The growth of ETFs and equity mutual fund investor activity

(1)

The growth of ETFs and equity mutual fund investor activity

Pauline Loijson

10587896 January 2017 Bachelor of Science Thesis

Thesis coordinator Dr. P.J.P.M. Versijp

Thesis supervisor

G. Vala Elias Pimentel de Oliveira

University of Amsterdam

Faculty: Economie en Bedrijfskunde Specialization: Financiering en Organisatie

Abstract

This paper researches whether active investors prefer ETFs over traditional equity mutual funds, lowering the turnover of investor for equity mutual funds. To answer this question, a pool of 5 ETFs and a pool of 5 equity mutual funds were researched, matched by 5 indexes. A VAR analysis was conducted to find for mutual

causality and interdependence over time, between the total market cap of the ETFs and the fluctuation rate of the NAV of the mutual funds (representing the turnover of investor). Impulse response functions were used to interpret the results. A multiple regression analysis was conducted to find whether the turnover of investor would indeed be lowered with the growth of ETFs. Evidence suggests that mutual causality is present, and that the growth of the total market cap of ETFs lagged by one week, indeed lowers the fluctuation rate of the NAV of the mutual funds, thus the turnover of investor.

(2)

2

This document is written by, Pauline Loijson, who declares to take full responsibility for the contents of this document. I declare that the text and the work presented in this document is original and that no sources other than those mentioned in the text and its references have been used in creating it. The Faculty of Economics and Business of the University of Amsterdam is responsible solely for the supervision of completion of the work, not for the contents.

(4)

4

1. Introduction

Exchange traded funds (ETFs) are a financial innovation that has been growing rapidly over the past years. ETFs use the same underlying assets as traditional open end equity mutual funds do, but rather than outperforming the benchmark ETFs are created to track the benchmark. ETFs are structured in a different way than mutual funds, and seem to satisfy a different investor type (Agapova, 2011). Literature suggests that ETFs are preferred by active investors. This paper examines whether active investors indeed prefer ETFs over mutual funds. Therefore the research question is:

“Does the turnover of investor for equity mutual funds lower with the growth of equity exchange traded funds (ETFs)?”

The present document becomes relevant by providing an empirical investigation on the impact of ETFs’ growth on the investor’s behavior. The findings suggest that the turnover of investor for mutual funds, indeed lowers by the growth of ETFs.

Even though mutual funds and ETFs seem similar, Agapova (2011) mentions that the investor’s choice between these two, largely depends on the circumstances of the investor. Despite the differences between mutual funds and ETFs, Agapova’s (2011) findings indicate that, even though the two investment vehicles are no perfect substitute, they are substitutes in attracting investor flows. It is stated that sorted by index, an inflow of one dollar into the ETF is expected to lower the open end index mutual fund inflow by 22 cents (Agapova, 2011).

This research paper examines whether ETFs indeed attract a different type of investor than traditional equity mutual funds do. Moreover, Poterba and Shoven (2002) write in their article that: “ETFs may be part of an emerging trend toward segmentation of the mutual fund marketplace, with investors who wish to trade frequently segregated into different products than low-turnover investors” (pp. 12). ETFs might attract investors that are active traders and that supposedly that would reduce the turnover rate for investors that continue to trade in the traditional equity mutual funds (Poterba & Shoven, 2002). This paper investigates if ETFs indeed attract active investors.

A way of measuring the choice of the investors is by measuring the turnover of investor. The turnover of investor determines the activity of investors between portfolios. High turnover of investor in traditional equity mutual funds means that capital gains and losses are realized constantly, due to the frequent trading (Bodie, Kane & Marcus, 2014). This means that active investors are high turnover investors.

The research is based on five mutual funds, five ETFs and five indexes which are each tracked by one ETF and one mutual fund. The ETFs and mutual funds are all in the equity

(5)

5

class. The research question is answered by use of two sub hypotheses: 1) there is mutual causality between the fluctuation rate of the NAV of the mutual funds and the growth of the market capitalization (market cap) of the ETFs, and 2) there is a negative effect of the growth of the total market cap of the ETFs on the fluctuation rate of the NAV of the mutual funds.

Two different analyses are used to test the hypotheses. The first one being a vector autoregression analysis (VAR) where the causality and interdependence of the two main variables over time is tested. In the VAR analysis, two regressions are conducted

simultaneously. Hereby, the main two variables will both be tested as dependent and

independent variables. The second analysis is a linear multiple regression to find the effect of the value and growth of the ETFs on the turnover of investor for the mutual funds. The dependent variable here is the fluctuation of the NAV of the mutual funds, and the main explanatory variable will be the total market cap of the ETFs. For both analyses the main two variables are transformed into a logarithmic scale, reflecting the percentage changes of the variables.

The results show that there is significant mutual causality between the ETFs’ market cap and the turnover rate for the mutual funds. The impulse response functions imply that a shock in the market cap results in a negative response of the fluctuation rate. More

specifically, the multiple regression points out that a 1% increase in the market cap of ETFs lagged by one week, results in a 9.81% decrease in fluctuation rate of the mutual funds. Thus, a drop in the turnover rate of investor. This indicates that ETFs’ growth indeed lowers the turnover rate of mutual funds, meaning that the investor type leaving the mutual funds are active investors. These results are in line with Poterba and Shoven’s (2002) formerly

mentioned expectations. It also shows that the reasons for active investors to prefer ETFs are strong enough for them to start investing in ETFs.

