Housing Finance Policy Center Lunchtime Data Talk Mortgage Modifications Using Principal Reduction: How Effective Are They?

Transcription

1 Housing Finance Policy Center Lunchtime Data Talk Mortgage Modifications Using Principal Reduction: How Effective Are They? Ben Keys, University of Chicago Tess Scharlesmann, Office of Financial Research, Dept. of Treasury Max Schmeiser, Federal Reserve Board August 11,

2 The Determinants of Subprime Mortgage Performance Following a Loan Modification Maximilian D. Schmeiser Federal Reserve Board Matthew B. Gross University of Michigan The opinions and conclusions expressed herein are solely my own and should not be construed as representing the opinions or policy of the Federal Reserve Board, Federal Reserve System, or any agency of the Federal government. August 11, 2015 Schmeiser (Federal Reserve Board) Mortgage Modifications August 11, / 40

3 State of the Housing Market Percentage of Loans Delinquent (LPS) Schmeiser (Federal Reserve Board) Mortgage Modifications August 11, / 40

4 State of the Housing Market Ongoing Problem of Foreclosure As of June 2015 Nationally: 3.37% of all mortgages 90+ days delinquent or in foreclosure (1.50% in foreclosure) DC: 3.68% of all mortgages 90+ days delinquent or in foreclosure (2.36% in foreclosure) In July 2007 Nationally: 1.30% of all mortgages 90+ days delinquent or in foreclosure (0.47% in foreclosure) DC: 0.99% of all mortgages 90+ days delinquent or in foreclosure (0.37% in foreclosure) Schmeiser (Federal Reserve Board) Mortgage Modifications August 11, / 40

5 State of the Housing Market Mortgage Delinquency and Foreclosure Nationwide: April 2008 Schmeiser (Federal Reserve Board) Mortgage Modifications August 11, / 40

6 State of the Housing Market Mortgage Delinquency and Foreclosure Nationwide: April 2009 Schmeiser (Federal Reserve Board) Mortgage Modifications August 11, / 40

7 State of the Housing Market Mortgage Delinquency and Foreclosure Nationwide: April 2010 Schmeiser (Federal Reserve Board) Mortgage Modifications August 11, / 40

8 State of the Housing Market Mortgage Delinquency and Foreclosure Nationwide: April 2011 Schmeiser (Federal Reserve Board) Mortgage Modifications August 11, / 40

9 State of the Housing Market Mortgage Delinquency and Foreclosure Nationwide: April 2012 Schmeiser (Federal Reserve Board) Mortgage Modifications August 11, / 40

10 State of the Housing Market Mortgage Delinquency and Foreclosure Nationwide: April 2013 Schmeiser (Federal Reserve Board) Mortgage Modifications August 11, / 40

11 State of the Housing Market Mortgage Delinquency and Foreclosure Nationwide: April 2014 Schmeiser (Federal Reserve Board) Mortgage Modifications August 11, / 40

12 State of the Housing Market Mortgage Delinquency and Foreclosure Nationwide: April 2015 Delinquent Payment: 90+ PD,Foreclosure, US, * Schmeiser (Federal Reserve Board) Mortgage Modifications August 11, / 40

13 State of the Housing Market Mortgage Delinquency and Foreclosure Nationwide: June 2015 Delinquent Payment: 90+ PD,Foreclosure, US, * Schmeiser (Federal Reserve Board) Mortgage Modifications August 11, / 40

14 Motivation Motivation Mortgage modifications have been the primary tool used to mitigate foreclosure Over 5 million mortgages had been modified as of Q (OCC 2015; U.S. Treasury 2015) 68K MHA modifications and over 100K proprietary modifications started in Q Typical modification terms have changed over time Passage of HAMP resulted in more standardized mods, particularly a reduction in P&I to 31% of income Still considerable variation in the exact terms of modifications Schmeiser (Federal Reserve Board) Mortgage Modifications August 11, / 40

15 Research Questions Research Questions What modification terms are most effective at reducing the probability of a subsequent default or foreclosure? Change in interest rate, change in P&I, capitalization of fees and interest, forbearance of principal, reduction of principal Further examine how loan, borrower, and servicer characteristics, and economic conditions affect subsequent performance Contribution: Unique dataset (Corelogic) with detailed information on loans, modification terms, current valuations and amount of junior liens; national focus; examine pre- and post-hamp mods Limitations: Have few borrower demographics or characteristics (such as income and employment) that could affect both modification terms and outcomes Schmeiser (Federal Reserve Board) Mortgage Modifications August 11, / 40

