0.925 -0.524** -0.446** -0.603(0.289) (0.197) (0.593) (0.208) (0.206) (0.395)
Pre-Year CLL 3.215*** 3.019*** 3.411** 2.852*** 2.591*** 3.112**(0.712) (0.529) (1.436) (0.699) (0.547) (1.392)
Below Threshold X -0.175 -0.605** 0.256 -0.467 -0.915*** -0.020Pre-Year CLL (0.351) (0.245) (0.625) (0.315) (0.130) (0.560)No. Obs. 217.804 88,613 129,191 162,788 62,997 99,791
Note: Table shows Fama McBeth coefficients computed from year by year regressions that usethree alternative measures of valuation as the dependent variable in each of the three panels. Thesample includes all transactions within USD 10,000 of each year's conforming loan limit, as wellas transactions of the same amount in the previous year (unlike the previous tables where we usethe subsequent year). In this table we include the results for all transactions, as well as those forthe sample that is restricted to having an LTV between 0.5 and 0.8. Below the Threshold refers totransactions up to USD 10,000 below the conforming loan limit at year t divided by 0.8 (i.e. thetransactions that were "eligible" to be bought with a conforming loan at a full 80 percent LTV in yeart , but were "ineligible" in year t-1) and Pre-Year CLL is the previous year in which the conformingloan limit is in effect. This specification makes the interaction coefficient directly comparable to themain regression on signs and magnitudes.
164
Table 3.14: Effect of the CLL on Valuation - Alternative Bands
Panel A: Value Per Square Foot
10K Ok to 5K 5K to 1OKAbove Threshold 1.261** 0.969 1.406***
(0.494) (0.722) (0.544)Year CLL -22.869*** -23.008*** -23.194***
(4.047) (3.988) (4.177)Above Threshold x -1.162*** -1.064* -1.181**Year CLL (0.264) (0.556) (0.581)No. Obs. 262,671 134,117 128,554
Panel B: Log of Transaction Value Residual from Hedonic Regressions
10K Ok to 5K 5K to 10K
Above Threshold 0.0129** 0.0071 0.0180***(0.0013) (0.0019) (0.0013)
Year CLL 0.0387*** 0.0384*** 0.0389***(0.0041) (0.0045) (0.0038)
Above Threshold x -0.0017*** -0.0015* -0.0023**Year CLL (0.0008) (0.0011) (0.0016)No. Obs. 251,431 128,429 123,002
Panel C: Value Per Square Foot Residual from Hedonic Regressions
10K Ok to 5K 5K to 1OK
Above Threshold 1.733*** 1.255* 2.110***(0.360) (0.700) (0.387)
Year CLL 4.103*** 4.052*** 3.946***(0.644) (0.678) (0.763)
Above Threshold x -0.651*** -0.712 -0.623***Year CLL (0.238) (0.508) (0.238)No. Obs. 251,764 128,601 123,163
Note: This table shows Fama MacBeth coefficients computed from year by year regressions thatuse 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 Section3.3.2. The sample for each year's regression includes all transactions within +/- USD 10,000 of thatyear's conforming loan limit, as well as transactions in the same band in the subsequent year. Allyear by year regressions include ZIP code fixed effects. Above the Threshold refers to transactionsup 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 yearin 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
1998-2005 1998-2005Above Threshold 1.261** 1.039*0.000 (0.625) (0.531)Year CLL -22.869*** -23.460***0.000 (5.119) (5.079)Above Threshold x -1.162*** -1.086***Year CLL (0.334) (0.393)Above Threshold x -0.213Year CLL x Post October (1.031)No. Obs. 262,671 262,671
Panel B: Log of Transaction Value Residual from Hedonic Regressions
1998-2005 1998-2005Above Threshold 0.0129*** 0.0132***
(0.0016) (0.0014)Year CLL 0.0387*** 0.0398***
(0.0052) (0.0056)Above Threshold x -0.0017* -0.0027**Year CLL (0.0010) (0.0013)Above Threshold x 0.0033Year CLL x Post October (0.0027)No. Obs. 251,431 251,431
Panel C: Value Per Square Foot Residual from Hedonic Regressions
1998-2005 1998-2005Above Threshold 1.733*** 1.751***
(0.456) (0.407)Year CLL 4.103*** 4.176***
(0.815) (0.813)Above Threshold x -0.651** -0.696**Year CLL (0.301) (0.277)Above Threshold x 0.031Year CLL x Post October (0.805)No. Obs. 251,764 251,764
Note: This table shows Fama MacBeth coefficients computed from year by year regressions thatuse 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'sconforming loan limit, as well as transactions in the same band in the subsequent year. Above theThreshold 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 percentLTV) and Year CLL is the year in which the conforming loan limit is in effect. This specificationinteracts the diff-in-diff specification with a dummy variable that is 1 in October, November andDecember of each year.
166
Table 3.16: Effect of the CLL on House Valuation with In-Sample Controls
Panel A: Value Per Square Foot
All years 1998-2001 2002-2005Above Threshold 2.926*** 3.272*** 2.581***
(0.366) (0.416) (0.612)Year CLL -15.158*** -9.681*** -20.634***
(2.706) (1.206) (3.567)Above Threshold x -0.771** -1.211*** -0.332Year CLL (0.299) (0.428) (0.327)No. Obs. 251,764 103,709 148,055
Panel B: Log of Transaction Value
All years 1998-2001 2002-2005Above Threshold 0.0281*** 0.0323*** 0.0239***
(0.0018) (0.0011) (0.0011)Year CLL -0.0004*** -0.0005*** -0.0004***
(0.0001) (0.0001) (0.0001)Above Threshold x 0.0000 -0.0001 0.0001Year CLL (0.0000) (0.0001) (0.0001)No. Obs. 251,431 103,535 147,896
Note: This table shows Fama MacBeth coefficients computed from year by year regressions that usetwo alternative measures of valuation as the dependent variable in each of the two panels. Insteadof using residuals from a hedonic regression, all characteristics of the houses are included as controlswithin 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 thesubsequent year. All year by year regressions include ZIP code fixed effects. Above the Thresholdrefers 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