THE JOURNAL OF. ETFs, ETPs & Indexing. The Voices of Influence iijournals.com. FALL 2016 VOLUME 7 NUMBER 2

Transcription

1 THE JOURNAL OF FALL 2016 VOLUME 7 NUMBER 2 ETFs, ETPs & Indexing The Voices of Influence iijournals.com

2 Long-Term Rewarded Equity Factors: What Can Investors Learn from Academic Research? Noël Amenc and Felix Goltz Noël Amenc is a professor of finance at EDHEC-Risk Institute and the CEO of ERI Scientific Beta in Singapore. noel.amenc@edhec-risk.com Felix Goltz is the head of applied research at EDHEC-Risk Institute and a research director at ERI Scientific Beta in Nice, France. felix.goltz@edhec.edu Equity index products that claim to provide exposure to factors that have been well documented in academic research, such as value and momentum, among others, have been proliferating in recent years. Interestingly, providers across the board put strong emphasis on the academic grounding of their factor indexes. 1 At the same time, product providers try to differentiate themselves using proprietary elements in their strategy, often leading to the creation of products using new factors or novel strategy construction approaches that may or may not be consistent with the broad consensual findings in the academic literature on empirical asset pricing. Moreover, discussion of the sources of performance is often based on provider-specific research rather than consensual findings in the academic literature. Given the wide availability of academic research on factor investing and the extensive external scrutiny it is exposed to, it is perhaps useful to inspect which key conclusions from this research and to analyze the implications for factor index design and for understanding the sources of performance. The objective of this article is not to provide new research results, introduce new factors or discuss yet another enhancement, but rather to survey the results across a vast array of peer-reviewed academic publications. Our aim is to analyze what academic research has to say on equity factors in order to understand what we can learn from such research on designing or evaluating factor indexes. When analyzing academic publications on equity factor investing, several important lessons emerge, which provide useful perspective on practical questions about factor indexes. A key result from analyzing the literature is that well-established factor premia are not simply based on a backtest similar to that used by product providers, but instead such premia have been subjected to extensive empirical analysis including an assessment over very long-term data, post-publication data, and cross-sample validation, notably when factors uncovered in the crosssection of U.S. stock returns have been confirmed in international data or in other asset classes. Moreover, common factor premia have been explained using formal economic models providing a rationale for the persistence of such premia. Recalling these results is useful in clarifying that novel strategies or enhancements with respect to such factors should be subjected to similar levels of scrutiny before deciding on their relative merits. Another key insight from academic research is that the basic lesson on diversification also applies to factor-tilted portfolios. Although practitioners often favor narrow portfolios and providers develop products with labels such as strong factor index, these narrow portfolios will add not just rewarded risk but also unrewarded (stock-specific) risk. Fall 2016 The Journal of Index InvesTIng

3 That the academic literature underlines the need for broad diversification for a given factor tilt both from an empirical and theoretical perspective should lead many investment professionals to reconsider which points are important to address in their factor strategies. Another lesson from academic research is that the practical implementation costs of academic factor strategies are relatively well understood, and many low turnover factors deliver significant premia net of transaction costs, especially when using straightforward implementation adjustments. Perhaps surprisingly, there are far more results on the impact of transaction costs on factor investing returns in the academic literature than can be found in product presentations, where index providers frequently omit such results with the argument that estimating transaction costs for their factor indexes is not something they can easily do. Here again, useful lessons can be learned from academic research. Concerning the sources of performance of factor strategies, the academic literature is particularly rich, and various formal explanations of premia and their persistence have been proposed. In contrast, discussions on the same topic by product providers often resort to mere storytelling, for example, involving vague references to historical market events (see Asness [2016] for a critique of such practices). More generally, there is a lot of debate about factor investing strategies among investment professionals. This article tries to contribute by tying such discussions to the academic evidence. Although one can obviously disagree about what the evidence is, a useful discussion should be about what the consensual evidence does or does not say. Ultimately, such a discussion of the broad academic evidence may lead to more general and more robust conclusions than looking at provider-specific research. Therefore, we believe that a review of the academic lessons on factor investing is of crucial importance. The outline of this article is as follows: We first look at the empirical analysis that is required to identify rewarded factors. Second, we turn to the economic rationale behind factors. Third, we look into the role of diversification for a given factor tilt. We then discuss the issue of implementation costs, and the subsequent section addresses the question of crowding risks. We conclude with a discussion of frequent mismatches of practical implementations with the academic groundings. LESSON ONE: BE SERIOUS WITH DATA When establishing which factors carry a reward by way of empirical analysis, it is important to understand that this is an almost daunting task. In fact, since Merton [1980], it is well known that we struggle to estimate expected returns reliably, simply because we rely on very few data points to estimate long-term expected returns: the starting price level and the end date price level. Of course, this is also true for factor returns. Given this difficulty, when testing whether a factor carries a positive premium, academic research conducts a thorough assessment, including the analysis of very longterm data (covering time spans of at least 40 years), analysis across different regions and asset classes, and various corrections for possible data-mining biases. Importantly, these studies are open to criticism. Numerous articles are written to question previous empirical results (see, for example, the debate on the low volatility puzzle ). For these reasons, academic research is much more capable of providing meaningful conclusions than a product backtest for a given factor index product. Even if a backtest is conducted very thoroughly by a product provider, it is hard to believe that the provider is able to conduct as thorough an analysis of the whole academic community, whose members have strong incentives not only to publish their own results but also to challenge the results of others by way of replicated tests. Therefore, factors that have undergone academic validation constitute a much stronger empirical justification than a mere product backtest. The first important characteristic of empirical evidence on factor premia, as mentioned earlier, is that this evidence is derived based on tests applied to long-term data. In fact, studies on U.S. equity data typically span at least 40 years of data, and in many cases, the data go as far back as the 1920s. Exhibit 1 summarizes the empirical literature on empirical tests of the six main factors we consider in this study: size, value, momentum, low risk, profitability, and investment. The main findings from the following seven articles are reported: 1. Fama and French [1992], 2. Jegadeesh and Titman [1993], 3. Ang et al. [2006], 4. Novy-Marx [2013]; 5. Frazzini and Pedersen [2014], 6. Fama and French [2015a] and 7. Hou, Xue, and Zhang [2015]. Long-Term Rewarded Equity Factors: What Can Investors Learn from AcademIC Research? FaLL 2016

