# The Many Uses of Automated Valuation Models (AVMs)

- Source: https://avm.digisavvy.dev/blog/many-uses-of-automated-valuation-models-avms/
- Author: Lee Kennedy
- Published: January 5, 2016
- Category: Whitepapers

---

By Lee Kennedy, Founder, AVMetrics LLC

## Introduction

In this paper we will focus on some of the many ways AVMs are being leveraged in mortgage banking.

First, a quick refresher.

An AVM is a predictive model designed to estimate a property’s value. The usefulness of an AVM can generally be determined by the extent to which it minimizes prediction error in a variety of contexts. Optimizing the selection of an AVM is best accomplished through analysis leading to the creation of a Model Preference Table (known as an MPT, and when combined with a set of business rules, as an AVM “Cascade”), which tells you the best model to use on a given property at a given point in time, and conversely, where AVMs should not be used.

## Applications for AVMs

Some of the most common internal uses for AVMs[1]:

· Portfolio Management and Valuation  
· Refinance Mortgage / Home Equity Origination Sales  
· Mortgage Marketing  
· Loss mitigation

## Portfolio Management and Valuation

Mortgage Servicing is a data-driven business, and in order to have accurate and up-to-date data on a loan portfolio, servicers must continuously update the value of their portfolio’s collateral. Not only does it make good business sense to do so, but it is also a regulatory requirement. This is a complicated task, fraught with regulatory, reputational and financial risks, and even though regulators provide guidance, there is no single “right” way to accomplish it. Even the largest industry players commonly seek external expertise to meet portfolio valuation challenges.

One objective of collateral valuation is to understand the risk-profile of the portfolio. Although many prefer to avoid it, certain mortgage servicers specialize in delinquent servicing, with its specialized products and services. Updating collateral valuation can help a servicer identify potentially delinquent loans, either to service them differently, such as accelerating collection activities, or to transfer servicing rights before delinquency issues arise.

The loans in a portfolio were originated according to policies, procedures, rates and terms that existed at the time of origination. At origination, there likely would have been analytics which used borrower credit strength, LTV, etc. to estimate a loan’s (and hence a portfolio’s) worth as well as its projected runoff. However, as time goes by, economies, markets and neighborhoods change, borrower circumstances also change, and those original estimates are replaced by actual results and new loan parameters, including collateral value. Periodically updating collateral valuation improves the estimates of the overall worth and payoff rates of loans in the portfolio.

For example, higher than expected appreciation in a property’s value could increase projected runoff rates. Servicers want to know how to appropriately value their assets and what retention strategies to use to prevent the runoff. In addition, updated collateral values enable the calculation of current LTV’s; these enable servicers to maintain their own longitudinal trending stats and conduct analyses on their own servicing footprint. This could enable a Servicer to adjust its retention strategies and improve its customers’ experiences. A well-thought-out strategy based on tailored analyses can be the foundation of a competitive advantage over other companies.

While there are a number of potential strategic advantages to be gained by continuously updating collateral valuation using AVMs, there is a more pressing reason for most servicers, and that is regulatory requirements. Interagency Appraisal and Evaluation Guidelines, adopted by all major U.S. regulators, require monitoring collateral risk by obtaining current collateral valuation information. Servicers should maintain sufficient information on current market values of collateral, and evaluations (AVMs) are sufficient in many cases. The need for updated collateral values will be greater in periods of increased credit risk, declining values, higher delinquency, etc., which may necessitate more frequent updates. Fortunately, bulk AVMs are not generally cost-prohibitive.

In all of these applications, minimizing AVM estimation errors is critical. The wider the dispersion of estimates, the more frequently the collateral value update will trigger an inappropriate and potentially much more costly response. For example, if an AVM inaccurately estimates a value that is actually 20% below current market value, it could initiate more expensive delinquent servicing, or even a sale of servicing rights.

The best way to improve the accuracy of estimates is to use a well-constructed AVM Model Preference Table (MPT), which ensures the selection of the best AVM for each specific property in the portfolio. Less error, variance and volatility results in fewer explanations to senior executives, boards of directors, and regulators.

## Refinance Mortgage and Equity Line Origination Sales

Perhaps one of the most widely-recognized best uses of AVMs is to service existing customers, either for the origination of refinance transactions or to cross-sell them on an Equity Line. For new customer acquisition, portfolio retention and product cross-selling, AVMs can play a key role in improving loan close rates by focusing officers on their best opportunities.

