Stochastic prediction in multinomial logit models

Stochastic prediction in multinomial logit models

0.00 Avg rating0 Votes
Article ID: iaor20012583
Country: United States
Volume: 46
Issue: 8
Start Page Number: 1137
End Page Number: 1144
Publication Date: Aug 2000
Journal: Management Science
Authors: ,
Keywords: marketing, statistics: sampling, stochastic processes
Abstract:

It is standard practice to form predictions from multinomial logit models by ignoring the estimation error associated with the parameter estimates and solving for the predicted endogeneous variable (market share) in terms of the exogenous variables and the point estimates of the parameters. It has long been recognized in the econometrics literature that this type of nonstochastic prediction, which ignores the sampling distribution of the parameter estimates, leads to incorrect inferences about the endogenous variable. We offer a simulation-based approach for approximating the exact stochastic prediction. We show that this approach provides very accurate approximations with minimal computation time and would be easy to implement in industrial applications.

Reviews

Required fields are marked *. Your email address will not be published.