The determinants of ETF returns and
regular mutual fund returns
“Finding true costs on the Dutch
market”
by
Emiel Hendrikus Antonius Logtenberg
Butjesstraat 17 k13 9712 EW
Groningen, the Netherlands e.h.a.logtenberg@student.rug.nl
S1628100 Masterthesis Finance Rijksuniversiteit Groningen Thesis Supervisor: Dr. A. Plantinga
Abstract
This paper studies the descriptive as well as the predictive ability of different determinants on fund performance in the Dutch market. The sample consists of 44 ETFs and 118 regular mutual funds, and the study is conducted over the volatile period 2006-2011. This study concludes that expensive ‘active’ fund management is not necessarily inefficient management, for ETFs expensive management even returns extra value. In addition, we find a negative predictive impact of persistence on both ETF performance as well as regular mutual fund performance. Finally, we could not find any relationship between turnover and the performance of the funds in the sample. Different robustness analyses support the aforementioned results.
Keywords: ETFs, mutual funds, risk-adjusted return, expense ratio,
active expense ratio, persistence and turnover
Preface
With proud, I hereby present my master thesis in Finance, which finalizes my 6 year long journey through the academic world of the Rijksuniversiteit of Groningen. During my master programme as well as my bachelor programme I have met inspiring people and have participated in many challenging events and assignments. I have experienced this master thesis as probably the most challenging of all. Therefore, I wish to express my gratitude to those who enabled me to complete this Master thesis.
First of all, my gratitude goes out to my 1st supervisor, Dr. Auke Plantinga, who supported me during the complete route. During difficult moments or when facing seemingly unsolvable problems, he always provided me a way out or a little help to remain on the right track. Without his help, the thesis would not have been what it is right now.
Secondly, I would like to thank my parents and little brothers for their mental support, their financial support during my study and most important of all, because they always support me in my choices.
Thirdly, my gratitude goes out to my friends with whom I have had and still have a great time. I think we inspired each other, and are capable of pushing each other to great achievements in the coming years. Furthermore, I would like to thank Tom, Maikel and Stefan in particular for their support on my thesis.
1.Introduction
Currently, Exchange Traded Funds (hereafter called: ETFs) are outpacing regular mutual funds in terms of growth rates on the investment market. Over 2011, the total amount invested in regular mutual funds in the US decreased by 1.7%, while on the other hand, the total amount invested in ETFs in the US increased with 5.6%1. A mutual fund is an investment vehicle that is made up of a pool of different securities, and it can only be traded once a day. An ETF is a particular type of a mutual fund, and can be defined as follows:
‘An exchange traded fund (ETF) is an investment vehicle that tracks specific indices, like index mutual funds. Like stocks, they can be bought and sold on an exchange throughout the trading day. In 2008, the SEC (US securities and exchange commission) authorized the creation of actively managed ETFs as well2.
The first ETF was launched in 1993, being the SPDR (the Standard & Poor’s Depository Receipts) to track the S & P 500 index. Less than 20 years later, at the end of 2011, the amount invested in ETF assets in the US was nearly $1.05 trillion compared to a total amount of$11.60 trillion in mutual funds1. There are two obvious explanations for the growth in ETFs. First explanation is the disappointing performance of many regular mutual funds in the past decades (Carhart, 1997;Wermers, 2000;Fama and French, 2010;Glode, 2011). The second explanation for ETFs being popular are their low expense ratios compared to the expense ratios of regular mutual funds (Gastineau, 2005;Kissel, 2006). Furthermore, the crisis caused a trend toward a back-to-basic environment in which simple products will prosper3. Next to being simple, the investment options in regions and sectors are numerous (Haslem and Gastineau, 2010). It is needless to say that in the US, as well as in Europe4, ETFs are perceived by investors as an increasingly attractive investment opportunity.
A problem is that a regular mutual fund might be an index fund in disguise. This phenomenon has been defined as ‘index hugging’. Index huggers are active managers who invest close to the index to avoid underperformance (Ineichen, 2001). These funds do demand active fees for their replicating strategies (Statman, 2004;Surz, 2010). As a result, the expense ratio of this fund might not reflect the actual activity of management of the fund.
1 http://www.ici.org/pdf/2012_factbook.pdf
2 Actively managed ETFs are not included in our sample.
An index hugger can exist due to intransparent fee structures (Statman, 2004). The expense ratio consists of management fees, marketing and distribution costs, and because its exact composition is often unknown, index huggers get away with high fees for doing ‘close to nothing’.
To measure the true costs of the fund, we will test the impact of the expense ratio and the active expense ratio on the risk-adjusted performance of a fund. The active expense ratio measures the expenses relative to the actively managed part of the portfolio. As a result, an index hugging fund has a high active expense ratio, since their little activity is relatively expensive. When these true costs of funds are published, index huggers can be confronted and better investment decisions can be made. We will test the impact of the above-mentioned variables on ETF performance as well as on regular mutual fund performance. Therefore, the first research question is:
1) What is the influence of the expense ratio and active expense ratio on the risk-adjusted performance of ETFs as well as on regular Dutch mutual funds?
The issuers of the ETFs, define their products as pure trackers, following an index or a sector. Therefore, you would expect a lower average turnover of ETFs in comparison to a regular mutual fund. A higher turnover does bring extra transaction costs however, this does not mean that a high turnover of a fund is wrong. An example is the Rock Canyon Top Flight fund, which in 2003 had a turnover ratio of 3111%, and ended the year with a 52.8% return, beating 91% of its peers5. Therefore, the second research question of this study is:
2) What is the influence of the turnover ratio on the risk-adjusted performance of ETFs and regular Dutch mutual funds?
Next to the influence of the turnover ratio, we will also test for persistence of performance. Amongst others, Ferreira et al. (2010) find a positive relation for the persistence of performance of mutual funds. However, the studies done so far have mostly covered timeframes without severe crises. The timeframe of our study runs from 2006 till 2011, and therefore includes some volatile years. The question is what the influence is of these volatile years on the sign of the persistence factor. This thesis will test, the impact of the persistence of performance on the risk-adjusted performance of the fund.
Therefore, the third research question of this study is:
3) What is the influence of the persistence of performance on the risk-adjusted performance of ETFs and regular Dutch mutual funds?
The strength of this paper is the comparison between the ETFs from iShares, tradable on the Dutch market, with regular Dutch-based mutual funds. All the regular mutual funds are actively managed. The results from examining the ETFs are directly benchmarked against results from regular mutual funds. The reason for doing this is two-folded. Firstly, comparable evidence from ETFs is rather scarce, since ETFs are relatively new. Secondly, the timeframe of our research runs from 2006 to 2010. This period is characterized byone of the largest financial crises the world has ever felt, with bull and bear markets of great severity.
This study is a follow-up on past research, in a sense that it expands the widely discussed active/passive management of mutual funds debate. As mentioned before, this study replaces index funds for the relatively new ETFs. This paper examines the descriptive as well as the predictive ability of different variables.The added value of this research can be found in different aspects of the study. Firstly, it appears that worldwide, except for Klee and Gup (2009), nobody examined the determinants of ETF returns. Secondly, there is no other paper on ETFs subjecting the Dutch market. Thirdly, this paper gives a better insight into the ‘true’ costs of funds. It also has practical significance, since it presents a framework to expose index hugging funds.
