Industry Momentum and Mutual Fund Performance

Transcription

1 Industry Momentum and Mutual Fund Performance Jun Wu September 2015 Abstract This paper argues that the abnormal returns of industry-concentrated mutual funds are significantly contributed by industry momentum trading rather than managerial skill advocated by previous studies. First, I provide empirical evidence that industry-concentrated mutual funds have significant loadings on the industry momentum factor and no longer significantly outperform diversified mutual funds when industry momentum profits are considered. Second, using fund holdings I show that industry-concentrated mutual funds deviate from random industry concentration and have similar industry holdings to portfolios trading industry momentum. In addition, industry-concentrated mutual funds actively purchase stocks in the winning industries, which is also consistent with an industry momentum strategy. JEL Classification: G11, G20, G23 Keywords: Mutual Fund Performance Heterogeneity, Industry Concentration, Industry Momentum Strategy, Euclidean Distance This draft is preliminary. I would like to thank Veronika Pool, Robert Jennings, Noah Stoffman, Ryan Israelsen for helpful suggestions. Kelley School of Business, Indiana University, 1275 E 10th St, Bloomington, IN 47405; junwu@indiana.edu. 1

2 1 Introduction Why some mutual funds outperform the others is an important topic in investments. The performance heterogeneity among actively managed mutual funds have been documented in many dimensions. According to Kacperczyk, Sialm, and Zheng (2005) [22], mutual fund managers who concentrate their portfolios in a small number of industries outperform managers who diversify across all industries because industry-concentrated managers may have information advantages. Following the industry-concentration dimension, this paper investigates the source of outperformance. While Kacperczyk, Sialm, and Zheng (2005) argue industry-concentrated mutual fund managers possess skills attributed to their industry level private information, this paper extends this strand of literature and shows that industry momentum strategies are an alternative to explain the outperformance of industry-concentrated mutual funds. Specially, I find empirical evidence that industry-concentrated mutual funds are active in trading industry momentum. Industry momentum signficantly explains up to 20 basis points per month for the excess returns of the most industry-concentrated mutual funds by decile from 1984 to Many investment strategies can contribute to high fund returns even in the absence of private information. This idea is reflected in the Carhart (1997) [4] four-factor model, which uses market beta, a small-minus-big (SMB) size factor, a high-minus-low (HML) book-to-market factor, and an individual stock momentum factor to explain mutual fund returns. Industry momentum strategy is defined as buying the winning industry and short-selling the losing industry based on sorted cumulative returns of industries in the past few months and holding the portfolio in the following several months. Moskowitz and Grinblatt (1999) [26] find that industry momentum generates positive and significant profits. The strategy profits mainly come from the buy side of the strategy in contrast to individual stock momentum profits, which are strong on the sell side. Since mutual funds generally cannot short-sell, 1 they would pursue such investment strategy to generate positive returns by buying stocks in the winning industry. Trading industry momentum requires the fund manager to deviate from a well-diversified portfolio and have high industry concentration. In addition, mutual fund managers who trade industry momentum will have high 1 Almazan et al. (2004) [1] find that only about 30% of mutual funds are allowed to sell short, and only 2% actually do sell short. 2

3 abnormal returns not explained by standard performance evaluation models such as the Carhart (1997) four-factor model. Investing in industry momentum is becoming increasingly popular with the prevalence of Exchange Traded Funds (ETFs) as industry ETFs allow investors to effectively track the performance of the entire industry. Industry momentum strategy, also known as industry rotation and sector rotation in the mutual funds industry, has brought high returns to investors and attracted public attention. As reported by a Wall Street Journal article, the strategy performance in 2013 was attractive. Industry-momentum portfolios - which jump into and out of various sector funds in an attempt to profit from those performing best at any given time - had a very good In fact, nine of the 10 most profitable mutual fund portfolios last year - out of the more than 250 monitored by the Hulbert Financial Digest - pursued such a strategy, also referred to as industry rotation. These nine produced an average gain of 41%, in contrast to a 32% return for the S&P 500, assuming dividends were reinvested. -Mark Hulbert, How to invest in the hottest stock sectors. Wall Street Journal Marketwatch, Jan 24th, 2014 I use two approaches to show that the abnormal returns of industry-concentrated mutual funds are contributed by industry momentum trading. First, in a returns-based approach, I regress the gross returns of a large sample of equity mutual funds on an industry momentum factor and the Carhart (1997) four factors. The industry momentum factor is constructed from screening for the most profitable industry momentum strategy during the sample period (unconditional)or at a given time (conditional). I find industry-concentrated mutual funds have significant loadings on the industry momentum factor. Additionally, they no longer significantly outperform diversified mutual funds when the industry momentum factor is added to the fourfactor model. These results suggest that the abnormal returns of industry-concentrated mutual funds are driven by industry momentum trading rather than private information advantages. One potential concern with the returns-based approach is that the industry momentum factor may be correlated with some unobserved factors that are correlated with mutual fund 3

4 returns. To address this issue, I also use a holdings-based approach to investigate the role of industry momentum in the portfolio of industry-concentrated mutual funds. In particular, I calculate the Euclidean distance between funds industry percentage holdings and those of a hypothetical portfolio trading industry momentum. 2 To interpret Euclidean distance by industry concentration deciles, I generate a random benchmark of Euclidean distance from a placebo test that assumes random industry concentration. 3 I find that industry-concentrated mutual funds significantly deviate from the random benchmark and have smaller Euclidean distance. In addition, industry-concentrated mutual funds purchase significantly more stocks in the winning industry compared to diversified mutual funds. The holdings evidence supports the finding that industry-concentrated mutual funds are active in trading industry momentum. The rest of the paper is organized as follows. Section 2 contains a review of related literature. Section 3 introduces the empirical methodology. The results from the returns- and holdingsbased tests are shown in Section 4. Section 5 concludes. 2 Literature My research connects the study of momentum strategies with mutual fund performance heterogeneity. Individual stock momentum strategy, defined as buying the winning stocks and short-selling the losing stocks based on sorted cumulative returns in the past few months, has been proven to be profitable consistently. 4 Moskowitz and Grinblatt (1999) find momentum also exists in industry portfolios. There are several differences between industry momentum and individual stock momentum. One of the most important distinctions is that industry momentum profits mainly come from the buying of winning industry stocks. 5 Therefore, it is attractive for mutual funds to invest using industry momentum. Despite the strategy s appeal and popularity in the mutual fund industry, industry momentum has not generally been incorporated in studies of fund performance in the mutual fund 2 Despite that holdings data from Thomson Reuters are quarterly, I assume the strategy is determined at month end. The hypothetical portfolio vector averages the winning industries in the holding months. 3 The Placebo test is motivated to avoid the convexity bias of the measure. 4 see Jegadeesh and Titman (1993) [19]. 5 In contrast, individual stock momentum strategy profits are greatly contributed by short-selling the losing stocks (Moskowitz and Grinblatt (1999)). 4

5 literature. The only exception is Busse and Tong (2012) [3] who add an industry momentum factor to the Carhart (1997) four-factor model to test the role of industry momentum in mutual fund returns. In contrast, my paper uses industry momentum to explain the outperformance of industry-concentrated mutual funds, which has been widely attributed to managerial skill in previous studies. I follow Moskowitz and Grinblatt (1999) to construct the industry momentum portfolios and screen for the most profitable strategy as a factor to explain mutual fund returns. 6 I find mutual funds have significant loadings on the industry momentum factor, indicating that mutual funds are active industry momentum traders. This finding still holds when the strategy profits from the lead-lag effects are excluded. 7 This paper is closely related to the literature on mutual fund performance heterogeneity. Several studies argue that mutual fund managers who follow active investment strategies possess stock picking skills. 8 Recent studies document that mutual fund performance heterogeneity exists in multiple dimensions. For example, Coval and Moskowitz (1999, 2001) [9] [10], and Pool, Stoffman and Yonker (2012) [27] show that mutual funds have a strong geographical bias for investing in locally headquartered firms and firms from managers home states where they may have information advantages. Cheng, Hong, Huang, and Kubik (2004) [6] find that fund size erodes fund performance because of association with liquidity and organization diseconomics. More importantly, Kacperczyk, Sialm, and Zheng (2005) document that fund managers who concentrate their holdings in a few industries outperform managers who diversify their holdings across many industries and conjecture that the outperformance is due to information advantages of industry-concentrated fund managers. My paper extends Kacperczyk, Sialm, and Zheng (2005) using both the orignal and extended sample. It proposes industry momentum as an alternative to explain the outperformance of industry-concentrated mutual funds. I find 6 I screen the most profitable industry momentum strategy because mutual funds can easily replicate this process and invest according to the most profitable strategy. This requires no skill or private information. 7 Lo and MacKinlay (1990) [25] find industry level portfolios exhibit strong and positive autocorrelation, which is also known as lead-lag effects. Moskowitz and Grinblatt (1990) show that industry momentum profits are positive and significant when lead-lag effects are excluded by skipping one month between portfolio formation and the start of holding the portfolio. 8 Evidence is from gross fund returns and holdings returns. For example, Grinblatt and Titman (1989,1993) [16] [17], Grinblatt et al. (1995) [18], Daniel, Grinblatt, Titman, and Wermers (DGTW) (1997) [11], Wermers (2000) [29], Frank et al. (2004) [14], and Busse and Tong (2012) show mutual funds can outperform their benchmarks based on gross returns. However, the literature of analyzing fund net returns documents that mutual funds underperform passive benchmarks (Carhart (1997)). Therefore, I focus on fund gross returns in my analysis. 5

