How ethical compliance affects portfolio performance and flows: Evidence from mutual funds. Sadok El Ghoul. Aymen Karoui

Transcription

1 How ethical compliance affects portfolio performance and flows: Evidence from mutual funds Sadok El Ghoul University of Alberta, Edmonton, AB T6C 4G9, Canada Aymen Karoui University of Quebec at Montreal, Montreal, Canada Abstract We study the effects of corporate social responsibility (CSR) on fund performance and flows. Using an asset-weighted composite CSR fund score based on firm-level ratings, we find that funds with high CSR scores display poor performance and strong performance reversal. Furthermore, high-csr funds exhibit weaker performance-flow relationships and slightly stronger flow persistence. All these findings are consistent with investors in high- CSR funds deriving utility from non-performance attributes. December 2015 Keywords: Socially responsible investments, ethical or sustainable investments JEL Classification: G11, G14, G23 Electronic copy available at:

2 1. Introduction Whether compliance of mutual funds with ethical standards improves or reduces performance is an important, long-standing question in the mutual fund literature. 1 This debate has further increased in intensity in light of the tremendous growth in socially responsible investment (SRI) funds. According to the Social Investment Forum (SIF), US SRI assets under management registered an increase from 639 $B in 1995 to 3,744 $B in 2012, and an increase in their relative market share from 9% to 11% over the same period. 2 Despite this striking growth, the literature has reported no astonishing return performance for US SRI funds. To examine the impact of compliance with ethical standards on fund performance, the mutual fund literature has adopted a dichotomous approach by comparing a group of funds that fully comply with ethical standards (i.e., SRI funds) to the remaining funds that are called conventional funds. However, in comparing SRI to conventional funds, the literature compares two samples with unequal sizes and dissimilar characteristics. It also ignores the amount of heterogeneity within each category (e.g., Barnett and Salomon, 2006). To overcome these issues, the current paper suggests a simple holdings-based measure to assess the level of corporate social responsibility (CSR) of a fund. The measure is a value-weighted score using firm-level ratings of portfolio holdings. Armed with this tool, we are able to examine the impact of ethical compliance on performance on a continuum basis, without resorting to the dichotomous categorization into SRI and conventional funds. Consequently, we are able to compare funds, based on the CSR criterion, cross-sectionally and across time. The measure can be used as an additional attribute by mutual fund investors. 1 Please refer to Renneboog, Ter Horst and Zhang (2008a) for a comprehensive review of the subject and a discussion on ethical investments. 2 Please refer to the report on Sustainable and responsible investing trends in the United States, 2012, (SIF, 2012, p. 11). 1 Electronic copy available at:

3 In examining the relationship between ethical compliance and fund performance, two hypotheses compete: On the one hand, investing in firms that comply with social responsibility practices is likely to reduce the set of investment opportunities (Geczy, Stambaugh and Levin, 2005; Renneboog, Ter Horst and Zhang, 2008b; Cortez, Silva and Areal, 2009) and to increase monitoring costs (Bauer, Koedijk and Otten, 2005). Ethical compliance would then negatively impact financial performance. On the other hand, fund managers that target socially responsible firms may in fact target firms with solid financial fundamentals, which in turn would translate into higher performance. In this case, investing in socially responsible stocks would be a valuegenerating strategy. At the empirical level, the literature has nonetheless reported mixed evidence as to the existence of a significant difference in performance between the SRI and conventional funds in the US market. For instance, Bauer, Derwall and Otten (2007) find that domestic ethical US funds significantly underperform conventional funds, while Gil-Bazo, Ruiz-Verdú and Santos (2010) report an opposite finding. Nofsinger and Varma (2014) find that SRI funds outperform their conventional peers during times of crisis and underperform at other times. However, most of the studies on US mutual funds find no statistical difference in performance between SRI and conventional funds (e.g., Hamilton, Joe and Statman, 1993; Goldreyer and Diltz, 1999; Statman, 2000; Schroder, 2004; Geczy, Stambaugh and Levin, 2005; Renneboog, Ter Horst and Zhang, 2008b). We contribute to this debate on the impact of ethical compliance on fund performance. However, instead of comparing SRI to conventional funds in a dichotomous way, we suggest a finer fund-level analysis using portfolio holdings, as defended by Borgers, Derwall, Koedijk and Ter Horst (2015). We are thus able to shed light on the potential interaction between CSR and fund performance without resorting to potentially biased matching procedures and conflicting SRI screening processes. 2

4 Using a sample of 2,168 US equity funds over the period , we examine the impact of CSR on the following fund-related issues: (i) performance, (ii) performance persistence, (iii) the flow-performance relationship and (iv) flow persistence. We first provide strong evidence that an increase in the level of CSR comes at the expense of a reduction in performance. Indeed, our empirical results reveal that the CSR level of the portfolio is negatively related to risk-adjusted performance. This result stands even when we control for common fund characteristics such as volatility, flows, the size of the assets under management, the number of stocks, the expense ratio and turnover. Furthermore, we find that the CSR level negatively predicts the next year s fund performance. These findings complement those of Borgers, Derwall, Koedijk and Ter Horst (2015), who find that funds that are more invested in sin stocks display higher risk-adjusted performance. The predictability of risk-adjusted performance shows no noticeable differences across quartiles when we sort funds based on their CSR score. The one-year-lagged CSR coefficient remains negative for all of the quartiles, albeit with various degrees of significance. Second, we examine the performance persistence of funds. Over the entire sample, we find a negative one-year-lagged performance coefficient, attesting to the presence of performance reversal. When we sort funds into quartiles based on their CSR level, we provide clear evidence that the risk-adjusted performance reversal increases with the level of CSR. We then examine the flow-performance relationship issue. Evidence in the mutual literature has shown that investor flows respond positively and significantly to past performance (e.g., Chevalier and Elisson, 1997; Sirri and Tufano, 1998). Our empirical tests confirm the positive relationship for the entire sample. Moreover, we find this relationship to weaken as the level of CSR increases. Therefore, we confirm that investors in funds with higher ethical standards become less responsive to past performance and derive their utility from nonfinancial attributes. This result is in line with Benson and Humphrey (2008) and Renneboog, 3

