Borrower protection and the supply of credit: Evidence from foreclosure laws

Accepted Manuscript

Borrower protection and the supply of credit: Evidence from foreclosure laws Jihad Dagher, Yangfan Sun PII: DOI: Reference:

S0304-405X(16)30003-4 10.1016/j.jfineco.2016.02.003 FINEC 2628

To appear in:

Journal of Financial Economics

Received date: Revised date: Accepted date:

27 July 2014 9 July 2015 6 August 2015

Please cite this article as: Jihad Dagher, Yangfan Sun, Borrower protection and the supply of credit: Evidence from foreclosure laws, Journal of Financial Economics (2016), doi: 10.1016/j.jfineco.2016.02.003

This is a PDF file of an unedited manuscript that has been accepted for publication. As a service to our customers we are providing this early version of the manuscript. The manuscript will undergo copyediting, typesetting, and review of the resulting proof before it is published in its final form. Please note that during the production process errors may be discovered which could affect the content, and all legal disclaimers that apply to the journal pertain.

ACCEPTED MANUSCRIPT

Borrower protection and the supply of credit: Evidence from foreclosure laws ✩ Jihad Dagher

Research Department, International Monetary Fund, 700 19th St NW, Washington, DC 20431, USA

Yangfan Sun b

b

CR IP T

a

a1

Department of Economics, Georgetown University, 3700 O St NW, Washington, DC 20057, USA

Abstract

AN US

Laws governing the foreclosure process can have direct consequences for the costs of foreclosure and, therefore could affect lending decisions. We exploit the heterogeneity in judicial requirements across US states to examine their impact on banks’ lending decisions in a sample of urban areas straddling state borders. A key feature of our study is the way it exploits an exogenous cutoff in loan eligibility to government-sponsored enterprises (GSEs) guarantees, which shift the burden of foreclosure costs onto the GSEs. We find that judicial requirements reduce the supply of credit

M

only for jumbo loans, which are ineligible for GSE guarantees, i.e., in the nonsubsidized segment of the market. Thus, while we find a significant effect on credit supply, the aggregate impact is

ED

muted by the indirect cross-subsidy by the GSEs to borrower-friendly states. JEL classification: D18, G21, G28

PT

Keywords: Borrower protection, Foreclosure laws, Credit supply, Regulation, GSEs.

AC

CE

✩ We gratefully acknowledge the helpful comments from an anonymous referee, Paul Calem, Stijn Claessens, Giovanni Dell’Ariccia, Luc Laeven, William Lang, participants at the International Banking, Economics, and Finance meeting, International Monetary Fund Research seminar, Federal Housing Finance Agency seminar, and the Financial Innovations and Bank Regulation Conference. This paper’s findings, interpretations, and conclusions are entirely ours and do not represent the views of the International Monetary Fund, its executive directors, or the countries they represent. 1 Corresponding author, e-mail: [email protected]

Preprint submitted to Journal of Financial Economics

February 13, 2016

ACCEPTED MANUSCRIPT

1. Introduction The foreclosure crisis and the ensuing regulatory reforms of the mortgage market have revived an old debate surrounding the trade-off between borrower protection and credit access.1 The importance of this issue also lies in the fact that regulations are persistent and thus their cumulative

CR IP T

costs could be considerable. Unfortunately, the impact of borrower protection laws on credit supply remains relatively understudied in the empirical literature.

In this paper, we investigate the impact of foreclosure laws on lending decisions by banks. Foreclosure laws govern the process through which creditors can take repossession of real estate following a default by the borrower. These laws have important implications to the foreclosure process, its duration, and the associated costs and risks to borrowers and creditors. Because

AN US

borrower-friendly foreclosure laws typically impose additional costs to creditors in the event of default and could increase borrowers’ incentive to default, one wonders whether and to what extent they also reduce the supply of credit. Our main focus is on judicial foreclosure laws, i.e., whether states require that the foreclosure process be handled by the court system. Judicial procedures are more costly than power-of-sale alternatives, as they are more time consuming and require the use of legal professionals [see, e.g., Clauretie and Herzog (1990) and Schill (1991)].

2

M

The foreclosure crisis has highlighted the stark differences in the foreclosure timelines between so-called judicial and nonjudicial states and their implication to foreclosure levels, house prices,

ED

and the economy, as recently shown in Mian, Sufi, and Trebbi (2015). To provide a convincing answer to our main question, one has to be able to trace the impact of the laws on the supply of credit, controlling for potentially confounding factors. We, therefore,

PT

design an empirical strategy that takes advantage of a quasi-experimental setting. We exploit two sources of exogenous variation. The first is the variation in foreclosure laws across state borders. The second variation is the discontinuity in loans’ eligibility to guarantees from government-sponsored

CE

enterprises (GSEs), guarantees that shift the financial burden of foreclosure costs onto the GSEs. The role of the GSEs cannot be ignored on the mortgage market. The GSEs guarantee a large

AC

share of mortgages in exchange for guarantee fees that are uniform across states. Whether a 1 See, e.g., Evans and Wright (2009), US House of Representatives (2012), Bipartisan Policy Center (2013), and Kupiec (2014). 2 These losses typically include foregone interest, attorney fees, court costs, property taxes, repairs, hazard insurance, and other indirect costs.

2

ACCEPTED MANUSCRIPT

loan is conforming, i.e., eligible for GSE guarantees, is determined by an exogenous regulatory loan-limit cutoff set by their regulator. Ignoring the role of GSEs could lead one to underestimate or misinterpret the effects of foreclosure laws. The use of this exogenous cutoff also allows us to considerably sharpen our empirical strategy to further address concerns related to unobservable factors.

CR IP T

While the costs associated with judicial foreclosure are directly related to the longer foreclosure timelines, the extant literature has so far relied on a binary variable to capture this difference. One of the contributions of this paper is that it brings in more direct and continuous measures of differences in foreclosure time frames. Our favorite variable is based on data collected by the US Foreclosure Network (USFN), which, through its legal expertise, provides an estimate of the number of days it takes to foreclose on a property solely based on state laws.

AN US

Using a comprehensive data set on mortgage lending in the US, we study the impact of foreclosure laws on banks’ probability of rejecting a loan application in urban areas straddling state borders while exploring the variation between conforming and non-conforming (jumbo) loans. One of the advantages of a loan-level analysis is that it allows us to study the decision by banks, which is more directly linked to the supply side of the market [see, e.g., Loutskina and Strahan (2009)], and to control for bank fixed effects. We show that the banking sector is highlighted in the literature.

M

heterogeneous across borders even within an urban area, an issue that has not been sufficiently

ED

Our findings point to a significant and economically meaningful impact of judicial foreclosure laws on the supply of credit. The aggregate effect of these laws on credit supply is muted, however, due to the GSE cross-subsidization of borrower-friendly states. We find that judicial

PT

foreclosure laws are associated with a significant increase in the relative rejection rate on jumbo loans. Their impact on the rejection rate of conforming loans is weak and overall not statistically

CE

significant. These results are in line with our hypothesis, as the foreclosure costs on GSE-securitized conforming loans are borne by the GSEs. While the supply of credit is unevenly affected by foreclosure laws around the jumbo cutoff, we show that the demand for loans does not exhibit

AC

such variation. The relative number, volume, and characteristics of jumbo loans in comparison with conforming loans do not correlate with judicial foreclosure laws across borders. All these results hold for either the standard binary measure used in the literature, or the foreclosure time

3

ACCEPTED MANUSCRIPT

frames we obtain from authoritative sources on state foreclosure requirements.3 We subject our findings to a battery of robustness and falsification tests (included in an Online Appendix). We find that they help strengthen our results and support our interpretation. While the literature on foreclosure laws is extensive, the impact of foreclosure laws on mortgage lending has received limited attention. An important exception is Pence (2006), which offers a

CR IP T

rigorous treatment of the subject using Home Mortgage Disclosure Act (HMDA) data focusing on urban areas that straddle state borders. Pence (2006) studies the impact of these laws on the size of the loan and finds that loan sizes are 3% to 7% smaller in defaulter-friendly states. Our paper differs along several important aspects. Key to our empirical analysis is the focus on isolating supply side factors and the distinction in loans’ eligibility to GSE guarantees based on the exogenous jumbo cutoff. Our analysis also differs in that we focus on the decision by banks to

AN US

reject a loan and in that we control for the heterogeneity in the banking landscape across borders. We also make use of new data that provide a more direct measure of the costs associated with judicial foreclosures.

