Abstract ’ s size and number of share classes on performance: Evidence from Luxembourg over 2004-2014 The impact of mutual fund

(1)

1

The impact of mutual fund’s size and number of share

classes on performance: Evidence from Luxembourg over

2004-2014

Author: Skevi Eleftheriou

1

Supervisor: Prof. Dr. A. Plantinga

University of Groningen

Faculty of Economics and Business

MSc Finance

Abstract

Fund management companies have the possibility to offer different share classes within a fund with purpose to satisfy investors with specific requirements. Therefore, they can increase their asset base, benefited from economies of scale and yield higher performance to the investors. Using a large sample free of survivor bias, the current study investigates the impact of fund’s number of share classes and size on risk-adjusted performance. The research focuses on the Luxembourg mutual fund market over the last ten years using both panel and portfolio regressions based on Fama and French (1993) three-factor model. The results have shown that the fund’s size and risk-adjusted performance are inversely related; whilst the number of share classes and funds’ risk-adjusted performance are positively related, which is proved to be highly significant based on the panel regressions but non-significant using the portfolio regressions.

Key words: mutual fund, share class, Luxembourg, size, risk-adjusted performance JEL Classification: G10, G11, G20, G23

1

(2)

2

Abbreviation List

ALFI Association of the Luxembourg Fund Industry

CSSF Commission de Surveillance du Secteur Financier

EFAMA European Fund and Asset Management Association

ESMA European Securities and Market Authority

FINRA Financial Industry Regulatory Authority

FMC Fund Management Company

NASD National Association of Securities Dealers

TNA Total Net Assets

UCITS Undertakings for Collective Investment of Transferable Securities

(3)

3

1. Introduction

Investors, considering to buy mutual funds, carefully take into account the investment objective, the risks as well as the policy of each fund in order to take the right decision as to which fund is the most appropriate for them. However, the last decades, mutual funds are permitted to structure their shares into different classes and, therefore, investors should also select which class suits best their investment goals. Mutual funds are investment companies that pool together the money of several investors. The funds are operated by portfolio managers who invest the pool of money in a common diversified portfolio of stocks, bonds or other assets. Share classes are categories of the same fund or sub-fund which share the same investment strategy and underlying assets. Classes which belong to the same fund/sub-fund have the same manager, board of directors, reporting requirements and performance before taking into consideration the specific features of each class. There are various types of share classes, some are simpler and some are more complex. Currently, the existing types of share classes in the European mutual fund industry differ from each other with respect to the cost structure2 (see Appendix A), the currency, the currency hedging, the minimum investment and holding amount, the targeted investors, the dividend distribution policy, the voting rights, the interest rate hedging and the volatility hedging.

The share classes in US, are permitted to differ according to their distribution or/and their shareholder servicing arrangements. The US classes are mainly differing in terms of their combination of front-end load, back-end load and 12b-1 fee3. In contrast with mutual funds in Europe, in US there exist three typical types of share classes, named A-class, B-class and C-class. Appendix B gives more details about the cost structure of the three typical share classes that exist in US.

Figure 1 depicts an example of a fund tree with two sub-funds, in which each has four and two share classes respectively.

2

Different combinations of entry charge, exit charge, ongoing charges and performance fee. 3

(5)

5

Fund Sub-funds Share Classes

Fig.1. Fund tree of the ‘MW Asset management’ fund. ‘MW Actions Europe’ sub-fund divided in four share classes named CI CAP, CI PCAP, CR CAP and CR P CAP and ‘MW Obligations Internationales’ sub-fund divided in two share classes named CI CAP and CR CAP.

The share classes are identified by abbreviations and combinations of abbreviations such as A, IB, CI USD and ACC in the end of the name of each fund. Each letter or combination of a letter and a number are used to declare the type of difference between share classes. For example, the fund named 'Nordea emerging consumer' has three share classes called ‘Nordea emerging consumer BI’, ‘Nordea emerging consumer BP EUR’ and ‘Nordea emerging consumer BI USD CAP’. Table 1 demonstrates typical abbreviations that have common meaning across different Fund Management Companies (FMC).

(6)

6

Table 1

Standard abbreviations in share classes’ names

These abbreviations have common meaning across different FMCs.

Abbreviation Meaning

I Stands for institutional. In most cases, it represents share classes that are offered only to institutional investors.

R Stands for retail. In most cases, it represents share classes that are offered to retail investors.

INC/DIS Stands for income or distribution. This type of share class gives dividends to investors.

CAP/ACC Stands for capitalisation or accumulation. This category of share class re-invests the income which is generated by a fund manager back to the fund.

HDG/Hedged In most cases, they are share classes which provide hedging against currency risk.

GBP/£ Share classes that are denominated in pounds. The currency of the share class is different from the base currency of the fund. EUR/€ Share classes that are denominated in euros. The currency of the

share class is different from the base currency of the fund. USD/$ Share classes that are denominated in US dollars. The currency

of the share class is different from the base currency of the fund.

However, beyond those typical abbreviations, the rest are randomly selected letters by FMCs which indicate some specific characteristics of each class. Each firm uses its own share class nomenclature. For instance, considering the fund 'Nordea', the letter 'A' in the end of the name of a share class represents the classes which distribute annual dividends to their investors. As for the fund 'Mirae asset global discovery', on the other hand, the letter ‘A’ entails the classes that are offered to all investors.

(7)

7

Norwegian law, the currency hedging as a separate type of a share class is not permitted but in the case of the UK, it started to be permitted in 2007 whilst in various other countries, it has long been available. Therefore, ESMA, recognizing the need for a common use of share classes, published a discussion paper on the 23rd of December 20144 on UCITS share classes which addresses different companies that are activated in asset management, legal and accountancy, investment services and others.5 The discussion paper includes ESMA's views on what is and what constitutes a UCITS share class and proposes a list with the permitted and non-permitted types of share classes. Chapter 2 gives a more detailed overview of the permitted and non-permitted share classes according to ESMA’s opinion.

The introduction of a new fund and the introduction of a new share class in an existing portfolio have different meaning for a FMC. A new fund/sub-fund essentially means that a FMC offers a new product to the investors, whilst a new share class within a fund entails the creation of more categories of an existing product. Sub-funds are exposed to different investment strategies and most of the times have legally segregated underlying assets. In contrast, classes have the same investment strategy and are exposed to the same pool of assets.

