In recent years, U.S. food prices have surged amid supply chain disruptions, labor shortages, rising input costs, and global shocks such as COVID 19 and the war in Ukraine. Because of this, there is renewed interest in forecasting food price inflation by economists, policymakers, and agribusiness firms alike. This study leverages machine learning methods and the availability of large economic databases to improve forecasts of U.S. food price inflation. We show that forecasts generated from machine learning models incorporating a large set of covariates—particularly using a modified version of the FRED-MD dataset—are more accurate than univariate benchmark forecasts over several alternative horizons. Notably, forecast combination strategies that combine forecasts generated from different machine learning methods as well simplistic univariate models often outperform forecasts generated from individual models and therefore warrant greater attention.
Publications