It is hard to successfully time any investment. Whether it’s deciding to pull money out of the stock market, overweight foreign stocks or a particular sector, or trim duration risk, adjusting a portfolio based on expectations about the future can easily backfire because the future is hard to predict. Yet, there is an emerging body of research that suggests it is possible to successfully time exposure to factors like value, momentum, size, quality, and low volatility. While each of these factors has a good long-term record, they all go through cycles of underperformance. If timing really works, it could help mitigate this cyclicality, which is one of the biggest drawbacks to factor investing.
A healthy dose of skepticism is in order. Much of the research done thus far has come from practitioners, rather than academia, who work for asset managers with a vested interest in bringing new products to market. As with most financial research, data mining is also a risk because there are many variables researchers could have tested to find a predictive relationship that worked in sample but may not work out of sample. Even if there is a return benefit from factor-timing, implementing it reduces diversification relative to a static multifactor portfolio, which may outweigh the benefit. And it’s important to bear in mind that even if a timing signal works on average, it won’t always get the calls right. There is no pain-free way to beat the market. That said, factor-timing warrants serious review.
Factor-Timing Signals
A recent paper from BlackRock1 suggests that there are four types of factor-timing signals that work: valuation, momentum, economic regime indicators, and dispersion. The authors found that each of the four types of signals work well on their own and even better together. BlackRock does not currently have any factor-timing exchange-traded funds on the market, though it did launch a factor-timing model in September 2016 based on these insights.
The few shops that do offer factor-timing ETFs rely on indicators that broadly fit into one of these four categories. For instance, Oppenheimer Russell 1000 Dynamic Multifactor ETF (OMFL, listed in the U.S.) relies on a blend of traditional economic and market sentiment indicators to gauge the economic regime and time its factor exposures accordingly. Global X Adaptive U.S. Factor ETF (AUSF, listed in the U.S.) uses a contrarian performance signal, which is a type of value signal because assets that underperform tend to become cheaper and may be poised to do better in the future. PIMCO RAFI Dynamic Multi-Factor U.S. Equity ETF (MFUS, listed in the U.S.) relies on momentum and contrarian (value) performance signals to time its exposures.
Let’s take a closer look at each type of timing signal.
Economic Regime
The idea that different factors tend to do better at different points in the business cycle is intuitive. BlackRock and Oppenheimer have both found that economic regime indicators were the strongest standalone predictors of factor performance in their back-tests. However, they use different metrics to define these periods and come to slightly different conclusions about when to overweight certain factors.
There are four stages in the business cycle: recovery, expansion, slowdown, and contraction. These are defined by whether the change in economic activity is positive or negative (Oppenheimer uses “above trend” or “below trend” instead) and whether it is accelerating or decelerating. Exhibit 1 summarizes the firms’ findings about when each factor tends to outperform.
.png)
Both firms found that the small-size and value factors tended to do the best during recoveries. Smaller stocks tend to be more cyclical than their larger counterparts, as their higher market betas attest. This is likely because fewer of them enjoy durable competitive advantages to insulate their profits from fluctuations in the business cycle.
The relationship between value and the business cycle is less intuitive—and in my view, more suspect. Broad value indexes, like the Russell 1000 Value Index, have a similar market beta to the broad market, which suggests they are not more cyclical. However, deeper-value portfolios tend to have higher betas. A possible explanation for value stocks’ observed cyclicality, which Andrew Ang of BlackRock posited, is that they have higher fixed costs and less flexibility than growth stocks, so their cash flows may be more sensitive to the business cycle. These stocks may also be more beaten-down than most during tough times and poised to outperform as conditions start to improve.
During expansions, as clearly defined trends emerge, momentum has been the best-performing factor (though Oppenheimer also found that small size and value continue to do well during those periods). Not surprising, low volatility and quality have tended to do the best during slowdowns.
The biggest difference in the findings between the two firms is about which factors have tended to do the best during contractions. Oppenheimer found that quality and low volatility continued to outperform as expected, as well as momentum, which benefits from clear trends in the market. In contrast, BlackRock found that all factors modestly outperform during contractions, but momentum less than the others, which was a bit surprising. However, it’s possible that market trends are less clear in contraction periods based on BlackRock’s definition because it only looks at traditional economic data, while Oppenheimer pairs economic data with market sentiment data to get a better read of the business cycle.
.png)
It is also a little surprising that BlackRock found that value and size tended to outperform in both contraction and recovery periods, as these two regimes represent opposite sides of business cycle trends.
While the relationship between the business cycle and factor performance is interesting, there are good reasons to be skeptical. In hindsight, it’s easy to identify each stage of the business cycles past, but it’s hard to know where we stand in real time. And although the signals that BlackRock and Oppenheimer tested avoid look-ahead bias, they could have been cherry-picked to look good in sample. There are thousands of datapoints that could be reasonable indicators of the economic cycle. By chance alone, some of those datapoints will likely appear to be predictive of factor performance.
In their paper, “The Promises and Pitfalls of Factor Timing,”2 a few researchers from State Street Global Advisors conducted an exercise to illustrate the dangers of data mining. They looked at which signals were most predictive of factor performance from 1970 through 1990. They found that most of the signals with predictive power in sample were not predictive over the next 20 years out of sample.
It’s also important to note that economic cycles are slow-moving, so there aren’t many full cycles to look at in the back-tests to infer a robust relationship between the stage of the cycle and factor performance. And every cycle is different.
The world is a different place than it used to be. Business has become increasingly global. So, it probably isn’t appropriate to look only at U.S. economic data. Even if there was a strong relationship between the U.S. business cycle in the past and factor performance, it may not be as strong now. That doesn’t mean that business cycle factor-timing will fail, just that more evidence is needed to build confidence in its efficacy.
In part 2 of this article, we will continue to look at the remaining three timing signals.
1 Hodges, P., Hogan, K., Pederson, J., & Ang, A. 2016. “Factor Timing with Cross-Sectional and Time-Series Predictors.” BlackRock, https://www.blackrock.com/institutions/en-nl/literature/whitepaper/factortiming-global-12-16.pdf
2 Bender, J., Sun, X., Thomas, R., & Zdorovtsov, V. 2017. “The Promises and Pitfalls of Factor Timing.” Univ. Pennsylvania, Wharton School of Business. https://jacobslevycenter.wharton.upenn.edu/wp-content/uploads/2017/08/The-Promises-and-Pitfalls-of-Factor-Timing-1.pdf
