Skip to Main Content
Thought Leadership
April 03, 2020

Factor Rotation: It’s About Time


The effectiveness of factor-timing strategies to generate alpha on a post-transaction cost, post-tax basis, has been heavily debated by academia and the investment community
over the last few years. While there is evidence that factor performance is cyclical and time-varying, alpha capture is problematic due to the many pitfalls associated with practical
implementation. During the last decade, factor momentum has degraded, favoring shorter-term factors and reversal, which is more difficult to capture at the portfolio level.
The implementation shortfall associated with factor timing strategies can be addressed by applying a time-weighting function to counter short-term cyclicality. Our approach works
to harness the power of longer-term factor rotation to generate higher risk-adjusted alpha accompanied by lower turnover.

Key Takeaways

  • The efficacy of factor-timing strategies has been the topic of contentious debate.
  • Much research has been published to make the case that factor performance is cyclical and varies over time, although it remains difficult to harness due to execution realities.
  • Recently, factor momentum has degraded in favor of shorter-term signals.
  • Time-weighting functions can be applied to address short-term cyclicality that has emerged in the past decade.

Debate Over Factor Timing

For several years, a debate has ensued among investment practitioners about the efficacy of factor-timing strategies. At the core of the controversy is whether risk premia factors such as value, momentum, growth, and volatility have become overvalued due to the rising adoption of smart beta and factor strategies. (Cliff Asness Blasts Rob Arnott on Factor Timing, 2017)
One practitioner group advocates for a “contrarian timing” approach that emphasizes factors that are trading cheaply relative to historical norms. Exhibiting behavior similar to that for stocks, factors can become expensive and investors typically work to time their exposures to buy low and sell high. (Arnott, Beck, & Kalesnik, PhD, September 2016)

On the other hand, critics of the contrarian approach argue that factor timing is “deceptively difficult” and dispute the contrarian group’s research findings. These critics suggest the focus should be on factor diversification, not factor timing, arguing that the best way to add alpha is through more factor exposures. (Asness, Chandra, Ilmanen, & Israel, 2017)

Rather than entering into this debate, we propose another approach. We explore whether the implementation shortfall associated with factor timing strategies can be overcome by applying a time-weighting function that inherently underweights recent periods when factors might become overbought, rather than trying to quantify precisely how expensive each factor is. We review this time-weighted approach as a possible method to counter short-term cyclicality and information decay, with the ultimate goal of harnessing the power of longer-term factor rotation.

The approach throughout the paper is to use the maximum amount of data possible for multifactor models and to look back over the entire postwar period for value factor analysis. As a result, starting dates for various exhibits may differ due to availability of data from a particular source. For example, monthly returns for the five Fama French factors begin in July 1963 while data in GIM’s factor library starts in 1993.

Evidence of Factor Cyclicality

Analyzing the rolling 12-month returns for value using price-to-book as a proxy, value’s time-varying cyclicality and factor persistence is clearly demonstrated in Exhibit 1. Book-to-price had 19 value cycles in the post-war era, lasting on average almost four years in duration. The average overall quintile return spread for the subsequent 12-month performance of the value factor was 10.4 percent as shown in the table in Exhibit 1. Factor Rotation_Book Price Performance Since 1946 Top vs Bottom Quintile

The cyclical magnitude of the value cycle peak to trough is quite pronounced as shown in Exhibit 2, with an average peak return of 30.7 percent and an average drawdown of 14.4 percent.

Factor Rotation_Durationa nd Magnitude of Book to Price Value Cycles 1946 to Present

Evidence of factor cyclicality is not limited to value factors. Analyzing a full factor library aggregated into value, fundamental, earnings and technical categories provides additional evidence that the excess return of these factors acts quite differently through time with respect to one another on a non-sector neutral basis. Exhibit 3 shows this analysis over a rolling 12-month interval from 1993, the start of GIM’s internal data set, through the end of 2019.

 Factor Rotation_Excess Return by Factor Type Rolling 12 Months Full Factor Library Non Sector Neutral

actor Rotation_Non Sector Neural Factor CorrelationsIn Exhibit 4 we display the non-sector neutral factor correlations to the style groups: Value, Fundamentals, Earnings, and Technicals.1 Looking at the correlations, we see that despite the obvious factor cyclicality displayed around economic or business cycle turning points, style factor returns remain relatively uncorrelated, making a strong case for factor diversification.

