Correcting sample selection bias with model averaging for consumer demand forecasting

Abstract

Sample selection bias exists in many consumer-level demand data. In this paper, we propose a new model averaging optimal correction (MAOC) method for correcting such bias. The averaged bias correction term is constructed from a set of candidate models to combat potential model uncertainty. The MAOC estimator is further proved to be asymptotically optimal in the sense of achieving the lowest possible mean squared error under mild regularity conditions. The simulation results demonstrate the superiority of MAOC estimator over many peer methods. In the empirical exercises, we study the movie open box office data and show that our MAOC method provides significant in-sample explanatory power and improves the out-of-sample performance as well. As the movie industry calls for more accurate box office predictions to control movie budgets, we believe our proposed method can help managerial decision making.