Adversity or Strategy? - Name - Tufts University

0 downloads 120 Views 531KB Size Report
Bianconi, in the pursuit of the master's program in economics at Tufts. ..... financial counseling has an effect in redu
Adversity or Strategy?: The Effects of Credit Constraint and Expectation on Mortgage Default and Personal Bankruptcy Decisions A thesis submitted by Yoshiyuki Miyoshi In partial fulfillment of the requirements for the degree of Master of Arts in Economics

TUFTS UNIVERSITY May 2008

Advisor: Yannis Ioannides

Abstract This thesis contributes to the literature on mortgage default and personal bankruptcy with empirical findings, which disentangle the two mainstream theories: the willingness-to-pay theory and the ability-to-pay theory. The emphasis is on credit constraints and insolvency in order to integrate these two theories. The estimation results of a model of simultaneous choices between mortgage and nonmortgage delinquency suggest that credit constraints and borrowers’ insolvency are the bridge between the two conflicting theories. I also find the importance of backward-looking expectations and the homestead exemption in households’ decisions on mortgage default and bankruptcy. In addition, with a model of bankruptcy decisions, I show that credit constrained and seriously financially distressed households act more “ruthlessly” based on their financial benefit from bankruptcy filings.

i

Acknowledgment I would like to express my sincerest gratitude to my advisor, Professor Yannis Ioannides, whose patience, enthusiasm, expertise, and guidance, have added considerably to my research experience. Beginning with a class of graduate macroeconomics under his direction, in which I wrote a research paper in economics for the first time in my life, he has helped me make significant academic progress in organizing and developing my ideas and thoughts. He was also the one who inspired me to work on mortgage default and personal bankruptcy, which I never knew that was such an exciting subject as it is. Without his encouragement and support, I would not have been able to write this thesis. I would like to thank Professor Gilbert Metcalf for all his guidance. While working as a research assistant for him, I learned what professional economic research is like. I would also like to thank Professor Jeffrey Zabel and Professor Edward Kutsoati for their insightful comments during the defense. I would also like to show my appreciation to Professor Joshua Fischman for his support for this project. I am also grateful for the guidance of my academic advisor, Professor Marcelo Bianconi, in the pursuit of the master’s program in economics at Tufts. I am greatly indebted to the Ministry of Land, Infrastructure, Transport and Tourism of the Japanese government for its financial help. Finally, I would like to thank my parents, my brother and my friends for all of their emotional support in the completion of this thesis.

ii

Table of Contents Abstract

i

Acknowledgment

ii

Introduction

1

Chapter 1: Mortgage default, delinquency and personal bankruptcy

3

1.1 Introduction

3

1.2 Definition

4

1.3 Literature review

7

1.4 Determinants of borrowers’ decisions of default, delinquency and bankruptcy

10

1.4.1 Dependent and explanatory variables in the SCF

10

1.4.2 Probit estimations for delinquency, default and bankruptcy

14

1.4.3 Ordered probit specification for delinquency and default

17

1.4.4 Structural change of the impact of ARMs

20

1.5 Conclusion

22

Chapter 2: Delinquency choice between mortgage and nonmortgage loans

32

2.1 Introduction

32

2.2 Theoretical framework

33

2.2.1 Delinquency, default and bankruptcy, revisited

33

2.2.2 Theoretical consideration I: financial profits

35

2.2.3 Theoretical consideration II: credit constraint

41

2.3 Estimation methodology and variables

43

2.3.1 Specification of multiple choice estimation

iii

43

2.3.2 Data and variables

45

2.4 Empirical results and some extensions

49

2.4.1 Result of the base specification

49

2.4.2 Protecting housing investments through bankruptcy

51

2.4.3 Households’ attitude toward risks

54

2.5 Conclusion

56

Chapter 3: Personal bankruptcy decisions of mortgage borrowers

63

3.1 Introduction

63

3.2 Theoretical model for financial benefit from bankruptcy

65

3.3 Data and estimation methodology

67

3.3.1 Recent bankruptcy experiences of the households in the SCF

67

3.3.2 Homestead exemption as a random variable

69

3.3.3 Estimation methodology for binary choice

70

3.3.4 Piecewise specification for homestead exemption

71

3.3.5 Explanatory variables

73

3.4 Estimation results

74

3.5 Bankruptcy decisions of mortgage borrowers in “dead-end delinquency”

76

3.5.1 Adjustment for dead-end delinquent households

76

3.5.2 Estimation results for dead-end delinquent households

78

3.6 Conclusion

81

References

96

iv

Introduction This thesis examines households’ decisions on mortgage default, personal bankruptcy and loan delinquency from several viewpoints by using a household-level dataset from the Survey of Consumer Finances (SCF). One of the main objectives of this thesis is to contribute to the literature on mortgage default and personal bankruptcy by disentangling the two mainstream theories: the willingness-to-pay theory and the ability-to-pay theory. The emphasis is on credit constraints and insolvency in order to integrate these two theories. Although mortgage default and personal bankruptcy are two major devices that financially distressed homeowners can choose, there are no empirical studies that directly address these simultaneous and multiple choices between mortgage default and personal bankruptcy. The empirical results reported in this thesis suggest that some of the determinants work in completely different ways for borrowers’ decisions on mortgage default and filing for bankruptcy respectively. Particularly, the effects of the amount of payments for different types of loans strongly support the validity of the hypothesis that credit constraints and borrowers’ insolvency are the bridge between the two conflicting theories. I also focus on borrowers’ backward-looking expectations of capital gains, which has attracted little attention in the previous studies of mortgage default and, as long as I know, no attention at all in the literature on households’ bankruptcy decisions. The empirical results in this thesis show that backward-looking expectations of capital gains play an important role in households’ decisions on loan payments, in particular interacting with a remarkable feature of the U.S. bankruptcy law: namely, the homestead exemption. Chapter 1 reviews the concepts, the terminology, the legal frameworks, and the literature that are relevant to the study of mortgage default, personal bankruptcy and loan delinquency.

1

Then, through the probit estimation by using the SCF, I confirm the findings of previous studies; both the willingness-to-pay theory and the ability-to-pay theory have predictive power for mortgage default, bankruptcy and loan delinquency. I also find an emerging importance of adjustable rate mortgages for mortgage default in the U.S. Chapter 2 formulates a multinomial choice model between mortgage and nonmortgage loan delinquency, based on an idea that these delinquency decisions are closely related to borrowers’ decisions between mortgage default and bankruptcy. With a model taking account of credit constrained households as well as households that just try to obtain financial profits from mortgage default or bankruptcy, the empirical results suggest that households’ decisions are firstly based on financial profits, but the combination of credit constraints and insolvency drives some households into mortgage default or bankruptcy against their will. I also find that backward-looking expectations of capital gains are playing an important roll along with the homestead exemption. In addition, I identify the impact of households’ heterogeneity towards financial risks, which some previous studies find indirectly, by explicitly using a parameter of households’ preferences. Chapter 3 concentrates on households’ bankruptcy decisions, emphasizing the effects of backward-looking expectations and the homestead exemption. In addition to the empirical support for a general case, the results in this chapter also reveal that credit constrained and seriously financially distressed households act more “ruthlessly” based on their financial benefit from bankruptcy filings. These findings give further grounds for the importance of credit constraints in reconciling the two theories; once households are caught by hopelessly severe financial distress due to a lack of ability to pay, they choose their actions among available alternatives primarily based on the financial benefit that they would be able to obtain.

2

Chapter 1: Mortgage default, delinquency and personal bankruptcy 1.1.

Introduction The housing finance market in the United States has experienced considerable structural

change in recent years. The percentage of Americans who own homes is 68.4 in 2007, having shown a sudden increase by more than four points during the last decade after years of being relatively constant at near 64 percent1. According to the SCF, a comprehensive survey on households’ financial circumstances conducted by the Federal Reserve, among households that have mortgage loans, the median amount of debt secured by residential properties increased by 47% in only 6 years, from 68,000 in 1998 to 100,000 in 20042. The share of households that took out home equity loans against their residential properties as collateral, which is often viewed as one of the major problem causes in the recent downturn in the housing finance market3, increased by 20.6% during the same period. After this expansion of the housing finance market, the recent situation which home mortgage debtors are facing during downturn in housing markets is drawing close attention by people in a variety of social quarters, such as business, finance, government and even international institutional agents4. The main concern of these people is how much mortgage loans are in trouble and why; since the secondary mortgage market has developed so well that many loans are bundled into mortgage-backed securities and traded on a world-wide market, all the

1

Housing Vacancy Survey, U.S. Census Bureau. (http://www.census.gov/hhes/www/housing/hvs/annual07/ann07t15.html) 2 Author’s calculation. 3 “Home Economics”, the New Yorker, March 10, 2008. (http://www.newyorker.com/talk/financial/2008/03/10/080310ta_talk_surowiecki) 4 For example, the Japanese government estimated the total value of the assets backed by subprime mortgages that Japanese financial institutions hold in February, 2008. (http://www.fsa.go.jp/news/19/ginkou/20080213-2/01.pdf)

3

people across the globe involved in financial activities are anxious about what would happen if a large fraction of the current mortgage balance could not be repaid5. Considering this importance of residential mortgage loans to the entire financial system of the United States, and as it appears of the international finance world as well, the focus in this paper is on the behavior of financially distressed households who have mortgage debts. There are several options for borrowers who cannot continue to repay their mortgage debts. Mortgage default and personal bankruptcy, if they have other unsecured loans, are such primary examples. Also they might be able to recover normal financial status just by selling their homes in the market. However, recent research in financial economics has revealed that borrowers, even including borrowers who are not facing serious payment problems, may exert “options” of mortgage default and file for personal bankruptcy strategically. These financial devices were originally intended, however, to help borrowers who suffer from severe financial distress. In particular, the consensus that the legal system regarding bankruptcy favored debtors and that some borrowers may have abused the system encouraged the U.S. Congress to amend the bankruptcy law in 2005 in order to tighten the requirements for bankruptcy. This paper will investigate, using the SCF, evidence of borrowers’ strategic behavior and determinants affecting their decisions.

1.2.

Definitions I would like to start with clarifying the definitions of the key concepts and associated

terms used in this paper. When a borrower misses any of her loan payments, we say that she is in delinquency. On the other hand, although mortgage default also refers to failures of borrowers to

5

“Could Risky Mortgage Lending Practices Prick the Housing Bubble?”, Knowledge@Wharton, September 21, 2005. (http://knowledge.wharton.upenn.edu/article.cfm?articleid=1280)

4

honor the terms of their loan agreements in general, there is no formal universally accepted definition of what constitutes mortgage default (U.S. Department of Housing and Urban Development 2007). This definition is so vague that some people may interpret skipping a single payment as default, which we define as delinquency above. Others also may define default as completion of foreclosing on, or acquisition of mortgaged properties by lenders6. This difference in the definition of mortgage default is unavoidable because usually there is a lapse of several months between start of missing payments and foreclosure (or other post-delinquency outcomes). Since legal proceedings of foreclosure are costly to lenders and delinquent mortgages are often collateralized by properties whose value fell below the loan balance, lenders are willing to work with nonpaying borrowers and renegotiate loan terms as long as it is less costly to them than resorting to foreclosure. Therefore, the number of periods after which foreclosure does take place depends on lenders’ decisions. Three months of delinquency is the prevailing standard to judge the probability of foreclosure in practice (Quercia and Stegman 1992). That is why three month of delinquency is sometimes termed “serious delinquency” (e.g. Ong, Sing and Teo 2007). Bankruptcy, or personal bankruptcy to be exact, is more straightforward to define, compared to mortgage default, since it is triggered by a deliberate action of borrowers, bankruptcy petitions. In the United States bankruptcy law, there are two chapters providing options for personal bankruptcy: Chapter 7 and 137. Although the legal treatment of bankruptcy underwent significant changes by the amendment in 2005 as mentioned above, the analysis in this paper is based entirely on the bankruptcy law before the 2005 changes, because the dataset that I use originates in 2004 at the latest.

6

Ambrose and Capone (1996) show that since foreclosure is the most costly post-delinquency outcome from lenders’ perspective, lenders prefer less costly outcomes, such as a deed-in-lieu-of-foreclosure or preforeclosure sale. 7 Mortgage borrowers can also file for Chapter 11 bankruptcy if they are business owners.

5

Chapter 7, or “straight bankruptcy”, is intended to offer borrowers a “fresh start” by discharging most unsecured debts and liquidating their nonexempt assets. However, secured loans (including mortgage loans), student loans, child support obligations, and debt incurred by fraud cannot be discharged according to Chapter 7. On the other hand, creditors are not qualified to liquidate any assets of debtors, but creditors are entitled to claim all of debtors’ assets above fixed levels of bankruptcy exemptions. An interim trustee, who is appointed by the court when bankruptcy is filed for, oversees the liquidation of nonexempt assets and distributes the proceeds to repay creditors. Although bankruptcy operates under federal law, states have the right to set their own bankruptcy exemptions. Most states have adopted different levels of exemptions for borrowers’ home equity, which is called homestead exemption, along with bankruptcy exemptions for a variety of other types of assets8. While borrowers do not have to pay anything from properties whose values are less than corresponding exemption levels, mortgage lenders have a priority on liquidation of mortgaged properties. Thus the trustee does not allow creditors to liquidate mortgaged properties if there would be no proceedings left beyond the sum of homestead exemption and mortgage balance. In this case, if debtors who are not behind in their mortgage payments wish to retain their properties, they can do so by reaffirming the mortgage debts. Chapter 13, which is also known as a wage earner proceeding9, requires borrowers to propose a plan to repay part of their debts from future earnings, usually over three to five years, instead of discharging unsecured debts immediately. Even though Chapter 13 discharges some types of debts which are not discharged under Chapter 7, liabilities incurred by fraud, for instance,

8 9

White (1998) presents a state-by-state list of bankruptcy exemptions. Brueggeman and Fisher (2005).

6

more than 70% of personal bankruptcy filings occur under Chapter 710. White (2007a) argues that this popularity of Chapter 7 should be attributed to a couple of reasons; borrowers are not obliged to repay from their future earnings under Chapter 7. Moreover, they can even protect their wealth by converting it from a nonexempt form to an exempt form before filing for Chapter 711. While the descriptions above have shown mortgage default and personal bankruptcy are different in their procedures to some extent, the number of people who use these devices is steadily increasing. Although I do not have access to data on the rates of mortgage foreclosures, LaCour-Little (2004) shows that the rates of foreclosure on government-insured loans have tripled during the period between 1986 and 2002, while the rates of bankruptcy per capita have also tripled during the same period12. Thus the long-term trends of mortgage default and personal bankruptcy filings are similar to one another and increasing at the same pace.

1.3.

Literature review There is a rich literature both on mortgage default and on personal bankruptcy. The

empirical study of personal bankruptcy seems slightly lagging behind that of mortgage default due to data availability. Indeed, the development of the literature in these two fields is quite similar to one another13. In both fields, there are two streams, the willingness-to-pay theory and the ability-to-pay theory14, and these two types of theories coexist being supported by empirical studies. Mortgage default can be viewed as a financial option that gives borrowers the right to 10

Bankruptcy Statistics in 2004, U.S. Courts. (http://www.uscourts.gov/bnkrpctystats/bankruptcystats.htm) Bahchieva, Wachter and Warren (2005) point out this possibility indicating that the loan-to-value ratio of mortgage borrowers who filed for bankruptcy is substantially increasing recently. 12 Author’s calculation based on Bankruptcy Statistics, U.S. Courts, and Population Estimates, U.S. Census Bureau. (http://www.census.gov/popest/estimates.php). 13 Lin and White (2001, pp.140) point out this fact. 14 This terminology follows U.S. Department of Housing and Urban Development (2007).