2. Literature review
This section discusses literature relevant for answering our research questions. First, the expense ratio and the active expense ratio and their impact on performance is discussed. Secondly, the other explanatory variables, respectively, turnover and persistence are also discussed. To conclude, an overview of the overarching debate about passive and active management is presented.
2.1 Expense ratio
The expense ratio consists of management fees, marketing costs and distribution costs. Transaction costs are not included in the expense ratio. There are large differences in fees between funds. Fund fees vary across countries and across different sizes (Ferreira et al., 2010). Furthermore, index funds, mix funds and funds offered cross-country, charge lower fees then funds distributed in more countries or funds domiciled in offshore locations (Khorana et al., 2006). Amongst others, Moran (2001) presents evidence that the average expense ratio of an ETF is lower than the average expense ratio of an actively managed mutual fund.
The empirical evidence on the relationship between fees and performance collected in the past is predominantly negative. Amongst others, Carhart (1997), Gruber (1996), Bogle (1998), Edelen et al. (2007), Gil-Bazo and Ruiz-Verdú (2009) and Ferreira et al. (2010) conclude that expense ratios have a negative impact on the performance of a fund. Otten and Bams (2002) find the same negative relation for Dutch mutual funds. For ETFs, Klee and Gup (2009) examine all US stock ETFs, and find a significant negative relationship between the expense ratio and performance.
Ippolito (1989) finds no relation between higher management fees and performance. On the other hand, Malhotra and McLeod (1997) conclude that bond funds with higher expense ratios achieved higher yields. Moreover, Droms and Walker (1996) also find a significant positive relation between fees and performance.
This paper tests the following 0 hypotheses:
H(a)0 : The expense ratio is not related to risk-adjusted ETF performance and
H(b)0 : The expense ratio is not related to risk-adjusted mutual fund performance
We will also examine the predictive ability of the expense ratio:
H(d)0 : The one-year lagged expense ratio is not related to risk-adjusted mutual fund performance
2.2 Active expense ratio
Not many papers have made an attempt to decompose the fee structure of a fund. Swedroe (2001) constructs an example that uses a variance decomposition to directly estimate the passive and active funds under management. However, this approach tends to generate passive and active shares that are inconsistent with a replicating portfolio, and therefore overestimate the implied expense ratio.
Miller (2007) also expands the concept of the expense ratio by developing a simple method that computes the implicit cost of active management. Miller (2007) takes a portfolio and decomposes it into a purely active and a purely passive component. By isolating the active component, and assigning an active expense ratio to this component, he claims the true cost of active management can be estimated. The active expense ratio compares the fund to the appropriate benchmark, and gives a certain price to the activity of the fund. Asness (2004) did the same, however he assumed that the asset-by-asset holdings of each fund were known. This method is then used to compare the fund to the benchmark. Miller (2007) does this by regressing the fund performance on the benchmark performance, and taking its R².
Miller (2007) applies the method to different mutual fund classes of the Morningstar universe, and for all classes the costs of active management are found to be substantial. A high active expense ratio should be interpreted as follows; it gives activity a certain price. It is important to notice that it is not a measure of how active a fund is. A fund can be very active, and having a low active expense ratio, which would be a reflection of ‘cheap activity’. This thesis tests for the effects of this proxy on the performance of the funds:
H(e)0 : The active expense ratio is not related to risk-adjusted ETF performance and
H(f)0 : The active expense ratio is not related to risk-adjusted mutual fund performance Again, next to testing its descriptive ability, its predictive significance is also tested.
H(g)0 : The one-year lagged active expense ratio is not related to risk-adjusted ETF performance
and
It appears, that next to the evidence discussed above, no one has decomposed the expense ratio, or find another way to expose any index hugging fund. Research so far, has only named index huggers and explain what they do (Ineichen, 2001;Statman, 2004;Surz, 2010).
2.3 Turnover
The turnover of a fund is a measure of activity in a fund. Turnover is a percentage, indicating the percentage of the assets of the fund that has changed over the year. It also gives an idea about the transaction costs of the fund. The more active the fund, the higher its transaction cost (Cuthbertson et al., 2010).
In what appears to be the first paper studying the impact of turnover ratio on performance, Ippolitto (1989) findsno relation between turnover ratio and risk-adjusted performance using a CAPM model to calculate for risk-adjusted performance. Chen et al. (2000) find a significant positive relation using a variety of benchmark portfolios. Huang et al. (2011) investigates the relationship between risk-shifting behaviour and fund performance, and they also control for turnover. They conclude that funds with higher turnover tend to exhibit slightly higher performance.
Kaushik and Barnhart (2009) conclude that higher turnover is in some occasions positively related to winner portfolios. According to them, a high turnover can be a winning formula; it all depends on the skills of the manager to pick the right stock. On the other hand, Elton et al. (1993), Carhart (1997) and Edelen et al. (2007) conclude that high turnover has a negative impact on fund performance. When trading is based on noise, or to attract investors, trading only results in higher costs, with as a consequence a negative impact on fund performance (Wermers, 2000).
Furthermore, amongst others, Droms and Walker (2001) reported no significant results when testing the persistence of mutual fund operating characteristics, with amongst these, the turnover ratio.
Klee and Gup (2009) examines all US stock ETFs and finds no significant relationship between turnover and ETF returns for the time period examined. Moran (2001) indicates that the average ETF turnover is lower than the turnover of an actively managed mutual fund. To our knowledge, the relationship between turnover ratio and performance is not examined on the Dutch market.
Our paper tests the following 0 hypothesis against its two-sided alternative hypothesis:
H(j)0 : Turnover ratio is not related to risk-adjusted mutual fund performance
Then, to find out whether turnover ratio has predictive value, the following hypothesis is also tested against its two-sided alternative hypothesis:
H(k)0 : The one-year lagged turnover ratio is not related to risk-adjusted ETF performance and
H(l)0 : The one-year lagged turnover ratio is not related to risk-adjusted mutual fund performance
2.4 Past performance
The persistence of performance, which is the ability to outperform over a longer period of time, has received considerable attention in the past. It is important to recognize the difference between the one-year momentum effect (Jegadeesh and Titman, 1993) and persistence of performance. Carhart (1997) re-models the Fama and French (1992) model by including the momentum factor (Jegadeesh and Titman, 1993) to estimate the risk-adjusted performance. The momentum factor captures the effect that the best performing stocks of one year, are often the best performing stocks of the following year (Wermers, 1997). Performance persistence on the other hand, is the ability of portfolio management to outperform over several years.
