Job Loss, Defaults, and Policy Implications Kyle Herkenhoff∗ February 20, 2012

Abstract I use new household level data to simultaneously test several competing hypothesis concerning mortgage default. Unlike past studies that use lender data, household data allows me to consider the potential interactions of (i) job loss, (ii) negative equity, and (iii) unsecured debt. I also consider the role of a fourth factor, which is often ignored in negative equity papers, (iv) prior mortgage modification. I find that the best absolute predictors of default are job loss, which makes someone 8.2% more likely to default, prior mortgage modification, which makes someone 14.0% more likely to default, and negative equity of at least -20%, which makes someone 11.2% more likely to default, ceteris paribus. Those people with both moderate negative equity and job loss are 37.0% more likely to default, than those with just negative equity. However, there is a non-monotonicity as the interaction between job loss and severe negative equity is insignificant, indicating that once someone reaches the point of -20% negative equity, job loss is irrelevant and they are likely to ruthlessly default regardless of whether or not they are out of work. Based on these results I propose several policy reforms with the overarching theme that it may be more efficient to stabilize the housing market through indirect economic channels that generate long run job growth rather than temporarily modifying mortgages.

∗

Herkenhoff: UCLA, 9273 Bunche Hall, Los Angeles, CA 90024, [email protected]. I am grateful for helpful comments made by Lee Ohanian. The first draft of this paper was written June 1, 2011

1

1

Introduction

Using household level data on mortgage delinquency, I compare several competing mortgage default hypotheses: 1. Negative equity is main trigger of default (Goodman, Ashworth, Landy, and Yin (2010)) 2. Illiquidity is the main trigger of default (Elul, Souleles, Chomsisengphet, Glennon, and Hunt (2010)) 3. Job loss is the main trigger of default (this paper) 4. Prior mortgage modification is the main trigger of default (this paper) I consider each of the hypothesis in two econometric frameworks, the Linear Probability Model (LPM) and the Logit model. In both frameworks, I find that the relationship between negative equity and default is non-monotonic with people ruthlessly defaulting only when the value of the house is at least 20% less than the mortgage debt owed, but at more moderate levels of negative equity only job loss triggers default, a result similar to Bhutta, Dokko and Shan (2010).1 The joint event of negative equity and job loss is a large and significant predictor of default, reaffirming the results in Foote, Gerardi, and Willen (2009) who hypothesize that double trigger events are the most likely culprit in predicting default since negative equity is a necessary but not sufficient condition for default.2 I also find that at high levels of negative equity, unsecured debt burden is a strong and significant predictor of default in the Linear Probability Model, but this result does not carry over to the Logit model. Not surprisingly, people who have modified their loans in the past are 14% more likely to default in either of the models discussed in this paper. This observation is in line with past work such as Adelino, Gerardi, and Willen (2009), Haghwout, Peach, and Tracy (2010) and Herkenhoff and Ohanian (2011). Mortgage modifications, as implemented by the Obama administration merely delayed the inevitable rather than stemming long term defaults and induce significant moral hazard. 1

Ruthless default is a term that refers to default solely on the basis of negative equity, which is when the property is worth less than what is owed. See Vandell (1995) for the basics of mortgage default 2 Double trigger events refer to job loss, divorce, negative equity, or any other negative life even occurring simultaneously

2

Based on the findings in this paper, policies that can better stem long term defaults involve stabilizing the labor market instead of temporarily reducing mortgage payments. To this extent, I propose several policy reforms including broad based tax reform to promote foreign direct investment and reinvestment, eliminating H1-B visa acquirements for skilled entrepreneurs that hire Americans, changing the structure of unemployment benefits to reward job finding and retraining instead of non-employment, and rolling back minimum wages to stop low skilled workers from being priced out of the labor market. Section 2 discusses the related literature. Section 3 describes the data and how I use the survey timing to generate time variation. Section 4 describes the initial trigger results in a linear probability model. For robustness, Section 5 includes the logit specification of the model. Section 6 provides policy recommendations based on these results, and Section 7 concludes.

2

Related Literature

The seminal papers by Mayer, Pence, and Sherlund (2009) and Foote, Gerardi, and Willen (2009) restarted the trend of empirically investigating the role of negative equity in the decision to default for the 2007-09 recession. Foote, Gerardi, and Willen use the Massachusetts Registry of Deeds to look at the affect of negative equity on the decision to default. In line with their theoretic model, they find that the majority of people with negative equity do not default. They find that the low default rates by homeowners reflect price expectations and that those who actually did default likely defaulted because of a double trigger event- negative equity and some adverse life event like job loss or health problems. To potentially capture these trigger effects, the authors use a local unemployment indicator, which has now become a standard in the literature. Goodman, Ashworth, Landy, and Yin (2010) try to disentangle the cause of defaults- whether its negative equity or unemployment- by using Loan Performance data and unemployment rates by city. They conclude that negative equity predicts default movements more so than regional unemployment, but they explicitly discuss the limits of using a regional unemployment rate and the bias it induces toward negative equity: “It is important to realize that we cannot tie the employment status of an individual loan to a particular borrower; we can only 3

