Understanding survey-based inflation expectations

This paper examines the behavior of inflation expectations in the United States. After documenting deviations from rationality in survey-based inflation expectations, I apply a model selection algorithm, boosting, to the inflation expectations of households and professionals. The algorithm builds a regression-like model of expected inflation using a large panel of macroeconomic data as possible covariates. The algorithm achieves a very strong fit in-sample, and finds that the inflation expectations of households correlate with different macroeconomic variables from the expectations of professionals. However, it is difficult to exploit the predictability of inflation expectations in order to improve forecasts of the realized inflation. Published by Elsevier B.V. on behalf of International Institute of Forecasters.