Discussion on an approach for identifying and predicting economic recessions in real-time using time-frequency functional models

摘要

In this well-written and thorough manuscript, the authors creatively combine statistical methodologies to improve identification and prediction of recessions and expansions of the US economy. This is a timely topic, and one with concrete impact for financial analysts and market economists. The key statistical contribution is the introduction of the time-frequency spectrogram, which is based on the short-term Fourier transform, in collaboration with macroeconomic variables. Empirical orthogonal functions (EOFs) are used to reduce the dimensionality of the time-frequency spectrogram computed from two quarters of a daily index of NASDAQ returns preceding the time point. Variable selection of the EOFs as well as economic variables, for purposes of recession prediction, is conducted using Bayesian model averaging. The end result is a successful predictor and classifier of recessions.