Inversion of the yield curve has come to be viewed as a leading recession indicator. Unsurprisingly, some recent instances of inversion have attracted attention from economic commentators and policymakers about possible impending recessions. Using a variety of time series models and recent innovations in econometric method, this paper conducts quasi-real-time forecasting exercises to investigate whether the predictive capability of the yield curve extends to forecasting economic activity in general and whether removing the term premium component from yields affects forecast accuracy. The empirical findings for the US, Australia, and New Zealand show that forecast performance is not improved either by augmenting simplistic models with information from the yield curve or by making such a decomposition of yields. Results from similar research exercises in previous work in the literature are mixed. The results of the present analysis suggest possible explanations that reconcile these conflicting results.