tie the unemployment rate of that MSA to a resident borrower. While we use a similar methodology to derive mark-to-market CLTV from original CLTV, the distortion is likely to be less dramatic for CLTVs. That is, if the unemployment rate in a particular area is only 10%, a particular borrower is only 10% likely to be unemployed. However, if homes in a given area have depreciated by 40%, that borrowers house is likely to have dropped a relatively similar amount.” (p. 4) There is a considerable amount of work, such as Mayer, Pence, and Sherlund (2008), and Haghwout, Tracy, and Peach (2008), studying the role of origination standards on default decisions; among the many factors they consider, these papers also control for the local unemployment rate, but the studies are based on mortgage data that cannot be linked to employment history. Nonetheless, Haghwout, Tracy, and Peach find that a 1% increase in unemployment increases the likelihood of default by .25%. Elul, Souleles, Chomsisengphet, Glennon, and Hunt (2010) use the Equifax database combined with loan-level mortgage data to predict default. They find that high credit card utilization rates (i.e. those who borrow up to their credit limits), large combined loan to value ratios (the first mortgage payment plus second mortgage payment over income) and negative equity are the most important factors in determining default. They control for county unemployment rates, and find that “county-level unemployment shocks are also associated with higher default risk (though less so than high utilization)” (p.1). I will refer to these results as the illiquidity results, since people who borrow over their limits at punitive interest rates must necessarily be cash constrained. Bhutta, Dokko, and Shan (2010) use lender level data from CoreLogic, restricted to people who originated loans with a Combined Loan to Value of 100% (although they do not perfectly see this information), and they combined it with the CoreLogic price information and the TrenData credit delinquency rates by zip code. Using a two step procedure, they find that people only strategically default when there is considerable negative equity (-60% or lower). They posit that for more moderate levels of negative equity, the role of trigger events is important. All of these recent papers are related to a long line of literature on mortgage default that began well before the 2007-09 recession. Campbell and Dietrich (1983) were among the first to estimate a logit model to predict 4

mortgage default. They noted the difficulties at that time of obtaining matched mortgage and demographic data, but they use regional data on unemployment rates and find that there is a statistically significant relationship between regional unemployment and default. Foster and Van Order (1985) describe the fact that wealth maximizing households will exercise the option to default whenever the value of the house plus transactions costs falls below the mortgage amount, “A key point about the model is that personal characteristics of the borrower (income, employment status, etc.) are irrelevant (p.7).”3 So long as equity exceeds the selling costs, there is no default. Riddiough (1991) is the first to hypothesize that trigger events such as divorce, loss of a job, or accident, influence default behaviour. He used a stochastic jump process to model the trigger event and was successful in replicating actual default behavior in simulations. Kau, Keenan, and Kim (1993) incorporate transactions costs and “suboptimal default,” which is just another name for trigger events. They find that these events must have a large and important role in option based models in order to match the data. Vandell (1995) reviews the literature on ruthless default, defined to be a default when the mortgage value exceeds the house value, and hypothesizes that if the borrower loses his or her job and has negative equity, that default is more likely to occur. Deng, Quigley, and Van Order (1996) use a competing hazards framework to analyze default. They use lender side data with controls for regional trigger events; in particular they use the quarterly unemployment rate and annual divorce rate as proxies for the triggers. However, they find that unemployment is not an important factor in their model. The sign of the unemployment coefficient is mixed and insignificant in several cases. Capozza, Kazarian, and Thomson (1997) look at regional unemployment rates in a prepayment model too. They find that regional unemployment rates are unimportant, “The trigger event variables... include unemployment, the divorce rate, and the moving rate... however, the economic impact [of these trigger variables] is secondary to the standard option model variables.”(p.18) Ambrose, Buttimer, and Capone (1997) added to the literature on ruthless default by considering foreclosure delay. Elmer and Seelig (1998) provide a theoretic three period model of trigger events and conclude that “negative equity is, a priori, neither a necessary nor a sufficient condition for default.” They explore the role of insolvency risk and the 3

Schwartz and Torous (1989) is a good example of the classic option theoretic view about mortgage defaults.

5

posit that default occurs when the costs of borrowing to remain solvent and maintain the option to sell the house in the future is negative.

3

PSID Supplement and Exploiting the Survey Timing

The delinquency data is taken from the 2009 PSID Supplement on Housing, Mortgage Distress, and Wealth Data. For the purposes of measuring high frequency default, it is difficult to use the PSID data since it is only administered every 2 years and the data is cross-sectional in nature. However, the recorded date of interview, and the recorded date of job loss allow me to tease out the desired time series relationships. Because the survey asks respondents about recent events as well as current mortgage delinquency status and current unemployment status, there is enough time variation to estimate a binary dependent variable model.