The limitations of this research are a lack of significant control variables. Moreover, the results only account for the funds used in the research, and cannot be generalized.

The rest of the paper includes a literature review and the stated hypotheses in section 2, data is presented in section 3, the empirical analyses and its results are presented in section 4 and 5, respectively. Section 6 concludes the paper.

2. Literature Review

In this section the characteristics of ETFs and mutual funds are described, as well as the advantages and disadvantages of ETFs relative to mutual funds. Finally, the hypotheses are stated.

(6)

6

Mutual funds have been in existence for more than seven decades (Agapova, 2011). A mutual fund is a portfolio of underlying securities. Barclay, Pearson and Weisbach (1998) explain that mutual funds are managed by an investment advisor (sponsor) or a management company. The sponsor purchases the shares that later become the underlying stock, when shares in the fund are sold in public at their net asset value (NAV). The sponsor makes a profit by charging a fee for managing the portfolio. Active equity fund managers for mutual funds aim to outperform the fund’s benchmark (Cremers & Petajisto, 2009), although some studies suggest that in general fund managers appear to be unable to actually outperform the fund’s benchmark (Malkiel, 1995). An equity index fund, first offered in 1972 (Agapova, 2011), is a kind of mutual fund that tracks a benchmark, rather than outperforming it.

Just like equity index funds, ETFs are designed to track a benchmark (Bernstein 2002), but they are a much newer innovation. In 1993, the first ETF in the United States was launched, called the Standard & Poor’s Depository Receipt (SPDR, ‘spider’), which was matching the S&P 500 index (Gastineau, 2010) and traded on the American Stock Exchange. Gastineau (2001) explains that when investors invest in an ETF, they are able to follow index returns without having to purchase all the shares, bonds or other investment types in that index. When depositing a ‘basket’ of securities with an ETF, an institutional investor will receive a fixed number of ETF shares in return. The institutional investor is then able to sell some or all of these ETF shares on the stock exchange. Individual investors can purchase these shares through a broker or an online trade account, and they can only buy or sell these when they are listed on an exchange (Bernstein, 2002). Institutional investors can redeem the number of creation ETF shares, received from the ETF, in return for the basket of underlying stock that was initially deposited (Bernstein, 2002). Broker-dealers can purchase ETF shares from individual investors, so that they are able to form a creation block. The broker-dealers can trade this block with the sponsor for a basket of underlying stock (Bernstein, 2002).

ETFs have grown a lot since they were first introduced. Poterba and Shoven (2002) mention that in 1993, virtually no assets were held by ETFs, but by the end of 2001, ETFs held $79 billion in assets. This means that the assets held by ETFs rose by a factor of 171. The two largest funds, the SPDR trust and the NASDAQ 100 trust accounted for $51 billion of these assets. During the same time frame the assets held by equity mutual funds roughly increased five-fold. At the end of December 1993, the assets held were 741 billion and by the end of November 2001, this number was 3,348 billion (Poterba & Shoven, 2002). Bernstein (2002) predicted that, due to the growth of ETFs, the amount of assets held by ETFs would rival the equity index fund’s amount of assets in the future. Agapova (2011) also mentions

(7)

7

that the loss of market share for traditional index mutual funds seems to primarily be due to the growth of the fund industry, also including the growth in ETFs. Data from the most recent years show that in the beginning of 2011, the total market cap of the five studied ETFs is $149 million and at the end of 2015 it is $269 million. This indicates a growth factor of 1.8 in five years.

An extensive pool of literature suggests that the key differences between ETFs and mutual funds are within costs, trading, arbitrage, tracking errors, liquidity, flexibility, and the difference mentioned most: tax events (Haslem, 2003). These differences are discussed in the following paragraphs, and are said to be important for the decision making process of

investors (Kostovetsky, 2003).

Most of the time ETFs are said to be cheaper than mutual funds. For example: an ETF does not need to market itself to smaller investors reducing management fees (Bodie et al., 2014), individual shareholder book-keeping is eliminated reducing operating costs (Agapova, 2011), and Kostovetsky (2003) suggests that transaction costs are lower for ETFs. The low costs of investing in ETFs is an advantage over mutual funds. However, ETFs are traded through brokers, resulting in broker fees (Poterba & Shoven, 2002). Furthermore, Haslem (2003) finds that trading costs of ETFs may actually exceed those of mutual funds over time, but that these costs can be reduced by usage of an online broker. Dellva (2001), points out that small investors are more attracted to mutual index funds than ETFs due to these broker fees. Finally, Haslem (2003) advises ETF investors to implement a buy-and-hold strategy. Here, Dellva (2001) and Haslem (2003) suggest that high turnover investors actually prefer mutual funds over ETFs, contradicting the previously mentioned studies.