A key question is what the motives for a FMC to introduce new share classes rather than to establish a separate fund are. Each share class satisfies a group of investors with special requirements, for instance, investors who seek different minimum investment amount, cost structure, etc. This allows a FMC to cater for different types of clients. For example, FMCs could offer a different share class which charges lower expenses to institutional investors as they tend to have larger portfolios requiring relative smaller marketing efforts. Furthermore, it is generally acceptable that the attraction of new investors can lead to an increase of asset flows and therefore, a FMC can benefit from economies of scale. This could not happen in the case of an introduction of a new sub-fund, because there is a legal segregation of assets compared to classes which share the same pool of assets. The aforementioned arguments are also supported by SEC, which adds that the increase of asset base also enhances the portfolio liquidity and diversification.6 European Fund and Asset Management Association (EFAMA) agrees that multi-class funds can achieve share class diversification through the attraction of different types of clients.7 In addition, it mentions that the permission for the creation of more share classes will increase the competitiveness of the UCITS funds against the US mutual funds by attracting more

4

ESMA discussion paper: Share Classes of UCITS (ESMA/2014/1577). 5

On 27 of March 2015, was the last day for to send their feedback to ESMA. 6

Adoption Release IC-20915 for Rule 18f-3.

(8)

8

European investors. Furthermore, share classes which are denominated in different currencies benefit foreign investors who can invest directly to a fund using their own currency instead of converting their money to the currency of a fund. Besides that, based on the responses of ESMA's discussion paper8, all the companies agree that the introduction of a new share class needs much less cost and time to market than the introduction of a new fund/sub-fund. For instance, the introduction of an additional share class in an existing fund requires the notification from the competent regulator and the preparation of a new KIID (Key Investor Information Document) compared to the introduction of a new fund that needs prior authorization, new contracts with depositary bank, appropriate changes to all the UCITS prospectuses, etc. Moreover, certain fixed costs can be avoided with the development of a share class instead of a sub-fund, such as accounting and audit fees.

Beyond the aforementioned benefits, we cannot underestimate the confusion created in the mutual fund market. Because of the dazzling array of share classes and the non-existence of standard abbreviations for labelling the meaning of each share class, an investment decision for an individual investor is becoming more complicated. This complexity by the advent of multi-class funds increases significantly the role of financial advisors in the mutual fund market. The investors are, therefore, 'forced' to solicit the help of a broker or a financial advisor in order to buy the most economical share class. Even the National Association of Securities Dealers (NASD) recognizes the complexity that share classes introduce (see NASD Notices to Members 95-80).

Nevertheless, Financial Industry Regulatory Authority (FINRA) alerts the retail investors that the primary differences between the several share classes are the charged expenses to the investors and the amount that the financial advisors receive for selling a specific share class. Hence, a conflict of interest can potentially arise since the compensation of a financial advisor differs depending on which share class the investors choose. The regulators are trying to address these issues through the enhancement of disclosures and the monitoring of sales practices of the brokers/financial intermediaries. Thus, a question arises on whether brokers promote the share class that is on their own interest rather than the one on behalf of their customers. Jones et al. (2005) find evidence that the recommendations of the US financial advisors concerning the appropriate share class for each investor are influenced by the commission structure of each class. Additionally, there are academic papers which claim that the benefits by the introduction of additional share classes do not eventually pass to the investors. The paper by Lesseig, Long and Smythe (2002) finds evidence that, multi-class funds experience lower expense ratio but the

8

(9)

9

benefits by this reduction do not ultimately pass to the investors and therefore, only sponsors take advantage of the multi-class structure.

The current paper focuses on the existence of share classes in the Luxembourg mutual fund market. This study investigates whether the number of classes decreases or increases the funds’ risk-adjusted performance. Since the number of share classes is interrelated with the portfolio size, I also examine the relation between fund’s size and risk-adjusted performance. Based on the proponents of share classes, multi-class funds attract more asset flows and therefore, they could be benefited from economies of scale, achieving higher performance for their investors. However, the creation of too many share classes causes confusion to the investors and therefore can have eventually adversely results for them.

The remainder of the paper proceeds as follows. Chapter 2 provides, firstly, a general overview of the permitted and non-permitted share classes according to ESMA and, secondly, gives information about the Luxembourg mutual fund market. Chapter 3 discusses the literature and hypotheses and Chapter 4 presents the data and the methodology. Chapter 5 describes and explains the empirical findings and finally, Chapter 6 summarizes the results and concludes.

2. Background information

This section is divided into two parts. The first part gives more information about ESMA's proposals with regard to permitted and non-permitted types of share classes. The second section provides an overview of the Luxembourg mutual fund market.

2.1 ESMA’s proposals

As I mention in the introduction, ESMA takes the initiative to publish a discussion paper on UCITS share classes with the aim to establish a common position on share class regime across the European jurisdictions. Generally, due to the inconsistency across member states concerning the share class regime, it is not possible to have an exhaustive list with all the types of share classes that exist in the European mutual fund industry.

ESMA assesses the legality of share classes under three main principles. Firstly, share classes within the same mutual fund should have the same investment strategy. Secondly, the characteristics of one share class should not have negative impact on any other share class. Thirdly, UCITS should disclose the differences of their existing share classes to the investors.

(10)

10

1) Their targeted clientele. These classes are available for sale only to institutional investors or to all retail investors.

2) The type and amount of fees and loads that are charged. For example, a mutual fund can consist of two different share classes; the first class is charged α front-end load and an annual ongoing fee, a second class is charged a back-end load and an annual ongoing fee.

3) Denominated currency. The difference with the other share classes is in matter of its price. The price of this class is calculated on the basic currency of the fund but it is quoted and paid in the designated currency. For instance, the investors in Sterling share class of a multi-class fund, in which the majority of its classes are denominated in euros, will receive distributions (if there are any) in sterling and they will hold units that are priced in sterling instead of euros. The value of the class is calculated in sterling.

4) Distribution policy (accumulation or distribution). A fund can have a class which will re-invest the income generated back to the fund and a class which will distribute the income of a fund, through dividends to their investors.

