Master Thesis
Economies of Scale in the European Mutual Fund Industry: An
Analysis of 16 Countries
Jeroen de Mooij 2787717 MSc Finance University of Groningen Faculty of Economics and Business
Supervisor J.V. Tinang 12/06/2019 Word count: 14,559
Abstract
This study examines the relation between size and performance of actively managed mutual funds. To investigate this relationship, a unique panel dataset has been composed with data from 16 different countries over the period from March 1999 to December 2018. Several regressions have been estimated investigating the effect of fund size and fund family size on performance. The analysis provides statistically significant results suggesting there exist some returns to scale for Europe equity mutual funds. No evidence is found for diminishing returns to scale for domestic equity mutual funds. This study further includes a range of country characteristics to control for differences between countries.
Keywords: Investment Strategies, Asset Allocation, Mutual Funds, Economies of Scale JEL: C33, G10, G23
2
I.
Introduction
Mutual funds are pooled investments providing liquidity and enabling investors to enjoy economies of scale from low-cost diversified portfolios which often are differentiated by fund styles (Cuthbertson et al., 2016). This allows an investor to gain exposure to particular characteristics. Examples of fund styles are small companies, equity-income, growth, growth and income and lastly aggressive growth. To compensate for their higher fees, compared to index or tracker funds, most funds are active. Rather than replicating the performance of the overall market, active funds research individual companies and developments in the overall markets (using macroeconomic indicators) and in this way attempt to create value for their investors by picking stocks and engaging in market timing.
According to the Investment Company Institute (2018), total worldwide assets invested in exchange-traded-, institutional- and mutual funds was 49.3 trillion dollars as of year-end 2017, up from 29.0 trillion dollars in 2010. Similarly, worldwide assets of these funds invested in equity increased to 21.8 trillion dollars as of year-end 2017, up from 11.9 trillion dollars in 2010. In the US, mutual fund total net assets were almost 19 trillion dollars of which more than half was invested in equity. Furthermore, 45.4% of US households own mutual funds. This widespread ownership of, and exposure to, mutual funds has led to considerable interest in mutual fund performance, not least in the academic literature (Cuthbertson et al., 2016). This research, however, has strongly focused on the US market. Research looking at European mutual funds has been sparse. This while funds in Europe are responsible for large amounts of assets under management (AUM). 36% of total worldwide assets invested in the aforementioned funds were under management by European funds, with 17.7 trillion dollars in total net assets compared to 10.9 trillion dollars in 2010. The percentage of households’ financial wealth invested in funds is much lower in the European Union at 8% compared to 23% in the US, but a substantial amount nonetheless. The increasing size of the mutual fund industry in Europe and the very large amount of available funds to choose from has made the decision which mutual fund to invest in more difficult. Investors will thus spend considerable resources to identify their best option.
Furthermore, developments in information technology have substantially reduced the barriers to deposit savings into the securities markets as well as into mutual funds. Investors who feel constructing portfolios of equities themselves is too difficult or time consuming will likely consider either index tracking instruments or mutual funds. Investors in mutual funds will rationally attempt to identify those mutual funds which will provide them with the best (risk-adjusted) returns.
3 and research expenses. These effects decrease average unit costs from increased production volume. Negative scaling effects come from organizational inefficiencies (Chen et al., 2004). Berk and Green (2004) furthermore on the basis of their model predict that funds with high returns during previous years attract disproportionate cash inflows and those funds are subject to diminishing returns as high fund returns cannot expand infinitely. At some point, the marginal dollar is invested in index funds when the marginal cost of finding undervalued securities is equal to the increased return on the securities. At that point, active portfolio management is unable to generate excess returns. Their model therefore hinges on the diminishing returns to scale being prevalent in the mutual fund industry.
Furthermore, the relationship between fund family size and performance is investigated. A fund family is a group of funds that are part of the same fund company. Examples of fund families are Vanguard funds and JP Morgan funds. Ferreira et al. (2013) find that fund performance improves with the size of its fund family, substantial economies in trading commissions and lending fees exist benefitting large fund families.
Even though the asset management market in the US is much bigger than that in Europe, the number of funds available for sale is about a third of the number of funds in Europe, while the average AUM per fund is roughly six times higher (Glow, 2018). The characteristics of the mutual fund industries are therefore different meaning that findings with respect to economies of scale of the US mutual fund industry might not carry over to the mutual fund industry in other countries. Ferreira et al. (2012) confirm this as they find that there are flow-performance differences across countries as a result of investor behavior in buying and selling mutual funds. Differences in regulation between the US and European countries (MiFID) might also mean US findings on economies of scale are not generalizable to Europe.
The effect of past fund and fund family size on performance has been examined using panel-data to run regression on a large sample of domestic equity mutual funds and international equity mutual funds investing in Europe. The sample consists of more than 2000 open-ended actively managed equity funds in 16 countries over the period 1999-2018. Fund characteristics, as well as country characteristics, have been added in the regressions as control variables to support the validity of the results. No diseconomies of scale are observed for domestic equity mutual funds. For the European equity mutual funds, positive economies of scale are actually observed suggesting the cost savings outweigh the possible organizational disadvantages. There appears to be no significant relationship between fund family size and performance. Furthermore, liquidity is a relevant factor for domestic equity mutual funds. Domestic funds investing in small-cap stocks therefore suffer from the limits of their investment universe hurting their performance partially explaining why there are no observed economies of scale for domestic equity funds.
The findings further suggest younger funds are able to generate higher returns. A long track record therefore does not suggest higher performance. Funds charging high expense ratios are generally unable to make up for them in returns. It is also observed that domestic equity mutual funds perform better when they are managed in a team rather than by one manager. This suggests that organizational inefficiencies might not be as prevalent in Europe and the increase in human capital benefits performance in team-managed funds. Lastly, European equity funds that are sold in multiple countries show better performance. This could be explained by a more stable AUM base allowing it to hold less liquid assets.
4 (Becker and Vaughn, 2001; Brown et al., 1996). European investors living in one of the countries where the mean alpha is negative might want to reconsider investing in a mutual fund. They should also be aware that past performance likely doesn’t carry over into the future. For European legislators, it is good to know that efforts of making funds available in multiple countries are likely positive for the performance of mutual funds. Furthermore, legislators should be aware of the effects new regulation has on the dynamics in the mutual fund industry, especially whether it reduces competition and makes it difficult for smaller funds to compete against large funds from large fund families.
For decision-making authorities within the mutual fund (families), the findings of this research could also be useful. Team-managed domestic equity funds achieve significantly higher returns, having multiple managers in charge of a fund could therefore be a good idea. The evidence of positive economies of scale for the European equity funds suggests that having a few funds with a large amount of AUM could enhance performance through lower relative fixed labor costs, information gathering, administrative and research expenses.
This study adds to the existing literature by looking at the relationship between fund size and performance in Europe. While there are ample studies looking at this relationship using US funds and a number of studies looking at individual European countries, there is no cross-country study investigating specifically the relationship for Europe. This paper confirms that findings for the size-performance relationship in the US are not generalizable to the EU. The relationship between fund size and performance using a sample of domestic equity mutual funds is insignificant, while a sample of European equity mutual funds show that positive economies of scale exist in this industry. This is in contrast to recent evidence of studies using US fund samples that seem to point towards diseconomies of scale.