Poterba and Shoven (2002) mention that ETF shares can be traded throughout the day, whereas mutual fund shares can only be traded at the end of the day. Unlike mutual funds, ETF shares can be bought on margin and sold short. These distinctions suggest that ETFs and equity mutual funds attract different types of investors. ETFs would be more preferable for investors that buy in large amounts and that demand short term liquidity, and equity mutual funds would be more preferable for investors that buy and sell smaller amounts and

administer less value to liquidity (Poterba & Shoven, 2002). Another difference is that, in contrary of equity mutual funds, ETF shares cannot be bought or redeemed directly from the fund by individual investors (Bernstein, 2002).

As mentioned previously, mutual funds trade at their closing NAV. ETFs on the other hand, come with the risk of selling at discounts, but the existence of arbitrage opportunities normally keeps these differences small. Arbitrage opportunities appear when there is a

(8)

8

substantial difference between the NAV and the price of an ETF. Arbitrageurs will then create or redeem shares, therefore bringing back the market to an equilibrium (Haslem, 2003). Wiandt and McClutsky (2002) conclude that ETFs provide portfolio transparency to enable arbitrageurs to find arbitrage opportunities. This transparency is not provided by mutual funds (Haslem, 2003).

Another distinction between ETFs and mutual funds are tracking errors. ETFs cannot immediately reinvest dividends. Until the dividends are distributed to the shareholders, they must be held in accounts that do not bear interest. This could lead to tracking errors (Haslem, 2003). Kostovetsky (2003) explains that for ETFs, a cash drag is present. The cash-balancing amount can also lead to tracking errors. Tracking errors make investing in ETFs less desirable than investing in mutual funds.

Guedj and Huang (2008) investigate whether an ETF is more efficient than an open ended mutual fund. They find that investors with high liquidity needs would rather invest in mutual funds, because mutual funds have a liquidity insurance benefit. However, the

insurance does not come without any cost, because moral hazard problems can occur,

reducing the fund performance (Guedj & Huang, 2008). The liquidity insurance benefit might not be clear to investors, because returns of ETFs and mutual funds are accounted for in a different way, leading to a possible lower reported return of the mutual funds as opposed to the ETFs (Guedj & Huang, 2008).

According to Haslem (2003), ETFs are just as flexible and liquid as common stock. The ETFs can be used to engage in similar strategies as common stock. They can also be used to quickly invest or receive cash. This suggests that in this aspect, ETFs are more beneficial than mutual funds to active investors.

One of the most important differences discussed in literature involve taxable accounts. ETFs are said to be more tax efficient than traditional equity mutual funds. When individual investors trade ETF shares, nothing in the underlying portfolio changes, and there is no tax event. Agapova (2011), researched whether ETFs and conventional funds are perfect substitutes for each other. This research showed they are not and that ETFs are preferred by tax sensitive investors with higher liquidity and trading needs (Agapova, 2011). Bodie et al. (2014) find that ahigh turnover of investor in traditional mutual funds means that capital gains and losses are realized constantly, due to the frequent trading . Therefore, investors are unable to time these realizations to manage their overall tax obligation. Bodie et al. (2014) and Agapova’s (2011) findings suggest that high turnover investors are tax sensitive and would prefer ETFs over traditional equity mutual funds. Another tax related difference is that

(9)

9

exchanges of redemption units for underlying stock are tax-free (Haslem, 2003), making this an attractive advantage for active institutional investors.

The differences in costs, trading, arbitrage, tracking errors, liquidity, flexibility, and tax events, give reasons for investor types to prefer ETFs over mutual funds, or vice versa. In this paper it will be researched if the active investor type will prefer ETFs over mutual funds based on the afore mentioned literature.

2.1. Hypotheses

According to the literature, a relationship is present between the turnover of investor for mutual funds and the growth of ETFs. The relationship depends on investor preferences and circumstances. Based on the literature, it is expected that active investors prefer ETFs over traditional equity mutual funds. This is mainly due to the large taxation advantages. This paper focuses on the interdependence between the growth of ETFs and the investor activity of equity mutual funds. Furthermore, it investigates whether the growth of ETFs lowers the investor activity for equity mutual funds. The theory leads to the following research question:“Does the turnover of investor for equity mutual funds lower with the growth of

equity exchange traded funds (ETFs)?”

The analysis of the interdependence between the growth of the ETFs and the investor activity of equity mutual funds leads to the statistical hypothesis (1). The analysis that tests whether the effect of the growth of ETFs on the turnover of investor for the mutual funds is negative, leads to the statistical hypothesis (2). These statistical hypotheses are formulated below. The models that test these hypotheses are explained in section 4.

1)

where p indicates the number of lags

2)

(10)

10

3. Data

In this section, the data description, sources and descriptive statistics are explained as well as the reason why they are incorporated. Firstly, the dependent and explanatory variables will be defined, and secondly, the control variables will be explained.

In this research five equity mutual funds, five exchange traded funds and five indexes are closely examined. The five mutual funds are on the list of ETFs that held over $1.5 billion in assets in 2001, provided by Poterba and Shoven (2002). These funds are currently in the top 100 list of the largest ETFs, compared by market cap, provided by ETF Database (2017). Table 1 presents the funds and indexes used. Each index is assigned to one mutual fund and one ETF as a benchmark. All data is on a weekly basis.