16 Background What is a Mortgage Modification? A mortgage modification is any change in the terms of a mortgage Modifications relatively rare prior to recent crisis However, magnitude of housing crisis necessitated widespread use of modifications to help borrowers in distress Goal of modification should be to improve affordability of mortgage payment Number of different options to reduce payments: Reduce interest rate Extend amortization period Use principal forbearance Reduce mortgage principal Schmeiser (Federal Reserve Board) Mortgage Modifications August 11, / 40

17 Background Early Mortgage Modifications At the start of the housing crisis, modifications were at the discretion of the servicer and terms varied substantially Modifications made in 2008 rarely lowered monthly payments and often resulted in increased payments Approximately half of all subprime and alt-a mods increased payments (White 2009) Among all 2008 modifications done by national banks, 32 percent resulted in increased payments (OCC 2009) Many modifications failed to improve affordability resulting in 12 month redefault rate over 60% Schmeiser (Federal Reserve Board) Mortgage Modifications August 11, / 40

18 Background Post-HAMP Modifications In March 2009 the HAMP program was introduced HAMP required servicers to adjust monthly P&I on first lien to 31 percent of total monthly income Follow waterfall to meet target: reduce interest rate, extend term, forbear part of principal interest free Servicers can continue to offer proprietary modifications to borrowers who don t qualify for HAMP Terms of proprietary modifications began to improve substantially Schmeiser (Federal Reserve Board) Mortgage Modifications August 11, / 40

19 Background Trend in Number of Mortgage Modifications Schmeiser (Federal Reserve Board) Mortgage Modifications August 11, / 40

20 Data Data Use loan-level data from FirstAmerican CoreLogic s Loan Performance Asset Backed Securities (ABS) Data on privately securitized mortgages ABS data include information on subprime and alt-a loans, but does not include information on agency backed securities or loans held in portfolio Data used here are only representative of privately securitized subprime and alt-a loans, not the entire US mortgage market Contain monthly performance history for about 20 million individual loans Schmeiser (Federal Reserve Board) Mortgage Modifications August 11, / 40

21 Data Data CoreLogic data contain detailed static and dynamic information on the loans and their performance Static data include information from origination such as date of origination, the state where the property is located, the borrower s FICO score, origination balance, interest rate, payment and interest amount, and servicer Dynamic data are updated monthly and include information on the current interest rate, mortgage balance, payment and interest, and loan performance Schmeiser (Federal Reserve Board) Mortgage Modifications August 11, / 40

22 Data Data CoreLogic also provides two supplemental files that are used in our analysis: The first contains detailed information on whether a borrower received a loan modification, as well as some parameters of the modification (e.g. reduction in principal, principal forbearance, change in P&I) Other modification terms are inferred from dynamic loan data (e.g. reduction in interest rate, change in amortization term) The second file is the CoreLogic TrueLTV Data, which matches the loans in the CoreLogic Loan Performance data to public records to obtain information on the presence and amount of any other liens on the property Also contain a monthly estimate of the property s value from their automated valuation model Schmeiser (Federal Reserve Board) Mortgage Modifications August 11, / 40

23 Data Data Given size of database select a five percent random sample of first-lien loans Our data on modifications and loan performance cover the period January 2008 through December 2013 Restrict our data to loans originated no earlier than January 2000 and modifications occurring after January 2008 To provide economic context for the loan performance we merge in state level unemployment rates obtained from the Bureau of Labor Statistics Yields sample of approximately 37,000 modified loans, 1.35 million loan-month observations Schmeiser (Federal Reserve Board) Mortgage Modifications August 11, / 40

24 Data Number of Modifications in Sample Schmeiser (Federal Reserve Board) Mortgage Modifications August 11, / 40

25 Data Share of Modifications with Payment Increase/Decrease 2008 to 2013 Schmeiser (Federal Reserve Board) Mortgage Modifications August 11, / 40

26 Data Share of Modifications with Balance Increase/Decrease 2008 to 2013 Schmeiser (Federal Reserve Board) Mortgage Modifications August 11, / 40

27 Data Modification Performance by Year Schmeiser (Federal Reserve Board) Mortgage Modifications August 11, / 40

28 Data Outcome for Modified Loans in Sample Schmeiser (Federal Reserve Board) Mortgage Modifications August 11, / 40

29 Empirical Methods Methods Use two different estimation strategies: Probit for 12 month redefault and Proportional Hazard Model estimated using a multinomial logit The probit model takes the form: Pr(Y izs = 1) = f (α + βx i + γmod i + δcltv i + πhpi z + θstate s + ɛ izs ) Y is an indicator for whether or not the loan becomes 60 plus days delinquent within 12 months of modification X is a vector of loan characteristics from origination (loan used for purchase, FICO score, owner-occupied, low or no documentation, and origination year) Mod is a vector of loan characteristics at the time of modification (loan servicer, payment status, junior lien, and modification year) Mod also includes key mortgage modification parameters (forbearance % principal, % change in P&I, interest rate, and principal balance, % capitalized fees and interest, HAMP mod) Schmeiser (Federal Reserve Board) Mortgage Modifications August 11, / 40

