Predicting US Recessions: A Dynamic Time Warping Exercise in Economics

Abstract

Dynamic Time Warping (DTW) is a widely used algorithm in speech recognition for measuring similarity between two time series. This non-parametric technique overcomes the problems associated with Pearson's correlation coefficient by allowing a non-linear mapping of one sequence to another obtained through the minimization of the distance between the two. Despite its superiority as a similarity measure, its application in the field of economics is almost non-existent. This paper seeks to fill this gap in the economics literature by providing a self-contained description of the method. In addition, it demonstrates an application of DTW to the prediction of recessions in US using Treasury term spread data. The exercise shows that DTW is successful in predicting the recessions of both 1999 and 2007. The predictions are stronger when asymmetric step-pattern is adopted. Also, compared to other non-parametric methods, DTW raises significantly fewer false signals of recessions. Finally, DTW concludes that given the current state of the economy there is a zero probability of a recession in the next one year.

Keywords: Dynamic Time Warping, similarity measure, non-parametric method, forecasting recession

JEL Classification: C01, C14, C53, E44