4 Exhibit 1 provides the methodology and underlying data for each article. The literature on factor premiums highlights the existence of statistically and economically significant premiums over the long term. A second important characteristic of empirical research on factor premia is the assessment across different regions and asset classes. In fact, merely deriving a result from U.S. data, even if it holds in long-term data, does not allow the findings to be generalized to other geographic or investment contexts. From the standpoint of generalization, it is therefore interesting if results can be confirmed on equity markets for other geographies or even in entirely different asset classes. Research has made considerable progress in this direction over the past decade, with surprisingly strong confirmation of the U.S. equity results in other investment universes (see Exhibit 2 for an overview). E x h i b i t 1 Summary of Empirical Tests on Risk Factor Premia (continued ) Fall 2016 The Journal of Index InvesTIng

5 E x h i b i t 1 (continued ) Summary of Empirical Tests on Risk Factor Premia A third important precaution empirical research takes before jumping to conclusions on the premium for a given factor is to adjust for data mining or so-called multiple testing. In fact, standard statistical tests are only valid when we test a given single hypothesis, such as that high book-to-market stocks carry a premium over low book-to-market stocks. However, in practice researchers may run several tests, for example, trying out a large number of metrics until they find one that leads to significant results. This is also known as data snooping or data mining. To understand why such multiple testing may lead to false inference, consider a simple example. Assume you simulate data for 100 variables (potential factors) that have zero mean. You would then expect to find about five variables with mean (premium) significantly different from zero. This means that, even though the true mean (premium) on all of the variables (factors) is zero in the simulation, the statistical inference will tell you that some of the means are significantly positive, as long as you run enough tests. In order to adjust for this problem, researchers have come up with tighter requirements for significance levels to take into account the possibilities of multiple testing. For example, Harvey, Liu and Zhu [2016] adjusted t-ratios that are used for evaluating the significance of factor premia to take into account the fact that researchers have run many tests across hundreds of factors to document their premia. Interestingly, when applying these methods to standard equity risk factors, researchers find that the main factors, such as value and momentum, among others, remain statistically significant. Despite the thorough evidence supporting the existence of premia for the main factors, there is continuous debate over the set of relevant equity factors. In fact, research often debates whether a factor has disappeared or a new factor has appeared. Although questioning the baseline results and discussing relevant actors is obviously useful, investors in practice should be prudent before making abrupt changes to their set of factors or the associated investment beliefs. As mentioned before, the measurement of a risk premium is highly sensitive to the chosen sample (Merton [1980]), and estimates of factor premia are subject to considerable uncertainty. Therefore, any conclusions based on empirical evidence should only be drawn from studying very long time periods and conducting tests across different datasets. Moreover, any arguments in favor of the disappearance Long-Term Rewarded Equity Factors: What Can Investors Learn from AcademIC Research? FaLL 2016