The paper is organized as follows. Section two provides a literature review. Section three covers the data, section four explains the methodology. Furthermore, section five discusses the results and, lastly, section six provides the conclusions of the paper.
II.
Literature review
A large amount of research has been done on mutual fund performance and the determinants of fund performance. Of course, mutual fund investors likewise want to identify those mutual funds that will provide them with high returns. Studies however have shown little evidence that mutual fund managers outperform passive benchmarks (Carhart, 1997; Gruber, 1996; Jensen, 1968 and Malkiel, 1995).
5 Larger funds are more likely to run into challenges in this respect as their large AUM means their own price impact is larger for the same fraction of total AUM invested in a stock. The marginal costs of actively managing funds become increasingly large decreasing expected excess returns. Fund managers also seem to be aware of the diminishing returns to scale as they might close the fund after strong fund inflows (Bris et al., 2006).
Chen et al. (2004) document that fund returns, both before and after fees and expenses, decrease with lagged fund size for their sample of US equity mutual funds, even after accounting for multiple performance benchmarks. Funds having to invest in small and illiquid stocks show the most pronounced association, which suggests these adverse scale effects are related to liquidity. Lastly, Chen et al. 2004 explore the idea that scale erodes fund performance as a result of the interaction of organizational diseconomies and liquidity.
Pollet and Wilson (2008) investigate the economies of scale in the mutual fund industry and find that funds are very reluctant to diversify in response to growth instead acquiring ever-larger ownership shares in the already owned companies. This appears to identify limits to the scalability of fund portfolios, such as liquidity constraints or price impact, as the proximate cause of diminishing returns to scale. Pollet and Wilson (2008) further provide evidence that diversification is associated with higher monthly risk-adjusted fund returns. Given the constraints of a fund style, holding small stakes in many stocks appears to be better for return than purchasing large stakes a few stocks. Small-cap sector funds benefit most from diversification with fund size being controlled for. The results support liquidity constraints as an explanation for why large-cap funds diversify more slowly in response to growth in AUM. Funds strongly constrained by high transaction costs, for example, small-cap funds will exhibit a positive association between diversification and subsequent performance, with fund size being controlled for. Funds less constrained by transaction costs, by contrast, will display a weaker relationship. Such funds are, for example, large-cap funds, funds with little AUM, or funds in large families benefiting from an improved trading environment. Their results further suggest that there are limits to the human capital that can be productively added to a fund. Berk and Green (2004) point out that there are also positive scale effects. Costs resulting from fixed labor, information gathering, administrative and research expenses are decreasing in average unit cost from increased production volume (in this case larger AUM).
Ferreira et al. (2012) show that marked differences exist in the flow-performance relationship across countries, which suggests that US findings concerning its convexity do not apply universally. Difference in behavior as a result of manager and investor sophistication means that flow-performance relationships of mutual funds might not be generalizable to other countries. This could potentially be extended to the size-performance relationship. The on average smaller size of funds in the European mutual fund industry compared to the US, manager behavior and investor sophistication differences could mean the observed diseconomies of scale in US mutual funds might not hold for European funds.
Lowenstein (1997) and Perold and Salomon (1991) also believe that a large asset base erodes performance of the fund because of trading costs associated with liquidity or price impact. This study will therefore test the liquidity hypothesis to determine the relevance of liquidity in explaining the relationship between fund size and performance.
6 strongly associated with the generation of additional investment ideas and rather than adding the investments to existing fund portfolios, new funds are created to hold them. These results on fund families are consistent with large fund families maintaining market share through having a broad range of different funds. This study will therefore also investigate the relation between fund family size and performance.
Control Variables
To appropriately determine the effect of fund size on performance, a number of characteristics variables should be controlled for in the regressions.
Fund age can be argued to be a determinant of performance and should therefore be controlled for. There are two lines of reasoning on how the age of a fund could influence its performance. Firstly, younger funds are likely more agile and eager to prove their legitimacy. At the same time, young funds may face higher costs and lack experience compared to older funds. Pástor et al. (2015) find a negative relationship between age and performance as do Ferreira et al. (2013) in their sample of non-US funds.
A fund’s expense ratio is the indirect measure of the value of the active management it should have. It’s the price paid to managers by investors to have their money invested. Khorana et al. (2008) find that funds charge higher fees in countries where there is limited investor protection. Carhart (1997) and Gil-Bazo and Ruiz-Verdu (2009) find a negative relationship between performance and expenses with samples of U.S. mutual funds. Dahlquist et al. (2000) also support this with evidence of Swedish mutual funds.
Total load is the sum of the initial and redemption charge. When investors buy into a fund they might have to pay an initial charge and when they want to sell their shares of the fund they might have to pay a redemption charge. High charges make it less attractive for investors to switch funds often. It gives mutual funds some more stability in their asset base allowing a riskier, less liquid portfolio. Previous research finds no significant relationship between total load and performance (Chen et al., 2004; Ferreira et al., 2013) or a negative relation (Carhart, 1997; Pollet and Wilson, 2008).
Gruber’s (1996) smart money hypothesis predicts investors are able to identify skilled managers and will invest in their funds. Fund flows could therefore have a positive relationship with future performance. This hypothesis is confirmed by Gruber (1996) and Zheng (1999) who find that funds experiencing net inflows perform significantly better than those experiencing outflows. Sapp and Tiwari (2004) argue that this observed effect is actually caused by momentum which wasn’t included in their models. Ferreira et al. (2013), for some specification of non-U.S. funds, do find evidence of the smart money hypothesis. The variable last year flow will therefore be added to test whether the smart money hypothesis is supported for this sample where momentum is already controlled for.
Performance persistence, or the positive relation between past and current performance, has been investigated in a number of studies. Studies employing US mutual fund samples find evidence of performance persistence (Brown and Goetzmann, 1995; Carhart, 1997; Ferreira et al., 2013; Grinblatt and Titman., 1994; Hendricks et al., 1993). Outside the US, however, Dahlquist et al. (2000) find no evidence of performance persistence in Sweden and Otten and Bams (2002) only for UK funds. Performance persistence will be tested for with a variable measuring average risk-adjusted return over the last 12 months.
7 The dummy variable management team could affect fund performance through organizational efficiency. Chen et al. (2004) discuss the implications of hierarchy costs on fund performance. They argue that in large organizations with hierarchies, the process of agents fighting for their ideas to be implemented will affect their ex-ante decisions of what ideas they want to work on. Stein (2002) argues that for organizations in which the processing of soft information is important, smaller might be better. This can be related to mutual funds deciding which stocks to invest in. When a fund has multiple managers, one manager might need to expend larger effort to prove the merits of an investment based on objective information, which might be an inefficient use of resources. Chen et al. (2004) find that solo managed funds are significantly more likely to invest in local stocks compared to team managed funds and do better at picking them. Lastly, they confirm the hypothesis that solo managed funds outperform team managed funds after controlling for fund size.
III.
Methodology
Performance measurement
Mutual fund performance is measured using risk-adjusted returns in Euro. To measure the actual skill of funds to generate a risk-adjusted return, or alpha, performance benchmarks have been used that control for widely accepted factors that affect returns.