Hendricks et al. (1993) and Brown and Goetzmann (1995) both findevidence for the so-called ‘hot hands effect’, meaning that funds with the best past 3-month performance perform significantly better the next year. Bollen and Busse (2005) use quarterly measurement periods, and daily returns, and find a positive relation for persistence on short-term returns. This disappears when funds are evaluated over longer periods. Next to the evidence on short-term performance persistence, Grinblatt and Titman (1992) and Elton et al. (1995) conclude that performance persist over longer periods of time as well. Furthermore, Otten and Bams (2002) find a significant relation between the one-year lagged performance and the performance of UK funds. More recently, Ferreira et al. (2010) present strong evidence in favor of performance persistence, especially for US funds.
It appears performance persistence is not studied in published papers on Dutch funds. Obviously, the influence of the persistence of performance can only be tested in a predictive model. Therefore, we will test for the predictive ability of performance persistence with the following hypotheses.
H(o)0 : The one-year lagged performance is not related to risk-adjusted ETF performance and
H(p)0 : The one-year lagged performance is not related to risk-adjusted mutual fund performance
2.5 Performance of Active versus Passive management
The influence of the above-mentioned fund characteristics on the risk-adjusted performance of funds is an integral part of the active or passive management debate. In general, actively managed funds charge higher fees than passively managed funds (Cuthbertson et al., 2010). Therefore, to remain an attractive investment vehicle, an actively managed fund should at least outperform the passive fund by such an amount that it covers for fees and expenses. Outperformance by active management is created by either picking the good-performing stocks or by picking and selling the stocks at the right time, or a combination of both (Cremers and Petajisto, 2009).
To invest passive or active, is one of the most important questions an investor has to consider (Surz, 2010). Wermers (2000) summarized USequity fund research as follows “the majority of the studies conclude that actively managed funds underperform their passively managed counterparts”. Carhart (1997) goes a step further, when showing that the more active a mutual fund manager trades, the lower the funds net return to investor. In 1975, far before ETFs were introduced, Ellis (1975) appointed the term ‘loser’s game’ to describe the function of an active fund manager. This view holds that it is impossible for managers to repeatedly beat the market, since it entails not only making good decisions, but also avoiding bad ones. Bogle (1998), the founder of the Vanguard Group6, compared active fund management with the child’s game tic-tac-toe. Each player can simply block an opponent’s previous move, which makes it a game that cannot be won. Bogle (1998) states that it is folly to believethat actively managed funds beat passively managed funds over a longer period of time.
In an earlier stadium, Grossman and Stiglitz (1980) concluded that active information-gathering investors can earn excess returns, however these are off-set by the costs of gathering this information, this is called the Grossman-Stiglitz rationality. Wermers (2000) find similar results, stocks of actively managed funds do outperform the
6
corresponding indices, however, due to expenses and transaction costs they underperform the market on net results with 1%.
Up to this point, the literature discussed all favored passive management over active management. However, evidence of active funds outperforming passive funds exists as well. Grinblatt and Titman (1993) who measured portfolio performance without a benchmark, conclude that aggressively active managed funds earned significant positive risk-adjusted returns. Next to Grinblatt and Titman (1993), Minor (2001) and Harlow and Brown (2008) present similar results showing active funds outperforming index funds. Cremers and Petajisto (2009) conclude that the most active stock pickers tend to create value. Barras et al. (2010) find that there is a minority of truly underperforming active funds, which are in a long-term struggle hoping to ultimately survive. However, most actively managed funds provide positive returns. Wermers and Yao (2010) more specifically state that active funds are drawn to the same stocks as passive funds, and active funds increase the price efficiency of stocks through their trades. Increasing price efficiency simply means selling high and buying low.
Papers discussing the active/passive debate and sampling the Dutch market are rather scarce. In 1998, Ter Horst et al. performed a style analysis over 289 Dutch mutual funds in the period 1990 – 1997. Although most funds show underperformance compared to their benchmark, Dutch funds with the objective to invest in ‘Dutch Equity’ outperform index funds tracking Dutch equity. It appears that the only other paper covering the Dutch market was published in 2002 by Otten & Bams. Otten & Bams (2002) investigate mutual fund performance and the influence of fund characteristics on risk-adjusted performance in the five most important European mutual fund markets. The significance of the results from the Dutch funds is questionable since only 9 Dutch funds are included in the sample. Their main conclusion is that except for the German funds, most European funds sampled outperform the market significantly. Especially, actively managed small cap funds are able to add value.
Investment Trust. This structure requires all received dividends to be held in a non-interest bearing account, this causes a ‘cash drag’ which regular mutual funds do not experience.
Furthermore, Gastineau (2004) presents evidence indicating that the ETFs of the Russel 2000 Small Cap and the S&Ps 500 Large Cap indices are underperforming their mutual funds counterparts.
Milonas and Rompotis (2006) and Blitz et al. (2010) come with evidence from the European market. Milonas and Rompotis for example, studied Swiss ETFs, and find the ETFs underperforming towards comparable index funds. Furthermore, ETFs are more risky since they can be traded more frequently than index funds. Results of Blitz et al. (2010) from the Irish and German market indicate that index funds as well as ETFs underperform their benchmarks with 50 to 150 basis points. Fund expenses as well as dividend taxes are explanations for this underperformance.
On the other hand, Svetina and Wahal (2008) study all ETFs at the end of 2007 worldwide. They findthat the ETFs that do compete with index funds deliver at least comparable performance and offer the advantage of immediacy. Immediacy can be defined as directly tradable at hours when the stock exchange is open. Harper et al. (2006) compare closed-end country funds with ETFs available for foreign markets, and conclude that ETFs exhibit a higher mean return then the closed-end funds. Furthermore, Klee & Gup (2009) find a significant positive relation between quantitative selection strategies/fundamental weighting strategies and ETF performance.
3. Methodology
As discussed in the literature section, this thesis will test for the influence of different fund characteristics on the risk-adjusted performance of the funds in the sample. In the sections 3.1 and 3.2 we will explain the different regressions we will run to find an answer to our research questions. We will also explain how the different characteristics are constructed. In section 3.3 we will explain how we have computed the depend variable, namely the risk-adjusted performance. This measure of performance is, as the name expresses, adjusted for risk.
3.1 Descriptive effect of non-risk variables
In order to investigate whether any of the fund characteristics has a significant influence on the risk-adjusted performance of a fund, either actively or passively managed, an OLS regression analysis will be executed.
Concerning the data, both time-series as well as cross-sectional dimensions of data are used. Not all the data is complete, therefore, our data is an unbalanced panel.
Panel data techniques can roughly be divided in three categories, a regression either allowing for no effects, fixed effects or for random effects. The ‘regular’ pooled regression can be executed when the necessary tests find no fixed or random effects in the sample. The fixed effects model allows the intercept in the regression to differ cross-sectional, however not over time. All slopes are fixed both cross-sectionally and over time. The time-fixed effects model is the other way around; it allows the intercept to differ over time, however not cross-sectional. To test whether it is necessary to allow fixed effects, a Redundant Fixed effects – Likelihood Ratio test will be performed. The other category, the random effect models assume the intercepts for each cross-sectional unit to arise from one common intercept. Since fewer parameters have to be estimated in this model, it saves degrees of freedom and can therefore be more efficient. However, a random effects model can only be used when the error term is uncorrelated with all of the independent variables. To test whether this is actually the case, the Hausman test is done (Brooks, 2008). After testing for fixed effects, it could be necessary to test for random effects with this Hausman test. The outcomes of these tests determine which technique is used for executing our regression.
