The Mid-Year Mortgage Portfolio Valuation and AVM Cascades
As financial institutions – including credit unions that offer mortgage lending – gear up for a mid-year analysis of their mortgage loan portfolios and the value of collateral assets, there are some important tools that can help. Certainly, automated valuation models (AVMs) are often utilized for this type of analysis. However, a look at the Interagency Appraisal and Evaluation Guidelines issued in December 2010 by the Department of Treasury – as well as the Supervisory Guidance on Model Risk Management OCC 2011-12 – clearly reflect an expectation of AVM independence. What does this mean, and how can this independence be achieved?
As a quick background, many AVMs on the market today are what might be called “closed systems.” AVM providers have developed their own “secret sauce” modeling logic, database sources and cascade partners (an AVM cascade uses the models of multiple AVM providers that are triggered sequentially until one is able to achieve the level of valuation confidence that is established by the financial institution). Since one AVM may not have the level of coverage in a particular market or neighborhood that another one does, a cascade model makes sense.
But here’s the rub: if all AVM providers claim that they are the best – or that their particular cascade group offers the most comprehensive coverage – then how is a credit union to know which one to select for their portfolio analysis? And if a particular AVM provider has established relationships with its own group of AVM partners – and determined the parameters for the order in which they are utilized in a cascade – is that true independence?
Perhaps not – but there is a better answer. AVM-agnostic testing providers like AVMetrics, LLC (http://www.avmetrics.net ) are able to conduct blind AVM testing versus property purchase prices in order to provide a reliable view into the performance of an AVM solution in a particular market area. To perform the testing, an AVM is utilized to determine the value of a given property, but the model is not given access to information on the most recent sales price of that property. This allows the testing firm to compare how closely the AVM “predicts” the value of a property versus what the property recently sold for in a particular market.
This approach is vital in the development of a truly independent AVM cascade solution. The testing firm not only helps to preserve the integrity of the AVM cascade product through continuous testing, but it also establishes the underlying cascade structure that determines which AVM is used, and in what order. In the case of AVMetrics, the testing firm can also hide the cascade structure from the AVM cascade provider to ensure an unbiased result.
When it comes to mid-year portfolio analyses, an AVM cascade solution can be tremendously helpful in determining the value of the assets in a mortgage portfolio. In fact, the performance of any well- constructed cascade that is based on accuracy and precision metrics will outperform a single or even multi-model solution. However, with the rigorous cascade model selection/elimination process driven by an independent testing firm like AVMetrics, credit unions are in a much better position to ensure that the results are as accurate as they can be, and are independently generated without bias toward any individual model.