Carhart’s (1997) four-factor model has been employed to compute the alphas. These alphas serve as the main measure of performance and are the dependent variable in most of the regressions. Two other benchmarks have been added to confirm the robustness of the findings. Risk-adjusted performance is therefore also determined using the capital asset pricing model and Fama-French’s three-factor model.
The following four risk factors are used; Market premium (MKT), small-minus-big (SMB), high-minus-low (HML), and prior 1-year momentum (MOM), the first two introduced by Fama and French (1993) and the last from Carhart (1997). MKT is the excess return on a value-weighted aggregate market proxy of European stocks. SMB is the difference in return of holding a small capitalization stock portfolio compared to a large capitalization stock portfolio. HML is the return difference between holding a high book-to-market equity portfolio and a low book-to-market equity portfolio. MOM is the return for holding a portfolio of stocks with high returns in the previous year over a stock portfolio with low returns in the previous year.
Performance Benchmarks
As discussed earlier, three approaches are used for calculating risk-adjusted performance. The first approach is the capital asset pricing model. The capital asset pricing model equation is the following:
(𝑅𝑖,𝑡− 𝑅𝑓,𝑡) = 𝛼𝑖+ 𝛽𝑖𝑀𝐾𝑇𝑡+ 𝜀𝑖,𝑡,
Where 𝑅𝑖,𝑡 is the mutual fund return net of expenses (calculated using net asset value (NAV)
and total expense ratio (TER) from the Lipper Mutual Fund database), 𝑅𝑓,𝑡 is the risk-free rate,
𝛼𝑖 is the excess return of the fund, 𝛽𝑖 is the beta of the fund measuring the systematic risk/
exposure to the market and 𝜀𝑖,𝑡 is a generic error term which is uncorrelated with all other
independent variables.
The second approach adds two factors, small-cap minus big cap (SMB) and high book-to-market value equity minus low book-to-book-to-market value of equity (HML). This is the Fama-French three-factor model:
8 The Carhart four-factor model adds one more factor, momentum, being prior 1 year winners minus prior 1 year losers (MOM). The full Carhart four-factor model is thus:
(𝑅𝑖,𝑡− 𝑅𝑓,𝑡) = 𝛼𝑖+ 𝛽𝑖,1𝑀𝐾𝑇𝑡+ 𝛽𝑖,2𝑆𝑀𝐵𝑡+ 𝛽𝑖,3𝐻𝑀𝐿𝑡+ 𝛽𝑖,4𝑀𝑂𝑀𝑡+ 𝜀𝑖,𝑡
The alpha is computed for each fund using monthly data. I employed a rolling window of the previous 36 months of data on the factors and returns to compute this alpha (24 months of return data is required at a minimum). This means that every following month the oldest return (t-37) is dropped and past month’s return (t-1) is added. Fund alpha is therefore only available for those funds with return data over a period of at least 37 months. Important to note is that the consequence of using this approach is that, although data has been collected from March 1999 to December 2018, my first estimate of a fund’s alpha is April 2002 as this is the first month providing us with enough data to estimate it. This approach is in line with Ferreira et al. (2013) their approach but differs in that they estimate alpha per quarter rather than monthly. Chen et al. (2004) instead sort by fund size and create five fund size portfolios. Alpha’s are then calculated per quintile and subsequently assigned to all funds in that quintile.
According to Chen et al. (2004), the technique employed in this paper is not as good as their main approach, they argue estimating loadings per fund leads to loadings that tend to be noisy. They note however that their results found are the same using either approach.
Regressions
Panel regressions will be run to investigate how fund performance varies with lagged fund size. As it is reasonable to expect a correlation of fund size with other fund characteristics, the (risk-adjusted) returns are regressed on fund size and a set of observable fund characteristics. Fund characteristics included in the regression are age, expense ratio, total load, past-year returns, past-year fund inflows, organization of management and number of countries in which a fund is sold. The explanatory variables are lagged one month. All regressions are estimated with time-fixed effects to control for those events affecting all funds at the same time. Regressions that don’t have country characteristics added are also estimated with both time-fixed effects and country-fixed effects.
Total net assets (TNA) are used to determine the size of a mutual fund, of which I will take the log to proxy for fund size in the regressions.
To investigate the effect of family size on performance to answer our second research question, a variable is added to the regression. The fund company variable in Lipper provides data on the company each fund belongs to. Fund companies with multiple branches listed separately have been consolidated into one fund company. Fund family is thus the collection of funds that are sponsored by one fund company. Size of the fund family, FAM, is defined as the aggregated TNA of all funds in the fund’s family with the TNA of the particular mutual fund substracted. The log of the combined TNA of all mutual funds in the family minus the TNA of the mutual fund itself is used to proxy for this.
𝐹𝐴𝑀𝑖,𝑡 = log(𝑇𝑁𝐴𝐹𝑎𝑚𝑖𝑙𝑦 𝑖,𝑡− 𝑇𝑁𝐴𝑚𝑢𝑡𝑢𝑎𝑙 𝑓𝑢𝑛𝑑 𝑖,𝑡)
Fund age constitutes the number of years the fund has been active/ years since inception. I expect a negative relation between fund age and performance as Ferreira et al. (2013) find younger funds to perform better than older funds for their sample of non-US funds. It is therefore expected that the agility and commitment of younger funds has a stronger positive effect on performance than the negative effect of higher costs and lack of experience that these funds might face. Pástor et al. (2015) also support this hypothesis, who find that younger funds outperform older funds in a typical month suggesting younger funds are more skilled.
9 Ferreira et al., 2013). Funds might charge higher expenses because they spend more on marketing to less informed investors. While funds charging higher expenses might be able to generate higher returns, they are generally unable to cover them. I therefore expect a negative sign for expense ratio in the regressions.
Total load is the sum of front-end and back-end load. Funds charge loads because they want to dissuade investors from switching between funds or redeeming their investments. By charging loads there is less volatility of the AUM, allowing funds to hold less in cash and invest in more risky portfolios. Carhart (1997) and Pollet and Wilson (2008) find a negative relation between loads and performance while others find no significant relation (Chen et al., 2004; Ferreira et al., 2013). I however expect a positive relation between total load and performance however based on the arguments mentioned above.
Fund flow, F, is calculated as 𝐹𝑖,𝑡=
𝑇𝑁𝐴𝑖,𝑡−𝑇𝑁𝐴𝑖,𝑡−1∗(1+𝑟)𝑖,𝑡
𝑇𝑁𝐴𝑖,𝑡−1 ,
Where TNAi,t is the total net assets in month t and TNAi,t-1 is total net assets of the previous
year (month t – 12) and (1+r)i,t the return of the fund between months t – 12 and t. The
corresponding variable in the regressions is last year flow. This variable tests the “smart money” hypothesis of Gruber (1996) which is the hypothesis that investors are able to detect skilled managers and will invest in those managers’ funds. Gruber (1996) and Zheng (1999) confirm the smart money hypothesis, Sapp and Tiwari (2004) however argue that the observed effect is explained by momentum. Ferreira et al. (2013) confirm the smart money hypothesis for their non-US fund sample even though they do control for momentum by using the Carhart model. I however hypothesize that there might be a negative relation between fund flow and performance following from the assumption that funds receiving large inflows won’t invest the inflows as efficiently as a fund with lower inflows. They likely aren’t able to generate enough attractive investment opportunities leading them to invest the inflows into already held securities or closet indexing. Closet indexing is the practice of pretending to actively manage but actually investing passively, in the overall market.