To test our hypotheses, the following regression is executed in EVIEWS:
RAPi,t = α + β1TUR i,t + β2TER i,t + β3LN(SIZE) i,t
+ β4LN(AGE) i,t + β5ATER i,t + β6RFLOW i,t + εi,t (1)
standard errors and sensitive coefficients, which make it difficult to draw any conclusions about significance (Brooks, 2008). Therefore, correlated explanatory variables will not be in the same regression. Consequently, multiple regressions have to be executed. This has as additional advantage that the variables are tested multiple times, and are therefore immediately tested on robustness. Risk-adjusted performance (hereafter named RAPi,t) is the alpha estimated with the risk-adjusted performance model. Furthermore, α is the constant and εi,t is the disturbance term. This term captures e.g., variables omitted from the model, outside influences and other variables.
In order to answer the first research question, the impact of the expense ratio and the active expense ratio on the RAPi,t will be examined. The TER is the published expense ratio of the fund. The factor ATER, is the active expense ratio of Miller (2007), who claims that this factor computes the implicit cost of active management. The factor consists of the fund published expense ratio, the expense ratio of the benchmark index and the funds R² relative to its benchmark index. The R² is derived from an OLS regression regressing the fund on its benchmark. It isolates the passively invested share of the fund, and thereby with 1-R² also the actively invested share of the fund. Miller (2007) himself, did not test the active expense ratio as an explanatory or predictive variable explaining risk-adjusted fund performance. However, when fund management is expensive, you would like to know whether you receive the appropriate premium for the fee paid. Therefore, this thesis tests the active share measure of Miller (2007) on its descriptive as well as its predictive ability.
The active expense ratio is defined as follows:
ATER = expense ratio of the benchmark + expense ratio premium * 100% (2) % actively managed
Secondly, our second research question examines the influence of the turnover ratio and persistence. The turnover ratio(TUR) is defined as follows:
TUR = (Annually holdings bought + sold of fund) – (issues + intakes) (3) Yearly average total holdings of fund
The influence of persistence can only be measured in predictive regressions. This factor will be explained in section 3.2. The variables SIZE (Net Asset Value of the fund), AGE (age of the fund) and RFLOW (relative flow) are used as controlling variables in the different regressions. The last factor, RFLOW, can be defined as the inflow or outflow of a fund in a certain year relative to the size of the fund. Relative flow is calculated as follows:
RFLOW = - ((Net Asset Value t-1 * ( 1 + annual return)) – Net asset value) (4)
SIZE
3.2 Predictive effect of non-risk variables
Next to the descriptive power of the fund characteristics, this paper also tests for the predictive effect of the different fund characteristics on returns. The influence of 1-year lagged values of the explanatory factors on the current risk-adjusted performance is measured.
As mentioned before, one factor is added to the regression, this is β7PERSt-1. This is the
persistence factor, which measures how persistent a fund is to previous performance. The persistence factor is the β7 measured by testing the influence of the risk-adjusted performance of the previous year on the risk-adjusted performance of this year. How this risk-adjusted performance is estimated is explained in section 3.3.
The regression examining the predictive ability of the characteristics can be expressed as follows:
RAPi,t = α + β1TUR t-1 + β2TERt-1 + β3LN(SIZE)t-1 + β4LN(AGE)t-1
+ β5ATERt-1 + β6RFLOWt-1 + β7PERSt-1 + εi,t (5)
3.3 Risk-adjusted performance
In order to explain differences in performance between funds, we need to correct for differences in systematic risk. In the past, several researchers like Jensen (1969), Sharpe (1964), Fama & French (1992) and Carhart (1997) have constructed models to estimate the risk-adjusted performance (Alpha (αI)). A positive alpha suggests that the fund has outperformed the market for a given riskiness, and a negative alpha suggests the fund has earned below the market required return for a fund given this riskiness. Past research has considered both the Carhart (1997) four-factor model and the Fama & French (1992) three-factor model as appropriate models for estimating risk-adjusted performance. Fama & French claim that their model can explain up to 90% of the diversified portfolio returns (Fama & French, 1992). Carhart (1997) claims that his model almost completely explains persistence in mean and risk-adjusted returns. This thesis will use the model which has the best fit (highest R²) with the sample. The Carhart (1997) model is given by equation 1. The Fama & French (1992) model is the same model without the (MOM) momentum factor:
Ri,t -Rf,t = αI +β1* (Rm,t -Rf,t)+ β2*SMB + β3*HML + β4*MOM + εi,t (6)
The factors can be defined as follows; Ri,t is the return of the fund at week t. Ri,t is calculated as a weekly continuously compounded return:
Ri,t = 100% * ln(RI t/RI t-1) (7)
Ri,t = Continuously compounded return at week t ln = Natural logarithm
RI t = Indexed return7 of the mutual fund at end of week t
Continuously compounded returns are used because they are time-additive. This means that a two period log return is identical to the sum of each periods log return. Furthermore, taking a log rescales the data in such a way it weakens extreme observations, and thereby brings the returns closer to normal distribution (Brooks, 2008).
Rm,t -Rf,t is the market return exceeding the risk-free rate, the risk-free rate being the one-week euribor rate. The market return is the return on the MSCI with the best fit with the sample.
SMB is the market cap risk factor; meaning the difference in return between small cap and large cap portfolios. SMB, or small minus big, measures the additional return investors have received in investing in stocks of companies with relatively small market capitalization. HML is the book-to-market risk factor, which can be defined as the difference in return between high book to value stocks and low book to value stocks. HML, or high minus low, measures the extra return earned by investing in high book-to-market stock. MOM, which is the momentum factor, is controlling for a strategy to buy last winners and sell last losers, and only appears in the Carhart model. MOM measures the extra earnings an investor has who is buying last year’s winners, and selling its losers. This captures the Jegadeesh and Titman (1993) momentum anomaly.
The funds in our sample are each investing in different parts of the world. Therefore, we will construct 4 different geographical models. The benchmarks considered are the MSCI Netherlands, the MSCI Europe, the MSCI World and the MSCI emerging markets. The one with the highest explanatory value (highest R²) with the sample is used to estimate the risk-adjusted performance of the funds. The idea is that each fund manager chooses their own benchmark to compare its performance with. When we compare each fund with its own benchmark, it is difficult to compare the different performances against each other. In order for us to compare the performance of the funds we choose for one benchmark, the one best fitting the funds in the sample. The other models are used for robustness checks.
3.4 Robustness checks
All regressions executed are tested on their robustness. Such a robustness check examines how certain ‘core’ regression coefficient estimates behave when the regression specification is modified by adding or removing regressors. If the coefficients are plausible and robust, this is commonly interpreted as evidence of structural validity (Brooks, 2008).
4. Data
This section discusses the data used in this study. First the construction of our sample is described. Secondly, the different fund characteristics are discussed, and their descriptive statistics are presented. Thirdly, the data used for estimating the risk-adjusted performance is explained.
