Can Style Momentum Be Optimized?

• Momentum is a well-documented investment strategy used across assets, industries and asset classes, as well as style factors.
• We combined past style performance with the risk and relationships of styles, to create a mean-variance optimized style-timing strategy.
• The optimized style-timing strategy generated an information ratio of 2.9 in a global universe of 19 styles (from January 1995 to October 2022), while a simple strategy generated a ratio of 1.4.

Momentum in style-factor performance suggests there could be value in style timing, but simple strategies for harnessing style momentum often only consider past performance. This is usually accomplished by going long the top performers and shorting the negative performers, while disregarding risk (variance) and relationships (covariance) between styles. Here we present an optimized style-timing strategy that accounts for both past performance and the variance-covariance of styles.

Construction methodology and performance

We used all 25 styles from the MSCI Global Equity Factor Trading Model and calculated past performance of each style over the last 12 months¹ as our forecast of alpha for each style. We then constructed an optimized long-short portfolio treating the styles as assets to evaluate the strategy. This minimized the predicted volatility of the resulting portfolio, while establishing unit exposure to the signal.

Cumulative monthly returns for the optimized and simple strategies are shown below. For comparison, we also detail performance of a simple strategy that does not consider risk, as it simply weighs styles by their previous 12-month return (all strategies were rebalanced monthly).

Cumulative return of the strategies

Cumulative returns for the optimized and simple style-momentum strategies with global-trading-model styles. Returns are scaled to 1% annualized volatility. January 1995 to October 2022.

We also list the information ratios (IR) of the optimized strategy, simple strategy and best-performing single style in the global, U.S. and Europe style universes in the table below.²

Information ratios for style universes

Information ratio (IR) is defined by annualized return divided by annualized risk and was calculated over the period from January 1995 to October 2022.

The IR of the optimized strategy with global long-term styles was 2.9, which was higher than any individual style and more than twice as high as the 1.4 of the simple strategy. We observed the same pattern in all the style universes we examined.

Which styles contributed most?

Then we examined the performance contribution of the optimized strategy relative to global trading styles. Almost 15% of the performance was attributed to the machine-learning (ML) factor. The strategy generally favored factors with high IR, such as the ML factor, or factors that could be hedged with other high-IR factors. All factors except beta and liquidity contributed positively to the performance (as detailed below).

Factors’ performance contribution

Contributions to the cumulative return of the optimized style-momentum strategy for global trading styles. Contribution’s total 100%. January 1995 to October 2022.

The optimized portfolio in detail

To further study the optimized portfolio, we show the alpha and weights for the global trading styles with largest positive and negative alphas in the table below. As detailed, the weights were not simply proportional to the alpha. For example, although residual volatility had a significantly negative alpha, it was assigned only a small relative weight because it is one of the riskiest factors. Similarly, momentum had the highest positive alpha but nearly zero weight, because of its high risk, while the ML factor had only the fifth-highest alpha but the largest weight.

Factor alpha and optimized portfolio weights

Alphas based on trailing-12-month return and weights on global trading styles in the optimized style-momentum strategy. As of Sept. 30, 2022.

We also examined the weights of the optimized strategy over time, and they are shown in the contour map below. We see that stock crowding and the ML factor had consistently large weights, while many traditional styles, such as momentum, residual volatility and earnings yield, had small relative weights. The sign of the weight was generally aligned with the sign of the risk premium for a given style. For example, styles with negative risk premia such as residual volatility, earnings variability and short interest typically received negative weight in the optimized strategy. The magnitude of the weight given to a style was inversely related to its risk and was also a function of how much that style could hedge the risks of other styles.

Optimized strategy weights

Average weight of global trading styles each year in the optimized strategy. Styles are ordered by average weight over the entire period of January 2012 to October 2022.

The efficacy of risk awareness

The elusive goal of style timing can be difficult because of the low predictability of returns. We leveraged the much greater predictability of risks and correlations between styles and created a strategy that outperformed an alternative that only considered return and ignored risk. This is an example of the benefit of being risk aware.

¹We use the pure daily factor return from the factor model as the style-factor performance. The trailing-12-month return is defined similarly to how we define the stock-level momentum style factor: trailing-12-month return with a six-month half-life.

²The MSCI Global and USA Long-Term Equity Factor Model universes contain 19 styles; the MSCI Global and USA Equity Factor Trading Models contain 25 styles; the MSCI Europe Long-Term Equity Factor Model contains 20 styles; and the MSCI Europe Equity Factor Trading Model contains 26 styles.

Further Reading

Markets in Focus: Is the Beta Pendulum an Edge or Hedge?

Inflation Sensitivity and Equity Returns

Crowd Control for Fund Managers