Determinants of mutual fund performance
-an empirical study on the Dutch market-
Daniël Andreas Kalkhuis* January 2010
Abstract
In this paper we study the relations between fund characteristics and fund performance for Dutch open-end active mutual funds. The sample consists of 81 domestic and international funds. Time-fixed effects analysis, with Fama and French alphas as measure of fund performance, shows that fund performance is negatively related to fund size and number of holdings and positively to fund age. In a robust analysis on a subsample of domestic funds only, these relations are confirmed, and show in addition, also a negative relation with turnover. For international funds we only find a significant, negative, relation between fund performance and fund size. Robustness analyses with five-factor alphas, as measure of fund performance, show for the full sample, only a negative relation between fund size and fund performance.
JEL classification: G11, G23
Keywords: Mutual fund; Fund performance; Fund characteristics; Dutch market
*
This study examines the relations between fund performance and fund characteristics for Dutch open-end active mutual funds. The Dutch mutual fund sector managed around $62.133 billion (Lipper Hindsight2) at the end of 2007, which is a sizeable market in Europe (e.g. Ferreira et al. 2009; Otten and Bams, 2002). Whereas most studies (e.g. Carhart, 1997; Chen et al., 2004; Kacperczyk, et al., 2005) on mutual fund performance focus their attention on U.S. funds, evidence on the determinants of fund performance for the Dutch market is limited to the study of Otten and Bams (2002). They studied the relations between fund performance and expense ratio, size, and age, for Dutch domestic funds as part of an European study. A disadvantage of their study is that they measure fund performance over the full sample period, and fund characteristics only at the end of the sample period, which result in possibly self-induced correlations3. Recently, Ferreira et al. (2009) have done a comprehensive international cross-country (the Netherlands included) study on fund performance. They studied the influence of fund characteristics and country characteristics on fund performance for U.S funds, non-U.S. fund, domestic and international funds. Their study however does not contain results for individual countries, so the relations between fund performance and fund characteristics are not clear for Dutch funds.
The relevance of research on mutual fund performance is given, by the important role that mutual funds play in the wealth dependence of investors, as addressed by Prather et al. (2004). This dependence underlines the interest of investors on information that support their funds selection process. Insight on the relations between fund performance and fund characteristics contributes to that interest.
2
Data from Lipper Hindsight is obtained from the study of Ferreira et al. (2009) 3
The aim of our study is to make a comprehensive analysis of the relations between fund characteristics and fund performance for Dutch open-end active mutual funds. We use the study of Ferreira et al. (2009) as starting point for our research, however in contrast, we focus only on fund characteristics. More specifically, we consider the following fund characteristics as determinants of fund performance: size, age, flow of money, expense ratio, turnover ratio, persistence and number of holdings. We estimate the relations with annual data about fund characteristics, which are one-year lagged to fund performance.
The research question of this study is:
Which fund characteristics influence performance of Dutch open-end active funds?
turnover as determinant of fund performance. Turnover ratio is also studied by several other studies (e.g. Dahlquist et al., 2000; Prather et al., 2004) as determinant of fund performance.
In order to examine the relations between fund performance and fund characteristics, we first estimate the risk-adjusted performance (alphas) of funds. We apply a Fama and French three-factor regression model (hereafter also: Fama and French model) with country/region-specific benchmarks, to estimated the risk-adjusted fund performance of funds individually. We use Ordinary Least Square (OLS) for the regressions. Second, we apply a time-fixed effects regression model to examine the relations between the estimated fund performance and one-year lagged fund characteristics data. As a robustness analysis, we run regression on subsamples and with five-factor alphas, as measures of fund performance. Results of our initial regression are used to test our hypotheses on the relation between fund performance and fund characteristics.
We use data for 81 Dutch open-end active mutual funds over the period 2005 to 2009. The sample contains both domestic funds (funds investing only in the home country) and international funds (investing in countries outside the home country). Furthermore we can classify the funds into sector funds, property funds, sustainability funds and country/region funds. Data about fund characteristics and persistence are one-year lagged in relation to fund performance. For 2009, we collect data about fund performance only.
performance is negatively related to size. Further, robustness analyses with five-factor alphas shows also several significant results. For the sample, containing domestic and international funds, we find again that fund performance is negatively related to size, we find this relation also for a subsample of international funds. Furthermore, we find for domestic funds, a negative relation with age.
I. Literature review
A. The relation between fund performance and expenses
The efficient market theory of Fama (1970) assumes that in highly efficient market all information is already incorporated in security prices. This suggests that there exist no opportunities for fund managers to earn back the expenses related to active fund management. Studies of Sharpe (1966), Gruber (1996) and Carhart (1997) all report a negative relation between fund performance and expenses.
setting of funds. The empirical study of Ferreira et al. (2009) find, for a sample of domestic and international funds worldwide, that fund performance is negatively related to expense ratio. Furthermore, when split out the sample into domestic and international funds, they find the negative relation holds for domestic funds only, while for international funds the (positive) relation is not significant. Otten and Bams (2002) find for Dutch domestic funds a negative relation between expense ratio and fund performance.
Given the discussed literature, we do not expect that funds are able to find enough underpriced securities to compensate expenses related with active management, resulting in a negative relation between fund performance and expense ratio. Therefore, we test the relation with the following null hypothesis (H1,0) against a one-sided alternative hypothesis (H1,A):
H1,0: Expense ratio is not related to fund performance; and
H1,A; Expense ratio is negatively related to fund performance.
B. The relations between fund performance and fund size
Ferreira et al. (2009) suggest that trading of relatively high volumes of stocks by large funds, possibly signals other market participants about the funds’ intentions, which results in high price impact costs.