4.1 Sample description
The sample consists of 162 funds traded on the Euronext Amsterdam. The data describes the time period from January 2006 untill December 2011. The dataset is constructed from an already existing database prepared by Mr. van Leeuwen and Dr. Plantinga. This database consisted of around 400 known mutual funds traded on the Dutch market. Furthermore, the expense ratios for most of these funds for the period 2003 – 2008 were already in this database.
For the purpose of our study, the existing database had to be adjusted in several ways. First of all, ETFs regularly trading on the Dutch market had to be added to the database. 44 ETFs from iShares, BlackRock, already present on the Dutch market before the 1st of January 2008 are included in the sample. The ETFs of iShares are chosen because these trackers are most frequently traded on the Dutch market and BlackRock is the worldwide number 1 issuer of ETFs.
Secondly, several funds were dropped for different reasons. For example, all funds that started after 1 January 2008 are excluded from the sample because not enough data is available. For mix funds, it sometimes was too difficult to find turnover and expense ratio information, and some of these funds are excluded.
Ultimately, enough information was available for 118 Dutch-based mutual funds and 44 iShares ETFs, resulting in a total sample of 162 funds. Eight of these funds ceased to exist during the time frame of our study. In previous research, many of the datasets were biased because they left out funds that ceased to exist during the timeframe of the research; this is called survivorship bias (Brown et al., 1992)(Otten & Bams, 2002). This bias can influence the results severely, simply because bad performing funds are frequently being liquidated or merged into another one. When funds disappear in our sample because they ceased to exist, the portfolios are re-weighted accordingly. An overview of the ETFs is provided in appendix 1, while the regular mutual funds overview is presented in appendix 2. As mentioned, all the regular mutual funds are actively managed mutual funds.
The main reason for choosing a timeframe from 2006 to 2011, is because it is rather difficult to find annual reports before 2006 for most of the funds in our dataset. Furthermore, most ETFs only exist since a few years. This time period encompasses bullish and bearish trends because of the financial crisis. This is taken into consideration, and this problem is somewhat circumvented by comparing the return of the funds with each other and with the corresponding market returns.
4.2 Fund characteristics
Table І. Definition of fund characteristics This table presents the definitions of the different fund characteristics, which are tested for their
potential influence on the return of the sample funds.
Fund Characteristic Definition
Dependent variable:
Risk-adjusted fund performance Alpha (% per year) estimated from the Fama and French model using weekly returns
Independent variables:
Expense ratio Total annual expenses measured as ratio of Net Asset Value. Expense ratio covers all operating expenses.
Active Expense ratio The expense ratio for the active part of the fund measured against its benchmark
Relative Flow The inflow or outflow of money into a fund on yearly basis divided by the size of the fund
Turnover ratio The difference of purchases and sales minus the difference of issuances and redemption of shares, as a percentage of the average assets over the year.
Ln(Net Asset Value (Size)) The natural logarithm of (total assets of fund – short term liabilities, measured in Euros)
Ln(Age) The natural logarithm of (number of years since establishment fund, measured in years)
Persistence Is the one-year lagged Alpha (% per year) estimated from the Fama &
French model Persistenceሺܽ௧ሻ = Fund performanceሺܽ௧ିଵሻ
This section continues with the descriptive statistics of both the returns of the funds in the sample as well as the data on their fund characteristics. Table 2 describes the statistics for the ETFs in the sample, and table 3 describes the statistics of the regular mutual funds. The overview of the descriptive statistics already presents large differences between ETFs and the regular Dutch mutual funds. The table displays a higher mean return over the whole period for ETFs compared to regular mutual funds. However, in the last 2 years, 2010 and 2011, regular mutual funds perform on average better than ETFs. This trend is supported by the data on persistence, which actually are the one-year lagged alphas.
Regarding the cost of funds, as expected, ETFs are indeed less expensive than the regular mutual funds. Therefore, on average, costs in general as well as the costs of activity are much higher for regular mutual funds than for ETFs. This evidence suggests that ETFs do track their indices as they are promising to. As mentioned, a high active expense ratio for an regular fund may indicate the presence of ‘index huggers’.
turnover in the last year for ETFs as well as regular mutual funds. This might suggest that fund managers have learnt a certain lesson from the crisis years, and start to prefer a more passive approach.
The impact of the financial crisis is obvious, e.g. the low mean returns in 2008 and the large decrease in average size of a regular mutual fund in that same year. The negative returns of 2011 might even suggest the presence of “the double dip effect”. The volatile years have their impact on the calculated alphas, which vary greatly in sign and value. This should be taken in consideration when interpreting the results of this thesis.
Table ІІ. Summary statistics of ETFs in the sample for each year This table presents the summary statistics of the ETFs in the sample. The data regarding returns is weekly data, and the data regarding fund characteristics is yearly data.
Funds in Sample 44 ETFs
Period 01/01/2006 – 31/12/2011
Weekly data ETFs
2006 2007 2008 2009 2010 2011 2006-2011 Mean Ret 0,30% 0,27% -0,71% 0,60% 0,21% -0,27% 0,04% Maximum 10,87% 18,36% 32,54% 53,18% 8,39% 11,88% 14,73% Minimum -13,00% -14,84% -42,06% -18,03% -15,13% -14,63% -24,51% Std. Dev. 2,28% 2,79% 6,07% 4,39% 2,82% 3,36% 3,83% Persistence NA 0,27% 0,15% 0,34% 0,08% 0,00% 0,21% Annual data
Av. Expense ratio 0,53% 0,53% 0,53% 0,53% 0,53% NA 0,53%
Av. Active Exp. rat. 0,04% 0,57% 0,80% 0,60% 0,61% NA 0,55%
Av. Turnover ratio NA 57,76% 57,09% 63,89% 45,04% NA 55,80%
Flow NA 85,87% 210,69% 230,19% 118,97% NA 165,92%
Av. Size (in mln Euros) 502,92 518,03 536,09 892,81 1110,02 NA 725,00
Av. Age (in years) 1,32 2,07 3,02 4,02 5,02 NA 3,10
Av. Number of funds in
Table ІІІ. Summary statistics of the regular mutual funds in the sample for each year
This table presents the summary statistics of the regular mutual funds in the sample. The data regarding returns is weekly data, and the data regarding fund characteristics is yearly data.
