We were learning about linear regressions in Data & Decisions a couple of weeks ago, and the topic of wine came up in lecture. Back in 1990, a Princeton Economics professor named Orley Ashenfelter developed a simple linear equation to predict the quality of Bordeaux vintages. Obviously wine critics laughed him off - Robert Parker called it "ludicrous and absurd", but Ashenfelter ended up predicting correctly that 1989 and 1990 vintages were the "greatest of the century".
Interestingly, Ashenfelter was able to isolate the determining factors of wine quality to weather. He modeled age and weather (temperature, rainfall, etc) as independent variables and selected auction price as the dependent variable. He ended up with a pretty good model, which is shown below:
Wine quality = 12.145 / 0.00117 * Winter Rainfall + 0.0614 average growing season temp – 0.00386 harvest rainfall
Crazy right? The coolest part is that one can make the predictions long before the wine is ready to drink which as all sorts of implications in wine futures. Makes me wonder what the wine industry is doing nowadays with technology and data. Does anyone know?
Hi Katie, very interesting post and thank you for sharing. I know that there was some initial push-backs when this predictive model first came out and I am curious as to whether that perception has changed in the recent years (given the wilder adoption of big data in many industries). One question I have thought: even though this model seems to make sense on capturing the basic year to year variations, how does it (if at all) capture the quality differences between estate-bottled wines of the same vintage?
ReplyDeleteI had totally forgotten about this! Thanks for resurfacing! I'd be super-interested to test the performance of this model on some of the investment-grade wines recommended in Liv-Ex. Will report back!
ReplyDelete