Will Your Factor Deliver? An Examination of Factor Robustness and Implementation Costs

Abstract

The multifactor investing framework has become very popular in the indexing community. Both academic and practitioner researchers have documented hundreds of equity factors. But which of these factors are likely to profit investors once implemented? We find that many of the documented factors lack robustness. Size and quality, two of the more prominent factors, show weak robustness, whereas value, momentum, illiquidity, and low beta are more robust. Further examining implementation characteristics, we find that liquidity-demanding factors, such as illiquidity and momentum, are associated with significantly higher trading costs than are other factors. Investors may be better off accessing these factors through active management rather than indexation.

Editor’s note: This article was reviewed and accepted by Executive Editor Stephen J. Brown.

Notes

¹ Harvey, Liu, and Zhu (2016) found 316 factors in the literature as of year-end 2014 and also found that approximately 40 new factors are published annually. These two facts suggest that as of year-end 2016, we will be at 300 + 80 or higher.

² John Cochrane coined the term “zoo of new factors” in his presidential address at an annual meeting of the American Finance Association (Cochrane 2011).

³ We limited the scope of our study to examining which factors can profit investors on a standalone basis. If we found that a certain factor lacks robustness, such a finding would not imply that this factor might not be important in the broader asset-pricing context (e.g., owing to its correlations with other factors).

⁴ Indexing is often associated with capitalization-weighted benchmarks. Smart beta, which breaks the link between asset prices and index weights, is another approach to index investing; it is designed to capture nonmarket sources of premiums.

⁵ We searched for keywords that included a factor name combined with the word factor and required at least 100 hits for a factor to be included in our study. Our rationale for using the word factor in each query was to home in on asset-pricing papers rather than, say, corporate finance papers.

⁶ The difference between our approach and that of Harvey et al. (2016) is that we further aggregated several of the distinct factors into groups that are more common among practitioners and that can be logically combined. For instance, we classified several ratios (e.g., P/E, P/B, and P/D) as value factors.

⁷ This phenomenon is often referred to as a flat or even inverted security market line (SML), which is often found in empirical studies. A flat SML means that average stock performance is largely unrelated to the riskiness of the stocks; an inverted SML means that low-risk stocks tend to slightly outperform riskier stocks.

⁸ For references, see Black, Jensen, and Scholes (1972) and Frazzini and Pedersen (2014).

⁹ For evidence of investors’ preference for gambling and the low-beta anomaly, see Blau, Hsu, and Whitby (2014); Bali, Brown, Murray, and Tang (2015); and Hsu and Viswanathan (2015). For evidence that investors in emerging markets use the stock market as a gambling substitute, see Gao and Lin (2015). For another reading of the preference-for-gambling hypothesis, see Baker, Bradley, and Wurgler (2011).

¹⁰ See Hsu, Kudoh, and Yamada (2013).

¹¹ See Baker, Bradley, and Wurgler (2011) and Brennan, Cheng, and Li (2012).

¹² For factor definitions, see Appendix A.

¹³ Frazzini and Pedersen (2014) constructed a long–short factor portfolio labeled “betting against beta” (BAB), which levers up the low-beta stocks (by about 1.4× for US stocks) on the long side and shorts the high-beta stocks (by only 0.7×), so this portfolio has a beta of approximately zero with respect to the equity market.

¹⁴ MSCI estimates that roughly $50 billion is tied to its minimum-volatility index, which is dwarfed by the estimated $7 trillion of assets tied to various cap-weighted market indexes.

¹⁵ For a full theoretical treatment of overreaction and momentum, see Hong and Stein (1999).

¹⁶ For an analysis of momentum transaction costs, see Grundy and Martin (2001).

¹⁷ See De Bondt and Thaler (1985).

¹⁸ We followed the definition in Fama and French (2012).

¹⁹ For other, more complicated definitions of illiquidity, see Amihud (2002) and Pástor and Stambaugh (2003).

²⁰ For a more exhaustive list of ways to define quality, see Appendix C (posted as supplemental material at www.cfapubs.org/toc/faj/).

²¹ In this analysis, we focused on US strategies for the sake of brevity. Because this trading-cost model concerns ADV characteristics of turnover, we saw similar patterns in trading costs in international markets where the behavior of these attributes is similar.

²² Although the linear model we chose is very simple, it is extremely easy to estimate for historical datasets because it does not require additional historical data beyond ADV. Moreover, despite its simplicity, it provides realistic estimates for trades that do not consume too much liquidity. Like Aked and Moroz (2015), Vangelisti (2006) estimated an impact of 30 bps per 10% of ADV. But the linear price impact model has limitations when trading volume is large. Gabaix, Gopikrishnan, Plerou, and Stanley (2006) found that for larger trades (equivalent to 200% or more of ADV), the price impact grows at a lower-than-linear rate (approximately the square root of trading volume). The slower-than-linear impact found by Gabaix et al. implies that the Aked–Moroz methodology we used overestimates trading costs for extremely high-turnover strategies. Nonetheless, in our study, the assumed level of assets under management ensured that most trades stayed below the 200% ADV level, where the nonlinearity effects start to matter.

²³ These techniques include (1) designing indexes with high weighted-average market caps, (2) eliminating unnecessary turnover, (3) placing bands or tolerance zones around index boundaries to reduce the expensive turnover that might otherwise arise from stocks jumping in and out of the index, and (4) spreading turnover with staggered rebalancing by using a methodology similar to the one introduced by Blitz, van der Grient, and van Vliet (2010). Such techniques can reduce transaction costs appreciably. In our simulation, we used simplified methods for constructing factor portfolios.

²⁴ There is a long literature on market factor timing, including Campbell and Shiller (1988); Welch and Goyal (2008); and Cochrane (2008). Other noteworthy studies that document the variability of factor premiums over time include Asness, Friedman, Krail, and Liew (2000; value); Daniel and Moskowitz (2013; momentum); and Li and Lawton (2014; low beta).