These days, most customers looking to refinance are already armed with a Zillow (or other comparable) estimate of their home’s value. In addition to pulling the Zillow estimate to learn what the customers likely have seen, pulling a validated AVM informs the loan officer as to the parameters she may have to work within, and helps to establish a credible first contact. A customer who believes in the competence of the loan officer, and who is satisfied with the loan product(s) being offered, is much less likely to continue shopping.

When working with borrowers already in the portfolio, the loan officer has the advantage of knowing a lot of detail about the original loan, its LTV, credit score, borrower’s income, loan performance, etc. After four to five years on the books, however (often the average life of a mortgage loan), most customer situations have changed and in particular, the Original Loan To Value (OLTV) no longer equates to the property’s true value. For both refinance and equity line prospects, pulling an AVM value estimate lessens the likelihood of portfolio run-off (due to terms that may no longer be as competitive) and enables the loan officer to prioritize the opportunities in the portfolio, whether refinances, equity lines or both. In addition, more time and energy can be devoted to those prospects that have a higher likelihood to close; this makes sense for everyone: the customer, the loan officer and the mortgage company.

However, not all AVMs are created equal! Many AVMs can produce estimates in some lending contexts that are within +/- 10%, or even +/-5% of the actual value. Those same AVMs, however, can be off by 20% or more in a different geography or price range. An accurate estimate is essential because even a 10% difference in value can completely change the loan program, the prospect’s expectations, and the likelihood of closing the loan. As valuable as the tool is, optimizing the expected accuracy of an AVM in an independently tested and proven MPT or cascade can be invaluable to the application of AVMs in mortgage and refinance sales; using a well-constructed cascade dramatically increases expected accuracy.

## Mortgage Marketing

As data-driven as mortgage servicing is, mortgage marketing may be more so. It is highly dependent on capturing data and using feedback to refine marketing methods. Good marketers know and continuously track even their most minute costs and conversion rates for every ad they run within every marketing channel, and they assiduously A-B test every strategy before each spend.

Critical to improving conversion rates is to qualify, qualify, qualify, and to screen out the least likely leads. For every one of the sub-segments the Marketers identify and manage, they typically break their leads into deciles of response and conversion likelihood. These are driven by as much data as they can find, including historical conversion rates, average loan amounts, geography, etc. One critical element used to qualify and rank portfolio leads is LTV.

Mortgage companies know the OLTV, and in fact, two or more mortgage companies often know each customer’s OLTV because servicing rights are so frequently sold. As a result, both the originator and the servicer have OLTV data. Several years down the road, however, current LTV data would be much more relevant to the Marketer’s conversion rate deciles; this is why marketing departments will typically use bulk AVM purchases to get an affordable and fairly accurate estimate of current property values to qualify and rank leads.

It is essential that valuation estimate error rates are minimized, or the entire lead ranking process will be compromised. Valuation errors come in two flavors: over-valuing and under-valuing, and both can be expensive. When property values are underestimated, leads may fall into lower deciles where they receive less attention. The deciles with the least likelihood of converting are commonly placed at the bottom of the lead list (ensuring that they are called less frequently), are assigned to less-experienced sales staff, or are skipped altogether. This can be viewed as a cost of lost opportunity…

On the other side, the deciles with the highest likelihood of converting go to the top of the sales queues and are worked by the highest converting sales staff. But when property values are overestimated, your production pipelines can be cluttered with leads that cannot be converted. As a result, accuracy is essential to making these marketing tactics successful. Using an independently tested and proven cascade is the best way to minimize those errors and maximize accuracy.

Competitors with more accurate valuations will reach the best prospects first. You need to be proactive in assessing your customers’ situations and reaching out to them. If you are not paying attention to the changing values of properties in your portfolio, you can be fairly assured that both your customers and your competitors are.

## Loss Mitigation

In the last decade Loss Mitigation has become much more sophisticated. There are more programs available, specialized departments are commonplace, and servicers have much more experience understanding which borrowers are likely to benefit and which are likely to continue down the delinquency waterfall. These specialized teams are capable of handling many different circumstances, but they need accurate data and prediction models to gain a clear understanding of each customer’s situation; this allows them to make the right response at each stage.

Servicers use current LTV along with FICO, payment history, loan program and other data to predict likely outcomes and to optimize their responses to different scenarios. If a property has equity, servicers need to take immediate action to get the loan back on track, starting with immediate collection activities, a possible refinance, forbearance, modification, etc. Properties without equity will follow a different path, probably with a team with different skill sets, since certain options are not feasible.

