A Comparison of Free Online AVMs
One of the most common questions that I get as an appraisal professional, from mortgage originators is “How do I tell what the value of the property is before I spend a Borrower’s money on appraisal?”
While many Appraisers can be offended by the concept of a “Comp Search”, this is a legitimate question from a mortgage originator.
As an Appraiser, I have always felt it’s my obligation to educate the users of appraisals so that they can better understand how and why we do our jobs.
Couple that with the fact that a mortgage originator is an advocate for their client, and I have a great excuse to write this short blog posting.
I have selected 5 of the more popular free online automated valuation models, or “AVM”s and I’ve run them through their paces on several different properties throughout California.
Zillow, Trulia, Bank of America’s Real Estate Center, HomeGain and eppraisal.com. There is also a plethora of fee-based AVMs. They vary in their accuracy as well, but that’s for another blog posting. If you know of others, post them at the bottom of this blog and I will play with them and report back.
The first thing I did was to set some criteria for selecting the homes to run the AVMs one. I selected sales that have occurred very recently, all after 3/1/2013. Each is a reasonably standard home, for it’s area, in typical to good condition for the neighborhood. I purposely stayed away from homes with pools, significant views or unusual design or appeal. In other words, I want homogeneous data. The goal of this exercise is not to showcase the AVMs shortcomings, which is a whole ‘nother blog post, but to see which ones are the most accurate and reliable.
Now come the numbers. I set up a spreadsheet with the:
Recent Sale Price
The differential after each AVM value is the AVM value divided by the recent sale price, expressed as a percentage. The closer to 1.00, the more accurate the AVM value. These are then averaged for a final indication of overall accuracy and reliability.
Finally, I calculated a differential ratio of the high to the low AVM value for each AVM. The narrower the range, and thus, the lower the final number for 5 properties, the more overall accuracy is indicated for that AVM. As a note, two of the AVMs express results as a range. The High and Low differentials are calculated and then averaged to compare to the other AVMs.
hint….use the bottom and side scroll bars to see all of the data.
So what do we see from these results?
Clearly, that there are good AVMs, relatively speaking, and not so good ones. In fact, if you use the wrong AVM, the results can be wildly misleading.
AVMs definitely have their place, but the user must understand their limitations and what they should be used for.
The best results got within 3% of the recent sales price of the sample properties. The Zillow Zestimate performed best on the San Francisco property, while Trulia’s AVM, which also got within 3% of an actual value, missed the San Francisco property by 20% too high. The Trulia AVM scored well on the Campbell property, but the Zillow Zestimate was off by 13% on that one. Although it gives a value range, the Bank of America free AVM was relatively accurate on the high side of it’s value ranges. In fact, the high side of the B of A value ranges ended up being the most accurate of the 5 AVMs, on average. eppraisal.com and HomeGain had very inconsistent results, including eppraisal not even finding the San Francisco property.
For the user of one of these AVM services, using the best of the three, Zillow, Trulia and B of A, will give you inaccurate and inconsistent results at times. The best way to mitigate this, in my humble opinion, is to use these 3 in concert and look for an average of the 3 results and look for consistency in the results among the 3.
So far, the holy grail of a free online AVM that is accurate within 5%, everywhere, is still just a pipe dream. But with a little analysis and comparison, a user of the these AVM services can expect an accuracy rate to within 10%+-, at best, on the simple properties in conforming areas. The trickier the property, however, the more variation and less confident your results will be.
What can throw an AVM off? Things that is can’t “see”. Remodeling, views, locational factors and other less tangible aspects of a property.
Have your results been different than mine? Maybe you have found a better free AVM? If so, Please leave a comment on this page.