Markov Switching Panel with Network Interaction Effects

Abstract

The paper introduces a new dynamic panel model for large data sets of time series, each of them characterized by a series-specific Markov switching process. By introducing a neighbourhood system based on a network structure, the model accounts for local and global interactions among the switching processes. We develop an efficient Markov Chain Monte Carlo (MCMC) algorithm for the posterior approximation based on the Metropolis adjusted Langevin sampling method. We study efficiency and convergence of the proposed MCMC algorithm through several simulation experiments. In the empirical application, we deal with US states coincident indices, produced by the Federal Reserve Bank of Philadelphia, and find evidence that local interactions of state-level cycles with geographically and economically networks play a substantial role in the common movements of US regional business cycles.