The Modeling Business [Mortgage Banking]

Quants-not fashionistas-rule in the automated valuation model (AVM) business. They are a super-smart bunch, and a number of big companies are angling for their know-how.

"Without a doubt, you have to be careful in the AVM [automated valuation model] space," cautions Darius Bozorgi, president and chief executive of Santa Ana, California-based Veros Real Estate Solutions. "You have to have good data and a good team; you don't want to have one guy sitting in the corner responsible for the AVM. That is a very dangerous prospect because if someone comes along and grabs that one guy, you are in trouble." * "Some very large organizations have had their key modelers stolen by other larger players," Bozorgi adds. "That causes a big ripple effect in the industry." * Stolen may be too harsh a word. Lured away is more like it. That kind of thing happens in all industries where the talent pool is limited. But when it comes to the specialized world of AVM modelers, talent is critical - as well as limited. So luring-away happens, * From the perspective of two such lured-away AVM modelers, they were just looking for new fields to conquer and harbored no hard feelings toward their old company. * In 2010, Renjie Chen, CMT, and Robert Walker, CMB, CMT,

two veteran - and, some might say, superstar - AVM modelers, left Santa Ana, California-based CoreLogic to join Jacksonville, Florida-based Lender Processing Services Inc. (LPS). Actually, Chen and Walker didn't really have to go across country - they simply moved across Orange County to LPS' offices in Irvine, California, where Walker is now managing director and Chen is senior vice president of AVM analytics.

The ripple effect described by Bozorgi hasn't fully played out in the AVM industry yet, but it will in the near future. For the last few years, change has been rife in the business due to new technologies and regulatory evolution. Increased competition will shake things up even more.

CoreLogic has been viewed by most as the big dog in the industry. LPS should have been or could have been, but wasn't. Now it wants to be.

"CoreLogic is the biggest and the best in the AVM space," notes Andrew Caplin, professor of economics and co-director of the Center for Experimental Social Science at New York University, and co-author of the research paper, The AVM Cascade Industry. Advanced Valuation Analytics Corporation.

Caplin got to know CoreLogic because it underwrote the A VM Cascade paper, and so he declares, "Everything that I saw tells me there is no one who could quite match CoreLogic in the AVM space. It doesn't mean there aren't local winners, but if you were constrained to pick one supplier for the country as a whole, you would take CoreLogic."

The principal reasons behind Caplin's singling out of CoreLogic are a long history in the AVM business and, more importantly, a massive database.

"There are two large groups that have massive data amounts - CoreLogic and LPS," Caplin affirms.

Here's the way Chen looks at it: "When I left CoreLogic in 2010, everything was in place. Here [LPS], they needed more work - so I had more challenges and opportunities."

"LPS had numerous AVM systems when Chen arrived, but the company wanted to take it to the next level," he says.

Chen won't get any argument from his boss, Dan Berman, president of LPS Applied Analytics, also based in Jacksonville, Florida. "We hired Chen and Walker because we felt it was critical to have a better model than we historically had at LPS. We are in the top three or four right now, from a performance standpoint, but we are putting ourselves in a position to dominate tomorrow."

Some of the innovation Walker and Chen are bringing to LPS and the industry is reflected in new AVM tools they are developing that suit current unique market niches.

One of these tools, LPS' Distressed Asset Review, combines the accuracy and consistency of an AVM with property condition information. Traditionally, AVMs have not accounted for property condition, which is why broker price opinions (BPOs) have become the valuation tool of choice among default servicers. Conversely, while BPOs evaluate property condition, the inherent subjectivity and inconsistency in these valuations can present challenges for default servicers. LPS' Distressed Asset Review taps the LPS ValueSure(TM) AVM - one of the industry's most accurate AVMs by independent testing, according to LPS - and then incorporates a real estate professional's property condition report to produce a more reliable value. It also meets both the letter and the spirit of the Treasury Department's Interagency Appraisal and Evaluation Guidelines.

"This new product gives default servicers the best of both worlds - values that take into account property condition and an AVM that is not subject to the inconsistencies inherent in BPOs over the distressed prop erty's life cycle," says Walker. "In this market, we know clients are eager for this more reliable data about underlying assets' values, so they can make critical loss-mitigation decisions."

Setting the stage

AVM, simply explained, is a technology that can provide property valuations using mathematical models.

