Depending on portfolio turnover, execution costs often have significant negative impact on a portfolio’s or a trading strategy’s profitability. There is often a trade-off between paper profits from a trading strategy and portfolio turnover: lower-frequency strategies with lower turnover tend to have lower paper profits (without considering costs) and higher-frequency strategies with higher turnover tend to have higher paper profits. For example, traditional value signals such as book to price ratio have very low turnover—often less than 25% annually—and short-term mean reversion strategies have much higher turnover—often as high as 25% daily. For strategies with high turnover, reducing trading costs is crucial. A strategy with a daily turnover of 25% and an assumed trading cost of 10 basis points requires 12.6% annualized before-cost return to break even. High trading costs are the reason why some apparent anomalies are hard to monetize and continue to exist.
Our overall approach has been to choose factors that have good performance and consistency in predicting portfolio returns balanced against turnover.
By design, some factors change slowly from one period to the next, whereas others change rapidly. For example, the size factor stays stable over months or even years. Therefore, the single-factor portfolio constructed using size factor has low turnover. In contrast, values of a short-term reversal factor change rapidly from day to day (if the signal works, recent losers become winners and recent winners become losers). Therefore, a single-factor portfolio constructed using a short-term reversal factor has high turnover. Since the turnover depends on how we use a factor to construct a portfolio, there is no standard approach to estimate turnover. One approach is to directly look at a zero-investment portfolio, e.g., the top quintile minus the bottom quintile, to calculate portfolio turnover. A different approach, independent of portfolio construction, measures the serial correlation of factor scores: the correlation between factor scores of the stocks at t and the factor scores of the same stocks at t + 1. Higher correlation means that factor scores are more stable over time and indicates lower factor turnover.
Factors with high turnover need to be, and sometimes are, compensated with higher predictive power. If we have a reasonable estimation of trading costs, they can be directly incorporated to estimate after-cost returns and information ratios. Mediocre predictable power and high turnover, however, do not automatically make the factor a poor signal. Investors usually use multiple factors in the portfolio construction and the final turnover depends on the interaction of the selected factors.
The hysteresis approach is a valuable tool for reducing turnover as well. After we enter a long position when the stock moves into the top quintile, we do not necessarily exit the position when it moves out of the top quintile. Instead, we only exit the position if the stock falls below the 60th percentile. Although the expected alpha is lower when the stock falls from the top quintile to the fourth quintile, holding the existing long position does not require extra trading and does not incur trading costs. Similarly, after we enter a short position when the stock moves into the bottom quintile, we only exit the position if the stock moves above the 40th percentile. By using different thresholds for entry and exit, we reduce the turnover of fast-moving signals and achieve a better trade-off between raw returns and trading costs.
This is just one of hundreds of active management approaches we use at ActiveAllocator.com to create better portfolios.