5) Characteristics: registered or bearer 6) Voting rights

7) Currency hedging. Classes that are denominated in a different currency can give to the investors the option of hedged or unhedged class.

8) Maximum/minimum investment amount and holding requirement.

In addition to the above, ESMA proposes a list with the types of share classes that exist in the market but they are not compatible with the three principles. The non-permitted classes according to ESMA’s view should be:

1) Share classes with hedging against market risk, either interest rate risk or volatility risk. These types of classes are incompatible with first ESMA's principle of sharing the same investment strategy.

2) Share classes which have different percentage of capital protection and performance payoff.

3) Classes with different leverage.

4) Share classes that are exposed to different pool of underlying asset.

5) Classes with portfolio which is swapped against different portfolio of assets.

(11)

11

not in line with the rule of common investment strategy across classes of the same fund. Nevertheless, the majority of companies9 disagree with the opinion of ESMA. They claim that classes, which provide partial or total hedging against any market risk, are still following the same investment objective, but differ in terms of the level of exposure and therefore should be permitted.

2.2 Luxembourg’s mutual fund market

My study focuses on the Luxembourg mutual fund market. Luxembourg is characterised as an offshore jurisdiction which manages to be a major financial hub. Figure 2 demonstrates the top European domiciles based on their investment fund net assets. It reveals that Luxembourg is the largest centre for investment funds in Europe.

Fig.2. Source: EFAMA. The percentage of net assets across the European investment industry at the end of December 2014.

According to Association of the Luxembourg Fund Industry (ALFI), Luxembourg is the first European country which adopts the UCITS I, III and IV directive in its national law. UCITS directive permits the cross-border distribution of funds globally. Luxembourg is recognised as a global leader for cross-border distribution since predominantly funds set up in Luxembourg are sold in various countries beyond Luxembourg, including European and non-European

9 _{Association Française de la Gestion financière (AFG), AIMA (Alternative Investment}

Management Association) , Amundi asset management, Blackrock, SKAGEN AS, the Investment Association, Deutche bank and State Street Corporation

(12)

12

jurisdictions. Because of the wide range of clients, Luxembourg's mutual funds have to satisfy several different investor requirements. Therefore, funds with multi-class structure make it easier for FMCs to accommodate the various investors’ specific needs. Based on a Commission de Surveillance du Secteur Financier’s (CSSF)10 database, which includes all the identifiable Luxembourg-based UCITS, there exist 48,065 share classes. Under the Luxembourg regulations, the permitted types of share classes are related to the cost structure, dividend policy, targeted customer base, minimum initial investment and holding requirement, denominated currency and hedging policy.

3 Literature review

Despite the vast majority of literature that investigates the performance of US funds, there are also studies that examine the performance of mutual funds in specific countries. For instance, Dahlquist, Engström, Söderlind (2000) examine Swedish funds and McDonald (1973) evaluates French funds. This study aims to investigate the risk-adjusted performance of mutual funds in Luxembourg and to consider the impact of the number of share classes and funds’ size on return. The domicile of a mutual fund is an important element of fund performance. Ferreira, Keswani, Miguel and Ramos (2012) examined the potential determinants of future performance of equity mutual funds in 27 countries from 1997 until 2007. Nevertheless, the funds domiciled in Luxembourg and Dublin are excluded from their study. They take into account several fund characteristics (fund’s size, age, fees and expenses, etc.) as well as country characteristics (economic development, quality of legal companies, etc.). Their paper concludes that non-US funds which attract more new money have higher performance. They also show that countries with higher investor protection, greater legal system and better economic development provide higher performance for the domestic mutual funds. Furthermore, Otten and Bams (2002) study the performance of European mutual funds, excluding from their sample, though, Luxembourg as an offshore region. Their results reveal that European mutual funds have positive risk-adjusted performance, especially the small capitalization funds which outperform the market.

The primary aim of the present research is to investigate if funds with more share classes have higher performance compared to funds with fewer share classes. There are two ways in which multi-class funds can yield higher expected return to the investors. Firstly, as many claim, an introduction of classes leads to the increase of fund's asset base and therefore, economies of scale can be achieved. Consequently, FMCs can benefit from lower expense ratio and can pass

10

(13)

13

higher performance to their investors. Secondly, multi-class funds might have lower cash flow volatility, which allows fund managers to hold less cash and invest more assets in highly return securities resulting in higher expected returns. However, even if fund managers benefit from the permission to create several share classes, there is still the need to investigate how much value eventually passes to investors through higher returns. It is essential to look into previous literature on how cash flow level and volatility change with the introduction of additional share classes and, afterwards, how this induces the performance of a fund to change. The share classes under the same fund have the same performance before taking into account the specific characteristics of each class, such as its cost structure.

3.1. Relation between share classes and fund’s size

According to share classes’ proponents, the creation of additional share classes increases the asset flows of mutual funds. Nanda, Wang and Zheng (2004) support that there is 4% annual growth in cash flows for multi-class funds in contrast with the single-class funds. Specifically, using a large sample of US equity funds which covers the period from 1993 to 2002, they conclude that the introduction of new share classes give rise to funds' cash flows especially during the second and third year after their introduction. The same result is also suggested by the study of Walsh (2004).

3.2. Relation between fund’s size and expense ratio

(14)

14 3.3. Relation between fund’s size and performance

Furthermore, previous studies, such as those of Chen et al. (2002), Ferreira, Keswani, Miguel and Ramos (2012) and Busse et al. (2014) report the negative relation between fund’s size and return. Chen et al. (2002) claim that fund size and performance have negative relation after controlling for other fund characteristics. Specifically, Busse et al.’s paper (2014) proxies fund size by aggregating the Total Net Assets (TNA) of all share classes among a portfolio and reveals that even after controlling for transaction cost, the funds with large TNA underperform the funds with small TNA. Ferreira et al. (2012) find evidence that fund performance decreases with the size of the fund in the US, but this is not the case for the non-US mutual funds. Their study presents positive and statistically significant relation between size and performance for non-US funds. In addition, Otten and Bams (2002) suggest a positive relation between fund’s size and risk-adjusted performance concerning the European mutual funds.