A related strand of literature has examined the impact of foreclosure laws on bank losses, borrower behavior, and foreclosure rates. For example, Clauretie and Herzog (1990) find that judicial foreclosure and the right of redemption increase the cost of foreclosure. Ghent and

M

Kudlyak (2011) find that recourse laws lower the sensitivity of default to negative equity. The literature also points to evidence that longer foreclosure duration, associated with judicial requirements,

ED

increases the incentive to default [see, e.g., Zhu and Pace (2011), Gerardi, Lambie-Hanson, and Willen (2013), and Calem, Jagtiani, and Lang (2014)]. These findings provide an additional channel through which judicial foreclosure can affect the supply of credit. Mian, Sufi, and

PT

Trebbi (2015) study the impact of judicial foreclosure on the incidence of foreclosure and use this as an instrumental variable to find that foreclosures lead to a large decline in house prices.

CE

A related strand of literature studies the impact of bankruptcy law on credit supply [see, e.g., Gropp, Scholz, and White (1997), Berkowitz and White (2004), and Goodman and Levitin (2014)]. Our paper is also related to the broader literature that studies the impact of state laws on credit

AC

markets [see, e.g., Huang (2008) and Favara and Imbs (2015)].

3 All our regressions control for two other weaker variations in foreclosure laws (discussed in Section 2) related to deficiency judgments and right of redemption, but we do not highlight them due to lack of sufficient variation in the cross-border sample.

4

ACCEPTED MANUSCRIPT

The paper is organized as follows. Section 2 briefly reviews state foreclosure laws in the United States. Section 3 presents the empirical strategy, data, and results. Section 4 concludes. 2. State foreclosure laws

CR IP T

When a borrower defaults on the payment of the debt, she has agreed to repay according to the terms of the mortgage. The loan documents usually contain an acceleration clause, which causes the entire indebtedness to become due and payable. The foreclosure process is the procedure through which a mortgaged property is repossessed by the creditor following default by the borrower and a failure to repay the outstanding debt. In the United States, the foreclosure process is mainly regulated at the state level. It provides protections to borrowers against the harsh

AN US

consequences of default and eviction, although with varying degrees across states.

A key characteristic of state foreclosure laws relates to whether the foreclosure process is handled by the court system. In states with judicial foreclosure, the lender initiates the foreclosure process with the court by bringing evidence of default on the mortgage. The court then issues a decree, stipulates what notices should be provided, and oversees the various steps of the procedure. In nonjudicial states, the liquidation of the property takes place relatively quickly

M

through a power-of-sale and is handled by a trustee. While the judicial process protects borrowers against potentially unfair practices, it imposes costs on the lender or insurer due to the delays

ED

associated with the process and the required legal resources. We follow recent work by Mian, Sufi, and Trebbi (2015) and classify states as judicial or nonjudicial based on publicly available data from RealtyTrac, a leading foreclosure data provider.4

PT

Fig. 1 shows a map of the US mainland in which the 20 states that require a judicial process are shaded in dark gray. The figure shows that judicial foreclosure states are concentrated in the eastern part of the United States.

CE

While the extant literature relies solely on a binary measure of judicial foreclosure laws, this has some well-known disadvantages. First, the categorization of states between judicial

AC

and nonjudicial is not straightforward in all cases, which explains some of the inconsistencies observed across papers on the topic. Second, as foreclosure experts and attorneys are well aware, 4 See

http://www.realtytrac.com/foreclosure-laws/foreclosure-laws-comparison.asp.

5

CR IP T

ACCEPTED MANUSCRIPT

AN US

Fig. 1. States with judicial requirement. This map of the US mainland indicates in dark gray the states that require judicial foreclosure based on RealtyTrac.

substantial variation exists in the processing timelines of foreclosures even within the groups of judicial and nonjudicial states. That is, some states impose lengthier procedures by law. To the best of our knowledge, this paper is the first to address these issues by bringing in data that capture this heterogeneity more accurately. We exploit data collected by USFN, which,

M

through its legal expertise, provides an estimate of the number of days it takes to foreclose on a property solely based on regulations, abstracting from additional operational delays. These estimates were constant between 2001 and 2006.

An index of the timelines (scaled by the

ED

longest timelines) is shown in Fig. 2, with red bars representing states that are categorized as being judicial.5 A strong correlation exists between judicial foreclosure and USFN timelines and substantial variation is evident within groups, indicating that one can improve on the dummy

PT

judicial variable by using these timelines. In the Online Appendix we supplement the USFN measure with a measure of processing timelines produced by the Federal Home Loan Mortgage

CE

Corporation (Freddie Mac).

Two other significant variations in foreclosure laws are usually considered in the literature. The first relates to the ability of the lender to pursue the borrowers’ other assets when the

AC

borrower’s debt exceeds the proceeds from the foreclosure sale. These are typically referred to as recourse laws. In the US, eight states allow the lender to pursue a deficiency judgment 5 The

timelines in days as reported in RealtyTrac are shown in Table A2 of the Online Appendix.

6

ACCEPTED MANUSCRIPT

1

0.6

0.4

0

T GX A TN V MA D D C N H M MI WO Y A CL T R AI MR WA V A MZ MS N VT A NK C N V C A KS F WL A N E U CT O KY ID M NT OD R SC SD IA L NA M O K D OE H H MI E IN N PAJ W I I NL Y

0.2

CR IP T

USFN Index

0.8

Requires judicial process

AN US

Does not require judicial process

Fig. 2. US Foreclosure Network timelines index. This figure plots the foreclosure timelines index we use in our regressions. The timelines are measured by the USFN, based on states’ regulatory requirements, which are shown in the Online Appendix. The red (green) bars indicate that the state does (does not) require judicial foreclosure.

M

against the borrower’s other assets (i.e., allow recourse). While such action is rarely pursued by the lender due to inherent costs and the availability of insurance on most mortgages, the lender

ED

has more power in the negotiation process, allowing her to obtain concessions from the borrower. The second variation relates to the borrower’s ability to regain the property after the foreclosure sale for the amount of the delinquent payments and the costs of foreclosure. In the nine states

PT

that give borrowers a statutory right of redemption, borrowers can legally regain the property within a period that typically ranges between three months and a year. Such a borrower-friendly statute also imposes uncertainty and costs on the lender.

CE

A recent in-depth historical analysis by Ghent (2013) supports the exogeneity of these foreclosure laws to current economic conditions. Ghent finds that with the exception of anti-deficiency

AC

statutes “most differences in state mortgage law date from the nineteenth century and in many cases from before the US Civil War. (p. 22)” Ghent shows that whether states require judicial foreclosure or make available nonjudicial options can be traced to idiosyncratic factors and individual decisions by judges instead of economic considerations. Recourse laws can, however,

7

ACCEPTED MANUSCRIPT

be partly traced to states’ experience with farm foreclosure in the early 1930s. 3. Empirical strategy, data, and results In this section, we discuss our empirical strategy, provide an overview of our data set together

CR IP T

with summary statistics, and present our empirical results. 3.1. Empirical strategy

Consumer-friendly foreclosure laws impose additional costs on banks in case of mortgage default. By providing additional protections to borrowers, they could also increase the incentive to default. We, therefore, ask whether and to what extent do these laws affect banks’ lending

AN US

decisions at the time of origination.

To answer this question one has to take on the traditional identification challenge of separating between demand and supply effects using the available cross-sectional variation in these laws. A key objective of the empirical strategy is to ensure that a comparison of credit supply is not tainted by regional economic factors. Further, one has to address the possibility that foreclosure laws could affect consumers’ incentive to borrow and default. One has to also control for bank-specific

M

factors, especially because the banking landscape can vary across state borders, as we show. These potentially confounding factors dictate the use of a granular approach because aggregate variables, such as the total volume of credit, are more easily tainted by demand factors.

ED

Our use of individual loan–level data has several advantages in that regard. First, we are able to study the impact of foreclosure laws on banks’ decision to reject a loan controlling for

PT

borrower characteristics and regional fixed effects. By having our dependent variable reflect the bank’s decision conditional on a loan application, we severely reduce the potential for variations in demand to affect our results. Second, we restrict our sample to a homogeneous category of

CE

loans originated in urban areas straddling state borders. That is, we compare lending within neighboring urban areas to minimize the variation in unobservable characteristics that could

AC

affect the demand for credit. Third, the loan–level approach allows us to control for bank fixed effects. Fourth, and crucially, we exploit an exogenous difference in banks’ exposure to foreclosure laws above a certain loan amount, which helps us considerably sharpen our identification strategy. This approach allows us to further ascertain the role of foreclosure laws as well as to address concerns related to impact of foreclosure laws on demand. We are also able to 8

ACCEPTED MANUSCRIPT

test the identifying assumption that underpins this methodology because the data we use allow us to observe the quantity and the quality of loan applications. The baseline regression takes the form (1)

CR IP T

Ri,k,u,s = αk + β u + θXi,k,u,s + γRegs + ρ1 di,k,u,s + ρ2 d2i,k,u,s + ei,k,u,s ,

where Ri,k,u,s is a dummy variable that takes the value one if a loan application i, at bank k, on a property in the urban area straddling state border u, in state s, was rejected by the bank. We control for the characteristics of the applications available from HMDA (X), including income, gender, race, income of the census tract (area size typically smaller than a zip code area) and the ratio of loan to income. The regulatory variables of interest are included in the vector

AN US

Regs . These are dummies for judicial foreclosure requirement, statutory right of redemption, and prohibition of deficiency judgment. We also replace the judicial dummy with continuous foreclosure timelines.

