Fant, L. Franklin, and Edward S. O'Neal, 2000, Temporal changes in the determinants of mutual fund flows ... Nanda, Vikram, Z. Jay Wang, and Lu Zheng, 2004.
Mutual Fund Flow-Performance Relationship under Volatile Market Condition
Abstract
We analyze the relationship between flows and performance of Chinese mutual funds that trade in a volatile market environment. Consistent with existing literature, we find that the net flow to a fund is positively related to past fund performance. Contrary to previous studies using samples in the U.S. and other countries, our results do not exhibit an asymmetric flow-performance relationship, nor do we find any significant star effect in China. These results imply that market volatility plays an important role in reducing the asymmetric flow-performance relationship. Furthermore, we find that the positive relationship is more pronounced during bull markets than during bear markets. Consistent with Thaler and Johnson’s (1990) house money effect and the overconfidence hypothesis proposed by Gervais and Odean (2001), this suggests that Chinese mutual fund investors are more confident and invest more aggressively when stock markets perform well.
JEL classification: G14 Key words: Chinese mutual funds; Flow-performance relationship; Asymmetric relationship; Disposition effect; House money effect; Star effect; Cognitive dissonance; Attribution bias; Overconfidence.
1
1. Introduction Prior studies find that mutual fund investors chase past performance by rewarding “winners” but failing to punish “losers.” As a result, the past top-performing funds attract disproportionately large fund inflows in subsequent periods, whereas past poor performers suffer minimal outflows (Ippolito, 1992; Goetzmann and Peles, 1997; Chevalier and Ellison, 1997; Sirri and Tufano, 1998; Fant and O’Neal, 2000; Del Guercio and Tkac, 2002). The conventional explanations for this asymmetric flow-performance relationship include: (1) transaction fees and switching costs (Ippolito, 1992); (2) marketing efforts and media attention (Sirri and Tufano, 1998); (3) investor participation costs (Huang, Wei and Yan, 2007); (4) strategy replacement (Lynch and Musto, 2003); (5) the disposition effect (Shefrin and Statman, 1985); (6) cognitive dissonance (Goetzmann and Peles, 1997); and (7) the investor clientele effect (Del Guercio and Tkac, 2002; Sawicki, 2001; Christoffersen and Musto, 2002). In the current paper, we further explore the flow-performance relationship with a focus on stock market condition. The theoretical models of Berk and Green (2004), Kim (2011a), and Huang, Wei and Yan (2012) suggest that as the noise in the observed fund return increases, investors learn less from past fund performance. Xie (2011) and Kim (2011b) find time-varying flow-performance sensitivity. More specifically, Kim (2011b) shows that the flow-performance relationship of U.S. mutual funds was convex during 1983-1999 and became less or no convex since earlier 2000s, and he contributes the change to volatile market conditions. However, it is unclear whether the variation is due to changes in market volatility or due to the evolution of other factors since Barberis and Xiong (2009) indicate that many factors, such as the reference points that investors use to judge gains or losses, expected risk returns, trading frequency, and holding periods, etc., affect fund flows and can result in different flow-performance relationships. In addition, Ferreira et al. (2012) show that there are marked differences in the flow-performance relationship across countries. Therefore the results observed in Kim’s (2011b) study could be just a U.S. phenomenon and does not exist in other markets. To further explore this issue, we test the impact of market volatility on flow-performance directly using Chinese mutual fund data since 2
Chinese financial markets are extremely volatile and the magnitude of volatility is far more severe compared with the volatility in other markets. Thus, the results based on Chinese mutual funds enhance our understanding on the impact of market volatility on flow performance relationship. The second issue we address is star effect in Chinese mutual fund markets. Researchers find that funds rated as star funds by Morningstar attract disproportionate inflows due to quality certification and search cost reduction (Nanda, Wang and Zheng, 2004; Del Guercio and Tkac, 2008). We conjecture that the star funds rated by Morningstar could have two contrary effects on the Chinese mutual funds. If search cost reduction is the driving force, we expect the star effect to be weak since the Chinese mutual markets are much smaller than the U.S. markets in both number of funds and the assets under management. In contrast, if quality certification plays a key role, we expect a strong star effect since Chinese mutual funds have a shorter history which makes it harder for investors to assess fund quality, especially under the volatile Chinese stock market environment. Similarly, the asymmetric information between Chinese mutual fund investors and fund management companies could be high because of the less transparent market environment. In this case, the star effect associated with quality certification is expected to be strong for Chinese mutual funds. The third issue we investigate is the effect of stock market return on fund flow-performance relationship. Olivier and Tay (2009) show that economic activities measured by GDP affect flow-performance relationship. We further explore the effect of stock market condition on the flow-performance relationship based on a joint implication of Thaler and Johnson’s (1990) house money effect and the overconfidence hypothesis proposed by Gervais and Odean (2001).1 We posit that the
1
Thaler and Johnson (1990) show that a prior gain serves as a loss reduction cushion, which mitigates the influence of loss aversion and increases the willingness of investors to take greater risks. Using gambling jargon, Thaler and Johnson call this “the house money effect.” They show that gamblers who have a prior win (i.e., they now have “house money”) tend to take larger risks by betting more because they treat subsequent losses as reductions in their prior gain; losing the “house money” is less distressing than losing their own money. In contrast, prior losses can increase risk aversion and decrease the willingness to take a risk. Gervais and Odean (2001) indicate that people have an attribution bias toward learning and tend to overestimate the degree to which they are responsible for their own successes. Gervais and Odean’s (2001) overconfidence model predicts that traders become more overconfident
3
flow-performance relationship is more pronounced during bull markets than during bear markets since public investors are more likely to experience capital gains in bull markets than in bear markets. Therefore, during bull markets, the house money effect is further strengthened by overconfidence in the upward market swinging, encouraging investors to invest more, and resulting in a stronger flow-performance relationship. In contrast, losses due to market downturns during bear markets would cause and/or increase risk aversion, consequently weakening the flow-performance relationship. We test these three issues using a novel sample of Chinese mutual funds. Chinese mutual fund markets and investors differ from those in other countries in several important ways. First, Chinese investors face limited investment opportunities and are more risk-tolerant than investors in the U.S. and other countries (Stulz and Wasserfallen, 1995; Sun and Tong, 2000). Second, Chinese investors trade much more frequently, hold fewer stocks, and have a shorter holding period than U.S. investors (Allen, Qian and Qian, 2005; Mei, Scheinkman and Xiong, 2009; Dhar and Zhu, 2006; Chen, Kim, Nofsinger and Rui, 2007). Third, Chinese in general are overconfident compared to Americans (Yates, Lee and Shinotsuka, 1996). Fourth, herding is more prevalent among Chinese investors than those in other markets (Tan, Chiang, Mason and Nelling, 2008; Chiang and Zheng, 2010). Fifth and finally, Chinese investors are vulnerable to market movements, and their investment sentiments and behavior exhibit oscillating patterns. Specifically, under bull markets, Chinese investors are over-optimistic and tend to overreact to positive news, whereas under bear markets, they are extremely pessimistic (Lu and Xu, 2004；Shi, Li, and Chen, 2009). These factors, as discussed by Barberis and Xiong (2009), could lead to different flow-performance relationships. In addition, Chinese stock markets are extremely volatile (see Figure 1), and Chinese mutual fund markets have a relatively short history and are much smaller compared to other developed markets.2 Thus, Chinese mutual funds offer an ideal
and trade more actively after market increases because traders attribute returns from the overall market increases to their own acumen when assessing their trading ability. 2
Massa (2003) reports that the number of mutual funds in the U.S. reached 8,171 in 2000. We retrieved the U.S. mutual fund data from the CRSP Mutual Fund Database and found that there were 26,708 funds at the end of 2009, among which 9,713 funds were domestic equity funds listed in the U.S. After deleting funds that hold less than 50%
4
environment for testing how volatile market condition affects fund flow-performance relationship and whether the asymmetric flow-performance relationship observed in studies using U.S. sample is sample sample specific or it exists in other countries. Using a sample of actively managed, open-ended equity mutual funds in China from 2004 to 2009, we establish several key findings. First, confirming previous studies, we find that the net fund flow is positively related to past fund performance, regardless of whether performance is measured by raw returns or by the risk-adjusted abnormal returns based on the capital asset pricing model (CAPM) or the Fama-French (1993) three-factor model. Second, the flow-performance is negatively related to stock market risk. Third, in sharp contrast with the studies on U.S. mutual funds, our results indicate that there exists no asymmetric flow-performance relationship between poorly performing funds and good performers. These results suggest that investors become less responsive to good performing funds when the stock market is noisy and when fund performance reflects mainly luck rather than skill. Fourth, the Morningstar rating is positively related to subsequent fund flows, but the difference in the flow-performance relationship is not significant between star and dog funds. This evidence suggests that the star effect due to search cost reduction is marginal in markets with a limited number of funds. Finally, we find that the flow-performance relationship varies significantly under different market conditions and that the positive flow-performance relationship is more pronounced in bull markets. This result is consistent with Thaler and Johnson’s (1990) house money effect and the overconfidence hypothesis proposed by Gervais and Odean (2001). The result also provides additional support that Chinese investors are over-optimistic during bull markets and they are extremely pessimistic under bear markets (Lu and Xu, 2004；Shi, Li, and Chen, 2009).
2. Literature Review and Background Information A. Explanations for the asymmetric flow-performance relationship in common shares, there were still 9,204 actively managed, open-ended equity funds in the U.S. at the end of 2009. In contrast, Chinese mutual fund markets are very small, and there were fewer than 350 actively managed, open-ended equity funds at the end of 2009.
