Five Concerns With The Five-Factor Model
Robeco Asset Management – Quantitative Strategies
Robeco Asset Management – Quantitative Strategies; Technische Universität München (TUM)
VU University Amsterdam, Finance; Robeco Asset Management
Robeco Asset Management – Quantitative Strategies
[drizzle]November 1, 2016
Fama and French (2015) propose to augment their classic (1993) 3-factor model with profitability and investment factors, resulting in a five-factor model, which is likely to become the new benchmark for asset pricing studies. Although the 5-factor model exhibits significantly improved explanatory power, we identify five concerns with regard to the new model. First, it maintains the CAPM relation between market beta and return, despite mounting evidence that the empirical relation is flat, or even negative. Second, it continues to ignore the, by now, widely accepted momentum effect. Third, there are a number of robustness concerns with regard to the two new factors. Fourth, whereas risk-based explanations were key for justifying the factors in the 3-factor model, the economic rationale for the two new factors is much less clear. Fifth and finally, it does not seem likely that the 5-factor model is going to settle the main asset pricing debates or lead to consensus.
Fama and French (1992) present strong evidence for the existence of size and value premiums in the cross-section of stock returns. Fama and French (1993) argue that these factors capture a dimension of systematic risk that is not captured by market beta in the Capital Asset Pricing Model (CAPM), and propose to extend the CAPM with size (SMB, small minus big) and value (HML, high minus low) factors, resulting in a 3-factor model. The 3-factor model has been very influential, and it has become common practice in the asset pricing literature to not only report 1-factor alphas, but also 3-factor alphas. However, many such studies report 3-factor alphas that are statistically significantly different from zero, which suggests that the 3-factor model is incomplete and that more factors are needed to accurately describe the cross-section of stock returns. Inspired by this mounting evidence that three factors do not suffice, Fama and French (2015) propose to augment their 3-factor model with two additional factors, namely profitability (RMW, robust minus weak) and investment (CMA, conservative minus aggressive). This new five-factor model is likely to become the new standard in asset pricing studies, which significantly raises the bar for new anomalies. In this paper, however, we argue that the five-factor model also raises a lot of questions. More specifically, we discuss five concerns with regard to the new model.
Our first concern is that, similar to the 3-factor model, the five-factor model retainsthe CAPM relation between market beta and return, which implies that a higher market beta should, ceteris paribus, result in a higher expected return. This assumption denies the existence of a low-beta or low-volatility premium, despite a wide body of literature showing otherwise. Fama and French (2016a) specifically address this issue and argue that, in fact, the low-beta anomaly is fully resolved by their five-factor model. However, we believe that this conclusion is premature, since direct evidence that a higher market beta exposure is rewarded with higher returns is still conspicuously absent.
Our second concern is that, similar to the 3-factor model, the five-factor model is unable to explain the momentum premium, and continues to ignore it. Because momentum is too pervasive and important to ignore, most studies do not only report 3-factor alphas, but also 4-factor alphas, based on the 3-factor model augmented with a momentum factor (WML, winner minus losers). For the same reason, many researchers will feel the need to augment the five-factor model with a momentum factor, resulting in a 6-factor model. However, even asset pricing models that include the standard momentum factor are unable to explain the related idiosyncratic momentum phenomenon documented by Guttierez and Pirinsky (2007), as shown by Vidojevic, Hanauer and Blitz (2016).
Our third concern is the robustness of the two new factors chosen by Fama and French (2015). Particularly surprising is that the investment factor is defined as asset growth, which they considered to be a “less robust” phenomenon in Fama and French (2008). More specifically, the five-factor model fails to explain a number of variables that are closely related to the two selected ones. Another robustness concern is that it is still unclear whether the two new factors are effective before 1963 or within other asset classes, as factors such as value and momentum are known to be.
Our fourth concern is the economic rationale of the new model. Fama and French (1993, 1996) still justified the addition of size and value factors by arguing that these could be seen as priced risk factors, hinting at the possibility that they might capture the risk of financial distress. Since then, however, studies such as Dichev (1998), Griffin and Lemon (2002), and Cambell, Hilscherm, and Szilagyi (2008) have shown that the direct relation between distress risk and return is actually negative, which is consistent with the existence of a low-risk premium. For the two new factors in the five-factor model, Fama and French do not even attempt to explain that these are plausible risk factors. Instead, their motivation for inclusion of these factors is that they should imply expected returns, which they derive from a rewritten dividend discount model. It remains unclear though if the higher expected returns for firms with high profitability or low investment, all else constant, are due to higher (distress) risk or mispricing. If the goal were to simply construct a model which fits the data best without the need to rationalize the chosen factors, the five-factor model does a pretty good job (apart from the concerns discussed above), but this is not how the model is motivated.
Our fifth and last concern is that the five-factor model is probably not going to put an end to empirical asset pricing discussions or lead to consensus. For instance, the classic size and value factors are still being challenged, as the size premium seems to have dwindled after it was first documented in the early nineteen eighties, while the HML value factor is also known to have robustness issues, especially in the large-cap segment of the market. Also, there is a lack of consensus that the five-factor model is the ultimate asset pricing model, and alternative, competing models are already out there; see, e.g. Hou, Xue, and Zhang (2015, 2016). Finally, just like over the last two decades, many studies documented asset pricing anomalies with significant 3- and 4-factor alphas, we expect that in the years ahead there will appear many studies that document anomalies with significant 3-, 4-, 5-, and 6-factor alphas.
In the following sections we discuss each of these five concerns in more detail. We conclude that although the five-factor model represents a major step forward in the asset pricing literature, it is not going to settle the asset pricing debate. It may well turn out to raise more questions than it answers.
Read full paper at SSRN.