In these illustrations we see the benefits of factor diversification, but we also see numerous examples of individual factor categories displaying reversion, with periods of extremely good or bad performance often followed by periods of correction. This was previously demonstrated in Exhibit 3 when value performance spiked sharply during the post-bubble and post-financial crisis periods and then reverted to the mean. Conversely during this period, earnings growth failed to perform well late in the economic cycle, but subsequently recovered.

While short-term factor cyclicality can be noisy and difficult to capture at the portfolio level, the case for longer-term reversion, based on 5-year time periods as depicted in Exhibit 5, appears more compelling. In a performance backtest of 5-year price reversions for a large cap universe of stocks, assuming 12-month holding periods, the total spread in excess return between quintiles 1 and 5 was 3.46 percent (+1.7 percent in quintile 1 and -1.76 percent in quintile 5), with an information ratio of 0.25 versus -0.44, respectively. The downside deviation was also significantly smaller in quintile 1 at 1.2 versus quintile 5 at 2.28, indicating that the longer-term reversion window resulted in more positive outcomes.*

Exhibit 5_ Factor Rotation_Five Year Price Reversion for Universe of Large Cap Stocks

The Pros and Cons Factor Timing

Farma French Five Factor Asset Pricing Model

While the potential for factor timing and factor diversification do exist, the ability to successfully time factors in order to generate better risk-adjusted returns remains open to debate.

One major issue associated with factor timing is that the causal links are dynamic and not consistent through time. As with all historical relationships observed with the benefit of hindsight, factor timing is subject to the tendencies of data-mining and overfitting. (Bender, Sun, Thomas, & Zdorovtsov, March 2018)

Furthermore, as noted in our previous paper “Aligning Value Investing with Today’s Businesses” (Kichula & Lavy, June 2019), the relevancy of specific factors may change over time. Price-to-book has become less relevant as a valuation metric as intangibles such as R&D and brand now comprise a greater percentage of company value than legacy balance sheet items such as property, plants, and equipment (PPE). (Ocean Tomo, 2016)

Trading strategies using macroeconomic inputs are complicated by the fact that government data is subject to historical revisions. Newer datasets have been developed that may avoid the restatement issue, but these tend to have relatively limited history.

Finally, there is evidence from the last decade that factor momentum has degraded in favor of shorter-term factor reversal which is more difficult to capture at the portfolio level, requiring more trading and turnover, resulting in higher transaction costs and reduced tax efficiency, according to Empirical Research Partners. Historically during the past three decades, 82 percent of the 88 stock selection factors tracked by Empirical Research Partners delivered positive alpha on an annual, serially correlated basis. However, over the last decade, in the period following the financial crisis era, only 57 percent of these factors have demonstrated favorable momentum and outperformed in consecutive years. Simultaneously, as returns from factor momentum have waned, performance from factor reversal has experienced gains. (Cahan & Liu, August 2019). One-month reversal was listed as the best-performing factor in Nomura Instinet’s factor library in the one year ending 12/31/2019, while traditional factor performance was quite mixed.

The Case For Time-Weighted Factor Rotation

When analyzing factor returns of the Fama French 5 Factor Model as shown in Exhibit 6, we note that factors have, on average, negative autocorrelations2 for two-year through seven-year periods since 1963, which is the inception of the relevant Fama French data. Over the last 20 years, there are significant negative autocorrelations between the two- and five-year periods. However, those autocorrelations turn positive over longer time frames, suggesting that reoptimizing over an extended time horizon may be better for capturing the benefits of factor rotation.

Farma French Five Factor Asset Pricing Model


Exhibit 6_2 Factor Rotation_ Farma French Autocorrelation

As shown in Exhibit 7, in order to create a factor weighting scheme that generally aligns with the Fama French 5 Factor autocorrelations, we gave the 2-5 year (12-60 months) periods a one-half weight, while giving all other periods a full weighting.