30 Methods Empirical Methods In each period, a loan can be in one of numerous mutually exclusive states: current, delinquent (30, 60, 90+ days), lis pendens, REO, foreclosure sale, prepaid, short-sale, and modification So estimate discrete time proportional hazard model using maximum-likelihood to account for these various potential states Similar specification to probit, but include time varying characteristics current combined loan to value ratio (CLTV) and state unemployment rate Also include number of months since modification Schmeiser (Federal Reserve Board) Mortgage Modifications August 11, / 40

31 Results Results: Probit 60+ Days Delinquent within 12 months (1) Junior lien (5.4123) Loan used for purchase (7.7463) FICO at Origination 580 to ( ) FICO at Origination 650 to ( ) FICO at Origination 720 and Above ( ) Not owner occupied (2.9566) 30 to 60 Days Delinquent at Mod ( ) 90 Days Delinquent at Mod ( ) In Foreclosure at Mod ( ) Year on Year Change in HPI ( ) Unemployment Rate (8.6610) Base Probability of 60 Day Delinquency: 38.9% t statistics in parentheses. p < 0.10, p < 0.05, p < 0.01 Schmeiser (Federal Reserve Board) Mortgage Modifications August 11, / 40

32 Results Results: Probit 60+ Days Delinquent within 12 months (cont) Principal Forbearance as Pct of Balance ( ) Percent Reduction in Principal ( ) Percent Capitalized Fees and Int (9.0034) Percent Reduction in Interest Rate ( ) Percent Reduction in P&I ( ) Percent Increase in P&I (3.4191) LTV 100 to 124 Percent (3.3038) LTV 125 to 149 Percent (4.8195) LTV 150 Percent and Above (5.6526) Base Probability of 60 Day Delinquency: 38.9% Observations t statistics in parentheses. p < 0.10, p < 0.05, p < 0.01 Schmeiser (Federal Reserve Board) Mortgage Modifications August 11, / 40

33 Results: Multinomial Logit Results Schmeiser (Federal Reserve Board) Mortgage Modifications August 11, / 40

34 Results: Multinomial Logit Results Schmeiser (Federal Reserve Board) Mortgage Modifications August 11, / 40

35 Results: Multinomial Logit Results Schmeiser (Federal Reserve Board) Mortgage Modifications August 11, / 40

36 Results: Multinomial Logit Results Schmeiser (Federal Reserve Board) Mortgage Modifications August 11, / 40

37 Results: Multinomial Logit Results Schmeiser (Federal Reserve Board) Mortgage Modifications August 11, / 40

38 Results: Multinomial Logit Results Schmeiser (Federal Reserve Board) Mortgage Modifications August 11, / 40

39 Conclusions Conclusions Any modification that improves the affordability of the loan reduces redefault and foreclosure Forbearance appears to reduce delinquency and foreclosure, but magnitude is one-third that of a principal reduction Same magnitude effect on REO The capitalization of fees and interest significantly increases redefault and foreclosure risk CLTV has very large effect on subsequent redefault, foreclosure filing, and termination in foreclosure FICO score at origination still highly predictive of loan performance even after a subsequent modification Delinquency status at time of modification also highly predictive of subsequent performance Schmeiser (Federal Reserve Board) Mortgage Modifications August 11, / 40

40 Conclusions Policy Implications Principal reductions may be quite effective as they affect redefault and foreclosure through multiple channels Direct effect through principal reduction coefficient Effect on P&I Reduces CLTV However, this does not necessarily mean they are most cost effective modifications from the investor s point of view Further analysis needed to determine relative cost of each type of modification Schmeiser (Federal Reserve Board) Mortgage Modifications August 11, / 40

41 Contact Information Contact Information Maximilian D. Schmeiser Senior Economist Federal Reserve Board Phone: Schmeiser (Federal Reserve Board) Mortgage Modifications August 11, / 40

42 The Effect of Negative Equity on Mortgage Default Evidence from HAMP s Principal Reduction Alternative Therese Scharlemann (U.S. Office of Financial Research) Stephen H. Shore (Georgia State University) Views expressed in this presentation are those of the speaker(s) and not necessarily of the Office of Financial Research.

43 Motivation HAMP Identification Results Conclusion Appendix Disclaimer: Views and opinions expressed are those of the authors and do not necessarily represent official positions or policy of the OFR or Treasury.

