Backloaded Mortgages and House Price Appreciation

Transcription

1 1 / 33 Backloaded Mortgages and House Price Appreciation Gadi Barlevy Jonas D. M. Fisher Chicago Fed Wisconsin-Fed HULM Conference April 9-10, 2010

2 2 / 33 Introduction: Motivation Widespread house price boom followed by bust Bust coincident with massive financial upheaval, severe recession, and unprecedented policy interventions Objective indicators of bubble-like behavior might help avert a recurrence Don t look for direct indicators: hard to identify if price expectations excessive relative to fundamentals Study house prices and mortgage terms in 389 US MSAs in : mortgages look like good indirect indicator of speculation amidst a bubble

3 Introduction: Theory 3 / 33 Theory suggests patterns which accompany bubbles Build on Barlevy (2009) Optimal contract for lenders when some agents they fund engage in speculation Lenders encourage speculators to unload their assets quickly by making loans with backloaded payments Adapt theory to case of housing: Interest only mortgage preferred by borrowers and lenders amidst a bubble Tradeoff: Borrower avoids building equity in risky asset, lender gets borrower to sell asset quickly Absent a bubble, borrowers and lenders prefer traditional mortgage

4 Introduction: Key Empirical Findings 4 / Larger IO shares where are larger, more rapid appreciations 2. IOs more robustly correlated with appreciations than other mortgage characteristics Not a subprime phenomenon. Subprimes may still explain Concurrent rise in home ownership, e.g. Chambers, Garriga, and Schlagenhauf (2009) Defaults during the subsequent downturn, e.g. Corbae and Quintin (2009), Mian and Sufi (2009) Not a securitization phenomenon Securitization could still be important for subprimes, e.g. Mian and Sufi (2009) Not a leverage phenomenon as in Lamont and Stein (1999) Low leverage associated with high appreciation Aware of bubble lenders make borrowers pay downpayments

5 5 / 33 Introduction: Key Empirical Findings, continued 3. Where IOs popular, their use is not a response to houses becoming more expensive: IO share renders affordability variables insignificant Use anticipates future appreciation Appreciation either does not predict IO share or predicts lower use of IOs 4. IOs and price appreciation may be connected through inelastic housing supply: inelasticity may make MSA susceptible to bubbles 5. IOs have shorter durations

6 Introduction: Prices vs. IOs in Two Cities 6 / 33 Phoenix, AZ FHFA Price IO share FHFA Price IO share 2000q3 2002q3 2004q3 2006q3 2008q3 Quarter Laredo, TX 2000q3 2002q3 2004q3 2006q3 2008q3 Quarter FHFA Price IO share FHFA Price IO share FHFA data from Haver Analytics, Mortgage data from McDash

7 Introduction: Prices vs. Subprimes in Two Cities 7 / 33 Phoenix, AZ FHFA Price Subprime Share FHFA Price Subprime Share 2000q3 2002q3 2004q3 2006q3 2008q3 Quarter Laredo, TX 2000q3 2002q3 2004q3 2006q3 2008q3 Quarter FHFA Price Subprime Share FHFA Price Subprime Share FHFA data from Haver Analytics, Mortgage data from McDash

8 Introduction: Peak IO Share vs. Peak Price 8 / 33 Max FHFA Price Max IO Share

9 Introduction: Peak IO Share vs. Fastest Price Growth 9 / 33 Max 4 Quarter Price Change Max IO Share

10 10 / 33 Introduction: Outline 1 Theory 2 Data 3 Cross section 4 Panel 5 Duration 6 Conclusion

11 11 / 33 Theory: Motivation Why relate credit market to house price appreciation? Barlevy (2009) argues bubbles encourages certain types of credit arrangements We consider implications for mortgage contracts

12 Theory: Sketch of Environment 12 / 33 Mortgage: sequence of payments {m t } T t=1 Traditional mortgage: m t = m Interest-only mortgage: { m m t = L < m for t = 1,..,τ m H > m for t = τ + 1,.., T Borrower may not be able to afford m H Given expectations of p t borrower can: 1. Pay off remaining debt and sell asset, payoff is p t L t 2. Make payment, obtain continuation payoff: βv t+1 m t 3. Default on loan, payoff is 0 Depending on supply of asset, different paths for p t If supply is constrained, can get a bubble: { (1 + g) pt 1 w/prob q p t = w/prob 1 q F t If supply is unconstrained, p t = F t

13 Theory: Predictions 13 / 33 In a bubble: Typically there is some interest-only mortgage both borrower and lender prefer to m t = m If borrower waits to sell, gains p t, but lender risks default Gains from trade: pay borrower to sell asset earlier IO gets borrower to sell by τ + 1, avoiding building equity in risky asset Absent a bubble: Borrower never sells early; holds asset, defaults if F t turns out to be low Hence, no gain to IO: since it can force costly default, it can make both borrower and lender worse off

14 14 / 33 Theory: Empirical Implications If rapid price increase in hot markets is a bubble, should also observe many backloaded contracts If increase due to fundamentals, no reason to expect preference for backloaded contracts Check on theory: backloaded contracts should also be paid off more quickly Setting aside question of whether these are bubbles, do mortgage market features predict house price growth?

