2016 Turrentine Crop Prediction Winners
Turrentine has announced the winners of their 2016 Crop Prediction Contest: https://www.turrentinebrokerage.com/blog/2016-crop-contest-winners/. I won the Statewide Chardonnay portion.
This was the first time I participated. For the category I won, I decomposed the problem into multiple parts. First, how many bearing acres will there be? Second, what will yields per acre be? Both of those questions were decomposed into an estimate of base-level rates and trend. That is, how many acres did we have last year and what was the average likely yield per acre last year; and how is that shifting over time. That took me too long to do for everything, so all other predictions were based of simple, linear regression. In no case did I adjust for optimal game theory to take into account that the contest is not won through pure accuracy, but in Price is Right style (closest to the number without going over.) Nor did I adjust based on knowledge from the field of how the harvest was looking.
It was interesting that the more complex model worked, even without the neglected adjustments, while the simple model failed to win in every case. I think next year I may make a more serious attempt to model all of the categories. It's an interesting exercise and I thank Turrentine for conducting it. It makes our industry better. But I think they should switch to a pure accuracy style. If there is not that wrinkle of being disqualified for going over, then they could actually use these predictions to create their own crowdsourced predictions.