The regression controls for bank (αk ) and state border (β u ) fixed effects allowing us to estimate the jump in the rejection rate at the state border. We also follow the recent tradition in the empirical literature and allow for the possibility that the dependent variable (in our case, the

M

rejection rate) to vary in function of the distance and squared distance to the border, where di,k,u,s is the distance of the centroid of the census tract (in which the property is located) to the state

ED

border. The distance takes negative values for judicial states [see, e.g., Mian, Sufi, and Trebbi (2015)]. To keep our sample as homogeneous as possible across a state border, we restrict our sample to census tracts that are within 25 miles from the border and further restrict this to ten

PT

miles for robustness.

Because we make use of two different measures of judicial foreclosure, one binary variable and one continuous USFN variable, we rely on two sets of alternative specifications. In the

CE

regressions in which the key regulatory variable is binary, we restrict our sample to those urban areas straddling state borders that separate a judicial from a nonjudicial state. When we use the

AC

continuous USFN variable, measuring delays in foreclosures due to regulations, we include all urban areas that straddle state borders. Our sample spans the 2001–2006 period. Because the literature finds a loosening of lending

standards around this period, we run our regressions for the following subsamples: 2001–2002,

9

ACCEPTED MANUSCRIPT

2003–2004, and 2005–2006. We use bi-yearly samples (as in Pence (2006)) to avoid excessively reducing the size of the sample especially because our key regressions rely on the availability of sufficient jumbo loan applications, which make up around 10% of total applications, and our sample includes smaller and less populated metropolitan statistical areas (MSAs). We estimate Eq. (1) with a linear probability model (LPM) to fit a binary dependent variable.

consistent using LPM [see, e.g., Wooldridge (2002)]. 3.1.1. Exploiting the segmentation of the mortgage market

CR IP T

In our setting, the LPM has an important advantage over probit and logit models when N → ∞ √ and T is fixed, because the estimates are generally inconsistent in probit or logit, but are N

One cannot study the US mortgage market without taking into consideration the role of the

AN US

government-sponsored enterprises [i.e., Fannie Mae and Freddie Mac]. The GSEs are at the heart of a securitization market that is a key feature of the deep mortgage markets in the United States. When a bank originates a mortgage with certain standard features, it has the option to sell the mortgage to the GSEs or to purchase credit guarantees from them. In both cases, the credit risk on the loan is transferred to the GSEs.

Importantly, and in case of default, the protection offered by the GSEs compensates not

M

only the creditors (banks or mortgage-backed security investors) but also the servicers for the associated foreclosure costs. Aware of the differences in foreclosure time frames across states, the

ED

GSEs issue guidelines to servicers that take into account these differences. Also, the guarantee fees charged by the GSEs do not vary across states. Further, the explicit and transparent eligibility criteria to GSE guarantees do not discriminate across states. This is no doubt due to their public

PT

charter and the mandate by Congress, prior to the crisis, for the GSEs to support homeownership. The fact that the GSEs did not price risk like a private insurer would have is a well-acknowledged

CE

fact.6

The market share of loans purchased or securitized by the GSEs has increased steadily since

AC

6 When the acting director of the Federal Housing Financing Agency (FHFA) announced a relative increase in guarantee fees on states with longer foreclosures delays in 2013, he acknowledged that: “The new pricing continues the gradual progression toward more market-based prices, closer to the pricing one might expect to see if mortgage credit risk was borne solely by private capital.” See http://www.fhfa.gov/Media/PublicAffairs/Pages/ FHFA-Takes-Further-Steps-to-Advance-Conservatorship-Strategic-Plan-by-Announcing-an-Increase-in-GuaranteeFees.aspx.

10

ACCEPTED MANUSCRIPT

the early 1990s. By 2003, around 45% of all outstanding mortgages had GSE guarantees.7 Because the GSE eligible market is roughly around 75% of the overall market, the share of the GSEs in their market was closer to 60%. This share, however, declined to some extent during 2004–2006 when the private label market boomed. These considerations imply that, to estimate the impact of borrower-friendly laws on banks’

CR IP T

lending decision, one needs to take into account the role of the GSEs. The availability of a credit protection on GSE-eligible loans, which also compensates for the uncertain foreclosure costs in exchange for a uniform guarantee fee across states, shields lenders from borrower-friendly laws. This identification problem can be overcome by acknowledging the fact that the GSEs operate under a special charter limiting the size of mortgages that they can purchase or securitize. Loans above the legal cutoff are typically called jumbo loans. As discussed in Loutskina and Strahan

AN US

(2009), jumbo loans are less liquid and are often held by banks or sold to private securitization pools. The existence of an exogenous discontinuity in the GSE subsidization of borrower friendly states suggests that the impact of judicial foreclosure on the supply of credit, if it is significant, should be more conspicuous in the jumbo segment of the market. By comparing jumbo loans with conforming loans, we can also difference out unobservable but potentially confounding demand-side factors.

M

Therefore, the following regression is key to our empirical strategy:

ED

Ri,k,u,s = αk + β u + θXi,k,u,s + γRegs + δJ + φRegs J + ρ1 di,k,u,s + ρ2 d2i,k,u,s + ei,k,u,s ,

(2)

cutoff).

PT

where J is a dummy variable that takes a value of one if the loan is jumbo (above the designated We estimate this regression in both the full sample and a subsample that includes applications

CE

that lie within a small interval of the jumbo cutoff to compare loans of similar size. If these borrower-friendly laws discouraged lending by banks, we would expect to observe a positive and significant coefficient on φ. Potentially confounding factors, such as regional variations in

AC

demand that are non-orthogonal to foreclosure laws, would not affect the coefficient θ in Eq. (2) as long as they affect demand evenly around the jumbo cutoff. We also test this identifying 7 See,

e.g., Acharya, Richardson, van Nieuwerburgh, and White (2011).

11

ACCEPTED MANUSCRIPT

assumption by examining the relation between foreclosure laws and the relative demand, both in quantity and quality, of jumbo to non-jumbo loans. 3.2. Data and summary statistics 3.2.1. Data

CR IP T

We use a comprehensive sample of individual mortgage application and origination data collected by the Federal Reserve system under the provision of the Home Mortgage Disclosure Act. Under this provision, the vast majority of mortgage lenders are required to report their housing-related lending activity. HMDA data offer the best coverage of mortgage loans originated in the US, with coverage nearing 95% in urban areas.

HMDA data provide information on the characteristics of applicants including income, ethnicity,

AN US

race, sex, and the median income of the census tract where the property is located. The data also provide the year of the application, the lender, the characteristics of the loan, and whether the loan was approved. We restrict our sample to either approved or denied loans, and the loan type to be conventional (excluding loans guaranteed by the Federal Housing Administration, Veterans Administration, Farm Service Agency, or Rural Housing Service), the property types to be single family, the loan purpose to be home purchase (excluding refinancing or home improvement), and

M

the occupancy status to be owner-occupied as principal dwelling.

Our empirical strategy dictates that we focus on metropolitan counties lying on state borders.

ED

We follow Pence (2006) and define an urban area as the collection of contiguous counties that belong to an MSA. Our sample is, therefore, made of 52 urban areas straddling state borders.8 The geographic sample is shown in Fig. 3. We choose to restrict our sample period to 2001–2006.

PT

We exclude years before 2001 due to lack of data on foreclosure processing days. We do not extend our sample beyond 2006 to avoid tainting our results with factors that are specific to

CE

the crisis.9 To minimize noise and outliers, from each year and county we include only banks that received more than 30 applications. All these restrictions leave around 1.3 million loan

AC

applications made at 699 banks, in 198 counties and 42 states. Our analysis exploits the changes 8 See

Table A1 in the Online Appendix for a list of these urban areas. to the unprecedented number of foreclosures during the crisis, foreclosure laws had a real impact on the local economy (Mian, Sufi, and Trebbi (2015)). In normal times, however, with foreclosure rates typically below 1%, differences in the way foreclosures are processed are unlikely to have meaningful general equilibrium effects. We further address this issue below. 9 Due

12

ACCEPTED MANUSCRIPT

CR IP T

in regulations across 50 state borders.

