Return Predictability and Market-Timing: A One-Month Model

Vol. 17, No. 3, 2019

Blair Hull, Xiao Qiao and Petra Bakosova
Our paper addresses an important question in portfolio management: tactical asset allocation. Asset allocation is a central theme in investment management, and a deeper understanding of when and how much to increase or decrease stock market exposure is important for many asset managers.We investi- gate tactical asset allocation through the lens of return predictability. In doing so, we provide a bridge between a large body of academic work and the investment management industry by demonstrating the real-world implications of predictability research.

In the academic literature, there has been lots of work on econometric models of return predictability, with emphasis on statistical results. Less work has focused on the portfolio management implications of predictability. In practice, the investment management aspect of return predictability may be more interesting than econometric analyses.We connect academic research and practice through the follow-ing research question: Can we form a viable trading strategy based on statistical models of stock return predictability?

We find that the answer is yes. Our paper illustrates how to construct such a trading strategy. Using weighted least squares (WLS) with stepwise variable selection, we combine 15 variables to forecast the one-month ahead market excess returns. We then propose a one-month market-timing strategy based on our statistical model. From 2003 to 2017, our strategy earned 16.6% annual returns with a 0.92 Sharpe ratio and a maximum drawdown of 20%. For the same period, the S&P 500 had 10% annual returns with a Sharpe ratio of 0.46 and a maximum drawdown of 55%. We also consider combining our one-month strategy with Hull and Qiao’s (2017) six-month market-timing strategy. The combined strategy had 15% annual return, Sharpe ratios of 1.12, and a maximum drawdown of 14%.We are the first to investigate the economic implications of stock market predictability at the one-month horizon, and we illustrate the practical benefits of model combination through combining two market-timing models.