Table 1 Funds and indexes

ETF Ticker Benchmark Equity Mutual fund Ticker

SPDR S&P 500 ETF SPY S&P 500 index Vanguard 500 Index Fund Investor Class VFINX SPDR S&P MidCap 400 ETF MDY S&P Midcap 400 index Fidelity Mid-Cap Stock Fund FMCSX

NASDAQ-100 Trust QQQ NASDAQ-100 Ultra OTC ProFund UOPIX

IShares Russell 2000 ETF IWM Russell 2000 index Schwab Small-Cap Index Fund SWSSX IShares Russell 3000 ETF IWV Russell 3000 index Schwab Total Stock Market Index Fund SWTSX

3.1. The dependent and explanatory variables

The dependent variable in this research is the average fluctuation rate of the NAV per share of the mutual funds. The NAV of a (mutual) fund is the value of the firm’s assets minus their liabilities. When there is a flow into the fund because investors invest, the fund is able to purchase more underlying securities and the value of the underlying portfolio, and therefore NAV, will then increase. When investors redeem their shares, the equity mutual fund is forced to sell underlying securities and the fund’s NAV lowers (Bodie et al., 2014). According to Agapova (2011), flows from and to funds indicate the investor preference, meaning the fluctuation of a fund’s NAV is important for psychological explanations of investors’ choices between mutual funds and ETFs (Ivković and Weisbenner, 2009). Therefore, in this research the turnover of investor is represented by the fluctuation rate of the NAV of the mutual funds indicating the fund’s flows.

(11)

11

For all five funds the daily NAV and their fluctuation were derived from Xignite Database (Xignite, 2016). The daily NAVs have been adjusted into weekly NAVs by using the weekly closing NAV. The net fluctuation was calculated by the absolute value of

and therefore the fluctuation rate is

To reflect the pool of funds, the data set contains the average fluctuation rate of the five equity mutual funds. Within this data set, the first data is as of January 10th 2011, and the last data is of December 28th 2015. The mean of the average fluctuation rate is 0.0227

meaning that on average the mutual funds had a weekly absolute flow of 2.27% of the previous week’s NAV with a standard deviation of 0.0233 (table 2). The standard deviation indicates how concentrated the data points are around the mean. Here, the standard deviation is larger than the mean. This shows that the data points are very wide spread from the mean. The main explanatory variable is the total market cap of the ETFs, reflecting the total market cap of the five ETFs in the pool. The market cap is the market value of the shares outstanding at a point in time. It is an indication of the value of ETFs, and is therefore used to illustrate the growth of ETFs. The presence of the growth of ETFs is shown in graph 1.

Graph 1

Total market capitalization of the 5 ETFs

0 50000000 100000000 150000000 200000000 250000000 300000000 350000000

Growth of the ETF market cap

(12)

12

The daily price and the shares outstanding were retrieved from Wharton Research Data Services (Wharton Research Data Services, 2016), and the market cap is a product of these two (Bodie et al., 2014). The data is converted into weekly data by using the weekly closing bid/ask average price and shares outstanding. This dataset starts at January 10th 2011, and ends at December 28th 2015. The mean of the total market cap is 212 million with a standard deviation of 53 million (table 2). For this variable, the standard deviation is not very concentrated, which could be caused by the formerly mentioned growth of the market cap. The descriptive statistics are a prerequisite for a deeper understanding of further statistical analyses.

Table 2 Descriptive statistics

Variable #Observations Mean

Standard

Deviation Min Max Average fluctuation rate of the NAV of the mutual funds 260 0.0227114 0.0233324 0.0025224 0.2253702 Total market cap of the ETFs 260 2.12e+08 5.32e+07 1.22e+08 3.06e+08

The main two variables are transformed into a logarithmic scale and are now percentage changes of the market cap and fluctuation rate and can be treated as elasticities.

3.2. The control variables

Ivković and Weisbenner (2009) point out that returns of the fund have an influence on the investor flow. Furthermore, Agapova (2011) states that the index returns and the returns of the mutual funds have an effect on the flow of funds towards conventional mutual funds. Hereby, it can be concluded that they should also have a significant effect on the fluctuation of the NAV of the funds, since this is the net value of flows in and to mutual funds. Therefore, the control variables used in this research are the average returns from the equity mutual funds to represent the pool of funds, and the average returns of the indexes to represent the pool of indexes. The stock prices are weekly and were derived on a weekly base from Yahoo Finance (Yahoo Finance, 2016). The adjusted closing prices are used, because they compensate for front-end and back-end load expenses. The returns were calculated by the following formula (Bodie et al., 2014):

(13)

13

The mean of the average index returns is 0.0021, meaning the weekly return of the index pool is on average 0.21% with a standard deviation of 0.0217 (table 3). The mean of the average equity mutual fund returns is 0.0029, meaning the weekly return of the equity mutual fund pool is on average 0.29% with a standard deviation of 0.0257 (table 3). These standard deviations are small meaning the data points are not very wide spread.

Table 3 Descriptive statistics

Variable #Observations Mean Standard Deviation Min Max

Average index return 260 0.002128 0.021657 -0.0859685 0.0817453

Average mutual fund return 260 0.0029054 0.0256501 -0.0985832 0.095045

The control variables are not transformed into a logarithmic scale, because the logarithms of negative values simply do not exist. When used in analysis, the logarithms of negative returns will be registered as missing values and will cause a biased result.

