Forecasting US Inflation in Real Time

We analyze real-time forecasts of US inflation over 1999Q3–2019Q4 and subsamples, investigating whether and how forecast accuracy and robustness can be improved with additional information such as expert judgment, additional macroeconomic variables, and forecast combination. The forecasts include those from the Federal Reserve Board’s Tealbook, the Survey of Professional Forecasters, dynamic models, and combinations thereof. While simple models remain hard to beat, additional information does improve forecasts, especially after 2009. Notably, forecast combination improves forecast accuracy over simpler models and robustifies against bad forecasts; aggregating forecasts of inflation’s components can improve performance compared to forecasting the aggregate directly; and judgmental forecasts, which may incorporate larger and more timely datasets in conjunction with model-based forecasts, improve forecasts at short horizons.