Exhibit 7_1 Factor Rotation_Time Weighting Schen vs Factor Auotcorrelation_1963-2019

Exhibit 7_2 Factor Rotation_Time Weighting Schen vs Factor Auotcorrelation_1999-2019

Fama French Factor Time-Weighting Analysis

In order to confirm the hypothesis suggested in Exhibit 7 that a time-weighting function would improve overall risk-adjusted returns, we ran empirical tests using a time-weighting function methodology. The first test was conducted using the Fama French factors, which have a longer history versus the GIM library.

In Exhibit 8 below, we display the results for three different optimizations using the Fama French Factors from 7/31/1983 to 12/31/2019. Each optimization uses a 20-year rolling history and is reoptimized on an annual basis for highest information ratio.

The Fama French Factor empirical tests conducted are as follows:

  • Equal Factor Weighted is the naïve model with equal weights (25% each) for the four Fama French Factors.
  • Equal Time Weighted is an annual reoptimization using an equal time-weighting scheme drawing on 20 years of equal weighted history. The results come from repeatedly applying a fixed model to 20-year periods going back to the start of the data set in 1963. As the approach is applied to each 20-year period, the weights in the model change each year based on the prior 20 years of known, rolling history.
  • UWL5 Time Weighted, as examined in Exhibit 7, is an annual reoptimization using a time-weighting scheme similar to that described above but emphasizing longer-term factors by underweighting four of the last five years. UWL5 places 50 percent weights on the periods between 12 months ago and 60 months ago, and equal weightings on the remaining months.Exhibit 8 Factor Rotation_Out of Sample Forward Returns Annual Reoptimization by Time Weighting Function Farma French Factors 1983-2019

Reviewing the results in Exhibit 8, the naïve, equal factor weighting resulted in a 2.65 percent excess return with a 4.84 percent standard deviation. There is a demonstrable improvement from rebalancing on an equal time weighted basis, which produced a 3.05 percent average excess return with a 5.2 percent standard deviation. The higher relative risk is rewarded with a significantly higher return, suggesting that the generally higher autocorrelations of Fama French Factors do in fact improve the end result. The information ratio increases from 0.55 to 0.59.

The third approach with the particular time weighting scheme which we referenced as UWL5, outperforms the other two models. It has the highest return of the three factors presented, at 3.11 percent. UWL5 also has a slightly lower standard deviation than the equal weighted, reoptimized model, giving it a superior risk-adjusted return. The information ratio for the UWL5 model was 0.61 as compared to 0.59 for the equal time-weighted model. Moreover, UWL5 model achieves this result with virtually the same factor turnover as the equal time weighted optimization.

Full Factor Library Factor Time-Weighting Analysis

A second test compared results from a more robust list of 84 investment factors, including multiple valuation, fundamental, earnings revision, and technical factors to the UWL5 model. In both cases, we optimized the models annually, but in the case of the full factor framework, we also optimized by sector.

Extending the reoptimization exercise to our full factor library at the sector level, we can see a similar result, as shown in Exhibit 9. The UWL5 model significantly outperforms an equally time-weighted approach, both on an excess return and information ratio basis, delivering an excess return of 1.75 percent versus the equally time-weighted optimization of 1.59 percent. The information ratio also was improved at 0.54 for the UWL5 model versus 0.42 for the equally time-weighted approach. Notably, the turnover in stocks and underlying factors is also lower than in the equal weighted approach. The naïve approach, which would have put an equal weight on each factor available with at least 20 years of history, would have only delivered 1.14 percent excess return, with a 0.25 information ratio and even higher turnover of more than 50 percent at the stock level.*


During the last decade, factor timing approaches have been the subject of much scrutiny and the efficacy of these strategies have been called into question. Most of the body of academic and practitioner research on the topic of factor timing and factor decay has focused on the time-varying risk premia of factors and on the practical implementation pitfalls.

We provide evidence, using the Fama French factors and a broader universe of factors, that applying a more practical, intuitive time-weighted approach may generate higher information ratios out of sample without increasing turnover. Our method was based on the annual re-optimization of 20-year historical periods with the underweightings of the most recent 13 to 60 month periods.

