Momentum, the tendency of past winner stocks to outperform past loser stocks over the next several months, is one of the most well-documented and well-researched asset pricing anomalies. (Page 1)
A momentum premium has been found for both cross-sectional momentum and time-series momentum (trend). (Page 2)
A slow time-series momentum (TSMOM) strategy tends to outperform a fast TSMOM strategy when market volatility is low. The opposite tends to occur when volatility is high. This pattern of relative performance exists in most global equity markets, including both developed and emerging markets. And both TSMOM strategies can suffer from turning points in market direction—reversals in trend from uptrend to a downtrend or vice versa. Strategies based on slower signals, such as the 12-month TSMOM, take more time to turn and therefore can be more greatly affected by turning points, while strategies based on faster signals, such as one-month returns, are more reactive and suffer less from turning points. (Page 2)
The fact that research, including the 2017 study “Tail Risk Mitigation with Managed Volatility Strategies,” has found that past volatility largely predicts future near-term volatility—volatility is persistent (it clusters) with high (low) volatility over the recent past, tending to be followed by high (low) volatility in the near future—and that volatility is negatively correlated with returns has led to the development of strategies that scale volatility inversely to past realized volatility. (Page 2)
Another strategy to address the problem of reversals is to use a blend of short-term, intermediate-term and long-term signals—an effective combination of signals from two or more speeds can take advantage of different market cycles and reduce exposure to the downside associated with turning points. (Page 3)
This led the authors to explore whether a decision tree (a supervised learning technique) to determine the optimal switching mechanism between fast and slow trend signals in different volatility regimes could find a more efficient way of choosing between the fast or slow signal in order to potentially improve performance. (Page 4)
•Fast tends to outperform when one-month S&P 500 Index volatility is above a volatility threshold (17 percent), and slow tends to perform better otherwise. (Page 4)
•The volatility threshold model successfully avoided catastrophic drawdowns during the most volatile periods. Although the 50/50 mix of slow and fast also did relatively well in preventing drawdowns, it was inactive, with a neutral position 30 percent of the time. (Page 4)
•While both fast and slow delivered positive volatility-timing alphas regardless of regime, slow had much higher market-timing alpha than fast in low-volatility regimes, whereas fast had much higher market-timing alpha than slow in high-volatility regimes—the slow signal predicts market movement better in low-volatility regimes, whereas fast signal performs better in high-volatility regimes. (Page 5)
Their findings led Cheng, Kostyuchyk, Lee, Liu and Ma to conclude that a dynamic momentum strategy that varies its speed based on volatility regimes delivers better out-of-sample risk-adjusted performance with less tail risk. (Page 5)
A large body of evidence demonstrates that momentum, including time-series momentum (trend following), has improved portfolio efficiency. Research has found that there are a few ways to improve on simple trend-following strategies. (Page 5)
The powerful tools and the easy access to data now available to researchers create the risk that machine learning studies will find correlations that have no causation and thus the findings could be nothing more than a result of torturing the data. To minimize that risk, it is important that findings not only have rational risk- or behavioral-based explanations for believing the patterns identified will persist in the future, but they also should be robust to many tests. (Page 5)
In closing, the following advice is offered. Investors must take note of the fact that, like all risk, or factor-based, strategies, time-series momentum strategies have gone through long periods of underperformance (see Wes Gray’s excellent article demonstrating this point here). (Page 6)