To test for performance persistence, past performance is included as a fund characteristic. Performance persistence is calculated as the average alpha over the last twelve months. While previous research using samples of US mutual funds seem to support evidence of performance persistence (Brown and Goetzmann, 1995; Grinblatt and Titman, 1994; Hendricks et al., 1993), research using samples of non-US funds is less convincing. Otten and Bams (2002) only find performance persistence for UK funds, while Dahlquist et al. (2000) find no evidence of performance using a Swedish sample and Ferreira et al. (2013) find only weak evidence. Weak evidence therefore suggests a positive relation between past and current performance. The organization of management could affect performance. The variable “Team managed” is a dummy variable which equals one when a fund is managed by multiple managers and zero when the fund is solo-managed. Chen et al. (2004) argue that there are organizational diseconomies in team-managed funds following mainly from hierarchy costs. They argue soft-information is important in mutual funds and as soft-soft-information isn’t easily transferred to others, managers will expend too much effort convincing others of the merits of their ideas. Evidence from mutual fund studies using US samples confirms this hypothesis (Chen et al., 2004; Massa et al., 2010). Bliss, Potter and Schwarz (2008) find no significant relation, Ferreira et al. (2013) find a negative relation for both their US sample and non-US sample.
10 The regression specification utilized to answer the main research question is
𝛼𝑖,𝑡= 𝜇 + 𝜙𝐿𝑂𝐺𝑇𝑁𝐴𝑖,𝑡−1+ 𝛾𝑋𝑖,𝑡−1+ 𝜖𝑖,𝑡 𝑖 = 1, … , 𝑁 ,
Where 𝛼𝑖,𝑡 is the risk-adjusted return of fund i in month t given by the performance benchmarks,
𝜇 is a constant, 𝐿𝑂𝐺𝑇𝑁𝐴𝑖,𝑡−1 is the measure of fund size, and 𝑋𝑖,𝑡−1 is the set of control
variables in month t-1 that includes the variables discussed above (age, expense ratio, flow, family size, past performance). 𝜀𝑝,𝑡 is the generic error term. 𝜙 is the coefficient of interest
capturing the relationship between fund size and fund performance with the fund characteristics being controlled for. 𝛾 is the vector of loadings on the control variables. To test the liquidity hypothesis, the regression is augmented by including a dummy indicator for loading above the median on SMB and an interaction term of LOGTNA with this dummy indicator.
Country Characteristics
Further, a number of country characteristics are added as a fund’s domicile country characteristics could play some role in explaining the performance of a fund.
Firstly some country characteristics proxying economic development are included.
GDP per capita is a proxy for the income of a country’s inhabitants. In countries with higher GDP per capita, it can be expected that there is more demand for mutual fund services. Furthermore, two proxies for the level of education in a country have been included.
Tertiary education measures the enrolment ratio to the age group corresponding to this level of education. The variable education is the percentage of a country’s GDP that is spent on education. Countries where the enrolment ratio is high can be expected to have more high human capital individuals, that could both function as fund managers or are more sophisticated investors pressuring fund managers to perform better. Countries in which relative expenditure on education is higher could have individuals that are better able to determine the quality and value of mutual funds also providing funds pressure to develop sophisticated strategies. Internet usage is the percentage of a country’s population that uses the internet. In countries where there is a higher share of people using the internet, more investors can be expected to do research on funds pressuring them to develop sophisticated strategies and to generate competitive returns.
Gross savings represents the difference between consumption and disposable income. In countries where the level of gross savings is higher, the mutual fund industry can arguably be expected to be larger.
Financial development of countries will also be controlled for in the sample of domestic equity mutual fund, these variables are therefore less relevant for European equity mutual funds and will therefore not be included in regressions employing that sample. Market capitalization/GDP measures the relative importance of equity markets for firms in a particular country. Publicly listed companies from those countries are therefore larger contributors to GDP. It can be argued that mutual funds in those countries might also play a larger role in capital allocation to companies.
The share turnover ratio of a country is a proxy for the liquidity and stock market development. It is defined as the value of stocks traded divided by market capitalization. For domestic equity funds, more liquidity translates into lower transaction costs which could positively affect performance of mutual funds.
11 Contract enforcement score measures the difficulty to have a legal contract enforced/ resolving a commercial dispute through court. It is a proxy for the quality and efficiency of the court system.
Minority shareholder protection score measures the strength of minority shareholder protections against misuse of corporate assets by directors.
For both these legal measures, it is expected that a higher score benefits mutual fund
investors because they are better protected against misconduct. I therefore expect a positive relation between the contract enforcement score and the minority shareholder protection score with mutual fund performance.
The last country characteristic is a proxy for mutual fund industry concentration. This variable measures the mutual fund industry Herfindahl index. It is the sum of squared market shares of parent management companies (fund families) for equity funds in each country. In a country where there is higher competition, the Herfindahl variable should be lower. Higher competition between fund families could be expected to improve fund performance as there is more pressure to generate high fund returns to avoid investors switching funds. A negative relation between the Herfindahl index and performance is therefore expected.
IV.
Data
Data sourcesFund data has been collected from the Thomson Reuters Lipper mutual fund database. Monthly data has been collected from March 1999 to December 2018. The data includes both active and defunct funds and is therefore survivorship-bias free.
There are two main samples used for regression, those are domestic funds and funds investing in European countries. Domestic funds are those funds that are investing solely in the country in which they are domiciled. Funds investing in European countries are funds domiciled in one of the 16 countries included in the sample that are investing in equities from European countries. All of the countries included are listed in Table 2. To determine this, variables domicile and geographic focus have been used, collected from the Lipper database. The domestic fund sample consists of funds from 16 countries that have as geographic focus their domicile country. The European sample consists of funds domiciled in one of the 16 countries, with geographic focus Europe, Eurozone, Europe Sm&Mid Cap or Europe exc UK. In total, slightly under a million firm-month observations were initially collected. The two samples contain roughly the same amount of funds.
Only equity investing mutual funds have been included. Bond, commodity, index, money market mutual funds and fund of funds have therefore been excluded. For each fund, I collected data on total net assets (TNA, also referred to as assets under management), net asset value (NAV), age, total expense ratio and TNA of the family the fund belongs to (TNA of the fund itself subtracted), front- and backend loads, number of countries in which the fund is sold and fund managers. The expense ratio, flows and total loads have been winsorized at the top and bottom 1% level.
The variable NAV has been used to compute the fund return. The NAV accounts for capital gains, dividends (reinvested) and administrative fees (subtracted) and therefore allows us to appropriately determine the fund’s return.
