Industry herding by hedge funds

Abstract

This paper investigates hedge fund herding at the industry level and its impact on industry returns. Although the level of industry herding on average is substantially weaker for hedge funds compared to non-hedge fund institutions, we find that industries that experience heavy herding by hedge funds experience return reversals in the long-run. We provide evidence that non-hedge funds especially follow hedge funds’ sell herding industries in following quarters, and the long-run return reversals observed in these industries are due to non-hedge funds’ failure to timely react to good news coming from these heavy hedge fund sell-herding industries in subsequent quarters.

KEYWORDS:

• Herding
• hedge funds
• industry returns
• mispricing

JEL Classifications:

• G10
• G11
• C13

Acknowledgement

We benefited from discussions with conference participants at 2019 Eastern Finance Association (EFA) meetings and seminar participants at Florida International University, Cleveland State University, and Duquesne University. This work was supported by Duquesne University (SOBA Summer Research Grant). All errors remain our responsibility

Disclosure statement

No potential conflict of interest was reported by the author(s).

Notes

1 Herding is usually defined as a group of investors trading the same securities at the same time or following each other’s trades. The most notable papers for the herding theory include: i) investigative herding: Froot, Scharfstein and Stein (1990) and Hirshleifer, Subrahmanyam, and Titman (1994); ii) informational cascades: Bikhchandani, Hirshleifer, and Welch (1992) and Welch (1992); iii) reputational herding: Scharfstein and Stein (1990) and Zwiebel (1995); iv) chasing certain characteristics or fads: Falkenstein (1996) and Friedman (1984).

2 The papers that provide empirical evidence of herding at the individual stock level include Lakonishok, Shleifer, and Vishny (1992, hereafter LSV), Grinblatt, Titman, and Wermers (1995), Wermers (1999), Sias (2004), Dasgupta, Prat, and Verardo (2011), and Brown, Wei, and Wermers (2014).

3 In addition to analyzing herding in the US equity markets, there are other studies that explore herding in other US markets and international markets as well. For example, Demirer and Kutan (2006) examine the herding in sectors of the Chinese stock market. Also, Demirer, Lee, and Lien (2015) focus on herding in different sectors by using futures contracts from US commodity markets.

4 Choi and Sias (2009) show that industry herding is not a manifestation of individual stock herding for all institutions. In addition, Celiker, Chowdhury, and Sonaer (2015) report similar evidence for mutual funds.

5 Choi and Sias (2009) report an average LSV (1992) herding measure of 1.39% (t-statistic = 34.66) for the total institutional investor sample during the 1983Q3 – 2005Q4 period.

6 While there are studies that argue that hedge funds may contribute to the destabilization of stock prices (e.g., Brunnermeier 2009; Stein 2009; Acharya et al. 2009; Khandani and Lo 2011), there is also a vast amount of literature which argues that hedge fund trading is informed and does not move prices away from fundamentals. Kokkonen and Suominen (2015), for example, provide evidence that the aggregate asset under management of hedge fund industry and flows to hedge funds are negatively related to the future mispricing at the market level. Sias, Turtle, and Zykaj (2016) find that hedge fund long equity portfolios are independent of each other and crowds of hedge funds does not destabilize the individual stock prices. In a similar vein, Cao, Liang et al. (2018) find evidence that hedge fund ownership improves efficiency except during liquidity crisis periods. Similarly, Cao, Chen et al. (2018) show that hedge funds contribute to the market efficiency through their stock picking ability in underpriced stocks.

7 More recently, Popescu and Xu (2014) provides empirical evidence that reputational concerns may contribute to the herding among institutional investors.

8 We filter out stocks with share codes other than 10 and 11. We also drop stocks with end of quarter share prices below $1.

9 It is worthwhile to point out that in this paper, we use TASS as the only source to identify hedge funds. Although TASS is one of the most comprehensive and widely used databases among the available hedge fund databases, such as CISDM/Morningstar, BarclayHedge, Bloomberg, and EurekaHedge, it does not include the whole universe of hedge funds. However, we do not expect the trading (herding) behavior at the industry level to be systematically different for hedge funds reporting to TASS compared to hedge funds reporting to other hedge fund databases. Furthermore, we also find evidence that hedge funds do not follow themselves as a whole group at the industry level in the next quarter. Therefore, we believe it is unlikely that the missing hedge funds in the TASS database introduce a significant bias to our results.