11

7

sell the mortgaged properties to the lenders in exchange for release of their debt contracts. A straightforward interpretation tells us that borrowers will default if their home equity is negative, or “in the money”. On the contrary, they will keep paying for their mortgages if the home equity is positive, or “out of the money”. This simple formulation may well describe borrowers’ willingness to pay off their mortgage. One of the seminal works in the mortgage default area is Foster and Van Order (1984), who formally interpret mortgage default as exercising a “put” option and apply option theory to their empirical study. Their empirical work shows the option-based model of mortgage default is so effective that it explains over 90 percent of the variance just by the home equity variables. On the other hand, they also find the default option is not exercised immediately, since lagged variables for households’ equity positions have statistically and economically significant effects. This result suggests that the simple option-based theory cannot explain borrowers’ behavior perfectly. Deng, Quigley and Van Order (2000) remedy this deficiency by introducing another option, prepayment of mortgage, for which a “call” option model seems appropriate. While the simultaneous option model of Deng, Quigley and Van Order (2000) works well, their contributions are not limited to the introduction of the prepayment option. Their empirical results also support the theory of borrowers’ ability to pay. Since mortgage loans require a stringent commitment by borrowers, in other words, their making regular payments, typically every month, researchers in the field of mortgage default have been paying attention to the affordability of mortgage payments, or borrowers’ ability to pay. While this ability is basically measured by the household’s income and the amount of mortgage payment, this theory also puts much emphasis on unexpected events, often called “trigger” events, that shock borrowers’ ability to pay, such as unemployment and divorce. These two events, unemployment and divorce, are

8

exactly what Deng, Quigley and Van Order (2000) find to have significant effects in their model. The other previous studies uncover the significant factors of this type. For instance, Vandell and Thibodeau (1985) the default risk of self-employed borrowers are higher probably due to their income volatility. It is important to note that the credit constraint will be a fundamental background to the ability-to-pay theory from the perspective of the consumer choice theory. There will be a case in which, even though borrowers hope to keep their homes expecting financial profits, they have to default because they run out of cash flow for regular payments and cannot borrow additional money to handle temporary financial adversities. In addition to unemployment and divorce, Elul (2006) argues that the explanatory power of original mortgage loan-to-value ratio found in the empirical study of Deng Quigley and Van Order (2000) provides evidence for the existence of credit constraints, since borrowers who have less wealth available for down payments are more likely to be credit constrained. Elmer and Seelig (1999) argue that the option-based willingness-to-pay theory has its limitations in describing borrowers’ behavior about mortgage default, and that mortgage default should be analyzed with a choice-theoretic framework that integrates the willingness-to-pay theory and the ability-to-pay theory. The situation is similar in the field of personal bankruptcy studies. Sullivan, Warren and Westbrook (1989) propose an ability-to-pay theory, namely that borrowers file for bankruptcy only when an unanticipated event occurs that reduces their ability to meet their loan payments. On the contrary, White (1998) and Fay, Hurst and White (2002) introduce a willingness-to-pay theory by establishing the framework to measure borrowers’ financial benefit from bankruptcy filings that incorporates their unsecured debts, nonexempt assets, bankruptcy exemptions and the costs of filing for bankruptcy, such as legal fees and limited access to credit in the future. The

9

empirical work of Fay, Hurst and White (2002) shows that the financial benefit from bankruptcy filings has significant impact on borrowers’ decisions on bankruptcy. However, they also find a household’s income, which is closely related to the ability-to-pay theory, affects decisions. Fisher (2005) reaffirms the empirical evidence for the effect of household’s income even with his more detailed specification. To summarize, in both fields of the study of mortgage default and personal bankruptcy, the willingness-to-pay theory is clearly well-defined, but leaves no room for nonfinancial factors to affect these decisions from a purely theoretical viewpoint. This theory is given plenty of strong empirical support. Nonetheless, the empirical works also suggest the validity of trigger events as well as financial variables, supporting the ability-to-pay theory. The credit constraint is one of the most promising candidates for reconciliation between those seemingly conflicting theories. Thus one of main objects of this thesis is to investigate how credit constraint functions for mortgage default and personal bankruptcy.

1.4.

Determinants of borrowers’ decisions of default, delinquency and bankruptcy

1.4.1. Dependent and explanatory variables in the SCF In the rest of this chapter, I will provide an overview of determinants of the probability of delinquency, default and bankruptcy filing by reduced-form models using a dataset from SCF in 1998, 2001 and 2004. The following two chapters also use datasets from the SCF. Although the SCF has some disadvantages, particularly compared to panel datasets, such as the Panel Study of Income Dynamics (PSID)15, the SCF has such significant advantages that make it the best choice to use the SCF datasets in this thesis. First of all, the SCF surveys households’ payment

15

I will discuss the disadvantages of the SCF and the measures that we can take in order to reduce the disadvantages in more detail in the following two chapters.

10

performance for each detailed category of debts as well as the amount of payments and current balance for those loan categories. In particular, these variables allow us to derive implications for borrowers’ simultaneous choices between mortgage default and bankruptcy, which is discussed in Chapter 2 in more detail. Second, since one of the main objectives of this thesis is the behavior of credit constrained households, information about credit constraints is crucial. In this regard, the questions about households’ loan applications and its denials that the SCF asks are very useful. The SCF also has another unique variable relevant to my study, which is about households’ attitude towards financial risks and profits. Lastly, the SCF contains very detailed information on the characteristics of each household’s residential mortgage, including whether it is subject to adjustable rates or not. Table 1-1 summarizes the variables regarding delinquency, default and bankruptcy that the SCF investigates. As for loan delinquency, the SCF asks the households if they are paying each type of loans, such as mortgage loans, car loans and consumer loans, as scheduled. It also inquires as to their loan payment history in general in the previous year. That is, whether they missed any loan payments at least once during the previous year, which is followed by an additional question whether they did so for three months or more. The latter is termed “serious delinquency”. The SCF started asking about personal bankruptcy experiences since its 1998 survey. Specifically, it asks whether the households have ever filed for bankruptcy and, if they had, when the most recent filings were. Table 1-2 reports the frequencies for households that are in delinquency or filed for bankruptcy in the SCF of 1998, 2001 and 2004. Since the plan of this thesis is to analyze behavior of households who have mortgage debts, the rates in Table 1-2 are about the mortgage-indebted households only. All of the three variables regarding delinquency show the

11

same pattern: they peaked in 2004 after dropping once in 2001. This trend agrees with the data published by U.S. Department of Housing and Urban Development (2007) and discussed by LaCour-Little (2004). The movement of the rates of bankruptcy filings does not follow precisely the trend in the United States during that period shown in Figure 1-1. This might be attributed to the fact that the data in Figure 1-1 are based on the entire population in the United States. My dataset from the SCF only includes the homeowners with mortgage debts; the bankruptcy filing trend of the mortgage debtors may be different from the nationwide trend. Previous studies have found a variety of factors affecting households’ decisions on default and bankruptcy, as discussed in the literature review. The relevant variables used in this thesis are summarized in Table 1-3. From the viewpoint of the theory of borrowers’ ability to repay, household’s income and the amount of mortgage payment should have important effects on borrowers’ decisions on default and bankruptcy, along with “trigger events” such as losing jobs and an unexpected decrease in income. Credit constraints will also be critical since credit constrained households could not mitigate such unexpected adversities by taking on additional loans. On the other hand, the theory of borrower’s willingness to repay emphasizes that the main determinants should be borrowers’ equity position in mortgaged property, which is measured by the variables such as loan-to-value ratio (LTV). To test the impact of these factors on borrowers’ decisions on default and bankruptcy filings, I perform several estimation for delinquency, serious delinquency and bankruptcy respectively, with the explanatory variables contained in the SCF, as described in Table 1-4. I include several variables that are not discussed above. For example, the non-labor income ratio may provide a cushion against unexpected shocks to wage income. Thus the effect will be negative on the probability of delinquency and bankruptcy.

12

I take advantage of the detailed information about the characteristics of mortgage loans that the SCF provide by including dummy variables for adjustable rate mortgages (ARMs), mortgage refinancing and home equity loans in the estimations to follow. Since ARMs change their interest rates according to the present economic situations, usually connected to the interest rates of other financial assets, they may cause shocks to borrowers’ ability to pay by increasing the interest rates. In addition to reducing borrowers’ home equity, home equity loans can be exploited to take more advantageous results out of bankruptcy filing. That is, borrowers can protect their home from liquidation by reducing their home equity below the level of homestead exemption. While ARM and home equity loans are often blamed for the recent jump in the rates of mortgage default, households who could manage to refinance their mortgages are more likely to be able to avoid giving up their homes. I will also look at the effects of the variables regarding household-level economic characteristics. In particular, Vandell and Thibodeau (1985) find self-employed households are more likely to default due to their volatility in income. The SCF enables us to test this effect. Because a variety of types of mortgage products are rapidly being developed and the origination of nontraditional mortgages accounts for a substantial fraction of the U.S. mortgage market (Edmiston and Zalneraitis, 2007), professional financial counseling became more important for borrowers, most of whom are not professionals in finance. I expect financial counseling will have some effect in preventing borrowers from financial distress, in other words, default or bankruptcy. Moreover, two distinctive variables of the SCF are used in the analysis: that is, dummy variables for credit constraints and households’ attitude towards financial risks. The former is generated based on the questions in the SCF about results or expectations of households’ loan

13

applications. On the other hand, households’ attitude towards risks is measured by a hypothetical question about financial opportunities. Since both variables play a key role in the argument in Chapter 2, I will give more detail, including precise description of relevant questions in the SCF, in that chapter. Lastly, I include a set of demographic variables regarding sex, age, race, marital status, family member and education. Among them, we may well expect marital status and supporting children influence borrowers’ ability to pay; married couples will have flexibility in supplying working hours, while supporting children implies more inflexible spending due to mandatory expenditures, such as education expenditure.

1.4.2. Probit estimations for delinquency, default and bankruptcy The explanatory variables discussed above are incorporated into a reduced-form model for the probabilities of delinquency, serious delinquency (delinquency for three months or more), mortgage delinquency and recent bankruptcy (bankruptcy filings within two years) respectively, as follows: Pr ( y i = 1 | x i ) = F (β ′x i ), i = 1,K, N

(1.1)

where yi is a binary variable for the i-th household that takes the value one, if the event corresponding to the regression, say delinquency, is true, and zero otherwise. x i is the vector of explanatory variables for the i-th household, which are listed in Table 1-4. β is the vector of the parameters to be estimated for each dependent variable. F (⋅) is an appropriate probability function of the model. If we assume the probit function is applied to F (⋅) , Equation (1.1) gives; Pr ( y i = 1 | xi ) =

β′x i

∫ φ (t )dt ,

i = 1, K , N ,

−∞

14

where φ (⋅) denotes the probability density function of the normal distribution. Table 1-5 presents the results of the maximum likelihood estimations of this probit specification for the probability of delinquency, serious delinquency, mortgage delinquency and recent bankruptcy, respectively. The first three columns, whose dependent variables are delinquency on any loans, serious delinquency and mortgage delinquency, share the basic features. The variables related to borrowers’ ability to pay such as income, mortgage payment-to-income ratio, unemployment and unexpected low income, have statistically significant coefficients and, more importantly, the magnitude of the coefficient on credit constraint is the most notable. Meanwhile, the variables corresponding to the other theory, the willingness-to-pay theory, also have significant impact. The LTV has consistently significant impact across all the three columns. In addition, for the probability of mortgage delinquency, original housing price has a negative and statistically significant effect, while original mortgage balance has a positive effect, even though the latter effect is marginally significant. This agrees with the result of Deng Quigley and Van Order (2000) that shows original LTV is positively related to mortgage default risk, even after controlling for current borrowers’ position in home equity16. The professional financial counseling has an effect in reducing the probability of delinquency. As for mortgage characteristics, adjustable rate mortgages and holding home equity loans increase the probability of delinquency significantly. Refinancing mortgages has a negative effect on the probability of delinquency, particularly serious delinquency and mortgage delinquency. Some of the

16

Although Elul (2006) interprets the effect of original LTV as an evidence for the existence of credit constraints in order to support the ability-to-pay theory, the result here shows that the effect of original LTV seems to persist even after, at least partially, controlling credit constraints. This strongly supports the argument of Yezer, Phillips, and Trost (1994); they argue that there is an asymmetric information problem between mortgage lenders and borrowers and that risky borrowers are more likely to finance their mortgage with higher LTV.

15

demographic variables, such as age, college degree and supporting children, also have explanatory powers as expected. However, there are some variables that are expected to play a key role but are not estimated as significant, or even opposite to the predictions. The payment-to-income ratio of nonmortgage loans has a statistically significant effect on the probability of delinquency only. I will address the relation between loan payments for mortgages and nonmortgage loans in Chapter 2. The value of home equity has a positive effect on the probability of mortgage default, which might be attributed to the correlation between home equity and LTV, which has strong positive effects as mentioned above. The last column in Table 1-5 presents the probit estimation for the probability of recent bankruptcy filings17. Although we can see several similarities between the results of delinquency and bankruptcy, there are clear differences as well. Most importantly, the coefficient on household’s income is not significant, which might suggest that households’ decisions on bankruptcy filings is more strategic, or more “ruthless”, than those on default, even though it would be impetuous to make that conclusion just based on these estimations. In addition to income, the effects of unexpectedly low income has no longer statistically significant coefficient, and its sign is even negative. This also supports weaker effects of borrowers’ ability to pay on bankruptcy filing decisions than on delinquency. Moreover, we can see a positive and statistically significant effect of holding home equity loan. That might suggest that bankruptcy petitioners are exploiting the homestead exemption to protect their homes by borrowing against their home

17

The probit model for recent bankruptcy filings does not include as explanatory variables share of non-labor income and payment-to-income ratio of unsecured debts. That is so because households’ assets, which are the main source of non-labor income, would have been liquidated and unsecured debts would have been discharged during bankruptcy proceedings. This issue is discussed in Chapter 3 in more detail.

16

equity so that their home equity falls below the level of homestead exemption18. This hypothesis provides another evidence for borrowers’ strategic behavior. I will investigate further this issue of their strategic bankruptcy decisions in Chapter 3. To sum up, the results of the probit estimations above support both of the theories, ability-to-pay and willingness-to-pay, of borrowers’ decisions on delinquency and bankruptcy. These agree exactly with the findings of previous studies, while the result for bankruptcy decisions appears to be closer to the side of the latter theory.

1.4.3. Ordered probit specification for delinquency and default The SCF does not inquire whether mortgage borrowers are in default in the sense of the definition made in most of the literature on mortgage default; those studies define mortgage default as foreclosure on mortgaged properties or lenders’ acquisition of them (e.g. U.S. Department of Housing and Urban Development, 2007). Nonetheless, since delinquency for three months or more is usually considered as a positive indicator of default in practice, we can use the “serious delinquency” in the SCF, which is defined as delinquency for three months or more, as a good proxy for default19. This means we are observing two distinct events, delinquency and default (proxied by “serious delinquency”). Although previous research on the relationship between mortgage delinquency and default is limited, compared to the rich literature on mortgage default decisions, recently some studies shed light on this relationship. For example, Ambrose

18

Barchieva, Wachter and Warren (2005) support this hypothesis more clearly from another viewpoint. They estimate the probability that households have high LTV when they file for bankruptcy. In that estimateion, the dummy variables for high-exemption (homestead exemption) states has a significant and negative effect. They argue that this result suggests households in states with lower homestead exemption have motivation to make their LTV higher so that the homestead exemptions of those states can fully cover their home equity. 19 In fact, the variables “delinquency” and “serious delinquency” in the present chapter represent missing payments for any types of loans rather than for mortgage loan only. Nonetheless, I assume that these two variables may well represent the mortgage payment trouble appropriately since LTV and mortgage payment-to-income ratio play an important role in the probit estimations for these two variables as shown in Table 1-4.

17

and Capone (1998) argue that the relationship between delinquency and default are continuous from borrowers’ perspective; for some reason, borrowers become delinquent as the first step for default, and then, default occurs when they are unable to take actions to avoid foreclosure or intentionally allows foreclosure to occur20. As a way to test this continuous link between delinquency and default, I perform the estimation of the ordered probit specification. Suppose the indirect utility functions of the i-th household who holds mortgage is represented as follows:

V1* = α 1 + β ′1 x i + ε 1,i ; V0* = α 0 + β ′0 x i + ε 0,i , where V1* denotes the i-th household’s optimal utility conditional on default that it realizes when it becomes delinquent, while V0* denotes its optimal utility conditional on continuing payment. Parameters α1 and α 0 are the choice-specific constants for default and continuing payment respectively. Similarly, β ′1 and β ′0 are the transposed vectors of choice-specific parameters, while ε 1,i and ε 0,i are error terms; x i is the vector of explanatory variables of i-th household. Then the i-th household chooses to delinquency if;

(β1′ − β ′0 )x i + (ε 1,i − ε 0,i ) > α 0 − α 1 . As mentioned above, when borrowers miss mortgage payments, lenders usually try to persuade them into resuming repaying by renegotiation over the terms of contracts for a certain period, typically three months. At the same time, financial distress that borrowers faced when they became delinquent may be alleviated during this period. Since such course of events makes

20

Ambrose and Capone (1998) argue that there are two distinctive reasons for borrowers to become delinquent: that is, to exercise the implicit put option or from a decision to use mortgage delinquency to finance other expenditures.

18

default unattractive to borrowers, the indirect utility conditional on default changes after this process as follows; ~ V1* = α~1 + β ′1 x i + ε 1,i ,

α~1 < α 1 ,

~ where V1* is the i-th household’s optimal utility conditional on default three months after its

delinquency started, and α~1 is a constant in this indirect utility function different from α1 . I assume here that the parameters for explanatory variables, β ′1 , never change because it is hypothesized that the incentives for delinquency and default are incorporated in the same utility function. Since now default is less attractive than used to, α~1 is smaller than α1 . Thus, the i-th household finally chooses to default if:

(β1′ − β ′0 )x i + (ε 1,i − ε 0,i ) > α 0 − α~1 . Then we can rewrite: yi = 0 if yi* ≤ µ1 ; yi = 1

if µ1 < yi* ≤ µ 2 ;

yi = 2 if µ 2 < y i* ,

where y i* = (β1′ − β ′0 )x i + (ε 1,i − ε 0,i ) , µ1 = α 0 − α 1 , and µ 2 = α 0 − α~1 . yi is a discrete-values variable, which is equal to one, when the household becomes delinquent,

equal to two, when the household defaults, and equal to zero otherwise. If we assume ε 1,i and

ε 0,i are normally distributed, then this gives the probabilities of these events as follows21: Pr ( y i = 0 | x i ) = Φ (µ1 − β ′x i ) ;

21

See Greene (2008, pp.831-832) for details.

19

Pr ( y i = 1 | x i ) = Φ (µ 2 − β ′x i ) − Φ (µ 1 − β ′x i ) ; Pr ( y i = 2 | x i ) = 1 − Φ(µ 2 − β ′x i ) ,

s

where Φ(s ) = ∫ φ (t )dt . −∞

Thus this specification requires the estimated cutoff values to satisfy the condition as follows:

µ1 < µ 2 .

(1.2)

Table 1-6 reports the result of this ordered probit specification. First of all, I obtain the cutoff values satisfying (1.2). In addition, the estimated coefficients are consistent with the outcomes of the last binomial probit estimations. The factors related to the ability to pay, the willingness to pay, mortgage characteristics and demographic characteristics have significant effects in this specification, agreeing with the predictions. This result implies that default and delinquency are basically motivated by the same factors.