44 Motivation HAMP Identification Results Conclusion Appendix Overview Research Question: By how much does reducing the amount of negative equity by reducing the principal balance reduce mortgage default? Data: Administrative data on HAMP PRA, a governmentsponsored program for delinquent borrowers. Method: Exploit quasi-experimental variation in amount of principal forgiveness granted under HAMP PRA program rules

45 Motivation HAMP Identification Results Conclusion Appendix Preview of Results Default program exit after 90+ days delinquency Principal forgiveness averages 28% of principal balance Observed quarterly default hazard: 3.1% Counterfactual hazard absent PF: 3.8% (CI: %)

46 Motivation HAMP Identification Results Conclusion Appendix Why might negative equity matter for default? Neg equity paying a price that is higher than realized value Vulnerable to economic shock; cannot sell home to satisfy mortgage balance. Holding current costs and benefits of defaulting constant (e.g., monthly mortgage payment, home value, home price expectations, etc). Principal forgiveness increases future expected benefits of not defaulting today.

47 Motivation HAMP Identification Results Conclusion Appendix Home Affordable Modification Program (HAMP) Government program to help delinquent borrowers Voluntary for servicers/lenders with government subsidy Delinquent borrowers are sent a letter about HAMP Qualified borrowers can apply for a HAMP modification Standard HAMP modifies mortgage terms until the mortgage is affordable to the borrower modification: reduce payments by reducing interest rate, extending loan terms, and/or forbearing principal affordability target: 31% debt-to-income (DTI) ratio temporary: rate subsidies phase out beginning in 5 years NPV test to ensure that modification benefits lender Borrower completes 3 month trial period at their new payment to receive "permanent modification".

48 Motivation HAMP Identification Results Conclusion Appendix HAMP PRA HAMP Principal Reduction Alternative (PRA) modification also includes principal forgiveness (PF) same affordability target as Standard HAMP LTV target: 115% LTV target (for 85% of loans) or a 100% LTV target (for 15% of loans) servicers with 100% LTV target limit PF to 30% initial UPB Applicants may be offered Standard HAMP or HAMP PRA Test to determine NPV to lenders. Servicers typically offer modification (or no modification) with highest NPV. 3 month trial period at new payment Program officially in place October Enrollment

49 Motivation HAMP Identification Results Conclusion Appendix Standard HAMP and HAMP PRA Waterfalls Waterfall continues until borrower hits affordability target of DTI=31% Standard HAMP HAMP PRA no reduction reduce mortgage balance in mortgage balance to hit LTV target typically 115% LTV reduce interest rate to 2% extend loan term to 40 years forbear principal as needed equivalent to zero rate balloon

50 Motivation HAMP Identification Results Conclusion Appendix Data overview Administrative records for all new permanent HAMP modifications enrolled 1/2011-6/ ,513 loans, 244,132 loan-quarters First cohort observable for 10 quarters. Outcome variable of interest: 90-day delinquency Borrowers disqualified from program if 90 days delinquent HAMP database does not track final resolution Measure default as quarterly hazard: new defaults as share of surviving loans in each quarter. Loans in sample have completed 3-month trial and become permanent Cannot reliably see mortgage characteristics of cancelled trials. About 10 percent of trial modifications are cancelled before they become permanent - about 2/3rds due to nonpayment.

51 Motivation HAMP Identification Results Conclusion Appendix Identification Regression Kink Design (RKD) Initial LTV Principal Forgiveness (PF) Final LTV 1 = max{0, min{amount needed to hit DTI target, amount needed to hit LTV target}} Initial Mortgage DTI PF needed to hit DTI target (PF DTI ): Target Mortgage DTI 1 Initial LTV PF needed to hit LTV target (PF LTV ): Target LTV 1 Servicers with LTV target 100% cap PF at 30% initial UPB. Can control flexibly for LTV and DTI terms separately LTV = loan-to-value (measures negative equity) DTI = debt-to-income (measures ability to pay) Total DTI = Mortgage DTI + Fixed DTI (taxes and insurance) Target Mortgage DTI = 31% - Initial Fixed DTI

52 Motivation HAMP Identification Results Conclusion Appendix Final LTV (red) Identification (Moderate Initial DTI) PR Amount 45º LTV Target DTI limits PR Initial LTV Source: Authors analysis

53 Motivation HAMP Identification Results Conclusion Appendix Final LTV (red) Identification (High Initial DTI) PR Amount 45º LTV Target DTI limits PR Initial LTV Source:Authors analysis

54 Motivation HAMP Identification Results Conclusion Appendix Final LTV (red) Identification (PRA-cap Kink) PR Amount 45º LTV Target Cap limits PR DTI Would Have Limited PR Initial LTV Source: Authors analysis

55 Motivation HAMP Identification Results Conclusion Appendix Identification: empirical model Pr(Default) = f (α + β PF ln (1 + PF) + β LTV ln(1 + PF LTV ) + β DTI ln(1 + PF DTI ) + β X X + ε) We are comparing the relationship between LTV and default on either side of the kink. RK treatment effect is calculated as the change in the slope at the kink (β PF ) divided by the change in the first derivative of the treatment at the kink (in this case, 1).