10 For the sake of brevity, in this paper, we refer to all 13F filers that we match with the TASS database as ‘hedge funds' rather than ‘hedge funds firms'. In order to identify and exclude fund families that offer both mutual funds and hedge funds from our analysis, we use the mutual fund management firm variable that is available in both TFN/CDA mutual fund and Thomson-Reuters’ 13F datasets. This variable is named MGRCOCD in TFN/CDA mutual fund dataset and MGRNO in Thomson-Reuters’ 13F dataset.

11 Yan and Zhang (2009) shows that trades of institutions with more frequent trading are better informed than long term institutional investors.

12 For each industry quarter, we calculate the expected value of $|p_{k,t}-p_t|$ under the null hypothesis of no herding by employing a binomial distribution for the number of buyers with $p_t$ as the probability of being a net buyer out of the active hedge funds (or non-hedge funds) in an industry.

13 As discussed in Section 3.3, different form LSV herding measure, this adjusted herding measure also indicates the direction of herding; a positive adjusted herding measure indicates that there are more buyers than expected, and a negative adjusted herding measure indicates that there are more sellers than expected. ADJHM makes sure that buy herding is always positive and sell herding is always negative, which makes it is easier to interpret the slope coefficients in Fama-MacBeth regressions.

14 In our analyses, we exclude observations (industry-quarters) where the number of traders is less than 5. There are only 8 industry-quarters during our sample period where the number of hedge fund traders is less than 5.

15 We also employ Fama-MacBeth regressions to examine the relationship between institutional herding and various industry characteristics, including size, book-to-market, past returns, sales growth, and asset growth for our sample of hedge funds and non-hedge funds separately. As LSV and Sias herding measures do not indicate the direction of herding, in this analysis we use current quarter (quarter Q₀) adjusted herding measure (ADJHM) of Brown, Wei, and Wermers (2014) and the ratio of number of buyers to total number buyers and sellers (p) as our dependent variables. We find that there is a positive and significant relationship between previous quarter industry returns and industry herding measures for our hedge funds sample. For non-hedge funds sample, we find that industry herding is negatively related to market-to-book ratio and positively related to sales growth. Our results also indicate that non-hedge funds do not pay attention to most recent returns but returns during the intermediate past. These results are not tabulated for the sake of brevity, but available upon request.

16 All 48 industries appear at least one time in the heavy-buy herding industries. Similarly, 47 industries appear at least one time in the heavy-sell herding industries. Probability of remaining in top buy (sell) herding industries for two quarters in a row is only 0.129 (0.104), suggesting that the persistence of the herded industries is quite low. These pieces of evidence indicate that the return pattern reported in this section is not driven by only a few industries.

17 Our inferences remain the same when we examine the returns over the immediate subsequent two quarters separately (as Q₁ and Q₂).

18 We also add the liquidity factor of Pastor and Stambaugh (2003) and the betting-against-the-beta factor of Frazzini and Pedersen (2014) to the Fama and French (1993) and Carhart (1997) four factors in the estimating the factor-adjusted abnormal returns of heavy-buy herding and heavy-sell herding industries, and find similar results to the ones reported in Table 4 Panel A. These results are provided in Table A1 of the Online Appendix.

19 We also compute the long-run industry returns over the Q₅–Q₈ period for the industries that experience heavy-buy and heavy-sell hedge fund herding in quarter Q₀. We find that the difference portfolio that buys heavy-buy herding and shorts heavy-sell herding industries delivers a return of −0.22% (t-stat=y-2.07) per month during the Q₅–Q₈ period. The corresponding CAPM and four-factor alphas for the difference portfolio are −0.20% (t-stat=−1.96) and −0.21% (t-stat=−2.02), respectively.

20 We also repeat this analysis by identifying the top five industries with the highest buy (sell) herding measures as heavy-buy (sell) herding industries rather than the top four. We get very similar results to those reported in Panel A of Table 4. For the sake of brevity these results are not tabulated, but they are available upon request.

21 These results are provided in Table A2 of the Online Appendix.

22 For a more detailed discussion on arbitrage asymmetry please see Miller (1977), Diether, Malloy, and Scherbina (2002), D'avolio (2002), Nagel (2005), Stambaugh, Yu, and Yuan (2012), and Antoniou, Doukas, and Subrahmanyam (2013).

