Mortgage default: Lattice model simulations and empirical analysis.

Abstract

This dissertation investigates the determinants of residential mortgage default by evaluating a large data base of loans issued from 1975-83 and followed through 1990. It has three primary objectives: (i) to better exploit insights from option pricing models in an empirical analysis, (ii) to assess some explanatory variables that have not been tested previously, and (iii) to help resolve some of the conflicting findings of determinants of mortgage default. The first objective is achieved by building a two stochastic state (house prices and interest rates) lattice model of default that is used to simulate default probabilities. This modeling extends beyond the work presented elsewhere by assessing the conditional probability of default on seasoned mortgages when holding current loan to value (Cltv) of the mortgage constant. It shows that changes in the mortgage age, house price volatility, and the house dividend rate (measured as the rent to price ratio) have fairly modest impacts on next period's default rate. Changes in interest rates are shown to the have a stronger impact, but the Cltv itself is the strongest determinant of default rates. The second objective tests the importance of a past decline in interest rates, changes in the rent to price ratio, differences in population age, and moving rates. All are found to be statistically significant default covariates, but the latter three are not very economically important. The third objective is achieved using a finer data grid than previous studies to increase the cross sectional variation in the data. Previous work has used data from a single locality, or has used regional or national data. Neither provides much cross sectional variation. This study measures its explanatory variables based on the 64 Metropolitan Statistical Areas over which the loans were originated. It confirms previous findings on the importance of Cltv and changes interest rates. After controlling for these, it finds a statistically significant but small economic effect for divorce and unemployment. As a further check the results are aggregated to the regional level. Using regional level data, covariates such as divorce and unemployment are not robust. The regional level analysis indicates that interest rate changes are more important than was shown in the MSA level analysis suggesting that regionally based studies underestimate the importance of Cltv.