4. Methodology

In this section the method used to conduct the research is defined, as well as the reason why this particular method is used. The regression analysis and its corresponding variables will be explained.

To be able to answer the research question: “Does the turnover of investor for equity

mutual funds lower with the growth of equity exchange traded funds (ETFs)?” a regression

analysis is used. The first regression analysis that will be conducted, is a vector autoregression (VAR) to inspect the interdependence of the two main variables over time. The main tool to interpret the VAR is by use of impulse response functions. Hereafter, a Granger Causality test will be conducted to find whether the past values of the independent variable can be used to forecast the future values of the dependent value. The second regression analysis is a linear multiple regression. This method is used to find the effect of the value and growth of the ETFs on the turnover of investor for the mutual funds.

4.1. VAR analysis

The two main variables are the market cap of the ETFs and the fluctuation rate of the NAVs of the mutual funds. In order to determine whether causal relationships between the two main variables are present, a VAR analysis with two time series variables is be executed. A VAR is a method of forecasting multiple variables with only one model (Stock & Watson, 2015).

(14)

14

The regression consists of two equations where both variables are independent variables. If only one equation is regressed in case of simultaneous causality, the main variable and the error term could correlate, leading to biased and inconsistent results (Stock & Watson, 2015). Therefore, the equations are regressed simultaneously. The explanatory variables used in both regressions are lagged values of the variables. The coefficients of the VAR analysis are estimated by ordinary least squares (OLS) (Stock & Watson, 2015). Including unrelated variables leads to estimation error without adding useful content (Stock & Watson, 2015). Thereby, adding unrelated variables will reduce forecast accuracy. The control variables appeared to be insignificant and are therefore removed from the VAR.

Before the VAR can be estimated, several conditions need to be met to capture all the dynamics of reality. Firstly, autocorrelation between residuals needs to be ruled out in order to determine the correct lag order. Secondly, the stability condition should be met to confirm stationarity. When these conditions are met the VAR can be estimated, and the coefficients can be interpreted as the responses that appear from impulses hitting the system (Lütkepohl & Krätzig, 2004).

In order to determine what the lag length of the VAR should be, autocorrelation between the residuals needs to be ruled out. To test for residual autocorrelation, VAR

Residual Serial Correlation LM tests need to be conducted (Lütkepohl & Krätzig, 2004). The null hypothesis of this test is that there is no serial autocorrelation in place. When the p-value of the LM-statistic is larger than the significance level for all lags used, no autocorrelation is in present, and the corresponding lag length can be used in the VAR. When there is still autocorrelation of the residuals the VAR Lag Order Selection Criteria is used to find what lag order to use. The selection criteria presented are: the Akaike information criterion (AIC), the Schwarz information criterion (SC), and the Hannan-Quinn information criterion (HQ).

A unit root test is conducted to find whether a unit root against non stationarity is present (Lütkepohl & Krätzig, 2004). When no AR roots lie outside the unit circle, the stability condition is met (Lütkepohl & Krätzig, 2004), and the VAR model is stationary. After all these conditions are met, the VAR can be estimated and the results can be interpreted.

In table 4, the AIC points out that the eighth lag order should be used. The SC and HQ show that the second and fifth lag orders, respectively, should be used (table 4). The second and fifth lag orders both resulted in autocorrelation between the residuals. Table 5 confirms that with a lag order of eight no autocorrelation between the residuals is in place, confirming the correct use of the eighth lag order. Lastly, the unit root test confirms that with the chosen

(15)

15

lag structure, the stability condition is met, implying the model is stationary (table 6).

Table 4

VAR Lag Order Selection Criteria

Lag AIC SC HQ 0 -3.163351 -3.135260 -3.152046 1 -3.398368 -3.314094 -3.364454 2 -3.489421 -3.348965* -3.432898 3 -3.534278 -3.337639 -3.455146 4 -3.550811 -3.297989 -3.449069 5 -3.603268 -3.294265 -3.478918* 6 -3.618201 -3.253014 -3.471240 7 -3.628839 -3.207470 -3.459270 8 3.669606* -3.192054 -3.477427

*Indicates lag order selected by criterion

Table 5

VAR Serial Correlation LM tests

Lags LM-statistic Prob 1 1.565703 0.8149 2 1.616547 0.8058 3 2.216140 0.6961 4 4.073686 0.3961 5 2.870947 0.5796 6 3.705038 0.4474 7 2.932950 0.5691 8 6.803725 0.1466

(16)

16

Table 6 Unit Root test

Root Modulus 0.642678 - 0.610322i 0.886301 0.642678 + 0.610322i 0.886301 -0.484195 - 0.689673i 0.842671 -0.484195 + 0.689673i 0.842671 0.146419 - 0.791915i 0.805338 0.146419 + 0.791915i 0.805338 -0.054278 - 0.798162i 0.800006 -0.054278 + 0.798162i 0.800006 -0.769894 - 0.210331i 0.798108 -0.769894 + 0.210331i 0.798108 0.607526 - 0.476291i 0.771972 0.607526 + 0.476291i 0.771972 -0.543069 - 0.537143i 0.763837 -0.543069 + 0.537143i 0.763837 -0.600125 0.600125 0.596151 0.596151

No root lies outside the unit circle VAR satisfies the stability condition

All conditions are met indeed, so the VAR can be estimated with a lag order of eight. The equations that meet all conditions, used in the VAR of this research are as follows:

1. 2.