5 Ter Horst and Zhang (2011), who find that SRI fund flows are less sensitive to past performance than those of their conventional counterparts. Finally, we look at the persistence of fund flows. Anecdotal evidence suggests that socially conscious investors may find fewer suitable investment opportunities than performance-chasing investors. Therefore, socially conscious investors may be less likely to switch from one fund to another. Our empirical results do not support this hypothesis. The sorts based on the CSR level reveal no evidence that high-csr funds exhibit stronger flow persistence than low-csr funds. This finding is further corroborated by the insignificant relationship between the CSR score and the standard deviation of fund flows. The remainder of the paper is organized as follows. Section 2 presents the literature review on the performance and flows of SRI and conventional funds. In Section 3, we describe the sample selection and the variables used in the study. Section 4 presents the portfolio's CSR score and its determinants. Section 5 highlights the relationship between the CSR score, performance and flows. We conclude in Section Related literature 1. Performance of SRI funds and conventional funds Two conflicting arguments have been proposed by the mutual fund literature to explain the relationship between performance and ethical compliance. On the one hand, investing in SRI funds might be costly as the set of investment opportunities is smaller for this category of funds. On the other hand, funds using an SRI screening process may in fact target firms with a more sustainable profitability and better long-run perspectives. Thus, the first argument aligns with SRI funds underperforming conventional funds, while the second argument advocates the opposite result. 4

6 To date, most of the studies on US mutual funds have reported no significant differences in performance between SRI and conventional funds. Hamilton, Jo and Statman (1993), Goldreyer and Diltz (1999), Statman (2000), Bello (2005), Bauer, Koedijk and Otten (2005) and Renneboog, Ter Horst and Zhang (2008b) find no evidence of a performance difference between SRI and conventional funds. However, Geczy, Stambaugh and Levin (2005) support that, because of the investments costs, SRI are likely to perform more poorly than their conventional peers. One of the main reasons for these conflicting results pertains to the use of different types of social screening for funds and the comparison of fund groups with different characteristics. The use of a fund-level measure of ethical compliance alleviates these two issues. Another strand of the literature has relied on hypothetical portfolios to compare the performance of high- and low-csr funds (e.g., Kempf and Osthoff, 2007). While such an approach is appealing because of its flexibility for testing a number of strategies and scenarios, it falls short of using real managers trades. The use of actual portfolios, however, perfectly depicts real conditions, as highlighted by Borgers, Derwall, Koedijk and Ter Horst (2015). 2. Flow characteristics of socially responsible funds and conventional funds The mutual fund literature has extensively examined the flow-performance relationship, which has proven to be largely significant for actively managed funds (e.g., Chevalier and Elisson, 1997; Sirri and Tufano, 1998). Focusing on the magnitude of the flow-performance relationship, the literature on SRI funds has compared them to conventional funds. On the one hand, as defended by Bollen (2007), SRI funds suffer from having a relatively short history and therefore being perceived as more uncertain by common investors. Hence, these investors are more worried about and sensitive to performance variations. In addition to this, SRI investors may display conditional utility in which they invest in SRI funds conditional on good performance. On the other hand, SRI investors may derive their utility from non-financial 5

7 attributes and as such should be less sensitive to past performance. Bollen (2007) finds that SRI fund flows display a lower volatility, and that, compared to conventional funds, SRI funds are more sensitive to lagged positive returns but less sensitive to lagged negative returns. Benson and Humphrey (2008) and Renneboog, Ter Horst and Zhang (2011) find, however, that the fund flows of US SRI funds are less sensitive to performance than are those of their conventional counterparts. A second issue that has attracted the attention of the SRI fund literature is the persistence of flows. Benson and Humphrey (2008) find that SRI flows exhibit stronger persistence than those of conventional funds and thus SRI investors are more reluctant to transfer their investments. This finding is consistent with SRI investors struggling to find investment alternatives and being likely to carry out smoother flow movements. 3. Data and variable construction We use the CRSP mutual fund database and the KLD database to construct a sample of mutual fund ratings. The CRSP mutual fund database includes mutual fund characteristics: expense ratios, turnover ratios, total net assets, returns and holdings. These fund characteristics are aggregated at the portfolio level using asset-weighted averages of share classes. We rely on the Lipper-style classifications to select actively managed equity funds. The KLD database includes ratings of individual companies based on their social and environmental performance. The latter are evaluated along seven dimensions: community, corporate governance, diversity, employee relations, environment, human rights, and product quality and safety. The KLD database has registered an important increase in the number of companies covered since 2003 to reach more than 3,000 (per year). In comparison, the coverage was limited to only 650 companies (per year) in the 1990s. This fact has guided the choice of start year for our sample. 6