23 For the sake of brevity, the results from the analysis using the value-weighted industry returns are not tabulated. These results can be obtained from the authors upon request. Moreover, the value-weighted return analysis results for our sample of non-hedge funds are also similar to the results obtained from equal-weighted returns, showing once again no evidence of long-term return reversals in industries that experience heavy herding by non-hedge funds.

24 As another robustness check, we also split our sample into two equal halves and check if our main findings hold true during sub-samples. Although not tabulated as a separate Table to save space, our inferences on long-term return reversals following the herds of hedge funds remain the same for these two sub-periods.

25 We perform the same Fama-MacBeth regression analysis for our sample of non-hedge funds. Consistent with the results of portfolio analysis, we find a significant positive relationship between institutional industry herding and concurrent industry returns, but no significant relationship between industry herding by non-hedge funds and Q₃ to Q₆ industry returns. For the sake of brevity, these results are not tabulated, but are available upon request.

26 We extend this analysis over the quarters Q₃ to Q₆ and find that non-hedge funds continue to herd in the same direction with hedge funds over this extended period, but not at the same level. On the other hand, we find no such evidence of herding by hedge funds in these heavy-buy and heavy-sell herding industries of non-hedge funds over the subsequent quarters Q₃-Q_6. For the sake of brevity, these results are not tabulated, but are available upon request.

27 Table A3 of the Online Appendix presents the percent of the appearance of the Fama-French 48 (FF48) industries in group 1 over our sample period. For convenience, the description section of Table A3 also provides a complete list of the 48 industries utilized in Fama-French 48 Industry Classification system. 32 industries appear at least one time in group 1 on the buy-side, and 39 industries appear at least one time on the sell-side. As an additional piece of information, we find that the probability for an industry being in group 1 in a given quarter given the industry was in group 1 during the previous quarter is only 0.078, for both heavy-buy herding and heavy-sell herding industries. Table A3 also reveals that a few industries that appear in group 1 multiple times on the buy-side never appear in group 1 on the sell-side (e.g., industry 32 (Communication), industry 44 (Banking)). We observe that these industries also appear among the hedge fund heavy-sell herding industries during several quarters over our sample period, but they are not followed by other institutions most intensely.

28 We also compute the long-run industry returns over the Q₅–Q₈ period for the industries that experience heavy-buy and heavy-sell hedge fund herding in quarter Q₀ that are most intensely followed by non-hedge funds in quarters Q₁ and Q₂. We find that the difference portfolio delivers a −0.43% (t-stat=−2.39) return per month during the Q₅–Q₈ period. The corresponding CAPM and four-factor alphas for the difference portfolio are −0.47% (t-stat=−2.46) and −0.48% (t-stat=−2.44) respectively.

29 These results are provided in Table A4 of the Online Appendix.

30 As a robustness check, we also include the liquidity factor of Pastor and Stambaugh (2003) and the betting-against-the-beta factor of Frazzini and Pedersen (2014) to the Fama and French (1993) and Carhart (1997) four factors in estimating the factor-adjusted abnormal returns. Alphas from these five-factor and six-factor models are similar to the ones reported in Table 7 Panel A. These results are tabulated in Table A5 of the Online Appendix.

31 Institutional managers managing $100 million or more must file 13F form (quarter-end holdings) within 45 days after the end of quarter. Therefore, for a more implementable strategy, we also compute the returns to these sell herding industries over the quarters Q₄ to Q₆ (skipping quarter Q₃). The average monthly raw and risk adjusted returns for this more practical strategy remain similar to those reported in Table 7 Panel A; 1.51% raw returns (t-stat=3.46), 0.63% CAPM alpha (t-stat=2.37), and 0.66% four-factor alpha (t-stat=2.88).

32 The return reversals observed over the quarters Q₃ to Q₆, which we report in Panel A of Table 4, do not necessarily indicate negative hedge fund skill for the following reasons. First, as presented in Table 1, hedge funds have substantially higher turnover rates than other institutions, suggesting that hedge funds do not hold on to their trades for a long time. Second, we show in Table 7 that only group 1 industries (the industries that other institutions follow hedge fund herding most intensely) experience return reversals. Third, we also examine the hedge fund herding in subsequent quarters (Q₁ and Q₂) for group 1 industries and find that the direction of hedge fund herding reverses in sign (negative adjusted herding measures for heavy-buy herding industries in quarters Q₁ (-0.41%, t-stat=−0.49) and Q₂ (-0.06%, t-stat=−0.07); and positive adjusted herding measures for heavy-sell herding industries in quarters Q₁ (0.83%, t-stat=0.94) and Q₂ (0.69%, t-stat=0.93)); but not statistically significant.