Empirical studies give mixed results for the relation between fund performance and fund size. There are several studies that find no relation between fund performance and fund size (e.g. Grinblatt and Titman, 1994; Droms and Walker, 1996; Golec, 1996). In contrast, Grinblatt and Titman (1989) find for the lowest size quintiles, negative abnormal returns. Furthermore, Indro et al. (1999) show that there exists an optimal fund size. They suggest that a minimum size is necessary to justify the costs of fund management, however, when funds grow too large, the positive effect of fund size turns into a negative effect. Furthermore, Yan (2008) finds that the negative impact of size is especially pronounced for funds that invest in illiquid stocks, and among growth funds and high turnover funds. Otten and Bams (2002) find that fund performance is positively related to fund size, for several European countries including the Netherlands. Ferreira et al. (2009) suggest that that there still exist economies of scale opportunities for non-U.S. funds, because non-non-U.S. funds are much smaller than non-U.S. funds. Furthermore, their study shows that US funds are more focused on small and illiquid stocks, than non-US funds. Thus, non-U.S. funds face less liquidity constraints. Results of their study indeed show that the relations between fund performance is negative for U.S. domestic funds and positive for non-U.S. funds.
Given the relative large size4 of Dutch funds from an international perspective (the U.S. excluded), we expect that Dutch fund performance is negatively related to size, as cause of
4
liquidity constraints. Therefore we test the null hypothesis against a one-side alternative hypothesis. Hypotheses are given by:
H2,0: Fund performance is not related to fund size; and
H2,A: Fund performance is negatively related to fund size.
C. The relation between fund performance and fund age
So far, the relation between fund performance and fund age has received little attention in the literature and there is no well developed theory about this relation. Ferreira et al. (2009) argue that younger funds are more agile than older funds, however older funds are more experienced than younger funds.
Studies (e.g. Chen et al., 2004; Golec et al., 1996; Prather et al., 2004; Yan, 2008) on the relation between fund performance and fund age show a negative relation, however not significant. Although, Otten and Bams (2002) find a significant negative relation for the U.K. and for Germany. For the Netherlands they find no significant relation. Cremers and Petajisto (2008) report a negative relation for U.S. funds as well. Ferreira et al. (2009) find a negative relation for a sample of international funds.
Since studies on the relation between fund performance and funds age seems to point towards a negative relation, although this is not always significant, we test the null hypothesis against a one-sided alternative hypothesis. Hypotheses are formulated as follows:
H3,0: Fund performance is not related to fund age; and
D. The relation between fund performance and net flow to funds
Gruber (1996) argues that if performance of funds is predictable, and if investors are aware of future performance, than investments and disinvestments into funds should express this knowledge of investors. This proposed relation is also called the “smart money effect”. Net flow to funds (investments minus disinvestments) has indeed been found to relate positively to fund performance. However, the effect is more contributable to disinvestments of investors in bad performing funds than new investments in well performing funds. Furthermore, the relation is more pronounced for a quarter holding period than for a one-year holding period. Zheng (1999) finds that U.S. domestic funds which attract high net inflows perform better in the subsequent period than funds with net outflows. However a large part of the effect is contributable to a repeat winner strategy (investing in funds that perform good in the previous period). Furthermore the smart money effect is more pronounced for small capitalization funds than for large capitalization funds. Ferreira et al. (2009) find that flow is positively related to performance, for a sample of domestic funds worldwide. However their findings are not robust for a sample of international funds. Dahlquist et al. (2002), the only European study on this relation, find no significant relation for Swedish domestic funds.
In this study we examine the “smart money effect” as proposed by Gruber (1996). Specifically, we test whether higher net inflow (lower net inflow) is positively (negatively) related to fund performance. Therefore we formulate the null hypothesis against a one-sided alternative hypothesis. Hypotheses are given by:
E. The relation between fund performance and past performance
The ability of funds to outperform the market over time, also called performance persistence, received a lot of attention. Studies of e.g. Grinblatt and Titman (1992), and Elton et al. (1996), find that performance persist over long time horizons. Furthermore, Hendricks et al. (1993) find that performance persist over short-time horizons for a sample of no-load, growth-oriented mutual funds. Funds with the best past three-month performance perform significantly better in the subsequent year. This is also called the “hot hands effect”. Furthermore Brown and Goetzmann (1995) find that performance persistence over a short-time horizon exists for both the best and worst managers. Carhart (1997) reexamines the ‘hot hand effect’ of Hendricks et al. (1993), and finds that the Jegadeesh and Titman’s (1993) one-year momentum effect (a strategy of buying last year winners and selling last year losers) is mainly contributable to the “hot hands effect”. Carhart (1997) controls for the momentum effect.
performance persistence for both, domestic and international funds. Their results show that the effect is stronger for U.S. funds than for non-U.S. funds.
In order to examine whether funds are able to realize to realize outperformance over time, we test the relation with a null hypothesis against a one-sided alternative hypothesis. Hypotheses are given by:
H5,0: Fund performance is not related to past performance; and
H5,A: Fund performance is positively related to past performance.
F. The relation between fund performance and portfolio turnover
Wermers (2000) states that: “A concept that is central to the idea of actively managed funds outperforming index funds is that higher levels of trading activity are associated with better stock-picking abilities”. In that case fund performance is positively related to fund performance. However when trading is based on noise, or to attract investors, than trading only results in higher trading cost, resulting in a negative impact on fund performance.
turnover with 2,7 % on a yearly basis. Dahlquist et al. (2002) find a positive relation for Swedish funds as well.
The discussion on the relation between fund performance and turnover, seem to point towards a negative relation. So we test the null hypothesis against a one-side alternative hypothesis. We hypothesize that:
H6,0: Fund performance is not related to portfolio turnover; and
H6,A: Fund performance is negatively related to portfolio turnover.
G. The relation between fund performance and number of stock holdings
and therefore trading of focused funds result in high price impact, even for liquidity motivated trades.
In contrast to Yan (2008), who finds a positive and significant relation between number of holdings and fund performance (net of fees), Prather et al. (2004) find no significant relation between a relative number of holdings and fund performance.
Taken into consideration the agency problems (Chevalier and Ellison, 1997) and price impact that focused funds face (Yan 2008), we expect that fund performance is negatively related to number of holdings. So, again we test the null hypothesis against an alternative hypothesis. Hypotheses are formulated as follows:
H7,0: Fund performance is not related to number of holdings; and
II. Methodology
In order to examine the extent to which fund performance is related to fund characteristics, we first estimate the risk-adjusted fund performance. Thereafter we apply panel regression to estimate the relations between fund performance and fund characteristics.
