Valuation Opinions Sometimes Less is More

WHITE PAPER October 2010 Valuation Opinions Sometimes Less is More by Susan Allen and Joni Pierce

Confidential The recipient of this document agrees that at all times and notwithstanding any other agreement or understanding, it will hold in strict confidence and not disclose the contents of this document to any third party and will use this document for no purpose other than evaluating or pursuing a business relationship with CoreLogic. No material herein may be reproduced, in whole or in part, by any means without the expressed written consent of CoreLogic. Unauthorized distribution is strictly prohibited. ii

Valuation Opinions Sometimes Less Is More By Susan Allen and Joni Pierce Collateral risk once considered to be less relevant than other forms of risk, such as credit, prepayment and fraud has been rediscovered by investors, lenders and servicers and is now firmly in the crosshairs of government regulators. In fact, the word appraisal appears more than 200 times in the Dodd-Frank Act. All of which has led to a more holistic view of collateral risk and the new emphasis on valuation quality, but is the pursuit of perfect valuations delaying good investment and loss mitigation decisions? The purpose of this article is to examine a few quality assurance alternatives and highlight recent innovations in valuations from two perspectives: a servicer trying to make sure that it is getting the right valuations on REO and an investor trying to improve a bidding strategy on a loan portfolio. In each case, we will see that new approaches and the use of analytics can significantly improve decision-making, often using less expensive or fewer valuation products. Before delving into the options, it may be helpful for readers who aren t in daily contact with valuations to discuss what all the fuss is about. Those of us who live and breathe property values hear a constant refrain: Why is it so difficult to produce an accurate property value? Properties are illiquid which means that the only time a true value is revealed is when a buyer and seller agree on a price 1. For any particular property, such a transaction usually occurs very infrequently. Every valuation outside of a purchase transaction is an estimate of what a reasonable buyer and seller should agree upon based on other property sales. And within each of those other property sales, the buyers and sellers and the conditions are unique. So properties truly exist in a range of values. Accepting Acceptable Ranges A few years ago, it was common for REO managers to order multiple independent values and have an internal reconciliation department review each pair. Often, a listing agent would also provide a Broker Price Opinion (BPO), which could be included in the analysis. If there was too much discrepancy among the values, another independent valuation could be ordered to break the tie. In this way, the multiple values served as an accuracy check too much discrepancy led to additional reviews and possibly another valuation product order. This practice is less ubiquitous now, although still deployed in many servicing shops. The primary disadvantages of this approach are obviously cost and time. Given the volumes in servicing shops, it is difficult to support the cost associated with buying multiple products when one should suffice. And each day a property is delayed from entering the market is another day of carrying costs. The less obvious disadvantage is accuracy. This may seem odd why would ordering multiple products lead to less accurate results? The answer lies in the subjective nature of valuations and the range of possible right answers. Academic studies on retail sales with no known condition bias have suggested that the defensible range of property values is at a minimum 10% - 15%. This means that two values on retail homes in good condition that are 10% - 15% apart are likely both defensible as representing a value that reasonable buyers and sellers would accept. Add in condition issues, and the defensible range of values is more realistically 15% 20%. This isn t an excuse for shoddy valuations based on poor information, but rather a rational expectation that competent professionals are likely to disagree within a defensible range. 1 Even a purchase is arguably not a sustainable value because a different buyer and/or seller may have arrived at a different value, even on the same date. A purchase transaction is, at best, one relevant valuation data point. 1