8 KLD assigns a binary (0/1) rating to a set of concerns and strengths indicators in each dimension. One can obtain a score for each dimension equal to the sum of strengths minus the sum of concerns, then obtain a raw overall CSR score equal to the sum of scores across all dimensions. However, this approach is complicated by the fact that KLD has been adding and removing strengths and concerns indicators over time. To make the overall CSR score comparable over time we follow the procedure in Deng, Kang and Low (2013). Specifically, each year we deflate each strength (concern) score by the number of strengths (concerns) in that year. The overall CSR score based on the adjusted strength and concern scores is comparable over time because it does not depend on the changing number of strengths and concerns from year to year. To construct a CSR score at the portfolio level, we match the CRSP fund holdings with the stock CSR scores from the KLD database based on the permanent number (permno). However, given that mutual funds have invested in 21,275 distinct stocks and the KLD database provides information on just 5,296 distinct stocks, a significant portion of stocks are left unrated. The average sum of stock weights of fund portfolios covered by the KLD database is indeed equal to 80.68%. To alleviate this issue, we follow Cremers and Petajisto (2009) and require that the sum of equity weights with a CSR rating accounts for at least 67% of the portfolio. 3 Doing so yields an average sum of stock weights equal to 87.66% for the selected fund portfolios. We then adjust these weights by dividing them by the sum of the weights, to make the weights sum to one. 4 Moreover, we require that funds have at least two years of reported holdings. The final sample comprises 2,168 US equity domestic funds and covers the period Similarly, Borgers, Derwall, Koedijk and Ter Horst (2015) required that at least 75% of the portfolio holdings match the stocks of the MSCI STATS ESG database. 4 The Pearson correlation coefficient between the adjusted and unadjusted weights is, however, equal to 99.55%. 7

9 To get fund-style dummies, we rank funds according to their self-reported investment objectives into one of the following seven style categories: aggressive growth, mid cap, small cap, growth and income, growth, equity income and maximum capital gains. In our sample, we also identify SRI funds using a list of funds collected from the SIF website. 3.1 The CSR score at the portfolio level We follow a widely used approach that relies on matching fund holdings with individual stock measures. This approach has been used, for example, to construct portfolio measures of performance (Daniel, Grinblatt, Titman and Wermers, 1997; Kacperczyk, Sialm and Zheng, 2008), systematic risk (Chevalier and Ellison, 1997), risk-shifting (Schwarz, 2012), liquidity (Idzorek, Xiong and Ibbotson, 2012) and exposure to socially sensitive stocks (Borgers, Derwall, Koedijk and Ter Horst, 2015). We follow the same approach to construct a CSR score at the portfolio level and at a yearly frequency using the following equation: CSR, ω, CSR, (1) where, is the weight of stock i at yearend t in fund j, Nj is the number of stocks held by fund j, and CSRi,t is the corporate social responsibility score at the firm level, computed using the same methodology as in Deng, Kang and Low (2013). 3.2 Other key variables Fund performance: In addition to considering fund raw returns, we compute the riskadjusted performance using daily returns and Carhart's (1997) model as follows: R, R, r r r r r ε,, (2) where, is fund j s net return on day t and Rf,t is the T-bill rate on day t. r r is the excess return of the equity market index over the one-month T-bill rate, r is the difference in returns between a small-cap portfolio and a large-cap portfolio, and r is the difference in returns between a portfolio with high book-to-market stocks and a portfolio with low book-to- 8

10 market stocks. r captures the one-year momentum factor. The parameter denotes the riskadjusted performance for fund j. R-square: Using equation (2), we can save the R-square and consider it a proxy for the weight of idiosyncratic risk relative to systematic risk. The R-square should gauge the selectivity and active management of the fund, according to Amihud and Goyenko (2013). Fund flows: Annual fund flows are measured as Flows,,,,,, (3) where TNAj,t and TNAj,t-1 are the total net assets for fund j in years t and t-1, respectively, and Rj,t is the return of the j th fund in year t. 3.3 Descriptive statistics Panel A of Table 1 reports descriptive statistics for our set of variables. The average fund CSR is equal to with a standard deviation equal to The average raw return of the sample is positive and equal to 0.074, while the annualized risk-adjusted return (net of fees) is negative and equal to Fund flows are on average positive (0.147), reflecting thereby substantial growth for some equity mutual funds over the considered sample period, while the mean flow standard deviation is equal to ****Insert Table 1 about here **** The last column of Panel A of Table 1 reports the correlation coefficients between the CSR variable and fund characteristics. The CSR shows significant correlations with all of the fund characteristics except for flows and flow standard deviations. The CSR variable displays a negative correlation with the two measures of performance, raw returns and alphas. These correlation coefficients are equal to and -0.06, respectively, and therefore support that an increase in ethical compliance comes at the cost of a reduction in financial performance. Return volatility is negatively correlated with the CSR level, implying that a better rating goes hand in 9

