AVMs Can NEVER Replace Appraisals
AVMs Can NEVER Replace Appraisals & why FNMAs vaunted CU AVM is flawed!
Yesterday (March 15, 2016), an appraiser called me to discuss a complex assignment he had. One of his comparable sales was reported in MLS as having sold for $3,469, 898. It was in range with his other comparable sales for his subject property. It was the most similar comparable property.
When he went to verify the sale price with our Los Angeles, California subscription-equivalent to “public records” he discovered a shocking discrepancy he couldn’t explain.
RealQuest (CoreLogic) showed the full value sale price as $17,665,455! So what one may ask? Just use another comparable sale, right? Even though it was the BEST sale available? It’s the data reliability that is the underlying issue.
IF you had five (5) sales in the $3 million to $4 million range and their actual average was $3.5 million, but one had been misreported by the SAME DATA SOURCE used for AVMS as indicated above, the “average” would appear as $6,333,091! OK I opted to use averages rather than medians because it’s easier to follow and calculate, but the same extreme result applies when medians are used. Certainly the amount skews any AVM ‘value’ calculations. The extreme degree of error remains whether it is a FNMA conforming loan limit amount or in a super jumbo loan range.
This is not an isolated occurrence, but relatively few appraisers are aware of WHY it happens. It significantly affects statistical reliability. We’ll pretend right now that 1004MC isn’t affected by this too.
“Public Records” data aggregators like CoreLogic don’t receive confirmed sale prices from the County Tax Assessors Office. The County does not compile THEIR records that way. What they and most title insurance companies DO receive daily is an electronic tape reel (or digital equivalent today) with the dollar amount of documentary transfer stamps paid for each sale.
THAT amount must be translated into a “verified full value sale price” by the aggregator.
So, in the above case the aggregator’s translation-clerk (or OCR algorithm software) correctly read doc stamps as $19,432. They also “know” that Los Angeles County transfer stamps are $1.10 per thousand dollars of sale price. In their mind [$19,432 / $1.10] x $1,000 = $17,665,455. So far, so good aside from being completely wrong.
The CITY of Los Angles ALSO assesses an ADDITIONAL $4.50 per $1,000 of value (sale price)! Combined with the County Transfer Tax, “doc stamps” are now $5.60 per $1,000. So [$19,432 / $5.60] x $1,000 = $3,470,000. This by the way is the rounded equivalent of the actual sale price as reported by MLS?
OK, FNMA and all the AVMs fans will argue that in the era of “Big Data” that the sample sizes are so huge that overstating the sale price in one city by more than $14,000,000 “doesn’t really skew the results enough to affect reliability.” Hmmm. Sounds like snake oil to me, but…
How about when OTHER cities also have special calculations the aggregator misses? For example Culver City has $4.50 per $1,000 (Plus the County rate of $1.10) just like L.A. City; Pomona has $2.20 per $1,000 (plus the $1.10), Redondo Beach has $1.10 PER $500 ($2.20 per $1,000 equivalent but stated differently in their ordinances-PLUS the $1.10). Santa Monica has $3.00 per $1,000 PLUS the $1.10. NOT listed in my chart is the City of Torrance with a $2.20 per thousand.
So, it is NOT just one ‘small’ city like L.A. after all.
FNMA (and other AVMs/BPOs proponents) wants us to believe the sample size adequately minimizes the impact of human errors I just pointed out.
The problem is that their Collateral Underwriter bases the geographic area of comparison to the same census tract as the subject! Yes, it can be over ridden in an AVM but the default comparison is the census tract or a half mile radius, and THAT is a very small area in which fourteen million dollar errors on individual sales WILL have a significant effect!
IF the only concern were doc stamps in Los Angeles, I’d say it is a problem that COULD be solved (in time), but how many OTHER Counties across America have transfer stamp data translated this way? Maybe there’s a REAL reason Zillow can’t reliably value real estate.
In Eckert, Colorado or Enid Oklahoma the appraisers may still drive down to Grand Junction or wherever (sorry Oklahoma!) and look at actual deed copies but I doubt it. How was the FNMA data entered or derived for non-disclosure states like Texas or New Mexico? How many other states are affected by anomalies in “big data”?
Back to California. Like I said IF these were the only issues I assume the greatest unlicensed “appraiser” minds at FNMA who invented the patented CU But are they even aware of the OTHER issues resulting in near zero credibility for AVMs?software process could also solve these kinds of glitches that directly affect the credibility of their initial AVM based screening.
But are they even aware of the OTHER issues resulting in near zero credibility for AVMs? Like the number of refinances being reported as much lower value sales? Rare? Hardly.
The trend in California (often) whenever a quit claim is involved in the chain of title (when property is purchased as sole and separate property by a married couple for instance, OR where it is put into an LLC or S Corp via QC) the LATER refinance MAY or MAY NOT REQUIRE grant deeding by the original parties into whatever entity holds title presently.
In California if a grant deed is involved along with a refinance deed of trust, a ‘sale’ is presumed for calculating doc stamps. So, a two million dollar property owned free and clear for many years with the Medical Doctor owner and his or her spouse frequently having it QC’d in and out of both names over the years has the title insurance people dictating that the chain of title be ‘cleaned up’ by new grant deed to record a new $400,000 refinance.
The transaction shows as a SALE for $400,000 even though title data aggregators will claim it does not! Certainly the AVM sees it that way.
How about REO’s that are subsequently flipped at quick sale prices?
No algorithm can adequately screen out huge anomalies in ‘Big Data’ that require specific, competently trained local expertise to recognize.
That is why we have full real estate appraisals and that those in turn are required to be performed to meet certain MINIMUM standards under USPAP. Circumventing those MINIMUM standards is the same as circumventing the will of the Congress of the United States of America as expressed under FIRREA 1989 Title XI.