3.4. Relation between expense ratio and performance

Based on the finding of the aforementioned papers that multiple class funds have higher expense ratio, I am expecting that these funds have lower performance. Fund managers, however, claim that higher level of fees does not erode the fund return since investors pay more due to better quality of management information. The paper by Otten and Bams (2002), though, studying the European mutual fund market, proves that expense ratio and fund performance have negative relation. This conclusion is also in line with the results of prior studies; Carhart (1997), Malkiel (1995), Edelen, Evans, and Kadlec (2013), Jensen (1968) and Elton et al. (1993). Specifically, Malkiel (1995) distinguishes the total expenses into two categories, the advisory and non-advisory expenses (advertising, marketing etc.), and investigates the relation between those two types of expenses and of fund performance. The findings of his study indicate that non-advisory fees have a significantly negative relation with the performance; in contrast, advisory fees have no statistically significant relation with the performance. Ferreira, Keswani, Miguel and Ramos (2012) have found statistically significant relation between loads (front-end load and back-end load) and fund’s performance.

Based on the above literature review, the first line of reasoning that funds with more share classes have lower expense ratio and higher performance does not hold.

3.5. Relation between share classes and cash flow volatility

(15)

15

yield higher expected return as stated in a previous section. Walsh (2004) investigates the possibility that multi-class funds decrease the cash flow volatility and lead to higher returns. In contrast with her expectations, she finds weak evidence supporting that multi-class funds actually increase flow volatility which could lead to lower gross returns. These results are consistent with Nanda, Wang and Zhang (2004). They notice that multi-class mutual funds attract investors with different preferences and needs. Therefore, they claim that this fact increases cash flow volatility and leads to significant reduction of fund performance.

The general conclusion drawn by all the above papers is that multi-class funds are successful at increasing the asset base; nevertheless, investors do not obtain any benefits in the form of lower expense ratio and higher performance. Additionally, mutual funds with multi-class structure increase cash flow volatility and therefore experience lower performance.

3.6. Other papers related to the existence of share classes

Other papers that investigate multi-class funds are those by Zhao (2002) and Nanda, Narayanan, Warther (2000). Zhao (2002) examines the factors that induce fund management companies to take two separate entry decisions; firstly, whether to add new share class in an existing fund and secondly, whether to introduce a new fund either single-class or multiple-class. The paper finds evidence that funds with both good and poor performance decide to launch a new fund. In contrast, the decision to launch a new share class in an existing fund or to introduce a new fund with multi-class structure is taken by poor performance funds searching for new investments. Furthermore, Zhao (2002) demonstrates that the decision of a fund to launch new fund with multiple-class structure is very risky introduction strategy.

(16)

16

To the best of our knowledge, this is the first empirical study in the academic literature that investigates the impact of the number of share classes on fund performance especially in the Luxembourg mutual fund market.

4. Data and methodology

4.1 Data Collection

The Thompson’s Datastream is the main database from which I collect the data for my research. The sample of the research is based on a sub-set of Luxembourg funds’ data set which is free of survivorship bias since it includes both active and inactive funds. Several papers show that survivorship issues can change the results in a great degree (Brown et al., 1992). The initial data set includes 3,776 share classes (both active and dead funds). However, 376 classes with missing observations concerning their ISIN number or their returns are excluded from the initial sample. Therefore, the final data set covers 3,437 share classes incorporated in 860 mutual funds. I acquire the name, the ISIN number, the monthly total return index and the total net asset value of each share class for a time period of ten years from 1st of September 2004 until 1st of September 2014 from Datastream. In addition, the monthly benchmark returns on RMRF (market return minus risk-free rate), SMB (Small Minus Big), and HML (High Minus Low) are obtained from Kenneth French's data library. Since the Luxembourg funds are distributed globally, I collect the data of the global factors which include 23 countries in four regions11. Finally I collect the Euribor one-month rate from the Datastream database as a proxy of the risk free rate. Table 2, depicts the measurement and source of all the parameters that are used in my analysis.

11_{Australia, Austria, Belgium, Canada, Denmark, Finland, France, Germany, Greece, Hong Kong,}

(17)

17

Table 2

Measurement and source of the regression variables

Variable Dependent/Independent Measurement Source

Ri-RF Dependent Excess Fund

Return

Ri: Datastream RF: Datastream

RM-RF (RMRF) Independent Excess market

Return RM: Kenneth R. French's data library RF: Datastream SMB (Small Minus Low)

Independent Small firm effect Kenneth R.

French's data library HML (High

Minus Low)

Independent Value premium

effect Kenneth R. French's data library LogSize (Logarithm of Size) Independent Logarithm of

fund’s size (sum of TNA)

Own Calculation

NOSC (Number Of Share

Classes)

Independent Number of share

classes of each fund

Own Calculation

4.2 Descriptive Statistics

(18)

18

which means higher probability to observe extreme values. The factors of SMB and HML have approximately similar standard deviations of 0.010 and 0.012 respectively. Looking at the average excess fund performance in Panels B and C, it is noticeable that multi-class funds have higher average excess return compared to single-class funds. Nonetheless, the abnormal return among the single-class fund sample is fluctuated in a smaller range of 1.436 compared to the case of multi-class funds which are fluctuated between 0.723 and -0.902. The higher variation on the excess portfolio return is also observable by studying the standard deviation of multi-class funds which stands at 0.062 compared to 0.040 of single-class funds.

Table 3

Descriptive statistics of the regression variables

Panel A, B and C includes descriptive statistics for my whole sample, only for single-class funds and only for multi-class funds respectively. The sample statistics covers the period 2004-2014. No of funds is the number of funds and Std. dev. is the standard deviation.

(19)

19

Figure 1 in Appendix C portrays the development of share classes and funds of my sample during the period of 2004 to 2014. The graph shows clearly that there is a significant difference between the two groups especially after 2010. In the beginning of the study period, mutual funds stand at 127 funds and, in 2014, they reached 860. In contrast, the amount of share classes experiences more dramatic rise of approximately eightfold the initial amount of 355 classes. More specifically, this number (i.e. 355 classes) increases significantly to 997 by 2009 and continues its upward trend more dramatically until 2014 reaching a peak of 3,340 share classes. Tables C1 and C2 in Appendix C illustrate the number of multi- and single-class funds respectively from 2004 until 2014. Both types of funds experience an increase over the years, with multi-class funds rising more significantly than single-class funds. Looking at Table C2, which gives information about the number of single-class funds, we notice the gradual increase of single-class funds by 253. Nevertheless, the multi-class funds rise by 479, reaching a peak of 547 funds in 2014. It is worth mentioning that the number of share classes in 2014 stands at 3,028 which belongs to 548 multi-class funds.

