A0227
Title: Identifying business cycles in real time with quantile dynamic factor Markov switching models
Authors: Jeremy Piger - University of Oregon (United States) [presenting]
Abstract: The global pandemic created extreme swings in macroeconomic data that complicate statistical analysis. Simply dummying out these observations is not a good option, as this would eliminate macroeconomic objects of interest, such as recessions. The existing literature has focused on model augmentation in the form of time-varying volatility (e.g., stochastic volatility, ARCH, GARCH, etc.) Such approaches have been taken to workhorse models in empirical macroeconomics, such as VARs and dynamic factor models. An alternative approach is evaluated for Markov-switching models, which are a popular class of models designed to identify recessions in economic data. Specifically, the performance of dynamic factor Markov switching (DFMS) models is investigated, which are specified in terms of conditional quantiles of recessions rather than conditional means. It is shown that these quantile DFMS models are capable of capturing recessions, both retrospectively and in real time, while not being too strongly influenced by extreme recessions and expansions, such as those observed during and following the March-April 2020 recession.