12 the (partial) fund name, the currency of the fund, country in which fund is active, fund company (family), and lastly its portfolio managers. For the other relevant variables (total load, returns and the expense ratio) weighted average values have been calculated using the TNA as the weighting factor. For fund age, the earliest inception date among the fund classes has been used. After merging funds with multiple share classes and dropping observations with missing data, 243,874 firm-month observations remain across 2003 funds, this means on average about 10 years of historical monthly data is included for each fund. This panel dataset is therefore unbalanced.
To determine whether a fund is managed by a team or individually managed, the variable portfolio manager from Thomson Reuters Lipper has been employed. The variable portfolio manager is not very consistent, sometimes listing the names of multiple managers, name of the management team, “null” or “undisclosed” for missing values. Based on several filters the dummy variable team managed has been created, with a value of 0 for individually managed funds and 1 for team managed funds, like has been included in the research of Bär et al. (2011) and Han et al. (2008).
The variable total load follows from summing front- and back end load.
Monthly four factors employed to calculate alphas have been downloaded from Kenneth R. French’s website1, factors for Europe have been used for domestic equity as well as Europe
equity investing mutual funds.
To compare the results of using European factors to country-specific factors, German factors were also collected from Stehle2 to calculate alphas, results of this are discussed in the
robustness section. Stehle provides data until June 2016, the sample used to compare is therefore slightly shorter. Lastly, UK factors were collected as computed by Tharayan and Christidis3.
Furthermore, country characteristics have been collected from the WDI database and The World Bank. Variables included are: share of people having enrolled in tertiary education, GDP per capita, government expenditure allocated to education as a share of GDP, gross savings, internet usage as a percent of the population, market capitalization of companies (as a share of GDP), stock turnover ratio, contract enforcement, minority shareholder protection. As these databases provide yearly data, the country variable values in a particular year have been assigned to all of the month-observations. The regressions with country characteristics are therefore also regressed with monthly time-series data per fund.
Using Lipper mutual fund database variables, a Herfindahl index measuring the concentration of fund holdings by fund families within a country has also been computed. This has been done by firstly aggregating the total TNA of all funds of a fund family and dividing it by the TNA of all funds domiciled in that country giving us the market share of a fund family in a country. Each fund family’s squared market share was then aggregated again to arrive at the Herfindahl index of the country.
Contract enforcement is a proxy for the quality and efficiency of the legal court system. Minority shareholder protection is a variable that measures the protection of shareholders from conflicts of interest. Data for these two variables was collected from the World Bank4, who based this
latter measure on the methodology of Djankov, La Porta et al. (2008).
Table 1 lists the definitions of the fund characteristics and country characteristics used in this
1 https://mba.tuck.dartmouth.edu/pages/faculty/ken.french/data_library.html
2 http://www.wiwi.hu-berlin.de/professuren/bwl/bb/data/fama-french-factors-germany 3 http://business-school.exeter.ac.uk/research/centres/xfi/famafrench/files/
13 paper.
Table 1
Definitions of Variables Used
Descriptive statistics
14
Table 2
Number of Funds and Size of Mutual Funds by Country
This table shows the number of funds and total net assets (TNA) under management of the sample of funds at the end of 2017. The minimum fund size for the fund to be included in the sample is €10 million and the sample includes only open-end actively managed equity funds.
Country Number of Funds Average TNA (€ mil) Std. Dev. (€ mil) Min Fund TNA (€ mil) Max Fund TNA (€ mil) Austria 48 68 97 13 591 Belgium 65 146 210 10 1,120 Denmark 32 96 81 11 419 Finland 13 155 119 28 379 France 650 159 287 10 3,260 Germany 60 229 491 10 2,630 Ireland 151 144 209 11 1,340 Italy 21 205 221 11 680 Netherlands 13 398 396 49 1,340 Norway 43 267 340 13 1,620 Poland 15 92 88 11 274 Portugal 10 45 34 11 100 Spain 123 132 167 11 933 Sweden 95 595 845 10 4,710 Switzerland 226 225 662 10 6,790 UK 438 413 947 10 11,600 Total 2003 242
15
Table 3
Mutual Fund Characteristics by Country
This table reports means of fund characteristics. The sample includes open-end actively managed equity funds with data collected in the period from 1999 to 2018. See table 1 for variable definitions.
Table 4 displays the correlation matrix of all the variables included in the regressions in this paper. Correlations between the fund characteristics are acceptably low with most correlations staying below 0.10. The highest correlation among fund characteristics is the correlation between fund size (TNA (log)) and family fund size (Family TNA (log)). Some of the country characteristics do show substantial correlation between them, however this will be mitigated in the regressions by having separate regressions per country characteristic type.
Table 5 shows the means of country characteristics. GDP per Capita is a proxy for wealth which likely relates to more clients for mutual funds. Gross savings is a proxy for prudence and potential fund management market size. Education, tertiary education enrolment and internet usage can be interpreted as proxy for investor sophistication. Contract enforcement score is a proxy for the quality and efficiency of the legal court system. Minority shareholders protection score measures the protection of minority shareholders from misuse of corporate assets or misconduct by directors and are proxies for the legal environment. Market Cap/GDP displays the importance of publicly listed companies for the economy of the country. The stock turnover ratio shows the amount of activity in markets. Herfindahl measures the concentration of the mutual fund industry in a country by fund family.
16
Table 5
Country Characteristics
This table reports means of country characteristics. The sample period is from 1999 to 2018. Variable definitions can be found in table 1.
V.
Results
In this section, the empirical findings from the regressions are presented. I will compare the effects found between the lagged explanatory variables and the dependent variable in the regressions with the hypothesized relation. Explanatory variables are lagged by one month unless specified otherwise.
Table 6 shows the first set of regression outputs for the set of domestic equity mutual funds. The dependent variable is the percentage alpha return where alpha is calculated with the Carhart four-factor model. Estimation (1) includes time fixed effects, estimation (2) includes time fixed effects as well as country fixed effects. The standard errors are estimated correcting for fund clustering, which means I correct for intra-fund correlations that might overstate the effects of the explanatory variables.
The main independent variable of interest here is TNA (log), the log of total net assets, which allows us to determine whether there are diminishing returns to scale in mutual fund size. I find a positive but insignificant coefficient for fund size, which means there is no evidence that fund size affects performance for our sample of domestic equity funds.
Secondly, Family TNA (log) allows us to figure out whether the size of the fund “family” is related to mutual fund performance. The coefficient of family size is positive but again insignificant. There is no evidence that family fund size affects performance for the sample of domestic equity funds.
Fund age (log) is a control variable that tests whether a fund’s age affects the fund’s performance. I find a negative, significant effect on performance. Younger funds therefore appear to perform better than older funds for domestic equity funds.
17 to affect its performance. Funds that charge higher fees therefore don’t compensate for their higher cost in higher risk-adjusted return.
The variable total load is the sum of a fund’s initial and redemption charge. This charge is a one-time cost for entering or exiting the fund, respectively. The coefficient for total load is positive but insignificant, total load therefore does not appear to affect a fund’s performance. Last year flow measures a fund’s net inflows or outflows over the past twelve months, with its return controlled for. This variable measures the “smart money” effect. The coefficient for this variable is negative but insignificant, fund flows over the past twelve months do not appear to affect performance and I am therefore unable to confirm the “smart money” effect.
