Factor Rotation: Possible, but Worth It?
In many ways, 2016 was a year of heavy factor rotation research. Rob Arnott (Research Affiliates) and Cliff Asness (AQR) sparred all year on the topic. To summarize their views:
- With significant research into factor rotation in 2016, we expect to see more factor rotation strategies in the market in 2017.
- Using six popular factors (Value, Size, Reinvestment, Operating Profitability, Momentum and Beta), we explore both switching and rotation based strategies.
- We find switching strategies to be largely unimpressive compared to simple buy-and-hold.
- We find factor rotation to have outperformed a passive multi-factor portfolio by 1.32% per year. However, we believe that after fees (i.e. manager fees and transaction costs), the added performance may not be worth the headaches of multi-year drawdowns relative to the simpler approach.If 2016 was the year of multi-factor products, we expect that 2017 will be the year of factor rotation products.
So who is right? Who is wrong? What is the expected efficacy of factor rotation strategies? Let’s dive in.
- The research from Rob and his team at Research Affiliates shows that valuation spreads within factors are important and can provide guidance on potential factor return premiums, if not flash a warning sign about outright factor crashes.For example, Research Affiliates would say that if low volatility stocks are significantly overvalued compared to historical levels (e.g. aggregate price-to-book versus historical average), it may not make sense to hold them.
- Cliff took exception to both the applicability of the research as well as the theory. Largely, his argument hinged on the fact that valuation provides long-term guidance on returns, and therefore says little about high turnover factors.For example, if we know that a small-cap index is trading at a 25 P/E today and our crystal ball tells us that it will be trading at a 15 P/E in five years, that would likely provide us very important information about drag created by valuation contraction.On the other hand, if we know that a momentum index is trading at a 25 P/E today and our crystal ball tells that it will be trading at a 15 P/E in five years, it tells us very little, because it is very unlikely that the portfolio will hold the same stocks a year from today, much less five years. So the contraction in portfolio P/E may be due to changes in holdings and not a contraction in valuation of those holdings.
Evidence-based research requires having a hypothesis before we do the research. Otherwise, we risk poking around in the dark, finding positive results, and creating a narrative to match those results.
So what is our general hypothesis about factor timing?
Well, if we believe that the factors exist for behavioral reasons (which we largely do), then those same behavioral biases should likely apply to the factors themselves. Therefore we expect that valuation-based timing should be effective for low-turnover factors (e.g. value and size) and momentum-based timing should be effective for all factors.
Well, at least these were our hypotheses when we first started investigating the topic early last year. At this point, we’ve already done a fair amount of research into the topic. In fact, we wrote a paper and submitted it to NAAIM for their annual Founders/Wagner Award competition....