Bayesian nonparametric panel Markov-switching GARCH models

TL;DR

This paper develops a Bayesian nonparametric panel Markov-switching GARCH model that captures regime-dependent volatility clustering and cross-sectional heterogeneity, effectively identifying latent clusters among assets.

Contribution

It introduces a novel hierarchical Bayesian framework with nonparametric priors for clustering in Markov-switching GARCH models, addressing high-dimensional parameter spaces.

Findings

Successfully recovers true parameters and group structures in simulations.

Identifies 2 clusters in the first regime and 3 in the second in empirical data.

Demonstrates the model's effectiveness on financial asset data.

Abstract

This paper introduces a new model for panel data with Markov-switching GARCH effects. The model incorporates a series-specific hidden Markov chain process that drives the GARCH parameters. To cope with the high-dimensionality of the parameter space, the paper exploits the cross-sectional clustering of the series by first assuming a soft parameter pooling through a hierarchical prior distribution with two-step procedure, and then introducing clustering effects in the parameter space through a nonparametric prior distribution. The model and the proposed inference are evaluated through a simulation experiment. The results suggest that the inference is able to recover the true value of the parameters and the number of groups in each regime. An empirical application to 78 assets of the SP\&100 index from $6^{th}$ January 2000 to $3^{rd}$ October 2020 is also carried out by using a two-regime Markov switching GARCH model. The findings shows the presence of 2 and 3 clusters among the constituents in the first and second regime, respectively.