56 Motivation HAMP Identification Results Conclusion Appendix Identification: Selection PF allocation: We predict PF very well. R 2 in first-stage regression is 0.99, with precise coefficient of Kink design: Difficult to argue that amount of PF is correlated with unobservable propensity to default (conditional on LTV, DTI, and servicer). Selection: Do servicers select borrowers into PRA with unobservably higher (or lower) propensity to default when those borrowers would get higher PF amounts? Servicer selection on unobservables is difficult for services because they must follow rules based on observables. Borrower selection at application on unobservables unlikely given kink design (borrowers don t know on which side they will land when they apply) Borrower selection possible between receiving trial modification or offer and before permanent modification

57 Motivation HAMP Identification Results Conclusion Appendix Table 1: Impact of Principal Forgiveness on Hazard (Logit) Dep. Var. Quarterly exit rate ln(1+pf) -0.85*** -0.54*** -0.61*** -0.65*** -0.87*** (0.13) (0.13) (0.14) (0.16) (0.25 ) ln(1 + PF LTV ) 0.38*** 0.23** *** (0.09) (0.09) (0.11) (0.13) (0.67 ) ln(1 + PF DTI ) -1.07*** 0.42*** 0.31*** 0.26** 1.77** (0.05) (0.05) (0.08) (0.10) (0.73) Dummy for LTV target = 115 percent YES YES YES YES YES Quarters of observation controls YES YES YES YES YES Quarter cohort and servicer dummies NO YES YES YES YES DTI before modification NO YES YES YES YES Interaction variable NO NO YES YES YES Other controls NO NO NO YES YES 10-ppt LTV and DTI bins NO NO NO NO YES Observations 244, , , , ,712 Loans 45,513 45,513 45,513 39,310 39,172 R (All specifications include an LTV target control and quarters-of-observation dummies. "Other Controls": FICO, ARM dummy, investor-owned mortgage dummy, log income, log pre-mod balance, length of trial (linear and squared), log NPV of HAMP mod over no mod, log NPV of HAMP PRA mod over no mod, dummy for whether the standard HAMP mod had a higher NPV than the PRA mod.) Source: Authors analysis.

58 Motivation HAMP Identification Results Conclusion Appendix Context for Results Logit coefficient for ln(pf) to predict the quarterly hazard is to The average quarterly hazard is is 3.1% for the HAMP PRA loans in this sample. A 10% PF at mean implies 0.2 to 0.3 percentage points drop in the quarterly hazard rate (e.g., from 3.8% to 3.5% or 3.6%). The average PF of 30% implies a drop in the quarterly hazard rate from 3.8% to 3.1%

59 Motivation HAMP Identification Results Conclusion Appendix Counterfactual: Cumulative default with and without PF Cumulative Default Rate by Quarter of Modification Average hazard with PR: 3.1% Average hazard without PR: 3.8% Confidence inverval: [3.5%,4.1%] :Q12011:Q22011:Q32011:Q42012:Q12012:Q22012:Q32012:Q42013:Q12013:Q2 Quarter of Modification Proportion of modifications that have exited through 2003:Q1 (green line) Counterfactual proportion of modifications exiting through 2013:Q1 (blue line) Green Observed Rate Blue Counterfactual Rate Absent Principal Forgiveness Source: Authors analysis

60 Motivation HAMP Identification Results Conclusion Appendix Figure: Estimates for 115 LTV Target Sample Principal forgiveness increases with pre modification LTV. Post modification LTV unchanged as LTV increases. Principal forgiveness unchanged as premodification LTV increases. Post modification LTV increases with pre modification LTV ln(pf LTV ) ln(pf DTI ) Source: Authors analysis

61 Motivation HAMP Identification Results Conclusion Appendix Table 2: Impact of Principal Forgiveness, kink analysis (logit) Dep. Var. Quarterly exit rate Sample Full Uncapped Kink 0.5 Kink 0.25 ln(1 + PF) *** *** ** (0.129) (0.145) (0.225) (0.519) ln(1 + PF LTV ) PR 0.380*** 0.308*** (0.089) (0.105) (0.138) (0.307) ln(1 + PF DTI ) *** *** *** *** (0.031) (0.039) (0.043) (0.056) Observations 244, , ,851 83,913 Loan count 45,513 41,339 27,641 16,200 R Source: Authors analysis