A. Performance measure
We estimate the risk-adjusted fund performance with a Fama and French (1992) three-factor model, with country/region-specific benchmarks, and with a five-three-factor model, using five different country/region-specific benchmarks. Although the estimated fund performance with the five-factor model is only used for robustness analyses, we will discuss the model in this section and in the following (sub)sections.
In both models, the risk-adjusted performance of funds is estimated as a weekly average alpha for each year during the period 2005 to 20095. A positive alpha indicates that a mutual fund outperforms a portfolio of risky assets, and a negative alpha indicates that a fund underperforms compared to a portfolio of risky assets. The estimated Fama and French alphas and the five-factor alphas are used as measure of fund performance, (ai,t), and as measure of the persistence variable (Persistencei,t) in the panel regressions (as specified by equation 4). We use the one-week EURIBOR as risk-free rate in both models. Since the one-week EURIBOR rate is annualized, we divide the rate by 52. Furthermore we calculate returns as continuously compounded weekly returns from the first week of 2005 to the last week of 2009, using total return prices. The formula is given in equation 1.
5
Ri,t=100%*ln(Pi,t/Pi,t-1), (1)
where Ri,t is the continuously compounded return of fund ifor week t, Pi,t denotes the
total return price of fund i for week t, and ln denotes the natural logarithm. Prices are net of operating expenses, fees and transaction costs, but gross of sales fees.
Fama and French three-factor model. The Fama and French (1992) three-factor model
extends the CAPM by including a size and a book-to-market risk factor, in addition to the market risk factor. The Fama and French three-factor model fits better than the CAPM with the cross-sectional variation of fund returns, resulting in more efficient alpha estimates. A disadvantage of the Fama and French model is that it does not control for the momentum factor of Jegadeesh and Titman (1993). This momentum factor controls for a strategy to buy last winners and sell last losers. Carhart (1997) proposes a four-factor model, which includes the momentum factor in addition to the three factors of the Fama and French model. However, due to data constraints we apply the Fama and French three-factor model. Since our sample consists of both domestic and international factors we use country/region-specific benchmarks for all the three risk factors. As argued earlier, we estimate the risk-adjusted performance, for every year, as an weekly average. The model is given by equation 2.
Ri,t-Rf,t= αi+β1,i*(Rm,t-Rf,t)+β1,i*SMBt+β2,i*HMLt+εi,t, (2)
portfolio of high book-to-value stocks and a portfolio of low book-to-value stocks for week t. Ri,t, Rm,t, the SMB factor and the HML factor are all calculated as continuously compounded returns.
As mentioned before our sample consists of both domestic and international funds, therefore we use country/region-specific benchmarks for the market return, SMB and HML. For Dutch domestic funds we use the MSCI Netherlands as market proxy. The difference in return between the MSCI Netherlands small capitalization index and the MSCI Netherlands large capitalization index form the SMB factor. The difference in return of the MSCI Netherlands value index and the MSCI Netherlands growth index forms the HML factor. International funds can be categorized in four country/region-specific groups: Europe, the USA, Far East and worldwide. For these groups we use respectively the MSCI Europe, MSCI USA, MSCI Far East, MSCI World as market indices. The corresponding small, large, value and growth MSCI style indices are used for forming the returns for the SMB factor and HML factor, respectively.
Five-factor model. The second model is an international version of the CAPM, which
covers the investment markets Netherlands, Europe, North America, Emerging markets and Asia. We use the following benchmarks respectively: MSCI Netherlands, MSCI Europe ex-Netherlands, MSCI North America, MSCI Emerging Markets and MSCI All Country Asia. These five benchmarks cover all the investment markets of Dutch mutual funds and are mutually exclusive. So potential benchmark errors are minimized. The five-factor model is given by equation 2.
Ri,t-Rf,t= αi+β1,i*(RNL,t-Rf,t)+β2,i*(REUR-NL,t-Rf,t)+β3,i*(RNA,t-Rf,t)+
where RNL,t stands for the Dutch market return, REUR-NL,t denotes the European (the Netherlands excluded) market return, RNA,t stands for the market return of North America, REM,t stands for the market return of emerging markets, and RASIA,t stands for the market return of Asia.
B. Panel regression analysis
In this section we qualify the data and describe the methodology we use to examine the relations between mutual fund performance and fund characteristics.
We use unbalanced panel data (data containing both time-series and cross-sectional data) for determining the determinants of mutual fund performance. Brooks (2008) argues that combining cross-sectional and time-series data, increases the number of degrees of freedom, and thus the power of the test. Also panel data mitigate problems of multicollinearity that possibly will occur if time-series are modeled separately. Furthermore, Brooks (2002) states that panel data are better than time-series or cross-sectional samples, for the study of the adjustment process of the dependent variable as reaction on changes in the independent variables.
A characteristic of the data is that some cross-sectional units have fewer observations, as consequence of funds merging or not surviving, or because cross-sectional elements are simply missing. Therefore the data are qualified as unbalanced panel data. The unbalanced data are no problem, since the software package will automatically account for missing data (Brooks, 2008).
random panel effects model. The entity-fixed effects model assumes varying intercepts between funds, which are constant over time. In this model dummies are used to capture the cross-sectional varying intercepts. The time-fixed effects model assumes varying intercepts over time, which are constant among funds. Again dummies are used, now to capture the time-varying intercepts. The random panel effects is comparable to the entity-fixed effects model, since it allows the intercept to vary cross-sectional, but not over time. The random effects model is preferred, since this model does not use dummies and therefore degrees of freedom are saved, resulting in a more efficient estimation than the (time- or entity-) fixed effects models. A disadvantage of the random model is that the model requires that the composite error term is not correlated with any of the independent variables. In case the composite error term is correlated, the fixed effects model is preferable. The assumption that the composite error term is not correlated to the individual variables is tested with the Hausman test as provided by Eviews. Results are given in table A of the appendix. The p-value shows that the composite error term is significantly correlated with one of the dependent variables, indicating that the random model is not appropriate for estimating the model6.