11 hand with a smaller risk. Investor fund flows respond negatively but insignificantly to an increase in the CSR level. The R-square is positively linked to CSR. Therefore funds with a high level of CSR are less active then funds with a low level of CSR, but invest in fewer stocks. The total net assets are positively related to CSR, while the expense ratio and turnover are negatively related to the CSR level. 4. The portfolio's corporate social responsibility (CSR) score 1. The CSR score and fund characteristics: a univariate analysis We sort funds each year into two groups based on whether they have an above- (i.e., high) or a below-median (i.e., low) CSR score. Then, we compute the average fund characteristics for the high- and low-csr funds and estimate the mean difference between the two groups. Table 2 reports the univariate results. The difference in performance between the high- and low-csr funds is negative when considering both raw returns and alphas. A high CSR score goes hand in hand with poorer performance. Furthermore, high-csr funds attract larger flows and exhibit lower volatility, a lower R-square, more assets under management, a smaller number of stocks, a smaller expense ratio and turnover ratio, and are younger. The lower R-square and smaller number of stocks for high-csr funds are consistent with the idea that these funds hold less diversified portfolios. Tests relying on the Mann-Whitney median statistics are overall consistent with those for the mean differences, except in the case of the turnover ratio, for which the difference turns out to be negative and significant. Overall, these results confirm the correlation results on the entire panel, except for the R-square. ****Insert Table 2 about here **** 2. The CSR score and fund characteristics: a multivariate analysis We examine the various fund characteristics that affect the level of ethical compliance as measured by the CSR level of the fund portfolio. We use the following panel regression model: 10

12 CSR j,t a 0 + a 1 return j,t + a 2 alpha j,t + a 3 flows j,t + a 4 std. flows j,t + a 5 volatility j,t + a 6 R-square j,t + a 7 TNA j,t + a 8 number of stocks j,t + a 9 expense ratio j,t + a 10 turnover j,t +a 11 fund age j,t +a 12 SRI dummy j,t + dummy year variables j,t + dummy style variables j,t + ε j,t (4) where CSRj,t is the corporate social responsibility score of the fund computed as in equation (1), returnj,t is the net return of fund j in year t, alphaj,t is the risk-adjusted return of fund j in year t and is computed as in equation (2) using daily returns, flowsj,t are the annual flows of fund j in year t and are computed as in equation (3), std. flowsj,t is the flow standard deviation in year t and is computed using monthly flow observations as in Bollen (2007), volatility is the annualized return volatility, R-square is computed using equation (2), TNA is the total net assets of the fund in millions of $, number of stocks refers to the number included in the portfolio, expense ratio is the annual fund fees, turnover is the annual fund turnover, fund age is the age of the fund in months since inception, and SRI dummy equals 1 if the fund is an SRI fund and 0 otherwise. The regression of equation (4) includes time (year) and style dummies, and standard deviations are clustered at the fund level. Table 3 shows that fund characteristics explain to a great extent the level of social responsibility of a fund. Fund returns load insignificantly on portfolio CSR. The risk-adjusted return loads negatively and significantly on fund CSR. The coefficient is equal to , suggesting that an increase of 10% in the fund risk-adjusted return induces a decrease of 5.78% in the portfolio CSR level, which is economically significant. High fund flows seem to be associated with a high level of CSR, though the statistical strength of this relationship is extremely weak. The flow standard deviation correlates negatively with CSR, albeit insignificantly. This is consistent with the results of Bollen (2007), who shows that SRI fund flows display a lower volatility than conventional funds. The last specification includes all fund 11

13 characteristics. Confirming the univariate analysis, we find that a high level of CSR is negatively related to the return volatility, the R-square, the number of stocks and the expense ratio, and is on the other hand positively related to fund age. Moreover, the SRI dummy loads positively on CSR, thereby confirming that the CSR score is higher for SRI funds. ****Insert Table 3 about here **** 3. The CSR score and Carhart s factor loadings Beyond the performance-csr relationship, it is worthwhile inspecting how the CSR measure relates to fund style exposures as measured by Carhart s (1997) factor loadings. To do this, we first compute Carhart s factor loadings using equation (2) at an annual frequency and then run a panel regression in which CSR is the dependent variable and Carhart s factor loadings are the main independent variables. We further include several control variables as in equation (4). Table 4 reports the regression results. The CSR level displays a significant relationship with all of the coefficients. In particular, a high level of CSR goes hand in hand with investment in small-beta stocks, large stocks, growth stocks and contrarian stocks. When considered jointly as in specification (6), all of the coefficients maintain their signs and their statistical significance. These results corroborate those of Bollen (2007) on the market, SMB, and UMD factors for which the author finds a positive average difference between the conventional and SRI funds. Only for the HML coefficient does Bollen (2007) find a negative difference. Furthermore, the significant relationship between the CSR score and the style factor loadings further justifies the use of risk-adjusted returns instead of raw returns as a measure of performance in our study. ****Insert Table 4 about here **** 5. Fund performance, fund flows and corporate social responsibility 1. Does a portfolio s CSR score predict fund performance? 12

