The R2 is between 0.5 and 0.7. The number of observations is quiteequally distributed among the credit grades. The number of observationsis much smaller than in the logistic regression, because here wecan use only completed listings. In the last column of Table 6, the finalrate estimates have been calculated with the following listingparameters: ‘starting rate’ 18%, ‘amount’ $5000, ‘debt-to-incomeratio’ 40% and zero ‘current delinquencies’. As we can see, the finalrate estimate quickly increases as the credit grade becomes worse. Asa whole the regression model predicts the final rate reasonably well.Looking at the associated residual plots, it appears there is unequal(increasing) variance associated with some of the variables implyingheteroskedasticity, in particular when outside the range of commonvalues. However because we are not calculating prediction intervals inthe BDA (point estimates are still unbiased), this appears less serious.