6 E x h i b i t 2 Empirical Evidence for Selected Factor Premia Note: N.A. = not available. of standard factors or the appearance of new factors should not be investigated based on empirical evidence alone but should also consider the underlying economic mechanisms, an issue we turn to in the next section. An important lesson from the empirical tests in the academic literature is that established factor premia have passed a much higher hurdle rate than a typical product backtest. Although it is quite likely that one can come up with a novel factor or investment strategy that has produced favorable results in a backtest for a single equity universe over a relatively brief time span of, say, 15 years, it is much less likely that one will uncover such effects when subjecting them to intense scrutiny by the academic community involving cross-sample validation and extensive tests for robustness. This suggests that investors can rely on a large degree of external scrutiny by simply following the consensual factors and definitions used in academic research. In turn, if they wish to deviate from consensual academic factor definitions, for example, it would be desirable that product providers apply similar levels of scrutiny to avoid ending up developing products based on results that will not be robust. LESSON TWO: BEING SERIOUS WITH DATA IS NOT ENOUGH In addition to convincing empirical evidence, the existence of a factor premium should be supported by a compelling economic rationale. Kogan and Tian [2012] made this point prominently when they wrote: We should place less weight on the data the models are able to match, and instead closely scrutinize the theoretical plausibility and empirical evidence in favor of or against their main economic mechanisms. To illustrate why the existence of an economic rationale is an important requirement for considering a factor to be rewarded, it is useful to take the equity market risk premium as an example. From an empirical perspective, the equity risk premium can be statistically indistinguishable from zero even for relatively long sample periods. However, economic reasoning suggests that stocks should have higher reward than bonds. Clearly, even if the premium for holding equity is welldocumented empirically, investors are reluctant to hold too much equity due to its risks. Similar reasoning can be applied to additional equity risk factors. Instead of focusing only on the empirical evidence, investors due diligence should look at why there should be a risk premium for a given factor in the first place. In other words, investors should ask what the economic rationale for a factor premium is, to form an opinion on its existence and persistence. The existence of factor premia can be explained in two different ways a risk-based explanation and a behavioral-bias explanation. The risk-based explanation premises that the risk premium is compensation to investors who are willing to take additional risk by being exposed to a particular factor. Additional risk exists Fall 2016 The Journal of Index InvesTIng

7 when assets that correspond to a given factor tilt tend to provide poor payoffs in bad times, thus exposing investors to a risk of losses in times when their economic situation is already poor, their consumption is low, and marginal utility of consumption is high. The behavioral explanation conceives that the factor premia exist because investors make systematic errors due to such behavioral biases as over- or underreaction to news on a stock. Whether such behavioral biases can persistently affect asset prices is a point of contention given the presence of smart market participants who do not suffer from these biases. For behavioral explanations to be relevant, it is necessary to assume that in addition to biases there are so-called limits to arbitrage, that is, some market characteristics, such as short-sales constraints and funding-liquidity constraints, that prevent smart investors from fully exploiting the opportunities arising from the irrational behavior of other investors. If the risk premium can only be explained by behavioral reasoning, it is expected to disappear in the absence of limits to arbitrage. Alternatively, a risk factor with a strong rational- or risk-based explanation is more likely to continue to have a premium in the future. Therefore, it is perhaps more reassuring for an investor to have a risk-based explanation. We refer to Exhibit 3 for a brief list of risk-based and behavioral explanations of each factor. Considering the economic explanation of factor premia should be an important part of the due diligence process of investors before they adopt such strategies. Factor indexes indeed offer asset owners the advantage of making explicit factor choices, rather than accepting the factor exposure choices made by an active manager for example (see Ang and Kjaer [2011]). To make such explicit choices, asset owners should consider what their investment beliefs for these factors are and, notably, whether they are confident that the premium associated with a factor is actually persistent. Given the limitations of empirical results, even when they are thoroughly scrutinized and confirmed externally, gaining an economic understanding of why a factor premium should be there in the first place is a crucial question when selecting the relevant factors. Considering the economic rationale behind a factor is an important complement to empirical analysis and, given the potential issues with datamining, may indeed be crucial to identifying persistent factor premia. Perhaps, reviewing why according to the academic literature a factor should matter is ultimately more important in the due diligence process than considering recently proposed enhancements and other provider-specific results about factor investing. LESSON THREE: DIVERSIFICATION MATTERS It should be emphasized that many smart beta strategies do not solely rely on tilting toward factors. Although many strategies labeled as smart beta effectively limit their strategy design to obtaining factor tilts, it should be noted that other strategies rely on diversification mechanisms to improve on cap-weighted indexes. E x h i b i t 3 Economic Mechanisms behind Main Factors Long-Term Rewarded Equity Factors: What Can Investors Learn from AcademIC Research? FaLL 2016