5
The asymmetric flow-performance relationship for mutual funds has attracted much attention and researchers have investigated this issue from different aspects. The major explanations can be summarized as follows: Transaction fees and switching costs. Ippolito (1992) develops a theoretical model and shows that rational investors do not automatically allocate new investments to funds that have recently performed well. These investors also do not routinely close old funds to open new accounts with recent strong performers because of the associated costs. In addition, the costs of selling existing shares exceed the costs of investing in new shares; therefore, investors require disproportionately poor performance to withdraw and reallocate existing investments from poorly performing funds to better-performing funds. As a result, past top-performing funds attract large inflows, whereas poorly performing funds suffer small outflows. Search costs, marketing efforts, and media attention. Because collecting and processing information on financial products are costly and most mutual fund retail investors are not well trained in portfolio analysis, Sirri and Tufano (1998) suggest that mutual fund investors purchase funds that are easier or less costly for them to identify, such as those with extensive marketing efforts, those receiving more media coverage, and those offered by well-known fund families. Drawing on U.S. mutual fund data from 1971 to 1990, Sirri and Tufano (1998) find that search costs and media attention play dominant roles in the asymmetric relationship between fund performance and flow. Investor participation costs. Huang, Wei and Yan (2007) model the effect of investor participation costs on the mutual fund flow-performance relationship, and they classify investor participation costs into two categories: the costs of collecting and analyzing information about funds and the transaction costs from purchasing and redeeming fund shares. The authors suggest that participation costs can lead to different flow responses at difference performance levels and, consequently, to an asymmetric flow-performance relationship. Strategy replacement. Heinkel and Stoughton (1994) develop a dynamic model of portfolio management contracts with a multi-period setting and they show that funds respond to poor performance 6
by replacing portfolio managers or the investment strategies that underperform the benchmarks. Building upon Heinkel and Stoughton’s (1994) implications, Lynch and Musto (2003) suggest that the past performance of poorly performing funds has less predictive power for future performance and hence, has little effect on investor decisions because the portfolio managers of these funds will be replaced or their investment strategies will be changed. Cognitive dissonance and disposition effect. Goetzmann and Peles (1997) show that the perceptions of U.S. mutual fund investors regarding past fund performance are consistently biased toward better-than-actual performance; moreover, the biased recollections regarding past fund performance cause investors to continue to hold poorly performing funds. Therefore, the cognitive dissonance explanation suggests that investors adjust their beliefs and seek support for their past investment decisions to reduce psychological costs and cognitive dissonance, which leads to the asymmetric and convex relationship between fund performance and flow. Similarly, Shefrin and Statman’s (1985) disposition effect indicates that investors sell winners too early and ride losers too long. The investor clientele effect. Del Guercio and Tkac (2002) compare the flow-performance relationships for investors in retail mutual funds and fiduciary pension funds. These authors find a systematic difference in the shape of the flow-performance between these two groups of investors. Pension fund clients punish poorly performing funds by withdrawing assets under management and do not flock disproportionally to recent winners. Their evidence implies that an approximately linear relationship exists between flow and performance. In sharp contrast, mutual fund investors chase and flock to past winners and do not withdraw assets from poorly performing funds. Sawicki (2001) investigates the flow-performance relationship using Australian wholesale funds, which are traded primarily by large, institutional investors. She finds that institutional investors in Australia react to recent performance, but the response is not asymmetric. Christoffersen and Musto (2002) argue that investors
7
have different demand curves and that the investors of bottom funds are relatively less sensitive to performance and price.3
B. Star funds and the flow-performance relationship Research on mutual fund flow-performance relationship has also led to growing attention regarding the impact of star funds. Nanda, Wang and Zheng (2004) find that new money growth for a star fund is approximately 13% higher than that for a non-star fund. In addition, star funds have significant spillover effects on the cash flow of other funds in a given fund family. Specifically, new money growth for star families (i.e., fund families with at least one star fund) is on average 4.4% higher (or $181 million more) than that of non-star fund families (i.e., fund families without any star funds). Meanwhile, the effect of dog funds (i.e., poorly performing funds) on the cash flow of other funds in the family is negative but not significant. Del Guercio and Tkac (2008) show that an upgrade from four to five stars results in $32 million in abnormal cash inflow (or 25% above normal flow) to the fund in the six months following the upgrade, whereas a downgrade from five to four stars does not have a significant effect on the fund cash flow.
C. Background information on the Chinese mutual fund industry The first Chinese open-ended mutual fund was traded in December 2001. By the end of 2009, the total number of open-ended funds reached 598, including actively managed funds, index funds, money market funds, bond funds, and Qualified Domestic Institutional Investor (QDII) funds. In addition to its short history and small scale, the Chinese mutual fund market differs from the mutual fund markets in the U.S. and other countries in several distinct ways:
3
Despite the prevalence of evidence for an asymmetric flow-performance relationship, several recent studies show that investors punish funds that have performed poorly in the past (see e.g., Shrider, 2009; O’Neal, 2004; Ivkovic and Weisbenner, 2009).
8
First, mutual funds in the U.S. are corporate entities, and each fund is overseen by a board of directors (or trustees) (Tufano and Sevick, 1997; Gil-Bazo and Ruiz-Verdu, 2009), whereas in China, mutual funds are contract funds and not corporate entities, and all issues are addressed by their funding companies. Second, mutual fund investors in the U.S. are the fund shareholders; they own shares that represent a part of their holdings, and their interests are represented by the board of directors. In contrast, Chinese mutual fund investors are the beneficiaries of funds rather than shareholders, and as such, their interests are not well served by the board of directors. Third, in the U.S. mutual fund industry, the fee structure is relatively flexible, fees are negotiated by the board of directors, and management fees vary depending on market competition and fund performance.4 In China, the management fees for most equity funds are fixed at 1.5% of total net assets under management since 2002, and hence, the fees do not directly reflect the fund’s performance. Finally, in the U.S., there are multiple channels for fund distribution, including (1) the direct channel; (2) the advice channel; (3) the retirement plan channel; (4) the supermarket channel; and (5) the institutional channel (Reid and Rea, 2003). However, in China, mutual funds are primarily distributed through commercial banks; securities companies and insurance firms play a minimal role.
3. Data Description and Sample Selection Chinese mutual fund data are collected from the CSMAR dataset compiled by GTA Data Services, which is one of the major financial data suppliers in China. Along with other data items, the CSMAR database includes each fund’s name, code, family name, monthly net asset value, total net assets (TNA) at the end of each quarter, and quarterly raw returns adjusted for dividend payments. Our initial sample includes all diversified open-ended equity funds listed in Chinese markets from 2004 to 2009. Following the literature on mutual funds (Sirri and Tufano, 1998; Nanda et al., 2004), we focus only on actively managed equity funds. Specifically, we exclude index funds, passively managed funds, and QDII,
4
funds
Wahal and Wang (2011) show that changes in management fees are negatively related to the overlap of securities held in incumbent and entrant funds, suggesting that funds compete with each other and attract new investors by changing management fees.
9
because QDII funds invest mainly in foreign stocks. This selection procedure reduces the total number of funds to 348. Similar to other studies (Huang, Wei and Yan, 2007; Oliver and Tay, 2009), we apply two additional filters when computing fund performance. First, a fund must have at least 12 months of historical data on the net asset value which allows us to compute raw returns; second, a fund must have at least 24 months of historical data on the net asset value in order to calculate risk-adjusted returns based on the CAPM and the Fama-French (1993) three-factor models.5 After applying these filters, there are 246 funds remaining in the sample as of the end of 2009. We cut off the sample period at 2004 because there were only a few mutual funds before 2004. Table 1 reports the summary statistics of the sample mutual funds listed on Chinese stock markets from 2004 to 2009. While the Chinese mutual fund market is relatively small, it has exhibited a substantial high growth rate since 2004. The total number of funds increased from 24 at the end of 2004 to 246 at the end of 2009, which is equivalent to a compounded annual growth rate of 52.4% during our test period. 6 The cross-sectional average annual raw return after dividend adjustment fluctuated considerably over the sample period, ranging from -48.52% in 2008 to 117.44% in 2007 in 2007, which is equivalent to a compounded annual return of 26.19% during the six-year test period. The cross-sectional standard deviation of raw returns increased substantially from 14.15% in 2004 to 30.09% in 2007. The large annual return and the high volatility of Chinese mutual funds mirror the volatile stock market environment in China. To get a general view about stock market volatility in China, we report the annual return and the annualized standard deviation of daily stock market return in the last two columns of Table 1. It is clear that the Chinese stock markets are highly volatile, especially during 2007 and 2008. The annualized standard deviation of market return is 35.80% and 39.22% in 2007 and 2008, respectively. Figures 1a and 5
Huang, Wei and Yan (2007) require that funds have at least 20 months of historical data, and Oliver and Tay (2009) select funds that have at least 24 months of historical data. We also filter for funds with at least 36 months of historical data, and the results are qualitatively similar to the results presented here. 6 According to Wahal and Wang (2011), the compounded annual growth rate of mutual funds was 16% in the U.S. between 1980 and 2008.
10
1b provide additional evidence about the volatile feature of Chinese stock markets. The Shanghai Stock Index (see Figure 1a), one of the major indices for Chinese stock markets, experienced an unusual growth, increasing from 1,060.7 points in May 2005 to a historical high of 6124 points in October 2007, and then it plunged to 1,728.0 points in August 2008. The Shenzhen stock index experienced a similar, highly volatile pattern (see Figure 1b). The average fund size measured by total assets increased from 1,221.78 million in Chinese Reminbi (RMB) (or about $147.6 million) at the end of 2004 to RMB 6,589.23 million (or $966.2 million) at the end of 2009, with a peak of RMB 10,299.67 million (or $1,297.5 million) at the end of 2007.7 The average total assets of fund family, which include only the composition funds of the current sample, range from RMB 5,190 million in 2005 to RMB 67,611 million in 2008. Similar to fund size, the distribution of fund family size has a humped shape, reflecting the 2007 bull market.
4. Empirical Results A. Measures of fund flow and performance Following the literature (Sirri and Tufano, 1998; Del Guercio and Tkac, 2002), we use two methods for measuring fund flows. The first measure is the quarterly net capital flow, which is defined as follows: Flowi,t = TNAi,t – TNAi,t-1× (1 + Ri,t),
(1)
where TNAi,t is the ith fund’s total net assets at the end of quarter t (measured in millions of RMB), and Ri,t is the ith fund's return in quarter t. Equation (1) measures the quarterly change in total net assets minus appreciation. The second measure is the percentage net flow, which is the net flow scaled by the fund’s total net assets at the end of quarter t – 1: Flow%i,t = Flowi,t / TNAi,t-1.
(2)
Flow%i,t is percentage net flow for the ith fund in quarter t and can be interpreted as the asset growth rate net of appreciation.
7
The exchange rate between the U.S. dollar and the Chinese RMB was about RMB 8.28 per dollar at the end of 2004 and declined to about RMB 7.37 at the end of 2007 and RMB 6.82 at the end of 2009.
11
To provide a better picture of past fund performance, we use three different performance measures, namely, raw return, the risk-adjusted abnormal return of the CAPM model (αCAPM), and the risk-adjusted abnormal return of the Fama-French (1993) three-factor model (αFF). The monthly raw return is adjusted for dividend distribution and retrieved from the CSMAR dataset compiled by GTA. For each fund in month t, we calculate the fund’s average return in the past 12 months (i.e., the 12-month moving average) and use it as a proxy for the fund’s performance in the past year.8 The risk-adjusted abnormal return of the CAPM model (αCAPM) is calculated in three steps. In the first step, we estimate the alpha (αi) and beta (βi) of the ith fund using the fund’s returns for the past 24 months. The ordinary least square (OLS) regression for estimating αCAPM is specified as: Ri,t – Rf,t = αi +βi(Rm,t – Rf,t) + ei,t ,
(3)
where Ri,t is the ith fund’s raw return in month t, Rf,t is the risk-free rate (i.e., the one-year interest rate for certified deposits in China) in month t, and Rm,t is the monthly return of the Chinese stock market composite index after dividend distribution adjustment. The data are obtained from the CSMAR dataset. In the second step, we compute the monthly risk-adjusted abnormal return as: αi,tCAPM = αi + ei,t.