Funds in Sample 118 Regular mutual
Funds
Period 01/01/2006 – 31/12/2011
Weekly data Mutual Funds
2006 2007 2008 2009 2010 2011 2006-2011 Mean Ret 0,19% -0,01% -0,85% 0,49% 0,26% -0,14% -0,01% Maximum 11,68% 11,01% 20,05% 14,14% 15,49% 9,99% 8,31% Minimum -12,07% -14,49% -25,53% -19,04% -20,43% -15,31% -12,35% Std. Dev. 01,67% 1,96% 3,73% 2,72% 1,99% 2,45% 2,30% Persistence NA -0,08% 0,02% -0,28% 0,07% 0,01% -0,05% Annual data
Av. Expense ratio 1,30% 1,29% 1,33% 1,30% 1,34% NA 1,31%
Av. Active Exp. rat. 2,12% 2,20% 2,33% 2,59% 2,13% NA 2,28%
Av. Turnover ratio 140,49% 119,45% 127,43% 128,16% 99,81% NA 123,14%
Flow NA 116,26% -58,23% -22,29% -3,04% NA 5,72%
Av. Size 401,45 472,93 247,35 280,91 327,30 NA 345,60
Av. Age 9,94 10,90 12,01 13,02 13,96 NA 11,95
Av. Number of funds
in sample 111,20 114,40 114,10 113,50 115,30 112,30 113,50
4.3 Risk-adjusted performance
The risk-adjusted performance is estimated using one of the models, described in the methodology section. The weekly returns of the funds and the benchmarks are provided by Thompson Financial Securities Data. The returns are calculated weekly since this creates enough observations for having sufficient degrees of freedom for properly executing a regression with 3 or 4 factors. Next to this, investors normally do not step in and out of a fund on daily basis, therefore weekly returns are appropriate for this study.
will use a European Fama & French model, which is displayed bold in table 3. The results of regressing the sample on this model are provided in appendix 3. As mentioned before, the other models, including a model with a momentum factor, are used as robustness analysis. As shown by the tables, the estimates are consistent amongst the models, and are therefore considered robust. The complete models for robustness checks can be found in appendix 4.
Table ІV. Summary statistics for the model estimating risk-adjusted performance In constructing the risk-adjusted performance estimating model, the different corresponding MSCI
benchmark returns are considered. These are weekly numbers for the period 2006 -2011.
Cross-correlations
Factor Portfolio Coefficient St. Error R²
Market Premium SMB HML Netherlands 0.41 Constant 0.0002 0.0001 Market Premium 0.5403* 0.0041 1,00 0,67 -0,10 SMB 0.2639* 0.0054 0,67 1,00 -0,21 HML 0.0195* 0.0045 -0,10 -0,21 1,00 Europe 0.44 Constant 0.0001 0.0001 Market Premium 0.6487* 0.0186 1,00 0,54 -0,25 SMB 0.2467* 0.0078 0.54 1.00 -0.19 HML -0.0248** 0.0112 -0,25 -0,19 1,00 Momentum -0.0260 0.0175 0.98 -0.16 -0.43 Europe 0.44 Constant 0.0001 0.0001 Market Premium 0.6216* 0.0037 1.00 0.54 -0.25 SMB 0.2459* 0.0078 0,54 1,00 -0,19 HML -0.0164 0.0097 -0.25 -0.19 1.00 World 0.41 Constant -0.0001 0.0001 Market Premium 0.5733* 0.0037 1,00 -0,30 0,27 SMB 0.1744* 0.0126 -0,30 1,00 0,28 HML 0.1133* 0.0126 0,27 0,28 1,00 Emerging Markets 0.38 Constant -0.0007* 0.0001 Market Premium 0.4733* 0.0030 1.00 -0.24 0.11 SMB 0.1335* 0.0078 -0,24 1,00 -0,40 HML 0.2180* 0.0174 0.11 -0.40 1.00
5. Results
This section covers the results from the regressions performed during this study. First, the final regressions ultimately conducted are explained. Thereafter the results from the descriptive as well as the predictive regressions are presented. The last sections of this chapter are different tests regarding the robustness of our results.
5.1 Regressions
As explained in the methodology, in order to draw conclusions about significance tests, the problem of near multicollinearity is ruled out. The correlation matrices of the explanatory variables of the ETFs as well as the regular mutual funds are displayed in appendix 5. The results show relatively high correlation between respectively size and age, and between the expense ratio and the active expense ratio. To solve the problem of multicollinearity, this thesis will run 4 different regressions, in which the correlated variables are not together in a regression. All variables are tested minimally twice in order to control for robustness. These regressions are separately run for the ETFs as well as the regular mutual funds. For the sake of clarity, table 4 displays these regressions.
Table V. The predictive regressions This table provides an overview of the conducted predictive regressions. The same regressions are conducted for descriptive purposes, only then without the persistence factor and with its regular factors instead of the one-year lagged value.
Predictive Regressions :
1) RAPi,t = α + β1TURt-1 + β2TERt-1 + β3LN(SIZE)t -1 + β4RFLOWt-1 + β5PERSt-1 + εi,t
2) RAPi,t = α + β1TURt-1 + β2ATERt-1 + β3LN(AGE) t-1 + β5PERSt-1 + εi,t
3) RAPi,t = α + β1TERt-1 + β2LN(AGE) -1 + β3RFLOWt-1 + εi,t
5.2 Results descriptive regressions
Table 5 presents the coefficients of the 4 descriptive regressions, with data corrected for skewness by means of dummies.
Table VІ. Explaining RAP from descriptive regressions This table presents the coefficients of the descriptive version of the regressions displayed in table 5. The
regressions are ran with a panel of 44 ETFs and 118 Dutch regular mutual funds with data from 2006 to 2010. The coefficients describe the impact of the variables on the depend variable, the risk-adjusted performance(RAP).
*significant at the 1% level. - Adj. R² = Adjusted R²
**significant at the 5% level. - DW = Durbin Watson statistic (of the panel)
The results of the descriptive regressions suggest a surprisingly positive impact of the expense ratio on the performance of the ETF. This means that the null hypothesis H(a)0, stating that the expense ratio is not related to risk-adjusted ETF performance, can be rejected. The result suggest that, more expensive ETFs are able to convert the paid fee into an actual higher return.
For the sample of ETFs, the active expense ratio influence is positive, though only in regression 4 this relation is significant. Since the results are not significant in regression 2, the null hypothesis H(e)0 cannot be rejected. The positive sign implies that the money paid for the activity is well-spent, since a higher return is received for the fee paid.
ETFs (Descriptive regressions)
Regressions 1 2 3 4 Constant -0.0002 0.0016 -0.0009 0.0001 p-value (0.8482) (0.0182)** (0.3368) (0.3278) TER 0.0037 0.0041 p-value (0.0011)* (0.0005)* ATER 0.0005 0.0007 p-value (0.0599) (0.0005) TUR - 0.0000 - 0.0000 - 0.0000 p-value (0.5216) (0.2045) (0.2352) RFLOW 0.0003 0.0002 p-value (0.2739) (0.4696) SIZE - 0.0000 - 0.0000 p-value (0.7517) (0.8344) AGE - 0.0005 - 0.0000 p-value (0.2155) (0.9068) Adj. R ² 0.35 0.36 0.31 0.32 DW 2.56 3.04 2.63 2.46
Regular mutual funds (Descriptive regressions)
Regarding these descriptive regressions, no significant impact of any fund characteristic is found on either ETF risk-adjusted returns or regular mutual fund returns. Therefore, all other null hypotheses regarding the descriptive ability of the different fund characteristics cannot be rejected.
5.3 Results predictive regressions
Table 6 provide the coefficients for the predictive regressions. A first glance teaches us that the different fund characteristics seem to have more predictive ability than descriptive ability.
