Bond Risk Premia with Machine Learning
75 Pages Posted: 26 Aug 2018 Last revised: 22 Jan 2019
Date Written: September 29, 2018
We propose, compare, and evaluate a variety of machine learning methods for bond return predictability in the context of regression-based forecasting and contribute to a growing literature that aims to understand the usefulness of machine learning in empirical asset pricing. The main results show that non-linear methods can be highly useful for the out-of-sample prediction of bond excess returns compared to benchmarking data compression techniques such as linear principal component regressions. Also, the empirical evidence show that macroeconomic information has substantial incremental out-of-sample forecasting power for bond excess returns across maturities, especially when complex non-linear features are introduced via ensembled deep neural networks.
Keywords: Machine Learning, Deep Neural Networks, Forecasting, Bond Returns Predictability, Empirical Asset Pricing, Ensembled Networks.
JEL Classification: C38, C45, C53, E43, G12, G17
Suggested Citation: Suggested Citation