In valuing wines, better to depend on machine learning than palates
For long, putting a price on a particular vintage of wine depended on who the connoisseur’s palate responded to a tasting. Of course, there is more to a bottle of wine, an oenophile will tell you, than just the vintage; its pedigree and provenance, even the matter of whether the vine the grapes came off was affected by a particular fungus or not—these and many other details go into the valuation of wine. But most tastings happen when the wine is still in the barrel, that is, when it is still to reach its prime. En primeur tasting, thus, makes wines a terribly sketchy investment, as one palate would expectedly differ a great deal from another. At the same time, if you missed buying the finest—a bottle of Domaine de la Romanee-Conti Romanee-Conti Grand Cru from Côte de Nuits in France now costs nearly Rs 8.2 lakh—you might as well hit a bottle of cheap port to drown your regret.
The Economist reports that this is soon to change—Tristan Fletcher, a professor at University College, London, is applying machine learning techniques to assess the value of a wine in a manner that will hold wider persuasion. Though there have been attempts to lasso the wine market with linear regression models—these quantify the many considerations going into the valuing of a wine and plot a straight line that, over the course of time, closely approximates the price—these have produced mixed results. Fletcher’s method however digs out correlations that most regression models miss, resulting in more nuanced valuation. He worked with historical data on prices of select wines listed in Liv-ex 100, an index of 100 wines valued the most. For wines whose prices fluctuated wildly over longer periods, the machine learning technique was closer to the approximate value than the regression techniques. Wine aficionados would raise their glasses to that!