A1462
Title: Bayesian group-shrinkage based estimation for panel vector autoregressive models with mixed frequency data
Authors: Nilanjana Chakraborty - Indian Institute Of Management Udaipur (India) [presenting]
Kshitij Khare - University of Florida (United States)
George Michailidis - University of California, Los Angeles (United States)
Abstract: Panel vector auto-regressive (VAR) models are effective tools for capturing temporal relationships between a set of variables (e.g., macroeconomic indicators of an economy) while accounting for interdependencies between a set of entities (e.g., market sectors or whole economies). In the case of modelling macroeconomic data, this challenge is often further accentuated by the fact that some variables are observed at different frequencies. Existing Bayesian approaches for linking entity-specific VAR models aim to fuse relevant coefficients of the corresponding VAR transition matrices to a common value. A balanced and less stringent Bayesian approach is developed for mixed frequency panel VAR models that use group shrinkage prior distributions to borrow strength across entities but, at the same time, provide enough flexibility for entity-specific idiosyncrasies. A key novel feature is the ability to incorporate and learn the interdependence structure between various entities through an inter-entity covariance (matrix) parameter. The performance of the proposed methodology is evaluated both on synthetic data and on two economic applications; the first focuses on employment indices across neighboring US states, and the second on macroeconomic indicators of tightly integrated European economies. Finally, the theoretical properties of the modelling approach are also established.
