How regional business cycles diffuse across space and time : evidence from a Bayesian Markov switching panel of GDP and unemployment in Poland

Abstract

We investigate the regional business cycles at NUTS-3 granularity in Poland (N=73) using two variables in parallel: GDP dynamics and unemployment. The model allows for both idiosyncratic business cycle fluctuations in a region in the form of 2-state Markov chain, as well as spatial interactions with other regions. The posterior distribution of the parameters is simulated with a Metropolis-within-Gibbs procedure. We find that the regions can be classified into business cycle setters and takers, and this classification exhibits a high degree of overlap with the line of division between metropolitan versus peripheral regions. We also find that, under large N, the fixed-effects methods, as proposed in the previous literature, are vulnerable to both identification issues and (MCMC) convergence problems, especially with short T, which is of critical importance in GDP on the considered spatial granularity level.