A comprehensive model for managing credit risk on home mortgage portfolios

The coefficients in Tables I through 3 and plots of the probability curves (including but not limited to those in Figure 2) reveal a greater tendency for transitions to take place into delinquency and default on loans with high loan-to-value ratios, in areas with high rates of unemployment, on larger loans, and on adjustable-rate mortgages. Even after accounting for these differences in characteristics, default rates were greater in Southern California, and for loans that were originated by another financial institution. Prepayments were more likely to occur on fixed-rate loans with high interest differentials, low loan-to-value ratios, and on properties located in areas with low unemployment rates. Ceteris paribus, high interest differentials motivate prepayment to save on interest costs; low loan-to-value ratios and low unemployment rates would increase the likelihood that a borrower would qualify for refinancing.

Models for transitions from the delinquent states are more parsimonious partly because a small fraction of the accounts is available for calibrating them. Also, once delinquency has occurred, the loan has shifted to a separate risk class in which some of the other predictive information becomes redundant. From the statistically significant coefficients in Table 2, one would infer that loans delinquent 30 to 89 days are more likely to default than to become current if the present loan-to-value ratio is high, the original loan amount is large, the home is located in Southern California, and the interest differential is high. From Table 3, loans delinquent 90+ days are simply indicated to be more likely to default if the estimated loan-to-value ratio is high. Each of these effects is consistent with theoretical expectations and with management's understanding of the behavior of the portfolio.

PROBABILITY OF INCURRING A LOSS ON A DEFAULTED LOAN

As previously stated, no losses were incurred on a substantial portion of first mortgages that terminated in default. Although data regarding second liens from other institutions were not available, the existence of second mortgages and a sufficient equity cushion on those properties were judged to account for this phenomenon. Second mortgages were popular in California in the 1980s as property values increased rapidly and high interest rates inhibited refinancing. Theoretically, the likelihood of incurring a loss on a defaulted loan is determined primarily by the expected amount of equity in the property (i.e., the difference between the expected market value and the current balance on the loan). On loans in their early years, changes in equity occur primarily through fluctuations in housing prices because most of the monthly payment is attributed to interest. As a loan approaches maturity, the principal balance drops rapidly with each payment and, in that phase of the loan's life, amortization has the primary impact on the loan-to-value ratio. We experimented with a variety of structural forms to predict the probability of incurring a loss in the event of default (IIt), and finally settled on a logistic function that used the current loan-to-value ratio to represent the equity position. The local unemployment rate was included to provide an indicator of the health of the local economy and potential demand for housing. Some loans were insured against a decline in property values under a shared-risk policy that protected the institution only against modest declines in the value of the collateral. Separate models were therefore developed for insured and uninsured loans.