A Markov-Switching Model of the Unemployment Rate: Working Paper 2022-05

By Michael McGrane

The unemployment rate has asymmetric dynamics: It increases rapidly in recessions and falls gradually in expansions. The Congressional Budget Office developed a Markov-switching model to help incorporate these dynamics into macroeconomic projections and cost estimates that require simulations of the national unemployment rate. The model produces simulations that match observed asymmetric business-cycle dynamics at a rate consistent with historical data. I also show that indirect duration dependence, in which transition probabilities are a function of the unemployment gap, creates significant distortions for statistical tests of duration dependence in the business cycle. I present evidence that the benchmark Markov-switching model produces forecasts superior to those of a simpler model with constant transition probabilities, in addition to the linear version of the model. Finally, I make adjustments to the model to account for the unique split between permanently separated unemployment and temporarily separated unemployment in the pandemic recession and recovery.