Researchers reveal two different attempts to out-smart the Rothko and Warhol markets with machine learning; Victoria Miro holds back.

This commentary by Marion Maneker is available to AMMpro subscribers. (The first month of AMMpro is free and subscribers are welcome to sign up for the first month and cancel before they are billed.)

The Rise of the Machines

The desire to invent a machine that can predict the future behavior of markets, random events like horse races or casino games seems to be nearly inexhaustible. The discovery of machine learning has only renewed interest in that desire. Predicting the prices of art works is now becoming a popular pastime for the mathematically minded. who believe that enough data run through our near limitless computing capacity can can possibly reveal patterns that will predict price better than knowledge and experience.

You might have thought these efforts were still somewhat in the realm of spec scripts for movie thrillers or academic test cases. But today we have two different attempts to predict prices for next week’s marquee auctions of Contemporary art.

On the one hand, we have a group looking at a few Mark Rothko prices; on the other, we have an art market start up running a bake on the whole Warhol market. We won’t know whether to take either of these seriously until next week. Even then, we should take care not to be fooled by randomness, as Nassim Taleb tells us, and assume that a single set of predictions constitutes a meaningful tool. Nonetheless, let’s start with Rothko and then move on to Warhol.

In a rather long-winded introduction to their work, Devin Liu and Doug Woodham explain that they chose Rothko because of the consistency of his main artistic output and the ability of a computer to read digital images of Rothko’s work. They claim the 118 works sold since 2000 were enough data for their algorithm to be able to predict prices within 5% accuracy—though it’s not clear how they came up with this 5% back-testing. Let’s let Liu and Woodham explain,

• “What makes this best-performing model so interesting is that its predictions are based simply on the digital image plus five variables: painting height and width, whether it’s a work on paper or canvas, the number of billionaires in the world, and the wealth they control. None of the other variables in our database appeared relevant to the price, including the date the painting was made.”

Using this model, Liu and Woodham turn their attention to the works on offer in the May sales. They start with the work being sold by SFMoMA at Sotheby’s which has a pre-sale estimate of $35m and an irrevocable bid. When you add the buyer’s premium, Liu and Woodham say, “Sotheby’s is forecasting the painting will sell for between $40.1-$57.2 million.” Here’s what the machine learning says:

• “After running this painting through our algorithm, the model predicts it will sell for $42.3 million (including buyer’s premium), toward the lower end of Sotheby’s auction estimates.”

When the de-accessioning was first announced, many looked at the $35m pre-sale low estimate as ambitious. Liu and Woodham think Sotheby’s played it too safe. They also think the works on paper from 1969 are under-estimated. They feel the ultimate buyer will pay $16.6m for the first one:

• “Untitled (Red and Burgundy Over Blue), has a pre-sale estimate of $9 million to $12 million, or $10.5 million to $13.9 with buyer’s premium.”

The second, smaller one they think will cost $13.7m:

• “Untitled (Red on Red), is smaller but notable for its bright red palette. Sotheby’s estimates it will sell for a hammer price between $7 million to $10 million, or $8.2 million to $11.7 million with buyer’s premium.”

Is this magic? Maybe not. Liu and Woodham make a strong case that the small number of Rothkos rise and fall on a sea of liquidity. When the wealthiest have more money, Rothkos are worth more than when the wealthiest have relatively less money.

Let’s turn to a more complex market. Trying to explain the drought of Warhol material on the market over the last few years, and especially this year, has been a fun parlor game in art market circles. Artbnk, a start-up trying to build a mechanism that can provide real time valuation (RTV), has released a report on the Warhol market today too. 

Artbnk doesn’t tell us much about how its machine learning works but they do give us a thorough read-out of the prices they’re expecting. Starting from the top, they think the Double Elvis [Ferus Type] is priced to the fullest at $50m. Since the work has a third-party guarantee, we can be sure that the final price paid will be at least 15% above what they predict. You have to work your way down through the beautiful blue Liz from the Meyer collection and the haunting Little Electric Chair that Si Newhouse owned and the large portrait of Russell Means to get to a Colored Campbell’s Soup Can that Artbnk believes will out-pace the estimates and overperform.

The Fiterman’s Two Coke Bottles and Yusaku Maezawa’s Flowers also get the thumbs up from Artbnk. (The Fiterman’s assembled matrix of flowers don’t fit into algorithm.) They also like the Map of USSR Missile Bases (Positive) and a 1962 Soup Can drawing. But not as much as they like the gold Jackie and a huge piss painting both of which Artbnk tags to double the high estimates. It’s not a surprise that the Dollar Signs get a wave from Artbnk. It is a bit of a shock that all three society portraits are slated to sell better than estimates. These have always been the dogs of Warhol’s oeuvre.

Artbnk promises to return to these predictions after the sale. They view this as a way to build their system’s accuracy. So we’ll all get a better sense after the sales of how well machine learning can guide bidders and their expectations.

Victoria Miro Gallery Sued for Failure to Deliver Painting

Process servers stalked Victoria Miro across New York last week finally catching up with an employee of the gallery at TEFAF. Miro had already gone home. The lawsuit is over a million-dollar deal. Miro sold a painting for $530k a year and a half ago. The buyer still hasn’t received it. Even so, the buyer paid for $637k for another work by the artist and, at the request of the gallery, donated the second work to the Tate.

This act of largesse earned the buyer the right to purchase a third work by the artist. But two months ago, things took a dark turn. According to the summons,

• “Notwithstanding the foregoing, the Gallery has not delivered the First Work to Plaintiff despite repeated demands, nor has it honored the Plaintiff’s decision to exercise the option to purchase the Third Work. Rendering even more troubling the gallery’s conduct, on March 19, 2019, the warehouse responsible for arranging shipment of the works to the Plaintiff advised recently that, ‘[w]e have been asked this morning to hold the shipment at the request of Victoria Miro.’”