(4)
In the third step, we calculate the moving average of αi,tCAPM for the past 12 months. The third measure is the risk-adjusted abnormal return based on the Fama-French (1993) three-factor model, which is calculated with a similar method as one used for αi,tCAPM. Specifically, in each month t, we estimate αi, βi, βi,SMB, and βi,HML for each fund using the returns over the past 24 months via the following OLS regression: Ri,t – Rf,t = ci +βi(Rm,t – Rf,t) + βi,SMBSMBt + βi,HMLHMLt + µi,t , (5)
8
As an alternative measure, we also compute the compounded annual return based on the monthly returns from the past 12 months. The flow-performance relationship-based compounded annual returns and the average returns for 12 months are qualitatively similar, and the results are not reported for brevity consideration.
12
where SMBt and HMLt denote the returns on portfolios based on firm size (i.e., return difference between small and large portfolios) and book-to-market value ratio (i.e., return difference between high and low boot-to-market portfolios) in month t, respectively. All other variables are as defined in previous equations. We then compute the monthly risk-adjusted return of the Fama-French three-factor model as: αitFF = ci +µi,t.
(6)
Next, we compute the moving average of αitFF for the past 12 months.
B. Basic model and preliminary analysis of the flow-performance relationship To provide a preliminary overview of the flow-performance relationship, we rank each fund at the beginning of quarter t based on the fund returns in the past 12 months and divide all funds into ten portfolios of approximately equal size in descending order. Each fund is assigned a ranking score from zero for the funds in the worst-performing group to nine for the funds in the best-performing group. Then, we compute the average flows for each of the ten portfolios in each quarter. Figure 2a plots the sample average for the net flow of the ten portfolios ranked by raw returns during our sample period. Although the net flow varies among the ten portfolios and the flow-performance does not exhibit a perfect linear relationship, the overall pattern indicated by the dashed trend line shows a positive relationship between flow and performance. Figure 2b plots the percentage net flow of the fund portfolios ranked by raw returns, and the overall pattern is similar to that reported in Figure 2a. Furthermore, unlike the results reported by Sirri and Tufao (1998), the net flows do not show a noticeable asymmetric pattern between the top and bottom funds. In addition, we also rank funds based on returns measured by αitCAPM and αitFF, and the unreported results show that the flow-performance relationship is consistent with that illustrated by Figures 2a and 2b. To test the flow-performance relationship, we first follow Nanda et al. (2004) and conduct a fixed effect regression using unbalanced panel data. The regression is specified as follows:
13
Flowi,t (Flow%) = αi + α1Ranki,[t-4,t-1] + α2 StdM,[t-4,t-1] +α3 ln(TNAi,t-1) + α4ln(Agei,t-1) + α5Stdi,[t-4,t-1] + α6ln(Familysizei,t-1) + α7 front-end loadi,t-1 + α8expenseratioi,t-1+µi,t,9
(7)
where the dependent variable is net flow (percentage net flow) to the ith fund in quarter t, and the independent variables are explained as follows: Ranki,[t-4, t-1] is the ranking score between zero and one based on the fund’s performance in the past year (i.e., from quarters t – 4 to t – 1); StdM,[t-4,t-1] is the annualized standard deviation of daily stock market return in the past year and it measures the overall stock market risk; Ln(TNAi,t-1) is the log transformation of the total net assets of the ith fund at the end of quarter t – 1; Ln(Agei,t-1) is the log transformation of age of the ith fund at the end of quarter t – 1, measured in number of years; Stdi,[t-4, t-1] is the annualized standard deviation of the ith fund’s monthly returns in the past year (i.e., from quarter t – 4 to t – 1); Ln(Familysizei,t-1) is the log transformation of total net assets under management in the fund family to which the ith fund belongs at the end of quarter t – 1, excluding the net assets of the ith fund; Front-End Loadi,t-1 is the ith fund’s front-end fee ratio at the end of quarter t – 1; Expense-ratioi,t-1 is the ith fund’s expense ratio at the end of quarter t – 1. We use the ranking score (Rank) instead of the actual performance measures in the regression because the ranking score explains fund growth much better than the cardinal measures, and it also reduces the impact of potential outliers (Patel et al., 1991). StdM,[t-4,t-1] measures the overall stock market volatility since Kim (2011b) indicates that stock market volatility affects investor’s decision on mutual fund. We use Ln(TNAi,t-1) to control for fund growth potential since large funds are generally more difficult to grow (Chevalier and Ellison, 1997; Sirri and Tufano, 1998). Given the fact that the age of a fund affects investor preferences because older funds grow more slowly than younger funds (Bergstresser 9
Note that in equation (7), the regression intercept and the fixed effect are combined and represented by α i to simplify the equation. Similar notation is used in all of the flow-performance regressions in the remainder of the paper.
14
and Poterba, 2002; Del Guercio and Tkac, 2002), the natural logarithm of the fund’s age is included in the regression. Because investors are generally risk-averse and the riskiness of a fund could negatively affect its growth, we use the annualized standard deviation of fund returns in the past 12 months (Stdi,[t-4, t-1]) as a proxy for risk.10 We include fund family size as a proxy for fund publicity because large fund families attract more publicity, which potentially affect fund flows (Goetzmann and Peles, 1997). Finally, we include front-end load and management fee ratios, which are found to affect flows (Sirri and Tufano, 1998; Nanda, Wang, and Zheng, 2004; Gil-Bazo and Ruiz-Verdu, 2009). The regression results of equation (7) are reported in Panel A of Table 2. In columns 1 and 2, the fund performance is measured by the ranking score (Rank) based on the raw returns for the past year, and the dependent variables are net flows (measured in ten millions of RMB) and percentage net flows, respectively. Our primary interest is to examine how past fund performance affects subsequent net flows. The coefficients on performance rank are all positive and significant at the 0.01 level, indicating that past performance is positively related to the subsequent fund flows, consistent with the findings in the literature. The coefficients on StdM,[t-4,t-1] are all negative and significant, indicating that high stock market volatility reduces fund flows, consistent with the results of Kim (2011b). Differing from StdM, the coefficient on Stdi is negative and significant at the 0.1 level in Flow regression (column 1) but positive and insignificant in Flow% regression (column 2), suggesting that Chinese mutual fund investors are more sensitive to the volatility of stock markets than to the volatility of individual fund. In addition, the coefficients on ln(TNA) and ln(Age) are all negative and significant at the 0.05 or higher level, suggesting that larger and older funds grow more slowly than smaller and younger funds, similar to the findings of Del Guercio and Tkac (2002) and Shrider (2009). The coefficients on Ln(Familysize) are positive and indicate that large fund families attract more flows. Differing from studies using U.S. samples (Sirri and Tufano, 1998; Nanda, Wang and Zheng, 2004; Gil-Bazo and Ruiz-Verdu, 2009), our results show that most of the coefficients on the front-end load and the expense ratios are insignificant, which is due to the rigid fee structure in China, as argued before. 10
Sirri and Tufano (1998) find a negative, but insignificant, relationship between fund riskiness and growth.
15
The results on the flow-performance relationship based on αCAPM and αFF are reported in columns 3 - 4 and 5 - 6, respectively. The sample size in the αCAPM and αFF regressions is smaller than that in the regressions using raw returns because we require the funds to have at least 24 months of data when estimating αCAPM and αFF. Regardless of which performance measure is used, past fund performance is always positively related to subsequent flows, while the market volatility is always negatively related to flows. These relationships are robust and significant after controlling for other factors. We conduct Hausman tests (1978) and F-tests to investigate whether a random or fixed effect model fits our sample better for the pooled data regression. The p-values of the F-tests are small and reject the null hypothesis of no fixed effects. Since p-values of the Hausman tests are close to zero, the null hypothesis that individual effects are uncorrelated with other regressors in the model is rejected. These results suggest that a random effect model produces biased estimates; as such, a fixed effect model is preferred (Greene, 2003). Therefore, we control for fixed effects in all regressions for the remainder of the paper. For brevity, the Hausman tests and the F-test results are not reported in the remaining tables. To provide additional evidence on the flow-performance relationship, we replace StdM with an interaction variable of Rank*StdM, which measures the sensitivity of fund flow to performance associated with stock market volatility, and the results are reported in Panel B of Table 2. The coefficients on performance (Rank) remain positive and significant at the 0.01 level. More importantly, all coefficients on the interaction variable of Rank*StdM are negative and significant at either 0.05 or 0.01 level. These results show that investors respond less positively to fund performance when the stock market is volatile, and the volatile market increases the noise of fund performance. The coefficients on all other variables are largely consistent with those reported in Panel A of Table 2. C. Test of asymmetric relationships between flow and performance Although we observe a positive relationship between flow and performance, this relationship does not necessarily mean that investor response to past fund performance is universal across different performance ranges. To test whether an asymmetric relationship exists between flow and performance, we 16
conduct two tests. First, we add a squared performance measure as Rank_square and an interaction variable of squared performance and market volatility, Rank_square*StdM, which captures the incremental effect of stock market volatility on the potential asymmetric flow-performance relationship. The regression is specified as follows: Flow(%)i,t = αi + α1Ranki,[t-4,t-1] + α2Rank_squarei,[t-4,t-1] + α3Rank_squarei,[t-4,t-1] *StdM,[t-4,t-1] +α4-9Controls +µi,t,
(8)
where, dependent variable is percentage of net flow (Flow%), and control variables include Ln(TNAi,t-1), Ln(Agei,t-1), Stdi,[t-4, t-1], Ln(Familysize i,t-1), Front-End Loadi,t-1, and Expense-ratioi,t-1, which are defined previously in equation (7). The results of equation (8) are reported in Table 3. In column (1), the coefficient on performance (Rank) is positive and significant at the 0.05 level, confirming a positive relationship between flow-performance. However, the coefficient on the squared performance measure (Rank_square) is negative but statistically insignificant, which indicates that flow-performance relation is linear, different from the convexity relationship documented in the previous studies using the U.S. data. The coefficients on other variables are similar to the results reported in Table 2. In column 2, we report the results with the interaction variable of Rank_square*StdM. The coefficient on Rank_square remains negative and insignificant, whereas the coefficient on Rank_square*StdM is negative and significant at the 0.1 level. We repeat the regression with αCAPM and αFF (see columns 3-4, and columns 5-6, respectively). The coefficients on Rank_square*StdM are consistently negative and significant at the 0.05 level. Although the coefficients on Rank are all positively, they are insignificant (see columns 5 and 6). Other results are similar to those reported in columns 1 and 2 and those reported in Table 2. As an additional test, we use new flows measured in RMB instead of percentage flows, the results (unreported) are qualitatively similar. In summary, these results indicate that investors become less responsive to high fund performance when market is noisy, which is the primary driving force for the linear flow-performance relationship.
17
To provide further evidence on the flow-performance relationship, we separately estimate the slopes for good and poor performers by incorporating two interaction dummy variables, High and Low, in the following regression:11 Flowi,t (Flow%) = βi+β1Ranki,[t-4,t-1]*Highi[t-4,t-1] +β2 Ranki,[t-4,t-1]*Lowi[t-4,t-1] + β3-8 Control Variables +µi,t.