6 empirical evidence that industry-concentrated mutual funds are active in trading industry momentum and they no longer significantly outperform diversified mutual funds when industry momentum profits are considered. My findings also contribute to the rising literature on mutual fund manager market timing skills. Jiang, Yao, and Yu (2007) [21] provide empirical evidence from bootstrapping fund holdings returns and find that mutual funds have market timing skill at intermediate horizons. Specially, market timing skill at 6-month horizon generates 0.27 % monthly returns. I obtain supporting evidence that the returns from market timing skills partially come from trading industry momentum. 9 My findings indicate that industry momentum trading significantly contributes to the high abnormal returns of industry-concentrated mutual funds. I am not trying to distinguish momentum from private information. Actually, momentum profits may be one consequence of private information 10 (see Vayanos and Woolley (2013) [28]). However, the way I construct the industry momentum factor makes it uncorrelated with contemporaneous private information. 11 I argue that the outperformance of industry-concentrated mutual funds is not from the private skill or information that fund managers possess but from a well-known trading strategy. 3 Empirical Methodology 3.1 Data and Industry Construction I use the Center for Research in Security Prices (CRSP) data to get stock returns, form industry portfolios, and construct the industry momentum factor. All stocks from NYSE, Nasdaq and AMEX are included. Stocks returns are value-weighted using lagged market capitalization at the end of previous month to form industry portfolios. Mutual fund size, returns, and expense 9 Please see the analysis in section 4.1 for more detail. 10 The source of industry momentum remains unknown. Research conjectures that excess stock covariation (Lewellen (2002) [24]), business cycle (Chordia and Lakshmanan (2002) [7]), analyst coverage (Boni and Womack (2006) [2]), and economic linkage among firms (Cohen and Frazzini (2008) [8]) may contribute to industry momentum profits. 11 In the context of Vayanos and Woolley (2013), private information has to drive industry portfolios returns continuously for 9 months to 12 months to affect the ranks of industry portfolios and correlate with the industry momentum factor. This is unlikely as private information would be soon incorporated into asset prices provided that the market is efficient. 6

7 ratios are collected from the CRSP Mutual Fund Database. Following Kacperczyk, Sialm, and Zheng (2005), I eliminate balanced, bond, index, international, and sector funds and focus on all equity mutual funds in the market. 12. Share class level data are value-weighted to fund level using lagged weights. The quarterly fund holdings are from Thomson Reuters Mutual Fund Holdings. 13 To get a sample free of index funds and sector funds, funds with over 300 stock holdings are dropped. A name search for key words for industries is conducted to eliminate unidentified sector funds. The observation is deleted if the holding record date (rdate) and file date (fdate) is not in the same quarter in Thomson Retuers Holdings to prevent using lagged holdings information. I provide empirical evidence using two sample periods. The first sample period is from 1984 to 1999 and aims to replicate the performance heterogeneity documented by Kacperczyk, Sialm, and Zheng (2005). The second sample period extends to 2013 and is reported as the main results. 14 From CRSP data, 10 value-weighted industry portfolios are formed monthly using CRSP twodigit Standard Industrial Classification (SIC) codes 15. Each industry portfolio is constructed by value-weighted stocks using lagged weights. I use a 10-industry classification because I find that mutual funds that clearly state that they use industry momentum strategies divide stock holdings into 8 to 14 different industries or sectors. Also, I want to replicate the main finding of Kacperczyk, Sialm, and Zheng (2005) in which they use a 10-industry classification. 16 Table I reports summary statistics for the 10 industries. The average number of firms, average market capitalization, and monthly returns in excess of the contemporaneous three-month t-bill rate are reported. The abnormal returns are the industry portfolio returns adjusted for matched size and book-to-market benchmarks where every stock within the industry is matched with a diversified portfolio of similar size and book-to-market ratio, and the stock returns in excess of 12 This is because I use stock characteristics-based factors to explain mutual fund returns. 13 I use MFLINKS to map CRSP Mutual Funds with Thomson Reuters Holdings Database. 14 The returns-based and holdings-based results for the original sample period ( ) are reported in section 4.4. If not specified, the results in the main section refer to the extended sample ( ). 15 CRSP reports time-series of industry classification codes, which allow for time-series variation in industry classification. 16 The 10-industry classification is the same as in Kacperczyk, Sialm, and Zheng (2005). Appendix B.1 provides detailed classifications. 7

8 the matched benchmark returns are value-weighted as industry abnormal returns. 17 According to Table I, the unconditional abnormal returns do not exist intrinsically for industry portfolios. Summary statistics of mutual funds are reported in Table II. The goal is to have a close sample compared to the sample of Kacperczyk, Sialm, and Zheng (2005). Comparing the replicated sample statistics with the one in KSZ (2005), fund age, expenses, turnover, load, quarterly returns and industry concentration index are very close. Average stock holdings is smaller as I eliminate funds with over 300 stock holdings to drop unidentified index fund. The replicated sample end up with about 200 more funds than KSZ (2005) has. This is because the traceback update of MFLINKS. 18 In general, fund characteristics are very close to the sample in Kacperczyk, Sialm, and Zheng (2005). 3.2 Industry Momentum Mutual Funds The goal of this paper is to answer whether industry-concentrated mutual funds trade industry momentum and whether industry momentum explains the outperformance of the industryconcentrated mutual funds. The main challenge is to identify mutual funds that trade industry momentum. Since high-frequency mutual fund trades cannot be directly observed, I need a proper methodology. I first search fund names and fund prospectuses with Industry Momentum as the key word and find several funds who specialize in trading industry momentum. An example of a fund prospectus is provided in Appendix A.1. Five funds from my sample are selected as they clearly indicate trading industry momentum via their fund name, fund prospectus, or a third party media report. 19 Table III provides the most recent performance and fund characteristics for the five selected 17 I adjust industry portfolio returns for size and book-to-market effects by first sorting individual stocks into size quintiles and then, sort stocks into book-to-market quintiles within each size quintile. The stocks are valueweighted in these quintiles. Then, stock j is matched with one of the 25 portfolios based on its size and bookto-market ratio at time t 1. I calculate the return of stock j minus the return of the matched portfolio at time t. I use this characteristic adjusted approach rather than a factor loadings approach because Daniel and Titman (1997) [13] find that characteristics better capture the cross-sectional variation in average returns than factor loadings. 18 MFLINKS has updated new matches since 2003 and the new matches show up in the earlier period. As I use the 2015 MFLINKS, the 200 new funds mainly come from the newest fund matches from 2003 to The selection of funds as a sample in Table III also depends on whether the fund has a recent filed fund prospectus. 8

9 funds from their prospectus. The asset scale of the selected funds ranges from 16 million to over 1 billion US Dollars. Four of the five funds outperform the S&P 500 benchmark in terms of most recent annual return. These funds have annual portfolio turnover of greater than 100%, which means they completely change their holdings every year. The self-reported alphas are positive and the sharpe ratios are greater than the benchmark, indicating the profitability of the selected funds. Fund inception years show that these funds have been in existence in the market for many years. Some funds clearly state that they invest using industry momentum in their fund prospectus, which makes them easy to be identified. For example, the Rydex Sector Rotation Fund states in their investment strategy: Seeks to provide long-term capital appreciation by investing in different sectors or industries using a momentum strategy. 20 However, I find few funds that clearly state industry momentum industry rotation or sector rotation. Some of them only state the fund has an investment objective of long-term capital appreciation in general. 21 Also, mutual funds that trade industry momentum belong to the categorization of technical allocation funds. Defined by Morningstar, technical allocation funds are mutual funds that actively shift holdings between different asset classes. However, these is no existing identification code for this category in the database. I try to see whether self-disclosed industry momentum mutual funds have high level of industry concentration. I start with a name search for funds whose name contains transector, sector rotation, sector rotational, sector analysis, select selectors, sector allocation, industry select/selection, industrial select/selection, IND SEL, and select fund. These funds are considered potential industry momentum traders as they disclose the key words in their fund names. According to Table III panel B, these funds have high level of industry concentration and belongs to high industry-concentration deciles as defined by Kacperczyk, Sialm, and Zheng (2005). As the population of mutual funds that actually trade industry momentum could be much bigger, I need a better identification strategy to investigate whether mutual funds, especially industry-concentrated mutual funds, are trading industry momentum. 20 The prospectus of Rydex Sector Rotation Funds is available here: 21 From the investment objective for Virturs AlphaSector Rotation Fund. Fund prospectus can be found here: AlphaSector%20Rotationsummary.pdf 9