This is why it is essential to have a well-constructed MPT or cascade which leverages the available AVMs in order to provide the most accurate value of the collateral. An incorrect estimate of value could put the delinquent servicing onto the wrong track. Time spent pursuing options that are not viable increases the delinquent balance, making every option less likely and increasing the odds of a long and expensive worst-case scenario. By minimizing error, an independently optimized cascade increases the return on investing in customized models, teams and procedures.

## Regulatory Considerations

AVMs can serve many valuable functions in mortgage banking, but anytime an AVM is used, it carries some regulatory implications. Most of these implications are described in the Interagency Appraisal and Evaluation Guidelines, which apply to all domestic banks and credit unions. Because AVMs are considered “models used for material decision making,” they must be used and validated in some specific ways.

In short, the Guidelines require four areas of compliance: staffing, policies and procedures, validation and documentation.

Staffing covers the skills, stature in the organization, and competencies required to perform the duties of selection, validation and enforcement of controls over AVM use. Selection should be performed by staff knowledgeable in AVM evaluation. Validation should be performed by a staff that is independent of production or model selection, that has the skills to perform competent testing, and has the authority to halt usage or change procedures as necessary. Enforcement of usage controls has to be credible, not just perfunctory or self-administered.

Policies and procedures should cover not just the controls to ensure the proper use of AVMs, but also their selection and validation. They should cover the criteria that employees would use to determine the adequacy of the AVM, when an AVM should not be used due to certain circumstances, and when an AVM should be supplemented with a condition report or other tool.

Validation of AVMs must be performed by a party independent of those who built the AVM, regardless of whether either function is outsourced or done in-house. Proper validation requires understanding the modeling techniques used to build the model, evaluating the data used and how it’s updated, its sources, and its quality control. Validations should be conducted regularly, and particularly after any relevant changes in the models or the economic environment, and especially in the case of any expansion of the business footprint.

In addition to policies and procedures covering the use of models, another key aspect of the Regulatory Guidance is the requirement for extensive documentation covering all aspects of the process, from development, validation and testing to deployment. Although documentation must be specific to your organization’s use and implementation of AVMs, the technical elements of this requirement (i.e. development, validation and testing) are well-suited to outsourcing, because it is detailed work that has been completed many times, and its reinvention from scratch is unnecessary and costly. AVMetrics can help you not only with the internal structuring of documentation, but with this technical aspect as well.

The regulatory requirements covering the use of AVMs are described in OCC 2010-42 (Interagency Appraisal and Evaluation Guidelines) and OCC 2011-12 (Supervisory Guidance on Model Risk Management). Keeping up with the regulatory requirements must be done regardless of how extensive AVM use is within your organization, but having said this, the comprehensiveness of the policies, procedures, documentation and validation should be commensurate with that level of usage. An organization that uses AVMs sparingly for marketing purposes only may need annual validation and light controls, etc. On the other hand, an organization that uses AVMs heavily and in many areas, including for important business decisions, should plan for much more comprehensive procedures, controls and documentation, as well as more frequent and more comprehensive validation.

## Conclusion

AVMs have a number of very cost-effective applications in mortgage banking. First, collateral value is a critical input into many mortgage banking decisions, from marketing to sales and servicing. We did not cover underwriting or appraisal review, but those are also areas where AVMs have valuable applications. Second, AVMs can provide an estimate of the collateral value at a fraction of the cost of an appraisal, and this opens up their use for many applications. To be useful, however, AVMs must be fairly accurate; as long as that error is low enough, AVMs can add tremendous value.

Any individual AVM will have strengths and weaknesses – regions of the country, for example, where it lags competitors in terms of reliability and accuracy. An independently tested MPT and optimized cascade seeks to mitigate those weaknesses by always positioning the strongest AVM in every context and location. AVMetrics uses over half a million benchmarks each year to test every commercially available AVM and build customized cascades for every market, no matter how large or small. If it makes sense to use an AVM, it makes sense to use an independently tested MPT or optimized cascade to maximize the effectiveness of the AVMs and the business processes they support.

*Lee Kennedy, principal and founder of AVMetrics in 2005, has specialized in collateral valuation, AVM testing and related regulation for over three decades. Over the years, AVMetrics has guided companies through regulatory challenges, helped them meet their AVM validation requirements, and commented on pending regulations. Lee is an author, speaker and expert witness on the testing and use of AVMs. Lee’s conviction is that independent, rigorous validation is the healthiest way to ensure that models serve their business purposes.*

[1] Outside of Mortgage Banking there are also other applications for AVMs, although we will not cover these in detail here. Two examples of these include law firms and their consultants using AVMs to value tracts of real estate that are involved in large class action lawsuits, such as environmental disasters, and government agencies using them during tax reassessments.