Formal and Heuristic Model Averaging Methods for Predicting the US Unemployment Rate

We consider a logistic transform of the monthly US unemployment rate. For this time series, a pseudo out-of-sample forecasting competition is held between linear and nonlinear models and averages of these models. To combine predictive densities, we use two complementary methods: Bayesian model averaging and optimal pooling. We select the individual models combined by these methods with the evolution of Bayes factors over time. Model estimation is carried out using Markov chain Monte Carlo algorithms and predictive densities are evaluated with statistical tests and log scores. The sophisticated averages of linear and nonlinear models turn out to be valuable tools for predicting the US unemployment rate in the short-term