8 Even in the area of factor indexes, one can distinguish between two approaches, namely, factor indexes that only deal with tilting toward stocks with favorable factor characteristics and well-diversified factor indexes (also termed smart factor indexes ), which not only tilt to a given factor, but also ensure diversification through alternative weighting schemes. Positive exposure to rewarded factors is obviously a strong and useful contributor to expected returns. However, products that aim to capture explicit riskfactor tilts often neglect adequate diversification. This is a serious issue because diversification has been described as the only free lunch in finance. It allows a given exposure to be captured with the lowest level of risk required. In contrast, gaining factor exposures exposes investors to additional types of risk, and therefore, such exposures do not constitute a free lunch. They instead constitute compensation for risk in the form of systematic factor exposures. Such capturing of risk premia associated with systematic factors is attractive for investors who can accept the systematic risk exposure in return for commensurate compensation. Factor-tilted strategies, when they are very concentrated, may also take on other, non-rewarded, risks. Unrewarded risks come in the form of idiosyncratic or firm-level risk, as well as potential risk for sector concentration. Financial theory does not provide any reason why such risk should be rewarded. Therefore, a sensible approach to factor investing should look not only at obtaining a factor tilt but also at achieving proper diversification within that factor tilt. To illustrate this point, we focus on the value factor as an example in the following, but the discussion carries over to other factors, too. In fact, if the objective were to obtain the most pronounced value tilt, for example, then the only unleveraged long-only strategy that corresponds to this objective is to hold 100% in a single stock, the one with the largest value tilt, as measured by, say, its estimated sensitivity to the value factor or its book-to-market ratio. This thought experiment clearly shows that the objective of maximizing the strength of a factor tilt is not reasonable. Moreover, this extreme case of a strong factor tilt indicates what the potential issues with highly concentrated factor indexes are. First, such an extreme strategy will allow the highest possible amount of return to be captured from the value premium, but it will necessarily come with a large amount of idiosyncratic risk, which is not rewarded and therefore should not be expected to lead to an attractive risk-adjusted return. Second, it is not likely that the same stock will persistently have the highest value exposure within a given investment universe. Therefore, a periodically rebalanced factor index with such an extreme level of concentration is likely to generate 100% one-way turnover at each rebalancing date, as the stock held previously in the strategy is replaced with a new stock that displays the highest current value exposure at the rebalancing date. Although practical implementations of concentrated factor-tilted indexes will be less extreme than this example, we can expect problems with high levels of idiosyncratic risk and high levels of turnover whenever index construction focuses too much on concentration and pays too little attention to diversification. Interestingly, the importance of diversification for a given factor tilt was outlined more than 40 years ago in Benjamin Graham s famous book on value investing: 2 In the investor s list of common stocks there are bound to be some that prove disappointing But the diversified list itself, based on the above principles of selection [ ] should perform well enough across the years. At least, long experience tells us so. Aiming at a highly concentrated value portfolio would be completely inconsistent not only with financial theory but also with the principles put forth by the early advocates of value investing. When constructing portfolios that contain, for example, a large number of value stocks, cap-weighted portfolios of value stock selections may at first seem to be more neutral implementations than equalweighted portfolios. It is well known, however, that cap weighting has a tendency to lead to very high concentration, given the heavy-tailed nature of the distribution of market cap across stocks in the same universe. It is well documented in the academic literature that simple cap-weighted, value-tilted portfolios have not led to attractive performance. In fact across different studies (see, e.g., Fama and French [2012, 2015a]), empirical results show that a value strategy needs to be well diversified to deliver a significant premium. For example, the standard Fama and French value factor includes a broad selection of stocks and uses a two-tiered weighting approach to obtain better diversification. In particular, the value factor is an equal-weighted combination of subportfolios for different market-cap ranges, Fall 2016 The Journal of Index InvesTIng