The first wave of AVMs appeared at the end of the iq8os and the start of the 1990s. "The acronym AVM didn't come into play until 1995," notes LPS' Walker, who started working in the industry in the early 1990s - which, even then, was based on creating a computerized program could value a residential program cheaper and more accurately than an appraisal.

In the early 1990s, as Walker sums it up, the early AVMs were all about cost and "reasonable accuracy."

Walker continued with numerous AVM companies through many product tions until late in this first decade, when he ended at First American Financial Corporation, a company that had employed him a couple of times before. First American was another Santa Ana-based firm that had a history of expanding in the real estate services industry through mergers, and in 2007 '* acquired CoreLogic and began operating it as First American CoreLogic. Three years later, CoreLogic was spun off from First American as a standalone company. The remaining company became First American Corporation, parent of the better-known First American Title Insurance Co.

When making the decision to join LPS, Walker says, "I was always looking for new challenges, and the big thing about LPS was it was the only company, with the exception of First American, that had the underlying data. Any other AVM company in the country is going to have to go to one of two repositories - First American and LPS - to get the underlying data."

When Walker arrived at LPS, his first impression was, "the company was a big gorilla with all kinds of assets and data horsepower, but it lacked focus."

A change in thinking at LPS happened because, says Walker, the company had secured a deal to build an AVM for the National Association of Realtors® (NAR), Chicago, which wanted a product as good as anything already in the market.

"Arguably they wanted the best, but the best is a tall order on any given day," Walker says. "LPS recruited Renjie [Chen] and myself as two of the leading AVM minds in the country to make this AVM for the NAR and strengthen LPS' AVM business."

"NAR's vision was to create a product with two defining characteristics: one, highest accuracy, based on the incorporation of market information or listings; and two, fully authorized use of the listing content which is our [NAR] members' intellectual property. Realtors have an obvious interest in the efficiency and quality of the mortgage process, but have not, until now, had a vehicle to contribute to that. Through NAR's wholly owned sub^ sidiary, Realtors Property Resource®, the Realtor organization chose to partner with LPS because of the flexibility and creativity they demonstrated and brought to both the design and execution of our concept," says Marty Frame, president of Realtors Property Resource LLC.

The future belongs to the data

The first generation of AVMs was all about establishing the product and the process, and that period lasted until early in the 2000s. Then from about 2002 until the end of the decade, the second generation of AVMs focused on models, new models and new methods of valuations.

Now, the industry is entering the third generation, which Bozorgi says is much more focused on data, data sources, data availability, data frequency and conñict resolution among the different data sources. One might want to include the usage of multiple listing service (MLS) data in the latter category as well.

"Access to data is critical to what we all are doing," says Bozorgi.

"It's critical for any stakeholder in the AVM chain. AVM developers and companies that distribute these products have been incorporating new data sets - whether they always had them and never knew what to do with them, are just incorporating them for the first time or have acquired different data sets over the last couple of years. These are good problems to have because they will push AVMs to a new level/' Bozorgi says.

Sid Som has been in the AVM industry for about a quarter-century, and today he is founder of HomeQuant LLC, New York, and produces a blog about AVMs called Valuation Matters (http://valuationmatters.blogspot.com).

Today, Som says, the big issue is primarily how to get data and how to make sure the data are good. "There are a lot of smart companies that are involved in the AVM business, but access to data is limited. The big guys have the national data warehouses, which they leverage to do AVMs in-house," he says.

The big companies don't collect the data - they just hold the data, warehouse it and then reformat it all, Som explains. "The advantage of getting data from the national warehouses is that it is already reformatted."

The problem is, some of the data get old, Som says, because the data generally come from the property tax assessing jurisdictions, and considering that the larger jurisdictions tend to recollect those data on four- to sixyear cycles, under the most optimistic scenario, at any time just one-fourth of the data will be fresh, Som maintains.

Historically, the problem with AVM underlying data was that they were heavily dependent on public records. Realizing this was a weakness, AVM companies, beginning in the last decade, started incorporating MLS data as well. CoreLogic began using listing data in its AVM products starting around 2008, and since then the company has reported its models now have data on 85 percent of U.S. homes while reducing error rates by half.

Last year, LPS announced a new valuation model for lenders from its LPS Applied Analytics division that utilizes listing and pending-sale data. Called the Realtor Valuation Model (RVM), LPS reports it is the only AVM with duly licensed MLS data drawn from the National Association of Realtors'Realtors Property Resource(TM) (RPR) database.