(17)

17

where the are intercepts and the and are error terms. The definition of the variables are given in the table below. The first differences of the variables were calculated by use of the statistical program Eviews. The following formula presents the calculation of the first differences of the variables:

All the first differences are noted with a delta sign (Δ). After the estimation of the model the impulse response functions will be interpreted. The impulse response functions rule out autocorrelation, and therefore contemporaneous shocks can only come through the residuals. The “Residual – one standard deviation” decomposition method is used. This explicitly assumes that no contemporaneous correlation between the residuals is present. The

autocorrelation was ruled out (table 5), and therefore the residual decomposition method may indeed be used.

Lastly, a Granger Causality test will be conducted to find whether the past values of the independent variable can be used to forecast the future values of the dependent value.

List of Variables

The log of the average fluctuation rate of the net asset value of the 5 mutual funds lagged with p weeks

The log of the total market capitalization of the 5 exchange traded funds in lagged with p weeks

4.2. Multiple regression analysis

When a variable is left out of the regression, but is: 1) a determinant of the dependent

variable, and 2) correlated with the regressor, this will lead to omitted variable bias (Stock & Watson, 2015). To lower the chance of omitted variable bias, control variables are used. Therefore, in this analysis, not a singular regression but a multiple regression is used. The regression equation is as follows:

1.

(18)

18

is the constant (intercept) of the model, are the slope coefficients and is the error term. The variables are explained in the following table:

List of Variables

The log of the average fluctuation rate of the net asset value of the 5 mutual funds lagged with p weeks

The log of the total market capitalization of the 5 exchange traded funds in lagged with p weeks

Average returns of the 5 indexes Average returns of the 5 mutual funds

The first differences of all variables are calculated and indicated as in section 4.1. In the multiple regression model, OLS is applied to estimate the intercept and the slope coefficients of the sample observations. The coefficients are estimated by a minimization of the sum of squared prediction mistakes (Stock & Watson, 2015). The null hypothesis in this test is that at least one of the regressors are unequal to zero, thus at least one of the regressors has an effect on the dependent variable.

5. Results

In this section the statistical results of the regression analyses are interpreted by discussing all the outcomes of the conducted tests. Firstly, the coefficients of the VAR analysis appear only significant in certain lags, and most of them indicate a negative effect. Secondly, the impulse response functions show significant effects in some lags. A shock in the first lagged value of the market cap results in a significant negative response of the fluctuation rate. Hereafter, the Granger Causality test shows that mutual causality is in place. Finally, the multiple regression also presents a negative effect of the first lagged value of the market cap, on the fluctuation rate. The control variables are insignificant.

5.1. The VAR analysis coefficients

Table 7 shows the coefficients and their standard errors of the VAR analysis. Firstly, the fluctuation rate equation (1) is considered. The fluctuation rate coefficients are significant in all lags, and the market cap coefficients are only significant in the first and third lag. All significant coefficients are negative, which implies that an increase in their past values, results in a decrease in the fluctuation rate.

(19)

19

Secondly, the market cap equation is considered (2). The fluctuation rate coefficients are significant in the third lag and the market cap coefficients are all insignificant. The significant coefficient is positive, meaning that an increase in its past value, results in an increase in the market cap.

Finally, the fit of the model is considered. With the addition of a new variable to a model, its unadjusted R² increases. However, this does not mean that the addition of another variable explicitly improves the fit of the model (Stock & Watson, 2015). Therefore, the adjusted R² is used in explaining the fit of the model, rather than the R². The adjusted R² of the first (1) model is 0.4613, meaning that the model accounts for 41.6% of the variation in the dependent variable. The adjusted R² of the second equation (2) is 0.0667 meaning that the model only explains 6.67% of the variation in the dependent variable, which is very low.

Table 7 Vector autoregression (1) (2) -0.823508*** (0.06399) 0.000745 (0.00115) -0.728398*** (0.08069) 0.000911 (0.00145) -0.633216*** (0.08959) 0.003767* (0.00161) -0.573397*** (0.09151) 0.000477 (0.00164) -0.549657*** (0.08989) -0.001006 (0.00161) -0.394772*** (0.08772) 0.000403 (0.00157) -0.327981*** (0.07984) 0.000143 (0.00143) -0.196578** (0.06384) 0.000782 (0.00114) -10.19256** (3.62711) -0.090094 (0.06504)

(20)

20 -4.833951 (3.67611) 0.007289 (0.06592) -6.794970* (3.67532) 0.039906 (0.06590) 3.506819 (3.66654) -0.106999 (0.06575) 2.832696 (3.67315) -0.061707 (0.06586) 3.855288 (3.65163) -0.104709 (0.06548) -4.843121 (3.61605) -0.053509 (0.06484) 4.730214 (3.59180) 0.112135 (0.06441) Intercept 0.016969 (0.04595) 0.001310 (0.00082) Number of observations 251 251 R-squared 0.461260 0.126424 Adjusted R-squared 0.424423 0.066692

Note: *p<0.05, **p<0.01, ***p<0.001, standard errors between parentheses

5.2. Impulse response functions

The interpretation of the estimates previously mentioned, does not provide much insight to the relations between the variables and how they interact. Many of the coefficients are

insignificant. The main tool of interpreting VAR results is by use of impulse response

functions. Impulse response functions give more insight on how the variables react in a shock environment.

