3.1 Base Sample and Variables

We obtain mutual fund data from the Korea Fund Ratings (KFR), which contains information on fund codes, fund names, inception date, expenses, total net assets (TNA) and net asset value (NAV) at the end of each month, and daily and monthly returns. The KFR also provides fund classifications for identifying ESG (SRI) funds. We include open-end equity funds and exclude index funds, passively managed funds (e.g., ETFs), and any funds with less than 12 months of return data.

Our sample period ranges from July 2012 to December 2020 because ESG ratings are being published since June 2012. To construct ESG scores at the fund level, we first obtain firm-level ratings in the overall ESG and ESG subcategories (E, S, and G) from the Korea Corporate Governance Service (KCGS). The KCGS publishes ESG ratings as S, A+, A, B+, B, C, and D; thus, we assign a score ranging from 7 (highest) to 1 (lowest). We standardize the firm-level ESG score by industry and then calculate the value-weighted fund-level ESG score using portfolio holdings. The KCGS has publicly disclosed ESG ratings on a yearly basis with different announcement dates.³⁾ Accordingly, we assign the most recent yearly ESG score to each month t. ESG score (E, S, and G score) represents the fund-level ESG score (environmental, social, and governance score).

Based on the standard measure used in the literature (Sirri and Tufano, 1998), we define fund flow as follows:

where TNA_i,t is the total net asset value of fund i at the end of month t, and r_i,t is the fund return at month t. Flow is winsorized at the 0.05% and 99.5% levels. To measure a fund’s performance, we use both raw fund returns and alpha. Return is the average monthly return during the 12-month period, ending at the end of the month. Alpha is estimated based on CAPM, Fama and French’s (1993) three-factor model, or Carhart’s (1997) four-factor model, depending on the model specification.⁴⁾ Taking Carhart’s (1997) four-factor model as an example, Alpha is calculated as follows:

where r_i,t is the fund’s return in month t, r_{f , t} is the risk-free rate, r_mkt,t is the value-weighted market return, and SMB_t, HML_t, and UMD_t are the size, value, and momentum factors,^5), respectively. For each month t, we estimate the factor loadings over the prior 36-month rolling window. Next, we compute the monthly abnormal returns using the estimated factor loadings. Alpha is the average abnormal return in the 12-month period ending at the end of the month. Rsquare is estimated from regression model (2); it captures the proportion of the fund’s return explained by Carhart’s (1997) four factors. A lower Rsquare may indicate a greater degree of active management (Amihud and Goyenko, 2013).

We include a series of fund-level control variables in our empirical analysis. LnTNA is defined as the natural logarithm of fund TNA in month t. LnAge is defined as the natural logarithm of the number of months observed since the fund’s inception. Volatility is the standard deviation of the monthly fund returns over the previous 12 months. Turnover is the maximum aggregated sales or aggregated purchases of securities divided by the average 12-month fund TNA. Expense is defined as total operating expenses divided by total NAV at the end of the previous month. LnNumStock is the total number of stocks in a fund portfolio.

3.2 ESG and non-ESG Funds

We obtain a list of ESG funds from the KFR^6), and construct a sample of matched conventional funds (non-ESG funds) based on the propensity score matching methodology (Alda, 2020; Ammann et al., 2019). The initial sample consists of 565 equity mutual funds: 24 ESG funds and 541 conventional funds (1,707 and 40,070 fund-month observations, respectively). We estimate the propensity score as the similarity measure among funds, considering a logistic regression on fund characteristics based on the ESG score, size, age, performance, expense ratio, turnover, and the number of stocks.^7), We conduct a 1:1 nearest neighbor matching, resulting in 24 ESG funds and 24 non-ESG funds (1,707 and 1,542 fund-month observations, respectively). <Figure 1> displays the distributions and QQ plots of the ESG score, suggesting that the matching procedure improves the baseline characteristics of ESG and non-ESG funds.

<Figure 1>

QQ-plots for the ESG Score.

The figure displays the distributions and QQ-plots of ESG scores. Panel A presents the full sample results, and Panel B presents the matching sample results.

<Table 1> presents the summary statistics for ESG and non-ESG funds. We find that ESG funds have significantly higher ESG, E, S, and G scores than non-ESG funds, even when the matching is based on ESG scores. This result is consistent with our premise that ESG funds preserve their sustainability objective. The average raw returns of ESG funds are not statistically different from those of conventional funds. However, the average alphas of ESG funds are significantly lower than those of non-ESG funds. The average ESG fund size is 35 billion KRW, which is much larger than that of non-ESG funds. The average age of ESG funds is 9 years, which is similar to that of non-ESG funds. Furthermore, ESG funds exhibit a lower R-square and a smaller number of stocks are included in the portfolio, implying that they are more selective and hold a less diversified portfolio than conventional funds.

