Fed paper finds ways to boost forecast performance

A study published by the Federal Reserve finds adding extra information can improve forecasts, even though simple models remain “hard to beat”.

The working paper Forecasting US inflation in real time explores highlights how “careful introduction” of additional information can make forecasts more accurate and more robust.

Chad Fulton and Kirstin Hubrich test whether additional macroeconomic variables, expert judgement or forecast combinations can improve projections of headline personal consumption expenditures inflation.

Three approaches seem to help, they find. First, combining different models can improve performance and “robustify” models, which is important in the presence of high uncertainty. Second, aggregating forecasts of inflation components can improve performance compared to forecasting the aggregate directly.

Third, the authors find the “large information set” available to professional forecasters and Fed staff can “substantially improve” forecasting performance, “especially at short horizons”. They suggest this means “multivariate models, including those capable of handling large datasets, can play an important role in inflation forecasting”.