(9)
Where High (Low) takes a value of 1 if the ith fund is ranked in the top (bottom) 50% based on its performance in the past year (i.e., from quarter t – 4 to t – 1) and 0 otherwise. The control variables include StdM,[t-4,t-1], Ln(TNAi,t-1), Ln(Agei,t-1), Stdi,[t-4,
t-1],
Ln(Familysize
i,t-1),
Front-End Loadi,t-1, and
Expense-ratioi,t-1. Specifically, β1 and β2, the coefficients on the two interaction variables, measure flow sensitivity to past performance for funds that perform well and those that perform poorly, respectively. If β1 – β2 =0, this indicates that investors respond to past performance symmetrically for good and bad funds. In contrast, if β1 – β2 >0, this implies that investor’s response to past performance is more sensitive for funds that perform well than for poor performers, which is the convex relationship documented in previous studies. Table 4 reports the regression results of equation (9). Regardless of which performance measure is utilized and whether the fund flows are measured in terms of net flow or percentage net flow, the coefficients on β1 are always positive and significant at the 0.01 level, suggesting that the net flow is positively correlated with past performance for funds that perform well. For poorly performing funds, similar results are observed. By large, the coefficients on β2 are significantly positive in the regressions. In all the regressions, the Wald tests fail to reject the null hypothesis of β1 – β2=0, suggesting that the difference in flow-performance relationships between funds that perform well and those that perform poorly is not statistically significant. The result contradicts to the asymmetric relationship observed in other studies on U.S. mutual funds.12
11
Many studies use a similar method to estimate separate slopes and test the differences (Lynch and Musto, 2003; Oliver and Tay, 2009). 12 To provide additional evidence, we rank and categorize all funds into three groups based on the past year’s performance: top, middle and bottom. Top (Bottom) takes a value of 1 if a fund is ranked in the top 25% (bottom
18
D. Test of flow sensitivity to the Morningstar rating Several studies observe the importance of the Morningstar rating on a fund’s flows. For instance, Nanda, Wang and Zheng (2004) report that new money flow to star funds rated by Morningstar is approximately 13% higher than that of similar non-star funds. Del Guercio and Tkac (2008) show that an upgrade from four stars to five stars increases the fund inflow by approximately 25%, whereas a downgrade from five to four stars does not have a significant effect on a fund’s cash flow. One of the explanations for the asymmetric relationship between the flows to star and non-star funds is investor search costs. Del Guercio and Tkac (2008) argue that the Morningstar rating offers investors a reputable and unbiased source for evaluating funds, reduces investor search costs, and provides valuable certification for financial professionals. We conjecture that the star effect due to search costs would be weak or diminished in Chinese markets because of a limited number of funds available in the markets. In addition, we expect that the effect of searching cost is further mitigated in the volatile Chinese markets because of the noise in performance. In contrast, the star effect is expected to be stronger (positive) and the quality certification of star rating is expected to be more important in the emerging Chinese markets, which are often plagued by severe asymmetric information problem. To carry out this test, we interact Morningstar ratings (Star, Medium and Dog) with fund performance (Rank) and conduct the following regression: Flowi,t (Flow%) = βi+ β1Ranki,[t-4,t-1]*Stari[t-4,t-1] + β2 Ranki,[t-4,t-1]*Mediumi[t-4,t-1] + β3Ranki,[t-4,t-1]*Dogi[t-4,t-1] + β 4-10 Control Variables +µi,t,
(10)
where Star takes a value of 1 if the ith fund is a star fund (i.e., the fund has five stars) rated by Morningstar at the beginning of quarter t based on performance in the past year (i.e., from quarter t – 4 to t – 1) and 0, 25%) in terms of the past year’s performance and 0 otherwise; Middle takes a value of 1 for all other funds. Similar to equation (9) and the method used by Sawicki (2001), we interact these dummy variables with performance rank and estimate flow sensitivity to performance separately for funds in the top, middle, and bottom groups. The unreported results show that the coefficients on the interaction variables Rank*Top and Rank*Middle are all positive and significant at the 0.05 or higher levels. However, the coefficients on the interaction variable Rank*Bottom are positive and significant at the 0.05 or higher level in only three out of six regressions. More importantly, the
difference in the coefficients between interaction variables Rank*Top and Rank*Bottom are insignificant, similar to the results reported in Table 4. For brevity, these results are not reported, but available from authors upon request. 19
otherwise; Medium takes a value of 1 for funds that have 2 to 4 stars rated by Morningstar and 0, otherwise; Dog takes a value of 1 if the fund is a dog fund (i.e., the fund has only one star) and 0, otherwise. The control variables include StdM,[t-4,t-1], Ln(TNAi,t-1), Ln(Agei,t-1), Stdi,[t-4, t-1], Ln(Familysizei,t-1), Front-End Loadi,t-1, and Expense-ratioi,t-1, which are defined in previous regressions. Table 5 reports flow-performance relationships based on the Morningstar ratings. The coefficients on the performance interaction variables β1 and β2 are all positive and significant at the 0.05 or higher level in all regressions, except for β1 in column 5. In contrast, only one out of six coefficients on β3 is significant at the 0.05 level (column 4). Furthermore, the Wald tests fail to reject the null hypothesis of β1 – β3 = 0, and the difference between β1 and β3 is not statistically significant at the 0.1 or higher levels. Confirming the results of Table 4, these results further indicate that the difference in flow sensitivity to past performance between stars and dogs is not statistically significant. The implication is that the role of the star rating in reducing search costs is minimal, especially in a relatively underdeveloped and highly volatile Chinese markets. Furthermore, consistent with the results reported in Tables 2 and 4, the coefficients on StdM are all negative and significant at the 0.05 or higher levels, suggesting that investors become less responsive to good performance when market is noisy. E. Test of flow and performance relationships under different stock market conditions The behavior studies show that investors are less risk aversive and invest more aggressively when they are more confident (Thaler and Johnson, 1990; Gervais and Odean, 2001). Accordingly, we posit that the flow-performance relationship varies under different market conditions and is more pronounced during bull markets. To test this hypothesis, we incorporate two dummy variables, Up and Down, to capture the stock market conditions of bull and bear markets, respectively. Up (Down) takes a value of 1 if the market return for the past year (i.e., from quarter t – 4 to t – 1) is greater (equal to or less) than 0, i.e.,
20
Rm,[t-4,t-1] >0 (Rm,[t-4,t-1] ≤0), and 0, otherwise.13 Similar to equations (9) and (10), we interact Up and Down with the performance ranking variable and modify the regression as follows: Flowi,t (Flow%) = βi+β1Ranki,[t-4,t-1]*Upi,[t-4,t-1]+β2Ranki,[t-4,t-1]*Downi,[t-4,t-1] + β3-8ContralVariables+µi,t.
(11)
If the flow-performance relationship is more pronounced during bull markets than bear markets, we expect β1 to be greater than β2, i.e., β1 – β2>0. Table 6 reports the regression results of equation (11) and the Wald tests for differences between β1 and β2. All of the coefficients on the interaction variable Rank*Up (β1) are positive and significant at the 0.01 level; meanwhile, the coefficients on Rank*Down (β2) are positive, but they are only significant in three out of six regressions. More importantly, the difference in β1 and β2 is positive and significant at the 0.01 level, and the Wald tests reject the null hypothesis of β1 - β2 = 0 in all regressions. The results suggest an asymmetric relationship between flow and performance in different market conditions. Consistent with our hypothesis and the extreme vulnerable Chinese investor behavior observed by Lu and Xu (2004) and Shi, Li and Chen (2009), these results indicate that Chinese investors respond more positively to past fund performance during bull markets than bear markets. To provide further evidence on the effects of market conditions on the flow-performance relationship, we incorporate interaction variables for market conditions (i.e., Up and Down) and Morningstar ratings (namely, Star, Medium, and Dog). The regression is specified as follows: Flowi,t (Flow%) = βi+ β1Ranki,[t-4,t-1]*Stari[t-4,t-1]*Upi[t-4,t-1]+ β2Ranki,[t-4,t-1]*Stari[t-4,t-1]* Downi[t-4,t-1] + β3Ranki,[t-4,t-1]*Mediumi[t-4,t-1]*Upi[t-4,t-1]+ β4Ranki,[t-4,t-1]* Medium[t-4,t-1]*Downi[t-4,t-1] + β5Ranki,[t-4,t-1]* Dogi[t-4,t-1]* Upi[t-4,t-1]+β6 Ranki,[t-4,t-1]*Dogi[t-4,t-1]*Downi[t-4,t-1]+ β7-12ControlVariables + µi,t.
(12)
The coefficients on the interaction variables Rank*Star*Up and Rank*Star*Down, denoted by β1 and β2, measure flow sensitivity to the past performance of star funds during bull and bear markets, respectively. We also conduct the Wald test for the difference between β1 and β2 and report the p-values in square 13
Many studies use a similar method to classify market conditions (Pettengill et al., 1995; Fletcher, 2000).
21
brackets in Table 7. As expected, the differences between β1 and β2 are all positive and significant at the 0.01 level, which suggests that for star funds, the positive relationship between flow and performance is stronger in bull markets than in bear markets. For funds rated as mid-level performers by Morningstar, β3 and β4 measure flow sensitivity to past performance during bull and bear markets, respectively. Consistent with the results observed for star funds, the Wald tests reject equality between β3 and β4. These results indicate that an asymmetric flow-performance relationship also exists for mid-level funds in different market conditions. Similarly, the coefficients on the interaction variables of Rank*Dog*Up and Rank*Dog*Down, denoted by β5 and β6, measure flow sensitivity to the past performance of dog funds during bull and bear markets, respectively. In sharp contrast to their top and mid-level counterparts, the insignificant differences between β5 and β6 indicate that for dog funds, there is no difference in flow sensitivity to past performance between bull and bear markets. For completeness, we also test the differences between β1 and β5 and between β2 and β6. The Wald tests fail to reject the null hypotheses of β1 – β5 = 0 and β2 – β6 = 0 in all regressions. This further confirms our observation in Table 5 that there is no significant difference in flow-performance relationship between star and dog funds, regardless of market conditions. As additional robustness checks, we replace Morningstar ratings with the Top, Middle and Bottom variables based on raw returns, αCAPM, αFF and interact these variables with the market condition variables Up and Down. The unreported results are qualitatively similar to the results reported here. The overall results indicate that market conditions directly affect investor response to past fund performance. Moreover, the flow-performance relationship varies under different market conditions, whereas the difference in the flow-performance relationship between funds that perform well and those that perform poorly is not significant.
