Mortgage Default and Bankruptcy: Theory and Empirical Evidence Wenli Li, FRB Philadelphia Michelle J. White, UCSD and NBER What we do: • Examine the interaction of homeowners’ decisions to default on their mortgages and file for bankruptcy. • We test: – Whether homeowners are more likely to default versus file for bankruptcy when they gain financially from either, and – Whether homeowners are more likely to default versus file for bankruptcy when they are liquidity-constrained. • We use new data that combines information on mortgage debt and other types of debt. – Previously, the literatures on mortgage default and bankruptcy were separate because of lack of combined data. How are mortgage default and bankruptcy decisions related? • Bankruptcy helps homeowners avoid mortgage default/keep their homes by discharging unsecured debt. • Bankruptcy helps homeowners keep their homes by delaying foreclosure and allows homeowners to repay mortgage arrears over five years. • But bankruptcy helps homeowners with high income or high assets less, since they must use repay from future income and assets. • Default helps homeowners preserve access to credit card loans—some choose default/avoid bankruptcy. • Bankruptcy helps homeowners give up their homes by discharging deficiency judgments. Homeowners’ predicted mortgage default and bankruptcy decisions Notes: • Diagram is separately calculated for each homeowner. • As shown it assumes that bankruptcy reform is in effect (means test), mortgage debt is fixed, and unsecured debt is fixed and high. • Homeowners are predicted to default and file for bankruptcy only when it is in their financial interest. – D/B predicted when house value is low and income is low. (House value is low enough that the cost of renting < cost of owning.) – D/NB predicted when V is low and Y is high. – ND/B predicted when V is higher and Y is high. (Here the income boundary between B and NB shifts to the right because of homeowners’ gains from filing for bankruptcy.) – ND/NB applies when V and Y are both high and when V is very high and Y is low. (Best not to default because must repay unsec debt from sale proceeds of the house.) Same, but some homeowners default due to liquidity constraints 0 100 200 300 400 500 Figure 3. Predicted Probability to Default and File for Bankruptcy 0 50 income (thousands) D/B D/NB ND/B ND/NB See the text for the value of the other variables we set for the hypothetic household 100 Notes: • Now some additional homeowners default even when it is against their financial interest b/c V is high. They default because of liquidity constraints. Data: We merge three datasets: – LPS: large sample of mortgages with information from the mortgage application, plus monthly updates on payment and bankruptcy. – Equifax: sample of individuals with information about all types of debt, plus quarterly updates on payment, credit scores, debt-to-income ratio. – HMDA: use it to merge LPS and Equifax based on date/location/principal of mortgage. Final dataset: • All mortgages originated 2004-2006. • They are followed quarterly until the mortgage is paid off or transferred, the homeowner defaults or files for bankruptcy, or at the end of 2009. • Currently, we include only prime, fixed rate mortgages. • Each quarter, we also have: – Amount owed and payment record for second mortgages, credit card debts, student loans, auto loans, and installment loans. (Half of each debt if homeowner married.) – Updated credit score and debt-to-income ratio. – Income at origination and homeowner’s age, sex, marital status. Specification: • We estimate a multi-probit model explaining: – – – – Default/no bankruptcy (aD/NB). No default/bankruptcy (aND/B). Relative to no default/no bankruptcy (aND/NB). We drop simultaneous default/bankruptcy because it’s very rare (aD/B). • Main variables of interest are the predicted decision variables D/NB, ND/B, D/B. • Control variables, quarter and state dummies. • Errors clustered by mortgage. Predicted signs: D/NB actual ND/B actual D/NB predicted Own effect (+) Cross effect (-) ND/B predicted Cross effect (-) Own effect (+) Sequence effect (+) Sequence effect (-) D/B predicted Summary statistics (quarterly) D/NB Predicted .034 Actual .0049 ND/B .0015 .0010 D/B .026 .0003 ND/NB .939 .994 Results w/o liquidity constraint: (% change when prediction changes) D/NB actual ND/B actual D/NB predicted ND/B predicted D/B predicted 38%*** (own effect) -38%*** (cross effect) 50%*** (sequence eff.) -60%*** (cross effect) 42%* (own effect) -36%** (sequence eff.) Add liquidity constraint: • Rerun the model with an additional dummy variable for homeowners who are liquidityconstrained—combined debt payments are more than 50% of income. • % of observations that are liquidityconstrained? • Everything else remains the same. Results with liquidity constraint: D/NB actual ND/B actual D/NB predicted ND/B predicted D/B predicted 47%*** (own effect) -54%*** (cross effect) 4%*** (sequence eff.) -52%*** (cross effect) 99%*** (own effect) -57%*** (sequence eff.) Conclusions: • Homeowners’ mortgage default and bankruptcy decisions respond strongly to financial benefit. are related. • The two decisions are related—homeowners are more likely to file for bankruptcy. • Liquidity constraints make homeowners more likely to do both. Future work: • Examine subprime mortgages and adjustable rate mortgages. • Compare results when default and bankruptcy decisions are independent.