14 Renneboog, Ter Horst and Zhang (2008b) posit that socially responsible funds invest within a smaller universe of stocks and this is likely to harm the performance of this category of funds. On the other hand, the multiple screening steps taken by these fund managers may well eliminate poorly managed companies with underperforming stocks. The latter argument aligns with SRI funds outperforming conventional funds. Here, we extend the argument over the entire sample of funds and verify whether a high CSR score is likely to be a value-generating or a value-destroying factor in the portfolio over the next year. To test such a hypothesis, we use the following regression model: alpha j,t+1 = a + a CSR j,t +a CSR j,t-1 CSR 0 j,t Q 1 alpha j,t a flows, + volatility j,t + a 10 R-square j,t + a 11 TNA j,t + a 12 number of stocks j,t +a 13 expense ratio j,t +a turnover + 15 fund age +a 16 SRI dummy j,t + dummy year variables j,t + dummy style variables j,t + ε j,t (5) where CSRj,t and CSRj,t-1 are the corporate social responsibility scores of the fund in years t+1 and t and are computed as in equation (1). alphaj,t+1 and alphaj,t are the risk-adjusted returns of fund j in years t+1 and t and are computed as in equation (2) and using daily fund returns. Qi is a dummy variable for each of the four quartiles of funds sorted on the CSR level. The flows, volatility, R-square, TNA, number of stocks, expense ratio, turnover ratio, fund age and SRI dummy are as defined in equation (4). Our estimations include two-year lags of the CSR variable in the first and last columns of Table 5. The results in the first column show that the one-year and the two-year-lagged CSR negatively predict the risk-adjusted performance. However, only the one-year-lagged CSR coefficient is significant. The latter remains significant even when we control for other fund characteristics as displayed in the last column of Table 5. We further examine the performance- 13

15 CSR relationship for each of the fund quartiles sorted by the level of CSR. The relationship is negative for all four quartiles, although the degree of significance varies. Among the control variables, flow is positive and significant. This hints to the existence of a smart money effect whereby investors are able to direct their investments on an ex-ante basis towards high-performing funds. Moreover, Table 5 shows that R-square, TNA, the expense ratio and the turnover ratio all relate negatively to the next year s performance. ****Insert Table 5 about here **** A natural extension to the performance predictability question is to examine performance persistence over two consecutive years. We next look at how performance persistence varies across CSR-sorted fund quartiles. To do this, we regress the risk-adjusted returns on the one-year-lagged risk-adjusted returns interacted with the CSR quartile dummies as defined in equation (5) and the one-year-lagged control variables. We use the following regression model: alpha j,t+1 = a + a alpha 0 j,t + alpha j,t Q 1 + volatility j,t + a 9 R-square j,t + a 10 TNA j,t + a 11 number of stocks j,t + a flows, a 11 expense ratio j,t + a turnover, + fund age j,t +a 14 SRI dummy j,t + dummy year variables j,t + dummy style variables j,t + ε j,t (6) where alphaj,t+1 and alphaj,t are the risk-adjusted returns of fund j in years t+1 and t and are computed as in equation (2) and using daily fund returns. Qi is a dummy variable for each of the four quartiles of funds sorted based on the CSR level. The flows, volatility, R-square, TNA, number of stocks, expense ratio, turnover ratio, fund age and SRI dummy are as defined in equation (4). 14

16 For the entire sample, we find no evidence of performance persistence over a horizon of one year, as the lagged alpha coefficient is negative. As it is significant, the coefficient hints rather at the presence of performance reversal. The sort based on the CSR level reveals a laggedalpha coefficient that is negative and increasingly significant across the CSR quartiles. Thus, funds with a high CSR score exhibit weaker persistence or higher reversal in their performance, a fact that, combined with the relative poor performance found previously, will make this category of funds less desirable to performance-chasing investors. ****Insert Table 6 about here **** 2. The performance-flow relationship and the CSR level The performance-flow relationship is a well-studied relationship in the mutual funds literature. Several papers have reported strong evidence that investor flows respond positively to performance and are more sensitive to good than to poor performance. We examine here the performance-flow relationship and investigate how it is affected by the CSR level. Benson and Humphrey (2008) and Renneboog, Ter Horst and Zhang (2011) find that SRI fund flows are less sensitive to performance than their conventional counterparts. Thus, as funds invest in more socially responsible stocks, they should attract investors that are less focused on performance. However, Bollen (2007) finds that, compared to conventional funds, SRI fund flows are more sensitive to lagged positive returns and less sensitive to lagged negative returns. Bollen (2007) argues that investors face more uncertainty in SRI funds due to their short history, and that investors are therefore more sensitive to performance variations. Our intuition, however, is that funds that invest in more socially responsible stocks will repel performance-conscious investors and attract socially conscious investors. Thus, as a fund invests more in ethical stocks, the investor s sensitivity to the fund s performance should weaken. As we find that the CSR score is negatively related to performance, funds with a high 15

17 CSR level should attract investors that are less sensitive to the performance criterion. To test this hypothesis, we use the following regression model: flows j,t+1 = a 0 + alpha j,t alpha j,t-1 alpha j,t Q a flows, a volatility j,t + a 9 R-square j,t + a 10 TNA j,t + a 11 number of stocks j,t +a 12 expense ratio j,t +a turnover, + 14 fund age, + a 15 SRI dummy j,t + dummy year variables j,t + dummy style variables j,t + ε j,t (7) where CSRj,t is the corporate social responsibility score of the fund, computed as in equation (1), alphaj,t+1 alphaj,t and alphaj,t-1 are the risk-adjusted returns of fund j in years t+1, t and t-1 and are computed as in equation (2) and using daily fund returns. Qi is a dummy variable for the four quartiles of funds sorted based on the CSR level. The flows, volatility, R-square, TNA, number of stocks, expense ratio, turnover ratio, fund age and SRI dummy are as defined in equation (4). Table 7 reports the estimation results of equation (7). The first column of Table 7 shows that fund flows respond positively and significantly to the one-year and two-year-lagged riskadjusted performances. Furthermore, we find that, as the level of CSR increases, the flowperformance relationship weakens. Hence, for more ethical funds, investors are less sensitive to the performance attribute. This result is in line with those of Benson and Humphrey (2008) and Renneboog, Ter Horst and Zhang (2011). As the level of ethical compliance increases, it becomes more difficult for investors to find similar investment alternatives and therefore they may be more reluctant to switch to other funds, even when these funds register a poor performance. 5 5 In unreported results, we also estimated the flow-performance relationship using raw returns. We confirmed the positive relationship between flows and past returns, and the same trend was observed across CSR quartiles: as the level of CSR increased, the flow-return relationship weakened. For brevity, we do not report these results. 16