62 Motivation HAMP Identification Results Conclusion Appendix Table 3: Placebo tests Dep. Var. Quarterly exit rate Sample Full Full Full Placebo Placebo ln(1+pf) *** *** *** (0.129) (0.133) (0.264) (0.128) (0.225) ln(1 + PF LTV ) 0.380*** 0.328*** *** (0.089) (0.094) (0.675) (0.085) (0.539) ln(1 + PF DTI ) *** *** 1.828** *** (0.045) (0.052) (0.737) (0.044) (0.609) ln(placebo PF) 0.282* (0.144) (0.252) Dummy for LTV target = 115 YES YES YES YES YES Quarters of observation controls? YES YES YES YES YES Full set of controls? NO NO YES NO YES Observations 244, , , , ,298 Loan count 45,513 45,513 39,172 52,862 48,378 R ("Full controls": cohort and servicer dummies, DTI before mod, DTI and LTV bins, FICO, ARM dummy, investor-owned mortgage dummy, log income, log pre-mod mortgage balance, trial mod length (linear and squared), log NPV of HAMP mod over no mod, log NPV of HAMP PRA mod over no mod, and a dummy indicating a standard HAMP mod had a higher NPV than a HAMP PRA mod.) Source: Authors analysis

63 Motivation HAMP Identification Results Conclusion Appendix Figure: Within Estimates for 115 LTV Target Sample - Placebo Principal forgiveness increases with pre modification LTV; Post modification LTV unchanged as LTV increases Principal forgiveness unchanged as premodification LTV increases; Post modification LTV increases with pre modification LTV ln(pf LTV ) ln(pf DTI ) Source: Authors analysis

64 Motivation HAMP Identification Results Conclusion Appendix Conclusion: Comparison with Existing Benchmarks Logit coefficient for ln(1 + PF) to quarterly hazard is to The average quarterly hazard is 3.1% for HAMP PRA. 30% PF reduces average hazard from 3.8% to 3.1%. 30% PF reduces cumulative default rate from 49% to 39% in earliest cohort. Impact of PF may not remain constant over time. Modifications with rate reduction get less generous after 5 years, so a spike in default for lower-pf (higher rate reduction) modifications is possible. Improvements in the housing market may make principal forgiveness less important.

65 Motivation HAMP Identification Results Conclusion Appendix Conclusion: A simplified discussion of costs vs. benefits of PF Cost of PF: 30% (1 lifetime re-default rate) Benefit of PF: (lifetime change in re-default rate) (difference between modified & recovery values) Cost-effectiveness of PF will increase/decrease if: default-reducing benefits of PF grow/narrow over time and/or Counterfactual default rate increases/decreases. First cohort: change in default rate implies $877k in writedowns per avoided foreclosure. Ten loans written down for every avoided foreclosure. Government subsidy of $262k per avoided foreclosure.

66 Motivation HAMP Identification Results Conclusion Appendix Evidence of bunching: Loans whose servicers do not apply cap Distribution of PRA Recipients Not Subject to Cap by Distance from Kink Density Distance from Kink

67 Motivation HAMP Identification Results Conclusion Appendix Evidence of bunching: Loans whose servicers apply cap Distribution of PRA Recipients Subject to Cap by Distance Between Uncapped and Capped PRA Amounts Density Distance Between Uncapped and Capped PRA

68 Motivation HAMP Identification Results Conclusion Appendix Context for Results return Cost of PF: (PF amount) (1 permanent default rate) Benefit of PF: (permanent change in default rate) (severity of loss in default, e.g. 50% of mortgage) 30% PF reduces average hazard from 3.8% to 3.1%. 30% PF reduces cumulative default rate from 49% to 39% in earliest cohort. To be cost effective absent subsidies or externalities default-reducing benefits of PF would have to grow over time relative to the baseline default rate and/or baseline lifetime default rate would have to be very high absent PF

69 Motivation HAMP Identification Results Conclusion Appendix Figure 2: HAMP Enrollment Cumulative enrollment of non GSE loans from Jan 2011 Dec 2012 Thousands HAMP with Principal Reduction Standard HAMP Jan Mar May Jul Sep Nov Jan Mar May Jul Sep Nov return

70 Motivation HAMP Identification Results Conclusion Appendix Table 1: Summary Statistics About Loans and Borrowers N ( 000s) Med. Mean SD Min Max Balance pre-mod ( 000s) 45.5 $292 $324 $173 $18 $1,279 Home value ( 000s) 45.5 $180 $205 $118 $10 $856 Gross monthly income ( 000s) 45.5 $4.5 $4.9 $2.39 $0.6 $22.2 Total mortgage payment ( 000s) 45.5 $2.0 $2.2 $1.05 $0.3 $9.5 Principal & interest payments ( 000s) 45.5 $1.6 $1.8 $0.90 $0.2 $9.0 FICO