3.1

Exploiting the Survey Timing

The 2009 Wave of the PSID was divided into 12 sub-waves, conducted over the course of the year. The interview date was recorded for each household. They recorded the layoff date of the last job held, and also recorded the current working status of the household. PSID interviewers recorded the number of missed mortgage payments as of the interview date, allowing me to determine if job loss was near the survey date, and if the delinquency spell was short enough to have been potentially caused by job loss. Here is the basic empirical strategy to assess job loss and delinquency: 1. Generate indicator variable for those who lost jobs within last 6 months of the interview date 2. Generate indicator for those who defaulted within the last 6 months of the interview date 3. Control for non-time varying characteristics of the mortgage as well as demographics 4. Use binary dependent variables models (Linear Probability Model and Logit Model) to characterize the relationship between recent job loss and recent default. 6

Restricted sample: the restricted sample includes all household heads who are mortgagors, labor force participants, and are less than 6 months delinquent as of the survey date.

3.2

Variable Definitions and Summary Statistics

Tables 1 and 2 show the variables used in the empirical portion of the paper, the associated definitions, and a numeric summary. The negative equity variable is constructed based on the reported home value (HV ) less the reported first mortgage principal (P R1 ) and the reported second mortgage principal (P R2 ). Keeping with the literature, I express negative equity as the combined loan to value ratio (CLTV): CLT V =

P R1 + P R2 HV

I use indicators for different levels of the combined loan to value ratio to capture potential non-linearities and to make the study comparable to the existing literature on mortgage defaults. I define default as at least 60+ days late. However, for the purposes of this study, I am interested in newly defaulted households, so I consider only those households who are between 2 and 6 months behind on the mortgage. And, as discussed above, I only consider job loss within 6 months of the survey date, in order to study trigger events. Because several variables such as liquid assets and unsecured debt are only measured cross sectionally, I must limit myself to job loss within a reasonable number of months of the survey date in order to make the comparison of hypotheses comparable. Liquid assets to income (LQTI) is defined as all savings, checking, and money market accounts (LQ) over total family income (I): LQT I =

LQ I

This variable is measured only once, as of the survey date. Likewise, unsecured debt to income (UDTI) is defined as credit card charges, student loans, medical or legal bills, and loans from relatives (UD) over total family income (I): UD U DT I = I 7

I control for various mortgage characteristics including the type of mortgage, the interest rate, the remaining term, the presence of a second mortgage, whether or not the mortgage refinanced, and whether or not the mortgage was modified. An oft forgotten facet of real estate law is the only purchase money mortgages (i.e. mortgages used to buy a home directly) are non-recourse loans, whereas refinanced mortgages (which are mortgages taken out to pay off another mortgage) are always treated as recourse loans. Therefore, it is more important to control for the refinance status than for the recourse status of a state. In the restricted sample, the majority of the heads of house are male (81%), and there are quite a few households who have missed 2-6 payments as of the interview date (2.6%). A relatively large number of households (10.72%) reported that they modified their loan at some point, potentially reflecting a poorly worded survey question (see the Appendix).4 Table 3 illustrates the fraction of households between 2 and 6 months delinquent. 11.7% of those who lost their jobs have defaulted within 2-6 months of the survey date, versus only 2.4% of those who have not lost their jobs. Likewise, for those with negative equity of at least -20%, roughly 20.4% have defaulted on their loan. For those with moderate or no negative equity only 2% have defaulted. Similarly, those with unsecured debt to income ratios greater than .75, 7.6% have defaulted compared to only 2.3% of defaults for those with lower unsecured debt to income ratios.

4

All statistics and regressions reported in this paper are based on weighted regressions, using the PSID weights that corrext for the sample bias. While the fraction modified is quite large considering the related literature on modifications, see Adelino, Gerardi, and Willen (2009) and Herkenhoff and Ohanian (2011), it is a cumulative variable and may have been misinterpreted by the respondents to mean any type of workout plan.

8

Table 1: Data Summary for Restricted Sample Symbol

Variable

Mean

CI, Lower

CI, per

Up-

Late Job Loss M odif ied 0 < LQT I ≤ .05 .05 < LQT I ≤ .1 0 < ILLQT I ≤ .05 .05 < ILLQT I ≤ .1 Hospital Divorce .8 < CLT V ≤ 1 1 < CLT V ≤ 1.2 1.2 < CLT V .25 < U DT I ≤ .5 .5 < U DT I ≤ .75 .75 < U DT I 2x M ortgage Ref inanced V ariable Rate Interest Rate > .07 T erm > 15 Y rs

2-6 Months Late Indicator Job Loss Indicator Modification Indicator 0 .1 Divorce Indicator .8< Combined Loan to Value Ratio ≤ 1 1< Combined Loan to Value Ratio ≤ 1.2 1.2< Combined Loan to Value Ratio .25< Unsecured Debt to Income Ratio ≤ .5 .5< Unsecured Debt to Income Ratio ≤ .75 .75< Unsecured Debt to Income Ratio Second Mortgage Indicator Refinance Indicator Variable Rate Indicator Interest Rate>.07 Remaining Term > 15 Yrs

0.02666 0.0322 0.1072 0.41948 0.17427 0.789 0.02588 0.03562 0.04079 0.23201 0.05498 0.03733 0.10892 0.03448 0.06543 0.20533 0.4659 0.0946 0.13306 0.68641

0.01946 0.02434 0.09381 0.39755 0.15745 0.77044 0.01913 0.02761 0.03059 0.21361 0.04429 0.02907 0.09518 0.02625 0.05413 0.18725 0.44359 0.08156 0.11813 0.66556