18 Looking at the other control variables, we find that funds exhibit persistence in their annual flows, as evidenced by the significant lagged-flow coefficient. Moreover, we find that TNA is negatively related to fund flows, which is consistent with the diseconomies of scale rationale. As funds become more mature, they find it more difficult to maintain a sustained growth rate. This is further corroborated by the negative relationship between fund age and flows, a fact that is in line with Chevalier and Ellison s (1997) results. Moreover, high-fee funds seem to be avoided by investors, as fees load negatively and significantly on fund flows. ****Insert Table 7 about here **** In unreported results, we also examined the asymmetry of the flow-performance relationship. Each year, we sorted funds based on their alphas into high and low groups and then estimated equation (7) while further interacting the alpha variable with an alpha dummy. Our results confirmed the positive relationship between performance and flows. We also found this relationship to be asymmetric as it was stronger for the higher-performance group. The finding corroborated Chevalier and Ellison (1997). However, across CSR-sorted quartiles, we found no major differences between the high- and low-performance groups. Both the high- and low-alpha groups displayed a decreasing trend as the CSR level increased. These results did not differ dramatically from that found in Table 7 for the entire sample and confirmed those of Benson and Humphrey (2008), who found no noticeable differences between the reactions of SRI and conventional fund flows to positive and negative returns, respectively The flow persistence and the CSR level We repeat the previous analysis by regressing fund flows on lagged versions of the explanatory variables, but we put the focus on the lagged flow variable in order to study the flow persistence question. Persistence of fund flows on an annual basis has been demonstrated 6 For brevity we do not display these results. They are available upon request. 17

19 by Patel, Zeckhauser and Hendricks (1994), Fant and O'Neal (2000) and Del Guercio and Tkac (2002). We contribute to this strand of the literature by examining whether flow persistence varies with the level of ethical compliance as measured by the CSR score. Persistence in fund flows shows the ability of funds to maintain sustainable flows across years. A strong persistence in flows indicates that a fund has a stable panel of investors that are loyal, whereas a weaker persistence may signal difficulty in preventing investors from diverting away. Our intuition is that, if a fund is more compliant with social responsibility principles, investors should be less concerned with other factors such as performance or risk and therefore more loyal to the fund. A positive relationship between the CSR level and flow persistence is therefore expected. To test this relationship, we regress fund flows on the one-year-lagged flows and the various control variables. The subsequent specifications include interaction variables between the flow variable and the dummy quartile variables of funds sorted based on their CSR level. The results of these estimations are reported in Table 8. The first and last columns of Table 8 show that fund flows exhibit persistence that is significant even when we control for common fund characteristics. We then examine the persistence while sorting funds based on their CSR level. We find it to be marginally stronger for the bottom two CSR funds than for the top two. Hence, funds with less ethical policies seem to have less time-varying flows. This result is also consistent with the weak evidence of a negative relationship between the flow standard deviation and the CSR level, found previously in the univariate section. ****Insert Table 8 about here **** 6. Concluding comments The current paper examines the impact of ethical compliance on fund performance and flows. To gauge such impacts, the literature has so far relied on a comparison between SRI and 18

20 conventional funds. While this approach has the advantage of comparing two distinct fund categories, it comes at the cost of ignoring the amount of heterogeneity within each one. Moreover, the fund categorization itself is still subject to debate as various screening processes exist and may sometimes provide conflicting fund classifications. Suggesting a continuum approach across all funds, the current paper offers a simple holdings-based measure to assess the level of CSR of the fund. The measure is a value-weighted score using firm-level ratings of portfolio holdings. We then test how the CSR level impacts fund performance. Our empirical estimations show strong evidence that the CSR score significantly and negatively relates to fund performance. We further find that the CSR score negatively predicts the next year s fund performance. This relationship shows no significant variations across quartiles when we sort funds based on their CSR level. We next examine the performance persistence and find evidence of performance reversal over the entire sample. The sorting of funds into CSR-based quartiles reveals that high-csr funds exhibit a more negative lagged performance coefficient, thereby confirming that these funds exhibit weaker persistence or stronger reversal in their performance. With relatively poor performance that is weakly persistent, high-csr funds may struggle to attract performance-chasing investors. Looking at the other determinants of the CSR score, we find that the CSR level of the portfolio is negatively related to fund flows, return volatility, R-square, the number of stocks and the expense ratio, and is positively related to fund age. Funds that invest in more socially responsible firms also display smaller market betas, invest in firms with large capitalization, in firms with low book-to-market ratios, and are contrarian. Finally, consistent with the idea that an increase in the CSR level of a portfolio attracts social investors and repels those that are performance-chasing, we find that the flowperformance relationship becomes weaker as the level of CSR increases. Therefore, high-csr 19