The impulse response functions of the VAR analysis are shown in graph 2. Graphs 2a and 2c start at zero. As mentioned in section 4, this has been imposed. The VAR only starts one period after. In the graphs, lag 0 is labeled as 1, lag 1 is labeled as 2, and so forth. The responses in the graphs are responses to a one standard deviation positive shock.

Graph 2a shows the response to an impulse of the fluctuation rate on itself. It predicts only the direct effect (lag 0) and the first lag to be significant. The direct effect is a positive response to a shock in the fluctuation rate. This is not surprising, because a positive shock at time zero will logically result in a positive effect at time zero. The first lag predicts a negative

(21)

21

response of the fluctuation rate to an impulse on itself. This means that a shock will result in a negative response, one week from the shock.

Graph 2b shows the results of an impulse of the market cap on the fluctuation rate. Only a shock in the first lag appears to be significant. The first lag predicts a negative

response of the fluctuation rate to a shock in the market cap. This indicates that a shock in the market cap, will result in a negative response of the fluctuation rate, one week from the shock.

Graph 2c shows the results of an impulse of the market cap to a shock in the fluctuation rate. Shocks in the third and fourth lags are significant. The third lag predicts a positive response of the market cap to a shock in the fluctuation rate. The fourth lag predicts a negative effect of a shock in the fluctuation rate. Following the same reasoning as the

formerly explained graphs, the shock results in a positive response, three weeks from the shock and a negative response four weeks from the shock.

Graph 2d shows the response of an impulse of the market cap to itself. The direct effect, sixth and tenth lags are significant. The direct effect and the tenth lags show a positive effect of a shock in the market cap on itself. The sixth lag predicts a negative effect of an impulse in the market cap on itself. Again, following the same reasoning: a shock in the market cap results in a direct positive response, and another positive effect after ten weeks from the shock. Six weeks after the shock a negative effect is predicted.

Graph 2

Response to Nonfactorized one S.D. Innovations ± 2 S.E.

(22)

22

c. Response of ΔlMC_ETFs to ΔlavfrNAV_MF d. Response of ΔlMC_ETFs to ΔlMC_ETFs

5.3. Granger Causality test results

Finally, the results from the Granger Causality test are presented in the tables below. Table 8 shows the Granger Causality test where the fluctuation rate is the dependent variable. The null hypothesis that the average fluctuation rate of the NAVs of the mutual funds was not Granger caused by the total market cap of the ETFs, is rejected at a significance level of 1%.

Following the same reasoning, table 9 shows that at a 5% significance level, the market cap was Granger Caused by the fluctuation rate. There is mutual causality in place, meaning it can be concluded that past values of the market cap indeed can be used to forecast future values of the fluctuation rate and vice versa, at the given significance levels.

Table 8

Granger Causality test 1

Dependent variable: lavfrNAV_MF

Excluded Chi-sq df p-value

lMC_ETFs 21.43620 8 0.0061

All 21.43620 8 0.0061

Table 9

Granger Causality test 2

Dependent variable: lMC_ETFs

Excluded Chi-sq df p-value

lavfrNAV_MF 18.25326 8 0.0194

(23)

23

5.4. The multiple regression results

The results of the regression, where the fluctuation rate is the dependent variable, are presented in table 10. The adjusted R² in this model is 0.4636, meaning that the model accounts for 46.36% of the variation in the dependent variable. The outcome of the F-test is 11.14 and its p-value is smaller than 0.001. This indicates that with a significance level of 0.1%, the null hypothesis can be rejected, and at least one of the regressors has a significant effect on the dependent variable.

The total market cap of the ETFs lagged by one week has a coefficient of -9.813, meaning that a 1% increase of the total market cap of the ETFs leads to a 9.81% decrease of the average fluctuation rate of the NAV of the mutual funds. The p-value of this coefficient is smaller than 0.01, meaning that with a 1% significance level the effect of this variable on the dependent variable is significant.

The variables of the fluctuation rate lagged by 1-7 weeks have a significant effect on the change of the fluctuation rate at a significance level of 0.01%, and the fluctuation rate lagged by 8 weeks has a significant effect on the change of the fluctuation rate at a 1% significance level. All these coefficients are negative, meaning they have a negative effect on the current change of the fluctuation rate.

The control variables used, appear to be insignificant. This goes against the

expectations that were formed, based on the literature. Due to the lack of significant control variables there could still be omitted variable bias in place. Adding more significant control variables to the model, could improve the fit of the model.

