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