Wednesday, October 20, 2004

The Fair Model

Jayson Javitz has a post on the Fair model, an extremely simple equation used to predict the incumbents popular vote percentage based on simple economic statistics. It has fit well in the past and is currently predicting a landslide for President.

I see an econometric problem with using it to say Bush is in good shape now, and that is the following: We have more information at this time than just Fair's inputs: growth rate of GDP, inflation, and how long we have been growing. We also have the polls themselves. Presumably, an optimal prediction function would use all the information available. To convince me that the polls are to be discarded, one would have to show that Fair's model predicts better (or as well as) a function that used two week out polling data in addition to Fair's inputs. If the prediction function didn't do any better with the polling data as it did without, then, yes, the polling data is useless. On the other hand, Fair's model might do well precisely because the economic data predicts the two week out polling data which then determines the election. In this scenario, the Fair model would still fit the past data and be useless for predicting this election.