Since we cannot use the random effects model, therefore we estimate the model with both, time-fixed and cross-sectional fixed effects. In order to determine whether these effects are necessary or not, we apply the redundant effect-likelihood ratio tests. These tests restrict the cross-sectional effects, the period-fixed effects and both types of fixed effects to zero. Table B in
6
the appendix7 reports the result of the test on the regression with Fama and French alphas. As shown, the time-fixed effects are significantly different form zero, the cross-sectional effects are not (as indicated by the p-value). Therefore we estimate a model with only time-fixed effects. The results for the test with five-factor alphas (used as robustness analysis) show a slightly different result. Again the time-fixed effect are significantly different from zero. For the cross-sectional fixed effects we find mixed results. The F-test version is not significantly different from zero, however the Chi-square indicates a significant result. The difference between the two tests is probably due to relatively small sample size. We ignore the significant result of the Chi-square, because the F-test version is in a more direct way related to sample size, and therefore mostly preferred in finite samples (Brooks, 2008). Consequently we also apply a model with period-fixed effects only for the robustness analysis with five-factor alphas.
In equation 4 the model is given, that we use to estimate the determinants of risk-adjusted fund performance. As argued before, the model allows for time-fixed effects. We run the model with Fama and French tree-factor alphas (as estimated with equation 2). Furthermore we run the model also with five-factor alphas (as estimated with equation 3) as robustness check.
ai,t = a0+β1 Persistencei,t+β2 ln(Size)i,t-1+β3 ln(Age)i,t-1+β4 Turnoveri,t-1+
β5 Flowi,t-1+β6 ln(Expenses ratio)i,t-1+β5 ln(Holdings)i,t-1+λt+vi,t (4)
where ai,t is the risk-adjusted performance of fund ifor year t measured as the Fama and French alpha, Persistencei,t is the performance persistence of fund ifor year t, ln(Size)i,t-1 is the
7
natural logarithm of fund size measured as the Net Asset Value (NAV), ln(Age)i,t-1 is the natural logarithm of fund age, Turnoveri,t-1 is the natural logarithm of portfolio turnover ratio, Flowi,t-1 is the net flow (inflow minus outflow) to funds, measured as ratio of NAV, ln(Expense ratio)i,t-1 is natural logarithm of the total expense ratio. Ln(Holdings)i,t-1 is the natural logarithm of number of holdings, λt is a time varying intercept, vi,t is the remaining disturbance term.
III. Data and descriptive statistics
First we describe the sample of funds for this study. Thereafter we discuss the data collection for estimating fund performance. Finally, we discuss the data collection for fund characteristics.
A. Sample description
We use data of 81 open-end Dutch active mutual funds for examining the relation between fund characteristics and fund performance. We collect weekly fund price data over the period 2005 to 2009, and yearly data about fund characteristics over the period 2005 to 2008. Data on fund characteristics are one-year lagged to fund returns and data on fund prices for 2005 are only used for estimating the persistence variable. The reason for the time period of our study, is the limited availability of data on fund characteristics before 2005.
relation between fund characteristics and fund performance too much, then we exclude the fund from the sample for one-year. Finally, there remain 81 funds for studying purpose, consisting of 8 dead funds and 73 living funds (70 international funds and 11 domestic funds). An overview of the funds included in the sample is given in table C of the appendix.
B. Data collection to estimate fund returns
We estimate the risk-adjusted fund performance with weekly total return fund prices. For indices we also use total return prices. Total return fund prices assume reinvestment of dividends, and therefore avoid possible biases as consequence of mutual fund’s dividend policy. Data are collected from the first week of 2005 to last week of 2009.
For the Fama and French three-factor model we collect data on the MSCI Netherlands, MSCI Europe, MSCI USA, MSCI Far East, MSCI World indices and on their corresponding small cap, large cap, value and growth style indices. For the five-factor model we collect data about respectively the MSCI Netherlands, MSCI Europe ex Netherlands, MSCI North America, MSCI Emerging Markets and MSCI All Country Asia. Finally, we use the one-week EURIBOR as risk-free rate. All data are provided by Thomson Datastream.
C. Data collection of fund characteristics
C.1. Description of variables
Table 1 provides an overview of the fund characteristics that we consider in this study as determinants of mutual fund performance. A definition of fund characteristics is given as well.
Table 1: Definitions of determinants of mutual fund performance
This table gives the definitions of the fund characteristics that are considered determinants of fund
performance. All determinants are measured at year-end. A definition of fund characteristics is given
as well.
Determinant Definition
Fund performance Alpha (% per year) estimated with the Fama and French model, or five-factor model (used for robustness analyses), using weekly total return prices in €.
Source: data are obtained from Thomson Datastream, alphas are estimated using Excel.
Age Number of years since the establishment of the fund, measured in full years. Source:
Morningstar and annual reports.
Size Net Asset Value (NAV) measured as the total assets minus short-term liabilities in millions €.
Source: annual reports
Flow Ratio of growth in fund size, the difference between fund inflow and outflow (controlled for
dividend) as fraction of a one-year lagged TNA. Source: annual reports, ratio is calculated in Excel.
Expense ratio Total annual expenses (performance fee included), measured as ratio of net asset value (NAV).
Source: annual reports.
Holdings Number of stock holdings. Source: annual reports.
Turnover Measured as the difference of purchases and sales minus the difference of purchases
ensuring from (re)issuances and redemption of shares of the fund, as a percentage of the average TNA. Source: annual reports.
Persistence Is the one-year lagged alpha (% per year) estimated with the Fama and French model,
or five-factor model. (Persistencet = Fund performancet-1 ). Source: data are obtained from
Thomson Datastream, one-year lagged alphas are estimated using Excel.
