On Testing Time Series Momentum Using Predictive Regressions

In studies of time series momentum (TSM), the Newey-West t-test has size distortion for linear predictive regression with excess returns because of non-stationarity, endogeneity due to correlated errors, and a lack of finite moments due to heavy tails. To solve these problems, we propose a new test that features log-returns, a model of the error correlations, and weighted least squares estimation. Simulations confirm the proper size and increased power of the new test. Empirically, we find weak support for TSM regardless of the predictor's time horizon and a different set of assets with TSM compared with using the Newey-West t-test.