Abstract
The generalized multinomial logit (GMNL) model accommodates scale heterogeneity to the random parameters logit (RPL) model. It has been often used to study people's preferences and predict people's decisions in many areas, such as health economics, marketing, agricultural studies, transportation research and public policy. However, there are few works studying the efficiency of this model estimator and the corresponding estimation and prediction risks. In this paper, we use a frequentist model averaging (FMA) estimator to reduce the estimation and prediction risks of the GMNL model estimator. We show that the asymptotic squared error risk of the FMA estimator dominates that of the GMNL model estimator, and it is consistent with the results of our Monte Carlo experiments. The accuracy of the predicted choices is also higher based on the FMA estimates compared to the results based on the GMNL estimates. In the empirical analyses, using the FMA estimator improves the percentage of correct predicted choices by 10% compared to the results with GMNL estimates. This paper provides a more efficient alternative to the GMNL model to capture people's preferences and predict people's choices.