0.03386 0.04005 0.12059 0.4414 0.19108 0.80755 0.03264 0.04363 0.051 0.25042 0.06567 0.04559 0.12266 0.04271 0.07673 0.22342 0.48821 0.10764 0.14799 0.70726

Number of obs

2915

Mean

CI, Lower

CI, per

45.9054 0.79249 0.65731 13.7681 0.07164 0.16187 0.33936

46.9212 0.83281 0.70294 13.9969 0.09588 0.19806 0.38279

Table 2: Data Summary for Restricted Sample, Cont. Symbol

Variable

Age M ale M arried Education Race Bust Recourse

Age of Head Male Indicator Married Indicator Years of Education of Head Race Dummy Housing Bust Dummy Recourse Dummy

46.4133 0.81265 0.68012 13.8825 0.08376 0.17997 0.36107

Number of obs

2915

9

Up-

Table 3: Comparison of Mean Default Rates Across Job Losers, High Negative Equity, and High Unsecured Debt to Income Ratio Fraction 2 to 6 Months Late

CI for Mean

With Job Loss With No Job Loss

0.117 0.024

[0.053,0.237] [0.018,0.031]

With Combined Loan to Value Ratio>1.2 With Combined Loan to Value Ratio<1.2

0.205

[0.133,0.302]

0.02

[0.015,0.027]

With Unsecured Debt to Income 0.076 Ratio>.75 With Unsecured Debt to Income 0.023 Ratio<.75

10

[0.042,0.133] [0.018,0.031]

4

Trigger Results

The first column of table 4 shows that job loss on its own is important in determining default. Likewise, severe negative equity, defined as negative equity of 20% or more, is also a good predictor of default on its own. Someone who just lost their job is 8.28% more likely to default, ceteris paribus, and someone with severe negative equity (combined loan to value (CLTV)>1.2) is 11.22% more likely to default than those with more home equity, ceteris paribus. This basic regression is a “single trigger” regression since each variable (negative equity, job loss, unsecured debt) is considered in isolation. Column (2) allows for the “double trigger” interaction between the variables of interest. In column (2) the interaction between job loss and moderate levels of negative equity (.8
Liquid assets to income (LQTI) is another important factor in the default decision. Those with low liquid assets (between zero and 5% of income) are 1.7% more likely to default. While this seems intimately related to the level of unsecured debt, surprisingly in the data, the correlation between low liquid assets and high unsecured debt is only .033. Those with high levels of unsecured debt to income (.75 < U DT I) are 3.9% more likely to default on their loans. This is roughly half of the magnitude of the coefficient on job loss, and one fifth of that of modification. A large unsecured credit burden, on its own, is a relatively mild predictor of default. In column (3) I allow for an interaction between unsecured debt and negative equity. For very high levels of negative equity, a high level of unsecured debt is an excellent predictor of default. d2 Late = .3405 d .75 < U DT I d 1.2 < CLT V The interpretation of this coefficient is that those with both severe negative equity and large unsecured debt burdens are 34% more likely to default than those with job severe negative equity, ceteris paribus. Those with just severe negative equity are 9.08% more likely to default then those with more home equity. Column (4) includes both interaction effects and shows that the interaction of job loss and moderate negative equity as well as the interaction of high unsecured debt and severe negative equity are still significant and of similar magnitudes. This means that there is a role for both shocks in explaining default at different levels of negative equity. It is interesting to see that the classic triggers of default such as divorce and high health bills are not significant in this framework. This is consistent with past default models that controlled for the aggregate divorce rate.

12

Table 4: Linear probability model (LPM); Dependent Variable Indicator Missed 2-6 Payments as of the Interview Date (1) LPM Int. Job Loss M odif ied 0 < LQT I ≤ .05 .05 < LQT I ≤ .1 Hospital Divorce .8 < CLT V ≤ 1 1 < CLT V ≤ 1.2 1.2 < CLT V .25 < U DT I ≤ .5 .5 < U DT I ≤ .75 .75 < U DT I

No

0.0828** (0.0418) 0.1400*** (0.0231) 0.0172** (0.0074) -0.0073 (0.0060) 0.0068 (0.0228) 0.0147 (0.0271) 0.0084 (0.0101) 0.0151 (0.0234) 0.1122*** (0.0391) 0.0082 (0.0124) 0.0380 (0.0330) 0.0393* (0.0207)

Job Loss × .8 < CLT V ≤ 1 Job Loss × 1 < CLT V ≤ 1.2 Job Loss × 1.2 < CLT V

(2) LPM Int.