Walker noted at the time of the Realtor Valuation Model launch, "With the introduction of the LPS RVM, users now have housing-supply information at their fingertips. By factoring listings and pending sales in the report, it becomes easier to better understand the external factors impacting the value of specific property."

According to an LPS press release at the time of the new product offering: "In testing, the LPS RVM has proven extraordinarily accurate when compared with comparative AVMs. In a sampling of Maricopa County, Arizona (Phoenix) properties, for example, the LPS RVM correctly predicted the selling price of a property 72 percent of the time. According to Walker, this high accuracy is the direct result of the addition of MLS data fully licensed by RPR."

He adds, "In the absence of MLS data, !prediction accuracy] might be in the range of 60 percent to 65 percent. The increase in accuracy is significant, from LPS' perspective."

There are a lot of flaws in public records, and AVM producers can only go so far with that kind of data, Walker opines. "We will get a lot more data from the MLS along the lines of property characteristics and square footage," he says.

Modeling as an art form

A lot of AVM creation is mathematical progression, Chen notes. "And the key thing for modeling is to understand the data. Even with public data, some of the information is inaccurate. Inaccurate information creates misleading results. Bad information coming in means bad information going out. The question is, how do you eliminate bad information?"

Traditionally, the people who created AVMs started with a statistical or mathematical formula and set up a model, but what Chen tries to do is start with the data or work from the bottom up instead of from the top down.

"Other AVM developers create the model and make everything fit the model," he says. "I look at the data and fit the model to the data."

AVM creators are constantly working through modeling to make improvements, Caplin adds. "It's a huge art form, and the question of how it all connects to fundamental economics remains opaque. Mostly, these are statistical filing exercises," he says.

Nevertheless, Caplin concedes current AVMs are much different from those of a decade ago. "These are massive, automated comparison engines with a lot of computing power. Depending on the definition of a house, recent history and nearby trades in similar houses, one can Come up with a fairly good estimator of what a house will trade for. Every day you can predict the trading value of any house and then see if you are right."

To which Caplin adds, "I don't think AVMs are any more based on fundamental economics than they ever were, but they have pattern recognition. As pattern recognition improves, AVMs will improve."

Perhaps the biggest flaw with the AVM is that it negates the human equation. That is, previously no one ever went out to eyeball the home - which is often not necessary, but in these tough times when homes have been abandoned, a lot of damage has been inflicted by vandals, or even in some cases, disgruntled former owners. That's to say nothing of the thousands of homes that have been damaged just this year by floods, tornadoes and wildfires.

AVM companies have been seeking to compromise with the human equation. LPS, as an example, introduced in February what it calls a condition-adjusted valuation model - when an AVM is requested, an order is simultaneously produced to get a property-condition report completed by a real estate sales professional. The findings are then factored back into the AVM.

AVMs are data-driven, but Walker says if there is damage related to flood, fire, hurricane or vandalism, LPS will make a note of that in the AVM with a message that says, "we cannot vouch for this property because of severe structural damage." So, there is a failsafe.

"The last thing I want is a regulated institution running an AVM on a property where the address exists, but the house was blown down yesterday," Walker says.

New regulatory environment

LPS' ValuSure Condition Adjusted (VSCA) valuation model was among the first AVM products to comply with the Federal Financial Institutions Examination Council's (FFIECs) Interagency Appraisal and Evaluation Guidelines (IAG), says Berman.

"The ValueSure Condition Adjusted valuation model goes beyond simply bundling a photograph of a property with a standard AVM," Berman said at the time of the product rollout in February 201 1.

The interagency guidelines indicate an AVM without a statement of condition is not good enough to underwrite a loan and not good enough to be the soie source for value, because AVMs assume average condition of a property, Berman explains. "What happens if the property is [in] below-average condition? That's why we launched the condition-adjusted valuation model, which takes condition of property into account."

Walker thinks one other vendor offers a conditionadjusted AVM product.

"Creating a condition-adjusted model requires access to a great deal of inspections and a highly advanced model capable of using and ingesting this type of information," he explains. "Therefore, it does not lend itself to product duplication."

Touting compliance with government regulations probably wasn't a big selling point through the first decades of AVM usage, probably because there wasn't a lot of oversight of the product. However, since the mortgage meltdown, the legitimacy and accuracy of the whole appraisal process has come under greater government scrutiny, and in some regards that includes the proper placement of AVM usage.

Beyond technology, the greatest change in the AVM world is the regulatory environment, notes Bozorgi.