1.4.4. Structural change of the impact of ARMs

Among a good number of the articles concerning the recent dramatic increase in the rate of mortgage foreclosure22, Edmiston and Zalneraitis (2007) attribute this unusual surge of foreclosure to the growing popularity of ARMs in the U.S. mortgage market along with subprime loans. They argue that the shock to borrowers’ ability to pay caused by the change of the interest rates applied to ARMs can be a strong default trigger, which they show by calculating hypothetical payment shocks based on the recent increase in the short-term interest rate. Even though, according to their data, the foreclosure rate of ARMs started to rise in 2006, which is a period that is out of the scope of my current study, I try to measure the structural change of the ARM’s impact on default by allowing the ARM term to take different coefficients across time

22

In the second quarter of 2007, the share of outstanding mortgages in some stage of foreclosure stood at 1.4 percent, near historic highs and up from less than 1 percent a year earlier (Edmiston and Zalneraitis, 2007).

20

between 1998 and 2004 in the ordered probit specification. Since the United States experienced a considerably low level of short-term interest rate between 2002 and 2004 along with booming housing markets23, that might have affected people’s expectations about economic situations in the future such that they were less aware of the risk of fluctuations of ARM payments. Table 1-7 reveals, indeed, the coefficient on ARM is statistically significant only in the most recent survey, 2004. In addition, the Wald test rejects the null hypothesis that all the ARM coefficients are equal across time at p-value of 0.0087. Therefore, we can see through this result that the impact of ARMs on mortgage default emerged even before 2006. Edmiston and Zalneraitis (2007) also point out that the deceptive attractiveness of ARMs, referring to “teaser rate”, may have generated payment shocks to borrowers who made incautious investments in housing, while the U.S. Department of Housing and Urban Development has been promoting education of mortgage borrowers and encouraging prospective mortgage borrowers to seek professional counseling. These stories imply that professional financial counseling seems effective in protecting mortgage borrowers from ending up involved in the wretched result, in other words, default. Thus, I also try to capture the importance of financial counseling in choosing ARMs. Table 1-8 reports the result of the ordered probit estimation for only the households who have ARMs, incorporating different slopes for the variable of financial advice across years24. Since now this model focuses on a specific subsample, that is, ARM holders, some of the coefficients change their signs and/or magnitudes. Nonetheless, the key variables, such as LTV,

23

For example, the nominal average rate of 3-month T-bill was less than 2% during this period, while that rate was moving around 5% in late 1990’s. (Board of Governors of the Federal Reserve System: http://www.federalreserve.gov/econresdata/default.htm) 24 This is a binary variable which takes the value one if the household says that when it makes decisions about credit or borrowing, it uses any information from professional people: namely, lawyers, accountants, bankers, brokers and financial planners.

21

payment-to-income ratio and credit constraint, remain as expected. The cutoff values are also consistent. In recent years, 2001 and 2004, professional advice has statistically significant impact on ARM borrowers, helping them stay away from default, while it has almost no effects in 1998. The Wald test also rejects the null hypothesis that the coefficients are equal coefficients across years (p-value = 0.0007). This result may reflect the fact that mortgage products, particularly ARMs, have been elaborately developed and getting too complicated for nonprofessional borrowers. These findings provide some evidence for the argument that the volatility of ARM payments and reckless mortgage originations without prudent plans are responsible for the recent rise in the rate of foreclosure.

1.5.

Conclusion

This chapter presents several estimation results to investigate explanatory variables for mortgage borrowers’ decisions on delinquency, default and bankruptcy filing along with reviewing research on mortgage default and personal bankruptcy. The empirical works suggest that both groups of variables related to the willingness-to-pay and the ability-to-pay theory have significant impacts on borrowers’ decisions on mortgage default and filing for bankruptcy. Across all the estimations reported in this chapter, credit constraint has persistent and the most significant effect on the probabilities of delinquency, default and bankruptcy. The issue of credit constraints is the main focus in the following two chapters. On the other hand, there are differences in determinants between delinquency and bankruptcy as well as a lot of similarities. For example, the household’s income has a much weaker effect on bankruptcy decisions, compared to mortgage delinquency decisions. Further investigation of the model of mortgage default decisions in terms of structural

22

changes implies that ARM borrowers and reckless housing investments without professional advice may account to some extent for the recent rapid increase in the foreclosure rate in the United States.

23

Figure 1-1 Bankruptcy filing rate 1996-2004 0.006 0.005 0.004 0.003 0.002 0.001 0 1996

1997

1998

1999

2000

2001

2002

Year Source: Bankruptcy Statistics, U.S. Courts. (http://www.uscourts.gov/bnkrpctystats/bankruptcystats.htm)

24

2003

2004

Table 1-1 Delinquency and bankruptcy variables in SCF Description

Name (1)

Time

Delinquency

Whether missed any loan payments

Previous year

Serious delinquency

Whether missed any loan payments for three months or more

Previous year

Mortgage delinquency

Whether paying for mortgage loans behind schedule (2)

Bankruptcy

Whether ever filed for bankruptcy

Time of Bankruptcy

The number of years since most recent bankruptcy filing (3)

Time of the survey Any time in the past Time of the survey

(1) Name in this thesis. (2) The SCF also asks the same questions about other types of loans, which are used in Chapter 2. (3) The numbers are coded as the nearest odd numbers.

Table 1-2 The percentage of delinquency and bankruptcy among households with mortgage debts 1998

2001

2004

Average

14.28

13.47

15.73

14.55

(10.27)

(10.69)

(12.40)

(12.40)

5.31

4.25

5.49

5.03

(3.26)

(3.19)

(3.64)

(3.37)

2.41

2.27

3.18

2.65

(1.47)

(1.76)

(2.03)

(1.76)

1.52

0.9

0.84

1.08

(2.24)

(1.31)

(1.20)

(1.55)

10.05

10.58

10.77

10.49

(7.73)

(8.08)

(8.51)

(8.12)

Delinquency

Serious delinquency

Mortgage delinquency

Bankruptcy within 2 years

Bankruptcy in the past Note: The weight provided in SCF are used. Unweighted values are inside the parentheses.

25

Table 1-3 List of main explanatory variables Predicted sign

Variable

Mortgage default

Bankruptcy

+ + ? -

? ? + + ?

+ ? + + + + + +

? + ? + + + + +

Willingness to pay Loan-to-value ratio (LTV) Home equity Home equity loan Unsecured debt Expected capital gains

Ability-to-pay Income Share of non-labor income Mortgage payment Nonmortgage payment Adjustable rate mortgages (ARM) Married Children Unemployed Self-employed Unexpectedly low income Credt constraint

26

Table 1-4 Summary statistics of explanatory variables Variable

Description

Mean

Std. Dev.

LTV

mortgage balance divided by current housing value

0.5016

0.2732

Home equity

current housing value less mortgage balance

367,745

956,768

Original housing price

original housing price

306,039

684,657

Original mortgage balance

original mortgage balance

218,769

377,618

Mortgage PTI

monthly mortgage payment divided by total income

0.0158

0.0267

Nonmortgage PTI

monthly nonmortgage payments divided by total income

0.0040

0.0138

ARM

1 if mortgage is an adjustable rate mortgage

0.1649

Home equity loan

1 if household holds home equity loan

0.1642

Refinance

1 if household ever refinanced mortgages

0.4716

Log income

log of total income

11.636

1.320

Share of non-labor income

non-labor income divided by total income

0.1157

0.2694

Age

household head's age

48.03

11.68

Age squared

household head's age squared

2443.63

1182.88

Female

1 if household's head is female

0.1165

Black

1 if household's head is black

0.0645

Hispanic

1 if household's head is Hispanic

0.0522

Married

1 if household is formed by a married couple

0.7559

Children

1 if household is supporting children

0.5751

College degree

1 if household's head has college degree

0.5833

Unemployment

1 if household's head experienced unemployment within 12 months

0.0614

Self employed

1 if household's head is self-employed

0.3236

Unexpected low income

1 if household's income was unexpectedly low in the previous year

0.1354

Professional advice

1 if household ever asked professional financial advice

0.5510

Credit constraint

1 if household was rejected for loan applications or expects rejection

0.1687

Willingness to take risk

1 if household is willing to take average financial risk or more risk

0.8093

Observations

26,870

Year 2004

9,445

Year 2001

8,912

Year 1998

8,513

27

Table 1-5 The probit model for mortgage borrowers' decisions Delinquency Coefficient t LTV Home equity

Serious delinquency Coefficient t

Mortgage delinquency Coefficient t

Recent bankruptcy Coefficient t

0.355

(7.38)***

0.490

(7.48)***

0.336

(4.44)***

5.51.E-01

(6.18)***

-1.73.E-07

(-2.37)**

1.37.E-07

(1.06)

1.60.E-07

(4.52)***

-1.38.E-06

(-2.66)***

Original housing price

1.90.E-07

(2.12)**

-6.45.E-07

(-1.75)*

-3.75.E-07

(-2.28)**

9.37.E-07

(3.26)***

Original mortgage balance

-9.57.E-07

(-6.21)***

-1.62.E-06

(-3.41)***

2.64.E-07

(1.60)

-1.83.E-06

(-3.41)***

Mortgage PTI

1.112

(2.66)***

1.862

(3.37)***

1.491

(3.04)***

1.264

(1.85)*

Nonmortgage PTI

2.050

(3.10)***

-0.711

(-0.60)

1.034

(0.89)

ARM

0.088

(2.78)***

0.166

(3.52)***

0.150

(2.69)***

-0.085

(-1.08)

Home equity loan

0.099

(3.21)***

0.121

(2.47)**

0.229

(3.99)***

0.296

(4.20)***

Refinance

-0.028

(-1.15)

-0.080

(-1.99)**

-0.174

(-3.45)***

-0.124

(-2.00)**

Log income

-0.067

(-3.31)***

-0.147

(-3.89)***

-0.158

(-4.20)***

-0.065

(-1.18)

Share of non-labor income

-0.084

(-1.23)

-0.285

(-1.76)*

0.163

(1.59)

Age Age squared

0.053

(7.60)***

0.058

(5.20)***

0.021

(1.74)*

0.061

(3.68)***

-5.77.E-04

(-7.98)***

-6.39.E-04

(-5.37)***

-2.59.E-04

(-2.05)**

-5.56.E-04

(-3.24)*** (0.40)

28

Female

0.015

(0.36)

-0.008

(-0.13)

-0.079

(-1.05)

0.036

Black

0.310

(8.02)***

0.050

(0.89)

0.350

(5.64)***

0.109

(1.38)

Hispanic

0.114

(2.60)***

-0.001

(-0.01)

0.129

(1.73)*

0.297

(3.64)***

Married

-0.113

(-3.33)***

0.002

(0.04)

-0.070

(-1.10)

-0.131

(-1.69)*

Children

0.299

(11.14)***

0.287

(6.58)***

0.250

(4.72)***

0.386

(5.70)***

College degree

-0.215

(-8.48)***

-0.083

(-2.07)**

-0.098

(-1.94)*

-0.242

(-3.75)***

Unemployment

0.290

(7.31)***

0.334

(6.41)***

0.292

(4.67)***

0.273

(3.54)***

Self employed

0.072

(2.54)**

0.024

(0.50)

-0.265

(-4.02)***

0.141

(1.95)*

Unexpected low income

0.087

(2.71)***

0.218

(4.77)***

0.182

(3.27)***

-0.094

(-1.23)

Professional advice

-0.054

(-2.33)**

-0.195

(-5.26)***

-0.159

(-3.44)***

-0.092

(-1.65)*

Credit constraint

0.778

(30.35)***

0.820

(21.99)***

0.607

(13.04)***

0.604

(10.77)***

Willingness to take risk

-0.066

(-2.34)**

-0.207

(-5.09)***

-0.253

(-5.15)***

-0.334

(-5.72)***

Year 2004

0.239

(8.31)***

0.211

(4.62)***

0.160

(2.83)***

-0.162

(-2.41)**

Year 2001

0.103

(3.58)***

0.093

(2.02)**

0.154

(2.74)***

-0.166

(-2.56)***

Constant

-2.045

(-7.98)***

-1.897

(-4.36)***

-1.199

(-2.67)***

-3.309

(-5.21)***

Observations

26870

26870

26870

26870

Log likelihood

-7626.17

-2916.89

-1844.42

-1208.72

0.184

0.259

0.221

0.235

Pseudo R squred

Table 1-6 The ordered probit model for mortgage borrowers' decisions on default Default Coefficient

t

LTV

4.02.E-01

(8.61)***

Home equity

-1.38.E-07

(-1.91)*

Original housing price

1.86.E-07

(2.18)**

Original mortgage balance

-1.08.E-06

(-7.06)***

Mortgage PTI

1.371

(3.50)***

Nonmortgage PTI

1.748

(2.66)***

ARM

0.101

(3.31)***

Home equity loan

0.095

(3.16)***

Refinance

-0.037

(-1.53)

Log income

-0.072

(-3.61)***

Share of non-labor income

-0.091

(-1.34)

Age

0.054

(7.93)***

-5.82.E-04

(-8.28)***

Female

0.005

(0.13)

Black

0.260

(7.00)***

Hispanic

0.092

(2.17)**

Married

-0.106

(-3.23)***

Children

0.295

(11.27)***

College degree

-0.191

(-7.72)***

Unemployment

0.309

(8.17)***

Self employed

0.057

(2.06)**

Unexpected low income

0.112

(3.61)***

Professional advice

-0.078

(-3.44)***

Credit constraint

0.785

(31.68)***

Willingness to take risk

-0.103

(-3.75)***

Year 2004

0.232

(8.29)***

Year 2001

0.101

(3.59)***

Cut off value 1

1.986

0.251

(1)

Cut off value 2

2.742

0.251

(1)

Observations

26870

Log likelihood

-9310.35

Age squared

Pseudo R squred

0.166

(1) reporting the standard deviations

29

Table 1-7 The ordered probit model for mortgage borrowers' decisions on default (structural change of the effect of ARM) Default Coefficient

t

LTV

4.03.E-01

(8.63)***

Home equity

-1.39.E-07

(-1.93)*

Original housing price

1.87.E-07

(2.18)**

Original mortgage balance

-1.08.E-06

(-7.08)***

Mortgage PTI

1.369

(3.50)***

Nonmortgage PTI

1.748

(2.66)***

in 2004

0.179

(3.80)***

in 2001

0.050

(0.88)

ARM

0.045

(0.83)

Home equity loan

in 1998

0.096

(3.20)***

Refinance

-0.038

(-1.57)

Log income

-0.072

(-3.61)***

Share of non-labor income

-0.090

(-1.32)

Age

0.053

(7.88)***

-5.80.E-04

(-8.24)***

Female

0.005

(0.13)

Black

0.261

(7.02)***

Hispanic

0.091

(2.15)**

Married

-0.108

(-3.28)***

Children

0.295

(11.26)***

College degree

-0.191

(-7.72)***

Unemployment

0.310

(8.20)***

Self employed

0.060

(2.16)**

Unexpected low income

0.112

(3.59)***

Professional advice

-0.077

(-3.42)***

Credit constraint

0.785

(31.69)***

Willingness to take risk

-0.103

(-3.76)***

Year 2004

0.210

(6.88)***

Year 2001

0.099

(3.24)***

Cut off value 1

1.970

0.251 (1)

Cut off value 2

2.725

0.251 (1)

Observations

26870

Log likelihood

-9308.02

Age squared

Pseudo R squred

0.167

(1) reporting the standard deviations

30

Table 1-8 The ordered probit model for ARM borrowers' decisions on default Default Coefficient

t

LTV

1.07.E+00

(7.51)***

Home equity

6.04.E-08

(0.55)

Original housing price

-2.19.E-07

(-0.93)

Original mortgage balance

-4.27.E-07

(-1.35)

Mortgage PTI

2.935

(2.82)***

Nonmortgage PTI

3.021

(1.93)*

Home equity loan

0.082

(1.08)

Refinance

0.050

(0.76)

Log income

-0.081

(-1.58)

Share of non-labor income

-0.260

(-1.70)*

Age

0.110

(5.93)***

-1.09.E-03

(-5.60)***

Female

0.190

(1.79)*

Black

0.229

(2.17)**

Hispanic

-0.231

(-1.87)*

Married

-0.049

(-0.55)

Children

0.253

(3.70)***

College degree

-0.272

(-3.98)***

Unemployment

0.658

(7.52)***

Self employed

-0.152

(-2.08)**

Unexpected low income

-0.028

(-0.33)

in 2004

-0.299

(-3.25)***

in 2001

-0.692

(-5.60)***

in 1998

Age squared

Professional advice

-0.045

(-0.43)

Credit constraint

0.842

(13.78)***

Willingness to take risk

0.008

(0.10)

Year 2004

0.442

(4.52)***

Year 2001

0.344

(3.36)***

Cut off value 1

3.871

0.677 (1)

Cut off value 2

4.651

0.679 (1)

Observations

4430

Log likelihood

-1426.07

Pseudo R squred

0.265

(1) reporting the standard deviations

31

Chapter 2: Delinquency choice between mortgage and nonmortgage loans 2.1.