Table VІІ. Explaining RAP from predictive regressions This table presents the coefficients of the predictive version of the regressions displayed in table 4. The
regressions are ran with a panel of 44 ETFs and 118 Dutch regular mutual funds with data from 2006 to 2011. The coefficients describe the impact of the one-year lagged variables on the depend variable, the risk-adjusted performance(RAP).
*significant at the 1% level. - Adj. R² = Adjusted R²
**significant at the 5% level. - DW = Durbin Watson statistic (of the panel)
The results in table 6 suggest the one-year lagged active expense ratio to have a predictive ability on ETF performance. We find a positive impact, and therefore the null hypothesis H(g)0, which states no relation between the one-year lagged active expense ratio and the risk-adjusted ETF performance, is rejected. This result suggest investors
ETFs (Predictive regressions)
Regressions 1 2 3 4 Constant 0.0946 -0.0013 0.1608 0.0038 p-value (0.4415) (0.4681) (0.2252) (0.1642) TER -0.1645 -0.3035 p-value (0.4844) (0.2307) ATER 0.0025 0.0020 p-value (0.0005)* (0.0045)* TUR - 0.0000 0.0000 0.0000 p-value (0.4168) (0.6836) (0.8127) PERSISTENCE -0.2439 -0.2432 p-value (0.0260)** (0.0024)* RFLOW 0.0003 -0.0001 p-value (0.3335) (0.7067) SIZE -0.0013 -0.0008 p-value (0.0039)* -0.0638 AGE 0.0004 - 0.0011 p-value (0.7691) (0.327) Adj. R ² 0.38 0.41 0.33 0.38 DW 2.27 2.46 2.90 2.72
Regular mutual funds (Predictive regressions)
could better invest a little extra for active management, because then they receive something in return. For the expense ratio, no consistent significant results are found.
Results suggest that the most significant variable predicting the alpha’s is the persistence variable. However, the impact is not positive, as would have been in line with previous research (Otten and Bams, 2002; Ferreira et al., 2010), it has a negative impact. The null hypotheses H(o)0 and H(p)0, stating no relation between persistence and performance, can therefore be rejected for ETFs as well as regular mutual funds. For the regular mutual funds the negative impact of persistence is even stronger than for ETFs. Actually, the negative relation might not be that surprising, since the financial crisis caused an extremely volatile era. Years of up and down movements in stock prices followed each other. This could be an explanation for the ‘reverse’ persistence found in this study.
For turnover ratio, no significant results are found, therefore the hypotheses regarding turnover cannot be rejected. This implies that the costs of a higher turnover can be offset by picking the right stocks.
Furthermore, in our sample, the controlling factors one-year lagged size and one-year lagged flow also seem to have a significant impact. The one-year lagged size of a fund seems negatively related to the corresponding alpha, this is not in line with previous research (Malhotra & McLeod, 1997) (Rea & Reid, 1998) (Latzko, 1999) (Otten & Bams, 2002). You would expect a larger fund to achieve economies of scale. The one-year lagged relative flow has a positive impact on risk-adjusted performance, this is more in line with expectations and past research (Gruber,1996) (Ferreira et al., 2010). This relation is better known as the so-called ‘smart money effect’.
5.4 Testing OLS assumptions
First, the OLS regressions ideally satisfy 5 assumptions(Brooks, 2008). In order to indeed satisfy these assumptions, we have performed tests for normality of the disturbances, heteroskedasticityand autocorrelation.
First, for most of the regressions, the DW (Durbin-Watson) statistic has a neat value of just above 2, implicating no evidence of autocorrelation. This statistic is displayed in tables 5 and 6.
Secondly, a diagnostic test requires a constant variance amongst the errors(homoscedasticity), therefore, a White’s test is done to test for heteroskedasticity for all the regressions. The coefficients remain the same under the White’s test as is displayed in appendix 8 and 11. This proves homoscedasticity respectively for regression 1 for regular mutual funds, and for regression 2 for ETFs.
Thirdly, the disturbances of actually all of the regressions allowing fixed effects, suffer from non-normality. This non-normality exists because of a non-normal kurtosis as well as a non-normal skewness. The skewness issue can be solved by removing outliers with adding some dummies to the different regressions. However, the kurtosis problem can, in this instance, not be solved. Normally, running a GARCH model is a solution for data suffering from kurtosis, however, with panel data this is not an option.
5.5 Robustness
To check whether the results survive the non-normality issue, 2 tests are performed. First, appendix 12 and 13 display the coefficients of the regressions without adjusting the data for skewness. The results are roughly the same, which can be interpreted as a sign of robustness. Next to this, an alternative robustness check is provided.
Table VІІІ. Robustness check
This table presents a robustness check for the signs of the coefficients of the different regressions.
Descriptive regressions Predictive regressions
ETFs Regular mutual
funds ETFs Regular mutual funds Active Expense ratio + = + = Log Age - + - = Log Size - + - - Persistence - - - - Relative Flow + = + + Expense ratio + - + - Turnover - - - +
6. Conclusion
This study present evidence about the impact of different fund characteristics on performance. This is done by examining the impact of different cost measures and other fund characteristics on the risk-adjusted performance of 162 funds. The sample includes 44 ETFs and 118 open-end actively managed mutual funds all traded on the Dutch market. Several fund characteristics are considered as potential determinants of fund performance: expense ratio, active expense ratio, turnover and persistence. The variables age, size and flow are included as controlling variables. This thesis tests for both the descriptive as well as the predictive ability of these fund characteristics.
The results suggest that the descriptive ability of the different determinants is rather low. Only the expense ratio, and to a lesser extent, the active expense ratio seem positively related to ETF performance. This is not in line with past research, which suggests a negative impact of the expense ratio on ETF performance (Klee and Gup, 2009). The negative impact could be due to rising interest in alternative investment options, because of bad results of regular investment options during the crisis years. The more alternative or exotic the ETF, the higher its expense ratio, this could be the reason for high expense ratios resulting in good performance. For the regular mutual funds, no significant relations were found.
The predictive regressions of this thesis generated more significant results. First, we found a positive predictive impact of the active expense ratio on risk-adjusted performance of ETFs. It implies that ETFs that demand higher fees have higher returns. Therefore, this study, with our version of the active expense ratio, generates an interesting input for the active/passive management debate. Despite higher active expense ratios for regular mutual funds than for ETFs, no relation is found with the risk-adjusted performance of the regular mutual funds. This suggests that index hugging funds do not perform worse than other funds.
In contrast with most of past research (Otten and Bams, 2002; Ferreira et al., 2010), our most significant result suggests a negative predictive ability of persistence on risk-adjusted performance of regular mutual funds; this is to a lesser extent also found for ETFs. As mentioned in the results, this outcome isprobably due to the financial crisis, which caused some volatile years. This is also in line with the descriptive statistics of our sample, which show negative average returns in 2008 and 2011, and positive returns in the other years. No significant results were found for the turnover ratio. To answer our second research question, only the persistence of performance has its influence on the risk-adjusted performance of ETFs as well as regular mutual funds. We think the negative sign is due to the volatile years of the crisis.
The other factors were used for controlling purposes in our regressions, and in some cases some surprising results occurred.
