Beyond Fama-French Factors: Alpha from Short-Term Signals

Abstract

Short-term alpha signals are generally dismissed in traditional asset pricing models, primarily due to market friction concerns. However, this paper demonstrates that investors can obtain a significant net alpha by applying a combination of signals to a liquid global universe and with advanced buy/sell trading rules that mitigate transaction costs. The composite model consists of short-term reversal, short-term momentum, short-term analyst revisions, short-term risk, and monthly seasonality signals. The resulting alpha is present in out-of-sample and post-publication periods and across regions, translates into long-only applications, is robust to incorporating implementation lags of several days, and is uncorrelated to traditional Fama-French factors.

Keywords:

• investments
• market efficiency
• market frictions
• portfolio construction
• short-term signals
• transaction costs

PL Credits: 2.0:

Acknowledgements

The authors thank Guido Baltussen, Executive Editor William Goetzmann, Associate Editor Daniel Giamouridis, Harald Lohre, two anonymous reviewers, and participants at the Maastricht University and Robeco research seminars for valuable comments and suggestions.

Disclosure statement

The authors disclose that they are employed by Robeco, a firm that offers various investment products. The construction of these products may, at times, draw on insights related to this research. The views and results presented in this article were not driven by the views or interests of Robeco and are not a reflection of its points of view.

Notes

1 The Fama and French (2015) five-factor model is often augmented with a mid-turnover momentum factor. Although the momentum premium has also turned out to be “pervasive” (Fama and French 2008, 1653), Fama and French (2018, 237) only “reluctantly” include it due to theoretical motivation concerns. Another reason could be that the model focuses on explaining long-term expected returns rather than short-term variation in returns (Fama and French 2016). Nevertheless, both for the U.S. and international markets, a modified Fama-French six-factor model seems to be the dominant factor model (cf. Barillas and Shanken 2018, Barillas et al. 2020, or Hanauer 2020).

2 The standard MSCI World index comprises large- and mid-cap stocks across developed markets countries and targets to cover approximately 85% of the free float-adjusted market capitalization in each country. Therefore, this size threshold is even stricter than the academic convention of “big” stocks, which are typically defined as the biggest stocks that account for 90% of the aggregated market capitalization per region (cf. Fama and French 2012). Before 2001, we do not have access to MSCI World constituents, so we use FTSE Developed as a proxy.

3 Another enhancement of the standard short-term reversal strategy is the residual short-term reversal strategy of Blitz et al. (2013). This approach adjusts the returns of a stock for its exposures to the market, size, and value factors of Fama and French (1993), estimated over the preceding 36 months, and then scales the residual returns by their volatility over the same period. Using this signal or the raw short-term reversal signal gives a very similar alpha and does not affect our main results (cf. Table A1 in the Appendix).

4 Earlier studies such as Stickel (1991) and Chan, Jegadeesh, and Lakonishok (1996) use the change in the consensus earnings forecast. However, this requires an appropriate scaling factor, which should not be a metric that can be zero or negative (like earnings itself), and which should not introduce other effects (like valuation effects when dividing by market capitalization). By using the definition of Van der Hart, Slagter, and van Dijk (2003), we prevent such issues.

5 The IBES database has been accused of rewriting history by Ljungqvist, Malloy, and Marston (2009), but it turns out that all major discrepancies stem from different ways of dealing with corporate actions and mergers between brokerage firms (cf. Thomson Reuters, 2016).

6 Alternative lead-lag effects were for instance also documented for customer-supplier links (Cohen and Frazzini, 2008) and for standalone versus more complex firms (Cohen and Lou 2012).

7 Other skewness-related signals are lottery-type features (Kumar, 2009) or expected skewness (Boyer, Mitton, and Vorkink, 2010).

8 Furthermore, our analysis starts in 1986, while the standard factor time series for developed markets, available on Kenneth French’s website, are only available from June 1990 onward. If we use the control factors from Kenneth French’s website and restrict our analysis to the June 1990 to December 2021 period for which these time series are available, our conclusions remain the same (cf. Table A1 in the Appendix).

9 For instance, when the fresh top quintile of strategy A is bought, the old top quintile of strategy B is sold, and the two strategies are independent, then about 20% of these trades can likely be crossed. Moreover, if a stock is very attractive according to the first strategy but very unattractive according to the second strategy then a neutral position might actually be most appropriate.

10 Please note that the break-even trading costs for the Fama-French six-factor model alpha are a conservative estimate as we use gross factor returns to compute this alpha. In reality, factors such as momentum also require a substantial amount of turnover, which decreases their net performance.

11 We thank an anonymous reviewer for raising this point.

12 The only exception is the break-even trading cost level for the analyst earnings revision signal that is just above 25 bps.

13 The figure should be read as follows. Each line presents one of the four portfolio construction approaches. The first dot at the top left of each line marks the gross six-factor alpha of the strategy. When following the line to the bottom right, the next two dots mark the net alpha and transaction cost levels when the strategies become insignificant at the 1% and 5% significant levels. The last dot for each strategy show at which trading cost level the net alpha becomes negative.

14 We thank two anonymous reviewers for raising this point.

15 The attenuation of prominent anomalies has also been documented by Chordia, Subrahmanyam, and Tong (2014), and Green et al. (2017) document a sharp decrease in return predictability after 2003.

16 As another robustness check, Table A2 in the Appendix reports net results for the second half of the sample, starting in January 2004. The 10/50 strategy is shown for global markets, as well as for the U.S. separately. Consistent with the out-of-sample and post-publication results, the alphas generally hold up well, with most t statistics remaining economically and statistically significant. However, for the U.S. market in isolation the net alpha (after 25 bps costs per trade) is insignificant over this recent period, with the net outperformance even being negative. This echoes the finding of Jacobs and Müller (2020) that post-publication performance decay is a major concern for the U.S. market but much less so for international markets. It also underlines the importance of further minimizing the implementation shortfall and continued innovation with new and improved signals.

17 The number of possible combinations with n signals is $\left(\genfrac{}{}{0pt}{}{5}{n}\right).$

18 We thank William Goetzmann for this suggestion.

Additional information

Notes on contributors

David Blitz

David Blitz is managing director of Quantitative Investments at Robeco Asset Management, Rotterdam, the Netherlands.

Matthias X. Hanauer

Matthias X. Hanauer, CFA, is director of Quantitative Investments at Robeco Asset Management, Rotterdam, the Netherlands, and postdoctoral researcher in finance at the Technical University of Munich, Munich, Germany.

Iman Honarvar

Iman Honarvar is director of Quantitative Investments at Robeco Asset Management and lecturer at Erasmus School of Economics, Rotterdam, the Netherlands.

Rob Huisman

Rob Huisman, CFA, is a quantitative researcher at Robeco Asset Management, Rotterdam, the Netherlands.

Pim van Vliet

Pim van Vliet is managing director of Quantitative Investments at Robeco Asset Management, Rotterdam, the Netherlands.