The recent downturn in the housing market has exposed certain weaknesses in mortgage credit risk models' ability to make accurate credit loss forecasts. Considering the significant volume of securities backed by subprime mortgage loans, and their diffusion throughout the global economy, these events have had a dramatic impact on global credit markets and on the financial institutions holding these assets.
At an institutional level, given the importance of mortgage credit and prepayment models to risk management and financial reporting processes, it is no surprise that the performance of these models is under closer scrutiny by management, auditors, and regulators. Additionally, the risk profiles of companies' financial estimates are growing as management addresses model weaknesses by implementing changes to these models that may or may not be done in a well-controlled manner.
Many mortgage credit and prepayment models were developed based on loan performance data that reflected periods of low interest rates, high growth in house prices, and relatively permissive underwriting standards. Current predictions of default and severity rates from such models may be significantly understated — while estimates of prepayment speeds may be overstated.
One way to assess the reasonability of these estimates in the current environment is by benchmarking the models' key outputs (prepayment rates, default rates, and loss severity rates) to the company's most recent experience, as well as to available third-party benchmarks for similar collateral segments. The company should ensure that this analysis is performed at a sufficiently granular level — that is, at least monthly and with a greater focus on vintages and important risk segments — to identify and respond in a timely manner to material trends in model forecast errors. Where appropriate, the company may wish to modify its model-based estimates in response to these benchmarks with well-documented and well-supported adjustments.
The deterioration in liquidity and prices in secondary markets for nonconforming mortgage-backed assets has led a number of companies to reclassify segments of their loan portfolio from "held for sale" to "held for investment" — increasing the population of loans on which loan-loss reserves are estimated.
In some cases, models being used as part of the loan-loss-reserve process may not have been validated for the products or characteristics of these held-for-sale loans. Companies should ensure that they can support the reasonable predictive performance of its credit and prepayment models for these new loans through methods such as benchmarking and back-testing.
Given the significant recent changes to the U.S. housing market, other key credit loss assumptions should be identified and revalidated; for example, preforeclosure expenses, average foreclosure time lines, loss-mitigation strategy mix, and counterparty credit risk associated with expected recoveries from credit enhancement or repurchase proceeds.
In some cases the breakdown of internal models may lead companies to license third-party credit or prepayment models for use in estimating their loan-loss reserve or mortgage-related valuations. Nevertheless, management still has to ensure the reasonability of these model outputs. For example, since nearly all vendor models contain settings for users to calibrate or tune the model's default, prepayment, and loss severity estimates, it is crucial that companies be able to support — through back-testing or benchmarking — the reasonability of its model calibration and, ultimately, the reasonability of the model's predictions for its specific portfolio. Companies generally should not be using vendor models solely with the vendor's default setup without a reasonable basis for doing so.
For a number of companies, on-top adjustments to model-based estimates tend to fall outside the scope of their model validation programs. Since these adjustments are typically quantified by modeling personnel and sometimes employ complex data processing and estimation methodologies, management should either scope the independent review of these adjustments into its model validation program, or employ appropriate control processes to ensure the reasonability and accuracy of these computations.
Management's responses to current events may result in an atypically high number of model changes. By opening up the models for these changes, companies create the potential for implementation errors and unauthorized changes to the model — highlighting the importance of effective pre-implementation testing and associated model change management controls.