Last year alpha measures performance persistence. It is calculated as the cumulative alpha of the past twelve months. Last year alpha is negative but insignificant, there appears to be no performance persistence for our sample of domestic equity funds.
The variable team managed is a dummy variable, it’s value is 0 for solo-managed funds and 1 for funds managed by multiple people. The coefficient is positive for both estimations and significant at the 10% level for the regression including only time fixed effects. When country fixed effects are added the relation is no longer significant. This provides weak evidence that funds that are solo-managed perform worse compared to team-managed funds.
No. countries sold measures the number of countries in which a fund is made available for investors to invest in. I expected funds that were sold in more countries could perform better because of lower liquidity needs. The coefficient is negative but insignificant however, the amount of countries in which a fund is sold does therefore not appear to have an effect on its risk-adjusted return.
Table 6
Regression of Domestic Equity Mutual Fund Performance using Net Alphas This table reports panel regressions of the performance of open-end actively managed domestic equity funds in 2002-2018. The dependent variable is the monthly net Carhart model alpha after subtracting expenses (percentage per month), estimated using monthly fund returns in Euros. Explanatory variables include fund characteristics and time fixed effects, (2) further adds country-fixed effects. Variable definitions can be found in table 1. T-statistics are robust for fund-level clustering.
(1) (2)
Alpha Alpha
TNA (log) 0.016 0.006
(0.011) (0.011)
Family TNA (log) 0.015 0.012
(0.011) (0.013) Age (log) -0.087*** -0.092*** (0.031) (0.032) Expense Ratio -0.027 -0.028 (0.021) (0.022) Total Load 0.011 0.013 (0.009) (0.011)
Last Year Flow -0.001 -0.005
(0.030) (0.030)
Last Year Alpha -0.057 -0.073
(0.305) (0.306)
Team managed 0.056* 0.022
(0.030) (0.031)
No. Countries Sold -0.011 -0.011
(0.011) (0.011)
18
(0.194) (0.246)
Obs. 57135 57135
R-squared 0.195 0.213
Standard errors are in parenthesis *** p<0.01, ** p<0.05, * p<0.1
The regressions in table 7 are equivalent to table 6 but instead uses the sample of equity funds that invest not only domestically but also in other European countries. These regressions are reported separately because the difference in investment universe could affect the relations found between the fund characteristics and the fund’s performance. (1) again estimates the regression including time fixed effects, where (2) includes both time fixed effects as well as country fixed effects.
TNA (log), the measure of fund size, positively affects fund performance with a 1% significance level. It can therefore be concluded that larger funds investing in Europe are able to achieve better risk-adjusted returns. This is intuitively appealing because these funds have a far larger investment universe, meaning they have the advantages of being a large fund but don’t suffer the disadvantages that large funds forced to invest domestically do. Family TNA (log), the measure for family size, and fund age are again not significantly related to performance. Expense ratio is negatively related to fund performance and significant at the 1% level. This means that funds charging higher fees actually perform worse. This is in line with the hypothesis that funds charging higher expenses are generally unable to compensate for their expenses in alpha.
The coefficient of total load is positive, and significantly related to performance at the 10% level only for estimation (2). This means there is weak evidence that funds charging higher loads are able to achieve better risk-adjusted returns. This is in line with our hypothesis that these funds have lower liquidity allowing them more discretion in their investment strategies with less fear of large outflows.
The coefficient of last year flow is again negative but insignificant.
The relation between last year alpha and performance for the sample of Europe equity investing funds is negative and significant at the 1% level. The interpretation of this is that funds that have had comparatively high performance in the last 12 months generally exhibit performance reversal in the following month.
Whether a fund is solo- or team- managed appears, like for the sample of domestic equity funds, to have no significant effect on its performance.
The number of countries in which a fund is sold positively affects its performance at the 1% significance level in estimation (1), this relation however becomes insignificant after adding country fixed effects. I conclude that there is some support for the hypothesis that a fund available in multiple countries is able to achieve better risk-adjusted returns.
Table 7
Regression of Europe investing Equity Mutual Fund Performance
19
(1) (2)
Alpha Alpha TNA (log) 0.051*** 0.051***
(0.011) (0.012)
Family TNA (log) 0.001 -0.000
(0.007) (0.007) Age (log) 0.011 0.011 (0.039) (0.044) Expense Ratio -0.099*** -0.097*** (0.014) (0.014) Total Load 0.008 0.017* (0.009) (0.010)
Last Year Flow -0.078 -0.081
(0.066) (0.066)
Last Year Alpha -0.793*** -0.799***
(0.090) (0.089)
Team managed -0.003 -0.002
(0.026) (0.027)
No. Countries Sold 0.016*** 0.008
(0.004) (0.005)
Constant -1.295*** -1.247***
(0.220) (0.248)
Obs. 57897 57897
R-squared 0.553 0.559
Standard errors are in parenthesis *** p<0.01, ** p<0.05, * p<0.1
Liquidity Hypothesis
Table 8, compared to table 6 and 7, adds two more explanatory variables. It adds a dummy that equals one when the fund loads higher than the median on SMB, meaning the fund invests more in large-cap companies. Secondly, an interaction variable of this dummy with the log of total net assets (fund size) is added. Adding these two variables allows me to test the liquidity hypothesis. This hypothesis predicts liquidity constraints are important in explaining the lack of scale-ability of fund investments.
Estimation (1) and (2) employ our domestic equity fund sample while (3) and (4) employ our European equity fund sample. (2) and (4), compared to (1) and (3) add country fixed effects. After controlling for liquidity, the relation between fund size and performance is still insignificant for the sample of domestic equity funds. In estimation (2) however, the large-cap fund variable positively affects fund performance at the 10% level. The interaction variable is also significant at the 10% level but shows a negative relation. This supports the liquidity hypothesis, funds investing more in small cap companies are less able to generate high performance. Recall that performance here is computed using the Carhart four-factor model which controls for systematic risk, size, value-vs-growth and momentum. The negative relation of the interaction variable with performance is also in line with expectation. There is a less adverse effect of fund size on performance in large-cap funds compared to small-cap funds.
20
Table 8
Regression of Domestic Equity Mutual Fund Performance: Importance of Fund Size and Liquidity
This table reports panel regressions of the performance of open-end actively managed domestic equity funds in 2002-2018. The dependent variable is the monthly Carhart model alpha, estimated using monthly fund returns in Euros. Estimations (1) and (2) are with the sample of domestic equity funds while estimations (3) and (4) are with the sample of European equity investing funds. Estimations (1) and (3) are estimated including time-fixed effects, (2) and (4) further add country-fixed effects.
Explanatory variables include fund characteristics as well as two additional variables to test the liquidity hypothesis. Large Cap Fund is a dummy variable equal to 1 when a fund has a SMB loading higher than the median. Large Cap Fund* Log TNA is an interaction variable between the dummy and the log of total net assets. Variable definitions can be found in table 1. Standard errors are corrected for fund clustering and therefore more robust.