AN US

Fig. 3. Urban areas in our cross-border sample. This map of the US mainland indicates in dark gray our sample of urban areas that are made of contiguous metropolitan counties straddling state borders.

Data on foreclosure laws are taken from various sources. We follow Mian, Sufi, and Trebbi (2015) and rely on RealtyTrac for the classification of states between judicial and nonjudicial. Our paper in addition relies on more direct measure of foreclosure timelines from the US Foreclosure

M

Network, which are also available online from RealtyTrac and are reproduced in the Online Appendix.10 The USFN measure is our favorite because it captures the wide variation in timelines across states. The fact that these timelines are constructed purely based on state regulations, i.e.,

ED

not incorporating additional delays that could be endogenous to market conditions, is ideal for our empirical strategy. We also supplement our measure of judicial requirements with numbers

PT

from Freddie Mac in the Online Appendix. We find that both the USFN and Freddie Mac measures remain constant between 2001 (first available year) and 2006 and that no material

CE

change has taken place until the foreclosure crisis. 3.2.2. Summary statistics Table 1 presents summary statistics from each year in our sample. The first three columns

AC

show the number of urban areas, state borders, and counties in the sample. We include only state borders along which an urban area is formed by MSAs from both states. The fourth column 10 The

USFN measure available from RealtyTrac corresponds to year 2004.

13

ACCEPTED MANUSCRIPT

Table 1 Summary statistics This table provides descriptive statistics of our final cross-border data set on individual loans from Home Mortgage Disclosure Act data. Columns 1-4 show the number of urban areas, state borders, counties and banks. The following columns show the average income of applicants, the average loan amount, and the ratio of the loan amount to income. The table also shows the ratio of jumbo (non-conforming) loans and the average rejection rate, which is the ratio of rejected applications to the total number of approved and rejected applications.

Year 2001 2002 2003 2004 2005 2006

Number of state borders (2) 50 50 50 50 50 50

Number of counties (3) 180 177 181 182 179 182

Number of banks (4) 417 422 430 397 395 355

Applicant income (thousands of dollars) (5) 93.57 95.89 91.92 94.95 100.05 105.29

Loan amount (thousands of dollars) (6) 145.56 160.43 169.15 201.66 212.69 203.12

Loan-to-income ratio (7) 1.97 2.12 2.26 2.49 2.45 2.26

Minority (percent) (8) 14.98 15.12 15.73 19.19 22.11 24.85

Jumbo loan (percent) (9) 9.02 10.11 9.92 14.30 14.98 8.78

Rejection rate (percent) (10) 11.92 11.43 11.06 10.20 12.02 12.61

CR IP T

2001 2002 2003 2004 2005 2006

Number of Urban areas (1) 53 53 53 53 53 53

AN US

Year

M

shows the number of banks. The decreasing number of banks is in line with the consolidation in the banking industry during the 2001–2006 period. Columns 5 and 6 show the average applicant

ED

income and loan amount, respectively. A key variable from the perspective of the lender is the ratio of loan-to-income (LTI), which is shown in Column 7. LTI has increased during the boom period, in line with the expansion in credit supply. Column 8 shows the percentage of minority

PT

borrowers. Column 9 shows that the jumbo loan market makes up around 10% of total loans, but that this share was higher at the peak of the boom. The final column shows the average rejection

CE

rate in our sample, which is decreasing between 2001 and 2004. Next, we motivate our cross-border analysis by examining differences in economic characteristics between counties within the selected urban areas lying on different sides of the state borders.

AC

Panel A of Table 2 compares county characteristics related to income, employment, minority, poverty, construction, mortgage delinquency, and foreclosure filings between the judicial and nonjudicial state subsamples. Based on the mean t-test, we find no significant differences in these characteristics. In Panel B, we run a regression of these characteristics on state border fixed effects

14

ACCEPTED MANUSCRIPT

Panel A

Panel B

ED

Nonjudicial states Mean 79.26 48.79 33.42 5.32 12.17 0.10 6.15 1.25 0.22

Judicial dummy as independent variable Coefficient p-value 0.732 0.861 0.571 0.828 0.361 0.808 -0.060 0.890 -0.693 0.320 0.028 0.271 0.797 0.389 -0.047 0.729 0.009 0.772

M

Dependent variables Applicant income (thousands of dollars) Median household income (thousands of dollars) Per capita personal income (thousands of dollars) Unemployment rate (percent) Poverty (percent) Minority (percent) New house built per 1,000 persons Delinquency ratio Foreclosure filing per 1,000 properties

Judicial states Mean 79.32 49.57 33.18 5.42 11.21 0.13 6.67 1.26 0.25

AN US

Variables Applicant income (thousands of dollars) Median household income (thousands of dollars) Per capita personal income (thousands of dollars) Unemployment rate (percent) Poverty (percent) Minority (percent) New house built per 1,000 persons Delinquency ratio Foreclosure filing per 1,000 properties

CR IP T

Table 2 Comparison of counties in judicial and nonjudicial states This table provides a comparison of county characteristics between judicial and nonjudicial states in the selected urban areas. Panel A compares county characteristics in the judicial and nonjudicial subsamples and presents a mean t-test. Panel B presents the coefficient and the p-value from a regression of these characteristics on the judicial dummy (first pair of columns) controlling for state border fixed effects. The second pair of columns repeats the exercise replacing the judicial dummy with the US Foreclosure Network (USFN) timelines. Mean test p-value 0.99 0.78 0.88 0.72 0.30 0.24 0.69 0.93 0.58

Delays (USFN) as independent variable Coefficient p-value 3.303 0.662 -5.319 0.180 -0.444 0.913 0.083 0.910 1.273 0.626 -0.040 0.385 0.985 0.548 0.334 0.180 0.001 0.989

and the judicial dummy (in the first pair of columns) or the USFN measure (the second pair of characteristics.

PT

columns). We continue to find no significant relation between foreclosure duration and county

CE

We next motivate the use of bank fixed effects (and thus loan-level data analysis) by comparing the banking landscape within urban areas but across state borders. We are interested in understanding the extent to which lenders are homogeneous across state borders. Because urban areas are

AC

typically asymmetric in terms of the number of counties on each side of the border, we focus on county-pairs straddling state borders within these urban areas. We look at banks’ operations across state borders in 2005. For the purpose of this exercise, we consider all banks, ignoring our minimum limit on the number of originations. There are 801 banks operating in 157 county pairs

15

ACCEPTED MANUSCRIPT

40

20

0 20 40 Market share in county A

60

AN US

0

CR IP T

Market share in county B

60

Fig. 4. The market share of banks in neighboring cross-border counties. For each county pair and each bank with positive lending in either of the two counties, we calculate the market share of the bank in each county. The figure shows these points for all county-pairs in our sample.

in our sample. Among all bank and county-pair combinations for which the bank is operating

M

in at least one of the counties, in only about 40% of cases do we see the bank operating in both counties. For those banks operating in both sides, market shares vary widely across borders. While many state banks not surprisingly operate on only one side of the border, our results

ED

suggest that the difference in the banking landscape even in terms of overall volume is significant. This finding is best summarized in Fig. 4. The figure plots a scatter of the market shares of bank k

PT

in counties A and B, which are part of a county-pair, for all banks and county-pairs in the sample. We find that the market shares of banks are very heterogeneous within county pairs. Numerous examples of banks with market shares above 40% in one county but with only modest presence

CE

in the neighboring county suggest that the banking landscape can vary significantly at the state border. This heterogeneity is not related to or affected by differences in foreclosure laws, as it also

AC

holds for state borders with similar laws. Therefore, by using bank fixed effects, our empirical methodology aims at studying the impact of state foreclosure laws on lending within the average bank.

16

ACCEPTED MANUSCRIPT

3.3. Results We next present the results from our estimations. The benchmark results are those obtained from estimating Eq.(1) and Eq.(2). Then follows several robustness tests. We also discuss the economic magnitude of our findings.

CR IP T

3.3.1. Benchmark results

The results from the benchmark regressions are shown in Table 3. All regressions include state border and bank fixed effects. Standards errors are clustered at both the state border and bank levels.

Columns 1–3 we present the coefficient estimates from the baseline regression (1). The first column shows the estimates for the 2001–2002 period; the second, for the 2003–2004 period;

AN US

and the third, for the 2005–2006 period. The first five rows show the coefficients on the X variables, which are the characteristics of the loan or borrower. The results are in line with expectations and the literature. We find that loan applications have a higher probability of rejection when they are made by applicants in lower-income census tracts, with lower income, and with higher loan-to-income ratios. The coefficient on the jumbo dummy is also positive, which is not surprising given the subsidy received by the non-jumbo loans. We find that the

M

coefficient on judicial dummy is positive but small and far from significant in each of the three periods.