9 effectively overweighting smaller stocks and increasing the effective number of stocks. In fact, across different studies on equity risk factors (see, e.g., Fama and French [2012, 2015a]), Fama and French emphasized the need to have a well-diversified portfolio as a proxy for a factor tilt. Fama and French [2012] said, for example, that they ensure that we have lots of stocks in each [factor-tilted] portfolio and argued that factor-tilted portfolios need to be well diversified in order to obtain factor tilts that are reliable in the sense that factor exposures can be estimated with precision. 3 The need for diversification is explicitly recognized by Hou, Xue, and Zhang [2015], who recalled that value-weighted portfolio returns can be dominated by a few big stocks (see also Fama and French [2015a]). This need for diversification is reflected in how the factors are constructed. Fama and French [2012] defined the size factor as the difference between the returns on diversified portfolios of small stocks and big stocks, and the value factor as the difference between the returns on diversified portfolios of high book-to market (value) stocks and low book-to-market (growth) stocks. Thus, the value and size factors are not based on concentrated portfolios with the maximum factorrelated characteristic. Furthermore, Asparouhova, Bessembinder, and Kalcheva [2013] reviewed the literature and summarized that examining articles published in only two premier outlets, the Journal of Finance and the Journal of Financial Economics, over a recent five-year (2005 to 2009) interval, we are able to identify 24 articles that report EW mean returns and compare them across portfolios (see also Plyakha, Uppal, and Vilkov [2014]). As a recent example, Hou, Xue, and Zhang [2015] addressed the diversification issue by forming factor portfolios that equally weight their component stocks, while excluding the smallest stocks due to implementation concerns. An important idea behind factor investing is that portfolios that tilt toward a range of well-documented factors have led to a reward in terms of higher returns. It is, however, well known that expected returns are notoriously hard to estimate (see Merton [1980]). Thus, concentrating portfolios in stocks with the strongest exposure to factors that deliver the highest expected return is a risky business as far as reliability of out-ofsample data is concerned. Moreover, estimating returns at the individual stock level is likely to lead to a large amount of noise. Black [1993] emphasized that expected returns cannot be reliably estimated for individual stocks. For this reason, studies that document factor premia rely on portfolio-sorting approaches. Rather than trying to determine differences in returns between individual stocks, researchers have created groups of stocks and tested broad differences in returns across them. Therefore, when considering whether to create a concentrated portfolio or a diversified one for a given factor tilt, one should keep in mind that the more concentrated the portfolio becomes, the more one relies on the existence of a detailed and strict relationship between stock returns and factor exposure. If we recognize that there is a lot of noise in estimating the link between returns and stocklevel characteristics, the standard way of addressing this issue is to make broad distinctions between groups of stocks. One would thus strive to build a well-diversified portfolio for a given tilt. Overall, it thus appears that the approach that proposes to construct concentrated factor indexes is supported neither by the academic literature nor, for that matter, by common sense. On the contrary, there is a strong theoretical motivation for constructing welldiversified factor-tilted portfolios. Cochrane [1999] emphasized that any portfolio should be constructed so as to provide efficient risk return trade-off in a mean variance sense at a given level of factor exposure. Fama [1996] showed that rewarded factors can be understood to be multifactor mean variance-efficient portfolios themselves. Thus, although there are different ways of achieving diversification for a given tilt in practice (most notably by including a large number of stocks in the selection process and by using a weighting scheme that avoids concentration), no form of factor investing should completely neglect diversification. The academic literature makes a clear case for constructing portfolios that are not merely exposed to a given factor but also efficient. In short, maximizing factor exposure without regard for diversification is not a reasonable objective. Amenc et al. [2016] analyzed concentrated factortilted portfolios that use a narrow stock selection and a concentrated weighting scheme (such as cap weighting or score weighting) with more diversified factor-tilted portfolios that use a broad stock selection and diversified weighting schemes (such as equal weighting or diversified multi-strategy weighting). Their results suggest that trying to improve the performance of factor-tilted portfolios by selecting fewer stocks that are most strongly Long-Term Rewarded Equity Factors: What Can Investors Learn from AcademIC Research? FaLL 2016

10 tilted toward the factor does not have any positive effect on the risk-adjusted performance. Narrow stock selections improve returns compared with broad selections, but this increase is accompanied by higher volatility and higher tracking error, which keeps performance ratios the Sharpe ratio and information ratio approximately unchanged. In addition, Amenc et al. [2016] showed that factor-tilted portfolios on narrow stock selections present implementation drawbacks, such as higher turnover. Conversely, if one focuses on deconcentration by using an alternative weighting scheme to weight stocks, better Sharpe ratios and information ratios can be achieved while incurring only marginally higher levels of turnover. Their results suggest that well-diversified factor portfolios or indexes outperform their highly concentrated counterparts in terms of risk-adjusted performance, because concentrated factors may be highly exposed to unrewarded factors. These findings are in line with the importance of diversification that is emphasized from both a conceptual and an empirical perspective in the literature, as discussed earlier. In a nutshell, a sensible approach to factor investing should perhaps not only look to obtain a factor tilt, but also to achieve proper diversification within that factor tilt. Or, to quote Eugene Fama [2015]: 4 All we really say in finance is: Hold diversified portfolios along whatever tilt you choose. When considering practical implementations of factor tilts, it thus appears to be important that one does not focus only on gaining exposure to systematic risk factors (factor premia). In addition, basic finance theory suggests that one needs to consider diversification in order to avoid unrewarded risks. Taking additional unrewarded risks at a given level of factor exposure would still preserve the long-term rewards of the relevant factor exposure but introduce additional risk that is not rewarded. Ultimately, gaining exposure to factors that carry a reward in the long term is not the same thing as being concentrated in the right stocks. Based on the insights about diversification for a given factor tilt in the literature, one can distinguish between broadly diversified factor tilts that are in line with empirical asset-pricing results and highly concentrated stock picking strategies repackaged as factor investing. Ultimately, more thought should be given in practical applications to sufficient diversification for a given factor tilt. LESSON FOUR: IMPLEMENTATION COSTS CAN BE MEASURED A common criticism of academic research on factor premia is the supposed impracticality of academic factor definitions, simply because most results in academic research abstract from transaction costs and other implementation issues, such as turnover. It is indeed the case that many academic studies do not necessarily aim to consider implementation issues. In fact, product providers often justify deviations from academic factors with implementation needs. But while early studies indeed abstract away from implementation issues, recent academic research addresses this shortcoming. In particular, recent research examines whether the premia to common equity risk factors survive net of transaction costs. Moreover, it assesses whether we can use mitigation strategies to ease implementation when harvesting these premia. Novy-Marx and Velikov [2014] assessed turnover and estimate transaction costs for common factor strategies. They found that the net-of-cost factor premia mostly remain significant. Exhibit 4 provides a summary of their findings. In addition to assessing whether the returns to simple strategies are robust to transaction costs, research has tested adjusted implementations of factor premium strategies that try to ease implementation. Novy-Marx E x h i b i t 4 Net-of-Cost Factor Premia, as Reported by Novy-Marx and Velikov [2014] Notes: Results are extracted from Novy-Marx and Velikov [2014]; see their Exhibit 3. All values are monthly. Factors are based on capweighted decile portfolios. Portfolios are rebalanced annually for most factors but monthly for low idiosyncratic volatility and momentum. Factors are return differences between two extreme decile portfolios (cap weighted). Time period is July 1963 to December Fall 2016 The Journal of Index InvesTIng