5. Additional tests
22
As indicated by Del Guercio and Tkac (2002), the difference in flow-performance relationships suggests that the main drivers of mutual fund flows vary in different market conditions and could be sample-specific due to clientele differences. However, due to the limited publicly available data, similar to Del Guercio and Tkac (2002) and Sawicki (2001), we do not have access to detailed transaction information at fund level and are unable to identify which aspects of the clientele differences contribute to these variations. As additional attempts to identify the aggregated differences between Chinese and US mutual fund trends, we test fund risk-taking and fund turnover for Chinese mutual funds. These results help elucidate how Chinese mutual funds differ from their U.S. counterpart. Following the literature (Chevalier and Ellison, 1997), we use the annualized standard deviation of a fund’s monthly abnormal returns (Ri -Rm) in the past 12 months to measure fund manager’s risk-taking. We compute the fund annual turnover ratio by dividing the minimum of a fund’s investment in new stocks or its sales of existing stock by the fund’s average net asset value during a given year. The average annualized standard deviation of monthly abnormal returns is 14.2% during the entire test period (Panel A, Table 8), which is significantly larger than the value of 6.3% for U.S. funds reported by Massa and Patgiri (2009) (also see Chevalier and Ellison, 1997). In addition, the difference in fund volatility between the top and bottom funds is not significant during the entire test period. Panel B of Table 8 reports the turnover ratio. The average annual turnover ratio for the entire sample is 327.6% during the entire test period, which is substantially higher than the average turnover ratio for U.S. funds.14 The difference in the turnover ratios between the top and bottom funds is significant at the 0.01 level, with 438.7% for the bottom funds and 275.6% for the top funds. Combined with the symmetrical flow-performance relationship between the top and bottom funds observed in the current study, the large volatility and higher turnover ratio (especially for bottom funds) imply that when market is highly volatile, good performing funds do not attract additional flows and poorly performing funds update their portfolios 14
Nanda et al. (2004) report that the annual turnover ratio is about 80% for U.S. funds from 1992 to 1998. Massan and Patgiri (2009) show that the average annual turnover ratio for U.S. funds is about 96.6% from 1996 to 2003.
23
more frequently to due to pressure from investor redemption. At a minimum, these additional results provide further evidence that Chinese mutual investors are different from U.S. investors; as such, the clientele effect and fund structure differences also appear to be the primary drivers of the different flow-performance relationships observed in the current study.
6. Conclusions and Discussions A growing literature has demonstrated an asymmetric flow-performance relationship in U.S. mutual fund markets, and many theories have emerged to explain this asymmetric relationship. The current study re-examines the flow-performance relationship by using a novel sample of Chinese mutual funds. Because Chinese stock markets are highly volatile and Chinese investors are extremely vulnerable to stock market conditions, this unique sample enables us to better understand how market condition affects flow-performance relationship. The results, which are based on a sample of actively managed, open-ended equity funds from Chinese markets, help us draw the following conclusions. First, net flows to funds are positively related to past fund performance. This positive relationship is consistent based on different performance and net flow measures, and it is robust after controlling for fund characteristics. In addition, the net flows to funds are negatively correlated with fund size and age, which suggests that larger and older funds grow more slowly than smaller and younger funds. Similar to the findings in previous studies on U.S. mutual funds, these results suggest that past fund performance affects the investment decisions of mutual fund investors. Second, stock market volatility have significant negative effect on fund flows, and flow sensitivity to past performance is not significantly different between funds that perform well and those that perform poorly. Third, differing from previous studies, we do not find a significant difference in flow sensitivity to performance between star and dog funds. This evidence suggests that the star effect due to search costs is marginal in markets with a limited number of funds. Fourth, our results indicate that stock market volatility reduces flow sensitivity to past fund performance, implying that investors respond less positively to the past performance when performance
24
reflects noise rather than stock picking skill and good performing funds do not attract disproportionately large flows. Finally, our results show that the flow-performance relationship varies under different market
conditions; namely, the positive flow-performance relationship is more pronounced during bull markets than bear markets. This finding is consistent with the house money effect, and it also indicates that investors are overconfident during bull markets. The current study also expands the literature by showing that differences in investor clientele could lead to different flow-performance relationships and, moreover, that market conditions affect mutual fund investor behavior. These new findings raise several issues that need to be addressed in future studies. For example, our results are based on aggregated net fund flows because of limited publicly available data. We believe that detailed transaction data at the investor level would provide more interesting findings as to which aspect of investor clientele contributes to the different flow-performance relationship. Is it due to gender, age, investment experience, or risk tolerance? In addition, previous studies show that the asymmetric flow-performance relationship between funds that perform well and those that perform poorly directly affects the incentives for fund managers; therefore, it would be interesting to investigate whether the incentives for fund managers and their risk-taking behavior change under different market conditions. Although these issues are beyond the scope of the current study, further tests would provide fruitful results to understand the dynamics of the flow-performance relationship.
25
Reference: Allen, Franklin, Jun Qian, and Meijun Qian, 2005. Law, finance, and economic growth in China, Journal of Financial Economics 77, 57-116. Barberis, Nicholas, and Wei Xiong, 2009. What drives the disposition effect? An analysis of long-standing preference-based explanation, Journal of Finance 64, 751-784. Bergstresser, Daniel, and James Poterba, 2002. Do After-tax returns affect mutu al fund inflows? Journal of Financial Economics 63, 381-414. Berk, Jonathan, and Richard Green, 2004. Mutual fund flows and performance in rational markets, Journal of Political Economy 112, 1269-1295. Chen, Congmeng, Kenneth Kim, John Nofsinger, and Oliver Rui, 2007. Trading performance, disposition effect, overconfidence, representativeness bias, and experience of emerging market investors, Journal of Behavioral Decision Making 20, 425-451. Chevalier, Judith, and Glenn Ellison, 1997. Risk taking by mutual funds as a response to incentives, Journal of Political Economy 105, 1167-1200. Chiang, Thomas C., and Dazhi Zheng, 2010. An empirical analysis of herd behavior in global stock markets, Journal of Banking and Finance 34, 1911-1921. Christoffersen, Susan K., and David K. Musto, 2002. Demand curves and the pricing of money management, Review of Financial Studies 15, 1499-1524. Del Guercio, Diane, and Puala A. Tkac, 2002. The determinants of the flow of funds of managed portfolios: mutual funds vs. pension funds, Journal of Financial and Quantitative Analysis 37, 523-557. Del Guercio, Diane, and Puala A. Tkac, 2008. Star power: the effect of Morningstar ratings on mutual fund flows, Journal of Financial and Quantitative Analysis 43, 907-936. Dhar, Ravi, and Ning Zhu, 2006. Up close and personal: investor sophistication and the disposition effect, Management Science 52, 726-740. Fama, Eugene F., and Kenneth R. French, 1993, Common risk factors in the returns on stocks and bonds, Journal of Financial Economics 33, 3-56. Fant, L. Franklin, and Edward S. O’Neal, 2000, Temporal changes in the determinants of mutual fund flows, Journal of Financial Research 23, 353-371. Ferreira, Miguel A., Aneel Keswani, Antonio F. Miguel, and Sofia Ramos, 2012. The flow-performance relationship around the world, Journal of Banking and Finance 36, 1759-1780. Fletcher, Jonathan, 2000. On the conditional relationship between beta and return in international stock returns, International Review of Financial Analysis 9, 235-245. Gervais, Simon, and Terrance Odean, 2001. Learning to be overconfident, Review of Financial Studies 14, 1-27.
26
Gil-Bazo, Javier, and Pablo Ruiz-Verdu, 2009. The relation between price and performance in the mutual fund industry, Journal of Finance 64, 2153-2183. Goetzmann, William N., and Nava Peles, 1997. Cognitive dissonance and mutual fund investors, Journal of Financial Research 20, 145-158. Greene, W. H., 2003. Econometric Analysis, 5th ed., Upper Saddle River: Prentice Hall. Hausman, J. A., 1978. Specification Tests in Econometrics, Econometrica 46, 1251-1271. Heinkel, Robert, and Neal M. Stoughton, 1994. The dynamics of portfolio management contracts, Review of Financial Studies 7, 351-388. Huang, Jennifer, Kelsey D. Wei, and Hong Yan, 2007. Participation costs and the sensitivity of fund flows to past performance, Journal of Finance 62, 1273-1311. Huang, Jennifer, Kelsey D. Wei, and Hong Yan, 2012. Investor learning and mutual fund flows, working paper, University of Texas at Austin. Ippolito, Richard A., 1992. Consumer reaction to measures of poor quality: evidence from the mutual fund industry, Journal of Law and Economics 35, 45-70. Ivkovic, Zoran, and Scott Weisbenner, 2009. Individual investor mutual fund flows, Journal of Financial Economics 92, 223-237. Kim, Min S., 2011a. Information asymmetry and incentives for active management. Working paper, University of New South Wales. Kim, Min S., 2011b. Changes in mutual fund flows and managerial incentives. Working paper, University of New South Wales. Lu, Rong, and Long-bing Xu, 2004. The analysis of asymmetric Information under bull and bear markets, Economics Research Journal 3, 65-72 (published in China). Lynch, Anthony W., and David K. Musto, 2003. How investors interpret past fund returns, Journal of Finance 58, 2033-2058. Massa, Massimo, 2003. How do family strategies affect fund performance? When performance maximization is not the only game in town, Journal of Financial economics 67, 249-304. Massa, Massimo, and Rajdeep Patgiri, 2009. Incentives and mutual fund performance: higher performance or just higher risk taking? Review of Financial Studies 22, 1777-1815. Mei, Jianping, Jose Scheinkman, and Wei Xiong, 2009. Speculative trading and stock prices: an analysis of Chinese A-B share premia, Annals of Economics and Finance 10, 225-255. Nanda, Vikram, Z. Jay Wang, and Lu Zheng, 2004. Family values and star phenomenon: strategies of mutual fund families, Review of Financial Studies 17, 667-698. Olivier, Jacques, and Anthony Tay, 2009. Time varying incentives in the mutual fund industry, Working paper, HEC Paris. 27