33 We repeat this analysis by splitting the top four heavy-buy and top four heavy-sell industries as 2 and 2 (as opposed to 1 and 3) such that group 1 industries consist of the two industries that non-hedge funds follow most intensely and group 2 industries consist of the remaining two industries. Our inference remains the same from this analysis as well. For the sake of brevity these results are not tabulated, but they are available upon request from the authors.

34 Note that generating SUEs at the industry level in different forms, either as the value-weighted average of the firm-level SUEs where weights reflect the market capitalization of firms in each industry, or calculating industry-level SUEs as the aggregate of firm-level SUEs in each industry, yield very similar results to industry-level SUEs obtained with the equal-weighted average of the firm-level SUEs.

35 When we examine the next quarter earnings of industries sorted by hedge fund herding (without conditioning on non-hedge fund trading), we find that industries that experience heavy-buy herding by hedge funds experience statistically significant positive average SUEs in the subsequent quarter consistent with the informed trading of hedge funds. However, we find no statistically significant average SUEs for industries that experience heavy-sell herding by hedge funds over the subsequent two quarters. On the other hand, we find that industries that experience heavy-sell herding by non-hedge funds have positive and statistically significant average SUEs over the subsequent quarters. In other words, industries subject to heavy-sell herding by hedge funds in quarter Q₀ experience positive earnings news in quarters Q₃ to Q₆ only when the same industries are subject to sell-herding by non-hedge funds in quarters Q₁ and Q₂. Panels A and B of Figure A1 in Online Appendix plot the average standardized unexpected earnings (SUEs) over the quarters Q₀ to Q₆ for industries that experience heavy-buy and heavy-sell herding by hedge funds and non-hedge funds in quarter Q_0, respectively.

36 To test whether non-hedge funds underreact to positive earnings news on the industries that hedge funds heavily sold in the previous quarter, we perform quarterly Fama-MacBeth regressions on non-hedge fund herding using the following model: NONHFADJHM _k,_q = b₀ + b₁ x HFADJHM_k,_q-1 + b₂ x SUE_k,,q + b₃ x HFADJHM_k,q-1 x SUE_k,,q + b₄ x HFADJHM _k,q-1 x SUE_k,q x Dummy_k,q, where NONHFADJHM _k,_q is the non-hedge fund adjusted herding measure for industry k in quarter q, HFADJHM_{k q-1} is the hedge fund adjusted herding measure for industry k in quarter q-1, SUE_k,_q is the SUE for industry k in quarter q, Dummy_k,q is a dummy variable that takes the value of zero when SUE_k,_q is positive and HFADJHM_k,q-1 is negative, and takes the value of 1 for all other three possible combinations for positive and negative values of HFADJHM_{k q-1} and SUE_k,_q., In this regression model, b₃ represents the relationship between non-hedge fund herding and the interaction term between HFADJHM_k,_q-1 and SUE_k,_q for the special cases when HFADJHM_k,_q-1 is negative and SUE_k,_q is positive. In line with our expectations, we find b₃ to be positive and marginally significant in this regression setup. The positive and significant b₃ coefficient suggests that when hedge funds sell an industry in the previous quarter and these industries receive positive news in the following quarter, non-hedge funds downplay the positive news by mimicking hedge funds’ previous quarter industry sell trades.

37 We also investigate whether the observed return reversals in the industries that experience heavy herding by hedge funds are just reversals in industry momentum. To examine this, we perform a similar analysis to that of section 5.1 (Table 4), however this time we identify top (bottom) four industries based on cumulative equal weighted returns over the quarters Q_-1 and Q₀ as winners (losers). We then compute the raw and abnormal returns to these winner and loser industries over the quarters Q₁ to Q₂ and Q₃ to Q₆ as in section 5.1. We find no evidence of return reversals in either winner or loser industries over the quarters Q₃ to Q₆, indicating that observed returns reversals are not due to reversals in industry momentum. These results are not tabulated for the sake of brevity, but available upon request.

38 We also examine the returns to the heavy-sell herding industries that non-hedge funds follow most intensely over the quarters Q₁ to Q₂ (i.e., group 1 industries), and find no evidence of negative abnormal returns in quarters Q₀, or Q₁ to Q₂.

Additional information

Funding

This work was supported by Duquesne University (SOBA Summer Research Grant). All errors remain our responsibility.