Forecasting commodity prices returns: The role of partial least squares approach

Abstract

This paper put forwards a novel aligned technical index, which eliminates a common noise component in technical indicators by employing the partial least squares (PLS) method, to investigate the predictability of commodity price returns.Our results show that the novel aligned technical index delivers powerful predictive ability statistically and economically under both in-sample and out-of-sample conditions. Specifically, the novel aligned technical index generates in-sample R2of 10.74%, much larger than the economically significant R2 benchmark of 0.5%. Meanwhile, the out-of-sample ROS2 of our new aligned technical index reaches 11.46%, while the highest ROS2 generated by the eight macroeconomic variables is only 5.26%. Furthermore, we also execute a range of robustness tests with respect to the business cycle, subsample periods, different risk aversion coefficients and the transaction cost. The new technical index is consistent and stable to various robustness checks.