Building a Credit Model Using
GSE Loan-Level Data

Abstract

This article outlines an approach for developing a loan-level model to predict the probability and timing of credit events that trigger investor losses in Fannie Mae and Freddie Mac’s recent credit risk transfer (CRT) transactions. The authors begin with background on why the government-sponsored enterprises (GSEs) released loan-level data and how this disclosure is different from their previous data disclosures. They compare the credit event, as defined in the GSEs’ CRT transactions, to other measures of credit performance and discuss the relevant population of data records to include in a development dataset. They show how a simple model based only on a limited set of loan characteristics can help explain variation in credit performance, and finally, they show how the inclusion of variables that capture the macroeconomic and credit-underwriting environment can improve the model fit over time and across different origination. The authors believe that this type of model can be used for evaluating the recent and future CRT transactions issued by the GSEs.