11 and Velikov [2014] tested several such mitigation strategies and found that such approaches can substantially ease implementation while sustaining most of the return benefits, which often results in improvements in net-ofcost factor premia. Frazzini, Israel, and Moskowitz [2012] conducted a similar analysis and found that after taking into account realistic transaction costs, factor premia remain significant, especially when making adjustments to ease implementation: We measure the real-world transaction costs and price impact function and apply them to size, value, momentum, and short-term reversal strategies. [ ] Strategies designed to reduce transaction costs can increase net returns and capacity substantially, without incurring significant style drift. We conclude that the main anomalies are robust, implementable and sizeable. Moreover, Amenc et al. [2012] provided a clear implementation framework for factor-tilted indexes in a long-only context with an aim of providing factortilted indexes that are not only implementable but also well diversified. Practical implementations of such welldiversified indexes lead to risk return improvements over simple cap-weighted quintile portfolios, 5 as well as considerable investability improvements through lower turnover and fewer average days to trade at rebalancing (Amenc et al. [2016]). In summary, although much of the early evidence did not consider practical implementation issues, more recent research confirms that the standard factors lead to rewards even net of implementation considerations. Moreover, straightforward adjustments to strategy design that ease implementation lead to even more pronounced premia net of transaction costs. Therefore, there is a strong case that academically grounded factors can be used to design implementable strategies. Given this evidence, when considering deviating from academic factor definitions, investors should be careful to not throw out the baby (academic grounding) with the bathwater (unrealistic assumptions on implementation issues). Instead, the available academic evidence can be used to assess which factors are expected to deliver significant premia net of transaction costs. Moreover, it is perhaps surprising that relatively little analysis is provided on transaction costs of many factor index strategies. In practice, index providers do not offer much information on net returns of their strategies. A key lesson from academic research is that reasonable estimates of transaction costs can be derived and applied to strategies to calculate net returns. Ultimately, making well-informed investment decisions requires that one consider such evidence when using standard factors and that one produce similar evidence when introducing enhancements. LESSON FIVE: ASSESSING CROWDING RISK REQUIRES RESEARCH, NOT JUST ANECDOTES Smart beta has been establishing a space in between traditional (cap-weighted) passive investments and traditional (proprietary and discretionary) active management. Smart beta draws fierce criticism from providers of both traditional active management and traditional passive management. Perhaps unsurprisingly, advocates of traditional active and passive management find that smart beta is not quite to their liking. In a nutshell, proponents of proprietary active strategies complain that smart beta is not active enough (see, for example, Yasenchak and Whitman [2015]), while proponents of traditional cap weighting say that smart beta is not passive enough (see, for example, Philips et al. [2015]). Among such critiques, a recurring issue is the presumption of a risk of crowding in smart beta strategies. Although crowding is commonly pointed to as a potential risk, it is rarely formalized or even defined. This absence of definition is an issue when one wants to draw founded conclusions. Indeed, if it is not clear how crowding is defined or how it can be measured, it is rather futile to talk about whether or not it has or will occur. The main idea behind a crowding risk is that, as everyone learns about successful smart beta strategies and increasingly invests in them, flows into these strategies will ultimately cancel out their benefits. If an increasing amount of money starts chasing the returns to a momentum strategy, for example, it is possible that the reward for holding this strategy which has been documented with historical data will ultimately disappear. Given that the most popular smart beta strategies already have a wide following, it should be feasible to establish evidence of the negative effects, if they exist, of being followed by increasing numbers of investors. It should be feasible to analyze whether popular smart beta indexes have led to overcrowding and come up with an empirical estimate of the magnitude of the drag Long-Term Rewarded Equity Factors: What Can Investors Learn from AcademIC Research? FaLL 2016

