On the Missing Disinflation Puzzle: A Data-Driven Approach

Orientador(a): Eduardo Zilberman

Co-orientador(a): Marcelo Medeiros

This paper examines the potential explanations for the Missing Disinflation Puzzle (MDP). We construct a data set containing only variables associated with the puzzle, and use of Machine Learning (ML) methods to compute estimates for U.S. Consumer Price Index inflation over the period of interest. These methods can handle large data sets, and perform variable selection. A model selection exercise using Model Confidence Set over pseudo-out-of-sample forecasts is proposed to assess forecasting performance and to analyze the variable selection pattern of these models. We analyze the variable selection performed by the best models and find evidence for explanations associated with different metrics for inflation expectations - in particular those linked to consumers’ surveys.