(20)

20

Fig.3. Distribution of my sample funds based on their number of share classes.

Table 4

Distribution of the sample funds into 4 categories based on the number of their share classes based on the last day of observation.

Number of share classes Number of Funds

1 312

2-3 247

4-6 138

7-30 159

In order to check for a possible existence of multicollinearity problem, I construct a correlation matrix with all the variables used in the methodology. Table 5 shows that variables with correlation above 0.7 do not exist, thus, no multicollinearity problem occurs. As it is expected, the variables of LogSize and NOSC exhibit positive relation at around 0.432, since bigger in size funds have more share classes.

(21)

21

Table 5

Correlation Matrix between regression variables

4.3 Methodology

Datastream treats each share class as a different unit of observation. Therefore, I have to identify which classes belong to which fund. I manually match the fund classes, by studying the share classes’ names, their ISIN number and using the website of Fundsquare market infrastructure12 which permits the investigation of dynamic data for each fund. The data set is composed of 3,437 share classes after the exclusion of 339 share classes with incomplete data. Based on the Fundsquare website, I realise that in my sample there are funds which actually have more share classes than those presented in Datastream. Fifty-one percent of the funds of my sample have more share classes than those that exist in Datastream. In order to continue with my analysis, I initially have to ensure that the percentage of share classes that are missing is not large enough that can alter my research results. Thus, for each fund I calculate which percentage of its share classes does not exist in Datastream by dividing the number of share classes that Datastream database provides by the number of classes that exist in Luxembourg Fundsquare website. Hence, I find that the Datastream provides a complete number of share classes for the 49% of funds in my sample while for the rest funds (i.e. the remaining 51%), it provides on average the 75% of their total number of share classes. In other words, for the 51% of the funds in my sample, a percentage of 25% of their share classes is missing. Thus, in the current study, I decided to include all the funds; those with complete and incomplete number of share classes. From Datastream, I obtained the total return index for each share class of my sample. Total return index is calculated assuming that dividends and distributions are reinvested; hence, it is proved to be a

12

Fundsquare Market Infrastructure is a wholly-owned subsidiary of the Luxembourg Stock Exchange, which offers a broad range of services in the order routing and information services space.

(22)

22

more accurate measurement of performance. Subsequently, the returns for each share class are calculated based on the formula below,

Ri,t = Pi,t – Pi,t-1 /Pi,t-1, (1) where Rit represents the monthly simple total return index, Pit is the total return index of share

class i in month t and Pi,t-1 is the total return index of share class i in month t-1. For the case of

single-class funds, the return of the class coincides with the return of the fund. On the other hand, for the calculation of multi-class funds' return, I apply the TNA-weighted average of the share classes’ returns which also considers the size of each class as a proxy by their TNA, following the methodology by Chen et al. (2002) and Nanda, Wang, Zheng (2004). Using the methodology of TNA-weighted average performance of mutual fund, the significance of any small-fund omissions can be minimized. Net asset value represents the per share market value of a fund.

Furthermore, for the calculation of funds’ size, I sum up the TNA of all the share classes that constitute a mutual fund. Following Chen et al. (2004), Christoffersen et al. (2008), Lang and Köhler (2011), Walsh (2004) and Yan (2008), I use the natural logarithm of TNA of a mutual fund as a proxy for fund size. In the case of multi-class funds, the size is the logarithm of the sum of TNA for all the classes within a mutual fund.

I measure the risk-adjusted fund performance, which indicates the percentage of the picking ability of a manager that contributes to the fund’s return. For the calculation of risk-adjusted performance, several studies such as Ippolito’s (1989) and Malkiel’s (1995) use the Capital Asset Pricing Model (CAPM). The CAPM explains the fund performance using only one risk factor named the market risk. Several papers such as Nanda, Wang and Zheng (2004), Chen et al. (2004), Busse et al. (2014) employ the Fama and French (1993) three-factor model to study the performance of equity funds,

𝑅it – 𝑅Ft = 𝛼i + 𝛽iRMRF(𝑅𝑀 − 𝑅𝐹t) + 𝛽𝑖SMB𝑆𝑀𝐵𝑡 + 𝛽iHML𝐻𝑀𝐿𝑡 + 𝑒it, (2) which approximates the fund performance by adding other two risk factors; the size and book-to-market. As a proxy of risk free rate, I use the Euribor one-month rate which is denominated in euro. The formula of geometric mean is used to convert the Euribor one-month rate from an annual basis to monthly,

(23)

23

where R is the Euribor one-month rate which in an annual basis. Additionally, the monthly global factors of SMB, HML and market return as obtained from the Kenneth French's data library are denominated in US dollars; therefore, I gather the monthly USD to EUR exchange rates from Datastream in order to convert them to euro. I investigate the impact of fund’s size on fund performance and the number of share classes on fund returns by adding two more variables to the Fama and French (1993) three-factor model. My regression is:

𝑅it − 𝑅Ft = 𝑎i+ 𝛽1 (𝑅Mt− 𝑅F) + 𝛽2 𝐻𝑀𝐿t+ 𝛽3 𝑆𝑀𝐵t + 𝛽4 𝐿𝑜𝑔𝑆𝑖𝑧𝑒i+ 𝛽5 𝑁𝑂𝑆𝐶i + 𝑒it, (4)

where Rit is the rate of return of fund i in month t, RFt is the one month Euribor (interbank) rate in

month t, Rit − RFt represents the excess fund return, RM-RFt represents the excess market return as

the market return minus the risk free and eit is the residual return of fund i in month t. In addition, SMBt is the difference in return between a small cap portfolio and a large cap portfolio and, HMLt

is the difference in return between a portfolio of high book to market stocks and a portfolio of low book to market stocks. The variables of NOSC and LogSize represent the number of share classes and the size of each fund respectively. Finally, alpha factor is the risk-adjusted performance which is an indicator of the contribution of fund manager to the performance of a fund.