12 associated with crowding that has occurred so far. As of today and to the best of our knowledge, there is no such evidence. This, of course, does not mean that such evidence may not be produced in the future, but it is important to ask what current claims about crowding are founded upon and the answer is often that we are in the sphere of unfounded assertions. Moreover, even when looking at the reasoning behind the supposed risk of crowding, one discovers several issues with the common wisdom about crowding. In the following, we first address the insights one can gain from considering the economic rationale of factor premia. We then turn to reviewing the empirical evidence on crowding and finally discussing practical considerations. Crowding Risk and Economic Explanations of Factor Premia Whether or not we should expect crowding in smart beta strategies is closely related to the economic explanations of the premia we observe in the data. If factor premia are explained by a rational risk premium, the factor premium is likely to persist, because some investors will rationally avoid a tilt despite the higher returns. Some investors may rationally choose to accept the lower returns of stocks that have good payoffs in bad times where marginal utility of consumption is high and may thus hold portfolios that go against rewarded factor tilts (e.g., by focusing on large-cap growth stocks). If, on the contrary, factor premia are because of systematic errors and investors learn over time to correct these errors, factor strategies may indeed see diminishing premia, except if there are limits to arbitrage that mean that many investors will not be able to benefit from the premium. This issue has been discussed extensively by, for example, Cochrane [1999]. Proponents of the crowding argument claim that crowding will occur in standard factors and factor premia will diminish, but this is not to be expected if factor returns are explained rationally. More specifically, the Lesson One section of this article discussed explanations that are available in the literature for why factor premia may exist on standard factors. Such economic explanations provide reasons for why factor premia should persist, even if investors are widely aware of them. Some investors will shy away from exploiting the premium even if they are convinced of its existence, simply because they are not willing to take the associated risks or because they are prevented by institutional constraints from going against biased behavior. Some who theorize about the existence of crowding argue that the losses occurring in a particular factor at some point in time are evidence of crowding. For example, Yasenchak and Whitman [2015] argued that given the increase in popularity of smart-beta strategies, there is a similarly increased overcrowding risk, which could result in factor crashes. Given such claims, one can ask whether the losses in a particular factor at a particular point in time are indeed evidence of crowding. Indeed, if a loss in a factor is proof of crowding one might as well claim that if the equity market factor has experienced crashes, there must be overcrowding in the cap-weighted equity index. And when long-term bonds severely underperform short-term bonds over a short period, is this then evidence of crowding by investors who are chasing the term premium? How can we conclude that such fluctuations are due to crowding rather than normal price fluctuations associated expected for risky assets that are completely uncrowded? Again, in the absence of a definition of crowding, it is not clear what one can conclude. If the argument is simply that returns vary over time and at times may be low, then it is not clear how factors are any different from, for example, the equity market. And the fact that returns vary over time does not mean that, in practice, investors are better off by accounting for such time variation through timing decisions rather than being exposed to the long-term premium (see Asness [2016]). In fact, claiming that there must be crowding in a factor because it suffers from losses completely ignores the nature of risk premia. A risk premium corresponds to a higher average return that is due to taking on additional risk. All risk factors will have returns that vary substantially over time, and only an analysis of longterm data can lead to any meaningful conclusions on the average premium. We should note with Black [1993] that we need decades of data for accurate estimates of average expected return. We need such a long period to estimate the average that we have little hope of seeing changes in expected returns. An example of the difficulties in concluding on changes of factor premia is the small-cap effect. There is a widely held belief that the size effect has disappeared after it has been widely published. Fall 2016 The Journal of Index InvesTIng

