سبک سرمایه گذاری،حرکت مشترک و قابلیت پیش بینی بازده

Barberis and Shleifer (2003) argue that style investing generates momentum and reversals in style and individual asset returns, as well as comovement between individual assets and their styles. Consistent with these predictions, in some specifications, past style returns help explain future stock returns after controlling for size, book-to-market and past stock returns. We also use comovement to identify style investing and assess its impact on momentum. High comovement momentum portfolios have significantly higher future returns than low comovement momentum portfolios. Overall, our results suggest that style investing plays a role in the predictability of asset returns.

Barberis and Shleifer (2003) present a parsimonious model in which investors allocate capital based on the relative performance of investment styles. Their model generates a rich set of predictions, some of which have received empirical attention. First, style-level return-chasing behavior generates both style and asset-level momentum. Barberis and Shleifer (2003) argue that the evidence in Moskowitz and Grinblatt (1999), Lewellen (2002), and Haugen and Baker (1996) is consistent with the profitability of style-level momentum (see also Teo and Woo, 2004). Second, they show that style investing generates excess comovement of assets within styles. Consistent with this, Barberis, Shleifer, and Wurgler (2005) show that when a stock is added to the Standard & Poor's 500 index, its comovement with the index increases (see also Greenwood, 2008 and Boyer, 2011). Finally, they show that style-based investing can generate momentum in individual asset returns at intermediate horizons and reversals at longer horizons. In their words: “If an asset performed well last period, there is a good chance that the outperformance was due to the asset's being a member of a ‘hot’ style... If so, the style is likely to keep attracting inflows from switchers next period, making it likely that the asset itself also does well next period” (pp. 183–184). It is this hitherto unexplored connection between style investing and asset-level return predictability that we investigate in this paper. A simple way to test whether style investing is responsible (at least in part) for asset-level return predictability is to see if past style returns have any predictive power in the cross section. We identify styles using the now ubiquitous size and value-growth grids, and then estimate Fama and MacBeth (1973) regressions of future stock returns on size, book-to-market ratios, past stock returns, and past style returns.2 We find that between 1965 and 2009, over one, three, six and 12-month future return horizons, style returns measured over the prior 12 months are significant predictors of future returns. We subject this basic result to a series of robustness checks. In some (but not all) specifications, style returns measured over the prior six months are also significant predictors. If we construct size breakpoints using NYSE stocks instead of all stocks, the slope coefficients on style returns are similar in magnitude and retain their statistical significance. If we limit our sample to all-but-tiny stocks (those above the 10th percentile in NYSE size), style returns remain statistically significant using 12-month prior style returns. However, if we use six-month prior returns, style returns are important only in explaining longer horizon future returns. We do not find predictability of past style returns among big stocks only (those above the median NYSE size), implying that style returns based on value-growth alone do not help explain cross-sectional variation in returns among large stocks. The slopes on style returns are stronger in the second half of our sample period (1988–2009). Prior to that, the coefficients of past style returns are mostly indistinguishable from zero. In that latter period, which coincides with increased use of size and value categorization in mutual funds and institutional portfolios, the slopes on style returns are large and reliably positive. Although the Fama-MacBeth regressions are suggestive of the role of style investing, a prediction of Barberis and Shleifer (2003) allows us to specifically identify its impact; namely, that style investing generates not only momentum but also comovement of a stock with its style. Comovement is an outcome of their model-not a primitive, but it serves as a valuable instrument for style investing. It frees us from treating all stocks as equally important to style investors because we can focus on stocks with extreme past returns and use a stock's comovement with its style to refine our assessment of the predictability induced by style investing. An added advantage is that comovement can be measured with precision, particularly compared with other measures of (aggregate) investor sentiment, behavioral biases, or style flows.3 Therefore, we implement a second set of tests that exploit this metric. If style-based investing generates asset-level momentum and comovement, then one should be able to use comovement to generate variation in momentum profits.4 Each month, we sort stocks into deciles (R1 through R10) based on past six-month returns (Jegadeesh and Titman, 1993). In the same month, we measure the comovement of a security with respect to its style by estimating its beta with respect to style returns over the prior three months (similar to Barberis, Shleifer, and Wurgler, 2005). Using these style betas, we independently sort all stocks into comovement terciles (C1 through C3). If the comovement metric is useful, a momentum portfolio that buys high comovement winners and sells high comovement losers should have higher returns than a momentum portfolio that buys low comovement winners and sells low comovement losers over intermediate horizons. We detect a monotonic relation between momentum profits and comovement. For example, for the six-month portfolio formation and evaluation period, the average winner minus loser (R10–R1) monthly portfolio return for the lowest comovement tercile (C1) is 0.71% per month. This increases to 0.96% for the second tercile (C2) and 1.15% for the highest comovement tercile (C3). The difference of momentum returns between C3 and C1 is large: 0.44% per month with a t-statistic of 2.98. Estimates of alphas based on the Fama and French (1993) model