This paper investigates which method best forecasts U.S. inflation out-of-sample: time series models, regressions using real activity measures, term structure models, or survey-based measures. The research examines the forecasting power of four distinct approaches, including ARIMA models, Phillips curve-motivated regressions, various term structure specifications, and survey data. It also explores forecast combination methods. The findings consistently indicate that survey-based measures are superior to the other examined methods, with term structure specifications performing poorly.

This academic paper presents a rigorous empirical comparison of inflation forecasting methods. The study's findings, which highlight the superior predictive power of survey data over traditional econometric models and term structure specifications, are of significant interest to economists and policymakers. The research also contributes to the understanding of optimal forecast combination strategies, emphasizing the dominance of survey information. The consistent results across different methodologies underscore the practical utility of survey data in anticipating inflation trends.