Our approach offers a possible solution to counter the problems of short-term momentum reversal and information decay which contribute to implementation shortfall at the portfolio level. Another advantage of this strategy, in addition to potentially higher risk-adjusted returns, is the benefit of lower turnover and associated transaction costs. While this approach is not a factor timing panacea, we offer it up as a common sense, practical solution that merits further research and investigation.

About Glenmede Investment Management LP

Glenmede Investment Management LP (“GIM”) is a boutique asset management firm offering actively managed equity, fixed-income and alternative strategies. The firm serves a global client base of institutions, consultants and advisors through tenured teams and consistent decision-making processes. GIM fulfills its commitment to clients by following core principles such as delivery of dynamic high-quality solutions, direct access to portfolio managers and transparency from investment philosophy through portfolio construction.

*See disclosures for further explanation regarding hypothetical data.

1. The style groups use averages from across the GIM factor library. Value factors measure a ratio of a company’s performance to its market price. Fundamentals measure a company’s business momentum, profitability, or capital structure. Earnings factors measure trends in professional analysts’ earnings estimates or targets. Technicals measure patterns in companies’ prices or liquidity in the capital markets.

2. Autocorrelation calculation shows the similarity between the time series data set and a lagged version of itself over 1-10 years.

All data is as of 12/31/2019 unless otherwise noted. Opinions represent those of Glenmede Investment Management, LP (GIM) as of the date of this report and are for general informational purposes only. This document is intended for sophisticated, institutional investors only and is not intended to predict or guarantee the future performance of any individual security, market sector or the markets generally. GIM’s opinions may change at any time without notice to you.

The charts referenced herein contain constructed or hypothetical data, which we are using purely for illustrative purposes. Though gathered or created in good faith, there is no guarantee that this information is accurate. No representation is being made that any individual account will or is likely to achieve the results similar to those shown. In fact, there are frequently sharp differences between hypothetical or constructed results and the actual results actually achieved in any particular program. Hypothetical results are prepared with the benefit of hindsight and don’t account for the financial risks of actual trading. There are numerous other factors which can adversely affect trading results which are not captured in these hypothetical scenarios. There can be no guarantee that the historical analysis contained herein will reflect future performance.

This report is not intended to be a client-specific suitability analysis or recommendation, an offer to participate in any investment, or a recommendation to buy, hold or sell securities. It may contain information which is not actionable or appropriate for every investor, and should only be used after consultation with professionals. References to risk controls do not imply that all risk is removed. All investments carry a certain degree of risk. Information contained herein is gathered from third party sources, which GIM believes to be reliable, but is not guaranteed for accuracy or completeness.

Works Cited
Cliff Asness Blasts Rob Arnott on Factor Timing. (2017, March 16). Institutional Investor.
Arnott, R., Beck, N., & Kalesnik, PhD, V. (September 2016). Timing “Smart Beta” Strategies? Of Course! Buy Low, Sell High! Research Affiliates.
Asness, C., Chandra, S., Ilmanen, A., & Israel, R. (2017). Contrarian Factor Timing is Deceptively Difficult. Journal of Portflio Management.
Polk, C., Haghbin, M., & de Longis, A. (April 2019). Time-Series Variation in Factor Premia: The Influence of the Business Cycle.
Jenkins, I. F. (May 2015). Using factors with different alpha decay times: The case for non-linear combination. Nomura Equity Research.
Kessler, S., Scherer, B., & Harries, J. P. (2020). Value by Design? The Journal of Portfolio Management, Vol. 46, Issue 3.
Bender, J., Sun, X., Thomas, R., & Zdorovtsov, V. (March 2018). The Promises and Pitfalls of Factor Timing. The Journal of Portfolio Management, 79-92.
Kichula, J., & Lavy, J. (June 2019). Aligning Value Investing with Today’s Businesses. Glenmede Investment Management LP.
Ocean Tomo. (2016). Intangible Asset Market Value Study. Retrieved from
Fama, E. F., & French, K. R. (September 2014, April 5). A Five-Factor Asset Pricing Model. Fama-Miller Working Paper, 52. Retrieved from QuantInsti:
Cahan, R., & Liu, Y. (August 2019). Factor Momentum: Still a Thing? Empirical Research Partners.