ED

The absence of a significant effect of state laws could be simply a result of the GSE guarantees (see Subsection 3.1). We, therefore, differentiate between conforming and jumbo loans, and we estimate Eq.(2), which interacts the jumbo dummy with state laws. The results are shown in

PT

Columns 4–6 for the three periods. We find a positive and significant coefficient on the interaction between the jumbo dummy and judicial foreclosure. The results show that, on average, the

CE

rejection rate of jumbo loans in judicial states is higher by 2.2 to 2.9 percentage points. This difference is statistically significant at the 5% level. A move from a nonjudicial to a judicial state is associated with a significant jump in rejection rates on jumbo loans by such orders of

AC

magnitude. Given the coefficients on the jumbo dummy, the increase in rejection rate at the jumbo cutoff, particularly after 2002, mostly takes place in judicial states. The coefficient on the interaction variable of interest is somewhat decreasing in magnitude

over time, although not significantly. This can reflect the decline in the expected probability of

17

AC

PT

CE

18

Number of observations R-squared

Distance squared

Distance

0.515** (2.16) 14.38 (0.99) 99313 0.142

0.0356 (0.15) 1.769 (0.13) 341781 0.092

Within 25 miles 2003–2004 2005–2006 (5) (6) -0.0441*** -0.0378*** (-5.52) (-5.27) -0.0313*** -0.0323*** (-3.89) (-5.45) 0.00870** 0.00615** (2.33) (2.08) 0.0595*** 0.0553*** (6.58) (5.89) 0.000926 -0.000682 (0.25) (-0.22) 0.0119 0.0170 (0.99) (1.43) -0.00128 -0.00203 (-0.22) (-0.29) 0.0287 -0.00274 (0.96) (-0.48) 0.0112 -0.00954 (0.83) (-1.45) 0.0241** 0.0217** (2.12) (2.22) -0.0202 -0.00167 (-0.93) (-0.15) 0.0137 -0.0293 (0.44) (-0.39) 0.474* 0.0280 (1.75) (0.12) 15.22 0.875 (0.95) (0.06) 215176 341781 0.097 0.092

CR IP T

2001–2002 (4) -0.0248*** (-2.61) -0.0605*** (-4.62) 0.00172** (2.02) 0.0678*** (4.57) -0.00956 (-1.43) 0.0475** (2.25) 0.00407 (0.35) 0.0872 (1.60) 0.0198 (0.55) 0.0293** (2.20) -0.121** (-2.21) 0.0146 (0.27) 0.506** (2.14) 12.67 (0.90) 99313 0.143

AN US

0.479* (1.78) 16.70 (1.03) 215176 0.096

M

2001–2002 (1) -0.0247** (-2.55) -0.0607*** (-4.63) 0.00171** (2.02) 0.0678*** (4.64) -0.00950 (-1.44) 0.0680*** (3.61) 0.00585 (0.50) 0.0836 (1.50) 0.0208 (0.57)

ED

Jumbo dummy*Deficiency prohibition dummy

Jumbo dummy*Redemption dummy

Jumbo dummy*Judicial dummy

Deficiency prohibition dummy

Redemption dummy

Judicial dummy

Jumbo dummy

Female

Minority

Loan-to-income ratio

Log of applicant income

Variable Log of tract median income in 2005

Within 25 miles 2003–2004 2005–2006 (2) (3) -0.0438*** -0.0376*** (-5.42) (-5.26) -0.0314*** -0.0325*** (-3.87) (-5.48) 0.00867** 0.00612** (2.30) (2.08) 0.0597*** 0.0556*** (6.64) (5.97) 0.000957 -0.000612 (0.27) (-0.19) 0.0286** 0.0323*** (2.22) (2.62) 0.00231 0.00156 (0.45) (0.27) 0.0257 -0.00517 (0.86) (-0.96) 0.0145 -0.00644 (1.11) (-1.10)

2001–2002 (7) -0.0246** (-2.13) -0.0560*** (-5.07) 0.00162*** (2.58) 0.0675*** (4.55) -0.0134** (-2.20) 0.0411** (2.16) 0.00343 (0.36) 0.0356 (1.38) 0.0242 (0.74) 0.0366** (2.41) -0.0618** (-2.26) -0.0145 (-0.65) -0.0602 (-0.12) -145.5 (-1.64) 65672 0.124

Within 10 miles 2003–2004 2005–2006 (8) (9) -0.0438*** -0.0357*** (-5.02) (-5.26) -0.0281*** -0.0342*** (-3.49) (-5.68) 0.00938** 0.00535* (2.15) (1.84) 0.0616*** 0.0568*** (5.31) (5.09) -0.00112 0.00146 (-0.24) (0.48) 0.00873 0.0151 (0.83) (1.18) 0.000362 -0.00359 (0.05) (-0.47) 0.00467 -0.00613 (0.53) (-0.83) 0.0205 -0.00815 (1.23) (-1.10) 0.0248* 0.0225** (1.66) (2.32) -0.00819 0.00422 (-0.49) (0.32) 0.00626 -0.0446 (0.17) (-0.73) 0.218 -0.114 (0.43) (-0.21) -48.09 -104.2 (-1.09) (-1.43) 137533 220601 0.094 0.097

Table 3 Benchmark regressions: judicial dummy This table presents the coefficient estimates from a linear probability model regressing the rejection dummy on application characteristics, foreclosure laws, and bank and state border fixed effects. The key explanatory variable is the judicial dummy. In Columns 1–3 we include only foreclosure laws and show the results for the period 2001–2002, 2003–2004, and 2005–2006, respectively. In Columns 4–6 we add the interaction of the jumbo dummy with foreclosure laws. In Columns 7–9 we restrict our sample to the loans that are within ten miles from the state border. t-statistics are reported in brackets. ***, **, * indicate statistical significance at the 10%, 5%, and 1% level, respectively.

ACCEPTED MANUSCRIPT

AC

PT

CE

19

Number of observations R-squared

Distance squared

Distance

0.116 (0.63) 8.648 (0.76) 197691 0.143

0.295* (1.87) -2.346 (-0.21) 673776 0.096

Within 25 miles 2003–2004 2005–2006 (5) (6) -0.0512*** -0.0448*** (-6.61) (-6.50) -0.0390*** -0.0304*** (-5.48) (-5.33) 0.00387* 0.00645*** (1.70) (3.09) 0.0644*** 0.0574*** (6.53) (6.79) -0.00201 -0.000518 (-0.86) (-0.22) 0.0210** 0.0208*** (2.17) (2.71) 0.00517 0.0305* (0.30) (1.68) 0.0208 -0.000254 (0.87) (-0.06) -0.000142 0.00841 (-0.01) (0.80) 0.0415*** 0.0319*** (2.82) (2.79) -0.0165 0.000235 (-0.90) (0.03) 0.0171 0.0397** (1.39) (2.47) 0.192* 0.299* (1.78) (1.96) -4.344 -2.473 (-0.46) (-0.23) 415396 673776 0.099 0.096

CR IP T

2001–2002 (4) -0.0358*** (-5.21) -0.0585*** (-5.12) 0.00144** (2.47) 0.0641*** (5.86) -0.00854** (-1.99) 0.0482*** (2.61) -0.000697 (-0.02) 0.0735* (1.78) -0.00923 (-0.24) 0.0445*** (3.12) -0.0946** (-2.19) 0.0324 (1.04) 0.114 (0.62) 8.066 (0.72) 197691 0.143

AN US

0.192* (1.84) -3.732 (-0.40) 415396 0.099

M

2001–2002 (1) -0.0354*** (-5.00) -0.0586*** (-5.12) 0.00145** (2.37) 0.0640*** (5.76) -0.00848** (-1.99) 0.0689*** (4.35) 0.00154 (0.05) 0.0705* (1.70) -0.00714 (-0.18)

ED

Jumbo dummy*Deficiency prohibition dummy

Jumbo dummy*Redemption dummy

Jumbo dummy*USFN timelines

Deficiency prohibition dummy

Redemption dummy

USFN timelines

Jumbo dummy

Female

Minority

Loan-to-income ratio

Log of applicant income

Variable Log of tract median income in 2005

Within 25 miles 2003–2004 2005–2006 (2) (3) -0.0508*** -0.0448*** (-6.46) (-6.47) -0.0388*** -0.0302*** (-5.43) (-5.27) 0.00386* 0.00645*** (1.71) (3.10) 0.0646*** 0.0576*** (6.61) (6.84) -0.00201 -0.000464 (-0.85) (-0.19) 0.0371*** 0.0339*** (3.57) (3.81) 0.0120 0.0360* (0.64) (1.78) 0.0194 -0.000630 (0.82) (-0.14) 0.00199 0.0122 (0.10) (1.07)