0.0096 (0.0214) 0.1399*** (0.0229) 0.0181** (0.0073) -0.0062 (0.0057) 0.0104 (0.0224) 0.0098 (0.0220) 0.0054 (0.0095) -0.0047 (0.0189) 0.1078*** (0.0398) 0.0100 (0.0124) 0.0293 (0.0274) 0.0393* (0.0208) 0.1282 (0.1179) 0.3722* (0.2120) 0.1203 (0.1934)

.75 < U DT I × .8 < CLT V ≤ 1 .75 < U DT I × 1 < CLT V ≤ 1.2 .75 < U DT I × 1.2 < CLT V

Demographic Controls State Controls Mortgage Characteristics Observations R-squared

Yes Yes Yes 2915 0.18

Job

Yes Yes Yes 2915 0.19

(3) (4) LMP Un- LPM Both sec. Int. Int. 0.0797* (0.0422) 0.1387*** (0.0230) 0.0166** (0.0076) -0.0074 (0.0060) 0.0120 (0.0230) 0.0154 (0.0268) 0.0054 (0.0097) 0.0186 (0.0248) 0.0908** (0.0397) 0.0083 (0.0125) 0.0386 (0.0329) 0.0144 (0.0235)

0.0459 (0.0544) -0.0095 (0.0687) 0.3405** (0.1693)

0.0095 (0.0215) 0.1384*** (0.0228) 0.0175** (0.0075) -0.0064 (0.0057) 0.0153 (0.0226) 0.0105 (0.0218) 0.0025 (0.0091) -0.0024 (0.0193) 0.0889** (0.0399) 0.0102 (0.0124) 0.0303 (0.0273) 0.0138 (0.0235) 0.1281 (0.1179) 0.3701* (0.2114) 0.0778 (0.2203) 0.0451 (0.0543) 0.0023 (0.0672) 0.3395** (0.1701)

Yes Yes Yes 2915 0.19

Yes Yes Yes 2915 0.20

Standard errors in parentheses * significant at 10%; ** significant at 5%; *** significant at 1%

13

5

Robustness Check

The linear probability model is ideal for interpretation of interaction effects. However, the model may be a poor fit if there are few default observations, and since the predictions are not restricted to lie between 0 and 1 there maybe out of sample problems. To verify the linear results, I fit a similar model based on a logistic function, which takes into account the fact that the predicted value must lie between 0 and 1. It also handles the case in which there are few default observations since the curvature of the logistic function is an estimation outcome. To verify the linear probability results, I ran the same set of regressions using a logistic function and maximum likelihood estimation. Table 5 summarizes these results. The point estimates of the interaction coefficients are significant for moderate levels of negative equity and job loss. However, the interaction between severe negative equity and high unsecured debt to income loses its significance; this may have occurred for several of the above reasons associated with the linear probability model. I will not provide an in depth interpretation the logit results since the same basic analysis of the linear probability model applies here, just with more complexity in the way to convey results. The main difference is that the coefficients are not directly translatable into probabilities of default; one must look at the various margins (i.e. the change in the probability of default for various values of independent variables), and these margins may be significant or insignificant depending on the different values of the independent variables that are chosen, especially for the interaction terms. These results are included merely to emphasize the consistency of the point estimate precision of coefficients across the two models, and the consistency of the relative magnitude of coefficients.

14

Table 5: Additional Controls; Dependent Variable Indicator Missed One Payment (1) Logit Int. Job Loss M odif ied 0 < LQT I ≤ .05 .05 < LQT I ≤ .1 Hospital Divorce .8 < CLT V ≤ 1 1 < CLT V ≤ 1.2 1.2 < CLT V .25 < U DT I ≤ .5 .5 < U DT I ≤ .75 .75 < U DT I

(2) (3) (4) No Logit Job LMP Un- Logit Both Int. sec. Int.

2.2704*** (0.5434) 2.8454*** (0.3809) 1.1460*** (0.4159) -0.4205 (0.7393) 0.6744 (0.6708) 0.0835 (0.9324) 0.5834 (0.4879) 0.6618 (0.6404) 1.6222*** (0.5758) 0.4471 (0.4304) 1.3737* (0.8077) 1.2477** (0.5070)

Job Loss × .8 < CLT V ≤ 1 Job Loss × 1 < CLT V ≤ 1.2 Job Loss × 1.2 < CLT V

1.1351 (0.8525) 2.9108*** (0.3890) 1.3477*** (0.4336) -0.1348 (0.7357) 0.7458 (0.6728) -0.1128 (0.8938) 0.4510 (0.4945) 0.1513 (0.7196) 1.7240*** (0.6011) 0.4733 (0.4376) 1.2791* (0.7695) 1.3003** (0.5204) 1.9230 (1.2420) 3.6851** (1.5128) -0.6970 (1.2034)

.75 < U DT I × .8 < CLT V ≤ 1 .75 < U DT I × 1 < CLT V ≤ 1.2 .75 < U DT I × 1.2 < CLT V

Demographic Controls State Controls Mortgage Characteristics Observations

Yes Yes Yes 2848

Yes Yes Yes 2848

2.2341*** (0.5466) 2.8371*** (0.3800) 1.1453*** (0.4240) -0.4693 (0.7656) 0.6820 (0.7304) 0.1072 (0.9182) 0.5586 (0.4800) 0.8422 (0.6202) 1.5352*** (0.5908) 0.4295 (0.4306) 1.3461* (0.7932) 1.2352 (1.1584)

0.1323 (1.3690) -1.1544 (1.7370) 0.9592 (1.5229)

