Forecasting the U.S. Unemployment Rate: Another Look

Abstract

In �Forecasting the U.S. Unemployment Rate: Another Look� John Guerard, Han Xiao and Rong Chen replicate and extend the MZTT analysis for the 1959 to 2019 time period. The authors report out-of-sample one-step to twelve-step ahead monthly prediction performance of various models for the 1990-2019 period, using a no-change (random walk) model as a forecasting benchmark. Results obtained from this study include: (1) weekly unemployment claims are indeed a useful and statistically significant input in a transfer function model to forecast the unemployment rate; (2) the leading economic indicators time series is a statistically significant input in a transfer function model to forecast the unemployment rate; (3) a seasonal ARIMA (SARIMA) model {outperforms} the no-change benchmark {for all forecasting horizons}; (4) the SARIMA and transfer function models are statistically significantly better forecasting models than a null, or no-change, forecast, particularly in the Global Financial Crisis (GFC), 2008-2019 time period. Improved upon MZTT, this paper provides a set of analysis that serves as an updated benchmark for comparison of forecasting methods and approaches on unemployment prediction.