I employ a Hausman test to check whether fixed effects or random effects are appropriate. The outcome of the test shows that the time fixed effects specification is preferred. I also follow a portfolio approach by separating the Luxembourg funds into four portfolios with the purpose to check if the variable of size changes the impact of share classes on performance. Running a regression for every fund based on Fama and French’s (1993) three-factor model, I come up with 860 different alphas that equal the amount of funds that my sample includes. Firstly, I divided the funds into two groups, the funds with many share classes and the ones with few share classes. Since the average fund in my sample has approximately three classes, I assume that funds with three classes or less are classified into the category with funds with few classes, while funds with four share classes or more are categorized as funds with many classes. Subsequently, both groups are sorted based on their TNA and then I take the 30% of the smaller and larger funds from each category.

Thus, I create four portfolios:

(24)

24

Afterwards, I calculate the average value of excess return in each portfolio. Based on Jarque-Bera test, the alphas presented in my sample do not follow normal distribution and therefore, in this case, the use of median instead of mean is more common, since median is less susceptible to the influence of skewed data and outliers.

Due to the fact that the population of alphas does not follow normal distribution, an alternative method of t-test is applied, called Wilcoxon rank-sum test13. The null hypothesis of this test checks if the median difference between the pairs is zero. The advantage that Wilcoxon rank-sum test provides is the fact that there is no need to asrank-sume that the population is normally distributed. I sort the funds into four different portfolios based on the number of their share classes and, I check whether there is a difference on risk-adjusted returns’ medians between the four portfolios.

Table 6 reports the estimation technics that some previous researches employ in order to investigate funds’ performance. Generally, the majority of studies perform portfolio regressions using the models of CAPM, Fama and French (1993) three-factor model and Carhart (1997) four-factor model. There are also papers that use pooled regressions controlling for portfolio attributes. The study by Nanda, Wang and Zheng (2004) uses at first a pooled regression with panel-corrected standard errors. Furthermore, as robustness check, they estimate portfolio regression, comparing the performance of multi-class funds with A-classes with that of single-class funds. The paper by Kacperczyk, Sialm and Zheng (2005) follows a portfolio approach, constructing deciles on the basis of past return gap. Moreover, their results are robust using panel regressions as well, controlling for other portfolio attributes and time fixed effects. In addition, Barber, Odean and Zheng (2002) employ portfolio approach, comparing the returns of two portfolios that are separated based on the front-end load.

13_{Wilcox is a non-parametric method appropriate for examining the difference in medians for 2}

(25)

25

Table 6

Methodology of previous studies

Author(s) and year Period Data Research Question Methodology Kacperczyk, Sialm and Zheng (2005) 1984-2002 U.S. Equity Funds The impact of unobserved actions on fund performance

1. Panel Regression (control for fund characteristics. (CAPM, Fama and French (1993), Carhart (1997)) 2. Portfolio Regressions

Nanda, Wang and Zheng (2004) 1993-2003 U.S Equity Funds The impact of introducing multiple share classes on fund performance

1. Pooled regression with panel-corrected Standard errors. (CAPM, Fama and French (1993), Carhart (1997))

Chen et al. (2004) 1962-1999 U.S Equity Mutual Funds The effect of fund size on fund performance

Portfolio Regression (CAPM, Fama-French (1993),

Carhart(1997))

Edelen, Evans, and Kadlec (2013) 1995–2006 Open-end domestic equity mutual funds The effect of funds’ cost on fund performance Carhart (1997) four-factor model

My first hypothesis supports the notion of inverse relation between fund’s performance and the variable of number of share classes. The second hypothesis predicts that fund’s size and risk- adjusted performance have negative relation based on the findings by Chen et al. (2002), Ferreira, Keswani, Miguel and Ramos (2012) and Busse et al. (2014).

5. Empirical results

5.1. Results of panel approach

(26)

26

suggest that higher market risk leads to higher risk-adjusted performance. What is more, a highly statistically significant positive SMB loading indicates that portfolio performance is driven mostly by smaller stocks than by larger. In addition, the results support the existence of negative coefficient of HML factor, implying that value stocks outperform the growth stocks during the study period.

Table 7

Results of panel regression coefficients

All Share Classes Multi-class Funds

a 0.013*** 0.010*** RM-RF 0.697*** 0.767*** SMB 0.124*** 0.074*** HML -0.086*** -0.053*** LogSize -0.003*** -0.002*** NOSC 0.001*** 0.001** Observations 55,492 33,779 Adjusted R2 0.338 0.377

Note: ***, **, and * represents the coefficients that are significant at the 1%, 5%, and 10% level respectively.

(27)

27

The results of Table 8 suggest that portfolios with seven to 30 share classes have the lowest returns across the four groups while single-class funds have the highest returns. Inconsistent with the previous results, this analysis implies that investors attract lower returns from portfolios with many share classes.

Table 8

Results of panel regressions-categories

Single 2-3 classes 4-6 classes 7-30 classes

C 0.075*** 0.021*** 0.018** 0.005*** RM-RF 0.593*** 0.757*** 0.878*** 0.712*** SMB 0.187*** 0.100* 0.101** -0.009 HML -0.128*** -0.141*** -0.092* -0.021 LogSize -0.019*** -0.004*** -0.003*** -0.001*** Adjusted R2 0.286412 0.153 0.283 0.362 Observations 21,736 14,732 8,428 10,512

Note: ***, **, and * represents the coefficients that are significant at the 1%, 5%, and 10% level respectively.

5.2. Results of portfolio approach

I employ a portfolio analysis constructing four different groups based on the number of share classes and fund’s size. The results of the portfolio approach can be seen in Table 9. The conclusion according to both mean and median alphas shows that, independently of portfolio size,

the excess performance is higher for funds with many share classes compared to those with few. Hence, the outcome supports the positive coefficient of NOSC that panel regression reveals. My

predictions for the negative effect of the number of share classes are not verified. A possible explanation is that multi-class funds benefit FMC and, those benefits also pass to the investors without being exploited only by the financial advisors. The paper by Ferreira, Keswani, Miguel and Ramos (2012) shows that funds domiciled in countries with strong legal environment and higher investor protection pass higher performance to the investors. Undoubtedly, Luxembourg represents a country with strong legal environment and high investor protection and this is enhanced by the fact that the majority of funds domiciled in Luxembourg adopt the UCITS directive.

