Title: Bayesian nonparametric density autoregression with lag selection
Authors: Matthew Heiner - Brigham Young University (United States) [presenting]
Athanasios Kottas - University of California at Santa Cruz (United States)
Abstract: The aim is to develop a Bayesian nonparametric autoregressive model applied to estimate transition densities exhibiting nonlinear lag dependence flexibly. The approach is related to Bayesian density regression using Dirichlet process mixtures, with the Markovian likelihood defined through the conditional distribution obtained from the mixture. This results in a Bayesian nonparametric extension of a mixtures-of-experts model formulation. We illustrate and explore inferences available through the base model by fitting to synthetic and real-time series. We then explore model extensions to include global and local selection among a pre-specified set of lags, and modifications to the kernel weight function to accommodate heterogeneous dynamics. We also compare transition density estimation performance for alternate configurations of the proposed model.