A1101
Title: Bayesian dynamic factor models for high-dimensional matrix-valued time series
Authors: Wei Zhang - Johns Hopkins University (United States) [presenting]
Abstract: High-dimensional matrix-valued time series are of significant interest in economics and finance, with prominent examples including cross-regional macroeconomic panels and firms' financial data panels. The aim is to introduce a class of Bayesian matrix dynamic factor models that utilize matrix structures to identify more interpretable factor patterns and factor impacts. The model accommodates time-varying volatility, adjusts for outliers, and allows cross-sectional correlations in the idiosyncratic components. For model comparison, an importance-sampling estimator is employed based on the cross-entropy method to inform decisions regarding: (1) the optimal dimension of the factor matrix; (2) the appropriate factor structure--whether vector-valued or matrix-valued; and (3) the suitability of an approximate versus exact factor model. Through a series of Monte Carlo experiments, the properties of the factor estimates and the performance of the marginal likelihood estimator in correctly identifying the true model are shown. Applying the model to a macroeconomic dataset and a financial dataset, its ability is demonstrated in unveiling interesting features within matrix-valued time series.
