Expert Judgment vs. Statistical Forecasting: Lake County Sauvignon Blanc Update
In August, I had predicted a price of $1300 per ton for Lake County Sauvignon Blanc for 2016. Single-year forecasts are the riskiest forecasts I make. The reason for this is that it typically pits purely quantitative methods against expert knowledge. The case of 2016 Lake County Sauvignon Blanc is a great case in point.
For this blog, I put out purely statistical, single-year forecasts. That is, I use only historical data to forecast a price. Someone out in the field, with enough knowledge would make such a forecast by asking around to find out what the contract prices for grapes are. The advantage of statistical methods is that, when properly done it can (a) remove all biases and political considerations and (b) is almost invariably better at assimilating multiple variables. That is, the numbers don't have an agenda and a mathematical calculation is better able than subjective consideration to reconcile multiple influences like, say, a large expected crop combined with rising demand. It is difficult to determine if increased demand will outweigh or be dwarfed by increased supply, unless you do some math.
On the other hand, domain knowledge, such as the information that a grape buyer has of the market is able to assimilate future information. A statistical model relies on past Crush and Acreage reports and other inputs to predict the future. Experts, however, know about specific contracts that will influence future price.
So, which is better? Well, it depends on the situation. For long-term forecasts, I'll take statistical methods any day of the week. Expert judgment has negligible accuracy at predicting prices five years out. Statistical methods, however, can be surprisingly accurate, since you can calculate the way prices respond to planting and the way planting responds to prices.
To forecast yields for a single, current harvest, on the other hand, you really want expert judgment. Historical yields are useful to guessing future yields, but actually visiting vineyards and seeing the crop offers stronger and complementary information. Ideally, domain experts will decompose their information into bite-size bits in a situation like this. This is simpler than it sounds. All it means is answering a few questions and penciling out your guess:
How many acres are out there?
What are historical per acre yields?
How much bigger or smaller is this year's yield going to be?
Acres x Tons/acre = Yield in tons
Often, the best course of action is to combine statistical methods and domain methods to leverage the best of both worlds. This, however, creates a complex process and so I don't do it for single-year predictions for my blog. OK, I know, I'm rambling without getting to the Lake County Sauvignon Blanc.
I was recently told by a domain expert that, when I made my purely statistical prediction of $1300/t in August, that lined up closely with his domain knowledge. He thought the prediction would turn out close to the target. Just a month later, though, he saw what may turn out to be the largest per acre yield in Lake County history for Sauvignon Blanc. This means that a great deal of excess fruit was sold for lower prices late in the year, which will bring down the average price.
One might revise the predicted average price to $1200 per ton in light of the new information, or maybe even lower. Ideally, if I were doing this forecast for a client, we would have sat down and combined his anecdotal evidence with the statistical forecast to update the prediction just prior to harvest. In any case, I'm looking forward to seeing where prices fell this year.
One last note that is interesting to me. Buyers often like to restrict yields to certain levels, in order to maintain quality. I sometimes take issue with that. In fact, while the data is sparse, my research shows some evidence that Lake County Sauvignon Blanc actually achieves higher scores when average yields per acre are higher, which is contrary to the conventional wisdom. These findings are very far from being conclusive or statistically significant. Still, if yields are record-breaking this year, I'll be keeping an eye on the scores.