Yield Spread Selection in Predicting Recession Probabilities: A Machine Learning Approach

TL;DR

This paper examines whether machine learning can enhance recession prediction using yield spreads, but finds limited improvement over traditional methods due to estimation errors.

Contribution

It introduces a machine learning approach to select optimal yield spread pairs for recession forecasting and evaluates its effectiveness.

Findings

Machine learning does not significantly outperform traditional spread use.

The 10-year--three-month spread remains a robust predictor.

Estimation error limits the gains from optimized spread selection.

Abstract

The literature on using yield curves to forecast recessions customarily uses 10-year--three-month Treasury yield spread without verification on the pair selection. This study investigates whether the predictive ability of spread can be improved by letting a machine learning algorithm identify the best maturity pair and coefficients. Our comprehensive analysis shows that, despite the likelihood gain, the machine learning approach does not significantly improve prediction, owing to the estimation error. This is robust to the forecasting horizon, control variable, sample period, and oversampling of the recession observations. Our finding supports the use of the 10-year--three-month spread.