10 3.2.1 Regression Approach To solve the identification problem, I use a returns-based regression approach. An industry momentum factor representing the profits of trades according to an industry momentum strategy is constructed. Fund gross returns are regressed on the industry momentum factor and the Carhart (1997) four factors. The regression equation is defined as: R i,t R F,t = α i +β i,m (R M,t R F,t )+β i,smb SMB t +β i,hml HML t +β i,mom MOM t +β i,im IM t +e i,t, (1) where i denotes mutual fund, t denotes month. R i,t R F,t is the monthly fund gross return minus three-month T-bill rate. The Carhart (1997) four factors are market beta (R M,t R F,t ), smallminus-big (size) portfolio returns (SMB t ), high-minus-low (book-to-market) portfolio returns (HML t ), and individual stock momentum portfolio returns (MOM t ). Consistent with the other factors as zero-cost portfolios, the industry momentum factor (IM t ) represents profits of buying the winning industry and short-selling the losing industry according to previous months sorted returns and holding the portfolio afterwards. The industry momentum factor is constructed from screening the most profitable industry momentum strategy. α i stands for the intercept, also known as the abnormal returns contributed by managers private skill or information advantages. The error term is denoted as e i,t. If a mutual fund trades industry momentum, the fund s returns would be significantly explained by the industry momentum factor. According to Kacperczyk, Sialm, and Zheng (2005), industry-concentrated mutual funds have significant abnormal returns in the Carhart (1997) four-factor model. Adding the industry momentum factor into the Carhat (1997) four-factor model, I expect the fund returns would be significantly explained by the industry momentum factor as strong evidence that industry-concentrated mutual funds are active industry momentum traders. Additionally, a reduction of significance of the intercept (abnormal returns) is expected. For robustness checks, the industry momentum factor would be constructed to consider leadlag effects (month skipped industry momentum factor) and forward-looking bias (conditional industry momentum factor). 10

11 3.2.2 Holdings Approach Although I may find that mutual funds have significant loadings on the industry momentum factor, it is possible that the industry momentum factor is correlated with an unobservable factor that determines fund returns. Therefore, I calculate the Euclidean distance of each fund s portfolio from that of a benchmark portfolio described in the next paragraph to provide more direct evidence from fund holdings. Euclidean distance is a mathematical term in geometry, also known as Euclidean matrices. It is commonly used to measure the distance between two vectors in a Euclidean space. I apply Euclidean distance to measure the distinction between quarterly fund holdings with an industry momentum benchmark vector. The industry momentum benchmark vector simulates holdings of a portfolio trading industry momentum. Euclidean distance has been widely used in empirical asset pricing studies. For example, Levy and Roll (2010) [23] calculate the Euclidean distance between sample parameters and true parameters to indicate market mean variance efficiency. To the best of my knowledge, this is the first time this measure is used in investments to compare holdings similarity. The Euclidean distance is defined as: d i,t (h i,t, h IM,t ) = 10 (h IM,j,t h i,j,t ) 2, (2) j=1 where h i,j,t denotes fund i s industry percentage holdings in industry j at quarter t. h IM,j,t is a benchmark vector that simulates the holdings of a portfolio trading industry momentum. The benchmark vector assumes that the industry momentum strategy is determined monthly and averages the winning industries during the holding months. 22 As the vector of industry percentage holdings sums up to one and spans a ten-dimensional Euclidean space, 23 the measure cannot be directly compared across different industry-concentration groups. The vector constitutes a convex cone that sets on the positive orthant of a ten- 22 I assume the winning industry is determined monthly and averages in the holding months. For example, using a IM(9,9) strategy requires the investor to sort the industry portfolios based on previous 9 months cumulative returns and hold the portfolio in the following 9 months. As the winning industry is determined monthly, each month s winning industry weights 1/9 of the benchmark vector. If industry 3, 4, and 5 each was the winning industry during the 9 month holding period, the benchmark vector is (0,0,1/3,1/3,1/3,0,0,0,0,0). This method captures the average effect of determining the winning industry when there might be a time difference of applying the strategy. 23 This is because I use the 10-industry classification. 11

12 dimensional Euclidean space. 24 I conduct a Placebo test which randomly shifts the ten industry percentage holdings for each fund-quarter observation and recalculate the Euclidean distance. This process is repeated for 1,000 times. The mean of the 1,000 Euclidean distance measures serves as a random benchmark for each industry-concentration decile. The random benchmark of the Euclidean distance gives a value that reflects random selection of industry concentration, which helps to interpret the Euclidean distance for various industry-concentration deciles. I use net purchases of stocks in the winning industry to provide additional, direct evidence. If mutual funds trade industry momentum, they have to actively shift their holdings to the winning industry when the winning industry changes from time to time. While Euclidean distance measures the state of the holdings, net purchases focus on the changes in quarterly holdings. I calculate stock net purchases as a percentage of winning industry market capitalization: Net P urchase i,t = S m=1 [(N m,t N m,t 1 ) P m,t ], (3) Market Cap J t,t 1 where stock m belongs to the wining industry at quarter t, which is denoted as J t. The numerator summarizes the change of total holdings of stocks in the winning industry from t 1 to t. The numerator is the lagged total market capitalization of the winning industry. This measure has been used widely in empirical investment studies (see Cella, Ellul and Giannetti (2013) [5]). The net purchase of fund i at quarter t is calculated for each fund and summed up for each industry-concentration decile. 4 Empirical Evidence I start with constructing the industry momentum factor and running the regression in equation (1). Then, I turn to the holdings-based approach using Euclidean distances and net purchases, which serve as direct evidence to support the regression findings. 24 The reason is industry percentage holdings are non-negative. For the proof of convex and closed cone, please see Glunt et al. (1990) [15]. 12

13 4.1 The Industry Momentum Factor Winners-minus-losers industry momentum investment strategies are formed by ranking the 10 industry portfolios based on their previous L-month cumulative returns and forming a zero-cost portfolio of buying the highest past L-month return industry funded by short-selling the lowest return industry. The portfolio is then held over the next H months. The average monthly returns in the holding period (H months) are reported as strategy profits. This approach is also used by Jegadeesh and Titman (1993) and Moskowitz and Grinblatt (1999). I choose L and H to be combinations of 1-month, 3-month, 6-month, 9-month and 12-month time periods because industry momentum profits decrease after one year. Therefore, I examine these 25 distinct industry momentum strategies. Furthermore, I follow Moskowitz and Grinblatt (1999) and skip one month between portfolio formation and the start of holding to eliminate potential lead-lag effects on profits. Table IV reports the trading profits of industry momentum strategies. In Panel A, I find that the most profitable industry momentum strategy is based on sorting by the previous 9- month cumulative industry returns and holding the portfolio in the following 9-month (denoted as IM(9,9)), which is close to the findings in Moskowitz and Grinblatt (1999). 25 An IM(9,9) strategy generates an average abnormal return of 0.87% per month from 1984 to Month skipped strategies aims to reduce the contribution of lead-lag effects. The most profitable strategy is based on sorting by the previous 12-month cumulative industry returns, forming the portfolio, skipping one month, and holding the portfolio in the following 6 months (denoted as IM*(12,6)). 26 As shown in Panel B of Table IV, 0.80% per month is generated by the most profitable month skipped industry momentum strategy. Since there are 25 distinct industry momentum trading strategies in each Panel of Table IV, I need to find the most profitable strategy and take it as the industry momentum factor to explain fund returns. The reason to focus on the most profitable strategy is that mutual funds managers are very likely to follow the same sorting procedure to identify the most profitable industry momentum strategy. The filters below are used to find the most profitable strategy: 25 They find that the most profitable strategy is IM(6,6) from a sample. 26 I use * to denote the strategies with skipping one month throughout the paper. 13

14 The strategy delivers positive and significant total profits and buy-side profits The strategy delivers the greatest total profits and buy-side profits among all strategies in each Panel The sell-side profits of the strategy are insignificant With the above filters, I end up with IM(9,9) and IM*(12,6) as the most profitable strategies. This approach is distinct from the industry momentum factor in Busse and Tong (2012) where they choose IM(12,12) as the industry momentum factor so that its horizon is consistent with the Carhart (1997) four factors. I report a summary of the profitability of IM(9,9) and IM*(12,6) in Panel A of Table V. The most profitable strategies for the original sample ( ) are IM(9,9) with 0.87% per month and IM*(12,6) with 0.80% per month. The most profitable strategies for the extended sample ( ) are IM(9,6) with 0.58% per month and IM*(9,6) with 0.48% per month. The buy-side profit significantly contributes to the strategy total profit. To further understand the profitability of the strategy, I plot the cumulative returns after the portfolio formation for strategies based on 6-month, 9-month, and 12-month sorted returns in Figure 1. As shown in Figure 1, I find the 9-month strategies (denoted as IM(9,X)) dominate other strategies in terms of average profitability. The strategy profits reach peak at around the 12-month horizon after portfolio formation with the highest average return of 8.47%. Then the profits start to decrease after one year and turn negative around 24 months since portfolio formation. My findings regarding the profitability of the industry momentum strategy is consistent with Moskowitz and Grinblatt (1999) where they find industry momentum profits decrease and become insignificant after one year. 27 Therefore, I focus on regressing mutual fund returns on the IM(9,6) industry momentum factor using the extended sample and the Carhart (1997) four factors. IM*(9,6) serves as a robustness check. Screening for the most profitable strategy as of today may suffer from forward-looking bias, which is caused by using future information to select the most profitable industry momentum strategy. To exclude forward-looking bias, I use information in the previous five years (t 19 to 27 Similarly, Jegadeesh and Titman (2001) [20] show that individual stock momentum cumulative returns reach their peaks at around 12 months and start decreasing after 12 months. But the cumulative returns become negative at about 60 months (See Figure 3 of Jegadeesh and Titman (2001)). Industry momentum profits exist in a shorter horizon compared to individual stock momentum profits. 14