Introduction

Chapter 1 shows that households’ decisions on delinquency, or missing loan payments, do not look perfectly strategic as the willingness-to-pay theory predicts, being greatly affected by their ability to pay, which is measured by income, mortgage payment-to-income ratio, unemployment and unexpected decreases in their income. These ability-to-pay variables also affect borrowers’ decisions on bankruptcy filings (see Table 1-5). However, separate regression estimations are not enough to investigate the mechanism of borrowers’ simultaneous choices between delinquency, default and bankruptcy, since borrowers are likely to weigh up these financial alternatives, that is, delinquency, default and bankruptcy, simultaneously. I will study further their decisions by focusing on a particular decision of borrowers: choice between mortgage and nonmortgage delinquency in order to investigate this subject. When households who hold both mortgage and nonmortgage loans, such as car loans, education loans or consumer loans, get into financial distress, or consider default or bankruptcy filing to be financially profitable, they may choose to stop payments either for mortgages, for nonmortgage loans, or for both. We expect mortgage delinquency and nonmortgage delinquency to have a strong connection with mortgage default and personal bankruptcy respectively, since mortgage default releases borrowers from mortgage liabilities, while personal bankruptcy discharges borrowers’ unsecured debts but not secured debts, particularly including mortgages. This multinomial choice may well be affected by the same variables as the binomial choice of mortgage default or of personal bankruptcy. Nonetheless, the study of simultaneous and multinomial choice of delinquency between two kinds of loans may provide more insight, since

32

the determinants of mortgage delinquency and those of nonmortgage delinquency are not likely to work in the same way. For example, the expected capital gain from the mortgage property is a financial factor behind mortgage default decisions and will be negatively correlated with the probability of default25. On the other hand, it will probably have an effect in a different way if borrowers plan to file for bankruptcy and expect they can retain their homes after the bankruptcy proceedings because of the homestead exemption. Since most of the existing studies pay attention only to borrowers’ decisions on mortgage default or on personal bankruptcy separately26, investigating borrowers’ choices between mortgage and nonmortgage delinquency can cast light on the theories from another angle, which the traditional literature has failed to capture. This chapter investigates borrowers’ choices with a multinomial choice model. The focus is particularly on credit constraints, capital gain expectations and borrowers’ heterogeneous sensitivity to financial profits. This chapter is organized as follows. The theoretical framework for delinquency decisions is developed in section 2. Section 3 describes the estimation methodology and variables. Estimation results are presented in Section 4. Finally, Section 5 concludes this chapter.

2.2.

Theoretical framework

2.2.1. Delinquency, default and bankruptcy, revisited

Although mortgage borrowers’ choices between mortgage default and bankruptcy (or both) is the main concern of us here, it is difficult to directly and empirically work on this subject,

25

Vandell and Thibodeau (1985) show this both theoretically and empirically by adopting backward-looking expectations. 26 Lin and White (2001) and Chomsisengphet and Elul (2006) formulate borrowers’ decisions between mortgage default and bankruptcy, but their empirical works deal with creditors’ reactions rather than borrowers’ decisions themselves.

33

mainly due to unavailability of data and to the retroactive nature of default decisions. That is, we cannnot distinguish between default and delinquency at the time when borrowers are missing payments, since borrowers may resume to meet their payments afterwards27. Instead, delinquency is easy to observe since it is defined as an action of failing to fulfill the terms of conracts within a particular period. In addition, it is still insightful to explore borrowers’ choices between mortgage and nonmortgage delinquency, since these two kinds of delinquency are closely related to mortgage default and personal bankruptcy, respectively. Given the facts that mortgage delinquency is the logical precursor to mortgage default and that lenders usually charge delinquent borrowers penalty fees (delinquency is not for free), we can expect a sizable fraction of delinquent borrowers at some time are thinking of default eventually. The empirical results of Campbell and Dietrich (1983) support this hypothesis since they show the determinants affect mortgage default and delinquency decisions in the same way28. Furthermore, the estimation results of the ordered probit specification in Chapter 1 gives us further grounds for believing the determinants of default and delinquency are closely related. On the other hand, delinquency on unsecured debts is not necessarily a precursor to bankuruptcy. Technically, households can file for bankruptcy before missing any payments29. However, since delinquency is a passive action, that is, just ignoring due dates, while filing for bankruptcy incurs substantial cost and time, such as legal fees and seeking financial counseling, it is not too restrictive to assume that households become delinquent before filing for bankruptcy in 27

For example, the SCF does not have information about mortgage default in the sense of foreclosure, but has information on delinquency and personal bankruptcy. 28 Although Campbell and Dietrich (1983) succeed in showing the effects of mortgage payment and interest rates work on the probability of delinquency in the same way as default, they fail to show the effect of loan-to-value ratio does so. 29 The amendment of the bankruptcy law in 2005 introduced some restrictions on filing for bankruptcy. For example, Borrowers should take a financial counseling course before filing. See White (2007b) for more information.

34

most cases. In addition, Dawsey and Ausubel (2002) explicitly address the households’ behavior of non-repayment without seeking the formal protection of the bankruptcy, which is termed “informal bankruptcy” in their study. In their model, borrowers make their decisions in two steps. First, they decide whether to stop loan payments or not, and then decide to whether to file for bankruptcy or to stay in informal bankruptcy, conditional on delinquency. Dawsey and Ausubel (2002) find empirically that the homestead exemption affects borrowers’ first-stage decisions since the bankruptcy exemptions apply to them regardless of whether they file for bankruptcy or not30. Thus examining nonmortgage delinquency is relevant to households’ bankruptcy decisions, particularly if we hope to study the effect of the homestead exemption on their decisions. With these relationships between delinquency, mortgage default and bankruptcy, as well as the empirical tractability of observing delinquency, we expect to derive some implications for the relationship between mortgage default and bankruptcy filings from studying borrowers’ choices between mortgage and nonmortgage loans. Figure 2-1 presents the borrower’s decision tree based on the argument above.

2.2.2. Theoretical consideration I: financial profits

In this section, I develop the model I adopt in this chapter. First, the model concentrates on the consequences that mortgage and nonmortgage delinquency often result in: mortgage default and personal bankruptcy. Since, as discussed above, mortgage default and bankruptcy filings occur after delinquency, the determinants of default and bankruptcy will also govern borrowers’ decisions on delinquency to some extent (see Figure 2-1).

30

On the other hand, Dawsey and Ausubel (2002) find that garnishment restrictions mainly affect the second-stage decisions, that is, whether to file for bankruptcy formally or not. That is because borrowers cannot seek garnishment restrictions without filing for bankruptcy.

35

The model assumes that the representative household that holds both mortgage and nonmortgage loans maximizes its utility over a vector of qualitative choices, considering expected financial gains from each choice within a two-period time horizon31. In the first period, the household holds exogenously determined amounts of mortgage and unsecured debt. Its income, housing wealth, and non-housing wealth are also given from outside the model. The household makes a decision between repaying all of its debts, defaulting on mortgage, or filing for bankruptcy to discharge its unsecured debt in the first period, taking account of each outcome that will occur in the second period. We may specify the household’s ex post wealth in the second period as follows: W0 ,t +1 = (Yt − Rt − M t − Dt + At )(1 + r0,t ) + Vt +1 ; W1,t +1 = (Yt − Rt − Dt + At )(1 + r0,t ) ; W2 ,t +1 = min[(Yt − Rt + At ), X ](1 + r0 ,t ) − M t (1 + r0 ,t ) + Vt +1 ,

where W0,t , W1,t and W2,t denote the household’s wealth in period 2 when it chooses repayment, mortgage default and bankruptcy, respectively. The variables are defined as follows: Yt : household’s income; Rt : required nondiscretionary expenditure for the household (assumed to be exogenous); At : household’s non-housing asset; Vt +1 : value of household’s home in the second period; M t : amount of mortgage balance; Dt : amount of unsecured debt;

31

Vandell and Thibodeau (1985), Zorn and Lea (1989) and Lin and White (2001) basically adopt this approach.

36

r0 : rate of return for non-housing investment; X : amount of bankruptcy exemptions in the state (this can be applied to all of the household’s

non-housing assets); where all variables are evaluated in the first period except Vt +1 . Since the household has to make this decision in the first period not knowing the value of its home in the next period, we need to assume expectations, as follows: Et [Vt +1 ] = (1 + Et [G ])Vt ,

where Et [G ] is the household’s expectation of house price appreciation in the first period. I am assuming the household can keep its home after bankruptcy proceedings, regardless of the level of homestead exemption. I will relax this assumption later. Subtracting W0,t +1 from W1,t +1 and W2,t +1 gives the expected benefits from mortgage default and filing for bankruptcy respectively,

as follows: E t [W1,t +1 − W0 ,t +1 ] = M t (1 + r0,t ) − E t [G ] ⋅ Vt − Vt = Vt {CLTVt ⋅ (1 + r0,t ) − E t [G ] − 1},

(2.1)

(expected financial benefit from mortgage default); E t [W2,t +1 − W0 ,t +1 ] = {Dt − max[Yt − Rt + At − X , 0]}(1 + r0,t ) ,

(expected financial benefit from bankruptcy filing); where CLTV is the current loan-to-value ratio defined as CLTVt = M t / Vt . From the results above, we can derive several predictions. The CLTV will have a positive effect on the probability of mortgage default, while expectations for capital gains will affect this probability negatively. As for bankruptcy, it seems hard to make precise predictions, because bankruptcy exemptions are considerably different across the states, as discussed in Chapter 1. Nonetheless, we can expect that the more households hold unsecured debts, the more likely they are to file for bankruptcy. 37

These are standard predictions in the previous studies. Even though we saw the consistent and strong positive effect of mortgage payment-to-income ratio on the probability of mortgage delinquency from the reduced-form regressions in Chapter 1 (see Table 1-5), the model of financial profits described above for mortgage default and bankruptcy tells nothing about the impact of mortgage payments, or borrowers’ ability to pay. The two-period model explains reasonably well borrowers’ motivation for the final outcomes, namely mortgage default and bankruptcy, but it does not allow us to address borrowers’ behavior with respect to loan payments before those final outcomes occur. Thus I expand the two-period model into a multi-period model to examine the impact of the mortgage payment variables on delinquency, rather than mortgage default, as follows:

(

)

(

[

)

(

]

)

W0,t +1 = Yt − Rt − Qtm − Qtn + At (1 + r0,t ) − Dt − Qtn (1 + rn ,t ) + Vt +1 − M t − Qtm (1 + rm ,t ) ;

(

)

(

[

)

]

W1,t +1 = Yt − Rt − Qtn + At (1 + r0,t ) − Dt − Qtn (1 + rn,t ) + Vt +1 − M t (1 + rm,t ) − Qtm ⋅ rm′ ,t ;

(

)

(

)

W1,default = Yt − Rt − Qtn + At (1 + r0,t ) − Dt − Qtn (1 + rn ,t ) ; t +1

(

)

[

] [

(

)

]

W2,t +1 = Yt − Rt − Qtm + At (1 + r0,t ) − Dt (1 + rn ,t ) + Qtn ⋅ rn′,t + Vt +1 − M t − Qtm (1 + rm,t ) ;

{ [(

) ]}

(

)

W2bankruptcy = min Yt − Rt − Qtm + At , X (1 + r0,t ) − M − Qtm (1 + rm ,t ) + Vt +1 , ,t +1 where W1,t +1 is the household’s wealth in the period t+1 if it becomes delinquent on its mortgage in the period t and plans to resume payments in the next period. On the other hand, W1,default t +1 denotes the household’s wealth, if it becomes delinquent for the purpose of default32. Similarly, W2,t +1 denotes the wealth after single-period delinquency on unsecured debt, while W2bankruptcy is ,t +1

the household’s wealth, if it plans to file for bankrupt. W0,t +1 denotes the wealth when the

32

For simplicity, here I assume that mortgage default and bankruptcy filing take place immediately in the next period, if the household chooses these alternatives in the current period.

38

household makes all of its required payments in period t. The variables that did not appear in the previous specification are defined as follows: Qtm : mortgage payment in period t; Qtn : unsecured debt payment in period t; rm ,t : interest rate for mortgage in period t; rn,t : interest rate for unsecured debt in period t, rm′ ,t : penalty rate for delinquency on mortgage in period t; rn′,t : penalty rate for delinquency on unsecured debt in period t.

I assume all the payments are applied to the principals of each loan. Now we can examine borrowers’ financial incentives for payment behavior. First, I look at the household’s financial benefit from temporary delinquency rather than default or bankruptcy. As I did before, subtracting one outcome from another gives; W1,t +1 − W0,t +1 = Qtm ⋅ r0,t − Qtm ⋅ rm,t − Qtm ⋅ rm′ ,t ,

(2.2)

(financial benefit from mortgage delinquency); W 2,t +1 − W0,t +1 = Qtn ⋅ r0,t − Qtn ⋅ rn ,t − Qtn ⋅ rn′,t ,

(2.3)

(financial benefit from unsecured loan delinquency). This conclusion is quite counterintuitive. Even if we assume the market rate of return for non-housing assets is as high as the interest rate for borrowing, r0,t = rm,t 33 and r0,t = rn,t , the financial benefit from temporary delinquency is negative since the penalty rates should be positive. Moreover, the magnitude of this negative benefit increases proportionally as the amount

33

This assumption is particularly unrealistic for the dominant type of mortgages in the U.S.: namely, fixed rate mortgages.

39

of payments goes up, suggesting that larger payments discourage borrowers from delinquency more from a financial viewpoint. In short, this formulation shows that “from a financial framework, delinquency may be viewed as borrowing … at a rate equal to the mortgage contract rate”, as Zorn and Lea (1989) argue34. Although this argument sounds theoretically reasonable, they find a completely opposite result, in which mortgage contract rates, rm ,t in my model, is positively correlated to the probability of mortgage delinquency. This hypothesis is also opposite to the positive effects of the payment-to-income ratio estimated in Chapter 1. In fact, we can also show that in this framework, larger payments are unlikely to have any positive effect, not only on the financial benefit from temporary delinquency, but also on that from mortgage default or bankruptcy filing. The financial benefit from mortgage default and bankruptcy are described respectively as follows: W1,default − W0,t +1 = Qtm (r0,t − rm,t ) − [Vt +1 − M t (1 + rm,t )] ; t +1

{ [

]}

W2bankruptcy − W0,t +1 = Dt (1 + rn,t ) − max Yt − Rt − Qtm + At − X , 0 (1 + r0,t ) + Qtn (r0,t − rn,t ) ,t +1 This result can be interpreted as the combination of Equations (2.1), (2.2), and (2.3). The terms regarding the payment variables, Qtm and Qtn , in the two equations above imply that missing payments can be viewed as borrowing at rates of loan contracts instead of the market rate of return for non-housing assets from a purely financial viewpoint, just as the case of temporary delinquency, even though mortgage default and bankruptcy do not incur the penalty rates, rm′ ,t and rn′,t , unlike temporary delinquency. The other terms in the equations correspond to the financial benefit from mortgage default or bankruptcy that we derived from the two-period model

34

Vandell and Thibodeau (1985) conclude that the effect of the payment variables on delinquency decisions is indeterminate based on their model which is similar to the model above.

40

in Equation (2.1)35. To sum up, it is almost impossible to explain the positive effects of loan payments on the probability of delinquency, default and bankruptcy by the financial profit theory.

2.2.3. Theoretical consideration II: credit constraint One possible explanation for the significant impact of loan payments on delinquency is the existence of credit constraints; credit constrained households might be forced to become delinquent when they are struck by financial distress, even if they understand that missing payments is not financially profitable, since they cannot borrow additional money to handle the situation. Thus, I expand the model by assuming that some of mortgage-indebted households are credit constrained and that their credit constraints might put them into delinquency36. The representative credit constrained household maximizes its wealth under a given level of credit constraint. The household becomes delinquent, if; Yt − Rt + At + C t < Qtm + Qtn ,

(2.4)

where Ct is the maximum amount that the household can borrow in period t, which is given exogenously. I am assuming that the household never sells its home, but this assumption is to be relaxed later in this chapter. Although the household has to become delinquent, if Condition (2.4) is satisfied, the household still has another choice: delinquency on mortgage, on nonmortgage, or both. As we see, since the financial benefit from delinquency is usually negative, it is reasonable to assume that the household decides on delinquency in order to minimize the financial cost of delinquency, defined as Equation (2.2) and (2.3). Then this specification determines the

35

This result basically remains same in the case in which the household default on the mortgage two or more periods later. 36 Vandell (1995) reviews the literature in order to emphasize the importance of mortgage default by credit constrained households rather than default driven by financial option values. Ambrose and Capone (1998) show empirically the difference between default of credit constrained households and that of non-constrained households.

41

household’s behavior under the condition of Equation (2.4) as follows: Case 1: if Qtm < Yt − Rt + At + C t < Qtn , then delinquent on Qtn ; Case 2: if Qtn < Yt − Rt + At + C t < Qtm , then delinquent on Qtm ; Case 3: if Yt − Rt + At + C t > Qtm and Yt − Rt + At + C t > Qtn , then delinquent on Qtm if Qtm ⋅ r0,t − Qtm ⋅ rm,t − Qtm ⋅ rm′ ,t > Qtn ⋅ r0,t − Qtn ⋅ rn,t − Qtn ⋅ rn′,t , delinquent on Qtn otherwise; Case 4: if Yt − Rt + At + C t < Qtm and Yt − Rt + At + C t < Qtn , then delinquent on both.

(2.5)

In the first case, the household can only pay for mortgage, Qtm , but cannot pay for the other loan, Qtn , even if they do not pay Qtm . Thus it skips the payment for nonmortgage loan, Qtn 37. The

second case is just the reverse situation to the first one. The household’s behavior is indeterminate in the third case; its decision depends on the market rate of return for the non-housing asset, the interest rates of both types of loans, and the penalty rates as well as the amount of payments. In the fourth case, the household virtually has no choice but to become delinquent on both loans since both payments are too large. Aside from Case 3, we can anticipate how the amount of loan payments affects the household’s decision on delinquency. The larger the payment of one type of loan is, the more likely the household is to become delinquent on that loan. On the other hand, the smaller the payment of one type of loan is, the more likely the

37

It is assumed here that lenders will not accept less than their contractual payments. This assumption is realistic in practice. See Zorn and Lea (1989, pp.134).