First, the results show a negative relation between the one year-lagged size factor and risk-adjusted performance for ETFs as well as for regular mutual funds. This is remarkable, since most past research show an opposite relation because of economies of scale. A potential explanation is suggested by Chen et al. (2004) and Indro et al. (1999). They show that when funds grow too large, they have to invest in less interesting investment opportunities, instead of only the best ideas. This causes liquidity constraints since these less interesting investment opportunities often mean small and illiquid funds. However, for ETFs, at least for the ones who are supposed to track an index, this is not be possible. Another explanation could be a consequence of the volatile years of the crisis. In bad years like 2008, most funds decreased in size. However, in 2009 performance was better than in 2008, and funds started growing again. This could be a reason for the negative sign of the size coefficient in the predictive regressions.
Furthermore, the results show a positive predictive ability of flow on ETF as well as regular mutual fund performance, just like Gruber (1996) and Ferreira et al. (2009). The descriptive statistics of our sample confirm the picture of a fast-growing ETF business with low visible costs compared to regular mutual funds.
6.1 Discussion
returns, and increases in size of a fund. A fund that performed very bad in 2008, might have done very well in 2009, just because they had a lot to recover. This explains the negative predictive ability of the factors persistence and size.
This paper concludes that in most occasions, qualitative and expensive fund management proves to be efficient management. This is interesting since it entails that active management does pay. Furthermore, the most-talented stock-pickers seem to be the ones that are the best paid-ones as well. Most past research concluded otherwise, favouring passive management over the, in general, more expensive active management. This paper presents a view favouring expensive and at the same time efficient management. For talented fund managers, this academic evidence justifies their demands for a good salary. Further research testing the active expense ratio should be able to support this.
In December 2011, the Dutch website “fondsnieuws.nl” published an article reporting that the growth of ETFs remained behind with goals set, mostly due to difficult economic circumstances. The goals of 20-30% growth proved not feasible. Furthermore, at the beginning of 2012 this same source reported that the American ETF market shows signs of saturation. New American ETFs are unable to attract enough investors to survive8. This trend is somewhat supported by the descriptive statistics of our sample. These statistics show that in 2010 and 2011, the regular mutual funds perform better than the ETFs. This might be another twist in the discussion about active or passive management and its true costs.
Another interesting fact is that ETFs sometimes display quite high turnover rates, which is also found in our sample. Although the strategy of traditional indexing has a number of attractive attributes, indices do change in composition. Therefore trackers do also change, and costs have to be made to bring about these changes. Therefore, despite being a passive fund, a high turnover might be required to actually keep up mimicking the benchmark. The point is that, where you would have expected a low turnover for passive funds, this might not always be the case. Turnover can certainly be considered as a good measure of activity of a fund. However, whether a fund is passively or actively managed, this does necessarily tell you anything about the turnover of the fund. Passively managed ETFs might still have quite high turnover rates. Or as Fuller, Han and Tung (2010) concluded, no investment strategy is truly passive. Further research of academics with accessibility to the necessary benchmark
8
information, creating a measure of activity outside “the benchmark mimicking activity” could prove to be very interesting.
Furthermore, in 2008 the SEC (US Securities and Exchange Commission) authorized the creation of actively managed ETFs. The first one was created in March 2008, and the group of actively managed ETFs have grown faster in its first 3 years than the indexed ETFs did in their first 3 years of existence9. Since the actively managed ETFs only came into existence in 2008, we could not include these ETFs into our sample. However, this could also be an interesting field of further research.
6.2 Limitations
For this study, the sample as well as the time-frame could be considered as quite small. This is mainly due to the very existence of ETFs. Further studies can go beyond this analysis by enlarging their sample and by adding years of data.
The different regressions lead to several statistical issues of multicollinearity, heteroskedasticity and non-normality. Multiple alternative tests, robustness checks were performed and dummies used to solve these issues. The majority of the issues were indeed mitigated. The kurtosis amongst the disturbances in the regressions could not be mitigated. In order to check the robustness of the non-normally divided estimators an alternative robustness check was performed. This test revealed that the main results are robust. In future, a model which could mitigate all issues could lead to even more reliable estimators.
This study attempts to obtain the active expense ratio of each fund. This is done by categorizing the funds in the sample into 6 groups and use this assigned group as benchmark for the fund. It is of critical importance to notice that the funds are not compared with their self-assigned benchmarks as Miller (2007) did. This has as advantage that funds cannot hide themselves behind the performance of their self-picked benchmark. However, it does give different results than when the regressions were ran with the self-assigned benchmarks and corresponding expense ratios. Especially for ETFs, which in our sample all track a sector or an index, the active expense ratio calculated does in most cases not represent the actual active expense ratio. Most probably, the active expense ratio as intended by Miller (2007) is very high for tracking ETFs. It would be very interesting to see whether the relationship between the active expense ratio and fund performance is still the same when funds are
7. References
7.1 Monographs
Brooks, Chris, 2008, Introductory Econometrics for Finance (2nd edition, Cambridge University Press, UK)
Gastineau, Gary, L., 2005, Someone will make money on your funds – Why not you?(Hoboken, NJ: John Wiley & Sons)
Haslem, John, A. & Gary L. Gastineau, 2010, Mutual funds: Portfolio structures, analysis, management and stewardship(John Wiley & Sons, Inc)
Swedroe, Larry, E., 2001, What Wall street doesn’t want you to know (New York: St. Martin’s Press)
7.2 Articles
Asness, Clifford, S., 2004, An alternative future’, Journal of Portfolio Management 31, 8-23.
Barras, Laurent, Olivier Scaillet & Russ Wermers, 2010, False discoveries in mutual fund performance: Measuring luck in estimated alphas, The journal of Finance 65, 179-216.
Blitz, David, Joop Huij & Laurens Swinkels, 2010, The performance of European Index Funds and Exchange-Traded Funds, European Financial Management,
DOI: 10.1111/j.1468-036X.2010.00550.x
Bogle, John, C., 1998, The implications of Style Analysis for mutual fund performance evaluation, The Journal of Portfolio Management 24, 34-42.
Bollen, Nicolas, P.B. & Jeffrey A. Busse, 2005, Short-term persistence in mutual fund performance, The Review of Financial Studies 18, 569-597.
Brown, Stephen, J. & William N. Goetzmann, 1995, Performance persistence, The Journal of Finance 2, 679-698.
Carhart, Mark, M., 1997, On persistence in mutual fund performance, Journal of Finance 52, 57-82.
Chen, Joseph, Harrison Hong, Ming Huang & Jeffrey D. Kubik, 2000, Does fund size erode mutual fund performance? The role of liquitdigy and organization, The American Economic Review 94, 1276-1302
Cremers, Martijn, K.J. & Antti Petajisto, 2009, How active is your fund manager? A new measure that predicts performance, Review of Financial Studies 22, 3329-3365.
Cuthbertson, Keith, Dirk Nitzsche & Niall O’Sullivan, 2010, Mutual fund performance: Measurement and evidence, Financial Markets, Institutions & Instruments 19, 95-187.0
