Share
 

The art of art analytics

 

Today I would like to talk about a topic that I occasionally find it very intriguing: How do we evaluate art? Can we use big data and artificial intelligence to price one of the “Tagli di Fontana” or a piece of Kusama? How do we determine whether a certain artist is overpriced? Do you like the picture below?

Figure 1

First things first. What is fundamental and what is irrational about asset evaluation? How do we identify whether an asset is in a bubble? When do bubbles burst? Financial economists have been debating these issues for decades. The outcome is a set of sophisticated analytical tools at the heart of algorithmic trading strategies. Discovering a nascent or a fading bubble has simple trading implications. Equally important is relative evaluation in a given asset class: which assets in this class are undervalued or overvalued? A long-short strategy (buy the cheap, sell the rich) would help investors regardless of the direction of the bubble. But, how do we know whether an asset is undervalued? Modeling relative value is an art, and the tools in this space may be quite sophisticated.

We may make use of comparable analytics in the fascinating art space, a space governed by amazing regularities despite the obvious unpredictable traits that underlie creative work. First, one would need to identify general trends in market evaluations for a given artist. One would, then, need to collect transaction data on this artist and extract general trends in these data, in spite of the unavoidable heterogeneity in the artist’s creations. Admittedly, though, this task might be expensive. Data vendors might require hundreds of thousand dollars subscription per year for having access to their unstructured data.

Figure 1 depicts an index of market evaluations that I have created for Yayoi Kusama, a celebrated Japanese artist. The index equals 100 in the first year for which we record transactions; the index, then, evolves over time, going up or down according to the average price direction in the universe of Kusama’s artworks over time.

The method of index construction is instructive. The methodology enables one to extract common trends in the artist’s artworks while also supplying tentative predictions of the value of each single artwork in the index universe through an approach known as “hedonic.” Consider Figure 2. Its left panel depicts the index values of Figure 1. The right panel depicts, instead, the price of sold artworks (i.e., observed prices) against the price of artworks predicted by the model. Shown in this right panel is also the 45-degree line, in red.

Figure 2

If the model were able to predict prices without errors, the blue circles would all lie on the red line. Alternatively, and in an idealized world in which the model holds perfectly, the points below the red line identify artworks that are cheap. The model, then, delivers an assessment of relative value. If we were allowed to sell short, we could implement a long-short strategy, selling the artworks identified by the circles above the red line, and buying the artworks identified by the circles below the red line. In practice, the model is only an approximation of the subtle market mechanisms underlying Kusama’s creations, and such trades are risky. Still, the model provides us with a useful trading recommendation that may complement traditional advice in this space.

Art analytics might truly help us understand art evaluations. An art-tech company might then have to resemble a fin-tech. Its dedicated team would have to aim at developing methods and processes to analyze artworks’ data for the purpose of indexing and pricing. Its data scientists, mathematicians and economists would have to dialogue with art experts and improve the provision of information and implement tasks such as:

  • Creation of a trading platform. Would need find regulatory clearance: not a trivial task.
  • Developing automatic index feeds and displays on the platform.
  • Provision of bespoke indexing and pricing services to UHNWI, such as portfolio selection, financial advice, trading recommendations.
  • Developing hedonic pricing algorithms comprising art expert categorical variables: the antecedent to big data analytics.
  • Designing financial products centered on artwork indices.
  • Contributing to mechanism design regarding pricing and trading protocols.
  • Designing listable products.

At the moment, these tasks seem to belong to science-fiction or, better, art-fiction…

Keywords: Art pricing models; hedonic models; big data analytics; artificial intelligence; bubbles; relative value trading; art-tech & fin-tech.