42

household is to become delinquent on the other type of loan. From this specification, we can also derive other implications which the theory of financial profits fails to explain. Cases 1 through 4 all suggest that the less a household’s income is, the more likely it is not be able to pay for its loan payments due to the existence of credit constraint, Ct . In a similar vein, this model supports the effect of income shocks, such as unemployment, on delinquency, too38. To integrate these two perspectives, the financial profit model and the credit constraint model, delinquency may be viewed as the mixture of two distinct borrowers’ actions; delinquency is the precursor to households’ attempts to gain financial profits either by mortgage default or by bankruptcy filing. At the same time, however, delinquency is an unavoidable response to financial distress caused by credit constraint. Therefore I expect the effects of determinants related to the credit constraint model as well as those related to the financial profit model will appear in the way predicted above, when we look into household-level data.

2.3.

Estimation methodology and variables

2.3.1.

Specification of multiple choice estimation The model used for estimation basically follows Campbell and Dietrich (1983), Vandell

and Thibodeau (1985) and Zorn and Lea (1989), but it is different with respect to borrowers’ choice. Borrowers have four choices as follows: 0) to pay the scheduled payments for both loans. 1) to become delinquent on mortgage loans;

38

This model also has another implication; that is, the financial asset, Wt , protects the household from delinquency. This agrees with Getter (2003), who argues the financial asset can be used as a buffer against negative shocks.

43

2) to become delinquent on nonmortgage loans; 3) to become delinquent on both loans. Then the i-th borrower’s utility-maximizing choice is described as the probability function F j of its explanatory variables x i as defined below: Pr (S j | x i ) = F j (x i ), i = 1, K , n,

∑ Pr (S 3

where

j

j = 0, K ,3 ,

| xi ) = 1 .

j =0

S j denotes each choice specified from (0) through (3) above. The probability function of F j is

defined for each outcome of S j . McFadden (1973) shows that, under reasonable assumptions, the consumer’s choice can be represented by the conditional logit function. Thus the probability functions for this multinomial choice may be specified as follows: Pij =

[

]

exp V j* (x i ) 3

[

* k

[

* k

]

1 + ∑ exp V (x i )

, i = 1, K, n,

j = 1,K,3 ;

, i = 1, K , n,

j = 0,

k =1

Pij =

1 3

]

1 + ∑ exp V (x i ) k =1

where Pij = P(S j | x i ), and V j* (x i ) is the indirect utility function concerning the outcome S j for the i-th borrower conditional on the vector of explanatory variables, x i . I set continuing payment (j = 0) as the base alternative. If we assume a reduced-form linear indirect utility function, the probabilities of these choices are rewritten as: ′ expβ j x i    , i = 1,K, n, Pij = 3 ′   1 + ∑ exp β k x i   k =1

j = 1,K,3 ;

44

(2.6)

Pij =

1 ′ 1 + ∑ exp β k x i     k =1 3

, i = 1, K , n,

j =0,

(2.7)

where β j is the vector of parameters of the indirect utility function for each outcome S j . From Equations (2.6) and (2.7) we can derive the log-odds ratios;

 Pij  ′ ln   = β j x i , i = 1, K , n,  Pi 0 

j = 1, K ,3 .

We can estimate the parameter vectors β j by the maximum likelihood method39. This method above matches the theoretical framework developed in the previous section, since the model describes the borrower’s behavior based on the differences between possible alternatives and the base alternative, that is, continuing payment.

2.3.2. Data and variables As in the previous chapter, SCF data for 1998, 2001 and 2004 are used to estimate the multinomial choice model. The sample is different from that of Chapter 1 since not every mortgage borrower holds nonmortgage debts. I select households in the SCF that hold nonmortgage debts and have regular payment contracts for those loans as well as payments for mortgages40. The number of households in the sample is 12,507 (out of 66,330). The rates of delinquency among them for each loan are presented in Table 2-1. The dependent variable is defined to take on one of four discrete values corresponding to households’ loan payment performances: continuing scheduled payments, in delinquency on mortgage, on any of

39

See Greene (2008, pp.842-847) for details. I include credit lines (not secured by home equity), car loans, loans for other vehicles, education loans and consumer loans in nonmortgage loans here. Excluded are credit lines secured by home equity, home improvement loans and loans concerning real estate investments. 40

45

nonmortgage loans or on both41. As we developed in the previous sections, the model suggests several variables have significant effects on borrowers’ decisions on delinquency from two perspectives: the financial profit theory and the credit constraint theory. Concerning the financial profit theory, I include the current loan-to-value ratio, that is, the mortgage balance divided by the house value, which has played a key role in most of the literature on mortgage default. Previous studies (e.g. Deng, Quigley and Van Order, 2000) consistently suggest that the effect of loan-to-value is non-linear and rapidly increases around 0.8 or 0.9. Therefore I also include the square term of this variable. Current balance of unsecured debts that would be discharged in bankruptcy proceedings is included in order to capture the financial benefit from filing for bankruptcy. Borrowers’ expectations about the future capital gains from their homes also play an important role in the financial profit theory, particularly for mortgage default (see Equation (2.1)). While there is a rich literature on house price expectations, the seminal work of Case and Shiller (1988) suggest that people seem to form their expectations on the basis of past price movements rather than any knowledge of fundamentals. In addition, Case and Shiller (1989) reveal predictable and persistent movements of house price index in cities. Poterba (1991) argues that housing market participants’ extrapolative expectations of this type, or “backward-looking” expectations, could account for the U.S. house price changes in 1980’s. As a recent study of the relationship between capital gains expectations and housing demands, Dusansky and Koç (2007) find a positive effect of current house price on housing demands in Florida housing markets, implying a significant impact of backward-looking expectations on households’ economic behavior. In the literature on mortgage default, there are few studies that explicitly examine the

41

The SCF asks questions for each type of loan as follows; “Are you paying off this loan ahead of schedule, behind schedule, or are the payments about on schedule?”

46

effect of backward-looking expectations on the exercise of mortgage default. However, Vandell and Thibodeau (1985) adopt backward-looking expectations and report a negative effect on the probability of default, which is consistent with their prediction as well as that of my model. As long as I know, no studies have been done on the relationship between backward-looking expectations and bankruptcy decisions. Thus, I will test the effect of backward-looking expectations in several specifications in this chapter and the next. Although Vandell and Thibodeau (1985) use a house price index based on Census tracts in their calculation of backward-looking expectations, we cannot use this method because no geographical data is available in the public version of the SCF. Since in the SCF, however, we can observe the purchase price of households’ homes and the year when they acquired them, as well as the current price, we can estimate the average rate of housing price appreciation that each household has actually experienced. The estimation formula adopted here is as follows: G0 =

ln (Vc ) − ln (Vo + I ) , Year

where G0 is the average rate of house price appreciation per year, Vc is current house price, Vo is original house price, I is the cost of home improvement (if any) and Year is the number of years since the households acquired their homes. Then I construct a proxy for households’ capital gain expectation ( Et [G ]⋅Vt in Equation (2.1)) as follows: CapGain = Vc ⋅ G0 , where CapGain is the proxy for households’ capital gain expectation that I use in the estimations to follow. While this approach has some disadvantages compared to the Census tract method, it arguably has advantages, as well. In particular, there is no worry about unobserved house-specific 47

characteristics, which local house price indices often suffer from, since this estimation is derived from a comparison of houses with basically the same characteristics. Gabriel and Rosenthal (1991) use this proxy variable to represent local housing market conditions in the study of borrowers’ choice between different types of mortgages and find reasonable results. Therefore, I use this variable, CapGain, as a proxy for backward-looking house price expectation of each household. On the other hand, the main factors for the credit constraint theory are borrowers’ ability to pay (see Equation (2.4)). Thus households’ income and the amount of each loan payment should also be included in the estimation model. Two kinds of shocks to the ability to pay, unemployment experience and unexpected low income, are also included as the explanatory variables. In addition to these variables for the ability to pay, the SCF allows us to detect credit constrained households directly to a great extent, since it asks the respondents some questions about their loan applications and denials42. Jappelli (1990) confirms the validity of these variables regarding credit constraint by examining the SCF in 1983. Thus these variables should be helpful to the estimations. I also include variables for age, sex, race, supporting children, which are selected for the estimation for credit constrained households in Jappelli (1990). While the dummy variable of credit constraint does not capture the degree of credit constraint of each household, that is, Ct in my model (Equation (2.4)), I expect these explanatory variables are correlated with the degree as well as the existence of credit constraints. The descriptions and summary statistics

42

The SCF has two variables concerning credit constraint; “In the past five years, has a particular lender or creditor turned down any request you or your (husband/wife/partner) made for credit, or not given you as much credit as you applied for?” and “Was there any time in the past five years that you thought of applying for credit at a particular place, but changed your mind because you thought you might be turned down?” The dummy variable used in this thesis takes a value one, if the household says yes for at least one of the questions above. This follows the definition used by Jappelli (1990).

48

for all the explanatory variables are presented in Table 2-2.

2.4.

Empirical results and some extensions

2.4.1. Result of the base specification Table 2-3 reports the result of the multinomial logit estimation. The first column is the coefficients for the probability of mortgage delinquency. Current loan-to-value ratio (CLTV) has a statistically significant effect as the financial profit theory predicts; its marginal effect turns to positive on the probability of default around CLTV of 0.7, and this effect increases rapidly as CLTV goes up towards negative equity, due to the quadratic specification. On the other hand, the coefficient on capital gain expectation is statistically significant and negative, supporting the prediction from Equation (2.1)43. This implies that backward-looking expectations play an important role in households’ decisions on mortgage default. At the same time, the coefficients on household’s income and the payment variables strongly support the possible borrowers’ strategy derived from credit constraint theory, as described in Equation (2.5). The larger the mortgage payment is and the smaller the nonmortgage payment is, the more likely households are to become delinquent on mortgage. The household’s income has a negative and statistically significant effect. The credit constraint dummy has a very strong effect, as expected. Race (black) and supporting children also have statistically significant effects in the predicted direction, based on the literature on credit constraint. In particular, the effect of supporting children can be interpreted as an increase in required nondiscretionary expenditure, Rt in Equation (2.4), as well. 43

Since, as expected, the current house value is correlated with the average rate of house price appreciation, G0 , (ρ = 0.0863), I added the current house value to all the estimation in this chapter. However, I found that it does neither have any statically significant effects nor change estimated coefficients for the other variables significantly. Thus I do not include the current house value in the tables hereafter.

49

The coefficients for the probability of nonmortgage delinquency reported in the second column support the model, too. In particular, the coefficients on the payment variables make a clear contrast to those of mortgage delinquency; the larger the nonmortgage payment is and the smaller the mortgage payment is, the more likely households are to become delinquent on nonmortgage loans. This result provides a strong evidence for the impact of credit constraints that force some households to become delinquent, along with the positive and statistically significant coefficient on the credit constraint dummy. Thus credit constraints are important as a determinant of mortgage default, as well as financial profits, as recent studies suggest (e.g. Ambrose and Capone (1996, 1998), Elmer and Seelig (1999) and Vandell (1995)). It might also suggest that households try to meet their loan payments as much as possible in order to minimize the financial losses from delaying payments, based on my model. Unemployment experience also has significant positive effects, but the effect of unexpected low income is negative and statistically significant. The variables of the financial profit theory are also consistent with the prediction; CLTV and capital gain expectation, both of which are important determinants of mortgage delinquency, do not have statistically significant effects. These results agree with the financial part of my model (Equation (2.1)), since CLTV and capital gain expectations are relevant only if barrowers seek mortgage default. However, the amount of debt has almost no impact on the probability of nonmortgage delinquency. This outcome may be due to my rough estimation of the financial benefit from bankruptcy. We cannot apply the actual bankruptcy exemptions to the calculation of the financial benefit without geographical information. It is reasonable to expect the coefficients for delinquency on both types of loans to be a combination of the predictions about mortgage and nonmortgage delinquency. The result in the third column basically agrees with this expectation. In particular, the payment variables of both

50

types of loans have positive coefficients, even though they are statistically insignificant.

2.4.2. Protecting housing investments through bankruptcy Next, I test an effect of backward-looking expectations of capital gains from another perspective in order to show that the effect is robust enough that we should not underestimate it in studying mortgage default and bankruptcy. As a preparation, I assume there are some credit constrained households that can get out of insolvency by selling their homes. Let us first assume the representative household satisfies the conditions as follows: Yt − Rt + At + C t < Qtm + Qtn ,

(2.4)

Yt − Rt + At + Vt − M t + C t > Qtm + Qtn .

(2.8)

This household can avoid delinquency and the associated financial losses (Equations (2.2) and (2.3)) by selling its home. If the household chooses to sell home, its expected wealth in the next period is described as follows44:

[

] (

)

(

)

(

)

m n n m Et W0sell ,t +1 = Yt − Rt − Qt − Qt + At + Vt (1 + r0 ,t ) − Dt − Qt (1 + rn ,t ) − M t − Qt (1 + rm ,t ) .

(2.9)

Then I examine the financial profits from mortgage default and filing for bankruptcy by using Equation (2.9) as a base outcome. If the household chooses to default on its mortgage or file for bankruptcy instead of selling its home on the market, the expected wealth in the next period is as follows:

[

] (

)

(

)

Et W1,default = Yt − Rt − Qtn + At (1 + r0,t ) − Dt − Qtn (1 + rn ,t ) ; t +1

[

] { [(

) ]}

(

)

Et W2bankruptcy = min Yt − Rt − Qtm + At , X (1 + r0,t ) − M t − Qtm (1 + rm ,t ) + Et [Vt +1 ] , ,t +1 if Vt − M t ≤ X h ;

44

Here I assume the household need not to borrow additional money to repay, if it sells its home.

51

[

]

Et W2bankruptcy = {min[(Yt − Rt + At ), X ] + X h }(1 + r0,t ) , ,t +1

[

where: Et W1,default t +1

]

[

otherwise,

]

and Et W2bankruptcy are the expected wealth in the next period when the ,t +1

household chooses mortgage default or bankruptcy filing, respectively; X h is the homestead exemption the household can claim in its state of residence. While the outcome of mortgage default never changes, I introduce the two distinct consequences of bankruptcy filing due to one of the remarkable features of the U.S. bankruptcy law: namely, homestead exemption. It relaxes the assumption that I made in the previous section. That is, the home will be liquidated by the bankruptcy trustee, if the household’s home equity is larger than the homestead exemption. In such a case, the household’s home equity would be protected only to the extent of the homestead exemption, at most. Then the subtraction between these outcomes gives;

[

]

m m Et W1,default − W0sell t +1 ,t +1 = Qt ⋅ r0 ,t − Qt ⋅ rm ,t − [Vt (1 + r0 ,t ) − M t (1 + rm ,t )] ;

[

]

{ [

]}

m Et W2bankruptcy − W0sell ,t +1 ,t +1 = Dt (1 + rn ,t ) − max Yt − Rt − Qt + At − X , 0 (1 + r0 ,t )

+ Qtn (r0,t − rn,t ) + Vt (Et [G ] − r0,t )

if Vt − M t ≤ X h ;

[

]

(2.10)

− W0sell Et W2bankruptcy ,t +1 ,t +1 = Dt (1 + rn ,t ) − {max[Yt − Rt + At − X , 0]}(1 + r0 ,t ) + Qtn (r0,t − rn ,t ) + Qtm (r0,t − rm,t )

+ X h (1 + r0,t ) − [Vt (1 + r0,t ) − M t (1 + rm,t )] otherwise. Compared with the previous model in which none of the credit constrained households are allowed to sell their home, these financial profits from mortgage default and bankruptcy have changed. Since this model focuses on insolvent and credit constrained households that sell their homes, their financial profits from mortgage default are not affected by capital gain expectations: they will lose their homes in both cases. Although, in the financial profit from bankruptcy, the 52

effects of the terms related to the rates of interest and bankruptcy exemption are not straightforward to interpret, one significant change occurred due to the possibility of retaining the homes after bankruptcy proceedings. The household that satisfies Equation (2.4) and (2.8) can keep its housing investment and is willing to do so if: (1) its home equity is low enough (below the homestead exemption); and (2) it expects the rate of return for the housing investment will be more preferable than that of non-housing assets (the last parenthesis in Equation (2.10)). This extension of the previous model implies that the larger capital gains (insolvent and credit constrained) households expect from their homes, the more likely the households are to file for bankruptcy, only if their home equity is less than the level of homestead exemption. We cannot directly test this hypothesis with the dataset of the SCF, because we do not know the homestead exemption that would be applied to each household due to a lack of geographical data in the SCF. However, if we treat the geographical variable (which state the household lives in) and associated homestead exemption as random variables in the model45, the marginal effect of the capital gain expectations on the expected financial profit from bankruptcy for this type of credit constrained households is described as follows: ∂ Et W2bankruptcy − W0sell ,t +1 ,t +1 = Pr ( X h > Vt − M t | Vt , M t ) ∂Et [G ] ⋅ Vt

[

]

where X h is defined as a random variable which follows a distribution which is known. The distribution of the level of homestead exemption weighted by the population of homeowners is presented in Table 3-3. I will discuss the homestead exemption in Chapter 3 again in more detail. Now I go back to the estimation. As we can see from Table 3-3, the level of homestead exemption is distributed in a wide range. Nonetheless, there are some values which have

45

However, the level of homestead exemption is unobservable only to us; the household itself is reasonably assumed to know the level of homestead exemption.

