Are Bond Returns Predictable with Real-Time Macro Data?
50 Pages Posted: 23 Jan 2018 Last revised: 1 Feb 2021
Date Written: April 30, 2020
We examine whether bond returns are predictable with real-time macro variables by using two supervised learning methods, scaled PCA (sPCA) and partial least squares (PLS) instead of the usual PCA. We find that the real-time sPCA and PLS factors can predict bond returns significantly both in- and out-of-sample, re-affirming recent studies on time-varying risk premia. We also find that the predictability is countercyclical and it yields substantial economic values to a mean-variance investor. Econometrically, we study the properties of the sPCA and PLS in a partially-relevant latent factor framework, providing insights on why they can extract relevant factors that contain predictive information about the target, and on conditions under which they can complement each other.
Keywords: Bond Return Predictability, Real Time Macro Data, Vintage, PCA, Big Data, Machine Learning
JEL Classification: C22, C53, G11, G12, G17
Suggested Citation: Suggested Citation