O’Neal, Edward S., 2004. Purchase and redemption patters of U.S. equity mutual funds, Financial Management 33, 63-90. Patel, Jayendu, Richard Zeckhauser, and Darryll Hendricks, 1991. The rationality struggle: illustration from financial markets, American Economics Review Papers and Proceedings 81, 232-236. Pettengill, Glenn N., Sridhar Sundaram, Ike Mathur, 1995. The conditional relation between beta and return, Journal of Financial and Quantitative Analysis 30, 607-618. Reid, Brian K., and John D. Rea, 2003. Mutual fund distribution channels and distribution costs, Investment Company Institute Perspective 9, 1-20. Sawicki, Julia, 2001. Investors’ differential response to managed fund performance, Journal of Finanical Research 24, 367-384. Shefrin, Hersh, and Meir Statman, 1985. The disposition to sell winners too early and ride losers too long: theory and evidence, Journal of Finance 40, 777-790. Shi, Yongdong, Zhuwei Li, and Wei Chen, 2009. The empirical research on trading strategies of Chinese security investors, Financial Research Journal 11, 129-142 (published in China). Shrider, David G., 2009. Does size matter? An analysis of mutual fund transaction size, Journal of Money, Investment and Banking 7, 29-42. Sirri, Erik, and Peter Tufano, 1998. Costly search and mutual fund flows, Journal of Finance 53, 1589-1622. Stulz, Rene M., and Walter Wasserfallen, 1995. Foreign equity investment restrictions, capital flight, and shareholder wealth maximization: theory and evidence, Review of Financial Studies 8, 1019-1057. Sun, Qian, and Wilson H.S. Tong, 2000. The effect of market segmentation on stock prices: the China syndrome, Journal of Banking and Finance 24, 1875-1902. Tan, Lin, Thomas C. Chiang, Joseph R. Mason, and Edward Nelling, 2008. Herding behavior in Chinese stock markets: an examination of A and B shares, Pacific-Basin Finance Journal 16, 61-77. Thaler, Richard H., and Eric J. Johnson, 1990. Gambling with the house money and trying to break even: the effects of prior outcomes on risky choice, Management Science 36, 643-660. Tufano, Peter, and Matthew Sevick, 1997. Board structure and fee-setting in the U.S. mutual fund industry, Journal of Financial Economics 46, 321-355. Wahal, Sunil, and Albert Wang, 2011. Competition among mutual funds, Journal of Financial Economics 99, 40-59. Xie, Lei, 2011. Time-varying mutual fund performance-flow sensitivity and managerial effort. Working paper, Yale School of Management. Yates, J. Frank, Ju-whei Lee, and Hiromi Shinotsuka, 1996. Beliefs about overconfidence, including its cross-national variation, Organizational Behavior and Human Decision Processes 65, 138-147. 28
Table 1: Summary statistics on mutual funds and stock market volatility Total assets of Stock market Std. dev. of Annual Std. dev. of Total assets fund Family Return Mkt. return fund return fund return of funds (millions RMB) (%) (%) (%) (millions RMB) （%） 20.44 2004 24 1.45 14.15 1,221.78 5,832.83 -15.48 21.82 2005 59 3.64 14.95 1,849.92 5,190.66 -6.95 28.61 2006 91 110.24 22.57 2,068.71 12,180.32 134.00 35.80 2007 143 117.44 30.09 10,299.67 67,611.12 122.17 39.22 2008 211 -48.52 28.42 4,964.55 29,087.38 -64.62 33.06 2009 246 63.22 28.36 6,589.23 42,703.15 89.21 This table reports descriptive statistics for actively managed, open-ended equity mutual funds from China at the end of each year during the test period from 2004 to 2009. The annual return is the cross-sectional average of annual raw returns after dividend adjustments at the end of each year. The data for raw returns are obtained from the CSMAR dataset compiled by GTA. The standard deviation of fund return is the annualized standard deviation of the sample funds at the end of each year based on the monthly returns after dividend adjustments for the past 12 months. The total assets of the fund family are the cross-sectional average of the total assets of the funds that comprise the current sample and belong to the same family. To provide a general view about the volatile Chinese stock markets, we report the annual return and the annualized standard deviation of daily stock market return in China in the last two columns. Year
Number of funds
29
Table 2: Tests of the flow-performance relationship Performance Measures
αCAPM
Raw Return
αFF
Panel A: include StdM to control for stock market volatility Dependent var. Independent Var. Ranki,[t-4,t-1]
Flow(RMB) (1) 712.52*** (4.54)
Flow(%) (2) 0.54*** (4.13)
Flow(RMB) (3) 952.76*** (3.24)
Flow(%) (4) 0.52*** (4.16)
Flow(RMB) (5) 685.50** (2.32)
Flow(%) (6) 0.37*** (2.94)
Control Variables StdM,[t-4,t-1] Ln(TNAi,t-1) Ln(Agei,t-1) Stdi,[t-4,t-1] Ln(Familysizei,t-1) Front-end-loadi,t-1 Expense-ratioi,t-1 P-value of Hausman test P-value of F-test Sample size(# of fund-quarters) R-square
-1833.77*** -2.32*** -1764.36** -2.03*** -1779.26** -2.03*** (-2.59) (-5.23) (-2.22) (-4.96) (-2.23) (-4.98) -1067.73*** -0.97*** -1160.78*** -0.83*** -1149.74*** -0.82*** (-9.12) (-6.30) (-7.20) (-5.71) (-7.14) (-5.67) () -464.51*** -0.25** -569.18*** -0.19 -552.83*** -0.19 () (-3.60) (-2.07) (-2.74) (-1.34) (-2.65) (-1.28) -1232.12* 0.38 -1079.45 -0.25 -1174.39 -0.31 (-1.95) (0.95) (-1.23) (-0.60) (-1.33) (-0.72) 723.23*** 0.50*** 735.24*** 0.48*** 727.15*** 0.48*** (7.62) (6.20) (5.75) (5.80) (5.72) (5.75) -2761.03 -5.16*** 992.19 0.69 963.46 0.68 (-1.43) (-2.84) (0.00) (0.00) (0.00) (0.00) -102.50 0.19 1908.91 1.33 1934.19 1.34 (-0.00) (0.00) (0.00) (0.00) (0.00) (0.00) 0.00
0.00
0.00
0.00
0.00
0.00
0.00
0.00
0.05
0.00
0.08
0.00
2711
2711
1846
1846
1846
1846
0.16
0.20
0.15
0.20
0.15
0.19
30
Table 2 (continued): Performance Measures Raw Return αCAPM αFF Dependent Var. Flow(RMB) Flow(%) Flow(RMB) Flow(%) Flow(RMB) Flow(%) Independent Var. (1) (2) (3) (4) (5) (6) Panel B: include interaction variable of Rank*StdM to control for stock market volatility Ranki,[t-4,t-1] Ranki,[t-4,t-1]*StdM,[t-4,t-1]
1682.94*** (3.44) -2788.73** (-2.55)
1.27*** (3.34) -2.13** (-2.25)
2171.14*** (2.97) -3468.67** (-2.44)
1.44*** (3.82) -2.63*** (-3.29)
1837.47** (2.50) -3307.23** (-2.29)
1.19*** (3.24) -2.37*** (-3.04)
Control Variables -1054.17*** -0.96*** -1154.35*** -0.82*** -1144.39*** -0.82*** (-9.07) (-6.17) (-7.21) (-5.66) (-7.15) (-5.62) -489.77*** -0.37*** -565.27*** -0.30* -552.86** -0.31* Ln(Agei,t-1) (-3.85) (-2.66) (-2.60) (-1.87) (-2.49) (-1.86) -1407.02** 0.17 -1118.77 -0.31 -1270.07 -0.41 Stdi,[t-4,t-1] (-2.21) (0.42) (-1.28) (-0.74) (-1.44) (-0.95) 714.54*** 0.47*** 733.56*** 0.47*** 722.91*** 0.46*** Ln(Familysizei,t-1) (7.72) (6.16) (5.81) (5.80) (5.78) (5.75) -2451.22 -5.08*** 859.21 0.50 647.00 0.35 Front-end-loadi,t-1 (-1.26) (-2.83) (0.00) (0.00) (0.00) (0.00) -313.95 -0.03 2072.17 1.44 2073.44 1.43 Expense-ratioi,t-1 (-0.00) (-0.00) (0.00) (0.00) (0.00) (0.00) Sample size 2711 2711 1846 1846 1846 1846 R-square 0.16 0.19 0.16 0.19 0.15 0.19 This table reports the results of a fixed effect regression on net flows measured in 10 millions of RMB (columns 1, 3, and 5) and percentage net flows (columns 2, 4, and 6) using unbalanced panel data. Ranki,[t-4, t-1] is a ranking score (between 0 and 1) based on the fund’s performance in the past year (i.e., from quarter t – 4 to t – 1) measured by raw returns (columns 1 and 2), risk-adjusted returns of the CAPM model, αCAPM (columns 3 and 4), and risk-adjusted returns of the Fama-French three-factor model, αFF(columns 5 and 6). Control variables are defined as follows. StdM,[t-4,t-1] is the annualized standard deviation of daily stock market return and measures the overall stock market risk. Ln(TNAi,t-1) is the log transformation of the total net assets of the ith fund at the end of quarter t – 1, and it measures fund size. Ln(Agei,t-1) is the age of the ith fund at the end of quarter t – 1 measured in number of years. Stdi,[t-4, t-1] is the annualized standard deviation of the ith fund’s monthly returns over the past year. Ln(Familysize i,t-1) is the total net assets under management in the fund family (excluding the total net asset of the ith fund) to which the ith fund belongs at the end of quarter t – 1. Front-End Loadi,t-1 and Expense-ratioi,t-1 are the ith fund’s front-end fee ratio and expense ratio in quarter t – 1, respectively. In Panel A, we report Hausman and F-tests. The small p-values of the F-tests reject the null hypothesis of no fixed effects, and the small p-values of the Hausman tests reject the null hypothesis that individual effects are uncorrelated with other regressors in the model; The F-test and Hausman test results suggest that a random effect model produces biased estimates, suggesting that a fixed effect model is preferred (Greene, 2003). For brevity, Hausman and F-test statistics are omitted from Panel B and all other tables. In Panel B, we include an interaction variable of Rank*StdM to control for the sensitivity of flow to performance associated with stock market volatility. The numbers reported in parentheses are regression t-statistics. ***, **, and * indicate 0.01, 0.05, and 0.1 significance levels, respectively. Ln(TNAi,t-1)
31
Table 3: Tests of flow-performance relationship and convexity Performance Measures Dependent Var. Independent Var. Ranki,[t-4,t-1] Rank_squarei,[t-4,t-1] Rank_square i,[t-4,t-1]* StdM,[t-4,t-1]
αCAPM
Raw Return Flow (%) (1) (2) 1.37** 1.41** (2.20) (2.30) -0.87 -0.19 (-1.47) (-0.22) -2.00* (-1.70)
αFF
Flow (%) (3) (4) 0.82** 0.85** (2.09) (2.16) -0.31 0.53 (-0.83) (0.96) -2.46** (-2.37)
Flow (%) (5) (6) 0.39 0.43 (1.10) (1.20) -0.03 0.69 (-0.09) (1.33) -2.18** (-2.15)
Control Variables Ln(TNAi,t-1) Ln(Agei,t-1) Stdi,[t-4,t-1] Ln(Familysizei,t-1) Front-end-loadi,t-1 Expense-ratioi,t-1 Sample size (# of fund-quarters) R-square
-0.98*** (-6.32) -0.51 (-4.30) 0.20 (0.50) 0.46*** (5.76) -5.43*** (-3.15) 0.02 (0.00)
-0.96*** (-6.17) -0.42*** (-3.04) 0.16 (0.40) 0.47*** (6.00) -5.04*** (-2.88) 0.01 (0.00)
-0.83*** (-5.71) -0.52*** (-3.54) -0.36 (-0.84) 0.45*** (5.62) 0.46 (0.00) 1.30 (0.00)
-0.83*** (-5.66) -0.38** (-2.32) -0.36 (-0.83) 0.46*** (5.75) 0.35 (0.00) 1.34 (0.00)
-0.83*** (-5.66) -0.51*** (-3.48) -0.40 (-0.93) 0.44*** (5.57) 0.41 (0.00) 1.30 (0.00)
-0.82*** (-5.62) -0.38** (-2.27) -0.44 (-1.01) 0.45*** (5.70) 0.35 (0.00) 1.44 (0.00)
2711
2711
1846
1846
1846
1846
0.19
0.19
0.19
0.19
0.19
0.19
This table reports fixed effect regressions on percentage net flows using unbalanced panel data. Ranki,[t-4,t-1] is a ranking score (between 0 and 1) based on the fund’s performance in the past year (i.e., from quarter t – 4 to t – 1) measured by raw returns (columns 1 and 2), risk-adjusted returns of the CAPM model, αCAPM (columns 3 and 4), and the Fama-French three-factor model, αFF(columns 5 and 6). Rank-square takes the squared value of Ranki,[t-4, t-1], which is used to capture the possible convexity in flow-performance relationship. StdM,[t-4,t-1] is the annualized standard deviation of daily stock market return and measures the overall stock market volatility. Rank_square i,[t-4,t-1]*StdM,[t-4,t-1] is an interaction variable of Rank-square and StdM,[t-4,t-1]. Control variables are defined as follows: Ln(TNAi,t-1) is the log transformation of the total net assets of the ith fund at the end of quarter t – 1, and it measures the fund size. Ln(Agei,t-1) is the age of the ith fund at the end of quarter t – 1 measured in number of years. Stdi,[t-4, t-1] is the annualized standard deviation of the ith fund’s monthly returns over the past year. Ln(Familysize i,t-1) is the total net assets under management in the fund family (excluding the total net asset of the ith fund) to which the ith fund belongs at the end of quarter t – 1. Front-End Loadi,t-1 and Expense-ratioi,t-1 are the ith fund’s front-end fee ratio and expense ratio in quarter t – 1, respectively. The numbers reported in parentheses are regression t-statistics. ***, **, and * indicate 0.01, 0.05, and 0.1 significance levels, respectively.
