Automated underwriting and mortgage pricing models are based on estimates of credit default and prepayment models that assume loan terms, such as loan-to-value ratio (LTV, and borrower characteristics, embodied in a credit score, are exogenous.  That is the expectation of future default probability cannot cause LTV or credit score.  In the extreme, it is obvious that borrowers who intend to default will choose as high an LTV as possible and will try to maximize their credit score in order to secure such a loan.

In contrast, models of rational borrowers predict that loan terms and borrower characteristics used in the application process will be the object of strategic manipulation by the borrower.  There is already evidence that choice of LTV is endogenous.  This paper adds empirical evidence that credit score is also manipulated strategically by applicants by showing that the change in credit score, after the mortgage is endorsed, is directly associated with the fall in APR realized when the credit score is raised.  Thus differences in incentive to raise credit score artificially explain the subsequent decline in credit score after endorsement.

It is easy to show that gullible lenders relying on credit risk models that assume LTV, credit scores, and other elements of their models are exogenous to the probability of default can easily be duped into underestimating credit risk on the mortgages that they endorse, sell, and purchase.  This provides and explanation for the high proportion of early payment defaults (EPDs often never make a single payment) and on the fact that default models substantially underpredict defaults on mortgages for 2007 and 2008 even when  the actual changes in economic conditions, including house prices, are fed into the models.