Recently, this phenomenon was demonstrated again when a number of mortgage risk professionals across the country were asked to participate in a challenge. They were divided into teams and asked to evaluate an REO property. Each team was given a package containing an appraisal, a BPO, an AVM and standard appraisal review reports. The actual sale price for the home was $235,000. Over 100 teams participated and came back with values ranging from $175,000 to $300,000. Granted it was a tough property to value, but the range of responses (and the number of teams reporting values in the very upper and lower tiers) surprised us. Coming Up With the Right Sales Price Now let s look at a real life example. A client recently asked us to help evaluate the efficacy of its valuation policies for establishing REO list prices. The policy was to order two BPOs (not necessarily from us) and reconcile any differences greater than 15%. Hypothetically, we assumed that what the first BPO returned was the primary result. Here are the initial findings: Total Number of Properties 800 Total BPOs Ordered 1,600 Properties where BPOs differed by 15% or more 200 Average adjustment from Primary BPO 13% The review department identified a full 25% of valuations as potentially being inaccurate by more than 15%, and they decided to adjust these values an average of 13% prior to listing the properties for sale. Of course, some adjustments were very large both up and down. At first blush, the process seemed to be catching a significant number of potential problems and allowing corrections before valuation errors influenced listing or sale decisions. Then we dug deeper. We used the eventual sale price (indexed back to the BPO date) as a benchmark of accuracy for both the BPOs and the reconciled value. Of the 200 properties reconciled, there were 62 where the direction of the reconciliation was inaccurate mostly based on the opinion of the second BPO ordered. In other words, the reconciliation process adjusted the property value up, but the ultimate sale price was lower than the primary BPO, or the other way around. In a further 41 cases, the adjusted value was in the right direction, but it overshot the ultimate sale price, making it less accurate than the primary BPO. For example, the primary BPO was $50,000, the reconciled value was $70,000 and the indexed sale price was $55,000. Of course, there were cases where the adjustment made a significant positive impact in the bank s handling of the property. Unfortunately, about half of the adjustments actually made matters worse occasionally by a large margin. And in some cases where the primary and secondary BPO matched, there were significant differences between the BPOs and the ultimate sale value, but the property was never subject to reconciliation. At that point it was agreed that improving the process would reduce a significant waste of time and money. What went wrong? We actually know this client to have very competent reviewers using solid review tools. The issue had more to do with the process and less to do with the capabilities of the staff. Ordering multiple, similar opinions of value just didn t lead the reviewers in the right direction enough of the time. So what s a servicer to do? Clearly, it is important to have a process in place to identify valuations that are potentially erroneous. Simply accepting all valuations provided isn t the answer. But it is important to have a process that efficiently and predictably reduces both Type 1 and Type 2 errors. Picking the Right Providers and Review Products There are several ways to reduce errors without ordering multiple valuation products. The key is to use a provider with strong review capabilities based on data and analytics. When we hire employees, we seek people whose skills can complement our own. The same is true for property valuations. The theories behind BPOs and appraisals are similar. Although specific professionals will select different comparable properties and make different adjustments, the methodology by which they perform the valuation is similar. Therefore, a good review process will look at the valuation from a different angle. 2

The creation of consistent BPO quality is a combination of having best in class panel management programs, combined with statistically sound data implementation frameworks, and quality-control-baseline transparency between BPO companies and their clients. Top BPO providers will consistently invest in quality along two key fronts including the effective use of data and analytics along with a robust and disciplined management program for provider panels. Automated Valuation Models (AVMs) are increasingly being used to review BPOs and other valuations. These products have come a long way, especially in the past few years. One of the consequences of a robust REO market is that some AVMs now have sufficient data to create robust distressed valuation tools. AVMs were originally created as a review tool for origination appraisals. Their main benefits for that purpose were cost, timeliness and predictable accuracy. The same is true now in the distressed AVM space. By testing distressed AVMs against known REO sales, users can deploy a review system with predictable outcomes. This is why quality BPO providers are now testing their results using more affordable options, like default AVMs. The Proof is in the Pudding When quality BPO providers incorporate analytics, the results are improved profitability for the servicer. Our client asked what could be done to reduce the wasted money stemming from the issues noted in the earlier study. We jointly created a specific BPO quality rating system so we could focus our analytics on both the Type I and Type II errors. Although the error rate on our BPOs was consistent with other providers when the study began, adding tailored analytics reduced the overall error rate by 75%. The client also leveraged the quality rating and analytics to inform policy decisions. When their servicing volumes increased, the client used the analytics and rating system to accept certain BPOs as is (no second BPO or review required.) The client absorbed twice the valuation volume and controlled risk without increasing review costs. This client s experience is a perfect example of how less can really be more in property valuation. Real value can be realized by selecting a BPO company that understands the use of relevant data. Without this, a chaotic assimilation of data can result in information overload and subsequent noise leading to further confusion. By using a provider who leverages analytics and a focused management program, servicers can have confidence that the first value is the right value. Using AVM Statistics to Sharpen Pool Bidding Now let s look at how new valuations techniques and analytics can help improve portfolio bidding and lower the risk on portfolio acquisition. Imagine you are a trader getting ready to bid on two seasoned, non-performing MBS loan pools with 1,000 loans in each pool. In each instance, the seller is providing BPOs on each property, and characterizing the BPOs as independent valuations. Even if the BPOs were provided by neutral and competent third parties, the seller s ordering and review policies can result in valuation bias. Having accurate, unbiased property values is critical to a successful bid, especially on a pool of non-performing loans. But how can you be sure? Although you may ultimately choose to order your own BPOs on all properties, it would expedite your decision process to first estimate the overall valuation risk before spending time and money on a pool you may ultimately reject. AVMs are an option, but by themselves they will only show you price discrepancies and won t answer the question of valuation bias. However, there is a promising new approach that relies on AVMs and AVM summary data to quickly and relatively inexpensively detect valuation bias. What the trader would do is request AVMs for each property from his or her favorite provider. Because these are nonperforming loans, the AVM should be a default AVM, which is calibrated to distressed properties. This is what the results looked like in our two pools: NPL Pool A NPL Pool B Default AVM Hit Rate 70% 62% Mean discrepancy between AVM and BPO -8% 3% 3