<Table 1>

Summary Statistics

This table presents the summary statistics of ESG and matched conventional funds (non-ESG funds). The fund variables included ESG score, raw return, CAPM alpha, Fama and French’s (1993) three-factor alpha, Carhart’s (1997) four-factor alpha, flow, TNA, age, volatility, turnover, R-square, expense, and number of stocks. The last column indicates the difference in the mean value between ESG and non-ESG funds. ***, **, and * indicate significance at the 10%, 5%, and 1% levels, respectively.

4.1 ESG Score and Fund Characteristics

We first examine the influence of fund characteristics on the ESG score of both ESG and non-ESG funds, using the following regression model:

where ESG score_i,t is the ESG (E, S, and G) score for fund i in month t. D(ESG)_i is a dummy variable equal to one if fund i is an ESG fund, and zero otherwise. The control variables include return, flow, size, age, return volatility, turnover, R-square, expenses, and the number of stocks.⁸⁾ The regressions include the year dummy; the standard errors are adjusted for clustering at the fund level.

<Table 2> examines the influence of fund characteristics on ESG scores. The first four columns show the results of Equation (3) for all the funds analyzed. Consistent with the results in <Table 1>, the ESG score is significantly higher for ESG funds than conventional funds. Moreover, ESG funds present a greater consideration of sub-category (environmental, social, and governance) issues separately. We also find some commonalities among funds that influence fund ESG scores. In general, funds with lower flow, lower volatility, lower turnover, higher R-square, and a smaller number of stocks present higher ESG scores. Moreover, funds with high expense ratios generally exhibit higher ESG scores, indicating that ESG investing requires greater ESG information costs. However, the relationship between expense and ESG score becomes negative for the ESG fund sample, possibly benefiting from the accumulated ESG information.

<Table 2>

ESG Scores and Fund Characteristics

This table presents the results from regressing the ESG score on the following fund characteristics: ESG dummy, raw return, flow, size, age, volatility, turnover, R-square, expense, and the number of stocks. The regressions include year dummy variables. t-statistics based on standard errors clustered at the fund level are reported in parentheses. ***, **, and * indicate significance at the 10%, 5%, and 1% levels, respectively.

Next, we investigate the relationship between fund ESG scores and factor loading. El Ghoul and Karoui (2017) show that SRI funds with high scores generally have a significant exposure to small beta, large capitalization, growth, and contrarian stocks. Similarly, we estimate Carhart’s (1997) factor loadings using Equation (2) and regress the fund ESG score on the factor loadings. <Table 3> shows the regression results. Unlike El Ghoul and Karoui (2017), ESG funds with high ESG scores do not have a significant exposure to low beta stocks, whereas non-ESG funds with high ESG scores have higher exposure to high beta stocks. For both ESG and non-ESG funds, the ESG score is significantly related to size factors. In other words, a high fund ESG score indicates a significant exposure to large-capitalization stocks. Unlike matched conventional funds, ESG funds with high ESG scores have a greater exposure to value stocks. Among the control variables, ESG funds with a smaller size, older age, and lower expenses tend to have higher ESG scores.

<Table 3>

ESG Scores and Fund Factor Loadings

This table reports the results of regressing the ESG score on Carhart’s (1997) factor loadings and the following control variables: flow, size, age, volatility, turnover, R-square, expense, and the number of stocks. The regressions include year dummy variables. t -statistics based on standard errors clustered at the fund level are reported in parentheses. ***, **, and * indicate significance at the 10%, 5%, and 1% levels, respectively.

Overall, our findings so far reveal that ESG funds share common features with non-ESG funds, based on their fund attributes. However, ESG funds preserve their ESG identity because they tilt toward a better ESG portfolio than conventional funds.

4.2 Fund Flow-return Relationship

In mutual fund literature, past performance is one of the most important determinants of fund flows (Chevalier and Ellison, 1997; Sirri and Tufano, 1998; Del Guercio and Tkac, 2002). However, ESG investors may drive their utility primarily from non-pecuniary attributes and, thus, are less sensitive to past performance (Benson and Humphrey, 2008; Bollen, 2007; Renneboog et al., 2011; Riedl and Smeets, 2017). To examine the fund flow-performance relationship, we estimate the following regression model:

where Flow_i,t is the flow of fund i in month t. D(ESG)_i is a dummy variable equal to one if fund i is an ESG fund, and zero otherwise. Return_i,t-1 is the average monthly return of fund i over the months t₋12 to t₋1, and R⁺ and R⁻ are indicator variables equal to one if Return_i,t-1 is non-negative or negative, respectively. Controls_i,t-1 is a vector of control variables, including flow, size, age, return volatility, turnover, R-square, expenses, and the number of stocks. The regressions include the year dummy; the standard errors are adjusted for clustering at the fund level.