1.1600 (0.8525) 2.9049*** (0.3889) 1.3348*** (0.4363) -0.1835 (0.7526) 0.7517 (0.7332) -0.0710 (0.8892) 0.4524 (0.4802) 0.3406 (0.7126) 1.6502*** (0.6040) 0.4573 (0.4367) 1.2606* (0.7631) 1.2788 (1.1216) 1.9144 (1.2483) 3.5783** (1.5126) -0.7879 (1.1903) 0.0390 (1.3413) -0.9063 (1.7485) 1.2951 (1.4851)

Yes Yes Yes 2848

Yes Yes Yes 2848

Standard errors in parentheses * significant at 10%; ** significant at 5%; *** significant at 1%

15

6

Policy Implications

The existing mortgage modification policy called Making Home Affordable (MHA) Program was put in place in March, 2009. The goal of the program was to help 3 to 4 millions struggling homeowners make payments. However, HAMP was a poorly designed program from the onset, helping relatively few homeowners and not actually addressing the problem of negative equity. The program also induced moral hazard by requiring pre-exiting default. I have listed the main criteria below: 1. The mortgage loan is already delinquent or default is reasonably foreseeable. 2. The borrower documents a financial hardship by completing a Home Affordable Modification Program Hardship Affidavit and provides the required income documentation. The documentation supporting income may not be more than 90 days old.( Roughly 57% of the permanent modifications were for people with employment problems, including outright unemployment.) 3. The borrower has a monthly mortgage payment ratio of greater than 31 percent (mortgage payment over gross income). 4. The loan must pass a standardized NPV test that compares the NPV result for a modification to the NPV result for no modification. If the NPV result for the modification scenario is greater than the NPV result for no modification, the servicer MUST offer the modification, otherwise the servicer has the option of performing the modification in its discretion. The fact that the loan had to be delinquent in order to obtain a modification induced many people to default ruthlessly. See Mayer, Morrison, Piskorski, and Gupta (2011) for a specific example of moral hazard among country wide mortgagors. The NPV test was the other facet of the loan modification program that limited the actual benefit of receiving a modification; with a constant or increasing NPV of mortgage payments, a modification can never actually reduce negative equity. Modifications are also temporary in nature, with payments increasing slowly after 5 years, back to, or close to, their old levels. The important modification process (often called the ‘waterfall’) is listed below: 16

1. Capitalize accrued interest, out-of-pocket escrow advances to third parties, and any other third party fees that are reasonable and necessary. [This often times increases the NPV, considering late fees and tax penalties are charged at roughly 3% per month] 2. Reduce the interest rate. The interest rate floor in all cases is 2.0 percent. The reduced rate will be in effect for the first five years followed by annual increases of one percent per year (or such lesser amount as may be needed) until the interest rate reaches the Interest Rate Cap, at which time it will be fixed for the remaining loan term.5 [Unless house prices rise in 5 years so the mortgagor can sell the house to pay off the mortgage, the mortgagor must improve their income situation to be able to afford the increasing payments] Thus, for those receiving modifications, it was nearly always the case the borrowers left the negotiating table with an increased NPV of payments, worsening their equity position.6 This stands in sharp contrast to a principal reduction, which is loan forgiveness in its most basic form; the bank merely takes a loss and reduces the amount owed by the mortgagor. This truly improves the equity position of a mortgagor, but comes at a large cost to the banks, especially if there is a reasonable chance of a cure (i.e. they start making payments again) or the prospect of redefault (i.e. they default again in the future). Rather than arguing that modifications were not performed correctly or arguing that there should be more principal reductions, I advise eliminating the modification program entirely. Considering that close to 60% of modified loans redefaulted within 2 years of the initial modification, these efforts are inefficient not only for the banks but for the tax payers that must fund this program.7 Instead, America should focus efforts on promoting job growth, which will automatically translate into a more stable housing market with price appreciation. Temporarily reducing negative equity doesn’t change the long run outcome of default and subsequent continued house price declines. I suggest that to truly fix these problems permanently, there needs to be active 5

The ‘Interest Rate Cap’ is the Freddie Mac Weekly Primary Mortgage Market Survey (PMMS) Rate for 30-year fixed rate conforming loans as of the modification date. 6 While payments are usually backloaded, Adelino, Gerardi, and Willen (2009) report that almost 20% of modifications resulted in immediate higher payments. 7 See the OTS OCC 2010-Q3 report, p. 34