(28)

28

hypothesis that the median alpha of a portfolio with few and large funds is equal to that of a portfolio with many and large funds. The result of Wilcoxon rank sum test reveals that none of median alphas is statistically significant. Panel B of Table 9 presents a square in which the right upper triangle illustrates the mean differences and the left down triangle illustrates the median differences among the four portfolios. The largest median margin is between the portfolio with many and large funds and the portfolio with few and small funds. In contrast, in the case of averages, the largest range is noticed between the portfolios with few and small funds and the portfolios with few and large funds.

Table 9

Results of portfolio approach

The funds are sorted into four portfolios on the basis of number of share classes and size on the last month of observation 1/08/2014. Panel A presents the mean and median risk-adjusted performance of each portfolio. Fama and French’s alpha is the intercept from a monthly time-series regression with dependent variable the mean monthly excess return of each sample portfolio and explanatory variables, the market excess return, the SMB and the HML. Panel B illustrates differences of alphas between the four portfolios. The upper triangle illustrates the mean differences and the left down triangle illustrates the median differences.

Panel A

Portfolios Mean Alpha Median Alpha

Many and large 0.000607 0.0009 Few and Large 0.001909 0.00055 Many and Small -0.00035 -0.00016 Few and Small -0.00171 -0.0012

Panel B

Portfolio I:ML Portfolio II: FL Portfolio III:MS Portfolio IV: FS

Portfolio I: ML - 0.0013 -0.0010 -0.0023

Portfolio II: FL 0.0003 - -0.0023 -0.0036

Portfolio III: MS 0.0011 0.0007 - -0.0014

Portfolio IV: FS 0.0021 0.0018 0.0010 -

(29)

29 5.3. Robustness checks

In this section, I employ a number of checks on the robustness of the results based on panel approach. The results are robust to different model specifications and different estimation techniques. The results are presented in Table D of Appendix D.

Robustness check 1

The first robustness test checks if my results are affected by missing reporting data. At this stage of the analysis, I use only the 423 funds which have complete data concerning their number of share classes. Panel A of Table D (see Appendix D) shows that the results are robust and, it is therefore proved that they are not affected by missing information.

The third robustness check is related to the factors of Fama and French’s (1993) three-factor model. I employ the same analysis using European instead of global factors to explain risk- adjusted returns. Panel B of Table D indicates that the performance results are robust to analyses including European instead of global factors concerning the parameters of Fama and French’s (1993) three-factor model.

I employ the Carhart (1997) four-factor model which is an extension of the Fama and French (1993) three-factor model, by adding a forth factor in order to capture the momentum anomaly. The study by Chan, Jegadeesh, and Lakonishok (1996) shows that momentum anomaly represents a market inefficiency that arises from the slow reaction of the information. As a robustness check, I also introduce the equity momentum factor (WML) to my basic equation,

𝑅it – 𝑅Ft = 𝛼i + 𝛽1(𝑅M− 𝑅Ft) + 𝛽2𝑆𝑀𝐵t + 𝛽3𝐻𝑀𝐿t + 𝛽4𝑊𝑀𝐿t+ 𝛽5 𝐿𝑜𝑔𝑆𝑖𝑧𝑒i + 𝛽6 𝑁𝑂𝑆𝐶i+ 𝑒it, (5) Panel C of Table D in Appendix D confirms that my basic results do not change after the introduction of the momentum factor.

6. Conclusion

(30)

30

the analysis shows the positive influence of the number of share classes on fund’s performance. A possible explanation of this outcome can be the higher regulation attributed to the existence of UCITS, especially in Luxembourg where the majority of mutual funds adopt the UCITS directive. Nevertheless, the panel approach denotes statistically significant positive impact whereas the portfolio approach indicates that the number of share classes is a non-significant parameter for the explanation of risk-adjusted return. Therefore, a non-strong result can be concluded for the influence of the number of classes.

While numerous studies examine the mutual fund market, the share classes receive a little research attention. This paper takes a step forward filling this gap by investigating the influence of funds’ share classes on performance. To the best of our knowledge, this is the first paper that investigates the impact of the number of share classes on the performance and especially in an offshore jurisdiction as the Luxembourg. Additionally, this study confirms that the negative relation between portfolio size and performance is also valid in the case of Luxembourg mutual funds.

(31)

31 Acknowledgements

I would like to express my special thanks and gratitude to my supervisor Dr. Auke Plantinga, for

his guidelines, helpful feedback, advice and understanding in order to achieve the objectives of

my research. He has been a source of knowledge and encouragement offering constant help.

Furthermore, a very warm thanks to my family and friends for their endless love, support,

encouragements, help, and insights until the accomplishment of this thesis. They are always being

there for me to support all my decisions.

(32)

32

7. References

Barber, B., Odean, T., Zheng, L., 2005. Out of sight, out of mind: The effects of expenses on mutual fund flows. Journal of Business 78, 2095-2119.

Brown, J. S., Goetzmann, W., Ibbotson, G. R., Ross, A. S., 1992. Survivorship bias in performance studies. Review of Financial Studies 5, 553–580.

Busse, J. A., Chordia, T., Jiang, L., Tang, Y., 2013. How does size affect mutual fund performance? Evidence from mutual fund trades. Unpublished working Paper. Emory

University, Tsinghua University and Singapore Management University.

Carhart, M. M., 1997. On persistence in mutual fund performance. Journal of Finance 52, 57–82. Chen, J., Hong, H., Huang, M., Kubic, J. D., 2004. Does fund size erode mutual fund

performance? Liquidity, organizational diseconomies, and active money management. American Economic Review 94, 1276-1302.

Christoffersen, S. E. K., Keim, D. B., Musto D. K., 2008. Valuable information and costly liquidity: Evidence from individual mutual fund trades. Unpublished working paper. University of Pennsylvania.

Dahlquist, M., Engström, S., Söderlind, P., 2000. Performance and characteristics of Swedish mutual funds. Journal of Financial and Quantitative Analysis 35, 409-423.

Edelen, R., Evans, R., Kadlec, G., 2013. Shedding light on “invisible” costs: Trading costs and mutual fund performance. Financial Analysts Journal 69, 33–44.

