Does accounting for preference heterogeneity improve the forecasting of a random utility model? A case study

This paper examines the out-of-sample forecasting performance of several random utility models of recreation behavior. The application is salmon fishing on Lake Michigan in 1996 and 1997, with the quality of fishing (as measured by catch rates) better in 1997 than in 1996. The models examined are a logit model, two random parameters logit (RPL) models, and a latent class logit (LCL) model. Results indicate that the RPL and LCL models forecast equally well, and by at least one measure the logit model outperforms them both. Results also raise the possibility that recreational fishing decisions have a dynamic component