display a similar pattern. These return differences are generated from both the short and long side of the portfolio strategy. Winner portfolio returns increase and loser portfolio returns decrease as comovement increases. Our comovement-based tests drop tiny stocks and are robust to using value-weighted returns, dependent sorts, and measuring comovement using various windows, lags, and style cut-offs. Perhaps the most serious concern with the comovement-based tests is that of spurious correlations with other variables known to influence momentum. Size and book-to-market ratios are obvious candidates (Hong et al., 2000, Lakonishok et al., 1994, Asness, 1997 and Fama and French, 1996). Another possibility is that our comovement–momentum relation is just the volume–momentum relation in disguise. The Lee and Swaminathan's (2000) momentum life-cycle hypothesis in which stocks cycle through attention and neglect can be viewed as an asset-level analog of the Barberis and Shleifer (2003) style-level story. Our results might also conceivably be contaminated by the relation between idiosyncratic volatility and returns (Ang, Hodrick, Xing, and Zhang, 2006), or we might be inadvertently double sorting on past returns. We control for all these confounding effects individually through triple-sort procedures and jointly in a regression framework. For the latter, we form portfolios based on the component of comovement that is orthogonal to all these factors (from a first-pass regression). Individual controls sometimes influence the magnitude of the return differences in comovement-based momentum portfolios, but the gist of our results and their statistical significance remains. More important, the component of comovement that is (jointly) orthogonal to all these factors continues to generate economically important and statistically robust results. To us, this suggests that comovement explains variation in momentum beyond spurious correlations with the above variables. Our tests can never perfectly and precisely pin down whether the return patterns that we find are due to differences in risk or style chasing. Nonetheless, we can provide circumstantial evidence on the issue. The fact that our results are present in both raw returns and alphas suggests that (constant) risk differences are not responsible. We also find similar results in conditional tests in which we allow for time series variation in loadings that changes with the style composition of our portfolios. Moreover, similar to Jegadeesh and Titman (2001) and inconsistent with a simple risk explanation, our portfolios experience return reversals after the first year. If style investing is responsible for the relation between comovement and momentum profits, then the high comovement tercile should experience larger reversals than the low comovement tercile. This is precisely what we find, offering a degree of consistency between long horizon reversals in style returns (Teo and Woo, 2004) and asset-level returns. We also decompose each stock's total return into a style component (by multiplying its style beta with the style return) and a residual and then generate comovement tercile returns based solely on the style component. We find that approximately 50% of the risk-adjusted return difference between high and low comovement portfolios is explained by style effects. In addition, we use deviations of the R2 of a stock from the long-run mean R2 of its style (based on the style regressions described earlier) to measure “excess” comovement. Even though such a test has a look-ahead bias, if the long-run average R2 of a style represents comovement in cash flows or discount rates or both, then positive deviations thereof could be caused by style investing. Using this metric continues to generate return differentials between comovement terciles, again consistent with style investing. Finally, we show that our results cannot be replicated by assigning stocks into arbitrary styles, indicating that we obtain results because our style definitions are followed by investors. Our results cannot unequivocally reject other stock-specific rational or behavioral explanations for return predictability cited earlier or portfolio-based lead-lag explanations propounded by Lewellen (2002). But we do not seek to, and this inability does not belie the importance of our results. Our purpose is simply to determine if style investing has a role to play in return predictability. Considering the totality of the evidence, that appears to be the case. The remainder of the paper is organized as follows. Section 2 discusses style definitions. Section 3 contains the Fama-MacBeth tests, and Section 4 shows comovement-based results. Section 5 discusses alternative explanations. Section 6 concludes.

Style investing is ubiquitous. As retail investors have reduced the fraction of directly held equity, they have concomitantly increased their holdings of mutual funds, almost all of which are classified based on investment styles. Similarly, the vast majority of plan sponsor (institutional) allocations to equity are also based on investment styles (see, for example, Goyal and Wahal, 2008). In retail as well as institutional arenas, investors, investment advisers, and plan fiduciaries use size and value-growth metrics in comparing investment alternatives. Yet, with the exception of the studies cited in the introduction, academic attention on the impact of style investing on asset prices does not appear to be commensurate with its apparent importance to investors. In this paper, we investigate the role of style-based investing on asset-level return predictability. Our motivation for this undertaking is the remarkably simple prediction provided by Barberis and Shleifer (2003), namely, that under certain conditions, style investing can generate predictability in returns. Consistent with this, the profits of winner, loser, and long–short momentum portfolios are directly related to the comovement of a stock with its style. Fama-MacBeth regressions also indicate that past style returns have some predictive power over and beyond stock's own past return. As we have recognized above, we cannot conclude that rational or stock-specific behavioral biases are not responsible for predictability in returns. We can, however, conclude that investing behavior in which investors chase style returns amplifies the waves in asset returns.