Elton, E., Gruber, M., Das, S., Hlavka, M., 1993. Efficiency with costly information: A reinterpretation of evidence from managed portfolios. Review of Financial Studies 6, 1-22. Fama, E., French, K.R., 1992. The cross-section of expected stock returns. Journal of Finance 47,

427-465.

Fama, E., French, K. R., 1993. Common risk factors in the returns on stocks and bonds. Journal of Financial Economics 33, 3-56.

Ferreira M. A., Keswani, A., Miguel, A. F., Ramos, S. B., 2012. The flow-performance relationship around the world. Journal of Banking and Finance 36, 1759–1780.

Jensen, M. C., 1968. The performance of mutual funds in the period 1945-1964. The Journal of Finance 23, 389-416.

Jones, .M.A, Lesseig, V.P., Smythe, T.I, 2005. Financial advisors and multiple share class mutual funds. Financial Services Review 14, 1-20.

(33)

33

Lang, G., Köhler, M., 2011. How does the domiciliation decision affect mutual fund fees?. ZEW Discussion Paper, 11-085.

Lesseig, V. P., Long, D. M., Smythe, T. I., 2002. Gains to mutual fund sponsors offering multiple share class funds. Journal of Financial Research 25, 81-98.

Malkiel, B. G., 1995. Returns from investing in equity mutual funds 1971-1991.Journal of Finance 50, 549-573.

McDonald, J., 1973.French mutual fund performance: evaluation of internationally-diversified portfolios. Journal of Finance 28, 1161-1180.

Nanda, V., Narayanan, M.P., Warther, V.A., 2000. Liquidity, investment ability, and mutual fund structure. Journal of Financial Economics 57, 417- 443.

Nanda, V., Wang J. Z., Zheng, L., 2004. The ABCs of mutual funds: A natural experiment on fund flows and performance. Journal of Financial Intermediation 18, 329-361.

National Association of Securities Dealers, Notice to Members 95-80.

Otten, R., Bams, D., 2002. European mutual fund performance. European Financial Management 8, 75-101.

Securities and Exchange Commission Rule 18f-3 Adoption Release, Release No. 33-7143, IC 20915, 1995

Walsh, L., 2004. The costs and benefits to fund shareholders of 12b-1 plans: An examination of fund flows, expenses and returns. Office of Economic Analysis, U.S. Securities and Exchange Commission.

Yan, X., 2008. Liquidity, investment style, and the relation between fund size and fund performance Journal of Financial and Quantitative Analysis 43, 741-768.

(34)

34

Appendices

Appendix A. Definition of share classes’ fees and loads

Each share class has different combination of front-end load, back-end load, ongoing charges and performance fee.

Entry charge/ Front-end load is defined as the percentage of the total invested amount that the consumers pay to the investment intermediaries at the time of the purchase as sales commission. The front-end load sometimes decreases as the amount of the investment increases.

Exit charge/ Back-end load is expressed as the percentage of the sale value of the share that the investors pay to the fund at the time of the redemption. This type of load is also known as contingent deferred sales charge which is paid if fund’s shares are redeemed before a specified holding period and this fee decreases yearly until the end of the holding period which equals to zero.

Ongoing Charges are also known as management expense ratio or Total Expense Ratios. It is an annual expense which is expressed as a proportion of the fund's net assets. The ongoing charges cover all the necessary operational expenses as well as the remunerations of any party which provides services. For example management fees, administration fees etc. Another ongoing fee is the 12b-1 fee (in US) or ongoing marketing fee (in EU), which cover the brokerages' commission and any other marketing expenses for advertising the fund.

Performance fee is paid to the fund manager as a reward for the positive return of the fund.

Appendix B. The three typical share classes in US mutual fund industry

The history of the multi-class funds began in October in 1980 when the SEC adopted the 12b-1 Rule, under the Investment Company Act of 1940, which allowed the US mutual funds to bear expenses concerning the distribution of fund shares. Fifteen years later, in 1995, the adoption of the 18f-3 Rule, permits the issue of multiple share classes by the open-end invstment companies. After those two developments, the Classses A, B and C created. A-class, B-class and C-class are the typical mutual fund classes in the US mutual fund market.

(35)

35

Class B shares charge a back-end load; which decreasing every year that investors hold the fund and a 12b-1 fees that are higher than A share class. Their particular characteristic is that B-classes converted to Class A shares several years later and therefore, the investors benefit from the lower annual expense ratio of A-class shares.

Class C shares charge a back-end load, typically 1%, provided that the investors redeem their shares prior of to the end of the first year from the date of purchase and a 12b-1 fee that is higher than class A funds. They are not converted to A class shares.

Appendix C. Development of share classes and funds based on my sample

during the study period of 2004-2014

Fig.1. Development of the number of share classes and portfolios during the period of 2004 to 2014. The figure illustrated based on my sample data.

(36)

36

Table C.1

Development of Multi-class funds during my study period of 2004-2014

Year Number of share classes Number of Funds

2004 296 68 2005 337 85 2006 556 133 2007 646 161 2008 722 181 2009 826 193 2010 1,060 240 2011 1,429 302 2012 1,812 370 2013 2,391 448 2014 3,028 548 Table C.2

Development of Single-class funds during my study period of 2004-2014

Year Number of Funds

(37)

37

Appendix D. Robustness Checks

Table D

Robustness checks

Panel A depicts the results of the robustness check of missing reporting data, by using only funds with complete (100%) number of share classes. Panel B shows the results of the robustness check of using European instead of global factors and Panel C displays the results the 3rd robustness check of adding the WML factor in my basic regression Eq.(4).

Panel A: 100% classes Panel B: European factors Panel C: WML

Alpha 0.02*** Alpha -0.013*** Alpha 0.013*** RM-RF 0.63*** RM-RF_European 0.566*** RM-RF 0.690*** SMB 0.152*** SMB_European 0.078*** SMB 0,.122*** HML -0.112*** HML_European -0.182*** HML -0.112*** LogSize -0.006*** LogSize -0.003*** WML -0.043***

NOSC 0.002*** NOSC 0.001*** LogSize

-0.003***

NOSC 0.001***

Adjusted R2 0.327 0.337 0.340

Observations 29,277 55,496 55,495