A1650
Title: Score-driven time-varying transition probabilities in a dynamic factor Markov-switching model
Authors: Bram van Os - Econometric Institute, Erasmus University Rotterdam (Netherlands) [presenting]
Dick van Dijk - Erasmus University Rotterdam (Netherlands)
Abstract: The business cycle is an important driver of many macroeconomic variables. The existing dynamic factor Markov-switching (DFMS) model has proven to be a powerful framework to measure the cycle. This model estimates the latent business cycle factor by exploiting the cross-sectional information in multiple observed variables. Furthermore, in line with the macroeconomic intuition of expansions and contractions phases in the cycle the evolution of the factor is allowed to be regime-dependent, with a hidden Markov process dictating the regime-switches. Allowing for time-varying parameters in univariate Markov-switching models has been found useful in the context of business cycle applications. In particular, previous literature has amounted substantial evidence that the assumption of time-invariant transition probabilities used may not be appropriate here. The authors enhance the DFMS model to allow for time-varying transition probabilities (TVTP) by combining the accelerated score-driven framework with a method for adding score-driven TVTP to a Markov-switching model, which they extend to the multivariate setting and where they allow for exogenous variables. In an empirical application using the four components of The Conference Boards Coincident Economic Index for the period 1959-2019, it is found that the proposed framework allows for superior dating of US business cycle peaks and troughs.
