A1445
Title: Bayesian copula factor autoregressive models for time series mixed data
Authors: S Yaser Samadi - Southern Illinois University Carbondale (United States) [presenting]
Samira Zaroudi - The City University of New York- John Jay College of Criminal Justice (United States)
Hadi Safari-Katesari - Hostos College, City University of New York (United States)
Abstract: The aim is to propose a Bayesian copula factor autoregressive (BCFAR) model for analyzing time series mixed data, accommodating main effects and interactions. The main motivation is to infer dynamic interactions between macroeconomic variables and stock market indices. The BCFAR model assumes conditional independence and applies latent factors in both response time series and high-dimensional mixed-type covariates in quadratic regression using copula functions. To complement this, a simpler time series Bayesian factor regression (TS-BFR) model is introduced, tailored for continuous Gaussian multivariate time series. Both models build on the quadratic autoregression (QAR) framework, employ latent factors for efficient dimension reduction, and capture main effects and interactions of covariates by integrating latent variables into the response. For computational efficiency, a semiparametric time series extended rank likelihood is used for explanatory-variable margins in the BCFAR model, reducing parameters and ensuring fast computation. To estimate latent factors and parameters, flexible Bayesian algorithms are designed, employing Metropolis-Hastings (MH) and forward filtering backward sampling (FFBS) within Gibbs sampling. The effectiveness of these methods is shown through simulation studies, and the approach is further validated with a macroeconomic dataset.