53

considerable frequencies. I pick up one of the highly frequent values, the homestead exemption level of 75,000 dollars (adopted by the state of California) as a breaking point in order to perform the multinomial logit estimation with the same explanatory variables, but allowing the coefficient on capital gain expectation to take a different value for the households with lower home equity from those with higher home equity. Thus we can test if capital gain expectations have a positive effect on bankruptcy decisions when households are likely to keep their homes even after bankruptcy proceedings. Table 2-4 shows the result of this estimation. Estimating the coefficients on capital gain expectation separately according to the home equity category clearly makes a difference for nonmortgage delinquency. The coefficient on capital gain expectation in the second column is positive and statistically significant for the group of households with the lower home equity; these households are likely to be able to protect their homes through bankruptcy proceedings. A similar estimation result is also obtained for the probability of delinquency on both loans. On the other hand, this specification brings about almost no changes in other coefficients. In particular, since the coefficients on household’s income and loan payments remain the same, compared to Table 2-3, introduction of home equity group seems to only affect households’ financial profits, but not the ability-to-pay determinants. This result provides another evidence for the important role of households’ backward-looking expectations in their financial incentives for loan terminations.

2.4.3. Households’ attitude towards financial risks When it comes to borrowers’ exercise of mortgage options, Deng, Quigley and Van Order (2000) emphasize the importance of the heterogeneity of their attitude towards financial

54

risks. They argue that some borrowers are astute enough to take advantage of the opportunities for financial profits, while some of them are less sensitive to such opportunities, and that the exercise of the mortgage default option is one of the primary examples in this context46. Their estimation of hazard functions provides strong evidence for the existence of this heterogeneity. If the heterogeneity of attitude towards financial risks is incorporated into the model of this chapter, the default decisions of households who are sensitive to the value of default option are likely to be greatly affected by the current loan-to-value ratio and their expectation of capital gains (see Equation (2.1))47. In other words, households who astutely take into account financial profits and risks are expected to exert their default option “ruthlessly”. In fact, the SCF contains a variable concerning the heterogeneity of attitude towards risks by asking the respondents a hypothetical question of risk-taking behavior48. Barsky, Juster, Kimball and Shapiro (1997) study a similar variable in the Health and Retirement Study to check its validity and its effect on risk-taking behaviors, such as holding stock. They find that the variable appropriately describes the heterogeneity of attitude towards risk and has predictive power for risk-taking behavior. Table 2-5 reports the estimation result of the specification which allows the coefficients on CLTV and capital gains expectations to take different values according to the groups based on attitudes towards financial risks. By this specification, we can test the hypothesis that the financial profit variables have different effects on households’ decisions on mortgage default due

46

Deng, Quigley and Van Order (2000) study the mortgage options of default and prepayment in a simultaneous framework, and find the effect of the heterogeneity is more remarkable for the exercise of prepayment option. 47 I am still assuming the backward-looking formulation of capital gain expectations. 48 The question is as follows: “Which of the statements on this page comes closest to the amount of financial risk that you and your (husband/wife/partner) are willing to take when you save or make investments?” The households who answered yes to the following question are labeled “willing to take risk”, otherwise labeled “risk averse”: “take substantial financial risks”, “take above average financial risks” or “take average financial risks”.

55

to their heterogeneity of attitudes towards risks. For mortgage delinquency, the coefficients on these variables fit the prediction; it implies the risk-tolerant group will exert the default option particularly based on their equity position in their properties and on expected capital gains. On the contrary, none of the coefficients on CLTV and expected capitals gains are statistically significant for the group of risk-averse households, even though their signs remain same. In addition, the F test cannot reject the null hypothesis that the coefficients for CLTV and CLTV square are jointly zero (p value = 0.263). Nonetheless, this specification does not change the effects of the ability-to-pay variables, such as income and loan payments. Thus we can see a crucial difference only in the financial factors in this specification of the heterogeneity of risk attitudes.

2.5.

Conclusion This chapter tries to disentangle the relationship between the determinants of

households’ decisions on mortgage default, bankruptcy and loan delinquency by focusing on the observable choice between mortgage and nonmortgage delinquency. First, I set up a theoretical model based on financial profits that households could obtain by mortgage default, bankruptcy or delinquency. This model describes well borrowers’ incentives for mortgage default and bankruptcy, but it turns out to be difficult to explain the effect of loan payments on delinquency decisions. Then a simple type of credit constraint model is introduced to show that for some households, loan delinquency is an unavoidable response due to credit constraints rather than any attempts to gain financial profits. Credit constraints give sound theoretical reasoning to the effects of household’s income and shocks to the ability to pay, or “trigger events”, as well. This mixture of the financial profit theory and the credit constraint theory is strongly

56

supported by the results of multinomial logit estimation. The significant effect of CLTV is consistent with the findings of the previous option-based studies. Although the impact of backward-looking expectations has drawn little attention in the traditional mortgage default studies, compared to other fields related to housing, such as housing demand, the estimation results suggest that borrowers’ backward-looking expectations affect their decisions both on mortgage default and on bankruptcy. On the other hand, household’s income, loan payments and trigger events also play an important role in households’ choice between mortgage and nonmortgage delinquency, supporting the credit constraint model. The effect of the homestead exemption and the importance of borrowers’ heterogeneity towards risks are also found empirically to be important. In conclusion, the results reported in this chapter imply there are certainly a good number of households that are forced to become delinquent against their will, or regardless of their optimal choice based on financial profits. Moreover, they might be even forced to default on their mortgage or file for bankruptcy beyond temporary loan delinquency, if they cannot escape from financial distress. These findings provide a solid empirical evidence to reconcile two conflicting theories in the literature on mortgage default and personal bankruptcy: the willingness-to-pay theory and the ability-to-pay theory. Since I make some strong assumptions, such as exogenously fixed expenditures, housing investment and credit constraints, we could derive more profound implications with more general models based on the consumer-choice theory. The model in this chapter also assumes all mortgages are non-recourse. As for empirical studies, examining households’ decisions between mortgage default and bankruptcy directly rather than their delinquency choice should be done, more effectively and persuasively with datasets of dynamic structure.

57

Figure 2-1

Effect

Mortgage default Mortgage delinquency Resume payment Effect

Personal bankruptcy Nonmortgage delinquency Resume payment

Table 2-1 Delinquency rates for mortgage and nonmortgage loans Paying for non-mortgage

Non-mortgage delinquency

Total

Paying for mortgage

11,962 (95.64)

290 (2.32)

12,252 (97.96)

Mortgage delinquency

185 (1.48)

70 (0.56)

255 (2.04)

Total

12,147 (97.12)

360 (2.88)

12,507 (100.00)

Note: percentage points inside parentheses

58

Table 2-2 Summary statistics of explanatory variables Description

Mean

Std. Dev.

CLTV

current mortgage balance divided by current housing value

0.5706

0.2557

CLTV squared

CLTV squared

0.3909

0.3125

Expected capital gain

average annual rate of house price appreciation times current housing value

21,063

71,773

Debt

the amount of debts that could be discharged by bankruptcy

61,008

944,214

Mortgage payment

monthly payment of mortgages

1,471

2,404

Nonmortgage payment

monthly payment of nonmortgage loans

1,357

10,955

Income

household's total income

291,361

3,652,404

Unemployment

monthly payment of mortgages

0.0648

0.2463

Unexpected low income

household head experienced unemployment within 12 month

0.1305

0.3369

Credit constraint

denial or expected denail of loan applications

0.2154

0.4111

Female

household head is female

0.1019

0.3025

Age

household head's age

45.49

11.03

Age squared

household head's age squared

2190.54

1050.84

Black

household head is black

0.0792

0.2700

Hispanic

household head is Hispanic

0.0514

0.2208

Children

household supporting children

0.6457

0.4783

Willing to take risks

answer yes to "take average financial risk" or more

0.8080

0.3939

Observations

12,507

Year 2004

4,699

Year 2001

4,060

Year 1998

3,748

59

Table 2-3 Borrowers' choice of delinquency: base model Mortgage delinquency Coefficient

t

CLTV

-3.589

(-4.82)***

CLTV squared

2.799

(6.10)***

Nonmortgage delinqency Coefficient

Delinquency on both

t

Coefficient

t

0.088

(0.10)

-2.096

(-1.74)*

0.294

(0.44)

2.466

(3.83)***

Debt

9.19.E-07

(0.18)

2.05.E-07

(0.68)

-2.89.E-06

(-0.56)

Expected capital gain

-2.48.E-05

(-2.70)***

7.89.E-07

(0.20)

6.76.E-07

(0.24)

Mortgage payment

2.06.E-04

(3.59)***

-8.03.E-04

(-4.65)***

3.64.E-05

(0.16)

Nonmortgage payment

-7.87.E-04

(-2.57)***

3.31.E-05

(3.22)***

8.00.E-05

(1.48)

Income

-1.86.E-05

(-5.69)***

-5.23.E-06

(-2.75)***

-9.95.E-06

(-2.22)**

Unemployment

0.023

(0.10)

0.678

(3.87)***

1.584

(5.59)***

Unexpected low income

0.541

(2.89)***

-0.513

(-2.57)***

0.265

(0.86)

Credit constraint

2.130

(11.07)***

1.805

(13.21)***

1.670

(5.85)***

Age

-0.075

(-1.76)*

0.102

(2.08)**

0.259

(2.19)**

Age squared

0.001

(1.25)

-0.001

(-2.30)**

-0.003

(-2.40)**

-2.12.E-01

(-1.00)

-1.84.E-01

(-1.00)

4.96.E-01

(1.57)

Black

0.634

(3.18)***

-0.060

(-0.32)

0.452

(1.33)

Hispanic

0.410

(1.39)

0.108

(0.44)

1.055

(2.84)***

Children

0.832

(4.07)***

1.013

(5.61)***

0.310

(1.01)

Female

year 2004

0.048

(0.24)

0.438

(2.79)***

1.036

(2.68)***

year 2001

-0.229

(-1.17)

0.160

(1.00)

0.927

(2.42)**

Constant

-1.441

(-1.46)

-6.305

(-5.96)***

-11.115

(-4.61)***

Observations Log likelihood Pseudo R square

12,507 -2184.42 0.211

60

Table 2-4 Borrowers' choice of delinquency: homestead exemption model Mortgage delinquency

Nonmortgage delinqency

Delinquency on both

Coefficient

t

Coefficient

t

Coefficient

CLTV

-4.029

(-5.36)***

-0.784

(-0.88)

-2.421

(-1.99)**

CLTV squared

3.007

(6.62)***

0.764

(1.17)

2.653

(4.10)***

8.35.E-07

(0.14)

2.97.E-07

(0.89)

-2.98.E-06

(-0.58)

Debt

t

Expected capital gain Home equity 75K

-4.13.E-05

(-3.11)***

-1.40.E-05

(-1.75)*

-1.33.E-07

(-0.02)

Mortgage payment

2.42.E-04

(4.07)***

-7.44.E-04

(-4.17)***

2.02.E-05

(0.08)

Nonmortgage payment

-7.71.E-04

(-2.52)**

3.46.E-05

(3.26)***

8.23.E-05

(1.52)

Income

-1.85.E-05

(-5.67)***

-4.89.E-06

(-2.52)**

-1.01.E-05

(-2.21)** (5.64)***

Unemployment

0.028

(0.11)

0.713

(4.06)***

1.600

Unexpected low income

0.529

(2.82)***

-0.567

(-2.81)***

0.223

(0.72)

Credit constraint

2.138

(11.08)***

1.808

(13.20)***

1.671

(5.84)***

Age

-0.070

(-1.65)*

0.099

(2.03)**

0.260

(2.19)**

Age squared

0.001

(1.17)

-0.001

(-2.20)**

-0.003

(-2.39)**

-1.77.E-01

(-0.84)

-1.21.E-01

(-0.65)

5.37.E-01

(1.69)*

Black

0.607

(3.03)***

-0.074

(-0.40)

0.458

(1.35)

Hispanic

0.471

(1.59)

0.211

(0.85)

1.105

(2.95)***

Children

0.846

(4.13)***

1.054

(5.81)***

0.334

(1.09)

year 2004

0.045

(0.23)

0.430

(2.72)***

0.999

(2.57)***

year 2001

-0.233

(-1.19)

0.151

(0.94)

0.910

(2.37)**

Constant

-1.467

(-1.49)

-6.138

(-5.82)***

-11.111

(-4.60)***

Female

Observations Log likelihood Pseudo R square

12,507 -2168.70 0.216

61

Table 2-5 Borrowers' choice of delinquency: heterogeneity model Mortgage delinquency

Nonmortgage delinqency

Delinquency on both

Coefficient

t

Coefficient

t

Coefficient

t

CLTV

-4.574

(-6.00)***

-0.025

(-0.03)

-0.984

(-0.52)

CLTV squared

3.157

(6.98)***

0.194

(0.27)

0.956

(0.64)

-2.96.E-05

(-2.45)**

1.61.E-06

(0.66)

3.24.E-07

(0.11)

CLTV

-1.415

(-1.36)

0.614

(0.56)

-2.654

(-1.80)*

CLTV squared

1.361

(1.59)

0.035

(0.04)

3.430

(3.72)***

-1.77.E-05

(-1.38)

-1.61.E-05

(-1.14)

-2.48.E-06

(-0.12)

1.55.E-06

(0.67)

1.84.E-07

(0.61)

-2.84.E-06

(-0.60)

Willing to take risk

Expected capital gain Risk averse

Expected capital gain Debt Mortgage payment

1.96.E-04

(3.37)***

-7.61.E-04

(-4.47)***

9.30.E-05

(0.51)

Nonmortgage payment

-8.27.E-04

(-2.57)***

3.19.E-05

(3.12)***

8.13.E-05

(1.67)*

Income

-1.39.E-05

(-4.43)***

-5.05.E-06

(-2.67)***

-1.07.E-05

(-2.27)**

-0.004

(-0.02)

0.653

(3.69)***

1.519

(5.27)***

Unemployment Unexpected low income

0.491

(2.56)***

-0.509

(-2.55)**

0.311

(1.01)

Credit constraint

2.125

(10.88)***

1.804

(13.18)***

1.560

(5.37)***

Age

-0.079

(-1.82)*

0.097

(2.00)**

0.253

(2.07)**

Age squared

0.001

(1.21)

-0.001

(-2.24)**

-0.003

(-2.31)**

-2.27.E-01

(-1.06)

-2.07.E-01

(-1.12)

5.13.E-01

(1.61)

Black

0.519

(2.56)***

-0.080

(-0.42)

0.496

(1.44)

Hispanic

0.148

(0.49)

0.089

(0.35)

0.962

(2.46)**

Children

0.799

(3.89)***

1.004

(5.55)***

0.425

(1.30)

year 2004

-0.144

(-0.72)

0.427

(2.71)***

0.956

(2.51)**

year 2001

-0.349

(-1.73)*

0.136

(0.85)

0.731

(1.89)*

Constant

-1.297

(-1.27)

-6.152

(-5.83)***

-10.827

(-4.36)***

Female

Observations Log likelihood Pseudo R square

12,507 -2148.71 0.224

62

Chapter 3: Personal bankruptcy decisions of mortgage borrowers 3.1.

Introduction Americans witnessed an unprecedented more than six-fold increase in the number of

personal bankruptcy filings, from 241,000 in 1980 to more than 1.6 million in 200549. While this explosive rise led to an amendment of the bankruptcy law in 2005, which basically made it harder for borrowers to file for bankruptcy, personal bankruptcy filings have drawn serious attention by many researchers from both economics and legal fields. In particular, several seminal empirical studies have been done focusing on borrowers’ decisions on bankruptcy filings. These empirical studies on bankruptcy mostly take advantage of a control variable offered by the legal system of bankruptcy in the United States: namely, bankruptcy exemptions, particularly homestead exemption, which states set at significantly different levels. For example, the homestead exemption in Delaware and Maryland is zero, while Arkansas, Florida, Iowa, Kansas, Oklahoma, South Dakota and Texas do not limit homestead exemptions (see Table 3-150). Recently, the literature on personal bankruptcy decisions is developing remarkably fast, mainly due to increasing availability of household-level bankruptcy filing data. Fay, Hurst and White (2002) describe households’ strategic behavior on filing for bankruptcy by explicitly formulating the financial benefit from bankruptcy. Roughly speaking, it is loans to be discharged less assets to be liquidated in bankruptcy proceedings. They show that the households in the PSID that could obtain greater financial benefit from bankruptcy are more likely to file for bankruptcy, implying the possibility that households in the United States are engaged in strategic

49 50

Bankruptcy Statistics, U.S. Courts. Table 3-1 summarizes the laws as of 1996 based on Hynes, Malani and Posner (2004).