17

initiatives in the labor market that generate jobs. Some potential initiatives include the following: • Broad based corporate tax reforms to reduce the employer cost of hiring and investing, including tax breaks for multinationals that invest in America and hire American workers. • Partially roll back minimum wage laws that price out many low skilled workers • Elimination of the H1-B visa requirement for skilled entrepreneurs who hire Americans. • Retraining and Job finding subsidies for unemployed persons, rather than the current benefit structure that incentivizes non-employment. Broad Based Tax Reform: Shifting to a lower flat corporate tax is tantamount in creating more long run job growth. Not only will it allow existing companies to retain more of their profits to invest in capital and hire more labor, it will encourage entrepreneurs to incorporate and tap broader capital markets. A concurrent elimination of the double taxation of dividends will also encourage investment from the household side, expanding the existing capital flows to growing companies. According to the CBO, the United States has the third highest top statutory tax rate of all OECD countries, and the second lowest tax deduction for new capital investment.8 By reducing the tax burden of investing in the United States, Americans will benefit from long run benefits of capital formation which primarily includes job growth. Roll Back Minimum Wages: According to recent research by Davis and Von Wachter (2010), displaced workers face earnings losses of close to 30%. According to Herkenhoff and Ohanian (2012), the median wage of an unemployed person was roughly 12$ per hour. Herkenhoff and Ohanian calculate a re-entry wage equal to 30% of the median wage of unemployed workers, which is the wage that a typical unemployed person can expect to earn upon finding a new job. The real federal minimum wage was hiked in 2007 to $5.85 from $5.15 and hiked again in 2008 to $6.55 and raised once more in 2009 to $7.25. Herkenhoff and Ohanian show that the re-entry wage 8

The statistics are from the November 2009 report by the Congressional Budget Office, “Corporate Income Tax Rates: International Comparisons”

18

and federal minimum wage are within $3 of each other as of 2011, and that a considerable number of people are potentially priced out of the labor market. Elimination of H1-B Visa Requirements for Entrepreneurs: According to the Survey of Earned Doctorates, as of 1966 roughly 20% of doctorate degrees were awarded to temporary residents, but as of 2008, over 40% of doctorate degrees were awarded to temporary residents. Any policy that helps entrepreneurs who start businesses that employ Americans is a win-win for the United States. Not only does it boost job creation, but it contributes to American innovation in the Science Technology Engineering and Mathematics (STEM) fields which are an important ingredient in long term growth. Retraining and Job Finding Subsidies: Reforming unemployment insurance would benefit the labor market by improving incentives to reward job-taking. Instead of subsidizing non-employment, “benefits” should be more explicitly focused on employment efforts and rewarding those who actually accept offers. The current 99 weeks of employment benefits has changed job search incentive for the nearly 7.6 million Americans using unemployment insurance. In particular, it is empirically documented that unemployment exit rate spikes as benefits expire.9 As Casey Mulligan (2011) has argued, replacement rates of income for non-employed persons have roughly doubled from 13% to 26% since the onset of the recession, which means the incentives to find a job after being laid are smaller. Most of these programs are meanstested, which favors those with lower income and acts as a tax on those who raise their income and are subsequently disqualified for certain benefits. He is able to explain as much as 2/3 of the reduction in employment through this channel suggesting that for many non-workers, the cost of taking a job, including the opportunity cost of their time, is greater than the pecuniary benefits. The idea is to reform unemployment benefits by shifting some of the existing benefits into retraining such as scholarships and financial aid for post secondary education, especially for those attempting to attain STEM jobs which historically generate economic growth. Job finding rewards, rather than the usual unemployment insurance, are also a better way of incentivizing re-employment than incentivizing non-employment. A combination of these policies will promote employment among those currently out of work and necessarily lead to a more stable housing market. 9

See Katz and Meyer (1990) for one such study

19

The overarching policy suggestion is that it may be more efficient to aid the housing market recovery through indirect economic channels. Changing the tax code many be the single most powerful tool to affect the long run stability of the housing market, drawing in foreign investment and capital required to make more jobs. The more incentives we provide for companies or entrepreneurs to reinvest those profits inside the US, the more jobs will be created. In terms of long run housing stability, temporary reductions in mortgage payments are a placebo, America needs long run income stability only capable of being maintained by consistent job creation.

7

Conclusion

The results presented in this paper support the past conclusions of Adelino, Gerardi, and Willen (2009), Foote, Gerardi, and Willen (2009), and Bhutta, Dokko and Shan (2010) who all find important roles for modifications and trigger events, respectively. My findings reveal the problems with existing mortgage modification programs but also reveal a new possible path for housing stability: job creation. Job loss is an important factor in the decision to default, and basic intuition suggests that temporary mortgage modifications are not a long run solution to defaults whereas stable household income from long term employment is. The policy prescriptions in this paper are predicated on the idea that indirect labor market stabilization may be a more efficient way of aiding the housing market, especially compared to existing mortgage modification programs. These results open further research opportunities to look at policy efficiency in general equilibrium models with labor markets and insolvency risk. Several inroads have been made toward incorporating labor markets and housing markets in a policy fungible framework, such as Herkenhoff and Ohanian (2012), however, in order to make efficiency arguments, the literature needs more detailed models capable of welfare analysis.

References [1] Adelino, Manuel, Kristopher Gerardi, and Paul Willen. 2009. ‘Why Don’t Lenders Renegotiate More Home Mortgages? Redefaults, SelfCures and Securitization.’ NBER Working Paper no. 15159.

