Rural Loan Credit Modeling
Client: United States Department of Agriculture (USDA) Rural Development (RD)
Developing Credit Models to Estimate Performance of USDA Rural Development Loan Guarantee Programs
Challenge: USDA RD needed better cost estimates for its $130 billion+ portfolio of guaranteed single-family mortgages (SFGH) and guaranteed business loans (B&I), and for these estimates to pass financial statement audit.
FI Solution: For each program, FI researched regulations, assessed available data, aggregated third-party data, and then began model development. We built a baseline model which used only historical performance to predict future performance. Upon completion of the baseline model, FI began hypothesis testing the factors that could influence loan performance. We developed econometric models to predict loan prepayments, delinquencies, defaults, and the loss severity on defaulted loans. FI tested several combinations of variables, evaluated their statistical validity, and analyzed how the projections compared to actuals, the baseline model, and previously tested specifications. Based on this iterative process, we selected the winning specification. We developed user-configurable cash flow engines to reflect each program’s unique structure.
We supported the models through RD’s and OMB’s approval process and deployed the models into production. After using the models to produce financial statement estimates, we supported the results through the financial statement auditor’s review.
FI Impact: The lifetime credit loss models that FI developed give USDA RD detailed insight into the risk and expected performance of its SFGH and B&I programs. The models are more responsive to changes in macroeconomic factors and portfolio characteristics, as compared to older model versions. The models give the agency information needed to set policy, understand risk, and accurately report budgets and financial statements.