15 t) for quarter t to find the conditional most profitable strategy. The monthly average holding period return of the conditional most profitable strategy is regarded as the conditional industry momentum return at quarter t. This approach allows strategy horizon to change over time and uses the most recent information without forward-looking bias. It is closer to reality of what a fund manager could do when he/she chooses the most profitable investment strategy at a given time. Finally, to consider the strong contribution of the buy-side profit and its attractiveness for mutual funds to buy and hold the winning industry stocks, I give equal weights to the total strategy profits (Wi-Lo) and the buy-side strategy profits (Wi-Mid) when select the most profitable strategy conditionally. Also, strategies with less than 3-month holding period horizon are dropped to consider transaction cost of frequent portfolio turnovers. Table V Panel B reports the average actual buy-side profit, the average sorted buy-side profit and the average sorted total profit with and without skipping one month using the original and the extended sample. The actual buy-side profits are constructed as the conditional industry momentum factors and denoted as IM(Cond) for no gap strategies and IM*(Cond) for month skipped strategies. 4.2 Regression Result I first replicate the main result of Kacperczyk, Sialm, and Zheng (2005) that industry-concentrated mutual funds significantly outperform diversified mutual funds. The quarterly Industry Concentration Index (ICI) is constructed as the sum of the squared deviations of the value weight held by the mutual fund (denoted as w j,t ) from the industry weight of the total market (denoted as w j,t ) for each industry j: ICI t = 10 j=1 (w j,t w j,t ) 2, (4) The index measures the deviation from the market portfolio. It equals to zero if the mutual fund has the same industry composition as the market portfolio. As the value increases, the mutual fund becomes concentrated in some industries. 28 The Industry Concentration Index for each fund is calculated using quarterly fund holdings data from Thomson Reuters. At 28 The Industry Concentration Index is closely related to Herfindahl Index. Please see Kacperczyk, Sialm, and Zheng (2005) for further explanation 15

16 each quarter, all funds are categorized into ten industry concentration deciles based on the fund s lagged Industry Concentration Index. 29 The monthly returns of a mutual fund in a quarter is value-weighted using lagged weights for each ten industry concentration decile. The industry-concentration deciles are rebalanced quarterly. The regression is conducted at industry concentration level using monthly value-weighted fund returns. Fund gross returns of each decile is regressed on the Carhart (1997) four factors, which are shown in equation (1) without the industry momentum factor. The results in Table VI shows that industry-concentrated mutual funds significantly outperform the diversified mutual funds by an average of 0.36% per month from 1984 to In the original sample (1984 to 1999, reported in appendix Table AIII), the outperformance is 0.52% per month, which is higher compared to that in KSZ (2005). This is because the replicated sample has slightly higher average industry concentration index and larger standard deviation of fund returns. In general, the four-factor model regression result is close to the result in Kacperczyk, Sialm, and Zheng (2005) Unconditional Result To investigate whether the abnormal returns are driven by investing in industry momentum, I regress fund gross returns on the industry momentum factor, 30 along with the Carhart (1997) four factors. As shown in Table VII, I find mutual funds, especially industry-concentrated mutual funds have significant loadings on the industry momentum factor, which indicates industryconcentrated mutual funds are active industry momentum traders. Industry momentum explains 20 basis points of the monthly excess returns of most industry-concentrated mutual funds. The difference in the abnormal returns between the concentrated and diversified deciles becomes insignificant and is significantly explained by the industry momentum factor. The regression result indicates that industry momentum profits significantly contribute to the abnormal returns of industry-concentrated mutual funds and industry-concentrated mutual funds no longer outperform diversified mutual funds when industry momentum profits are considered. 29 I lag one quarter to avoid contemporaneous bias. 30 The correlations between the industry momentum factor and Market beta, SMB, HML are lower than The correlation between the industry momentum factor and the individual stock momentum factor is This is because industry momentum profits explain individual stock momentum profits and are more powerful than individual stock momentum to explain stock returns (Moskowitz and Grinblatt (1999)). 16

17 Table IX uses the industry momentum IM*(9,6) strategy profits as the industry momentum factor. The month skipped industry momentum factor reduces the lead-lag effects of the momentum strategy. Table VIII shows that the difference in abnormal returns between concentrated and that diversified mutual funds can be explained by the IM*(9,6) factor and the industry-concentrated funds no longer significantly outperform diversified funds when the industry momentum factor is included in the regression. The robustness check indicates that the findings also hold when lead-lag effects are excluded in the industry momentum factor Conditional Result Although both Table VII and Table IX show a significant result, the construction of the unconditional industry momentum factor contains forward-looking bias. To eliminate the forwardlooking bias, Table VIII and Table X report the regression outputs using conditional industry momentum factor for no gap strategies and month skipped strategies accordingly. The two conditional tables indicates that look-forward bias has little impact on the loadings of the industry momentum factor and the regression-based results still hold Industry Stock Selectivity Kacperczyk, Sialm, and Zheng (2005) provide evidence that there is a positive relation between level of industry concentration and ability to select higher return stocks within the industries. The industry stock selectivity is measured by IS: IS t = j [w j,t 1 [R j,t IR t (j, t 1)]], (5) where w j,t 1 is the weight of the stock j in the portfolio at quarter t 1, R j,t is the return of stock j at quarter t, and IR t (j, t 1) is the quarter t industry return that the stock j belongs to at the end of previous quarter. The positive relation between IS and industry concentration index serves as the evidence that industry-concentrated mutual funds have skill or information advantage to select stocks with in the industries. Table XI Panel A reproduced their OLS result and find a stronger positive and linear relation compared to KSZ(2005). Panel B of Table XI reports average IS by Industry 17

18 Concentration Index (ICI) deciles. It shows the relation is not linear and the difference of IS is not significant between Decile 10 and Decile 1. OLS regression cannot perfectly fit the relation of ICI and IS. Indeed, the standard deviation of IS for the most industry-concentrated mutual funds (Decile 10) is almost three times as large as the standard deviation of IS of the diversified mutual funds (Decile 1), indicating a strong heterogeneity of stock selectivity within industries for the concentrated mutual funds. Therefore, abilities to select the superior stocks within the industry is not critical for industry-concentrated mutual funds. Then, how could we explain the abnormal returns contributed by industry-concentrated funds with inferior stocks selections? Industry momentum does not require the investor to have ability to select winning stocks within the winning industry. Indeed, buying losing stocks in the winning industry and shortselling winning stocks in the losing industry also generates positive and significant strategy profits, documented by Moskowitz and Grinblatt (1999). Thus, industry momentum trading could explain the contribution to the abnormal returns by industry-concentrated mutual funds that don t have abilities to select superior stocks within the industries. 4.3 Holdings Result Euclidean Distance I calculate the Euclidean distance for each fund and the measures are value-weighted for each industry concentration decile. Since industry weights sum up to one and the Euclidean distance constitutes a closed convex cone in a ten-dimensional space, it cannot be compared directly across industry-concentration deciles. To solve this problem, I compare the Euclidean distance against a random benchmark for each decile. The random benchmark is generated by a Placebo test that randomly shifts industry weights for each fund. This approach produces benchmark values for the Euclidean distance that assume random selection of industry concentration. The Placebo test is repeated 1,000 times and the mean of the randomly shuffled Euclidean distance is used as the benchmark. 31 Table XII reports the difference between calculated Euclidean distance and the random benchmark. As shown in Table XII, I find evidence that industry-concentrated mutual funds have sig- 31 For concerns that 10! 1, 000, I redo the Placebo test five times and find little variation in the findings. 18

19 nificantly smaller Euclidean distance compared to their random benchmark. The industries in which those funds concentrate coincide with the winning industries in an industry momentum strategy. It indicates that industry-concentrated mutual funds are very likely to be industry momentum investors, which is consistent with the regression findings Net Purchases/Sales While the Euclidean distance captures the level of the holdings every quarter, net purchases/sales are focused on the changes of fund holdings from quarter to quarter. I first look at net purchases/sales of stocks in the winning industry for each fund. If a mutual fund invests in industry momentum, it has to actively purchase the stocks in the winning industry based on sorting by previous industry portfolio returns. The net purchases/sales are summed up for each industry concentration decile and compared between the top and bottom deciles. Stock splits and share repurchases are adjusted using CRSP adjustment factor. Panel A of Table XIII reports the net purchases by industry-concentration deciles and the difference between the most concentrated decile and the most diversified decile. Industry concentrated mutual funds actively purchase 0.37% (0.35%) of the market capitalization of stocks in the winning industry every quarter based on the industry momentum IM(9,6) (IM*(9,6)) strategy. Their net purchases are 0.41% (0.40%) more than those of diversified mutual funds. Given the average industry market capitalization from 1984 to 2013 is about $95.6 billion, Industryconcentrated funds (Decile 10) actively purchase an average of $ million of stocks in the winning industry every quarter, while diversified funds (Decile 1) sell an average of million of stocks in the winning industry every quarter. In terms of number of shares, Decile 10 funds purchase an average of million shares of stocks in the winning industry every quarter and Decile 1 funds sell an average of million shares of stock every quarter. To understand the rotation of holdings from industry to industry according to industry momentum, I also calculate the net purchases/sales of stocks in the winning industry H months ago, where H depends on the length of the holding period of the strategy. For example, IM(9,6) has a holding period of 6 months, therefore the net purchases/sales of stocks in the winning industry 6 months ago are calculated. Ideally, investing in industry momentum requires the 19

