Robust Model Fitting for Two-Part Model Within Bayesian Semiparametric Framework: Variational Approach

Accepted - May 2024

Abstract

A Bayesian semiparametric fitting to guard against the distribution deviations is developed for two-part model (TPM). The parametric assumption on the continuous process is relaxed to be the normal mixture of Dirichlet process (DP). By taking advantage of the sparseness of DP, the semiparametric fitting automatically adjusts the random weights to accommodate the heterogeneity such as the skewness, multimodality and/or extreme observations. We develop a variational Bayesian inference procedure. A proper variational density is constructed to approximate posterior. Our empirical results show that the proposed method is more robust against the distributional departures and can efficiently improve model fitting.