C.2. Descriptive statistics
Table 2 reports the descriptive statistics of the fund characteristic independent variables. The average fund size is almost €300 million, and the average TER is 1,4%. The five-factor persistence variable is only used for the robustness analysis with five-factor alphas as independent variables. The significant Jargue-Beras indicates that the error terms for all the variables are not in line with a normal distribution. In the remaining study we use the natural logarithm for age, number of holdings, size and TER. There are three reasons to use logarithms (Brooks 2008). First, they can rescale the data, resulting in a more constant variance. Second, logarithms can make a skewed distribution more normally distributed. And third, logarithms can make a non-linear, multiplicative relationship between variables more linear. In Table E of the appendix, we report the descriptive statistics of the fund characteristics using natural logarithms. This table shows that using natural logarithm improves both the skewness and the Jargue-Beras of the variables age, holdings, size and TER. However only the Jargue-Bera for fund size is not longer significant.
C.3. Correlations
relation between flow and persistence means that outperforming funds attract more money from investors in the subsequent period than underperforming funds. Age is negatively related to turnover, expense ratio, and positively related to size. Indicating that older funds trade less, have lower expenses and are larger.
Table 2: Descriptive statistics of fund characteristics
This table reports the descriptive statistics of the mutual fund characteristics variables and the persistence variable, for all funds. Values are calculated as yearly average over 2005 to 2008, using year-end data. Age of fund measures the number of years since the founding of funds. Flow, is the aggregate of money inflow and outflow, adjusted for dividend payments, as percentage of last year’s NAV. Holding measures the number of equity holdings. Size measures fund size (NAV). TER is measured as all operating cost, as percentage of NAV. Turnover measures the portfolio turnover of funds. Fama and French persistence is estimated as the one-year lagged Fama and French alpha (see equation 2), five-factor persistence is estimated as the on-year lagged five-factor alpha (see equation 3). The five-factor alpha persistence variable is only used for the robustness analyses. The Jargue-Beras of the variables and persistence are all significant, indicating that the errors terms are non-normally distributed. ***
indicates significance at the 1% level.
Age (years) Flow (%) Holdings (number) Size (NAV, in € millions) TER (%) Turnover (%) Fama and French
Table 3: Correlation matrix of independent variables
This table shows the correlation coefficients of independent variables. Panel A contains the correlation coefficients of independent variables including the Fama and French persistence variable and Panel B contains the independent variables including the five-factor persistence variable. Except for the persistence variable, the sample of panel B is equal to the sample of panel A. The sample period is from 2005 to 2008. The variables age, holdings, size and TER are measured as the natural logarithm values. *** indicates significance at the 1% level, ** indicates significance at the 5% level and * indicates significance at the 10% level.
Panel A: Sample including Fama and Frenh persistence variable
1 2 3 4 5 6 7 (1) Flow (%) 1.00 (2) Ln Age (Years) -0.21*** 1.00 (3) Ln Holdings 0.01 0.06 1.00 (4) Ln Size (million €) 0.01 0.33*** 0.05 1.00 (5) Ln TER (%) 0.03 -0.19*** -0.23*** -0.21*** 1.00 (6) Turnover (%) -0.05 -0.18*** 0.18*** -0.18*** 0.02 1.00
(7) Fama and French persistence (%) 0.18*** -0.11* -0.06 0.12** -0.12** -0.10 1.00
Panel B: Sample including five factor persistence variable
IV. Empirical results
A. Results of the main analysis
In this section we report the main results from the analysis with the time-fixed effects model, whereby we use Fama and French alphas as independent variables. Table 4 reports the empirical results of the time-fixed effects model as specified by equation 4. As shown from the R-squared, the panel regression with the Fama and French alpha estimates explains almost half (0.41) of the variation in risk-adjusted returns. Furthermore, the adjusted R-squared indicates that the variables included in the model are justified. The table reveals that fund performance is not significantly related to expense ratio. This indicates that we cannot reject the null hypothesis (H1,0) that states that fund performance is not related to expense ratio. The insignificant relation suggests that funds with higher expense ratios, are able to find enough tradable underpriced securities to offset their costs of fund management. This finding is in line with the theory of Grossman and Stiglitz (1980) that in a market where information and implementation is costly, there are profitable trading opportunities for informed investors to fully compensate their costs.
constraints because these are significant smaller that U.S. funds. Given the relative large size of Dutch funds, in comparison to other non-U.S. funds, the result is not unexpected.
The table reports a significant positive coefficient for age, which means that older funds perform better than younger funds. Hence, we reject the null hypothesis (H3,0) that states that fund age is not related to fund performance. Also this result is not in line with our alternative hypothesis (H3,A) that propose a negative relation. The negative relation is quite unique, because most studies find no relation (e.g. Chen et al., 2004; Yan,2008), or find a negative relation between fund age and fund performance (Ferreira et al., 2009).
The table show an insignificant (negative) coefficient for flow, so there is no evidence that funds with large inflow during the previous period, have higher future returns. Our result show no evidence about the “smart money effect” as proposed by Gruber (1996). Hence, the null hypothesis (H4,0) stating that fund performance is not related with net flow to funds, cannot be rejected.
As shown, for the persistence variable we also find no significant coefficient. Therefore we cannot reject the null hypothesis (H5,0), which states that fund performance is not related to future fund performance. This finding indicates that there is no evidence that good (or bad) performance persist over time.
find that funds with high turnover underperform funds with low turnover. Our study supports the idea that funds are able to find enough underpriced securities to compensate the trading costs.
We find a negative coefficient for holdings, which is significant at the 10% level. Hence, we reject the null hypothesis (H7,0), which states that fund performance is not related to the number of holdings. However this finding is also not in line with our alternative hypothesis (H7,A) that hypothesize a negative relation. Our result indicates that focused funds outperform diversified funds. This finding contrasts with Chan (2008), who suggest that focused funds underperform compared to diversified funds as a result of agency problems and price impact.
