A mortgage lender currently uses one of several processes to measure so-called "credit risk" for first-mortgage products. Washington Mutual, using an independent group of researchers, completed a study of best practices to help the practitioner and the regulator evaluate the results of alternative models. This article is an executive summary of that study.

A number of models are currently in use to estimate unexpected losses for mortgages. A study sponsored by Washington Mutual compares four basic classes of models and benchmarks their results against those of best-practice institutions.

The classes include:

1. The "panel data" models.

2. The first-generation Oliver-Wyman (OW) model.

3. The Merton models.

4. The "roll rate" (RR), or "state-transition'' models.

The purpose of the study is to help both practitioners and regulators evaluate the results of these models through two separate empirical experiments. The first experiment involved establishing estimates of the loss distributions for important cohorts of the market portfolio, using two large databases of prime and subprime loans. Shortcomings were observed in the panel data and OW models, and the most complete was found to be the State-Transition models. The second experiment uses a more detailed analysis on a randomly drawn sample to establish parameters for the model adopted by regulatory agencies, using the economic capital results from a best-practice model.

Principles of Credit Risk Measurement and Economic Capital

Risk practitioners use a number of common terms. An understanding of these terms helps set the stage for any discussion of the various credit risk models used for first mortgage products.