20 manager to sell stocks that belong to the winning industry H months ago and buy the stocks of the current winning industry. As shown in Panel B of Table XIII, industry concentrated mutual funds sell 0.19% (0.14%) more stocks that belong to the winning industry (month skipped*) 6 months ago than diversified mutual funds. The net sales results are negative but not significant and may caused by second-best choices by some funds to consider transaction costs. The previous winning industry could be the industry with the second or third highest return industry in the following quarters and turning over from one industry to another would generate high transaction costs. Thus, some cost-sensitive funds may opt not to turn over to the current winning industry. Industry-concentrated mutual funds actively purchase stocks that belong to the current winning industry and some funds sell no-longer winning industry stocks, revealing that they rotate industries according to an industry momentum strategy. 4.4 Original Sample Returns-based and Holdings-based Results I report the regression-based results and holdings-based results using the original sample ( ) in Appendix A.2. There is little variation in my findings. The main results are: Table AI: Higher values in industry average firms, industry portfolio excess returns, and abnormal returns are observed. Table AII: Industry momentum profits are lower the extended sample because the extended sample includes the financial crisis. Momentum strategy profits are sensitive to market volatility (Daniel and Moskowitz (2013) [12]). The most profitable industry momentum strategies are IM(9,6) and IM*(9,6) for the extended sample. The returns-based results and holdings-based results for the extended sample ( ) are reported as the main results. Returns-based results: Industry concentrated mutual funds significantly outperform diversified mutual funds by 0.52 % per month (Table AIII), which is higher than the outperformance in the extended sample. When an industry momentum factor is added to the Carhart (1997) four-factor model in Table AIV, industry-concentrated mutual funds no longer significantly outperform diversified mutual funds. Their outperformance is signifi- 20

21 cantly explained by the buy side profits of an industry momentum strategy. The loadings on month skipped industry momentum factor is significant (Table AVI). Although the abnormal return of Decile 10 is still significant, the significance and the level of abnormal return of Decile 10 are significantly reduced. It implies that industry momentum factor is an important pricing factor to control for. The conditional month skipped result (Table AVII) has the strongest result, suggesting that lead-lag effects and forward-looking bias are necessary to adjust for. Holdings results: The Euclidean distance of industry-concentrated mutual funds significantly deviates from the random industry concentration (Table AVIII). Industry-concentrated mutual funds have industry selectivity and concentrate in the winning industries according to the industry momentum strategy. Net purchases/sales of the current and past winning industry stocks (Table AIX) show that concentrated mutual funds significantly purchase more winning industry stocks than diversified mutual funds do. The net sales result is negative but insignificant, indicating some funds are turning over while the others maybe cost-sensitive. 5 Conclusion This paper provides empirical evidence that industry momentum strategy trading as an alternative to explain why industry-concentrated mutual funds outperform diversified mutual funds. By extending Kacperzyk, Sialm, and Zheng (2005) using a recent sample from 1984 to 2013, I show that industry-concentrated mutual funds have significant loadings on the industry momentum factor and they no longer have significant abnormal returns when the industry momentum factor is added to the Carhart (1997) four-factor model. The outperformance of industry-concentrated mutual funds is significantly explained by the industry momentum factor. Holdings-based results indicate that industry-concentrated mutual funds have a significantly smaller Euclidean distance compared to the random benchmark. They deviate from a random selection of industry concentration and purse an industry momentum strategy. Net purchases/sales show industryconcentrated mutual funds rotate industry holdings by purchasing current winning industry 21

22 stocks and some funds actively turnover past winning industry stocks. In sum, industry momentum is an important motivation for mutual fund managers to deviate from a well-diversified portfolio and invest in a concentrated portfolio of stocks from a small number of industries. It is necessary to include industry momentum profits as a pricing kernel when evaluate the performance of mutual funds on the industry-concentration dimension. 22

23 References [1] Andres Almazan, Keith C Brown, Murray Carlson, and David A Chapman. Why constrain your mutual fund manager? Journal of Financial Economics, 73(2): , [2] Leslie Boni and Kent L Womack. Analysts, industries, and price momentum. Journal of Financial and Quantitative Analysis, 41(01):85 109, [3] Jeffrey A Busse and Qing Tong. Mutual fund industry selection and persistence. Review of Asset Pricing Studies, 2(2): , [4] Mark M Carhart. On persistence in mutual fund performance. The Journal of finance, 52(1):57 82, [5] Cristina Cella, Andrew Ellul, and Mariassunta Giannetti. Investors horizons and the amplification of market shocks. Review of Financial Studies, page hht023, [6] Joseph Chen, Harrison Hong, Ming Huang, and Jeffrey D Kubik. Does fund size erode mutual fund performance? the role of liquidity and organization. American Economic Review, pages , [7] Tarun Chordia and Lakshmanan Shivakumar. Momentum, business cycle, and time-varying expected returns. The Journal of Finance, 57(2): , [8] Lauren Cohen and Andrea Frazzini. Economic links and predictable returns. The Journal of Finance, 63(4): , [9] Joshua D Coval and Tobias J Moskowitz. Home bias at home: Local equity preference in domestic portfolios. The Journal of Finance, 54(6): , [10] Joshua D Coval and Tobias J Moskowitz. The geography of investment: Informed trading and asset prices. Journal of Political Economy, 109(4): , [11] Kent Daniel, Mark Grinblatt, Sheridan Titman, and Russ Wermers. Measuring mutual fund performance with characteristic-based benchmarks. The Journal of finance, 52(3): ,

24 [12] Kent Daniel and Tobias J Moskowitz. Momentum crashes. Working Paper Series, [13] Kent Daniel and Sheridan Titman. Evidence on the characteristics of cross sectional variation in stock returns. The Journal of Finance, 52(1):1 33, [14] Mary Margaret Frank, James M Poterba, Douglas A Shackelford, and John B Shoven. Copycat funds: Information disclosure regulation and the returns to active management in the mutual fund industry. Journal of Law and Economics, 47(2): , [15] W Glunt, Tom L Hayden, S Hong, and J Wells. An alternating projection algorithm for computing the nearest euclidean distance matrix. SIAM Journal on Matrix Analysis and Applications, 11(4): , [16] Mark Grinblatt and Sheridan Titman. Mutual fund performance: An analysis of quarterly portfolio holdings. Journal of business, pages , [17] Mark Grinblatt and Sheridan Titman. Performance measurement without benchmarks: An examination of mutual fund returns. Journal of Business, pages 47 68, [18] Mark Grinblatt, Sheridan Titman, and Russ Wermers. Momentum investment strategies, portfolio performance, and herding: A study of mutual fund behavior. The American economic review, pages , [19] Narasimhan Jegadeesh and Sheridan Titman. Returns to buying winners and selling losers: Implications for stock market efficiency. The Journal of Finance, 48(1):65 91, [20] Narasimhan Jegadeesh and Sheridan Titman. Profitability of momentum strategies: An evaluation of alternative explanations. The Journal of Finance, 56(2): , [21] George J Jiang, Tong Yao, and Tong Yu. Do mutual funds time the market? evidence from portfolio holdings. Journal of Financial Economics, 86(3): , [22] Marcin Kacperczyk, Clemens Sialm, and Lu Zheng. On the industry concentration of actively managed equity mutual funds. The Journal of Finance, 60(4): , [23] Moshe Levy and Richard Roll. The market portfolio may be mean/variance efficient after all. Review of Financial Studies, page hhp119,

25 [24] Jonathan Lewellen. Momentum and autocorrelation in stock returns. Review of Financial Studies, 15(2): , [25] Andrew W Lo and Archie Craig MacKinlay. When are contrarian profits due to stock market overreaction? Review of Financial studies, 3(2): , [26] Tobias J Moskowitz and Mark Grinblatt. Do industries explain momentum? The Journal of Finance, 54(4): , [27] Veronika K Pool, Noah Stoffman, and Scott E Yonker. No place like home: Familiarity in mutual fund manager portfolio choice. Review of Financial Studies, 25(8): , [28] Dimitri Vayanos and Paul Woolley. An institutional theory of momentum and reversal. Review of Financial Studies, page hht014, [29] Russ Wermers. Mutual fund performance: An empirical decomposition into stock-picking talent, style, transactions costs, and expenses. The Journal of Finance, 55(4): ,

