Predicting Corporate Bond Yield Term Structure

Abstract

This paper investigates the monthly out-of-sample predictability of corporate bond yield using BVAL curve values for the period April 2011 to June 2020 for three tickers of JPM, GS and IBM and a rolling time window. Multiple univariate and multivariate models using statistical and machine learning approaches are developed and the best models are selected based on the performance measure. The univariate random forest model has a better performance when taking logarithm of forward rates as the explanatory variable. In contrast, the autoregressive integrated moving average models demonstrate a higher performance measure values when the independent variable is considered as forward-spot spreads.

Keywords: Corporate bond, Yield predictability, Financial time series, Regression, Machine Learning

JEL Classification: C32, C53, G12