But what have you learned? Perhaps the trade would negotiate a better price for Pool A because the seller is underestimating property values (evidenced by the -8% mean discrepancy between the AVM and the BPO). Perhaps the values in Pool B are more likely to be reliable because the mean discrepancy is closer to zero. Or are the differences between the pools simply a reflection of the volatility one would expect when using automated valuations on distressed properties? At this stage, although the trader has more information, he s still bidding blindly. If, on the other hand, the trader can establish a baseline using a large enough sample that is known to be unbiased, then the advantage shifts. Creating a Baseline Start with a sample of known, unbiased property values across the country: for example distressed properties that this firm has had valued in an unbiased manner, or public record data on known REO purchase transactions. By running an AVM on that sample, he creates the AVM statistics as a baseline. For this discussion, we will assume that our baseline AVM sample results in the following summary statistics: Baseline Default AVM Hit Rate 71% Mean discrepancy between AVM and BPO -7% Mean absolute discrepancy between AVM and BPO 15% % of properties where AVM and BPO differ by 20% or more 17% Average collateral risk score 4.2 % of properties labeled high risk by collateral risk score 19% Remember, the purpose here isn t to assess the accuracy of either the AVM or the BPO. We already presume that the BPOs in the baseline are accurate and unbiased because we specifically selected a sample of unbiased transactions. Rather, we want to know what pool-level AVM statistics look like when distressed properties are valued in an unbiased manner. Now let s look at the pools we are evaluating compared against the baseline. NPL Pool A NPL Pool B Baseline Default AVM Hit Rate 70% 62% 71% Mean discrepancy between AVM and BPO -8% 3% -7% Mean absolute discrepancy between AVM and BPO 16% 20% 15% % of properties where AVM and BPO differ by 20% or more 20% 25% 17% Average collateral risk score 4.0 9.3 4.2 % of properties labeled high risk by collateral risk score 21% 40% 19% What a difference a baseline makes! Pool A looks similar to the baseline. So you can infer from the data that the valuation approach used by the seller of Pool A is likely to be unbiased. We can also see from the collateral risk score that the neighborhoods in which Pool A properties are located are no more or less risky than the baseline population. Without digging into seller A s property valuation methodology or investigating any individual property valuations, you have an independent view that indicates a balanced valuation approach and average collateral risk, given the neighborhoods in which the Pool A properties reside. Pool B is a different story. Although the AVM values are 3% higher than BPOs, we now know that the AVMs should be lower by an average of 7%. This means that the method of requesting and reviewing BPOs by seller B potentially overvalues the properties by 10%. Comparing against the baseline doesn t tell you why the Pool B valuation approach may lead to higher values, but a 10% variance is meaningful. The lower AVM hit rate in Pool B is also informative. AVMs require sufficient transaction volume in order to produce results. So a lower hit rate means that more properties are in areas with limited transactions. The higher collateral risk scores indicate that more properties in Pool B are in neighborhoods with higher property valuation risk (such as higher foreclosure rates) than in the baseline case. If you were still interested in Pool B, 4

additional digging into the seller s valuation methodology and the geographic concentration of the properties would be appropriate. This case demonstrated the application of a fairly simplistic baseline. The concept can be expanded to include more granular geographic data and the incorporation of additional data such as forecasts, shadow inventory, etc. Using these baselines can create competitive advantage by establishing proprietary views that can be used to immediately evaluate the quality of loan pools. Susan Allen is Vice President, Strategic Relationships and Joni Pierce is Senior Vice President, Valuation Services, for CoreLogic. 5

About CoreLogic CoreLogic is a leading provider of consumer, financial and property information, analytics and services to business and government. The company combines public, contributory and proprietary data to develop predictive decision analytics and provide business services that bring dynamic insight and transparency to the markets it serves. CoreLogic has built the largest and most comprehensive U.S. real estate, mortgage application, fraud, and loan performance databases and is a recognized leading provider of mortgage and automotive credit reporting, property tax, valuation, flood determination, and geospatial analytics and services. More than one million users rely on CoreLogic to assess risk, support underwriting, investment and marketing decisions, prevent fraud, and improve business performance in their daily operations. Formerly the information solutions group of The First American Corporation, CoreLogic began trading under the ticker CLGX on the NYSE on June 2, 2010. The company, headquartered in Santa Ana, Calif., has more than 10,000 employees globally with 2009 revenues of $2 billion. For more information visit www.corelogic.com CoreLogic 4 First American Way Santa Ana, CA 92707 FOR MORE INFORMATION PLEASE CALL 1-866-774-3282 CORELOGIC is a registered trademark of CoreLogic 17-VALLESS-1010-00 corelogic.com