The Illusion of Oil Return Predictability: The Choice of Data Matters!

Abstract

Previous studies document statistically significant evidence of crude oil return predictability by several forecasting variables. We suggest that this evidence is misleading and follows from the common use of within-month averages of daily oil prices in calculating returns used in predictive regressions. Averaging introduces a bias in the estimates of the first-order autocorrelation coefficient and variance of returns. Consequently, estimates of regression coefficients are inefficient and associated t-statistics are overstated, leading to false inference about the true extent of return predictability. On the contrary, using end-of-month data, we do not find convincing evidence for the predictability of oil returns. Our results highlight and provide a cautionary tale on how the choice of data could influence hypothesis testing for return predictability.