13 This is mainly based on empirical results over the period 1980 to 2000, however, and thus is sample-specific. For example, Schwert [2003] stated that it seems that the small-firm anomaly has disappeared since the initial publication of the articles that discovered it, and Hirshleifer [2001] stated that the U.S. small firm effect has been weak or absent in the last 15 years. Over the long term, however, Fama and French [2015] showed that the size factor is empirically important whether it is for the period or To illustrate that conclusions on disappearance of the small-cap premium can be sample-specific, Exhibit 5 shows the size premium (annualized return) and its associated t-statistic over rolling periods of 15 years. The results in Exhibit 5 suggest that, at times, one will tend to conclude that the premium has disappeared when looking at such time periods of 15 years. However, the widely popularized belief that the size premium has disappeared is not found in the most recent data. Moreover, while the size premium was lower during the 15 years after the publication of the seminal study by Banz [1981], the premium had also been low or negative in earlier times, such as the early 1960s. Therefore, the conclusion that the size premium has disappeared because of the wide publication of this phenomenon is not obvious from such empirical results. Perhaps such results merely suggest that returns to factors vary over time, and we need very long time spans of data to conclude on the significance of a premium. Even over relatively long time spans, it is difficult to conclude reliably that a factor has truly disappeared, given the variation in premia. A fortiori, claiming that factor premia have disappeared due to crowding based on short-term events is a risky business as far as the reliability of such conclusions is concerned. Indeed, due to fluctuations in average returns, it is expected that we will observe periods with low returns, and given the uncertainty in estimating returns reliably, any sample-specific conclusions should be handled with care. As an example, it is noteworthy that while the size factor is often claimed to have disappeared, and the value factor has been argued to be redundant based on sample-specific analysis, more general findings typically conclude that such factors are still relevant. For example, Fama and French [2015a] concluded based on a U.S. sample that the value factor is redundant, but Fama and E x h i b i t 5 Rolling 15-Year Average Annualized Return Premium of Size Factor Notes: The analysis is based on daily returns from July 1, 1926, to December 31, Starting from 1926, 15-year average annual returns are calculated by rolling each year forward till 2015 and the corresponding t-statistics are calculated. Source: U.S. size (SMB) factor returns are obtained from Kenneth R. French s data library ( data_library.html). Long-Term Rewarded Equity Factors: What Can Investors Learn from AcademIC Research? FaLL 2016

14 French [2015b] did not find evidence of the redundancy of the value factor in global data and caution that their earlier results on redundancy may be sample-specific. Moreover, comprehensive comparisons of multi-factor models including different sets of factors show that such factors as value and size need to be included to successfully explain the cross-section of expected returns (see Barillas and Shanken [2015]). In a nutshell, focusing on specific time periods is ill-suited to drawing inferences on the long-term behavior of factors. In fact, losses to any factor strategy over any particular period do not necessarily suggest that the long-term premium has disappeared because of crowding into a fashionable factor. Such losses may simply suggest that the reward for holding the factor comes with associated risk. Where Is the Evidence? Although there is no specific evidence on the crowding effects in particular smart beta indexes, a small number of recent studies examined the potential effects of wide use of common factors for which a reward has been broadly documented. Even though proponents sometimes cite such studies to substantiate their claims about crowding risk, it should be emphasized that recent studies do not provide clear evidence to suggest that factor premia are likely to disappear because of crowding. When inspecting the results in the unpublished working article of Yost-Bremm [2014], which is sometimes cited in support of the crowding theory, one does not find conclusive evidence that crowding effects impose any meaningful cost on factor investors. Even though the article finds evidence of abnormal trading volume for stocks that switch across thresholds of standard factor portfolios, the results do not necessarily imply a heavy burden or cost to strategies following standard factors. In fact, the evidence presented is strong for effects on trading volume but much weaker for effects on stock returns. In fact, if one considers, for example, the effects around stocks that switch into the value portfolio, the results suggest the following. The study reports an effect on trading volume that is significant and consistent. Volume in switching stocks tends to increase consistently and in a statistically significant manner across the different model specifications the author tests. However, the return effect is not very consistent. Thus, while the volume effects are consistently shown as positive and significant for stocks switching to the value portfolio, return effects are often insignificantly different from zero across the different model specifications, which is hardly strong evidence of an abnormal return phenomenon. Moreover, results in the article show that a small percentage of firms actually switch into the value portfolio so that any abnormal returns of switching stocks only apply to a small fraction of assets held. The overall effect on a value portfolio investor would be muted by the fact that most of the assets held are not switching stocks. McLean and Pontiff [2016] addressed the question of whether the publication of results showing a return premium associated with an equity factor destroys this premium going forward. Specifically, they analyzed the returns to almost 100 different strategies that tilt toward single or composite variables, such as accounting variables or return-based variables. It should be noted that the study included both consensual factors and less standard factors. Such nonstandard factors are based on variables such as firm age, corporate governance measures, inventory-related measures, seasonality, revenue surprises, changes in R&D spending, and analyst earnings forecasts. 6 The authors analyze the in-sample result for a return premium over the period used in the original study. They contrast this premium with the premium observed out of sample but before publication and with the post-publication premium up to today. If investors automatically crowd into factors once they know about the documented reward, one would expect the premia to decline after publication of the respective article. McLean and Pontiff attributed a 32% drop in returns to the publication effect. However, the authors also reject the hypothesis that post-publication anomaly returns decay entirely. The key conclusion is thus that while the publication of academic research tends to lower returns going forward, these premia do not disappear. It is noteworthy that this result is obtained when analyzing a large number of almost 100 factors, which include nonstandard factors. As one increases the number of factors, it may indeed be plausible that this may include ad-hoc factors with no clear economic rationale. Persistence of premia may arguably be even stronger when constraining the analysis to factors with a strong risk-based explanation. That the authors reject the hypothesis of disappearing rewards Fall 2016 The Journal of Index InvesTIng