2001–2002 (7) -0.0366*** (-3.38) -0.0560*** (-5.59) 0.00110** (1.96) 0.0641*** (4.80) -0.0117** (-2.35) 0.0424** (2.00) -0.00779 (-0.24) 0.0448** (2.26) -0.000822 (-0.03) 0.0502*** (2.84) -0.0377 (-1.47) 0.0146 (0.61) -0.0858 (-0.17) -72.22 (-1.08) 116107 0.130

Within 10 miles 2003–2004 2005–2006 (8) (9) -0.0545*** -0.0440*** (-5.23) (-5.02) -0.0355*** -0.0316*** (-5.13) (-5.21) 0.00547** 0.00630*** (2.06) (2.65) 0.0663*** 0.0587*** (5.83) (6.47) -0.00366 0.000349 (-1.16) (0.14) 0.0140 0.00955 (1.29) (0.87) 0.0267* 0.000919 (1.66) (0.05) 0.00973 -0.00423 (1.13) (-0.90) 0.0101 0.00859 (0.67) (0.68) 0.0473*** 0.0417** (2.79) (2.55) -0.0104 0.000336 (-0.49) (0.03) 0.0145 0.0414** (0.99) (2.11) 0.370 0.0345 (1.22) (0.09) -51.39* -50.02 (-1.83) (-0.96) 242236 394996 0.097 0.099

Table 4 Benchmark regressions: US Forecolsure Network (USFN) timelines This table presents the coefficient estimates from a linear probability model regressing the rejection dummy on application characteristics, foreclosure laws, and bank and state border fixed effects. The key explanatory variable is the USFN timelines. In Columns 1–3 we include only foreclosure laws and show the results for the period 2001–2002, 2003–2004, and 2005–2006, respectively. In Columns 4–6 we add the interaction of the jumbo dummy with foreclosure laws. In Columns 7–9 we restrict our sample to the loans that are within ten miles from the state border. t-statistics are reported in brackets. ***, **, * indicate statistical significance at the 10%, 5%, and 1% level, respectively.

ACCEPTED MANUSCRIPT

ACCEPTED MANUSCRIPT

default [as argued by, for example, Brueckner, Calem, and Nakamura (2012)], which correlated with an acceleration of house price growth). This can also explain the decline over time of the coefficient on the jumbo dummy alone. Other related factors could also be at play, such as the deepening of the private securitization market [see, e.g., Keys, Seru, and Vig (2012)]. The coefficients on the interaction between jumbo and the two other foreclosure laws are in all but

CR IP T

one case not significantly different from zero. The weak variation in these laws in our sample suggests that one should be careful in interpreting the coefficients on redemption and deficiency. The coefficients on the distance variables are non-significant in all but one case. More important, the inclusion of these distance regressors does not affect the coefficients of interest, as shown in the Online Appendix.

In the last three columns, we restrict our sample to census tracts that are within ten miles of

AN US

the border, reducing the sample by roughly a third. The results are very similar. We continue to find that the key coefficient on the interaction term between the judicial dummy and the jumbo dummy is positive and significant.

The judicial foreclosure dummy is not a sufficient statistic of the differences in foreclosure timelines associated with the heterogeneity in the judicial proceedings across states. We, therefore, reestimate the regression replacing the judicial dummy variable with the continuous measure

M

taken from the USFN. Because these timelines vary continuously across states, we make use of the full sample of urban areas that straddle any state border. For the purpose of comparison, we

ED

re-scale the USFN measure for it to lie in the [0, 1] interval. The results, which are broadly similar, are shown in Table 4. Columns 1–3 show the results from the baseline regressions. Again, the evidence that foreclosure laws have an impact on

PT

lending decisions for the average loan is weak, with the coefficient being significant only in the 2005–2006 period at 10%. But, as can be seen in Columns 4–6, and the results for the reduced

CE

sample in Columns 7–9, the coefficients on the interaction between the foreclosure delays measure and the jumbo dummy are consistently positive and significant, mostly at the 1% level. The magnitude of the coefficient varies between 0.032 to 0.05, suggesting that a hypothetical move

AC

from the least regulated state (Texas scores 0.06 on the USFN rescaled measure) to the most regulated state (NY scores one) would result in a significant jump in rejection rates by such orders of magnitude. Given the coefficients on the jumbo dummy, the increase in rejection rate at the jumbo cutoff is twice as large in the most regulated state. We continue to find that the 20

ACCEPTED MANUSCRIPT

magnitude of the coefficient on the key term is slightly decreasing over time. 3.3.2. Band samples and falsification The estimation of the impact of foreclosure laws on the relative rejection rate of jumbo loans could be further sharpened by focusing on a subset of loans that are near the jumbo cutoff.

CR IP T

This would zoom in on the comparison with loans that are more similar in size (and thus to borrowers with more similar incomes) and help address the concern that the jumbo cutoff is simply capturing the difference in loan amounts. We, therefore, repeat Eq. (2) on subsamples of loans that are within 20%, 10%, and 5% of the jumbo cutoff. Because our focus is on the interaction term, we are also able to control for county fixed effects instead of the state border fixed effect, which allows us to further address unobservable factors at the county level.11 Before

AN US

presenting the results from the regressions, we use a simple figure to illustrate the idea. Figure 5 shows a box-plot comparison of the distribution of those rejection rates for both conforming loans and jumbo loans in the 20% band sample, in nonjudicial (Panel A) and judicial states (Panel B). In states not requiring judicial process, the rejection rates for conforming loans and jumbo loans are very close. However, the difference of rejection rate between jumbo and conforming loans is larger in judicial states by 2.5 percentage points for the median county (the mean rejection rate is

M

also higher by around 1 percentage point).

The results from the band regressions are shown in Table 5. The regressions combine all

ED

years in the sample given the significantly reduced sample and the fact that the coefficients from the different time periods are not significantly different from each other. The first three columns show the results when we use the judicial requirement dummy and restrict our sample to borders

PT

that separate a Judicial from a nonJudicial state. The next three columns replace the dummy with the USFN measures. We find a consistently positive and significant coefficient on the interaction

CE

between the jumbo dummy and the judicial requirement. The magnitude of the coefficients is consistent, albeit slightly smaller, than Tables 3 and 4. The coefficients on the interactions between jumbo dummy and redemption or prohibit deficiency dummies are not significant.

AC

To further confirm that our results are due to the discontinuation of the GSE guarantees at

the jumbo cutoff, we conduct a falsification test. We assume an imaginary jumbo cutoff that is at 11 We continue using the border sample because one can argue that unobservable factors could interact with the jumbo dummy. Hence, it is best if we minimize imbalances in the sample by focusing on these border urban areas.

21

PT

CE

ED

22

Number of observations R-squared

Distance squared

Distance

0.000111 (0.00) -0.0515 (-0.86) 0.119 (0.41) -13.41 (-0.71) 40744 0.134

0.0286*** (3.59) -0.000460 (-0.03) 0.00754 (0.52) 0.222 (1.38) -15.95* (-1.77) 150981 0.135

0.0226** (2.03) 0.00124 (0.05) 0.00969 (0.48) 0.0862 (0.44) -12.95 (-1.10) 68981 0.142

-0.00462 (-0.15) 0.00931 (0.33) -0.575 (-0.97) -29.98 (-1.01) 19897 0.145

0.0289* (1.72) -0.00352 (-0.17) -0.00286 (-0.10) -0.123 (-0.84) -18.72 (-1.22) 33580 0.153

All MSA sample 20% band 10% band 5% band (4) (5) (6) -0.0355*** -0.0340*** -0.0341*** (-4.69) (-4.69) (-3.29) -0.0421*** -0.0464*** -0.0489*** (-5.20) (-6.29) (-5.42) 0.00269 0.00263 0.00207 (1.57) (1.50) (0.94) 0.0582*** 0.0581*** 0.0709*** (4.64) (4.80) (6.09) 0.00515 0.00885* 0.0102* (1.13) (1.73) (1.83) 0.0113 0.0112 0.00692 (1.63) (1.60) (0.78)

CR IP T

AN US

-0.0179 (-0.74) 0.00499 (0.13) 0.254 (1.50) -10.81 (-1.10) 88136 0.126

M

Jumbo dummy*Deficiency prohibition dummy

Jumbo dummy*Redemption dummy

Jumbo dummy*USFN timelines

Jumbo dummy*Judicial dummy

Jumbo dummy

Female

Minority

Loan-to-income ratio

Log of applicant income

Variable Log of tract median income in 2005

Judicial/nonjudicial sample 20% band 10% band 5% band (1) (2) (3) -0.0299*** -0.0290*** -0.0263** (-3.10) (-2.96) (-2.27) -0.0346*** -0.0414*** -0.0452*** (-3.67) (-4.38) (-3.76) 0.00441 0.00326 0.00223 (1.61) (1.19) (0.82) 0.0553*** 0.0541*** 0.0706*** (3.87) (4.16) (5.34) 0.00567 0.0109* 0.0123 (1.37) (1.92) (1.55) 0.00502 0.00581 0.00143 (0.74) (0.94) (0.15) 0.0217*** 0.0181** 0.0206** (3.04) (2.39) (2.51)

Table 5 Robustness check: band regression This table presents the coefficient estimates from the linear probability model regression of the rejection dummy on application characteristics and the interaction between foreclosure laws and jumbo dummy. The regression also control for county, bank, and year fixed effects. The regressions are run on subsamples of applications that fall within a band of the jumbo cutoff. For example, the 20% band sample includes only loans with size between 80% and 120% of the jumbo cutoff. In Columns 1–3 we use the judicial dummy. In Columns 4–6 we replace the judicial dummy with the US Foreclosure Network (USFN) timelines. We cluster standard errors at both state border and bank level. t-statistics are reported in brackets. ***, **, * indicate statistical significance at the 10%, 5%, and 1% level, respectively.

