A comprehensive model for managing credit risk on home mortgage portfolios

The model is used in simulation mode to examine the effects of key factors on the expected behavior of a loan. As mentioned earlier, to project the performance of a hypothetical new loan over its life, a single loan is created with the desired characteristics. Values are set for the loan amount, appraised value of the property, term to maturity, interest rate, mortgage product (fixed, ARM, or hybrid), etc. The comprehensive model with its attendant economic assumptions is applied to the hypothetical loan and the summary reports of performance are produced. In Table 7, we illustrate the results of a series of simulations intended to show the effects of initial loan-to-value, mortgage insurance, mortgage product, and loan size upon forecasted performance. The first three cases show the increasing risk associated with higher loan-to-value ratios. The next case shows the beneficial effect of the shared-risk mortgage insurance. The next two cases suggest higher risks associated with borrowers who selected ARM versus hybrid versus fixed-rate mortgages. The last three cases show increasing risks with size of loan. Figure 4 illustrates a graph that is produced to show managers the net effect of projected account transitions and loss severity on annual rates of loss (dollars as a percent of outstanding principal) on a new loan over its life. In this instance, Figure 4 compares the annual risk of loss projected by the comprehensive model for the three mortgage-product comparisons in the middle of Table 7. The hybrid mortgage performs quite similarly to a fixed-rate mortgage (albeit with slightly greater risk) during the interval when its interest rate is fixed. Then it is converted to an ARM and projected to perform accordingly. The shapes, of course, depend on the economic assumptions--especially on housing price changes. Higher appreciation in housing prices would cause the curves to shift downward and the loss rates to decline more rapidly; conversely, sustained depreciation in housing prices would shift the curves upward and cause the loss rates to be sustained at higher levels until the loan nears maturity. The curves as presented assume modest appreciation after the first year and are in consonance with experience in the industry, which shows credit risk increasing for about 5 years after origination, and then diminishing toward maturity.

With information of this type, the institution reviews its pricing practices. The risk premium charged according to size of loan (e.g., "points" assessed on the origination of "jumbo" loans) is examined in relationship to differences in estimates of lifetime credit losses. Consideration is given to points charged or requirements for mortgage insurance on loans that originate with high loan-to-value ratios. The results of such cost-based approaches to pricing of products are, of course, weighed against the need to offer favorable terms to compete for business.

FURTHER RESEARCH AND EXTENSIONS

Within the Markovian structure, there is considerable flexibility for refinement and enhancement of the model. When the outlook for interest rates became less certain in early 1994, we began to calibrate fixed and ARM mortgages separately. For fixed-rate mortgages, we now use interest rate differentials to emphasize the effects of refinancing incentives on prepayment rates. For ARMs, we use absolute interest rates to emphasize the effects of payment shock on default rates as interest rates change. Since the model was originally calibrated on loans issued in a period of declining interest rates and was later applied in a period of possibly rising interest rates, this refinement produced better behavior for projections of ARMs under future scenarios. The severity model has also been restructured to pattern the foreclosure process, adding a stage for transfer of ownership of the property to the institution (to a status called "other real estate owned" or ORE). The final forecasts are not affected materially by this change in structure of the severity models, but the information for managerial review is enhanced. The effect is to add logistic models for each dichotomous branch in the decision hierarchy.