A0414
Title: A Bayesian group sparsity approach to multiple structural breaks of an AR(p) process
Authors: Yi-Chi Chen - National Cheng Kung University (Taiwan) [presenting]
Kuo-Jung Lee - National Cheng Kung University (Taiwan)
Abstract: A Bayesian group sparsity approach is introduced for modeling non-stationary time series with structural breaks in autoregressive (AR) processes. Unlike conventional methods, this approach treats structural breaks as unknown parameters inferred directly from the data. By embedding break detection within a Bayesian model selection framework, it jointly estimates breakpoints and AR structures, ensuring a fully integrated, data-driven inference process. A key contribution is the unification of break detection and parameter estimation through hierarchical Bayesian variable selection techniques, identifying significant regime changes and within-segment AR structures. The group-wise Gibbs sampling (GWGS) method facilitates posterior sampling while maintaining computational feasibility, making it suitable for macroeconomic and financial time series analysis. Simulation studies confirm the method's effectiveness in detecting structural changes and estimating regime-specific dynamics. An empirical application to economic time series demonstrates its ability to identify regime shifts and capture evolving economic conditions. This approach provides a robust tool for understanding regime-dependent economic dynamics, with significant implications for macroeconomic forecasting, risk assessment, and policy evaluation.