AC

ACCEPTED MANUSCRIPT

PT

CE

ED

23

M

Number of observations R-squared

Distance squared

Distance

False jumbo dummy*Deficiency prohibition dummy

False jumbo dummy*Redemption dummy

False jumbo dummy*USFN timelines

Flase jumbo dummy*Judicial dummy

False Jumbo dummy

Female

Minority

Loan-to-income ratio

Log of applicant income

Variable Log of tract median income in 2005

-0.00640 (-0.56) 0.00279 (0.16) 0.194 (1.08) -2.424 (-0.23) 116059 0.111

0.00790 (1.32) 0.00214 (0.35) -0.0123* (-1.81) 0.252 (1.19) -21.60 (-1.35) 216058 0.121

0.00910 (0.97) 0.00206 (0.12) 0.0190 (1.46) 0.201 (1.08) -3.531 (-0.22) 117977 0.139

All MSA sample 80% of actual 120% of actual jumbo cutoff jumbo cutoff (3) (4) -0.0329*** -0.0338*** (-4.45) (-7.47) -0.0379*** -0.0454*** (-4.94) (-5.36) 0.00347* 0.00264* (1.68) (1.69) 0.0554*** 0.0592*** (6.40) (4.95) 0.00493** 0.00823 (2.03) (1.52) 0.00701* 0.0159** (1.75) (2.52)

-0.0425** (-2.57) 0.0444 (1.22) 0.181 (0.55) 16.34 (0.91) 71919 0.131

CR IP T

AN US

Judicial/nonjudicial sample 80% of actual 120% of actual jumbo cutoff jumbo cutoff (1) (2) -0.0247*** -0.0325*** (-2.64) (-6.64) -0.0351*** -0.0374*** (-3.77) (-2.58) 0.00402 0.00583 (1.42) (1.24) 0.0542*** 0.0525*** (5.89) (4.08) 0.00383 0.00900** (1.45) (1.99) 0.00350 0.0165** (0.51) (2.07) 0.00798 0.00118 (1.10) (0.16)

Table 6 Robustness check: falsification test This table presents the coefficient estimates from the same regressions in Table 5 with the exception that jumbo cutoff is chosen to be, as a falsification test, at 80% or 120% of the actual cutoff. We continue to control for application characteristics and county, bank, and year fixed effects and cluster standard errors at both state border and bank level. We only show the results for the 20% band sample. t-statistics are reported in brackets. ***, **, * indicate statistical significance at the 10%, 5%, and 1% level, respectively.

AC

ACCEPTED MANUSCRIPT

ACCEPTED MANUSCRIPT

20

20

15

15

10

5

10

5

0

0 1
0.8
1
AN US

0.8
CR IP T

Panel B: States requiring judicial process 25

Rejection rate

Rejection rate

Panel A: States not requiring judicial process 25

Fig. 5. The distribution of rejection rates within a band of the jumbo cutoff. This figure shows the distribution of rejection rates for both conforming loans and jumbo loans in the 20% bandwidth sample, in non-judicial states (Panel A) and judicial states (Panel B). In each panel the left box shows the distribution of rejection rate on loans that are above 80% and below 100% of the jumbo cutoff, and the right box shows the distribution of rejection rate on loans that are above 100% and below 120% of the jumbo cutoff.

80% or 120% of the actual cutoff. We then create a new jumbo dummy based on the respective

M

imaginary cutoff. We then repeat the band estimation using these false cutoffs (Table 6). We only show the results for the 20% band. (The full results are available in the Online Appendix.) The

ED

first two columns show the results for the judicial dummy and the corresponding sample. We find that the key coefficient is no longer significant in either the 80% or 120% sample. The false jumbo dummy is significant in the 120% sample, but that just reflects a higher rejection rate on

PT

loans with higher amounts. In the full sample, in which we explore the interaction between the jumbo dummy and the USFN variable (Columns 3 and 4), we also find that the coefficients on

CE

this interaction are not significant. The results strongly support our hypothesis that the impact of judicial requirement is more pronounced for jumbo loans due to the legal constraint on GSEs, not other confounding factors such as loan size.

AC

3.3.3. Demand around the jumbo cutoff Our empirical strategy has thus far focused on addressing potentially confounding demand

factors in several ways. First, we have chosen the dependent variable to be the bank’s decision to reject a loan (not the volume of loans or their average size), a variable that is more directly linked 24

ACCEPTED MANUSCRIPT

to the supply side of the market. Second, we control for loan and borrower characteristics, as well as state border (in some cases county) fixed effects in our regressions. Third and crucially, our results from the jumbo cutoff analysis further minimize such concerns because for these results to be demand-driven there has to be a significant shift in demand around the jumbo cutoff and the extent of these changes in demand must correlate with foreclosure laws.

CR IP T

HMDA data allow us to investigate our identifying assumption, as we can observe the total number of applications, i.e., the demand for loans. Thus, to address concerns that foreclosure laws, by lowering the costs of foreclosure to households, can somehow tilt the relative demand of jumbo loans and subsequently impact the rejection rate, we can examine whether one can observe a correlation between foreclosure laws and relative changes in demand.12

We, therefore, proceed to compare demand characteristics within a band around the jumbo

AN US

cutoff in line with our earlier exercise. We focus on the 20% band sample, but similar results are obtained using the smaller bands. We compute the ratio of the number and dollar volume of jumbo to conforming applications. We also compute the ratio of average applicant income and the ratio of the average of census tract income (HMDA reports the median census tract income). We estimate the regression

J N J c,u,s,t

(3)

is the ratio of jumbo to non-jumbo loans (or other demand measures) in county

ED

where

M

J = αu + ρt + γRegs + ec,u,s,t , N J c,u,s,t

c, state border u, state s, and time t. We control for both state border and time fixed effects and control for state regulations (Regs ). The results are shown in Table 7. We first look at the results

PT

from the specification using the judicial foreclosure dummy as a control (Columns 1–4). The first column shows the results from the regression of the relative number of jumbo to non-jumbo loans

CE

on foreclosure laws, controlling for state border and year fixed effects, and clustering standard errors at the state border level. In the second column we replace the number of loans by the dollar volume of loans. In Column 3 the left-hand-side variable is the ratio of the average

AC

applicant income of jumbo to non-jumbo loans. This is replaced by the ratio of the average median tract income in Column 4. All coefficients on foreclosure laws, including the judicial 12 Even in the presence of relative changes in demand, absent capacity constraints, it is not clear how higher demand should lead to a higher rejection rate, controlling for borrower characteristics.

25

ACCEPTED MANUSCRIPT

foreclosure dummy, are far from significance levels. We repeat these same four regressions replacing the judicial foreclosure dummy with the USFN timelines (Columns 5–8). The results across all these specifications indicate no significant correlation between foreclosure requirements and quantitative or qualitative differences in demand around the jumbo cutoff. Table 7 brings direct evidence in support of our identifying assumption, and hence of our interpretation of the

CR IP T

earlier results as being evidence of a response in the supply of credit to foreclosure costs. 3.3.4. Economic magnitude

Our findings naturally raise the question of whether the overall effect of judicial foreclosure on credit supply is economically meaningful. While estimating the general equilibrium effects in the context of our empirical model is impossible, we can nevertheless illustrate the magnitude

AN US

of the effect. We choose the simplest and most tractable approach to do so. Our earlier results suggest that judicial foreclosure is associated with at least a 2 percentage point relative increase in the rejection rate on jumbo loans. Similarly, we find that processing delays increase the rejection rate by a slightly larger magnitude. We can, therefore, illustrate the overall effect on credit, using a back-of-the-envelope calculation. We focus on the results from the dummy variable (judicial versus nonjudicial) for simplicity.