<Table 4> shows the regression results. First, fund flows are sensitive to past returns. In the first column, using raw returns, the coefficient of R⁺ ×Return is 1.032 (t-statistic = 4.18) and statistically significant, implying that fund flow is sensitive to positive past returns. However, fund flows are not significantly responsive to negative past returns. When we replace raw returns with alphas based on CAPM, Fama and French’s (1993) three-factor model, and Carhart’s (1997) four-factor model, we find that fund flows are positively associated with both positive and negative past performance. Second, in all specifications, we find that the flow-performance sensitivity is not statistically different from that of conventional funds. For example, in the first column, the coefficients of R⁺×Return×D(ESG) and R^-×Return×D(ESG) are 0.187 and -0.050, respectively, but both are statistically insignificant.

<Table 4>

Flow-return Relationship and ESG Fund

This table presents the flow-return sensitivity. The dependent variable Flow_i,t is the flow of fund i in month t. D(ESG)_i is a dummy variable equal to one if fund i is an ESG fund. Return_i,t-1 is the raw return, the CAPM alpha, the Fama and French’s (1993) three-factor alpha, or the Carhart’s (1997) four-factor alpha over the previous 12 months depending on the specifications. R⁺ and R^- are indicator variables that equal one if the average raw return is non-negative or negative, respectively. The control variables are return, flow, size, age, volatility, turnover, R-square, expenses, and the number of stocks. The regressions include year dummy variables. t -statistics based on standard errors clustered at the fund level are reported in parentheses. ***, **, and * indicate significance at the 10%, 5%, and 1% levels, respectively.

<Table 5> further examines whether the flow-return sensitivity is affected by the holdings-based measure of the ESG score. The regression model specification is the same as that of Equation (4), except for the introduction of the ESG dummy variable D(ESG) _i,t-1^High, which is a dummy variable equal to one if an ESG fund has a higher ESG score than the median value for each month t-1, or zero otherwise. The first column reports that the coefficient of R⁺×Return×D(ESG)^High is -0.781 (t-statistic = -1.92) and significant at the 10% level. The results are robust to using individual E and S scores in the second and third columns, respectively. These findings imply that the flow-performance relationship may depend on the level of ESG attributes. ESG funds with higher scores significantly weaken the flow-performance sensitivity.

<Table 5>

Flow–return Relationship with high ESG Scores

This table presents the flow–return sensitivity. The dependent variable flow_i,t is the flow of fund i in month t. D(ESG)^High_i,t−1 is a dummy variable equal to one if an ESG fund has a higher ESG score than the median value for each month t -1, and zero otherwise. Return_i,t−1 is the average raw return of fund i over months t -12 to t -1, and R⁺ and R⁻ and are indicator variables that equal one if the average raw return is non-negative or negative, respectively. The control variables are return, flow, size, age, volatility, turnover, R-square, expenses, and the number of stocks. The regressions include year dummy variables. t -statistics based on standard errors clustered at the fund level are reported in parentheses. ***, **, and * indicate significance at the 10%, 5%, and 1% levels, respectively.

4.3 Fund Performance

To examine the risk and return characteristics of ESG and non-ESG funds, we test the performance of ESG and non-ESG funds using the monthly returns of an equally weighted portfolio of funds based on the CAPM, Fama and French’s (1993) three-factor model, and Carhart’s (1997) four-factor model. <Table 6> presents the alpha and factor loadings of ESG and non-ESG fund portfolios. The estimates of the alphas are annualized for the presentation. The results show that the alphas for the ESG funds are not significantly different from those of non-ESG funds. Taking Carhart’s (1997) four-factor alpha as an example, the ESG funds are observed to have an average risk-adjusted return of -1.050%, whereas the non-ESG funds have an average return of -0.726%. The difference in alphas between the ESG and non-ESG funds is -0.323% but statistically insignificant, which is consistent with Renneboog et al. (2008, 2011). Among the factor loadings, we find that ESG funds are significantly exposed to market risk and invest relatively more in high-beta stocks than in matched conventional funds.