Table 4. Results of the time-fixed effect regressions
This table shows the results of the time-fixed effects analysis with Fama and French alphas as independent variable. Relations between fund performance and fund characteristics are estimated with the following model: ai,t= a0+β1 Persistencei,t+β2 ln(Size)i,t-1+β3 ln(Age)i,t-1+β4 Turnoveri,t-1+β5 Flowi,t-1+β6 ln(Exp) i,t-1+β5 ln(Positions)i,t-1+λt+vi,t. We use yearly data for our regression. Because alphas are estimated as
weekly alphas for each year, we multiply these by 52. The sample consist of 278 observations, from 81 funds over the period 2005 to 2008. Prob. show whether the relation is significant or not. ***Indicates significance at 1% level, ** indicates significance at 5% level and * indicates significance at 10% level.
Independent variable: Fama and French alpha
Variable Coefficient Std. Error Prob.
B. Results of robustness analyses
In this subsection we perform several robustness analyses with the regression models specified in equation 4. We perform robustness analyses with five-factor alphas (estimated with equation 3) as independent variables. We perform an analysis with before-fee (gross) performance, which is estimated by dividing the expense ratio by 52, and adding it back to weekly fund returns. Whereas the results of the initial analysis do not indicate a significant relation between fund performance and expense ratio, we expect that if funds invest their money efficiently (Ippolito, 1980), that than the expense ratio is positively related to before-fee performance. Furthermore, we perform robustness analyses on three different subsamples, a sample that consist of only domestic funds, a sample that consists of only international funds, and a sample which excludes sector funds, property funds and sustainability funds. Following the reasoning of Ferreira et al. (2009), we expect that international funds face relative smaller liquidity constraints than domestic funds, because international funds have more investment opportunities. Furthermore we run regressions on a sample, which excludes sector funds, property funds and sustainability funds, because these funds are less diversified which will possibly influence the analyses.
Analyses with gross returns, show no noteworthy differences for both the regressions with Fama and French alphas as well for the regression with five-factor alphas. Relations between expense ratio and before-fee performance remain insignificant, for the regression with Fama and French alphas as well for the regression with five-factor alphas as independent variables. So there is no evidence that funds with higher expense ratios, find more underpriced securities than funds with lower expense ratios. This finding contradicts with the theory of Grossman and Stiglitz (1980) that funds will be fully compensated for their costs, and with the theory that funds invest their money efficiently, as proposed by Ippolito (1989). We suggest that the insignificant relation is due to the fact that expense ratio contains for a large part of e.g. marketing fees, which adds no value to investors. Furthermore Gil-Bazo and Ruiz-Verdú (2009) argued that fees do not necessarily represent the services provided to investors, but also are the result of strategic fee setting of funds.
The regression with five-factor alphas on the sample of domestic funds, only results in one significant coefficient, for fund age (positive).
Table 5: Results of the robustness analyses
This table contains the results of the robustness analyses with the following model ai,t= a0+β1 Persistencei,t+β2 ln(Size)i,t-1+β3 ln(Age)i,t-1+β4
Turnoveri,t-1+β5 Flowi,t-1+β6 ln(Exp)i,t-1+β5 ln(Positions)i,t-1+λt+vi,t. We perform the analyses on different (sub)samples as specified on the first row.
On the second row, the independent variable of the regressions is specified. Fama and French alphas and five-factor alphas are estimated with equation 2 and 3, respectively. Gross returns are estimated by adding back the expense ratio to the estimated performance. The last row contains the effects specification of the regressions. As shown, for the regressions on the sample of domestic funds only, we apply no time varying intercept (λt). The table also contains the number of cross-sections, and the observations. *** Indicates significance at the 1% level, ** indicates significance
at the 5% level and * indicates a significance at the 10% level.
Sample of domestic Sample of Sample of Sample without sector-, sustainability
and international funds domestic funds international funds and property funds
Independent variable: Five-factor alpha Gross Fama and French alpha Gross five-factor alpha Fama and French alpha Five-factor alpha Fama and French alpha Five-factor alpha Fama and French alpha Five-factor alpha Cross-sections: 81 81 81 11 11 70 70 54 54 Observations 278 278 278 39 39 239 239 188 188
V. Conclusion
In this study we examine the relation between fund performance and fund characteristics. More specifically, we considered age, size, holdings, flow, turnover ratio, expenses ratio and persistence as determinants of future mutual fund performance. The sample of this study consists of 81 Dutch domestic and international open-end active mutual funds. Data about fund characteristics are collected over the period 2005 to 2008 and are one-year lagged to fund performance. Fund performance is estimated with a Fama and French three-factor model with country/region-specific benchmarks. Relations between fund performance and fund characteristics are estimated with a time-fixed effects model.
performance. The founded significant negative relation between fund performance and holdings. This evidence indicates that focused funds outperform diversified funds and refutes the general idea that focused funds suffer from underperformance as a consequence of price impact and agency conflicts. Results are robust for the analysis on domestic funds, when we measure fund performance as five-factor alphas we do not find a significant coefficient for age.
Robustness analysis show further that turnover is negatively related to fund performance of international funds. However, this relations is not supported by the regression with five-factor alphas. No evidence is founded that performance persist over time. Also we find that expense ratio has no explanation power on fund performance. Even before-fee performance show no relation with expense ratio. These result show no supporting evidence that funds with higher priced management have better stock picking abilities than funds with lower expense ratios .
References
Books
Brooks, C., 2004, Introductory Econometrics for Finance, Cambridge University Press, first edition.
Brooks, C., 2008, Introductory Econometrics for Finance, Cambridge University Press, second edition.
Journals
Brennan, M., and P. Hughes, 1991, Stock prices and the supply of information, Journal of Finance, 46, 1665-1691.
Carhart, M., 1997, On persistence in mutual fund performance, Journal of Finance, 52, 57-82.
Chen, J., H. Hong, M. Huang, and J. Kubik, 2004. Does fund size erode performance? Liquidity, organizational diseconomies and active money management, American Economic Review, 94, 1276-1302.
Chevalier, J. A., and G. D. Ellison, 1997, Risk taking by mutual funds as a response to incentives, Journal of Political Economy, 105, 1167–1200.
Cremers, M., and A. Petajisto, 2008, How active is your fund manager? A new measure that predicts performance, Review of Financial Studies, 22, 3329-3365.