20

[2] Ambrose, Brent W., Richard J. Buttimer, Jr., Charles A. Capone. “Pricing Mortgage Default and Foreclosure Delay.” Journal of Money, Credit and Banking, 29(3): pp. 314-325 [3] Bhutta, Neil, Jane Dokko, and Hui Shan. 2010. “The Depth of Negative Equity and Mortgage Default Decisions.” Working Paper 2010-35. [4] Campbell, Tim S. and J. Kimball Dietrich. 1983. “The Determinants of Default on Insured Conventional Residential Mortgage Loans.” The Journal of Finance, 38(5): 1569-1581. [5] Capozza, D.R., D. Kazarian, and T.A. Thomson. 1997. “Mortgage Default in Local Markets.” Real Estate Economics, 25(4):631-. [6] Congressional Oversight Panel. 2010. “A Review of Treasurys Foreclosure Prevention.” Pub. L. No. 110-343. [7] Davis, S.J. and T.V. Wachter. 2011. “Recessions and the Cost of Job Loss.” NBER Working Paper No. 17638. [8] Elmer, Peter J., and Steven A. Seelig. 1998. “Insolvency, Trigger Events, and Consumer Risk Posture in the Theory of Single-Family Mortgage Default.” Journal of Housing Research, Volume 10, Issue 1 1. [9] Deng, Yongheng, John M. Quigley, and Robert Van Order. 1996. “Mortgage Default and Low Downpayment Loans: The Costs of Public Subsidy.” Regional Science and Urban Economics, 26(3): 263-285. [10] Elul, Ronel, Nicholas S. Souleles, Souphala Chomsisengphet, Dennis Glennon, and Robert Hunt. “What Triggers Mortgage Default.” American Economic Review, 100(2):490-94. [11] Foote, Chris L., Kristopher Gerardi, and Paul S. Willen. 2008. “Negative Equity and Foreclosure: Theory and Evidence.” Journal of Urban Economics, 64(2): 234-245. [12] Goodman, Laurie S., Robert Ashworth, Brian Landy, and Ke Yin. 2010. “Negative Equity Trumps Unemployment in Predicting Defaults.” The Journal of Fixed Income, 19(4):67-72.

21

[13] Haghwout, Andrew, Richard W. Peach, Joseph Tracy. 2008. “Juvenile Delinquent Mortgages: Bad Credit or Bad Economy?” Federal Reserve Bank of New York Staff Reports, Staff Report no. 341. [14] Herkenhoff, K.F., and L.E. Ohanian. 2011. “Labor Market Dysfunction During The Great Recession.” NBER Working Paper 17313. [15] Herkenhoff, K.F., and L.E. Ohanian. 2012. “Foreclosure Delay and US Unemployment.” Unpublished. [16] Foster, Chester and Robert Van Order. 1985. “FHA Terminations: A Prelude to Rational Mortgage Pricing.” Real Estate Economics 13(3): 273291 [17] Katz, L., and B.D. Meyer. 1990. “The Impact of the Potential Duration of Unemployment Benefits on the Duration of Unemployment.” Journal of Public Economics, 41: 4572. [18] Mayer, Christopher J., Edward Morrison, Tomasz Piskorski, and Arpit Gupta. 2001. “Mortgage Modification and Strategic Default: Evidence from a Legal Settlement with Countrywide.” Manuscript. [19] Mayer, Christopher J., Karen M. Pence, and Shane M. Sherlund. 2009. “The Rise in Mortgage Defaults.” Journal of Economic Perspectives, 23(1):27-50. [20] Mulligan, C.B. 2011. “Means-Tested Subsidies and Economic Performance Since 2007.” NBER Working Paper 17445. [21] Riddiough, Tom. J. 1991. “Equilibrium Mortgage Default Pricing with Non-Optimal Borrower Behavior.” University of Wisconsin Dissertation. [22] Schwartz, Eduardo S. and Walter N. Torous. 1989. “Prepayment and the Valuation of Mortgage-Backed Securities. ” Journal of Finance, 44(2): 1989. [23] Vandell, Kerry D. 1995. “How Ruthless is Mortgage Default? A Review and Synthesis of the Evidence.” Journal of Housing Research 6(2):245265.

22

A

PSID Interview Questions

I include some of the important PSID interview questions and question numbers in order to help future researchers locate important variables. • A20. Could you tell me what the present value of your (house/apartment) is–I mean about how much would it bring if you sold it today? • A23b. Is that the original loan and terms or have you refinanced?– FIRST MORTGAGE • A24. About how much is the remaining principal on this mortgage?– FIRST MORTGAGE • A27FOR2. How many months are you behind?–FIRST MORTGAGE • A27FOR5. Have you worked with your bank or lender to restructure or modify your (mortgage/loan) • W38. Aside from the debts that we have already talked about, like any mortgage on your main home or vehicle loans – do you (or anyone in your family living there) currently have any other debts such as credit card charges, student loans, medical or legal bills, or loans from relatives? • W39. If you added up all of these [other debts] (for all of your family living there), about how much would they amount to right now? • About how much did you (and your family) pay out of pocket for nursing home and hospital bills in 2007 and 2008 combined? • BC1. We would like to know about what you do–are you (HEAD) working now, looking for work, retired, keeping house, a student, or what?–FIRST MENTION • BC6. When did you (HEAD) start and when did you stop working for this employer? Please give me all of the start and stop dates if you have gone to work for (this employer/yourself) more than once.– MONTH FOR MOST RECENT MAIN JOB

23