26 Figure 1. Cumulative Industry Momentum Profits. This figure presents cumulative industry momentum portfolio returns for the 6-month, 9-month, and 12-month strategies from 1984 to The portfolio is formed based on sorting by the previous 6-month, 9-month, or 12-month industry returns, buying the winning industry and shorting the losing industry. The cumulative returns after the portfolio formation are graphed for the following 30 months. 26

27 Table I Summary Statistics of Industries Summary statistics of the 10 industry portfolios are reported. The industries are formed monthly using lagged weights, from January 1984 to December 2013 using CRSP SIC codes. I report the average number of stocks in each industry portfolio every month and the minimum number of stocks in each portfolio at any point of time (in parentheses). The average market capitalization and the average return in excess of the corresponding three-month T-bill rate are reported The abnormal returns adjust the industry portfolios for matched size and book-to-market characteristics, with t-statistic reported in parentheses. Industry Avg. No. of Firms (min) Avg. % of Market Cap Excess Returns Abnormal Returns (t-stat) 1 Consumer non-durables (133) 8.38% (1.18) 2 Consumer durables (52) 9.13% (1.37) 3 Healthcare (20) 7.61% (1.58) 4 Manufacturing (372) 17.33% (1.42) 5 Energy (53) 11.16% (1.01) 6 Utilities (98) 6.72% (1.07) 7 Telecom 93.99(8) 6.42% (1.54) 8 Business equipment and services (28) 13.17% (1.55) 9 Wholesale and retail (84) 7.46% (1.04) 10 Finance (31) 12.63% (1.19) Average (87.90) 10.00% (1.49) 27

28 Table II Summary Statistics of Mutual Funds Summary statistics of mutual funds are reported. Following Kacperzyk, Sialm, and Zheng (2005), I only include actively managed US equity funds. Balanced, bond, index international and sector funds are eliminated. To better filter the sample and exclude index and sector funds, a name key word search based on fund name in conducted. In addition, funds with more than 300 stock holdings are dropped. The observation is included if the file date (fdate) and the report date (rdate) of the fund in Thomson Reuters Database are in the same quarter. Share class level data are value-weighted to fund level using lagged weights. Number of funds, average stock holdings, fund size (TNA), fund age (since inception), fund expenses, turnover, load, quarterly raw returns, and industry concentration index are reported in this table. Panel A reports the fund summary statistics of the replicated original sample. Panel B is from Kacperzyk, Sialm, and Zheng (2005) for comparison. Panel C reports the fund number and average Industry Concentration Index of the extended sample. Panel A Replicated Sample ( ) Mean Median Min Max Num. of Funds 2,008 Num. of Stocks Held by Fund TNA ($m) , Age (years) Expenses (%) Turnover (%) Total Load (%) Quarterly Raw Return (%) Industry Concentration Index (%) Panel B Original Sample by KSZ(2005) Mean Median Min Max Num. of Funds 1,771 Num. of Stocks Held by Fund ,439 TNA ($m) ,594 Age (years) Expenses (%) Turnover (%) ,263 Total Load (%) Quarterly Raw Return (%) Industry Concentration Index (%) Panel C Extended Sample ( ) Mean Median Min Max Num. of Funds 3,386 Industry Concentration Index (%)

29 Table III A Sample of Industry Momentum Mutual Funds Fund characteristics for a sample of five mutual funds that clearly state they invest in industry momentum are reported in Panel A. These funds come from the sample and contain the key word industry momentum industry rotation or sector rotation in their fund name, fund prospectus, or a third party media report. Fund characteristics including assets under management, most recent annual return, and portfolio turnover are collected from the most recent fund prospectus (2014 Q2 report) and reported. Also listed are self-reported fund alphas, sharp ratios, and fund inception years. Panel B extends to a wider fund name-based key word search for funds whose name contains transector, sector rotation, sector rotational, sector analysis, select selectors, sector allocation, industry select/selection, industrial select/selection, IND SEL, and select fund. A heat map for Industry Concentration Index (ICI) rank of these funds is reported in Panel B. The time-series average of ICI rank, minimal ICI rank and maximum ICI rank are reported. Mutual Fund Name Asset under Management (in $m) Panel A Example Funds Annual Annual Return Portfolio Alpha (1Yr) Turnover Sharpe Ratio Inception Year Rydex Sector Rotation % 186% Leuthold Select Industries Fund % 153% Virtus AlphaSector Rotation Fund % 123% Powershares DWA Momentum Portfolio 1, % 134% Fidelity Sector Rotation Fund % 146% Benchmark: S&P % 0.27 Panel B Identified Industry Momentum Fund Heat Map Mean Medean Std Min Max Num. of Identified Funds ( ) 53 Average ICI Rank Min ICI Rank Max ICI Rank Num. of Identified Funds ( ) 169 Average ICI Rank Min ICI Rank Max ICI Rank

30 Table IV Industry Momentum Trading Profits Average monthly profits of the industry momentum trading strategies from January 1984 to December 1999 (T=192) are reported below. The industry momentum portfolios are formed based on L-month sorted cumulative industry portfolio returns and held for the following H months. Results are reported for the IM (L,H) industry momentum treading strategy, where the winner portfolio is the return of the highest momentum industry, the middle portfolio is the equal-weighted return of the middle momentum industries, and the loser portfolio is the return of the lowest momentum industry. The returns for the winners (Wi), losers (Lo), and winners minus losers (Wi-Lo) are reported, as well as the winners minus middle (Wi-Mid) and middle minus losers (Mid-Lo), where the middle portfolio is the equal-weighted average return of the two industries ranked 5 and 6 based on past L-month sorts. t-statistics for the zero-cost strategies are in parentheses. Panel A reports the profits when no gap exists between the portfolio formation period and the holding period (i.e., sort on t-l to t-1 returns). In Panel B, one month is skipped between portfolio formation and holding periods (i.e., sort on t-l-1 to t-2 returns). Panel A: No Gap Panel B: Month Skipped L H= Wi Lo Wi-Lo (3.55) (2.19) (1.21) (1.31) (2.63) (0.94) (0.53) (-0.01) (0.55) (2.29) buy Wi-Mid (1.42) (-0.25) (-0.03) (0.27) (1.08) (-0.44) (-0.18) (-0.88) (0.15) (0.78) sell Mid-Lo (3.39) (3.07) (1.64) (1.23) (1.70) (1.75) (0.93) (0.88) (0.46) (1.67) (Continued on next page) 30

31 Industry Momentum Trading Profits continued from previous page Panel A: No Gap Panel B: Month Skipped L H= Wi Lo Wi-Lo (1.63) (-0.12) (-0.59) (0.76) (1.62) (-0.26) (-1.66) (-0.53) (0.84) (1.18) buy Wi-Mid (1.93) (0.43) (-0.25) (1.03) (1.09) (0.66) (-1.26) (-0.26) (0.85) (0.78) sell Mid-Lo (-0.06) (-0.62) (-0.56) (0.03) (1.04) (-0.98) (-0.81) (-0.47) (0.32) (0.79) 6 Wi Lo Wi-Lo (1.44) (0.34) (0.88) (1.83) (2.09) (0.27) (-0.16) (1.15) (1.79) (1.88) buy Wi-Mid (0.17) (-0.11) (1.15) (1.54) (1.96) (-0.04) (0.05) (1.64) (1.61) (2.08) sell Mid-Lo (1.80) (0.58) (0.19) (1.12) (1.11) (0.40) (-0.25) (0.17) (1.01) (0.73) 9 Wi Lo Wi-Lo (Continued on next page) 31

32 Industry Momentum Trading Profits continued from previous page Panel A: No Gap Panel B: Month Skipped L H= (2.85) (1.88) (2.41) (2.69) (2.28) (1.27) (1.41) (2.23) (2.39) (2.01) buy Wi-Mid (2.21) (1.91) (2.28) (2.72) (2.37) (1.25) (1.58) (2.18) (2.5) (2.28) sell Mid-Lo (1.96) (0.88) (1.47) (1.47) (1.18) (0.47) (0.58) (1.25) (1.19) (0.82) 12 Wi Lo Wi-Lo (1.76) (1.89) (2.21) (1.92) (1.57) (1.70) (2.04) (2.35) (1.81) (1.26) buy Wi-Mid (1.71) (1.59) (2.23) (2.27) (1.96) (1.59) (1.57) (2.61) (2.12) (1.72) sell Mid-Lo (0.82) (1.27) (1.16) (0.68) (0.45) (0.81) (1.43) (1.05) (0.62) (0.24) t-statistics in parentheses 32