63

use of bankruptcy. Bahchieva, Wachter and Warren (2005) find households living in states with high homestead exemptions are likely to have large home equity when they file for bankruptcy by a survey of bankruptcy petitioners. Their result also suggests households’ strategic behavior to protect their homeownerships through the bankruptcy law. On the other hand, another hypothesis has been proposed for personal bankruptcy decisions. Sullivan, Warren and Westbrook (1989) discuss another type of framework to explain the incentive for bankruptcy, namely that households file for bankruptcy when unexpected adverse events occur which reduce their ability to repay their debts. Fay, Hurst and White (2002) attempt to refute the latter model of borrowers’ ability to pay by their empirical results, but do not completely succeed. Household income, which should be irrelevant in the strategic behavior model, still has a significant effect in discouraging households from filing for bankruptcy. Fisher (2006) reexamines this effect of household income by using a greater number of explanatory variables in the PSID to obtain the same result. These two theories in the field of personal bankruptcy decisions parallel the two theories in the literature on mortgage default: the option-based theory and the theory of trigger events. We can call these two types of theories “the willingness-to-pay theory” and “the ability-to-pay theory”, as I did in Chapter 1. While the last chapter examined borrowers’ choices between mortgage default and bankruptcy, the present chapter focuses only on households’ bankruptcy decisions and investigates this subject in more detail. The aim of this chapter is to extend the literature by focusing on bankruptcy decisions of households with mortgage debts and on the effects of homestead exemption and capital gains expectations. As in Chapter 2, I also try to disentangle the two conflicting theories, the willingness-to-pay theory and the ability-to-pay theory, by looking into decisions made by households who are suffering from the combination of severe financial

64

distress and credit constraints. This chapter is composed of six sections. The next section develops a model which analyzes households’ decisions on bankruptcy filings. The Section 3 describes the data used in the empirical analysis and presents empirical specifications. The section 4 reports the empirical result. Another empirical result focusing on a smaller subsample of financially distressed households is examined in Section 5. Lastly, the Section 6 concludes the findings in this chapter.

3.2.

Theoretical model for financial benefit from bankruptcy In this section, I set up a theoretical framework to address the motivation for households

to file for bankruptcy. Based on the willingness-to-pay theory, I assume that households file for bankruptcy if they could obtain any financial benefit from bankruptcy, following Fay, Hurst and White (2002). The financial benefit of the representative household from bankruptcy can be represented as follows51: FinBen = max[D − max[H − X h ,0] − max[W − X w ,0],0] − S ,

(3.1)

where D is the value of the household’s unsecured debt that would be discharged in a bankruptcy proceeding, H is the value of the home equity that the household owns, which is defined as the current housing value less the mortgage balance that the household still owes52, X h is the homestead exemption in the household’s state of residence, W is the household’s non-housing wealth net of secured debts such as car loans, X w is the bankruptcy exemption that could be applied to W in the state where the household resides, and S is the cost of filing bankruptcy, including legal and filing fees, the cost of bankruptcy stigma, and the cost of reduced access to

51

Although this formulation is originally set for Chapter 7 bankruptcy, Fay, Hurst and White argue that it can also be used as a proxy for the household’s financial benefit from Chapter 13 bankruptcy. 52 It is represented as H = V − M by using the notation of Chapter 2.

65

credit following bankruptcy53. If the household filed for bankruptcy, it could discharge its unsecured debt to which the bankruptcy law should be applied, but at some expense. The household should give up the value by which their properties exceed the relevant bankruptcy exemptions, namely, its housing equity, less the homestead exemption, and its other non-housing wealth, less the personal property exemptions that are applicable to each category54. However, the household does not need to give up any part of its home equity, if the homestead exemption exceeds the current value of its home equity. Moreover, the financial benefit defined above cannot take negative values since the household would never file for bankruptcy, if their assets net of bankruptcy exemptions exceeded the amount of debts to be discharged. In addition, it is also reasonable to assume that the bankruptcy trustee sells household’s mortgage property in order to distribute the proceeds to the creditors, only if the home equity exceeds the homestead exemption that the household could claim in its state of residence. Then Equation (3.1) should be divided into two cases: FinBen1 = max[D − max[W − X w ,0],0] − S ,

if X h ≥ H ;

FinBen0 = max[D − max[H − X h ,0] − max[W − X w ,0],0] − S − E[G ] ⋅ V ,

otherwise,

(3.2)

where FinBen1 denotes the financial benefit if the household keeps its home after bankruptcy proceedings, while FinBen0 denotes the benefit if it loses its home. E [G ] is the expected rate of price appreciation of the household’s mortgage property, while V is the current value of the household’s home. So E [G ] ⋅ V represents the future capital gain that the household expects from its home. If the household files for bankruptcy, it would lose the mortgage property if the 53

See White (2007a) for more information. In equation (1), I am dealing with non-housing wealth and the exemptions other than homestead exemption collectively, for simplicity. 54

66

home equity were to exceed the homestead exemption. In such a case, the household would not be able to obtain the capital gain that it would have received, if it retained the home. On the other hand, if the homestead exemption fully covers the household’s home equity, the household can protect its mortgaged property from liquidation and obtain the capital gain later as long as it keeps repaying the mortgage. Thus expected capital gains only negatively affect bankruptcy decisions of households that have home equity larger than the homestead exemption.

3.3.

Data and estimation methodology

3.3.1. Recent bankruptcy experiences of the households in the SCF In this chapter, I use the subsamples of the SCF in 1998, 2001 and 2004 that consist of mortgage-indebted households in order to estimate the effects of a variety of determinants on households’ bankruptcy decisions. Although the SCF provides detailed information on household-level financial situations, particularly information about mortgaged properties and about credit constraints, the fact that it is a cross-section dataset prevents us from carrying out dynamic estimations. Ideally, it would be the best econometric approach to estimate the probability of bankruptcy filing based on households’ characteristics, including their financial benefit from bankruptcy, just before their decisions of filing (or not filing). Even though the SCF asks the respondents about experiences of filing for bankruptcy, we can only observe their current situations, in other words, post-bankruptcy characteristics rather than pre-bankruptcy ones. Moreover, since the SCF does not ask any questions about whether the households ever lost their mortgaged properties through bankruptcy proceedings, we cannot distinguish households that lost their homes because of bankruptcy filings from households that have never owned homes or that lost their homes for other reasons.

67

The SCF, however, asks the households when they filed for bankruptcy (or never). It allows us to concentrate on the most recent bankruptcy experience of households, which is filing for bankruptcy within two years in the SCF55, so that most of the current characteristics of the households, including households that filed for bankruptcy, can be considered to reflect their pre-bankruptcy characteristics accurately enough56. Thus, in this chapter, I use the subsamples of the SCF in 1998, 2001 and 2004, which consist of mortgage-indebted households that acquired their mortgaged properties no less than two years ago, in order to look into their bankruptcy decisions during recent two years. The dependent variable here is a dummy variable indicating whether or not the household has filed for bankruptcy within two years of the time of the survey. The number of households in this subsample is 23,534 (out of 66,330 households) and the rate of recent bankruptcy experiences is 1.126%, as shown in Table 3-2. Since this subsample does not capture households that used to own mortgaged properties but lost them during bankruptcy proceedings ( H > X h in Equation (3.2)), the results presented in this chapter must be interpreted carefully. Nonetheless, we can expect the bias caused by this omission would be small, considering that the fraction of homeowners who have large home equity among all the homeowners who filed for bankruptcy is very small, according to the Survey of Bankruptcy Petitioners. That survey is used by Bahchieva, Wachter and Warren (2005)57. The discussion above is illustrated as Figures 3-1 and 3-2.

55

Two years is the smallest number that the variable of years since filing for bankruptcy takes in the public dataset from the SCF. I will call bankruptcy filed within two years “recent bankruptcy” hereafter in this chapter. 56 We should be particularly careful about the variables greatly affected by bankruptcy proceedings, such as non-housing assets and unsecured debts. 57 According to this survey conducted in 2001, the average home equity among the homeowners who filed for bankruptcy was 3,217 dollars, which is smaller than the homestead exemption in most states in the U.S (see Table 3-3). Moreover, it reports that the loan-to-value ratio (LTV) of 60% of the homeowners who filed for bankruptcy was 90 percent or above and the average LTV was 0.91. These facts suggest the impact of omission of homeowners with large home equity in the estimation will be small.

68

3.3.2. Homestead exemption as a random variable Another problem with the SCF is that recently it does not disclose any geographical information of respondents in the public datasets. Therefore we cannot observe the exact amount of the bankruptcy exemptions, particularly the homestead exemption, which would be applied to each household58. Let us look at the distribution of the homestead exemption in 1996 weighted by the home-owning population of each state, which is presented in Table 3-3. Significant frequencies appear at zero, 10,000, 15,000, 20,000 and 75,000 dollars while the unlimited exemption also has a large frequency. Since we do not observe which state the households in the SCF reside in, we treat a household’s state of residence and associated homestead exemption as random variables drawn from this distribution. Then we know the larger the household’s home equity is, the more likely the household is to fall into the region where H > X h holds, which implies that the household would lose its home if it filed for bankruptcy. From Equation (3.2), this specification of the homestead exemption gives an alternative representation of the financial benefit from bankruptcy for the purpose of the estimation as follows: FinBen = Pr ( X h ≥ H | H ) ⋅ Finben1 + Pr ( X h < H | H ) ⋅ Finben0 , where FinBen1 = max[D − max[W − X w ,0],0] − S ; FinBen0 = max[D − max[H − X h ,0] − max[W − X w ,0],0] − S − E [G ] ⋅ V .

(3.3)

Now X h is a random variable following a known distribution. It is worth noting that; ∂ FinBen = − Pr ( X h < H | H ) ; ∂H

58

The unavailability of geographical information causes another problem when we estimate the rate of house price appreciation. See the discussion about this issue in Chapter 2.

69

∂ FinBen = − Pr ( X h < H | H ) . ∂E [G ] ⋅ V

(3.4)

Thus, under the specification of the level of homestead exemption as a random variable, the larger a household’s home equity is, (1) the stronger the negative effect of home equity on the financial benefit is, and (2) the stronger the negative effect of expected capital gains on the financial benefit is. For illustrative purposes, Figure 3-3 shows the effect of home equity on the financial benefit from bankruptcy in three examples of different values of the homestead exemption.

3.3.3. Estimation methodology for binary choice I describe my empirical specification here. We can reasonably assume that households compare their respective optimal utilities for each case, whether to file for bankruptcy or not, and choose the most satisfactory one, conditional their economic and demographic characteristics. I further assume that the indirect utility function for the i-th household over the j-th bankruptcy choice (j = 1 stands for filing for bankruptcy, j = 0 otherwise) is specified as follows: Vij* (x i ) = α j + β ′j x i , i = 1,K N , j = 0,1 , where Vij* (x i ) is the optimal utility of the i-th household for the choice j conditional on x i , α j is a choice-specific constant, β ′j is the (transposed) vector of the model parameters for each choice j, and x i is the vector of the explanatory variables that are considered to affect the i-th household’s indirect utilities. Now define: Vi* = Vi *,1 − Vi*, 0 , i = 1, K , N . Then Vi * = α + β ′x i , i = 1, K N ,

70

where α = α 1 − α 0 and β = β 1 − β 0 (vector). Therefore the i-th household makes the decision on filing for bankruptcy based on the value of Vi * such that it files for bankruptcy if Vi * > 0 while it does not do so otherwise. This leads to a general specification for the probability of observed bankruptcy filings as follows:

(

)

Pr ( y i = 1) = Pr Vi* > 0 = F (α + β ′x i ), i = 1, K , N

where yi is the binary dependent variable which represents the choice of the i-th individual on bankruptcy filing, taking a value 1 if the household i files for bankruptcy, and 0 otherwise. F is an appropriate distribution function. I use the standard probit function for F as follows: Pr ( y i = 1) =

α + β′xi

∫ φ (t )dt ,

i = 1, K , N

−∞

The maximum likelihood method is used to estimate the parameter vector β in the sections that follow.

3.3.4. Piecewise specification for homestead exemption Since the main purpose of the present chapter is to address the effects of the homestead exemption and capital gains expectations on households’ decisions on bankruptcy, I need a more careful specification about home equity and capital gain expectations. While households’ home equity is straightforward to calculate with the information that the SCF provides, their expectations about capital gains from their mortgaged properties are not so. As in the literature on mortgage default, previous studies of personal bankruptcy have not conducted much empirical examination of a possible effect of households’ backward-looking expectation on their bankruptcy decisions. In fact, as long as I know, there are no previous studies that investigate this subject. Thus I will test this effect by including a proxy variable for 71

backward-looking expectations into the estimation model. Since I see the convincing performance of the average rate of house price appreciation as a proxy for households’ backward-looking expectation in Chapter 2, I continue to rely on this variable. Following Chapter 2, this proxy variable is defined as follows: G0 =

ln (Vc ) − ln (Vo + I ) , Year

where G0 is the average rate of house price appreciation per year, Vc is current house price, Vo is original house price, I is the cost of home improvement (if any) and Year is the number of years since the households acquired their homes. Then, CapGain = Vc ⋅ G0 , where CapGain is the proxy for households’ capital gain expectation. From Equation (3.4), we expect that the negative effects of home equity and capital gain expectations on the financial benefit from bankruptcy will increase as home equity goes up. Since the distribution of the homestead exemption is discrete and has spikes at several values as we see in Table 3-3, I apply a piecewise specification to the variables of home equity and capital gains expectations. I pick up the two most frequent values in the homestead exemption distribution, that is, 10,000 and 75,000, to divide the subsample into three groups according to households’ home equity: households with home equity smaller than 10,000, those with home equity between 10,000 and 75,000, and those with home equity greater than 75,000. The number of households that belong to each group is shown in Table 3-4. Then I allow these three groups to take different values of the coefficients on home equity, capital gains expectations and the intercept term. From this specification and Equation (3.4), we expect that the greater home equity the group is associated with, the stronger the negative coefficient of the group will be. This prediction can be

72

illustrated as Figure 3-4.

3.3.5. Explanatory variables While borrowers’ home equity and capital gains expectations are major determinants of their financial benefit from bankruptcy, the other variables in Equation (1.1), that is, unsecured debts, D, and non-housing assets, W, are also important components of their financial benefit from bankruptcy. However, I do not include these variables because the values of these variables for households that filed for bankruptcy recently will be totally different between before and after filing; bankruptcy proceedings would have discharged unsecured debts while non-housing assets would have been liquidated if their value exceeds the bankruptcy exemption59. Another objective of this chapter is to test the second theory of personal bankruptcy decision, the ability-to-pay theory. From a strict viewpoint of the theory of borrowers’ willingness to pay, households only care about the financial profits that they would be able to obtain if they filed for bankruptcy. Thus the explanatory variables that measure household’s ability to pay, such as household income and the amount of loan payments, should be irrelevant to their decisions on bankruptcy. In particular, exogenous shocks to their ability to pay, or “trigger events”, such as unemployment, should not have any predictive power in the estimation of the probability of bankruptcy filings. However, even though Fay, Hurst and White (2002) calculate the financial benefit carefully by making use of all the information available in the PSID, they still find household income has a negative effect on the probability of bankruptcy. To test this issue by the dataset from the SCF, I include the variables of the ability-to-pay theory: household income, the

59

I added the variable of unsecured debts to the estimations presented later. Contrary to the theory, its effect on the probability of bankruptcy is strongly negative, implying the discussion here is relevant. Thus I do not include the variables of unsecured debts and of non-housing assets hereafter.

73

amount of mortgage payment, and recent unemployment experience. Moreover, as the model developed in Chapter 2 suggests, it is reasonable to consider credit constraints to be closely related to the ability-to-pay theory: credit constrained households may be forced to file for bankruptcy even if they do not think bankruptcy is financially profitable. Thus I also include the dummy variable for credit constraints, which is defined in Chapter 2, and some other variables that are relevant to credit constraint, namely, age, sex, race (black) and Hispanic, supporting children, education (college degree), following the argument in Chapter 2. Lastly, the year dummies are included in the estimation to follow. The descriptions and summary statistics of these explanatory variables are presented in Table 3-5.

3.4.

Estimation results Table 3-6 presents the estimation results of the probit model specified in the previous

section. As specified in that section, the coefficients of home equity and expected capital gain are estimated separately for three groups of home equity: home equity smaller than 10,000, between 10,000 and 75,000, and greater than 75,000. Instead of setting a base category, this table reports estimated coefficients for each group separately in order to illustrate the effects clearly and to test their statistical significance directly. On the other hand, the intercepts for these groups are reported using the group with home equity of greater than 75,000 as the base group. Even though the estimated coefficients for home equity are all negative as predicted, the relationship of its magnitude between different home equity groups is not consistent with the model shown in Equation (3.4); the group with the greatest home equity is subject to the smallest marginal effect of home equity among three groups. This might be caused by an endogeneity problem. If households with greater home equity than homestead exemption plan to file for bankruptcy and

74

want to keep their home after bankruptcy proceedings for some reasons, they have an incentive to borrow more against their home equity in order to reduce their home equity below the level of homestead exemption60. Thus households’ bankruptcy decisions may affect their home equity, while home equity is one of the main determinants of financial benefit from bankruptcy at the same time. The coefficients for expected capital gains, however, completely agree with the prediction of the model. Although all the coefficients on expected capital gain are negative, only the coefficient for the group with the greatest home equity is estimated to be statistically significant. This result implies that households are taking into account backward-looking expectations of capital gains as a part of financial benefit from bankruptcy filing when they make a decision on bankruptcy. The results reported in Table 3-6 strongly support the ability-to-pay theory, too. Household income has a negative effect on the probability of bankruptcy, while the amount of regular mortgage payments has a positive effect. Unemployment seems to work as a trigger event. Credit constraints also has a positive effect, along with the predicted effects of the variables related to credit constraints, such as age, sex, race, supporting children and education. All the coefficients of the ability-to-pay theory are statistically significant at the 1% level, except for female, which is statistically significant at the 5% level, and black. Overall, the results are basically consistent with the literature on personal bankruptcy, in which the willingness-to-pay theory and the ability-to-pay theory coexist, while it finds that backward-looking expectations of housing price play some role in households’ decisions on bankruptcy.

60

Bahchieva, Wachter and Warren (2005, pp.107) argue for this possibility.

75

3.5.