71 Motivation HAMP Identification Results Conclusion Appendix Table 2: Summary Statistics About Modifications N ( 000s) Med. Mean SD Min Max Total DTI pre-mod % 47.0% 12.3% 24.0% 100.0% Total DTI post-mod % 31.0% 0.4% 12.4% 33.0% Total payment reduction % 30.0% 16.0% 0.0% 69.0% LTV before modification % 164.2% 33.5% 109.8% 240.0% LTV after modification % 121.2% 16.1% 100.0% 237.7% Principal balance reduction % 28.9% 17.4% 0.0% 73.5% Rate pre-mod % 6.3% 2.0% 0.8% 15.1% Rate post-mod % 3.9% 2.1% 1.0% 15.0% Term pre-mod (months) Term post-mod (months)

72 Motivation HAMP Identification Results Conclusion Appendix Figure: Treatment for 115 LTV Target Sample

73 Motivation HAMP Identification Results Conclusion Appendix Table 4: Impact of PF, Between and Within Estimates Dep. Var. Quarterly default rate Sample Full LTV Target of Sample Sample 115% 100% Identification Full Between Within Within ln(1+pf) *** *** *** (0.129) (0.224) (0.163) ln(1 + PF uncapped ) ** (0.387) ln(uncapped PF) ln(capped PF) (0.481) ln(1 + PF LTV ) 0.380*** *** (0.089) (0.124) (0.183) ln(1 + PF DTI ) *** *** *** (0.045) (0.659) (0.058) (0.115) ln(ltv initial) (0.670) LTV bin x PRA from DTI bin No Yes No No Observations 244, , ,146 43,876 Loan count 45,513 45,513 38,513 7,000 R

74 Motivation HAMP Identification Results Conclusion Appendix Table 7:Estimates in subsamples (logit) Dep. Var. Quarterly exit rate Sample Full ARM FRM PLS Portfolio ln(1+pf) *** *** *** *** *** (0.129) (0.210) (0.167) (0.165) (0.218) ln(1 + PF LTV ) 0.380*** 0.504*** 0.352*** 0.518*** (0.089) (0.145) (0.115) (0.108) (0.161) ln(1 + PF DTI ) *** *** *** *** *** (0.045) (0.071) (0.063) (0.056) (0.079) Dummy for LTV target = 115 YES YES YES YES YES Quarters of observation controls? YES YES YES YES YES Observations 244, , , ,271 97,861 Loan count 45,513 18,867 26,646 29,412 16,101 R

75 Motivation HAMP Identification Results Conclusion Appendix Table 6: Variation in estimates (logit) Dep. Var. Quarterly exit rate ln(1+pf) *** *** *** *** (0.129) (0.325) (0.187) (0.203) (0.284) ln(1+pf) x ln(initial TDTI) (0.349) ln(1+pf) x quarters since mod (0.039) (0.041) quarter of mod *** *** (0.013) (0.014) ln(1+pf) x quarter of mod (0.045) (0.047) Dummy for LTV target = 115 YES YES YES YES YES Quarters of observation controls? YES YES YES YES YES Observations 244, , , , ,132 Loan count 45,513 45,513 45,513 45,513 45,513 R NB: ln(predicted PF) ln(pre-mod LTV) ln(predicted Post-Mod LTV)

76 Motivation HAMP Identification Results Conclusion Appendix Counterfactual: hazard by modification duration, with and without PF Hazard by Quarters Since Modification (Quarterly rate of exit from program) Quarters since Modification Observed hazard (green) Counterfactual hazard absent principal reduction (blue)

77 Motivation HAMP Identification Results Conclusion Appendix Counterfactual: hazard by calendar quarter, with and without PF 0.07 Hazard by Calendar Quarter (Quarterly rate of exit from program) :Q2 2011:Q3 2011:Q4 2012:Q1 2012:Q2 2012:Q3 2012:Q4 2013:Q1 2013:Q2 2013:Q3 Calendar Quarter Observed hazard (green line) Counterfactual hazard absent principal reduction (blue line)

78 Mortgage Modifications using Principal Reduction: How Effective Are They? Ben Keys Harris School of Public Policy University of Chicago August 11, 2015

79 What we talk about when we talk about mortgage modification Goals: Help struggling homeowners continue to maintain consumption Effect of losing $5 trillion (!) in home equity Avoid foreclosure by curing delinquent borrowers Prevent neighborhood-level spillovers from neglect/abandonment Campbell, Giglio, and Pathak (2011) Why do homeowners default? Income disruptions or shocks to liquidity Bad underwriting or contract-related shocks Strategic incentives to walk away when sufficiently underwater Bhutta, Dokko, and Shan (2010) suggest relevant only when LTV>150%

80 Key Research Questions How do we help homeowners stop the transition from default to foreclosure? These papers unpack the mechanisms of modification effectiveness Are modifications cost-effective? Huge costs to the government to facilitate modifications Depends on which features are used (e.g. principal reduction) Depends on how we estimate the benefits Are principal reductions the solution if temporary income disruptions are the problem?