32
Table 4 Tests of flow sensitivity to past performance for high and low performance funds Performance Measure
αCAPM
Raw Return
αFF
Dependent Var. Independent Var. Ranki,[t-4,t-1] * Highi[t-4,t-1] (β1) Ranki,[t-4,t-1] * Lowi[t-4,t-1] (β2)
Flow(RMB) (1) 742.97*** （4.79） 861.04** （2.30）
Flow(%) (2) 0.56*** （3.97） 0.63** （2.28）
Flow(RMB) (3) 1048.47*** （3.67） 1405.94*** （3.21）
Flow(%) (4) 0.58*** （4.30） 0.81*** （2.91）
Flow(RMB) (5) 736.26** （2.44） 926.43** （2.04）
β1 – β2 [Wald P-value]
-118.07 （0.70）
-0.07 （0.70）
-357.47 （0.26）
-0.23 （0.25）
-190.17 （0.50）
Flow(%) (6) 0.37*** （2.82） 0.38* （1.65） -0.01 （0.92）
Control Variables -1829.98*** -2.31*** -1774.70** -2.03*** -1791.84** -2.03*** （-2.59） （-5.22） （-2.23） （-4.97） （-2.24） （-5.01） -1068.29*** -0.97*** -1159.30*** -0.83*** -1150.09*** -0.82*** Ln(TNAi,t-1) （-9.12） （-6.30） （-7.19） （-5.71） （-7.14） （-5.67） -465.58*** -0.25** -571.90*** -0.20 -555.22*** -0.19 Ln(Agei,t-1) （-3.61） （-2.07） （-2.75） （-1.35） （-2.66） （-1.28） -1219.87* 0.39 -1083.30 -0.26 -1180.26 -0.31 Stdi,[t-4,t-1] （-1.93） （0.97） （-1.23） （-0.61） （-1.34） （-0.72） 723.31*** 0.50*** 736.87*** 0.48*** 728.56*** 0.48*** Ln(Familysizei,t-1) （7.61） （6.20） （5.77） （5.80） （5.72） （5.76） -2752.98 -5.16*** 921.89 0.65 895.92 0.64 Front-end loadi,t-1 （-1.42） （-2.83） （0.00） （0.00） （0.00） （0.00） -49.50 0.23 1916.60 1.33 1929.39 1.33 Expenseratioi,t-1 （-0.00） （0.00） （0.00） （0.00） （0.00） （0.00） Sample size 2,711 2,711 1,846 1,846 1,846 1,846 R-square 0.16 0.20 0.16 0.20 0.15 0.19 This table reports fixed effect regressions on net flows measured in 10 millions of RMB (columns 1, 3, and 5) and percentage net flows (columns 2, 4, and 6) using unbalanced panel data. Ranki,[t-4, t-1] is a ranking score (between 0 and 1) based on the fund’s performance over the past year (i.e., from quarter t – 4 to t – 1) measured by raw returns (columns 1 and 2), risk-adjusted returns of the CAPM model, αCAPM (columns 3 and 4), and the Fama-French three-factor model, αFF (columns 5 and 6). High (Low) takes a value of 1 if the fund is ranked in the top (bottom) 50% based on the fund’s past year performance and 0, otherwise. The coefficient β1 measures flow sensitivity to performance for funds that perform well (i.e., in the top 50%), and the coefficient β2 measures flow sensitivity to performance for poorly performing funds (i.e., in the bottom 50%). We conduct the Wald test for the difference between β1 and β2 and report the p-values in square brackets. Control variables include the following: StdM,[t-4,t-1] is the annualized standard deviation of daily market return and it measures the overall stock market volatility. Ln(TNAi,t-1) is the log transformation of the total net assets of the ith fund at the end of quarter t – 1, and it measures fund size. Ln(Agei,t-1) is the age of the ith fund at the end of quarter t – 1 measured in number of years. Stdi,[t-4, t-1] is the annualized standard deviation of the ith fund’s monthly returns in the past year. Ln(Familysize i,t-1) is the total net assets under management in the fund family (excluding the total net assets of the ith fund) to which the ith fund belongs at the end of quarter t – 1. Front-End Loadi,t-1 and Expense-ratioi,t-1 are the ith fund’s front-end fee ratio and expense ratio in quarter t – 1, respectively. The numbers reported in parentheses are regression t-statistics. ***, **, and * indicate 0.01, 0.05, and 0.1 significance levels, respectively. StdM,[t-4,t-1]
33
Table 5: Test of flow sensitivity to past performance based on the Morningstar ratings Performance Measure Dependent Var. Independent Var. Ranki,[t-4,t-1]*Stari[t-4,t-1] (β1) Ranki,[t-4,t-1]* Mediumi[t-4,t-1] (β2) Ranki,[t-4,t-1]*Dogi[t-4,t-1] (β3) β1 – β3 [Wald P-value]
αCAPM
Raw Return Flow(RMB) (1) 720.02*** （3.06） 746.25*** （4.44） 217.11 （0.38） 502.91 （0.41）
Flow(%) (2) 0.67*** （3.32） 0.52*** （3.78） 0.37 （0.98） 0.30 （0.44）
Flow(RMB) (3) 835.17** （2.32） 1058.03*** （3.31） -72.14 （-0.13） 907.31 （0.15）
αFF Flow(%) (4) 0.63*** （3.28） 0.49*** （4.03） 0.55** （2.10） 0.08 （0.77）
Flow(RMB) (5) 588.37 （1.59） 768.60** （2.43） -416.01 （-0.74） 1004.38 （0.12）
Flow(%) (6) 0.51*** （2.67） 0.33*** （2.68） 0.27 （1.07） 0.24 （0.36）
Control Variables -1795.16** -2.27*** -1847.26** -2.04*** -1916.45** -2.07*** （-2.53） （-5.12） （-2.30） （-4.94） （-2.36） （-5.01） -1142.63*** -1.03*** -1173.62*** -0.84*** -1162.75*** -0.84*** Ln(TNAi,t-1) （-9.15） （-6.39） （-7.10） （-5.71） （-7.03） （-5.67） -557.18*** -0.37*** -562.53*** -0.22 -535.16** -0.21 Ln(Agei,t-1) （-3.88） （-2.72） （-2.66） （-1.49） （-2.51） （-1.41） -1245.78* 0.46 -1136.88 -0.19 -1238.00 -0.24 Stdi,[t-4,t-1] （-1.93） （1.09） （-1.27） （-0.44） （-1.37） （-0.54） 788.88*** 0.55*** 752.47*** 0.49*** 746.65*** 0.49*** Ln(Familysizei,t-1) （7.73） （6.29） （5.71） （5.75） （5.68） （5.70） -2666.17 -5.24*** -462.25 -0.17 -403.91 -0.15 Front-end loadi,t-1 （-1.34） （-2.83） （-0.00） （-0.00） （-0.00） （-0.00） 1414.78 1.47 237.20 0.19 167.55 0.13 Expenseratioi,t-1 （0.00） （0.00） （0.00） （0.00） （0.00） （0.00） Sample Size 2711 2711 1846 1846 1846 1846 R-square 0.17 0.21 0.16 0.20 0.15 0.19 This table reports fixed effect regressions on net flows measured in 10 millions of RMB (columns 1, 3, and 5) and percentage net flows (columns 2, 4, and 6) using unbalanced panel data. Ranki,[t-4, t-1] is a ranking score (between 0 and 1) based on the fund’s performance over the past year (i.e., from quarters t – 4 to t – 1) measured by raw returns (columns 1 and 2), risk-adjusted returns of the CAPM model, αCAPM (columns 3 and 4), and the Fama-French three-factor model αFF (columns 5 and 6). Star takes a value of 1 if the ith fund is a star fund (i.e., the fund has five stars) rated by Morningstar at the beginning of quarter t based on performance in the past year (i.e., from quarter t – 4 to t – 1) and 0, otherwise; Medium takes a value of 1 for funds that have 2 to 4 stars and 0, otherwise; Dog takes a value of 1 if the fund is a dog fund (i.e., the fund has only one star) and 0, otherwise. The coefficients β1, β2, and β3 measure flow sensitivity to performance for star, medium and dog funds in the past year, respectively. We conduct the Wald test for the difference between β1 and β3 and report the p-values in square brackets. Control variables include StdM,[t-4,t-1], Ln(TNAi,t-1), Ln(Agei,t-1, Stdi,[t-4, t-1], Ln(Familysize i,t-1), Front-End Loadi,t-1, and Expense-ratioi,t-1, which are defined in the previous tables. The numbers reported in parentheses are regression t-statistics. ***, **, and * indicate 0.01, 0.05, and 0.1 significance levels, respectively. StdM,[t-4,t-1]
34
Table 6: Test of flow sensitivity to past performance under different market conditions Performance Measure Dependent Var. Independent Var. Ranki,[t-4,t-1]*Upi[t-4,t-1] (β1) Ranki,[t-4,t-1]* Downi[t-4,t-1] (β2) β1 – β2 [Wald P-value]
αCAPM
Raw Return
αFF
Flow(RMB) (1)
Flow(%) (2)
Flow(RMB) (3)
Flow(%) (4)
Flow(RMB) (5)
Flow(%) (6)
1,022.37*** (5.36) 318.27** (2.43) 704.10*** [0.00]
0.82*** (5.14) 0.16 (1.50) 0.66*** [0.00]
1,167.94*** ( 3.49) 613.49** ( 2.51) 554.45*** [0.00]
0.69*** ( 4.90) 0.24** (2.21 ) 0.45*** [0.00]
872.84*** (2.64 ) 376.29 (1.49) 496.55*** [0.00]
0.52*** (3.81 ) 0.10 ( 0.89) 0.42*** [0.00]
-1,080.73*** （-9.16） -590.38*** （-4.85） -1,301.87** （-2.06） 657.93*** （7.81） -3,085.44 （-1.53） -1,22.27 （-0.00） 2,711 0.16
-0.98*** （-6.39） -0.43*** （-3.95） 0.27 （0.65） 0.43*** （5.58） -5.57*** （-2.94） 0.13 （0.00） 2,711 0.20
-1,179.94*** （-7.19） -802.53*** （-3.65） -913.90 （-1.05） 688.25*** （5.90） 822.69 （0.00） 1819.05 （0.00） 1,846 0.16
-0.84*** （-5.77） -0.48*** （-3.33） -0.14 （-0.33） 0.43*** （5.56） 0.48 （0.00） 1.26 （0.00） 1,846 0.20
-1,166.73*** （-7.13） -783.41*** （-3.56） -1,095.19 （-1.25） 681.31*** （5.87） 733.43 （0.00） 1,779.45 （0.00） 1,846 0.15
-0.84*** （-5.73） -0.46*** （-3.25） -0.26 （-0.61） 0.43*** （5.50） 0.42 （0.00） 1.23 （0.00） 1,846 0.19