(1) (2) (3) (4)
Alpha Alpha Alpha Alpha
Domestic Europe
TNA (log) 0.018 0.022 0.046*** 0.044***
(0.014) (0.015) (0.012) (0.013)
Family TNA (log) 0.015 0.012 0.000 -0.002
(0.011) (0.013) (0.007) (0.007) Age (log) -0.085*** -0.090*** 0.010 0.009 (0.031) (0.031) (0.040) (0.045) Expense Ratio -0.026 -0.029 -0.098*** -0.095*** (0.021) (0.023) (0.014) (0.014) Total Load 0.009 0.013 0.009 0.018* (0.009) (0.011) (0.009) (0.010)
Last Year Flow -0.001 -0.005 -0.078 -0.081
(0.030) (0.030) (0.066) (0.066)
Last Year Alpha -0.058 -0.072 -0.793*** -0.800***
(0.305) (0.306) (0.090) (0.089)
Team managed 0.055* 0.022 -0.001 -0.000
(0.029) (0.031) (0.026) (0.027)
No. Countries Sold -0.012 -0.010 0.016*** 0.008
(0.011) (0.011) (0.004) (0.006)
Large Cap Fund 0.107 0.489* -0.283 -0.346
(0.245) (0.256) (0.343) (0.338)
LCF * TNA (log) -0.004 -0.025* 0.014 0.017
(0.013) (0.013) (0.019) (0.019)
Constant -0.174 -0.346 -1.179*** -1.098***
(0.265) (0.318) (0.255) (0.284)
Time-fixed effects Yes Yes Yes Yes
Country-fixed effects No Yes No Yes
Obs. 57135 57135 57897 57897
R-squared 0.195 0.214 0.554 0.559
Standard errors are in parenthesis *** p<0.01, ** p<0.05, * p<0.1
Country Characteristics
21 Country characteristics added are proxies for economic development, financial development, investor protection and legal institutions and lastly mutual fund industry concentration. Table 9 shows the regressions using the domestic equity fund sample using all the country characteristics. Table 10 shows the regression using the European equity fund sample, for this sample only the economic development and industry concentration variables are relevant as they affect the relationship between mutual fund investors and mutual fund managers.
Economic development
A country’s GDP per capita is a proxy for the income of the country’s inhabitants. It can be expected that in countries with a higher GDP per capita, there is more demand for the services offered by mutual fund and hence a larger market.
Looking at column 1 of tables 9 and 10, A weak negative relation between GDP per capita and performance is found, with the relation only being significant at the 10% level for the European equity investing fund sample. Domestic funds domiciled in countries with higher GDP per capita show no significant differential ability in achieving higher risk-adjusted returns however.
Education and tertiary education are both proxies for investor as well as manager sophistication. I hypothesize that funds domiciled in a country with higher educated inhabitants could perform better as a manager’s higher education could lead it to be better able to find interesting investing opportunities. This relation is expected to be stronger for the Europe sample as in the domestic sample these managers will mostly be competing against each other rather than competing against managers from other countries. Additionally, funds domiciled in a country with investors that have a higher level of education have clients who are more sophisticated in their decision in which fund to invest. This could also lead to higher quality mutual funds as investors will be more critical in determining which funds to invest in. While education appears to have no significant effect on fund performance, the variable tertiary education negatively affects the performance of funds, both those that invest domestically as well as internationally.
The variable internet usage is the percentage of a country’s population that uses the internet. In countries where there is a higher share of people using the internet, it can be hypothesized that more investors conduct research on their fund and investing opportunities. The variable internet usage does not significantly affect performance in column 1 of either table. In the fifth column of the sample with domestic funds, the relation with fund performance is positively significant. This provides us with weak evidence that funds domiciled in a country with higher internet usage perform better but is likely explained by endogeneity instead.
22
Table 9
Regression of Domestic Equity Mutual Fund Performance: Importance of Fund Size and Country Characteristics
This table reports panel regressions of the performance of open-end actively managed domestic equity funds from 1999 to 2018. The dependent variable is the monthly Carhart model alpha (percentage per month) using monthly fund returns in Euros. The independent variables include fund characteristics as well as country characteristics. Estimation (1) is with economic development variables added, (2) with financial development variables, (3) with legal proxies, (4) with the mutual fund industry concentration proxy, lastly (5) adds all country characteristics The regressions are estimated with time-series fixed effects. Standard errors are corrected for fund clustering and therefore more robust.
(1) (2) (3) (4) (5)
Alpha Alpha Alpha Alpha Alpha
TNA (log) 0.017 0.003 0.012 0.016 0.014
(0.012) (0.019) (0.010) (0.011) (0.020)
Family TNA (log) 0.012 0.010 0.013 0.015 0.001
(0.013) (0.012) (0.011) (0.011) (0.012) Age (log) -0.075** -0.032 -0.089*** -0.087*** -0.028 (0.034) (0.028) (0.031) (0.030) (0.029) Expense Ratio -0.031 -0.013 -0.032 -0.027 -0.018 (0.024) (0.028) (0.021) (0.020) (0.030) Total Load 0.017 0.017 0.001 0.011 0.011 (0.010) (0.014) (0.010) (0.008) (0.012)
Last Year Flow -0.024 -0.006 -0.006 -0.001 -0.000
(0.046) (0.039) (0.029) (0.030) (0.058)
Last Year Alpha 0.085 0.205 -0.071 -0.057 0.309
(0.347) (0.340) (0.308) (0.305) (0.379)
Team managed 0.043 0.064* 0.065** 0.057* 0.036
(0.036) (0.035) (0.031) (0.031) (0.033)
No. Countries Sold -0.013 -0.014 -0.015 -0.011 -0.039**
(0.017) (0.013) (0.010) (0.011) (0.018) GDP/Capita -0.000 -0.000 (0.000) (0.000) Gross Savings 0.002 -0.007 (0.003) (0.004) Education -0.038 -0.013 (0.037) (0.035) Tertiary Education -0.002** 0.009*** (0.001) (0.002) Internet 0.004 0.017*** (0.003) (0.005) Market Cap/GDP 0.000 0.002** (0.000) (0.001) Turnover 0.000 0.001 (0.000) (0.001) Contract Enforcement -0.008*** -0.014*** (0.003) (0.005) Minority Investors Protection 0.004*** 0.007 (0.001) (0.005) Herfindahl Index -0.020 -0.335 (0.213) (0.310) Constant 0.110 -0.032 0.095 -0.091 -1.120 (0.336) (0.233) (0.259) (0.219) (0.740) Obs. 34988 29456 57084 57135 20909 R-squared 0.182 0.280 0.201 0.195 0.310
23
Financial development
Market capitalization/GDP measures the relative importance of equity markets for firms in a particular country. Mutual fund’s in those countries likely play a bigger role in the capital allocation to companies compared to countries in which the equity markets are less significant. This variable appears to insignificantly affect fund performance for the domestic fund sample. Turnover, more specifically the share turnover ratio, is a proxy for the liquidity and stock market development. It is defined as the total value of stocks traded divided by market capitalization. Turnover does not significantly affect fund performance in domestic equity funds.
Legal environment
Contract enforcement is a variable that measures the difficulty to have a legal contract enforced. It is a proxy for the quality of the legal system in the country. For the sample of domestic equity funds, contract enforcement is related significantly negatively to fund performance, at the 1% level.
