Macroeconomic Regime Identification Using a Two-Step Approach With Independent Component Analysis and Hidden Markov Models

Abstract

Hidden Markov models are often used to identify different regimes. However, in a multivariate setting, correlations between variables may skew the results, leading to potentially flawed analyses. This paper proposes a two-step approach to better identify hidden regimes in macroeconomic time series. In the first step, independent components are extracted from nine macroeconomic time series using second order blind identification (SOBI). In the second step, the independent components are used in a hidden Markov model to identify macroeconomic regimes. The results from the two-step process show increased regime persistence compared with a pure hidden Markov model, suggesting clearer identification of regimes when dealing with correlated time series. The paper also introduces two new measures of the quality of regime classification.