21 funds are likely to attract investors that are less sensitive to the performance attribute. We also examine the persistence of fund flows for funds sorted based on their CSR level. We find evidence that flows are positively persistent across all quartiles of sorted funds, and thus the level of ethical compliance seems not to dramatically affect the loyalty of fund investors. 20

22 References Amihud, Y., Goyenko, R., Mutual fund s R2 as predictor of performance. Review of Financial Studies 26, Barnett, M.L., Salomon, R.M., Beyond Dichotomy: The Curvilinear Relationship between Social Responsibility and Financial Performance. Strategic Management Journal 27, Bauer, R., Derwal, J., Otten, R., The ethical mutual funds performance debate: New evidence for Canada. Journal of Business Ethics 70, Bauer, R., Koedijk, K., Otten, R., International evidence on ethical mutual fund performance and investment style. Journal of Banking & Finance 29, Bello, Z.Y., Socially Responsible Investing and Portfolio Diversification. Journal of Financial Research 28, Benson, K.L., Humphrey, J.E., Socially responsible investment funds: Investor reaction to current and past returns. Journal of Banking & Finance 32, Bollen, N.P.B., Mutual fund attributes and investor behavior. Journal of Financial and Quantitative Analysis 42, Borgers, A., Derwall, J., Koedijk, K., Ter Horst, J., Do social factors influence investment behavior and performance? Evidence from mutual fund holdings. Journal of Banking & Finance 60, Carhart, M.M., On Persistence in Mutual Fund Performance. Journal of Finance 52, Chevalier, J., Ellison, G., Risk taking by mutual funds as a response to incentives. Journal of Political Economy 105, Cortez, M., Silva, F., Areal, N., The Performance of European Socially Responsible Funds. Journal of Business Ethics 87,

23 Cremers, K.J.M., Petajisto, A., How active is your fund manager? A new measure that predicts performance. Review of Financial Studies 22, Daniel, K., Grinblatt, M., Titman, S., Wermers, R., Measuring Mutual Fund Performance with Characteristic-Based Benchmarks. The Journal of Finance 52, Del Guercio, D.D., Tkac, P.A., The Determinants of the Flow of Funds of Managed Portfolios: Mutual Funds vs. Pension Funds. Journal of Financial and Quantitative Analysis 37, Deng, X., Kang, J.-K., Low, B.S., Corporate social responsibility and stakeholder value maximization: Evidence from mergers. Journal of Financial Economics 110, Fant, L.F., O'Neal, E.S., Temporal Changes in the Determinants of Mutual Fund Flows. Journal of Financial Research 23, Geczy, C., Stambaugh, R.F., Levin, D., Investing in Socially Responsible Mutual Funds. University of Pennsylvania, working paper. Gil-Bazo, J., Ruiz-Verdú, P., Santos, A.P., The Performance of Socially Responsible Mutual Funds: The Role of Fees and Management Companies. Journal of Business Ethics 94, Goldreyer, E.F., Diltz, J.D., The performance of socially responsible mutual funds: incorporating sociopolitical information in portfolio selection. Managerial Finance 25, Hamilton, S., Jo, H., Statman, M., Doing Well while Doing Good? The Investment Performance of Socially Responsible Mutual Funds. Financial Analysts Journal 49, Idzorek, T.M., Xiong, J.X., Ibbotson, R.G., The Liquidity Style of Mutual Funds. Financial Analysts Journal 68,

24 Kacperczyk, M., Sialm, C., Zheng, L., Unobserved Actions of Mutual Funds. Review of Financial Studies 21, Kempf, A., Osthoff, P., The Effect of Socially Responsible Investing on Portfolio Performance. European Financial Management 13, Nofsinger, J., Varma, A., Socially responsible funds and market crises. Journal of Banking & Finance 48, Patel, J., Zeckhauser, R.J., Hendricks, D., Investment flows and performance: Evidence from mutual funds, cross-border investments, and new issues. In Sato, R., Levitch, R., Y Ramachandran, R. (Ed.): Japan, Europe and the International Financial Markets: Analytical and Empirical Perspectives, New York: Cambridge University Press. Renneboog, L., Ter Horst, J., Zhang, C., 2008a. The price of ethics and stakeholder governance: The performance of socially responsible mutual funds. Journal of Corporate Finance 14, Renneboog, L., Ter Horst, J., Zhang, C., 2008b. Socially responsible investments: Institutional aspects, performance, and investor behavior. Journal of Banking & Finance 32, Renneboog, L., Ter Horst, J., Zhang, C., Is ethical money financially smart? Nonfinancial attributes and money flows of socially responsible investment funds. Journal of Financial Intermediation 20, Schroder, M., The performance of socially responsible investments: Investment funds and indices. Swiss Society for Financial Market Research 18, Schwarz, C.G., Mutual Fund Tournaments: The Sorting Bias and New Evidence. Review of Financial Studies 25, Sirri, E.R., Tufano, P., Costly Search and Mutual Fund Flows. The Journal of Finance 53,

25 Social Investment Forum, SIF, Report on responsible investing trends in the US. Statman, M., Socially responsible mutual funds. Financial Analysts Journal 56,

