Equity and Bond Comovements: A Machine Learning Perspective

Abstract

We study the comovements between stock and Treasury bonds from a machine learning perspective. Employing cutting-edge machine learning techniques and an extensive panel of characteristics, we assess the effectiveness of various machine learning models and identify the primary drivers of stock-Treasury correlation. All machine learning methods outperform traditional OLS regression, with dimension reduction techniques being the best-performing linear approaches and neural networks excelling among nonlinear methods. Stock illiquidity emerges as the most influential characteristic driving the negative stock-Treasury correlation, while two inflation-related measures are central to the positive stock-Treasury correlation. The positive stock Treasury correlation is closely tied to high inflation, whereas the subsequent negative correlation largely reflects a cross-market hedging dynamic.

Keywords: Stock-bond comovement, Machine Learning, Cross-market hedging

JEL Classification: G12, G17, P16, E44