Minority shareholder protection is a variable that measures the protection of shareholders from conflicts of interest. Data for both variables were collected from the World Bank, the latter is based this measure on the methodology of Djankov, La Porta et al. (2008). Minority shareholder protection positively affects fund performance, with a 1% significance level for the domestic equity fund sample.
Mutual fund industry concentration
The variable Herfindahl index is a proxy for the concentration of the mutual fund industry in a country. Countries in which most mutual fund TNA are held by a few large fund families will have a higher Herfindahl index. This variable can therefore be interpreted as the level of competition in the mutual fund industry in a country.
24
Table 10
Regression of European Equity Mutual Fund Performance: Importance of Fund Size and Country Characteristics
This table reports panel regressions of the performance of open-end actively managed domestic equity funds from 1999 to 2018. The dependent variable is the monthly Carhart model alpha (percentage per month) using monthly fund returns in Euros. The independent variables include fund characteristics as well as country characteristics. Estimation (1) is with economic development variables added, (2) with the mutual fund industry concentration proxy, lastly (3) adds all relevant country characteristics. The regressions are estimated with time-series fixed effects. Standard errors are corrected for fund clustering and therefore more robust.
(1) (2) (3)
Alpha Alpha Alpha
TNA (log) 0.042*** 0.052*** 0.043***
(0.014) (0.011) (0.014)
Family TNA (log) -0.009 -0.000 -0.009
(0.009) (0.007) (0.009) Age (log) 0.041 0.016 0.044 (0.056) (0.039) (0.058) Expense ratio -0.095*** -0.102*** -0.095*** (0.017) (0.014) (0.017) Total Load 0.023** 0.012 0.024** (0.011) (0.009) (0.011)
Last Year Flow -0.073 -0.079 -0.073
(0.085) (0.066) (0.085)
Last Year Alpha -0.703*** -0.794*** -0.703***
(0.097) (0.090) (0.097)
Team managed 0.026 -0.002 0.025
(0.031) (0.026) (0.031)
No. Countries Sold 0.039*** 0.015*** 0.039***
(0.006) (0.004) (0.006) GDP/Capita -0.000* -0.000* (0.000) (0.000) Gross Savings -0.001 -0.000 (0.006) (0.006) Education 0.039 0.039 (0.027) (0.027) Tertiary Education -0.006*** -0.006*** (0.002) (0.002) Internet usage 0.002 0.002 (0.003) (0.003) Herfindahl -0.374** -0.120 (0.154) (0.223) Constant -0.760** -1.241*** -0.768** (0.333) (0.219) (0.337) Obs. 34732 57897 34732 R-squared 0.537 0.554 0.537
25
Discussion
While all specifications used to determine the relation between size and performance of the domestic equity fund find a positive relation between them, none of relations is significant. Notably, however, the relation between size and performance for the European equity fund sample provide evidence that economies of scale exist for internationally investing equity funds.
The difference between the scale relationship of domestic and European equity funds could be explained by a difference in the investment universe the funds face. While a domestic equity fund enjoys the same benefits from scale, having more resources for research and lower relative overhead expenses, the disadvantages for domestic funds are stronger. Domestic equity funds are more constrained in the securities in which it can invest. This might lead it to take larger positions in stocks than is optimal resulting from increased transaction costs or to invest in stocks that aren’t as attractive. Because European equity funds enjoy all the same benefits of scale but are far less constrained in the securities and assuming research resources are used efficiently, more attractive stocks can be identified. The disadvantages of price impact and liquidity potentially of European equity funds is therefore far lower. While the benefits and disadvantages for larger domestic equity funds seem to cancel out, for European equity funds the benefits outweigh the disadvantages. The results from the regression including the dummy variable of funds that invest more in large-cap stocks (higher than median SMB loading) and the interaction variable of this dummy variable with fund size support this evidence.
Interestingly, there appears to be no significant relation between fund family size and
performance. Benefits of a larger fund family were argued to be lower commissions/ bid-ask spreads reducing the costs of trading. The shrinking trading costs could explain the reason why fund family size no longer appears to significantly affect performance. As this study uses a more recent sample compared to previous research, the delta in commissions/ bid-ask spreads that small vs large mutual funds are paying might have become largely irrelevant. The findings suggest younger funds are able to generate higher returns. A long track record therefore does not suggest higher performance. Funds charging high expense ratios are generally unable to make up for them in returns. Mutual fund investors might therefore want to avoid funds charging high expense ratios. European equity funds charging a higher load are able to achieve better performance. This is in line with the hypothesis that these funds suffer less from volatility in fund size requiring them to hold relatively more liquid assets. The findings provide no evidence of the “smart-money” hypothesis. Funds with large inflows vs funds with large outflows do not achieve significantly different performance. The variable last year alpha is generally negative and significantly so for the European equity fund sample. This suggests there is performance reversal rather than performance persistence. Regression results further suggest domestic equity funds that are managed by multiple managers display better performance than funds that are solo-managed. This means that in European countries funds might be better organized, not suffering from some of the
26
Robustness checks
In order to investigate the robustness of the findings, several tables have been added with different methods of determining fund performance. These tables can be found in the appendices and will be discussed in turn.
Appendix I introduces two alternative performance benchmarks to compare the effect the addition of the Fama-French and Carhart factors for calculating alpha has on the relations found. Columns (1) to (3) employ the domestic equity sample while columns (4) to (6) employ the European equity sample. The relations found between the fund characteristics and fund performance found are robust to performance benchmark used. For almost all explanatory variables, the signs are consistent for all three performance benchmarks, with a few insignificant exceptions. The most interesting observation from this table however is that domestic equity fund size is positively significantly related to fund performance for both the CAPM model alpha and Fama-French three-factor model alpha, at the 5% level. Accounting for momentum strategies therefore has resulted in this significant relation disappearing. When using the CAPM model alpha as the performance benchmark, the relation between expense ratio and performance also becomes significant for the domestic equity sample, albeit only at the 10% level.
Furthermore, the relation between total load and performance becomes significant for both the domestic as well European sample when using CAPM model alphas, and significant for the European equity sample when using Fama-French model alphas.
Last year flow also becomes weakly negatively significant for the European equity sample using both CAPM and Fama-French model alphas.
Appendix II and III introduce country-specific factors to investigate the robustness of using EU factors for all funds. Appendix II uses German funds and adds Germany specific factors. Column 1 shows the relation of the domestic sample with alpha calculated using German factors. As the amount of observations for those samples is far smaller given we look at funds domiciled in Germany in appendix II, most fund characteristics lack significance in explaining fund performance. A few significant relations are observed for the regressions using German factor computed alphas which are insignificant when using EU factor
computed alphas. These are last year flow and the team managed dummy for the domestic funds and age, last year flow and number of countries sold for European equity funds. The signs of these variables are consistent with the EU factor found relations.
Appendix III is similar to appendix II but differs in the use of a UK sample and UK specific factors. While the UK sample is larger than the German given the larger amount of funds in the UK, many fund characteristics are again insignificantly able to explain fund performance. The expense ratio becomes insignificant for the domestic sample when using UK specific factors. For the European equity sample, last year alpha becomes insignificant in explaining fund performance when using UK specific factors.
