Predicting Presidential Elections with Equally Weighted Regressors in Fair's Equation and the Fiscal Model

Abstract

Three-decade-old research suggests that although regression coefficients obtained with ordinary least squares (OLS) are optimal for fitting a model to a sample, unless the N over which the model was estimated is large, they are generally not very much superior and frequently inferior to equal weights or unit weights for making predictions in a validating sample. Yet, that research has yet to make an impact on presidential elections forecasting, where models are estimated with fewer than 25 elections, and often no more than 15. In this research note, we apply equal weights to generate out-of-sample and one-step-ahead predictions in two sets of related presidential elections models, Fair's presidential equation and the fiscal model. We find that most of the time, using equal weights coefficients does improve the forecasting performance of both.