It’s no secret the lending industry is aggressively pursuing alternatives to traditional appraisals. While most participants in the industry will concede an appraisal is still the gold standard for collateral valuation, they increasingly see the appraisal as something to be steered around or avoided altogether. An array of products/alternatives are being bandied in lieu of a traditional appraisal: Hybrids, Evaluations, BPOs (broker price opinions), Desktop Appraisals, and outright Appraisal Waivers. The industry is also turning its attention more and more to AVMs (automated valuation models).
AVMs have been around for nearly three decades now. The first AVMs had a hard time with consistency and accuracy, but as the world has become increasingly digital and data driven, the accuracy of the models has been steadily increasing. In conforming tract home subdivisions with sufficient recent sales, their value may be extremely accurate.
An appraiser should keep an eye on alternatives to the services he/she offers, so I recently downloaded a number of white papers on AVMs from different websites, including Clear Capital’s. I was somewhat surprised that Clear Capital’s white paper freely admitted that an AVM is not as good as an appraisal, but could be a starting point in risk assessment and collateral valuation. I had to provide my email address and phone number to access the white paper. This resulted in me receiving a follow-up call from Clear Capital. (Full disclosure: I previously worked for Clear Capital for a couple of years doing quality assurance work on incoming appraisals.) The gentleman on the phone was very nice. We chatted a bit and I asked him how long the product has been offered and he mentioned they took their previous AVM off the market and just released a new and improved model a couple of months ago. I asked him if they needed beta testers and he said no, but he did email me a code to sample five AVM valuations for free on their website.
That day I had just completed an appraisal report and, after completing it, ran the subject property’s address through their portal. The AVM is highly specific in its dollar output: $416,771, which I thought was a bit weird. It might be their way of emphasizing this is an algorithmic valuation. If so, that is the opposite of what we as appraisers do, which is round an opinion of value to emphasize a measure of uncertainty. (I sure hope no licensed appraiser is signing their name to an opinion of value as specific as $416,771!) For reference, my developed opinion of value was: $400,000, so the AVM was ~4.2% higher.
The AVM output is very bare bones: just three pages and mainly a list of addresses. I got 30 addresses in total; 15 apparent sales and 15 apparent listings; I say apparent because, oddly enough, there is no legend anywhere in the document explaining what the colors denote. The AVM gives a high estimated value of $451,500; a low estimate of $390,000; and the official, highly specific, estimate mentioned previously. It also provides a confidence score with a letter next to it: “H,” “M,” or “L” to presumably denote either high, medium, or low confidence. (My AVM report had an “H” next to the confidence score.) A map of the properties is provided on the second page. The third page consists of just two paragraphs of disclaimers: first warning the product is not an appraisal and then warning the reader not to reverse engineer their work or infringe on their intellectual property. With so little explanatory material in the document itself, I don’t think they need to worry about anyone being able to reverse engineer their system.
All the sales listed are within approximately one third of a mile of the subject property and the dates of sale within ten months of the date of the AVM report. (The listings are selected from within three quarters of a mile.) Five of the seven comparables I used in my report are found on the list, but none of the comparables I used cracked the top ten on either the sales or listings tally. My subject is a single-level home and all the properties I gridded in the report are single-level as well. Of the top ten sales the Clear Capital AVM used: five are two-story structures; one was not listed in the local MLS; another was listed but had no photos and no agent comments at all; one was a bit older and smaller; and the other two would have made acceptable alternative comparables in my report.
As I mentioned before, there is almost no explanatory material with the AVM output, so I really don’t have a clue how it derived its value estimate. I checked to make sure the AVM didn’t do something as simple as average the 15 sales (or the 15 sales with the 15 listings). It didn’t.
So, is something like this useful? I think so. While many of the properties it listed were rejected by me in my comparable search, the results were within 4% of my opinion of value. As their own white paper stresses—and the disclaimers in the AVM report itself reiterates—the AVM output is not an appraisal. The cost of the AVM would have been $10 if I didn’t have the promo code. I can see lenders using products like this when the loan-to-value ratio is low and the credit score of the applicant is high. (A full-blown appraisal may be overkill in that situation.) A $10 charge is also nominal considering the total fees a mortgage or other real estate loan product may generate. A quick peak at the AVM early on may give a loan officer an idea of what they’re in for or if the deal is feasible or not. (It is hoped said loan officer is also paying mighty close attention to the confidence score.)
(I am glossing over the bigger question as to how these products will be checked for accuracy, data integrity, tamper resistance, etc. But that is a blog entry for another day.)
While my opinion may be considered biased for obvious reasons, I genuinely believe appraisers offer an extremely valuable service to the public and are a pillar in sound collateral valuation and risk mitigation. However, it is clear that the future of valuation will include increased utilization of AVM products such as the one Clear Capital offers or the free Zillow “Zestimate.” Appraisers who work in areas with highly conforming properties will need to make sure they level-up their skill set to specialize in the valuation of complex properties; the type that currently gives computer algorithms nervous breakdowns. I strongly suspect the appraiser of the future will be, essentially, a data scientist. One that has strong Excel, R, Python, and/or SAS proficiency. There may be fewer of us in the future, but the ones that remain will be extremely talented professionals who offer local, boots-on the-ground expertise with high-level analytical skills. And we can offer what no computer currently can: a friendly smile and a listening ear as the homeowner tells the story of their property!