81 Schmeiser and Gross (2015) Data from CoreLogic on subprime and Alt-A loans Modifications between 2008 and 2013 (HAMP and non-hamp) Many mods increase principal balance by rolling in fees/interest Almost 20% of mods increase payments (!) through post-due capitalization Hazard model approach to estimate impact of modification features on re-default Baseline characteristics at time of modification matter a lot Delinquency status, FICO and especially CLTV Forbearance effect is 1/3 that of principal reduction Principal reduction helps through multiple channels Reduces CLTV (strategic incentive) Reduces monthly payment at same time

82 Challenges for Studying Modification Size and elements of modification are endogenous to premodification characteristics of borrower and housing market e.g. why is initial CLTV so predictive of subsequent default? Strategic default, or Homeowners who reached high CLTV did so through low down payments, bad contracts (e.g. NegAm), or aggressive extraction combined with negative regional shocks Servicers may be strategically selecting those loans most likely to succeed Results are conditional on choice to modify Servicers can design their own program or opt into HAMP program

83 Scharlemann and Shore (2015) Use data from HAMP PRA program and kinks in program design used to determine amount of principal forgiveness Non-GSE loans enrolled in : $3.8b in total forgiveness Average loan receives 28% reduction in principal Kinks in design affect different borrowers in terms of the amount of principal reduction Based on distance from kink in principal reduction schedule Avoids endogeneity problems Requires assumptions about similar borrowers on either side of the kink, and relatively farther away as well to estimate slopes Useful placebo test where default increases monotonically in pre-mod LTV Cost estimate: $877,000 per foreclosure avoided (!) Cost to government: $262,000 per foreclosure avoided

84 Placebo test figure nails effect of PRA:

85 Consensus View: Larger Payment Reduction, Fewer Defaults This is true almost by construction Need a better understanding of why people default in order to target modifications and modification features If people are liquidity constrained, reducing principal with no change in payment should not matter on the margin Only affects wealth, not liquidity Holy grail of data: 1)Debts, 2)Assets, 3)Income, 4)Consumption If we knew more about income shocks, we could better design temporary assistance programs If we knew more about consumption, we can see whether programs had their intended effect outside of the mortgage contract alone

86 Efficient Credit Policies in a Housing Debt Crisis Eberly and Krishnamurthy (BPEA 2014) model: Government s priority should be helping cash-strapped homeowners 1) Maintain spending in a weak economy 2) Avoid foreclosures Some role for principal reduction to reduce strategic default motives But costly and spreads benefits over much longer period of time Much larger role for temporary efforts to lower monthly payments: Defer/forbear mortgage payments Reduce interest rates Extend mortgage terms Propose redesigning mortgages with automatic adjustments Allow for disproportionately lower payments when borrowing constraints bind

87 Concluding Thoughts Two fantastic examples of careful empirical work to answer hugely important policy question Building consensus around view that principal forgiveness received limited bang for its buck Alternatives Large-scale forbearance and refinancing programs Automatic refinancing contracts Letting some homeowners fail in a low-cost way (own-to-rent?) Final note that mortgage programs missed some of the neediest households e.g. NPV-negative because unemployed, failed trial mod, failing to takeup HARP

88 More notes on Schmeiser and Gross Unemployment rate coefficient is negative in paper (positive in slides) MNL results on unemployment and HPI flip sign depending out mortgage outcome, this is puzzling How to think about static vs. time-varying characteristics Most importantly CLTV, HPI, and unemployment rate Why are HAMP modifications more effective even after conditioning on terms? Has to be something about selection, right? Or omitted variables? Can be more up-front about endogeneity concerns

89 More notes on Scharlemann and Shore Introduction undersells the result Doesn t fully explain how important the finding is or put it in a broader context of the literature, and varies across too much detail and too little Identification could use a clearer simpler example Terms of forbearance: How long does it last? 5 years? Why does PRA have the reverse waterfall relative to HAMP? How does 2MP fit in? Median FICO at time of modification is only 560: Should we expect any of these people to cure? Why are the signs different on PF from LTV and PF from DTI? Shouldn t a dollar in PF be a dollar regardless of where it comes from? This is still unclear Is counterfactual exercise of no principal forgiveness but same payment reduction way out of sample? Is this all just substituting forgiveness for forbearance?