33 Table V The Unconditional and Conditional Most Profitable Industry Momentum Strategy The IM(9,6) industry momentum portfolios are formed based on 9-month sorted cumulative industry portfolio returns and held for the following 6 months. The IM*(9,6) industry momentum portfolios are formed based on 9-month sorted cumulative industry portfolio returns, skip one month, and held for the following 6 months. The monthly average profits of winners (Wi), losers (Lo), winners minus losers (Wi-Lo), the buy side profit (Wi-Mid), and the sell side profit (Mid-Lo) are reported. See Table IV for definition. The unconditional most profitable industry momentum strategy is reported in Panel A. It sorts all the strategy combinations in table IV and select the one with highest profit. The conditional result sorts and select the most profitable strategy for a given time using the strategy returns in the past five years. The sorting weights equally the total profit (Sorted Wi-Lo) and the buy side profit (Sorted Wi-Mid) and their t-statistics and select the first strategy in the rank. The following holding period return is recorded as the conditional industry momentum factor (actual Wi-Mid). In the conditional result, strategy is allowed to change horizon combinations. Any strategy with a holding horizon less than one quarter is eliminated to consider transaction cost. t-statistics for the zero-cost strategies are in parentheses. Panel A Unconditional Most Profitable Industry Momentum Strategy Original Sample ( ) Extended Sample ( ) No Gap Month Skipped No Gap Month Skipped IM(9,9) IM*(12,6) IM(9,6) IM*(9,6) Wi Lo Wi-Lo (2.69) (2.35) (2.50) (2.14) buy Wi-Mid (2.72) (2.61) (2.11) (1.75) sell Mid-Lo (1.47) (0.99) (1.65) (1.44) Panel B Conditional Most Profitable Industry Momentum Strategy Original Sample ( ) Extended Sample ( ) No Gap Month Skipped No Gap Month Skipped IM(Cond) IM*(Cond) IM(Cond) IM*(Cond) buy Wi-Mid (Actual) (2.28) (2.82) (1.70) (1.73) buy Wi-Mid (Sorted) (40.95) (24.56) (24.79) (22.05) Wi-Lo (Sorted) (72.74) (58.40) (35.36) (34.51) t-statistics in parentheses 33

34 Table VI Performance Heterogeneity by Industry Concentration Deciles This table presents abnormal returns and factor loadings on the Carhart (1997) four factors for mutual funds from 1984 to The first column reports the abnormal returns by industry concentration deciles. All mutual funds in the sample are divided into industry concentration deciles based on quarterly industry concentration index ICI t = 10 j=1 (w j,t w j,t ) 2, where w j,t is the fund s weight in industry j and w j,t is the market weight of industry j. The fund returns are value-weighted as decile returns every month and the industry concentration deciles are rebalanced quarterly. t-statistics are in parentheses. The table includes the difference in abnormal returns between the top and the bottom deciles. Abnormal Return Factor Loadings (% per month) Market Size Value Stock Mom All Funds 0.13* 0.91*** 0.11*** * (1.74) (117.23) (11.28) (0.77) (1.69) Decile *** -0.06*** 0.09*** 0.01 (Diversified) (0.43) (149.66) (-7.27) (10.45) (1.07) Decile *** ** -0.02*** (0.36) (112.50) (-1.18) (2.43) (-3.13) Decile *** 0.02** 0.08*** 0.01 (1.01) (115.22) (1.98) (6.18) (1.64) Decile *** 0.08*** 0.06*** 0.01 (0.82) (95.72) (5.86) (4.61) (1.00) Decile *** 0.11*** 0.04*** 0.02*** (0.44) (98.17) (8.53) (2.88) (2.64) Decile *** 0.07*** (1.49) (102.47) (5.75) (1.13) (-0.40) Decile *** 0.13*** 0.05*** (1.16) (81.54) (8.54) (3.44) (-0.38) Decile *** 0.16*** * (1.54) (70.82) (8.59) (1.36) (-1.87) Decile * 0.91*** 0.18*** (1.77) (62.90) (8.95) (-0.90) (-0.50) Decile ** 0.98*** 0.22*** -0.15*** (Concentrated) (2.58) (34.55) (5.40) (-3.56) (-0.84) 10 th decile - 1 st decile 0.36** 0.09*** 0.28*** -0.25*** (2.03) (3.22) (6.96) (-5.78) (-1.07) t-statistics in parentheses, p 0.01, p 0.05, p

35 Table VII Industry Momentum and Performance Heterogeneity (Unconditional) This table presents abnormal returns and factor loadings on the industry momentum factor and the Carhart (1997) four factors. The first column reports the abnormal returns by industry concentration deciles. The industry concentration deciles are the same as in Table VI. Industry momentum factor is the buy side profit of an industry momentum (9,6) strategy. The IM (9,6) strategy buy-side factor is constructed as sorting the ten industry portfolios based on previous 9-month cumulative returns, buying the winning industry and shorting the middle industries, and holding the portfolio in the following 6 months. The average returns in the holding period represent the strategy profits. The table includes the difference in abnormal returns between the top and the bottom deciles. t-statistics are in parentheses. Abnormal Return IM(9,6) Factor Loadings (% per month) Buy-side Market Size Value Stock Mom All Funds * 0.89*** 0.11*** (1.56) (1.84) (113.60) (12.18) (0.79) (1.04) Decile *** -0.06*** 0.09*** 0.00 (Diversified) (0.43) (0.69) (147.58) (-7.37) (10.52) (0.32) Decile *** *** -0.03*** (0.37) (1.34) (111.14) (-1.45) (2.60) (-3.70) Decile *** 0.02* 0.07*** 0.01 (1.02) (1.43) (113.64) (1.76) (6.21) (0.69) Decile *** 0.07*** 0.06*** 0.00 (0.83) (1.51) (94.39) (5.60) (4.74) (0.10) Decile *** 0.11*** 0.04*** 0.01 (0.45) (0.98) (97.48) (8.12) (3.16) (0.80) Decile *** 0.08*** (1.28) (0.38) (100.92) (5.63) (1.16) (-0.54) Decile *** 0.13*** 0.06*** (1.16) (1.05) (80.31) (8.31) (3.52) (-0.86) Decile *** 0.16*** (1.52) (-0.31) (69.85) (8.54) (1.33) (-1.45) Decile * 0.05* 0.91*** 0.18*** (1.65) (1.73) (61.91) (8.72) (-0.81) (-0.96) Decile *** 0.96*** 0.19*** -0.13*** -0.10*** (Concentrated) (1.51) (5.01) (34.44) (4.82) (-3.23) (-3.28) 10 th decile - 1 st decile *** 0.07** 0.26*** -0.23*** -0.10*** (1.40) (4.76) (2.51) (6.43) (-5.51) (-3.35) t-statistics in parentheses, p 0.01, p 0.05, p

36 Table VIII Robustness: Industry Momentum and Performance Heterogeneity (Conditional) This table presents abnormal returns and factor loadings on the conditional industry momentum factor and the Carhart (1997) four factors. Except for the industry momentum factor, the other variables are the same as in Table VI. The conditional industry momentum factor represents the profits of the most profitable industry momentum strategy in the previous five years. It ranks all industry momentum strategies and select the highest ranked strategy considering the total profit and the buy side profit of the strategy. The IM(Cond) buy-side factor is constructed as the average monthly actual return of the most profitable conditional strategy in the following holding period according to the strategy holding horizon. The conditional industry momentum strategy tracks the exact investment choices available to fund managers at a give time without forward-looking bias. Abnormal Return IM(Cond) Factor Loadings (% per month) Buy-side Market Size Value Stock Mom All Funds ** 0.90*** 0.10*** (1.60) (2.01) (115.31) (11.33) (0.85) (0.92) Decile *** -0.06*** 0.09*** 0.00 (Diversified) (0.53) (1.29) (146.36) (-7.23) (10.48) (0.30) Decile *** ** -0.04*** (0.60) (0.99) (110.15) (-1.34) (2.54) (-4.84) Decile *** 0.02* 0.07*** 0.01 (0.96) (1.53) (113.84) (1.74) (6.47) (1.15) Decile *** 0.08*** 0.07*** (0.64) (1.01) (95.18) (5.97) (4.57) (-0.88) Decile *** 0.10*** 0.04*** 0.01 (0.87) (1.39) (97.44) (7.70) (3.23) (0.67) Decile *** 0.08*** (1.37) (1.21) (102.38) (5.66) (1.14) (-1.14) Decile * 0.91*** 0.13*** 0.06*** (1.11) (1.79) (81.15) (8.22) (3.70) (-1.31) Decile * 0.92*** 0.15*** ** (1.35) (1.71) (71.47) (8.13) (1.63) (-2.37) Decile * 0.04** 0.90*** 0.17*** (1.68) (2.14) (62.51) (8.39) (-0.81) (-1.26) Decile *** 0.97*** 0.19*** -0.15*** -0.10*** (Concentrated) (1.52) (5.57) (35.31) (4.75) (-3.59) (-3.56) 10 th decile - 1 st decile *** 0.07** 0.25*** -0.24*** -0.11*** (1.50) (5.28) (2.57) (6.37) (-5.93) (-3.63) t-statistics in parentheses, p 0.01, p 0.05, p