<Table 6>

Fund Performance and Factor Loadings

This table presents the returns of the equally weighted portfolios of ESG and non-ESG funds based on the CAPM, the Fama and French’s (1993) three-factor model, and the Carhart’s (1997) four-factor model. Difference is a portfolio constructed by subtracting non-ESG from ESG fund returns. The alphas’ estimates are annualized for the presentation. t -statistics are computed with Newey and West (1987) standard errors and are reported in parentheses. ***, **, and * indicate significance at the 10%, 5%, and 1% levels, respectively.

To examine the determinants of the future fund performance for ESG and non-ESG funds, we use the following regression model:

where Alpha_i,t is the average abnormal return estimated from CAPM, Fama and French’s (1993) three-factor model, and Carhart’s (1997) four-factor model over the previous 12 months. ESG score_i,t-1 indicates the fund ESG score for fund i in month t-1. Controls_i,t-1 are the same as in Equation (3). The regressions include the year dummy; the standard errors are adjusted for clustering at the fund level.

Panel A in <Table 7> presents the regression results with different asset pricing models. For ESG funds, we find that the ESG attributes are negatively associated with future performance. In the first column, based on the CAPM alpha, the coefficient of ESG score is -0.154 (t-statistic = -4.35), which is statistically significant at the 1% level. The coefficients are also statistically significant when using alphas based on Fama and French’s (1993) three-factor model and Carhart’s (1997) four-factor model. However, this is not the case for the matched conventional funds, as the coefficient of is statistically insignificant under all specifications. These findings are consistent with the notion that the ESG screening process may constrain the investment universe, thus decreasing future performance (El Ghoul and Karoui, 2017; Renneboog et al., 2008, 2011). Unlike non-ESG funds, the coefficients of LnTNA and LnAge are significantly negative. These results suggest that larger and older ESG funds are more likely to undergo intensive screening and thus significantly underperform with regard to the matched conventional funds. Moreover, the coefficient of LnNumStock is significantly positive, further supporting the notion that ESG screens narrow the investment universe and decrease future performance. Panel B further reports the results of ESG funds based on the ESG sub-category using the Carhart alpha. The coefficients of E score and S score are negative and statistically significant at the 1% level. However, the coefficient result of the G score is not statistically significant, suggesting that the screening intensity of governance factor in the case of ESG fund is relatively low compared to the environmental or social factors.

<Table 7>

Determinants of Fund Future Performance

This table presents the estimates of the determinants of future performance. The dependent variable Alpha_i,t is the abnormal return estimated from the CAPM, the Fama and French’s (1993) three-factor model, or the Carhart’s (1997) four-factor model over the previous 12 months. The control variables are return, flow, size, age, fund return volatility, turnover, R-square, expenses, and the number of stocks. The regressions include year dummy variables. t -statistics based on standard errors clustered at the fund level are reported in parentheses. ***, **, and * indicate significance at the 10%, 5%, and 1% levels, respectively.

Panel A: Based on different asset pricing model

The “smart money” effect (Gruber, 1996; Zheng, 1999; Muñoz, 2019) indicates that investors can select mutual funds to generate higher future returns. To further test whether ESG investors are financially smart (Renneboog et al., 2011), we use the following regression:

where Alpha_i,t is the average abnormal return estimated from Carhart’s (1997) four-factor model over the previous 12 months. D(ESG)_i and Controls_i,t-1 are defined using Equation (3). Column (1) in <Table 8>, based on the CAPM alpha, shows that fund flows are positively related to future fund performance for all sample funds (Muñoz, 2019; Yoo and Kim, 2012). However, we find no evidence to support that ESG fund investors have better selection skills than non-ESG fund investors. As shown in Column (2), the coefficient of the interaction term between Flow and D(ESG) is insignificant. In Columns (3) to (6), the results are robust with the alphas based on Fama and French’s (1993) three-factor model and Carhart’s (1997) four-factor model. These results suggest that ESG fund investors’ fund-selection skills are not significantly different from those of non-ESG fund investors.

<Table 8>

Smart Money Effect

This table presents the results for the smart money effect. The dependent variable Alpha_i is the abnormal return estimated from the CAPM, the Fama and French’s (1993) three-factor model, or the Carhart’s (1997) four-factor model. The alphas and returns are converted from decimals to percentages for regression. D(ESG)_i is a dummy variable equal to one if fund is an ESG fund. The control variables include return, flow, size, age, fund return volatility, turnover, R-square, expenses, and the number of stocks. The regressions include year dummy variables. t -statistics based on standard errors clustered at the fund level are reported in parentheses. ***, **, and * indicate significance at the 10%, 5%, and 1% levels, respectively.