Droms, W.G., and D. Walker, 1994, Investment performance of international mutual funds, Journal of Financial Research, 27, 1-14.
Edelen, R.M., R. Evans, and G.B. Kadlec, 2007, Scale effects in mutual fund performance: The role of trading costs, working paper, Boston College.
Elton, E.J., M.J. Gruber, J. Martin, S. Das, and M. Hlavka, 1993, Efficiency with costly information: a reinterpretation of evidence from managed portfolios, Review of Financial Studies, 6, 1-22.
Elton. E.J., M.J. Gruber, and C.R. Blake, 1996, The Persistence of Risk-Adjusted Mutual Fund Performance, Journal of Business, 69, 113-157.
Fama, E.F., 1970, Efficient Capital Markets A Review Of Theory And Empirical Work, Journal of Finance, 25, 383-417.
Fama, E., and K. French, 1992, The cross-section of expected stock returns, Journal of Finance, 47, 427-465.
Ferreira, M. A., A.F. Miguel and S.B. Ramos, 2009, The Determinants of Mutual Fund Performance: A Cross-Country Study, Working paper, ISCTE Business School.
Gil-Bazo, J., and P. Ruiz-Verdú, 2009, The Relation between Price and Performance in the Mutual Fund Industry, Journal of Finance, 64, 2153-2183.
Grinblatt, M., and S. Titman, 1992, The persistence of mutual fund performance, Journal of Finance, 47, 1977-1984.
Grinblatt, M., and S. Titman, 1994, A study of monthly mutual fund returns and portfolio performance evaluation techniques, Journal of Financial and Quantitative Analysis, 29, 419-444.
Grossman, S., and J. Stiglitz, 1980, On the impossibility of informational efficient markets, American Economic Review, 70, 393-408.
Gruber, M., 1996, Another puzzle: The growth in actively managed mutual funds, Journal of Finance, 51, 783-807.
Harlow, W. V., and K.C. Brown, 2005, The Right Answer to the Wrong Question: Identifying Superior Active Portfolio Management, Journal of Investment Management, 34, 15-40.
Hendricks, D., J. Patel, and R. Zeckhauser, 1993, Hot hands in mutual funds: Short-run persistence of relative performance 1974-1988, Journal of Finance, 48, 93-130.
Ippolito, R., 1989, Efficiency with costly information: A study of mutual fund performance, Quarterly Journal of Economics, 104, 1-23.
Jegadeesh, N., and S. Titman, 1993, Returns to buying winner and selling losers: Implications for stock market efficiency, Journal of Finance, 48, 65-91.
Kacperczyk, M., C. Sialm, and L. Zheng, 2005, On the industry concentration of actively managed equity mutual funds, Journal of Finance 60, 1983–2011.
Prather, L., and W.J. Bertin, 2004, Mutual Fund Characteristics, Managerial Attributes, and Fund Performance, Review of Financial Economics, 13, 305-326.
Sharpe, W. F., 1966, Mutual fund performance, Journal of Business, 39, 119-138.
Wermers, R., 2000, Mutual fund performance: An empirical decomposition into stock-picking talent, style, transactions costs, and expenses, Journal of Finance, 55, 1655–1695.
Yan, X., 2008, Security concentration and active fund management: Do focused funds offer superior performance, The Financial Review, 43, 27-49.
Appendix
Table A: Results of the Hausman test
This table shows the results of the Hausman test. ***Indicates significance at the 1% level.
Test Summary Chi-Sq. Statistic Chi-Sq. d.f. Prob. (P )
Cross-section random 33.93 7 0.00***
Table B: Redundant fixed effects-likelihood tests
This table shows the results of the redundant fixed effects-likelihood ratio tests. ***Indicates significance at the 1% level.
Effects Test Statistic d.f. Prob. (P.)
Cross-section F 1.06 -80,19 0.36 Cross-section Chi-square 104.15 80 0.04 Period F 11.10 -3,19 0.00*** Period Chi-square 45.56 3 0.00*** Cross-Section/Period F 1.55 -83,19 0.01*** Cross-Section/Period Chi-square 145.27 83 0.00***
Table C: Overview of the funds included in the sample
The funds that are included in the sample and the time period of the collection of fund characteristics are in this table. In case funds use a financial year which not coincides with calendar year then this is specified.