Bankruptcy decisions of mortgage borrowers in “dead-end delinquency”

3.5.1. Adjustment for dead-end delinquent households In this section, I define a smaller group of mortgage borrowers to investigate further the bankruptcy decision model, in particular, in order to test the effect of backward-looking expectations of capital gains. This selection is motivated by a recent article in the New York Times61. The article reports the story about the struggles of households who cannot refinance their mortgage any longer, even though they could do so before the recent downturn in the housing market. The article claims that they are now forced to file for bankruptcy. This story suggests that there exists a situation in which households have to file for bankruptcy against their will, due to shortage of ability to repay, probably caused by unexpected adversities and changes in their circumstances, such as unemployment or a sudden increase in payments of ARMs. Credit constraints play a key role in this type of situation; borrowers cannot escape from financial distress because they cannot borrow additional money. This argument is one specific case of the discussion about borrowers’ choice that I examined in Chapter 2, in which the estimation result implies that some borrowers are forced to become delinquent due to credit constraints. Suppose a credit constrained household with mortgage debt is caught by such a severe and persistent financial distress that makes it impossible for the household to keep making all mandatory loan payments. Let us call this situation “dead-end delinquency”. I assume the only possible ways for the household to escape from this dead-end delinquency are as follows: 0) Sell its home; 1) Default on its mortgage62;

61

“Mortgage Crisis Spreads Past Subprime Loans”, The New York Times, February 12, 2008. (http://www.nytimes.com/2008/02/12/business/12credit.html) 62 Some recent news articles, such as an article in the New York Times, February 29, 2008, report a new borrowers’

76

2) File for bankruptcy; 3) Combination of (0) and (2) or (1) and (2). The decision tree is displayed in Figure 3-5. As in Chapter 2, I assume that the determinants that put households into dead-end delinquency, such as income, payments, credit constraints and adverse events, are exogenously given to households, but keep assuming that households behave based on financial profits: they will make a choice between the four alternatives specified above from (0) through (3) comparing the financial profits that they would gain (or loss) for each outcome. Thus the model developed in Section 2 can also be applied to the bankruptcy decisions of dead-end delinquent households, but it needs some adjustment since the base outcome is now selling their homes rather than continuing repayment to keep their homes; they cannot avoid filing for bankruptcy without selling their homes to recover financial health63. The financial benefit of a dead-end delinquent household from filing for bankruptcy changes from Equation (3.3) as follows: FinBen = Pr ( X h ≥ H | H ) ⋅ Finben1 + Pr ( X h < H | H ) ⋅ Finben0 , where FinBen1 = max[D − max[W − X w ,0],0] − S + E[G ] ⋅ V ; FinBen0 = max[D − max[H − X h ,0] − max[W − X w ,0],0] − S . The difference from the previous specification in Equation (3.2) is the role of expected capital

option called “walk away”. Essentially, however, this option can be viewed as a form of mortgage default since borrowers abandon their home equity in exchange for release from mortgage debts. Nonetheless, this option should be examined by comparison with the traditional mortgage default, particularly from the perspective of transaction costs for borrowers. (http://www.nytimes.com/2008/02/29/us/29walks.html) 63 From the viewpoint of the willingness-to-pay theory, mortgage default can be viewed as a “put option” to give the mortgage property back to the lender at the price of current mortgage balance (less the sanction cost of future credit) as discussed in Chapter 2, so I will not distinguish mortgage default from selling the house in the market hereafter for the purpose of concentrating on households’ bankruptcy decisions. In particular, in this theoretical framework, it never causes any changes in the prediction about the effect of capital gains expectations by a similar argument as shown in Equation (2.10) in Chapter 2.

77

gains, E [G ]⋅V . Since even if the household does not file for bankruptcy, it will lose its home either by selling it on the market or due to mortgage default, the loss of expected capital gains from the mortgaged property due to bankruptcy proceedings is no longer a cost of bankruptcy. On the contrary, the household could protect its home through bankruptcy, if its home equity is below the homestead exemption. Therefore, expected capital gains should be added to the financial benefit in this case. We can check the marginal effects of home equity and capital gain expectation in the same way as Equation (3.4); ∂ FinBen = − Pr ( X h < H | H ) ; ∂H

∂ FinBen = Pr ( X h ≥ H | H ) . ∂E [G ] ⋅ V

(3.5)

The prediction from this model of dead-end delinquent households about home equity remains the same as the previous model; the greater their home equity is, the more likely home equity is to have a negative marginal effect on their financial benefit from bankruptcy. On the other hand, the smaller their home equity is, the more likely households are to be able to obtain capital gains from their homes, as a part of the financial benefit from bankruptcy. Households with smaller home equity are more likely to protect their home through bankruptcy proceedings, since their home equity is more likely to be below the homestead exemption. In other words, the smaller their home equity is, the more likely they are to be able to continue their potentially profitable housing investment by taking advantage of the U.S. bankruptcy law.

3.5.2. Estimation results for dead-end delinquent households The major problem with testing these predictions empirically is how to define the subsample of the dead-end delinquent households in the dataset. Although the SCF does not ask 78

questions which correspond to my definition of dead-end delinquency, I set two criteria to characterize the dead-end delinquent households as follows: 1) Households that missed any of their loan payments for at least three months; 2) Households that are credit constrained. More than three months of delinquency are considered to be the situation in which borrowers cannot return to a sound financial status in the business practice (U.S. Department of Housing and Urban Development, 2007). Although households with mortgage debts can handle financial problems by refinancing or additional borrowing, these options would not be available if their loan applications were rejected, in other words, they were credit constrained. Since the SCF asks questions about the households’ loan applications as discussed in Chapter2, we can specify households that cannot borrow. Thus we can admit the subsample of the households satisfying these two conditions above are a good approximation of the group of households that suffer from dead-end delinquency, in which the households either have to sell its home, file for bankruptcy or both. Therefore I choose the subsample based on the criteria described above out of all the households in the SCF in 1998, 2001 and 200464. The number of observation in this subsample is 539 (see Table 3-7). The rate of recent bankruptcy experiences is 13.0%, which is not surprisingly much higher than the previous subsample (1.1%). Table 3-9 presents the probit estimation for the probability of bankruptcy filing among this subsample, along with the summary statistics in Table 3-8. The effect of home equity is now consistent with the model described as Equation (3.5).

64

In addition to the questions about denial of loan applications described in Chapter 2 as a variable for credit constraints, I choose households that answer yes to the question as follows: “Were you ever behind in your payments by two months or more during last year?” (defined as “serious delinquency” in Table 1-1).

79

While its effect for the group of the smallest home equity is not statistically significant, the probability of bankruptcy filings for the other two groups is negatively affected by home equity and the coefficients are statistically significant. The endogeneity problem that I am worried about in the previous estimation seems to be less important in this case; credit constrained households will not be able to reduce their home equity by borrowing against their home in order to take advantage of the homestead exemption. The coefficients on capital gain expectation dramatically changes from the previous estimation presented in Table 3-6. The positive and statistically significant coefficient for the group of the smallest home equity implies that households with small home equity are more likely to file for bankruptcy if they expect larger capital gains from their homes. This result matches the prediction in Equation (3.5), strongly suggesting that even households in dead-end delinquency are engaged in strategic use of the bankruptcy law. Although the negative and marginally statistically significant coefficient for the group of the middle-level home equity is remarkable, it may be attributed to the imperfect separation of the dead-end delinquency subsample from the whole sample. The estimated coefficients for the ability-to-pay variables are quite different from the previous estimation, too. In particular, the main indicators of households’ ability to pay, namely household’s income and mortgage payment, no longer have statistically significant effects; the sign of the coefficients for income is even opposite to the previous one. Moreover, contrary to the prediction, recent unemployment experience, which is one of the primary examples of trigger events, has a negative and statistically significant effect on the probability of bankruptcy. It may imply that if households fall into dead-end delinquency due to temporary unemployment, they expect that they will be able to endure financial distress until they find a job and get back to

80

healthy financial balance. The coefficients for the other demographic variables have the predicted sign, but some of them are not statistically significant in this estimation. In short, the impact of the ability-to-pay variables is considerably diminished if we restrict the sample to households that are in serious delinquency and credit constrained. Chapter 2 presents the theoretical framework composed of two aspects for bankruptcy and mortgage default; some households choose to file for bankruptcy (or mortgage default) in order to gain financial profits, while some households are forced to file for bankruptcy because they cannot continue to make regular payments due to insolvency and credit constraints. The estimation results about the ability-to-pay variables in this section reinforce the validity of this theory. Once households get into hopelessly severe insolvency, their ability to pay is no longer relevant to their decisions on bankruptcy. Instead, they will make a decision between selling their homes (or default on their mortgages) and filing for bankruptcy, in view of financial profits that they would obtain from each alternative.

3.6.

Conclusion In this chapter, I construct a model of bankruptcy decisions of households with mortgage

debts, focusing on the effects of homestead exemption and backward-looking expectations of capital gains. The interaction of homestead exemption and backward-looking expectations shows a significant effect. Households that expect large capital gains are less likely to file for bankruptcy in order to avoid risking their homes. I also specify a financial circumstance in which borrowers cannot continue repayment unless they liquidate their housing assets: termed “dead-end delinquency”. Dead-end delinquent households with small home equity seem to try to protect their homes through bankruptcy if they expect capital gains. Along with the findings in Chapter 2, backward-looking expectations of capital gains play an important role in households’

81

decisions on loan termination, both mortgage default and bankruptcy. Contrary to the willingness-to-pay theory, the ability-to-pay variables have statistically significant effects, but those effects greatly diminish if we concentrate on the behavior of households that are already in dead-end delinquency. This result supports my theoretical considerations in Chapter 2, which decompose households’ incentives for bankruptcy into two distinct aspects: financial profits and insolvency due to credit constraints. Since dead-end delinquent households have to give up meeting their liabilities of loan contracts, their ability to pay does not matter any longer. They make a decision based on financial profits. The next step of the research on this subject should be the relationship between credit constraints and excessive borrowing. If a sizable fraction of loan delinquency, mortgage default, and personal bankruptcy are caused by credit constraints that are imposed on households, against their will based on financial incentives, those consequences could be reduced by removing credit constraints. Therefore, if there is imperfection in credit markets, as Stiglitz and Weiss (1981) argue, then such imperfection results not only in limited access to credit for potential borrowers, but also in deterioration in payment performances of existing borrowers. On the other hand, credit constraints may well reflect lenders’ appropriate unwillingness to lend money to borrowers who have already owed a lot of monely, in other words, households who have borrowed more than needed for smoothing their lifetime income profile. Since my model treats credit constraints, income, loan balance and the stream of loan payments as exogenous, it cannot judge the nature of credit constraints as discussed here65. As for empirical aspects, studies with richer datasets, particularly panel datasets, should be done to obtain more precise estimations of households’ decisions on loan delinquency,

65

Getter (2003) concludes that delinquency risk is more likely to increase as a result of unexpected shocks rather than an excessive burden of payments that households take on by using the 1998 SCF.

82

mortgage default and personal bankruptcy. More information about laws and loan contracts, such as precise differences in levels of homestead exemption and in availability of deficiency judgment by state, will be needed for accurate estimations.

83

Figure 3-1 Time of the survey

Equity ≤ X h

Keep home

(observable)

Lose home

(unobserved)

Bankruptcy Equity > X h

Homeowners Used in the regression

No bankruptcy experience (observable)

Figure 3-2 Recent Bankruptcy

Equity ≤ Eh

No Bankruptcy Experience

- Recent bankruptcy record - Mortgage debtor

- No bankruptcy record - Mortgage debtor

- Recent bankruptcy record

- No bankruptcy record - Mortgage debtor

Equity > Eh - Lost home (unobservable unobservable) unobservable - Not mortgage debtor

This group is unobservable

84

Figure 3-3 Financial benefit

(3) X h = X h 3 (1) X h = X h1

K + X h1

X h1

( 2) X h = X h 2

X h2

K + X h2

H (Home equity)

Note: K = D − max[W − X w ,0] X h1 < X h 2 < X h 3 = ∞ (unlimited)

Figure 3-4 Financial benefit

H 10,000

75,000

85

(Home equity)

Figure 3-5 Equity ≤ X h

Keep home

Equity > X h

Lose home

Bankruptcy

Equity ≤ X h

Homeowners

“Dead-end” delinquency

Keep home

Bankruptcy Equity > X h

Lose home

- serious delinquency - credit constrained

Sell home

Lose home

Default

Lose home Keep payment

86

Table 3-1 Homestead exemption, population, homeownership by state

Alabama Alaska Arizona Arkansas California Colorodo Connecticut Delaware Florida Georgia Hawaii Idaho Illinois Indiana Iowa Kansas Kentucky Louisiana Maine Maryland Massachusetts Michigan Minnesota Mississippi Missouri Montana Nebraska Nevada New Hampshire New Jersey New Mexico New York North Carolina North Dakota Ohio Oklahoma Oregon Penn Rhode Island South Carolina South Dakota Tennessee Texas Utah Vermont Virginia Washington West Virginia Wisconsin Wyoming Mean

Homestead exemption

Population

homeownership rate

10,000 62,100 100,000 Unlimited 75,000 60,000 150,000 0 Unlimited 10,000 50,000 100,000 15,000 15,000 Unlimited Unlimited 10,000 15,000 25,000 0 100,000 7,000 200,000 150,000 8,000 80,000 20,000 125,000 60,000 0 60,000 20,000 20,000 160,000 10,000 Unlimited 33,000 0 0 10,000 Unlimited 7,500 Unlimited 11,000 150,000 11,000 60,000 30,000 40,000 20,000

4,290,403 604,918 4,432,308 2,504,858 31,780,829 3,812,716 3,267,030 727,090 14,426,911 7,332,225 1,184,434 1,187,706 11,953,003 5,834,908 2,848,473 2,598,266 3,881,051 4,338,763 1,241,436 5,057,142 6,085,393 9,739,184 4,647,723 2,709,925 5,367,888 876,656 1,647,657 1,596,476 1,160,768 8,009,624 1,706,151 18,143,805 7,307,658 642,858 11,187,032 3,289,634 3,195,087 12,038,008 987,858 3,738,974 730,699 5,313,576 19,006,240 2,022,253 586,352 6,665,491 5,509,963 1,818,983 5,173,828 480,085

71 62.9 62 66.6 55 64.5 69 71.5 67.1 69.3 50.6 71.4 68.2 74.2 72.8 67.5 73.2 64.9 76.5 66.9 61.7 73.3 75.4 73 70.2 68.6 66.8 61.1 65 64.6 67.1 52.7 70.4 68.2 69.2 68.4 63.1 71.7 56.6 72.9 67.8 68.8 61.8 72.7 70.3 68.5 63.1 74.3 68.2 68

48,595

5,293,806

67

87

Table 3-2 Bankruptcy experience rate among mortgage debtors

Observations Bankruptcy within 2 years Percentage

1998

2001

2004

Total

7,557

7,867

8,110

23,534

115

70

80

265

1.522%

0.890%

0.986%

1.126%

88

Table 3-3 Homestead exemptions in 1996 (weighted by the population of homeowners) Homestead exemption

Population weight (1)

Frequency

Cumulative

0

18,267,694

10.56

10.56

7,000

7,138,822

4.13

14.68

7,500

3,655,740

2.11

16.79

8,000

3,768,257

2.18

18.97

10,000

21,435,485

12.39

31.36

11,000

6,036,039

3.49

34.85

15,000

15,297,307

8.84

43.68

20,000

16,133,469

9.32

53.01

25,000

949,699

0.55

53.56

30,000

1,351,504

0.78

54.34

33,000

2,016,100

1.16

55.5

40,000

3,528,551

2.04

57.54

50,000

599,324

0.35

57.89

60,000

7,835,315

4.53

62.42

62,100

380,493

0.22

62.64

75,000

17,479,456

10.1

72.74

80,000

601,386

0.35

73.08

100,000

7,350,740

4.25

77.33

125,000

975,447

0.56

77.89

150,000

4,644,701

2.68

80.58

160,000

438,429

0.25

80.83

200,000

3,504,383

2.03

82.86

Unlimited

29,667,590

17.14

100

Total

173,055,931

100

(1) Author's calculation based on U.S. Census Bureau (2007)

89

Table 3-4 The distribution of home equity in the subsample Home equity

Observations

Percentage

Cumulative

~0

277

1.18%

1.18%

1 ~ 10,000

846

3.59%

4.77%

10,001 ~ 75,000

7,978

33.90%

38.67%

75,001 ~

14,433

61.33%

100%

Total

23,534

100%

90

Table 3-5 Summary statistics of explanatory variables Description

Mean

Std. Dev.

387,684

965,637

35,252

121,914

Home equity

current housing value less mortgage balance

Expected capital gain

average annual rate of house price appreciation times current housing value

Current housing value

(estimated by respondents)

581,399

1,207,249

Income

household's wage income

155,826

497,380

Mortgage payment

monthly payment of mortgages

Credit constrained

denial or expected denail of loan applications

Age

household head's age

Age squared

1,860

4,289

0.1575

0.3643

49.03

11.36

household head's age squared

2533.42

1178.53

Female

household head is female

0.1144

0.3183

Black

household head is black

0.0616

0.2405

Hispanic

household head is Hispanic

0.0504

0.2188

Children

household supporting children

0.5750

0.4944

College degree

household head has a college degree

0.5871

0.4924

Unemployment experience

household head experienced unemployment within 12 month

0.0582

0.2342

Observations

23,534

Year 2004

8,110

Year 2001

7,867

Year 1998

7,557

91

Table 3-6 Probit model for bankruptcy filings Probability of bankruptcy Coefficient

Marginal effect (1)

t

0< Equity