Mathematics and wine

The wine futures market hinges on predicting the price of wine long after its been placed in barrels to aid the aging process. And the price of wine hinges on the taste of the aged wine before it becomes aged. And the tasting hinges on the palate of so-called expert wine connoisseurs. What an inefficient predictive model!

Tristan Fletcher of the University College in London thinks the predictive model is inefficient too, and wants to change it, using the application of artificial intelligence (AI) – i.e. the application of mathematics and wine (The Economist, August 8, 2015).

Previous mathematics and wine analyses were based on linear regression models. Linear regression (I hope my students remember this) applies factors related to specific vintages – such as weather, soil, precipitation, history and so on – and plots these on a graph over time to form a straight line that approximates the price. Pick a point and there’s an approximated price (this is where the term ‘price point’ comes from). I would have used this model too. But Fletcher has another model.

Instead of regression, Fletcher applies AI. AI uses correlations – instead of a straight line, the result is price curve, which Fletcher believes yields stronger predictions. To do this, Fletcher needed a lot of information. He gathered the information from the Liv-ex100, which is the global marketplace for professional buyers and sellers of fine wine. He looked only at the historical price information of lots of wines. He found two distinct groups (published in the Journal of Wine Economics).

Half of the wines Fletcher had information on fluctuated in price over short periods, settling towards the mean return quickly – he called this the calm group. The other half of the wines trended up and down more erratically over longer periods. Therefore Fletcher conducted two types of AI (computer-based) algorithms, one on each group, and compared them to the results of the linear regression model.

The calm group under the AI algorithm produced similar results to the linear regression model (it actually out-performed it a little, but not significantly). The erratic group under the AI algorithm outperformed the linear regression model – and this is the group where wine speculators could make more money on the futures market.

But let’s remember that this is the futures market. This is the ‘speculative game’ of agreeing to buy and sell an asset (in this case, wine) at a certain date in the future at a certain price. This price must be honoured (it must be paid) whether the price goes up or down. The buyer must take the wine at that date at that price. Even with mathematical models, the price can still be volatile and risky. Wine in the futures market, like all else in the futures market, is still a risky business.