Art AVMs

Fellow appraiser and Appraisal Foundation ASB member Tim Luke, ISA AM sent me an interesting article from Forbes on fine valuations unis an AVM (automated valuation model). Real estate AVMs have been around for some time, and with technological advances with artificial intelligence there are more opportunities for AVMs to be applied to fine art appraisals. Of course how accurate and how conclusions are reviewed, interpreted and applied are of utmost importance.

Tim pointed out that when AVM search auction databases for comps there are concerns with incorrect reporting or use of sales which may not have been completed. A few years ago when ISA was working with Collectrium (collection management software owned by Christie's) they were attempting some form of an AVM, although it never came to fruition.

I am soon to discuss this topic with an AI art program vendor and an art trend/index developer and hopefully will have more insight on the topic in a few weeks. I will be posting on my discussion, the website and data usage shortly.

Forbes reports

At the RICS WBEF held in New York last May, I was asked by the moderator if an AVM could ever be developed to appraise paintings.

2019 RICS WBEF in New York, panel on AVM's
Panel on AVM's at the 2019 RICS WBEF in New YorkRICS
(Acronyms decoded: RICS=Royal Institution of Chartered Surveyors, WBEF=World Built Environment Forum & AVM=Automated Valuation Model.) With AI all the rage, HR gurus are taking stock of what its impact will be on those of us who are mere wetware.

AVM’s are already infiltrating the world of real estate valuation, but a quick Google search did not uncover any reference at all to an AVM which can appraise your Warhol. Part of the reason, naturally, is that the average building is worth far more than the average painting. Hence, the fees are higher, and the amount which can be invested in R&D is commensurately greater. Yet, there may be another explanation.

Today In: Lifestyle
If you have ever reviewed a real estate appraisal, you would have seen a section marked “sales comparables”. Therein, the appraiser lists the sales of houses like yours, near yours. Most neighborhoods are fairly homogeneous, so the comparable buildings selected are most likely not too different or too far from the subject structure.

Paintings, though, are substantially unique. Sometimes, there are no true comparables whatsoever, or frustratingly, if there is one, the sale took place 32 years ago. The data which personal property appraisers need cannot be located nearly as easily as those for the real property sector. In appreciation of this, it is generally accepted that two art appraisers can disagree by as much as +/- 10% without either being deemed off-base.

I would say, though, that it should be possible to develop an art AVM. The easiest prey would be prints. If Warhol printed 250 examples of his 1969 screenprint Campbell’s Soup “Chicken ‘N Dumplings”, they would all be practically identical.

Andy Warhol, Chicken 'N Dumplings, silkscreen, edition of 250
Andy Warhol, "Chicken 'N Dumplings", silkscreen, edition of 250LIVEAUCTIONEERS
Odds would be high that an AVM would be able to search the art databases and identify numerous recent sales of that very image to use as comparables. All a human would really need to do would be to correct for the condition of the comparables, and, perhaps, the different sale locations and dates. That would mainly be deemed fine-tuning.

As I see it, such an AVM could set the baseline in legal disputes, such as divorce or probate. The expert witness on either side would then have to justify to the court why his/her figure might have deviated from that generated by the AVM. If one appraiser submits a figure which turns out to be 10% less than the AVM’s figure, then minimal justification would be necessary. If the other’s number is 30% greater than the AVM’s figure, then that would call for a lengthy explanation.

Similarly, if the IRS chose to adopt an AVM, estate and donation appraisals could be analyzed by the system to generate a baseline figure. If the appraiser’s number is within +/-10% of the AVM’s, then it could be automatically accepted. If not, the IRS in-house appraisers could then review the appraisal in depth and decide whether or not to challenge it.

Of course, an AVM in any realm is only as good as the data fed to it; (garbage in, garbage out). If the appraiser has misdescribed the subject work in any way, the system will choose the wrong comparables and miscalculate the value. For prints, as noted earlier, most of the details are readily apparent. For original drawings and paintings, though, far more interpretation is required. Is the artist known? If so, is this his typical piece? Is it from his most desirable period? Is it certainly authentic?

Ideally, as a proof-of-concept, a system would first cut its teeth on prints, using machine learning and a feedback loop. The auction realm offers a perfect training ground. Myriad pieces are offered each season, and their ultimate sale prices are easily obtained. The system would appraise each piece when the auction is publicized and then compare its calculated value to the actual sale price weeks later. It could then adjust its algorithm for its errors, and refine the system ad infinitum. From prints, it could move on to sculptures which were cast in editions, and then on to original works of fine art, and then decorative art, gems, etc..

For decades, the vision of an art appraiser was that of a bespectacled, accented gentleman who put his hand to his forehead and magically set a value on the spot.

AVM’s can help bring the field into the 21st C..

Source: Forbes 

---