Abstract: We evaluate the ability of economic uncertainty measures to forecast recessions in real time. We find that including uncertainty increases the predictive power of both in sample and out-of-sample forecast models relative to a baseline set of financial variables. A nonlinear maximum transformation of uncertainty, which captures whether a measure exceeds its maximum over the past year, improves forecast performance for some measures. Adding a contemporaneous indicator like GDP growth alongside uncertainty yields additional predictive gains. Lastly, ex post Bayesian model averaging outperforms individual uncertainty models and ex ante factors of uncertainty generated using principal component analysis.
