Idiosyncratic Market Effects in Global Time Series Momentum

Abstract

Time series momentum is a pervasive explanatory factor in financial markets whereby a market's recent historical average return positively predicts future returns. Recent applications of statistical learning augment the traditional linear model definition with a more descriptive global nonlinear factor model. We extend the work of \citet{MoskowitzNonlinear} and demonstrate that models fit at the individual market level exhibit response curves that are distinct from the global factor model. We highlight the heterogeneity of forward market returns as a function of time series momentum and provide a framework that enables each feature-market pair to express its unique prediction behaviour. Empirical results demonstrate that, despite substantially smaller sample sizes, fitting nonlinear models to individual markets is a robust and often overlooked modelling approach that can result in a marginal improvement in out-of-sample performance.