Time period of the collection of fund characteristics
Number Fund name Start End
Domestic funds
1 BNP Paribas Netherlands Fund 2005/2006 2008/2009
2 BNP Paribas Small Companies Netherlands Fund 2005/2006 2008/2009
3 Allianz Holland fund 2004/2005 2007/2008
4 Avero Achmea Nederland Aandelenfonds 2005 2006
5 Delta Lloyd Deelnemingen Fonds 2005 2008
6 Delta Lloyd Nederland Fonds 2005 2008
7 ING Dutch Fund 2005 2008
8 Kempen Orange fund 2005 2008
9 Kempen Oranje Participaties 2005 2008
10 Robeco Hollands Bezit 2005 2008
11 SNS Nederlands Aandelenfonds 2005 2007
International funds
12 BNP Paribas Asia Pacific High Income Equity Fund 2007/2008 2008/2009
13 BNP Paribas Global Property Security Fund 2005/2006 2008/2009
14 BNP Paribas High Income Equity Fund 2005/2006 2008/2009
15 BNP Paribas High Income Property Fund 2005/2006 2008/2009
16 BNP Paribas Property Securities Fund America 2005/2006 2008/2009
17 BNP Paribas Property Securities Fund Europe 2005/2006 2008/2009
18 BNP Paribas Property Securities Fund Far East 2005/2006 2008/2009
19 Allianz Holland Amerika Fonds 2004/2005 2007/2008
20 Allianz Holland Europe Fonds 2004/2005 2007/2008
21 Allianz Holland Pacific Fonds 2004/2005 2007/2008
22 ASN Aandelenfonds 2005 2008
23 ASN Milieu & Waterfonds 2005 2008
24 ASN Small & Midcap Fund 2007 2008
25 Avero Achmea Euro Aandelenfonds 2005 2006
26 Avero Achmea Noord Amerika Fonds 2005 2006
27 Delta Lloyd Donau Fonds 2005 2008
28 Delta Lloyd Europa Fonds 2005 2008
29 Delta Lloyd Investment Fund 2005 2008
30 Delta Lloyd Jade Fonds 2005 2007
31 Delta Lloyd select Dividend Fonds 2006 2008
32 Friesland Bank Aandelen Fonds 2005 2008
33 ING Basic Materials Fund 2005 2008
34 ING Biotechnology Fund 2005 2008
35 ING Daily Consumer Goods Fund 2005 2008
37 ING Emerging East Europe Fund 2005 2008
38 ING Energy Fund 2005 2008
39 ING Europe Fund 2005 2008
40 ING Europe Growth Fund 2005 2008
41 ING Europe Small Caps Fund 2005 2008
42 ING Far East Fund 2005 2008
43 ING Financials Fund 2005 2008
44 ING Global Emerging Markets Fund 2005 2008
45 ING Global Fund 2005 2008
46 ING Global Opportunities 2005 2008
47 ING Global Real Estate Fund 2005 2008
48 ING Health Care Fund 2005 2008
49 ING Hoog Dividend Aandelen Fonds 2005 2008
50 ING Industrials Fund 2005 2008
51 ING Information Technology Fund 2005 2008
52 ING Internetfund 2005 2008
53 ING Japan 2005 2008
54 ING Luxury Consumer Goods Fund 2005 2008
55 ING North America Fund 2005 2008
56 ING Premium Dividend Fund 2006 2008
57 ING Telecom Services Fund 2005 2008
58 ING Utilities 2005 2008
59 Kempen Best Selection European Property Fund N.V. 2004/2005 2007/2008
60 Kempen European Smallcap fund 2004/2005 2007/2008
61 kempen Sense Fund 2004/2005 2007/2008
62 Ohra Aandelen Fonds 2005 2008
63 Ohra Care Fonds 2005 2006
64 Ohra Communicatie Technology Fonds 2005 2006
65 Ohra Internet Fonds 2005 2006
66 Ohra Medical Technology Fonds 2005 2006
67 Ohra Milieu Technologie 2005 2008
68 Ohra Multimedia Fonds 2005 2006
69 Ohra New Energy 2005 2008
70 Ohra Onroerend Goed fonds 2005 2008
71 Robeco 2006 2008
72 Robeco Duurzaam Aandelenfonds 2007 2008
73 Robeco Young Dynamic 2006 2008
74 SNS Amerika Aandelenfonds 2005 2008 75 SNS Azie Aandelenfonds 2005 2008 76 SNS Duurzaam Aandelenfonds 2005 2008 77 SNS Euro Aandelenfonds 2005 2008 78 SNS Euro Vastgoedfonds 2005 2008 79 SNS Hoogdividend Aandelenfonds 2005 2008
80 Triodos Meerwaarde Aandelenfonds 2005 2008
Table D: Yearly estimated alphas over and for the period 2005 to 2009
This appendix shows the alphas estimated with the Fama and French three-factor model
(Ri,t-Rf,t=αi+β1,i*(Rm,t-Rf,t)+β1,i*SMBt+β2,i*HMLt+εi,t,) and the alphas estimated with the five-factor model
(Ri,t-Rf,t=αi+β1,i*(RNL,t-Rf,t)+β2,i*(REUR-NL,t-Rf,t)+β3,i*(RNA,t-Rf,t)+β4,i*(REM,t-Rf,t)+β5,i*(RASIA,t-Rf,t)+εi,t,).
Alphas are reported as yearly averages, these are obtained by multiplying the estimated weekly averages by 52. ***Indicates significance at the 1% level.
2006 to 2009 2005 2006 2007 2008 2009 Fama and French alpha Five- factor alpha Fama and French alpha Five- factor alpha Fama and French alpha Five- factor alpha Fama and French alpha Five- factor alpha Fama and French alpha Five- factor alpha Fama and French alpha Five- factor alpha Mean -5.67 -6.06 7.33 -4.99 -2.08 -3.02 -2.55 -8.63 -21.94 -11.96 3.74 -1.35 Median -3.93 -6.36 7.75 1.77 -3.17 -2.76 -2.08 -8.89 -18.98 -10.66 2.08 -0.78 Maximum 56.10 30.33 32.29 23.19 18.88 23.27 44.82 24.28 9.62 17.21 56.11 30.32 Minimum -71.25 -62.07 -13.36 -574.91 -24.65 -33.07 -36.76 -34.22 -71.24 -62.09 -29.12 -24.96 Std. Dev. 16.11 11.83 8.42 66.352 9.31 10.82 13.36 9.20 14.51 12.38 12.43 12.01 Skewness -0.45 -0.35 0.25 -8.382 0.39 -0.14 0.35 0.20 -1.18 -1.41 1.05 0.31 Kurtosis 5.41 5.70 3.24 72.52 2.93 3.13 5.02 4.93 5.05 7.74 6.79 2.76 Jarque-Bera 76.73 90.28 1.18 16408.87 2.04 0.31 15.38 13.10 29.64 92.44 56.22 1.31 Probability 0.00*** 0.00*** 0.56 0.00*** 0.36 0.86 0.00*** 0.00*** 0.00*** 0.00*** 0.00*** 0.52 Observations 278 278 77 77 80 80 81 81 73 73 72 72
Table E: Descriptive statistics using logarithms
This table shows the descriptive statistics of the mutual fund characteristics variables and the persistence variable, for all funds. In comparison to descriptive statistics reported in Table 3, now we use the natural logarithm for fund age, number of holdings, fund size and TER. Data are measured as yearly average over the period 2005 to 2008, using year-end data. Despite the improved Jargue-Beras for the natural logarithm values of fund characteristics, only fund size is not longer significant at the 1% level. *** Indicates significance at the 1% level.
Ln Age (years) Flow (%) Ln Holdings Ln Size (in € millions) Ln TER (%) Turnover (ratio)
