Computer Scientists Are Using Artificial Intelligence to Predict The Price of Fine Wine

Most people look at a bottle of wine and think about the taste—or its ability to get you totally white-girl wasted. Others, however, look at wine as a serious investment, a financial commodity that could increase greatly in value over time.

Any wine investor worth their decanter knows that the Liv-ex Fine Wine 100 is an index that represents the price movement of 100 of the most sought-after fine wines. The wines listed in the Liv-ex 100 are traded on a secondary market the way stocks and bonds are. Fine wine investors are, obviously, looking for those wines whose prices are most likely to go up.

But how do they determine that? Well, they can do it the old-fashioned way, using meticulous research, worldly experience, or intangible luck. Or, now, they can use more modern, computer-based models for predicting the future price of wine, just like the so-called "flash boys" do on Wall Street.

According to a report in Phys.org, researchers at University College London have developed a model that predicts the price fluctuation of fine wines using artificial intelligence. Their study is being published today in the Journal of Wine Economics.

The new model uses complex machine-learning methods and has been shown to outperform other simpler processes. By determining which data is most important in predicting wine prices, the new approach predicts prices with greater accuracy on the Liv-ex 100 index.

Professor John Shawe-Taylor—co-director of the UCL Centre for Computational Statistics & Machine Learning and Head of UCL Computer Science, who co-wrote the paper—said that their algorithms "learn from new data without human intervention."

UCL has developed machine-learning algorithms for other industries, including medical and financial ones, but this is the first time they have dabbled in wine.

Incidentally, we unequivocally need an AI sequel where Haley Joel Osment plays a fast-talking, oenophile boydroid.

The new system searches the data that is available for useful information—stuff that actually has been predictive of price in the past. This information is then extracted and used to predict the future values of wines. The computer scientists used two forms of machine learning: "Gaussian process regression" and the more complex "multi-task feature learning." Both worked better than traditional methods of price prediction, but where multi-task feature learning could be applied—it can't always be used—the accuracy of predictions increased by 98 percent compared to the old benchmarks.

Primary author Michelle Yeo, had this to say: "We're pleased we were able to develop models applicable to fine wines and we hope our findings give the industry confidence to start adopting machine learning methods as a tool for investment decisions."

One of the authors of the paper, Tristan Fletcher, who is a research associate in computer science, is also the founder of a quantitative wine asset management firm called Invinio. He plans to continue to work with the others to refine the algorithms so that existing and potential wine investors can use them through his company's website.

Next up? Phys.org reports that the authors are planning to apply the new investment techniques to another so-called "alternative asset": classic cars. That's all well and good, but when are they going to get to the real moneymaker? Potato chips that look like celebrities, of course.

Who can say? But maybe we should start putting our money on a future where sommeliers double as card-counting precogs, fighting crime from massive vats of syrupy Barolo.