26 Table 1 Descriptive statistics Table 1 presents descriptive statistics for our sample of equity mutual funds over the period. In particular, it reports the number of observations (N), mean, median, standard deviation, minimum and maximum of the CSR score, fund raw return, volatility, fund riskadjusted performance (alpha), R-square, flows, flow standard deviation, total net assets (TNA) in millions of $, number of stocks, expense ratio, turnover ratio, fund age and SRI dummy, which equals 1 for SRI funds and 0 otherwise. The last column reports the Pearson's coefficient of correlation of our main variable (CSR score) with each of the aforementioned variables. ***, ** and * indicate significance at the 1%, 5% and 10% levels, respectively. Variable N Mean Median Standard Min Max Corr deviation CSR Return Volatility *** Alpha *** R-square *** Flows Std flows TNA (in $M) *** Number of stocks *** Expense ratio *** Turnover *** Fund age *** SRI dummy *** 25

27 Table 2 Corporate social responsibility (CSR) score and fund characteristics: A univariate analysis Table 2 reports the univariate results on fund characteristics. Each year, we sort funds based on their corporate social responsibility (CSR) score and categorize funds into high- or low-csr groups according to whether they have an above-median or below-median CSR figure. We next compute the average for each characteristic and for each of the two groups (i.e., high and low). The fund characteristics are the fund raw return, volatility, fund risk-adjusted performance (alpha), R-square, flows, standard deviation of flows, total net assets (TNA) in millions of $, number of stocks, expense ratio, turnover ratio and fund age. The third line reports the mean difference in characteristics between the high- and low-csr funds. The T-statistic for the difference in means and the Mann-Whitney Z-statistic for the difference in medians are also reported. ***, ** and * indicate significance at the 1%, 5% and 10% levels, respectively. Return Volatility Alpha R-square Flows Std flows TNA (in $M) Number of stocks Expense ratio Turnover Fund age Low-CSR funds High-CSR funds High Low *** *** *** *** *** *** T-statistic (-0.53) (-6.15) (-5.81) (-4.18) (0.51) (-0.28) (1.04) (-7.14) (-2.65) (0.05) (2.66) Z-statistic (-2.77) (-5.57) (-7.07) (-3.22) (-0.54) (-1.44) (-1.23) (-11.24) (-3.61) (-1.82) (2.07) 26

28 Table 3 Corporate social responsibility (CSR) and fund characteristics: a multivariate analysis Table 3 reports the regression results of the corporate social responsibility (CSR) score on the following fund characteristics: fund raw return, volatility, fund risk-adjusted performance (alpha), R-square, flows, standard deviation of flows, the natural logarithm of the total net assets (TNA), the natural logarithm of the number of stocks, the expense ratio, the turnover ratio, fund age and the SRI dummy, which equals 1 for SRI funds and 0 otherwise. All the variables used are winsorized at the 0.5% and 99.5% levels in all of the regression specifications. The regressions include time (year) and style dummies, and standard deviations are clustered at the fund level. ***, ** and * indicate significance at the 1%, 5% and 10% levels, respectively. CSR t (1) (2) (3) (4) (5) (6) Return t (0.90) (-0.51) Alpha t *** *** (-8.85) (-10.25) Flows t * (0.15) (0.06) (1.76) Std flows t (-0.52) (-0.52) (-1.04) Volatility t *** *** (-4.67) (-7.03) R-square t *** *** (-4.58) (-4.91) Ln(TNA t) (-1.52) (-1.40) Ln(Number of stocks t) ** ** (-2.41) (-2.18) Expense ratio t *** *** (-3.26) (-3.60) Turnover t (1.14) (0.06) Fund age (0.81) (0.73) SRI dummy 0.142** 0.137** (2.44) (2.30) Intercept *** *** *** *** 0.718*** 0.717*** (-5.68) (-9.09) (-5.22) (-8.75) (5.15) (5.25) N Adj. R

29 Table 4 Corporate social responsibility (CSR) and style factors Table 4 reports the regression results of the CSR score on the Carhart (1997) model coefficients: alpha, βmkt, βsmb, βhml and βumd. The regression includes the following control variables: volatility, R-square, the natural logarithm of the total net assets (TNA), the natural logarithm of the number of stocks, the expense ratio, the turnover ratio, fund age and the SRI dummy, which equals 1 for SRI funds and 0 otherwise. The regression includes time (year) and style dummies, and standard deviations are clustered at the fund level. ***, ** and * indicate significance at the 1%, 5% and 10% levels, respectively. CSR t (1) (2) (3) (4) (5) (6) (7) Alpha t *** *** *** (-8.85) (-10.21) (-9.61) β MKT,t *** *** ** (-4.13) (-5.19) (-2.07) β SMB,t *** *** * (-4.64) (-4.33) (-1.91) β HML,t *** *** *** (-6.72) (-8.24) (-8.27) β UMD,t *** *** *** (-5.29) (-6.12) (-5.54) Volatility t *** (-5.69) R-square t *** (-4.74) Ln(TNA t) * (-1.67) Ln(Nb. of stocks t) (-0.84) Expense ratio t *** (-3.47) Turnover t (1.38) Fund age (0.16) SRI dummy 0.130** (2.19) Intercept *** *** *** *** *** (-9.09) (-0.63) (-7.49) (-8.79) (-8.34) (0.37) (5.26) N Adj. R