Scalable Bayesian Estimation in the Multinomial Probit Model

The multinomial probit (MNP) model is a popular tool for analyzing choice behavior as it allows correlation between alternatives. Because current specifications employ full covariance matrix of the latent utilities alternatives, they are not scalable to large number This article proposes factor structure on matrix, which makes sets. main challenge in estimating this that parameters require identifying restrictions. We identify by trace-restriction imposed through reparameterization structure. specify interpretable prior distributions and develop an MCMC sampler parameter estimation. proposed approach significantly improves performance sets relative existing MNP specifications. Applications purchase data show economic importance including alternatives consumer analysis.