37 Table IX Robustness: Month Skipped Industry Momentum Strategy (Unconditional) This table presents abnormal returns and factor loadings on the industry momentum factor and the Carhart (1997) four factors. Except for the industry momentum factor, the other variables are the same as in Table VI. The industry momentum factor represents the profits of trading an industry momentum month skipped (9,6) strategy. The IM*(9,6) buy-side factor is constructed as sorting the ten industry portfolios based on their previous 9 months returns, buying the winning industry and shorting the middle industries, skipping one month, and holding the portfolio in the following 6 months. Skipping one month between portfolio formation and the start of holding reduces the strategy profits contributed by lead-lag effects. Abnormal Return IM*(9,6) Factor Loadings (% per month) Buy-side Market Size Value Stock Mom All Funds ** 0.88*** 0.10*** (1.51) (1.98) (110.61) (11.09) (0.88) (0.83) Decile *** -0.07*** 0.10*** 0.00 (Diversified) (0.47) (0.82) (147.78) (-7.49) (10.64) (0.16) Decile *** *** -0.03*** (0.41) (1.25) (111.22) (-1.53) (2.72) (-3.83) Decile *** 0.02* 0.08*** 0.01 (1.04) (1.46) (113.58) (1.74) (6.33) (0.83) Decile *** 0.08*** 0.07*** 0.01 (0.84) (0.85) (94.23) (5.66) (4.69) (0.53) Decile *** 0.11*** 0.05*** 0.01 (0.51) (1.06) (97.46) (8.06) (3.29) (1.05) Decile *** 0.07*** (1.48) (0.54) (100.87) (5.59) (1.19) (-0.60) Decile *** 0.13*** 0.06*** (1.18) (0.73) (80.25) (8.32) (3.51) (-0.67) Decile *** 0.16*** (1.48) (-0.70) (69.92) (8.59) (1.26) (-1.37) Decile * *** 0.18*** (1.69) (0.81) (61.87) (8.72) (-0.79) (-0.81) Decile *** 0.96*** 0.19*** -0.12*** -0.08*** (Concentrated) (1.50) (5.21) (34.46) (4.73) (-2.97) (-3.07) 10 th decile - 1 st decile *** 0.07** 0.25*** -0.22*** -0.08*** (1.41) (4.81) (2.48) (6.35) (-5.27) (-3.11) t-statistics in parentheses, p 0.01, p 0.05, p

38 Table X Robustness: Month Skipped Industry Momentum Strategy (Conditional) This table presents abnormal returns and factor loadings on the conditional industry momentum month skipped factor and the Carhart (1997) four factors. Except for the industry momentum factor, the other variables are the same as in Table VI. The conditional industry momentum month skipped factor represents the profits of the most profitable industry momentum month skipped strategy in the previous five years. It ranks all industry momentum strategies and select the highest ranked strategy considering the total profit and the buy side profit of the strategy. The IM*(Cond) buy-side factor is constructed as the average monthly actual return of finding the most profitable conditional strategy, skipping one month, and holding the portfolio in the following holding period according to the strategy holding horizon. The conditional industry momentum strategy tracks the exact investment choices available to fund managers at a give time without forward-looking bias. Skipping one month between portfolio formation and the start of holding reduces the strategy profits contributed by lead-lag effects. Abnormal Return IM*(Cond) Factor Loadings (% per month) Buy-side Market Size Value Stock Mom All Funds ** 0.90*** 0.10*** (1.59) (1.93) (118.47) (12.33) (0.91) (0.81) Decile *** -0.07*** 0.10*** 0.00 (Diversified) (0.47) (1.38) (146.86) (-7.35) (10.62) (0.19) Decile *** *** -0.04*** (0.52) (1.62) (109.49) (-1.21) (2.65) (-4.15) Decile *** 0.02* 0.08*** 0.01 (0.91) (1.24) (113.93) (1.69) (6.53) (1.42) Decile *** 0.08*** 0.07*** (0.54) (1.41) (94.98) (5.84) (4.78) (-0.43) Decile *** 0.10*** 0.05*** 0.01 (0.74) (1.44) (97.32) (7.47) (3.60) (1.11) Decile *** 0.07*** (1.45) (0.94) (102.48) (5.59) (1.23) (-0.97) Decile *** 0.13*** 0.06*** (1.16) (0.72) (81.08) (8.23) (3.69) (-0.75) Decile *** 0.16*** (1.39) (1.08) (71.67) (8.49) (1.39) (-1.08) Decile * *** 0.17*** (1.71) (1.47) (62.47) (8.30) (-0.63) (-0.85) Decile * 0.16*** 0.97*** 0.18*** -0.12*** -0.07*** (Concentrated) (1.65) (4.44) (35.00) (4.55) (-2.98) (-2.77) 10 th decile - 1 st decile *** 0.08*** 0.25*** -0.23*** -0.08*** (1.52) (4.04) (2.71) (6.16) (-5.31) (-2.81) t-statistics in parentheses, p 0.01, p 0.05, p

39 Table XI Industry Stock Selectivity by ICI Deciles This table presents the industry stock selectivity measure (IS) by industry concentration index deciles from 1984 to IS is calculated as IS t = j [w j,t 1[R j,t IR t (j, t 1)]] where w j,t 1 is the weight of the stock j in the portfolio at quarter t 1, R is the return of stock j, and IR is the industry return that the stock belongs to. It measures a manager s ability to select superior stock within industries. Panel A replicates the linear regression output from KSZ (2005). Panel B looks into industry concentration index deciles as a deeper look into the linear relation. t-statistics are in parentheses. Panel A OLS Regression Replication KSZ(2005) IS(fund level) IS(fund level) ICI 2.369*** ICI 1.89*** t-stat (5.90) t-stat (4.02) Intercept 0.555*** Intercept Not Reported (12.39) Not Reported Controls YES Controls YES Panel B Industry Selectivity by ICI Deciles IS Mean(%) IS Std Decile *** (2.92) Decile *** (2.68) Decile * (1.73) Decile ** (2.18) Decile *** (2.92) Decile *** (2.66) Decile ** (2.09) Decile ** (2.03) Decile *** (2.73) Decile * (1.84) 10 th decile - 1 st decile (0.84) t-statistics in parentheses, p 0.01, p 0.05, p

40 Table XII Euclidean Distance against Random Benchmark This table shows the difference between Euclidean distance and its random benchmark by industry concentration deciles from 1984 to Euclidean 10 distance of fund i at quarter t is calculated as di,t = j=1 (h IM,j,t hi,j,t) 2, where hi,j,t is the industry weights and him,j,t is the industry weights of random benchmark trading industry momentum using IM(9,6) strategy (Panel A) or IM*(9,6) strategy (Panel B). The random benchmark is generated from a Placebo test in which the industry weights for each fund is randomly reshuffled. The random process is repeated for 1,000 times and the mean of random Euclidean distance is taken as random benchmark. The difference between Euclidean distance and its benchmark by industry concentration deciles are reported. t-statistics are in parentheses. Panel A IM(9,6) di,t- di,t Euclidean Distance di,t Placebo Benchmark di,t Mean t-stat Mean Std Min Max Mean Std Min Max (Diversified) Decile (-1.29) Decile (-0.36) Decile (-0.53) Decile (-0.65) Decile (-0.69) Decile (0.02) Decile (-0.70) Decile (-1.63) Decile (-1.35) (Concentrated) Decile *** (-2.73) di,t- di,t Panel B IM*(9,6) Euclidean Distance di,t Placebo Benchmark di,t Mean t-stat Mean Std Min Max Mean Std Min Max (Diversified) Decile (-1.19) Decile (-0.31) Decile (-0.65) Decile (-0.42) Decile (-0.80) Decile (0.05) Decile (-0.55) Decile (-1.27) Decile (-1.43) (Concentrated) Decile *** (-2.86) t-statistics in parentheses, p 0.01, p 0.05, p

41 Table XIII Net Purchases/Sales of Winning Industry Stocks Net purchases/sales of stocks in the winning industries are reported from 1984 to Net purchases/sales are calculated as changes of holdings in the stocks belonging to the winning industry at quarter t: Net P urchasei,t = S m=1 [(N m,t Nm,t 1) Pm,t]/Market CapJ t,t 1 for fund i at quarter t, where Nm,t denotes shares held of stock m at quarter t and Pm,t is the stock price. Net purchases/sales of stocks in the current winning industry and the winning industry H months (H=6 for IM(9,6) and IM*(9,6)) ago are reported in Panel A and Panel B respectively. The differences of net purchases/sales between top and bottom industry concentration deciles are reported. t-statistics are in parentheses. Panel A Net Purchases/sales of Stocks in the Winning Industry at time t IM(9,6) IM*(9,6) Mean Std Min Max Mean Std Min Max (Diversified) Decile Decile Decile Decile Decile Decile Decile Decile Decile (Concentrated) Decile th decile - 1 st decile ** ** t-stat (2.34) (2.26) Panel B Net Purchases/sales of Stocks in the Winning Industry at time t-h IM(9,6) IM*(9,6) Mean Std Min Max Mean Std Min Max (Diversified) Decile Decile Decile Decile Decile Decile Decile Decile Decile (Concentrated) Decile th decile - 1 st decile t-stat (-1.44) (-0.87) 41

42 Appendix A A.1 Fund Prospectus Example Table III reports the fund characteristics of the five selected mutual funds that clearly state they trade industry momentum. Below is an example of a fund prospectus. The key word industry rotation can be found in the fund prospectus. 42