In December 2010, the FFIEC released a new version of what is referred to as the Interagency Appraisal and Evaluation Guidelines, which, says Bozorgi, "is a document that had not been formally touched since around 1994. The revised document specifically referenced how the agencies ]that are members of FFIEC] expect the lenders to deal with all types of valuations, appraisals, BPOs and AVMs," he says.

The latter was specifically addressed in terms of model dations and when an AVM can be used as a method of evaluation.

"This is really the first time lenders got some specifics about the use of AVMs from the regulators," says Bozorgi.

"They did a good job in terms of providing broad strokes as opposed to being too prescriptive. They set some general guidelines around expectations and let the lenders conclude specifically how they can best meet those guidelines," he says.

The Dodd-Frank Wall Street Reform and Consumer Protection Act also addresses appraisal reform, including such subjects as appraisal independence, oversight of appraisal management companies (AMCs) and changes to the Financial Institutions Reform, Recovery and Enforcement Act of 1989 (FIRREA). Under Dodd-Frank, AVMs are required to have quality-control standards that achieve a high level of confidence in estimates produced by AVMs; protect against manipulation of data; avoid conflicts of interest; and require random sample testing and reviews.

Dodd-Frank certainly discusses AVMs, but some oversight might end up under the new Consumer Financial Protection Bureau (CFPB), according to Bozorgi - "so there are still a lot of unknowns," he says.

All this government interference makes people like Caplin queasy.

"Laws can be written insisting on A, B, C or D, but those writing the laws know next to nothing about AVMs," he says.

"There's no research, and the field is treated as if its numbers don't mean anything. This stuff is crying out to be understood before it is regulated, but that's not how people have been moving. Understanding is difficult; regulation is easy," Caplin says.

Government shouldn't be making decisions on this, because it doesn't know how to judge the good and the bad AVMs, Caplin adds. "Sometimes the government writes regulations and tries to identify which type of AVM to be used, which means it prevents the AVM industry from advancing. If government approves AVM III, what does it do about AVM IV, which might be better? It's hard to move forward when you have regulators who don't understand."

What's ahead

Back in 2008, LPS introduced REO Market Analysis, an AVM and market-analysis report that helps mortgage servicers identify default risks and trends in sales of real estate-owned (REO) properties. Chen is looking forward to getting his hands on it. "Time to upgrade," he says. "This is more for loss mitigation; the regular AVM is for the lending process. Most of it is the same data, just focusing more on a smaller market."

The market is going to see a lot more AVM specialty products, predicts Bozorgi. "The best example is the REO AVM or default AVM. The market will demand a specific type of valuation or valuation for specific scenarios, and ask AVM developers to create a model for that situation."

Bozorgi points out that AVM-backed loans have historically performed better than their counterparts, but AVMs still have an inherent weakness, which is why the major trend is to use multiple products - appraisals and BPOs - to get a more accurate valuation.

Som, however, maintains that's not going to be necessary because the industry is moving toward real econometric and mathematical models.

"Appraisal-assisted and BPO-assisted valuations are non-econometrical in nature," Som points out. In other words, there is a tremendous amount of subjectivity. "If you don't apply econometrics, if you don't apply a real quantitative solution, you are talking about a tremendous amount of subjectivity," he adds.

Som recognizes there are circumstances where data might show a property in decent condition while just the day prior, the property might have been demolished or burnt down - but these are marginal cases, he argues. "In 95 percent of the cases, the mathematical model will be just fine," he says.

The AVM industry is going to become large and even more important than it is currently, Som predicts. "It is going to be more scientific and quantitative in nature. Appraisals and BPOs will be used at a lower level."

Even though Walker would agree with Som, he has a slightly different take on the future trend line. "Ten years from now, the current model of appraisal will not be the same," he asserts. The reason: demographics.

"Today, the average age of an appraiser is somewhere between 55 and 58," Walker notes. "And there are fewer and fewer people coming into the industry because there is no money in it. Once these current appraisers retire, there is no new group coming in to replace them."

AVM, anyone?

In the early 1990s as Walker SUIIlS it up, the early AVMs Were all about cost and "reasonable accuracy."

The AVM industry is going to become large and even more important than it is currently, Som concludes.

Steve Bergsman is a freelance writer based in Mesa, Arizona, and the author of Growing Up Levittown: In a Time of Conformity. Controversy and Cultural Crisis, available on Amazon.com. He can be reached at [email protected].