15 15 / 33 The Data LPS Applied Analytics (McDash) Loan level data from 9 out of the top 10 servicers covering 60% of all mortgages by value in Undercounts sub-primes Coverage varies over sample. From 2004H2: new servicers with relatively high share of IO added to sample Backfill to avoid composition bias If IO shorter duration, backfill understates true share of IOs earlier in the data due to survivorship bias Focus on first lien for purchase

16 McDash Coverage and IO Shares, / 33

17 McDash Coverage and IO Shares, / 33

18 Descriptive Statistics of McDash Data 18 / 33 Mortgage Characteristics Private Type Year All Owner Long Term Sec. Subprime P.P. Pen IO % 90.9% 62.9% 80.0% 0.5% 0.3% % 85.8% 99.4% 65.6% 9.6% 16.7% Fixed % 93.4% 85.8% 16.1% 1.7% 5.4% % 90.4% 82.7% 19.7% 3.6% 4.0% O ARM % 97.0% 98.5% 20.6% 1.6% 32.9% % 85.6% 99.9% 55.1% 17.4% 69.0% ARM % 94.2% 93.7% 33.4% 11.7% 17.9% % 88.7% 99.7% 53.8% 32.9% 30.0%

19 19 / 33 Cross-Section Analysis What variables account for rapid price appreciation in MSAs from 2000 to 2005? OLS regressions with LHS variable Maximum 4 Quarter Change Very similar results for Peak Price Fundamentals include Level and growth rates: population, unemployment, per capita income, property taxes Levels: employment/population, housing units per person Later work will include WRI; so far it does not seem to affect results

20 Cross-Section: Explanatory Power of IOs / Max IO Share (0.052) (0.052) (0.048) Fundamentals N Y Y Y State Fixed Effects N N N Y R^ N Remaining cross-section regressions include fundamentals

21 Cross-Section: Explanatory Power of IOs / Max IO Share (0.087) (0.088) (0.104) House Prices in (0.136) (0.136) Price/Income in (6.359) (5.912) Undevelopable Land (Saiz QJE 2010) (0.054) (0.042) R^ N

22 Cross-Section: Explanatory Power of IOs max share of hybrids max share of 30+ years max share of subprime max share private securitized average nonprime CLTV max share of jumbo none Max IO Share (0.048) (0.057) (0.054) (0.049) (0.059) (0.045) (0.065) Other Variable (0.084) (0.227) (0.080) (0.042) (0.070) (0.053) R^ N Other Variable (0.082) (0.214) (0.093) (0.039) (0.082) (0.040) R^ N State Fixed Effects included CLTV is from subprime dominated LP dataset 22 / 33

23 Panel Analysis Theory predicts IOs favored when agents anticipate price appreciation (above that predicted by fundamentals) But, IOs could be favored because prices are high Which is it? Which effect is driving the data? Of independent interest: can we use IOs to predict potential future price appreciation Use Granger-causality tests to address these questions If IOs chosen because agents anticipate price appreciation, then increases in IOs should Granger-cause increases in prices If IOs chosen because prices have increased to make houses unaffordable using a traditional mortgage, then increases in prices should Granger-cause IOs Should we treat all cities the same, or separate out high IO use cities? 23 / 33

24 Panel: Corr(dioshare t, dprice t ) vs Max IO Share 24 / 33 Max Interest Only Share of Mortgages Correlation of Change in IO share with Change in House Prices

25 Panel: IOs Lead Prices (All Cities) 25 / Correlation(dioshare t, dprice t+j ) Correlation J 2000:1 2008:4 2000:1 2005:4

26 Panel: IOs Lead Prices (High IO Shares) 26 / Corr(dioshare t, dprice t+j ), IO Share > Correlation :1 2008:4 2000:1 2005: J

27 Panel: IOs Forecast Prices / 33 Test if coef's on dhpi same across msa's Test if coef's on dios same across msa's Test if dhpi Granger causes dios Test if dios Granger causes dhpi * coef on dhpi_{t 1} varies by MSA Left Hand Side Variable dhpi dios dhpi dios dhpi dhpi_{t 1} * (0.072) (0.043) dios_{t 1} (0.025) (0.034) (0.029) F stat p value F stat p value F stat 0.07 p value F stat p value

28 Panel: IOs Forecast Prices / 33 Left Hand Side Variable dhpi dios dhpi dios dios dhpi_{t 1} (0.060) (0.050) (0.050) dhpi_{t 2} (0.029) (0.057) (0.069) dios_{t 1} * (0.040) (0.055) dios_{t 2} * (0.032) (0.057) Test if coef's on dhpi F stat same across msa's p value Test if coef's on dios F stat same across msa's p value Test if dhpi Granger causes dios Test if dios Granger causes dhpi F stat p value F stat p value * coef on dios_{t 1} and dios_{t 2} varies by MSA

29 29 / 33 Mortgage Durations Theory relies on backloading encouraging earlier sales Examine Kaplan-Meier survival rates (one minus probability of legitimate exit conditional on surviving) to see if this is in data Could be that households who intend to move choose IO mortgages so this cannot prove theory Focus on data after 2005 since this is where have most data on IOs and there is less backfilling. Data for 2004 look similar. Less similar for data in 2003 where the sample is small.

30 Durations: Originations in 2004:1 30 / 33 Kaplan Meier Survival Rates Loans Originated in 2004Q Quarters After Origination All Non IO Loans Interest Only Loans

31 Durations: Originations in 2005:1 31 / 33 Kaplan Meier Survival Rates Loans Originated in 2005Q Quarters After Origination All Non IO Loans Interest Only Loans

32 Durations: Originations in 2006:1 32 / 33 Kaplan Meier Survival Rates Loans Originated in 2006Q Quarters After Origination All Non IO Loans Interest Only Loans

33 33 / 33 Conclusions IO usage is a robust predictor of high price appreciation Subprimes, teasers, securitization do not seem important for price appreciation IO usage appears to anticipate future price appreciation where usage is high; based on theory this suggests prices expected to appreciate beyond the level justified by fundamentals Verified that IOs have shorter duration Does not prove there were bubbles, but begs the question of why borrowers and lenders chose backloaded financing disproportionately in places with bubble-like price dynamics