Chapter 2 Combining Predictors & Combining Information in Modelling: Forecasting US Recession Probabilities and Output Growth

Abstract

We consider combining forecasts versus combining information in modelling, in the context of predicting regime probabilities and output growth in the US. The simple models whose forecasts we combine each use one of the leading indicators that comprise the Conference Board Composite Leading Indicator as explanatory variables. Combining this information set in modelling is achieved by using a relatively simple model selection strategy. For predicting output growth, our findings support pooling the forecasts of the single-indicator models, while the results are more mixed for predicting recessions and recession probabilities. Our results are not affected by allowing for non-linearities in the output growth regressions, although issues to do with the vintages of data used to estimate the models and evaluate the forecasts are important.