M

The uniform guarantees by the GSEs have shielded the conforming market from the adverse effect of foreclosure laws on credit supply. We, therefore, limit our attention to the jumbo market.

ED

Because we are interested in shedding light on the potential aggregate effect, we consider lending in all counties in the mainland United States, and not just border counties. Between 2001 and 2006, there were around 300,000 applications to jumbo loans in judicial states, with around

PT

$150 billion in 2006 US dollars. Therefore, a 2 percentage point increase in the rejection rate corresponds to around six thousand loans with a value of around $3 billion.

CE

Importantly, however, this number would have been much larger according to our results had the GSEs not indirectly subsidized conforming loans in judicial states. Applying, as a counterfactual example, this 2 percentage point rejection rate on the total number of loans in

AC

judicial states (around three million conforming and jumbo loans) during that period yields around 60 thousand loans with a total volume of around $10 billion. These calculations provide a useful illustration of the impact of GSEs’ cross-subsidization on overall lending. They suggest that absent the GSEs’ cross-subsidization, and everything else equal, mortgage lending would

26

PT

CE

ED

27

Number of observations R-squared

Deficiency prohibition dummy

Redemption dummy

USFN timelines

Variable Judicial dummy

0.0311 (0.49) -0.0196 (-0.32) 1106 0.153

Number of loans (1) 0.0318 (0.98)

M

0.0309 (0.68) 0.0311 (0.55) 1123 0.141

Volume of loans (2) 0.000469 (0.02) 0.00650 (1.43) 0.000853 (0.08) 916 0.069

0.000540 (0.27) -0.00128 (-0.30) 898 0.077

Tract median income (4) -0.000277 (-0.19)

-0.0657 (-1.34) 0.0256 (0.58) 0.0203 (0.32) 1123 0.143

0.00471 (0.56) 0.00713 (1.39) 0.00123 (0.13) 916 0.069

Applicant income (7)

CR IP T

-0.0609 (-0.68) 0.0360 (0.67) -0.0475 (-0.72) 1106 0.152

Number of loans (5)

AN US

Applicant income (3) 0.000835 (0.23)

Volume of loans (6)

-0.00312 (-0.99) 0.000218 (0.10) -0.00169 (-0.45) 898 0.077

Tract median income (8)

Table 7 Robustness check: relative demand This table presents the coefficient estimates from regressions of ratios of jumbo to conforming loans on foreclosure laws, and state border and year fixed effects. The subsample used is made of loans that are within the 20% band of the jumbo cutoff. In Columns 1–4 we use the judicial dummy as the key explanatory variable. The dependent variables in these columns are the ratio of the number of jumbo loans to that of conforming loans, the ratio of their volume, the ratio of the average applicant income in each of these segments, and the ratio of the average tract median income. In Columns 5–8 we replace the judicial dummy with the US Foreclosure Network (USFN) timelines. We cluster standard errors at the state border level. t-statistics are reported in brackets. ***, **, * indicate statistical significance at the 10%, 5%, and 1% level, respectively.

AC

ACCEPTED MANUSCRIPT

ACCEPTED MANUSCRIPT

have been smaller by $7 billion over that period (computed as the difference between the two scenarios).13 Clearly, the cross-subsidization by the GSEs became very costly during the foreclosure crisis when foreclosure delays in states such as New York reached more than eight hundred days. This has prompted the Federal Housing Finance Agency in 2013 to plan an increase in guarantee fees on a number of states with the longest processing delays (all of them requiring judicial

CR IP T

foreclosure).14 4. Conclusion

This paper studies the impact of foreclosure laws on the supply of mortgage credit. We seek to understand whether and to what extent borrower-friendly foreclosure laws can have a negative

AN US

impact on the willingness of banks to accept a loan application, everything else constant. Our empirical methodology addresses the usual challenges stemming from potentially confounding demand factors. It also takes into account the institutional characteristics of the mortgage market in the US, specifically the role of the GSEs, using an exogenous cutoff in loan eligibility for purchase and securitization by these entities. Overlooking the role of the GSEs could lead one to underestimate the impact of borrower-friendly laws on the supply of credit, given the

M

cross-subsidization by the GSEs of borrower-friendly states. Our findings point to a significant and meaningful impact of judicial foreclosure delays on the supply of mortgages. Our findings

ED

illustrate the negative indirect effect of borrower protection laws on credit and, therefore, help inform future cost-benefit analyses of these regulations. They also illustrate how the indirect subsidy by the GSEs to borrower-friendly states, which is currently being reconsidered, has

AC

CE

PT

helped mitigate the overall impact of borrower-friendly laws on aggregate credit supply.

13 It is important not to confuse the impact of the cross-subsidization with its net cost to the GSEs, which is clearly much more challenging to estimate. 14 The decision was later put on hold for further study following the appointment of a new director.

28

ACCEPTED MANUSCRIPT

References Acharya, V. V., Richardson, M., van Nieuwerburgh, S., White, L. J., 2011. Guaranteed to fail: Fannie Mae, Freddie Mac, and the debacle of mortgage finance. Princeton University Press, Princeton, NJ.

CR IP T

Berkowitz, J., White, M. J., 2004. Bankruptcy and small firms’ access to credit. RAND Journal of Economics 35 (1), 69–84.

Bipartisan Policy Center, 2013. The consumer financial protection bureau: measuring the progress of a new agency. Financial Regulatory Reform Initiative, Washington, DC.

Brueckner, J. K., Calem, P. S., Nakamura, L. I., 2012. Subprime mortgages and the housing bubble.

AN US

Journal of Urban Economics 71 (2), 230–243.

Calem, P. S., Jagtiani, J., Lang, W. W., 2014. Foreclosure delay and consumer credit performance. working paper no. 14-8, Federal Reserve Bank of Philadelphia, PA.

Clauretie, T. M., Herzog, T., 1990. The effect of state foreclosure laws on loan losses: evidence

M

from the mortgage insurance industry. Journal of Monetary, Credit and Banking 22 (2), 221–233. Evans, D. S., Wright, J. D., 2009. Effect of the consumer financial protection agency act of 2009 on

ED

consumer credit. Loyola Consumer Law Review 22 (3), 277–335. Favara, G., Imbs, J. M., 2015. Credit supply and the price of housing. American Economic Review 105 (3), 958–992.

PT

Gerardi, K., Lambie-Hanson, L., Willen, P. S., 2013. Do borrower rights improve borrower outcomes? evidence from the foreclosure process. Journal of Urban Economics 73 (1), 1–17.

CE

Ghent, A. C., 2013. America’s mortgage laws in historical perspective. unpublished working paper, Arizona State University, Phoenix, AZ.

AC

Ghent, A. C., Kudlyak, M., 2011. Recourse and residential mortgage default: evidence from us states. Review of Financial Studies 27 (6), 3139–3186.

Goodman, J., Levitin, A., 2014. Bankruptcy law and the cost of credit: the impact of cramdown on mortgage interest rates. Journal of Law and Economics 57 (1), 139–158. 29

ACCEPTED MANUSCRIPT

Gropp, R., Scholz, J. K., White, M., 1997. Personal bankruptcy and credit supply and demand. Quarterly Journal of Economics 112 (1), 217–251. Huang, R. R., 2008. Evaluating the real effect of bank branching deregulation: comparing contiguous counties across us state border. Journal of Financial Economics 87 (3), 678–705.

CR IP T

Keys, B. J., Seru, A., Vig, V., 2012. Lender screening and the role of securitization: evidence from prime and subprime mortgage markets. Review of Financial Studies 25, 2071–2108.

Kupiec, P. H., 2014. Government financial policy and credit availability. Statement before the House Financial Service Subcommittee on Monetary Policy and Trade, Washington, DC.

Loutskina, E., Strahan, P. E., 2009. Securitization and the declining impact of bank finance on loan

AN US

supply: evidence from mortgage originations. Journal of Finance 64 (2), 861–889.

Mian, A., Sufi, A., Trebbi, F., 2015. Foreclosures, house prices, and the real economy. Journal of Finance, Forthcoming 70 (6), 2587–2634.

Pence, K. M., 2006. Foreclosing on opportunity: state laws and mortgage credit. Review of

M

Economics and Statistics 88 (1), 177–182.

Schill, M. H., 1991. An economic analysis of mortgagor protection laws. Virginia Law Review

ED

77 (3), 489–538.

US House of Representatives, 2012. The consumer financial protection bureau’s threat to credit access in the united states. Committee on Oversight and Government Reform, Washington, DC.

PT

Wooldridge, J. M., 2002. Econometric analysis of cross section and panel data. MIT Press, Cambridge, MA.

CE

Zhu, S., Pace, R. K., 2011. The influence of foreclosure delays on borrowers default behavior. 46th

AC

Annual AREUEA Conference Paper.

30