Control Variables Ln(TNAi,t-1) Ln(Agei,t-1) Stdi,[t-4,t-1] Ln(Familysizei,t-1) Front-end loadi,t-1 Expenseratioi,t-1 Sample size R-square
This table reports the results of a fixed effect regression on net flows measured in 10 millions of RMB (columns 1, 3, and 5) and percentage net flows (columns 2, 4, and 6) using unbalanced panel data. Ranki,[t-4, t-1] is a ranking score (between 0 and 1) based on the fund’s performance over the past year (i.e., from quarters t – 4 to t – 1) measured by raw returns (columns 1 and 2), risk-adjusted returns from the CAPM model, αCAPM (columns 3 and 4), and risk-adjusted returns from the Fama-French three-factor model, αFF (columns 5 and 6). Up (Down) takes a value of 1 if the market return in the past year is greater (equal to or less) than 0, i.e., Rm,[t-4,t-1] >0 (Rm,[t-4,t-1] ≤0), and 0, otherwise. The coefficient β1 on the interaction variable Ranki,[t-4,t-1]* Upi[t-4,t-1] measures flow sensitivity to past performance during bull markets, while the coefficient β2 on the interaction variable Ranki,[t-4,t-1]*Downi[t-4,t-1] measures flow sensitivity to past performance during bear markets. We conduct the Wald test for the difference between β1 and β3 and report the p-values in square brackets. Control variables include the following: Ln(TNAi,t-1) is the log transformation of the total net assets of the ith fund at the end of quarter t – 1, and it measures fund size. Ln(Agei,t-1) is the age of the ith fund at the end of quarter t – 1 measured in number of years. Stdi,[t-4, t-1] is the annualized standard deviation of the ith fund’s monthly returns in the past year. Ln(Familysize i,t-1) is the total net assets under management in the fund family (excluding the total net assets of the ith fund) to which the ith fund belongs at the end of quarter t – 1. Front-End Loadi,t-1 and Expense-ratioi,t-1 are the ith fund’s front-end fee ratio and expense ratio in quarter t – 1, respectively. The numbers reported in parentheses are regression t-statistics. ***, **, and * indicate 0.01, 0.05, and 0.1 significance levels, respectively.
35
Table 7: Test of flow sensitivity to past performance rated by Morningstar and market conditions Performance Measure
αCAPM
Raw Return
αFF
Dependent Var. Independent Var.
Flow(RMB) Flow(%) (1) (2)
Flow(RMB) Flow(%) Flow(RMB) Flow(%) (3) (4) (5) (6)
Ranki,[t-4,t-1]* Stari,[t-4,t-1]* Upi,[t-4,t-1] (β1)
1103.40*** (3.64)
0.91*** (3.36)
1122.64*** (2.73)
0.77*** (3.18)
856.26** (2.07)
0.64*** (2.72)
Ranki,[t-4,t-1]* Stari[t-4,t-1]* Downi,[t-4,t-1] (β2)
221.37 (1.10)
0.37** (2.39)
377.73 (1.16)
0.42*** (3.05)
137.61 (0.40)
0.31** (2.16)
882.03*** [0.00]
0.54*** [0.00]
744.91*** [0.00]
0.35** [0.03]
718.65*** [0.01]
0.33** [0.03]
Ranki,[t-4,t-1]*Mediumi,[t-4,t-1]* Upi,[t-4,t-1] (β3)
993.09*** (4.90)
0.78*** (4.64)
1259.18*** (3.38)
0.68*** (4.93)
942.12** (2.57)
0.50*** (3.68)
Ranki,[t-4,t-1]*Mediumi,[t-4,t-1]* Downi,[t-4,t-1] (β4)
386.46*** (2.86)
0.12 (1.10)
714.98*** (2.91)
0.17 (1.56)
456.79* (1.84)
0.02 (0.21)
606.63*** [0.00]
0.66*** [0.00]
544.20*** [0.00]
0.51*** [0.00]
485.33*** [0.01]
0.48*** [0.00]
Ranki,[t-4,t-1]*Dogi,[t-4,t-1]* Upi,[t-4,t-1] (β5)
277.01 (0.22)
0.54 (1.00)
-236.28 (-0.34)
0.59** (2.30)
-594.94 (-0.86)
0.31 (1.24)
Ranki,[t-4,t-1]*Dogi,[t-4,t-1]* Downi,[t-4,t-1] (β6)
90.02 (0.36)
0.13 (0.27)
419.53 (1.28)
0.53 (0.74)
202.74 (0.55)
0.26 (0.40)
β 5 – β6 [Wald P-value]
186.99 [0.88]
0.41 [0.56]
-655.81 [0.35]
0.06 [0.94]
-797.68 [0.25]
0.05 [0.94]
β 1 – β5 [Wald P-value]
826.39 [0.52]
0.37 [0.52]
1358.92* [0.09]
0.18 [0.55]
1451.20* [0.07]
0.33 [0.27]
β 2 – β6 [Wald P-value]
131.35 [0.66]
0.24 [0.64]
-41.80 [0.91]
-0.11 [0.88]
-65.13 [0.88]
0.05 [0.94]
β1 – β2 [Wald P-value]
β 3 – β4 [Wald P-value]
Control Variables Yes Yes Yes Yes Yes Yes Sample size 2,624 2,624 1,826 1,826 1,826 1,826 R-square 0.17 0.21 0.16 0.20 0.16 0.20 This table reports the results of a fixed effect regression on net fund flows measured in 10 millions of RMB (columns 1, 3, and 5) and percentage net flows (columns 2, 4, and 6) using unbalanced panel data. Ranki,[t-4, t-1] is a ranking score (between 0 and 1) based on the fund’s performance over the past year (i.e., from quarters t – 4 to t – 1) measured by raw returns (columns 1 and 2), risk-adjusted returns from the CAPM model, αCAPM (columns 3 and 4), and risk-adjusted returns from the Fama-French three-factor model, αFF (columns 5 and 6). Star (Dog) takes a value of 1 if the ith fund is rated as a star (dog) fund by Morningstar at the beginning of quarter t based on the performance in the past year and 0, otherwise. Medium take a value of 1 for other funds that are not rated as either a star or a dog in the past year. Up (Down) takes a value of 1 if the market return in the past year is greater (equal to or less) than 0, i.e., Rm,[t-4,t-1] >0 (Rm,[t-4,t-1] ≤0), and 0, otherwise. β1 and β2 measure flow sensitivity to the performance of star funds during bull and bear markets, respectively. Similarly, β3 and β4 measure flow sensitivity to the performance of medium funds during bull and bear markets, respectively; and β5 and β6 measure flow 36
sensitivity to the performance of dog funds during bull and bear markets, respectively. We conduct the Wald test for the difference between β coefficients and report the p-values in square brackets. Control variables include Ln(TNAi,t-1), Ln(Agei,t-1), Stdi,[t-4, t-1], Ln(Familysize i,t-1), Front-End Loadi,t-1, and Expense-ratioi,t-1, which are defined in the previous tables. For brevity, the results on the control variables are omitted. The numbers reported in parentheses are regression t-statistics. ***, **, and * indicate 0.01, 0.05, and 0.1 significance levels, respectively.
37
Table 8: Return volatility and turnover ratio Panel A: Annualized standard deviation of monthly abnormal returns (Ri – Rm) (%) Difference (Top – Bottom) Year Whole sample Top 25% funds Bottom 25% funds (t-stat) 0.99 2004 10.51 11.08 10.09 (0.83) 3.43*** 2005 14.73 16.01 12.58 (4.25) -3.45*** 2006 13.32 13.07 16.52 (-3.41) -2.90*** 2007 17.17 15.95 18.85 (-3.10) 9.29*** 2008 17.49 22.73 13.44 (9.68) -3.66*** 2009 10.20 8.94 12.60 (-4.29) 0.76 Whole period 14.20 15.11 14.35 (1.25) Panel B: Fund annual turnover ratio (%) -21.91 2005 198.63 187.68 209.59 (-0.49) -131.41 2006 339.15 273.44 404.85 (-1.43) -328.54*** 2007 355.18 195.74 524.29 (-4.09) 153.16*** 2008 237.76 316.76 163.60 (3.61) -377.28*** 2009 501.78 313.14 690.42 (-4.82) -163.07*** Entire period 327.56 275.59 438.66 (-4.31) Panel A of this table reports the annualized standard deviation of a fund’s monthly abnormal returns (Ri – Rm) for the past 12 months, where Ri and Rm are the ith fund and the composite market index return in month t, respectively. Panel B reports the fund annual turnover ratio, turnover ratio = min {the fund’s investment in new stocks or sales of existing stock}/fund average net asset in a given year. The numbers reported in parentheses are t-statistics that test differences between top and bottom funds. ***, **, and * indicate 0.01, 0.05, and 0.1 significance levels, respectively.
38
Figure 1a: Shanghai Stock Index 6,500 5,954.8 Jul 2007
Shanghai Stock Index
5,500
4,500
3,500
2,500
1,728.9 Aug 2008
1,500
1,060.7 May 2005
500
Test period
Figure 1b: Shenzhen Stock Index 20,000 19,035 Oct 2007 18,000
16,000
Shenzheng Stock Index
14,000
12,000
10,000
8,000
6,000
4,000
5,839 Oct 2008
2,662 Oct 2005
2,000
Test Period
39
Figure 2a: The flow-performance relationship based on fund net flow 650
550
Fund flows (RMB in millions)
450
350
250
150
50
0
1
2
3
4
5
6
7
8
9
-50
-150 Ranking portfolios based on raw return
-250
Figure 2a: The flow-performance relationship based on percentage net flow 40.0%
35.0%
30.0%
Fund Flow (%)
25.0%
20.0%
15.0%
10.0%
5.0%
0.0%
0
-5.0%
1
2
3
4
5
6
Ranking portfolios based on raw return
40
7
8
9