Table 10

Multiple Regression Analysis

(1) -0.821830*** (0.064236) -0.734209*** (0.081070) -0.629340*** (0.089921) -0.573343*** (0.091978) -0.550651*** (0.090095) -0.393700***

(24)

24 (0.087909) -0.330594*** (0.080109) -0.191790** (0.064159) -9.813056** (3.655509) -5.011490 (3.688297) -6.845193 (3.699060) 3.825293 (3.709124) 2.518847 (3.706898) 4.105910 (3.681416) -5.109830 (3.637424) 4.997326 (3.609331) -11.28197 (14.54508) 10.22395 (12.25682) Intercept 0.016506 (0.046046) Number of observations 251 R-squared 0.463605 Adjusted R-squared 0.421988 F-statistic 11.13984*** Note: *p<0.05, **p<0.01, ***p<0.001

6. Conclusion

In this paper the causality between the growth of ETFs and the investor activity in equity mutual funds, and the potential effect of the growth of ETFs on the investor activity have been researched. The research was based on weekly data of five years, starting January 10th 2011, and ending December 28th 2015. The effects have been researched for a pool of five mutual funds and five ETFs, matched to 5 different indexes. The main research question was: “Does

the turnover of investor for equity mutual funds lower with the growth of equity exchange traded funds (ETFs)?”.

To examine this question, a literature review was conducted. The literature review resulted in the following hypotheses: 1) there is mutual causality between the fluctuation rate

(25)

25

of the NAV of the mutual funds and the growth of the market cap of the ETFs, and 2) There is a negative effect of the growth of the total market cap of the ETFs on the fluctuation rate of the NAV of the mutual funds. Hereafter, the statistical analyses were conducted.

Firstly, the a VAR analysis was conducted. The coefficients do not give enough insight, and therefore the main tool used to interpret the results were impulse response functions. The impulse response functions predict that an impulse in the first lagged value of the market cap of the ETFs, has a negative effect on the response of the fluctuation rate of the NAVs of the mutual funds. This indicates that a sudden increase in the market cap, results in a decrease of the fluctuation rate, one week after the shock. This result follows the expectations that were formed, based on the literature. The Granger Causality test showed that mutual causality between the fluctuation rate and the market cap is in place.

The results of the multiple regression analysis indicate that at a 0.1% significance level a 1% increase in the total market cap of the ETFs, reduces the fluctuation rate of the NAV of the mutual funds one week after by 9.81%, meaning that the alternative hypothesis was met. The results of the research indicate that the growth of the total market cap of the ETFs lagged by one week seems to indeed lower the fluctuation rate of the NAV of the mutual funds, thus the turnover of investor. This answers the research question of this study. The control

variables appeared to have no significant effect on the fluctuation rate of the NAV.

This research has proven that active investors prefer ETFs over mutual funds. Further research could build upon the results of this study, by distinguishing which of the advantages of ETFs over mutual funds have the most effect on the investor’s choice to switch.

(26)

26

7. Reference list

Agapova, A. (2011). Conventional mutual index funds versus exchange-traded funds. Journal

of Financial Markets, 14, 323-343.

Barclay, J., D. Pearson, N., & S. Weisbach, M. (1998). Open-end mutual funds and capital-gains taxes. Journal of Financial Economics, 49(1), 3-43.

Bernstein, P. J. (2002, January). A primer on exchange-traded funds: CPAs should know the difference between ETFs and mutual funds. Journal of Accountancy, 193(1), 38-41. Bodie, Z., Kane, A., & Marcus, A., J. (2014). Investments (10th ed.). Maidenhead, United

Kingdom: McGraw-Hill Education.

Cremers, K. J. M., & Petajisto, A. (2009). How Active Is Your Fund Manager? A New Measure That Predicts Performance. Review of Financial Studies, 22(9), 3329-3365. Dellva, W. (2001). Exchange-Traded Funds Not For Everyone. Journal of Financial

Planning, 110-124.

ETF Data Base (2017). Largest ETFs: Top 100 ETFs by Assets. Retrieved January 25, 2017, from http://etfdb.com/compare/market-cap/

Gastineau, G. L. (2001). Exchange Traded Funds. Handbook of finance.

Gastineau, G. L. (2010). The exchange-traded funds manual (Vol. 186). John Wiley & Sons. Guedj, I., & Huang, J. (2008). Are ETFs replacing index mutual funds. Working paper.

University of Texas at Austin.

Haslem, J. A. (2003). Exchange-Traded Funds: Nature, Developments, and Implications. ETFS and Indexing, 2003(1), 116-126.

Ivković, Z., & Weisbenner, S. (2009). Individual investor mutual fund flows. Journal of

Financial Economics, 92(2), 223-237.

Kostovetsky, L., (2003). Index mutual funds and exchange-traded funds. Journal of Portfolio

Management 29 (4), 80–93.

Lütkepohl, H., & Krätzig, M. (2004). Applied Time Series Econometrics. New York, United States of America: Camebridge University Press.

McClatchy, W., & Wiandt, J. (2002). Exchange Traded Funds. John Wiley & Sons.

Poterba, J., & Shoven, J. (2002, May). Exchange Traded Funds: a New Investment Option for Taxable Investors. American Economic Review, 92(2), 422-427.

Stock, J.H., & Watson, M.W. (2015). Introduction to Econometrics (10th). Harlow, United Kingdom: Pearson Education.

Wharton Research Data Services (2016). CRSP Daily Stock. Retrieved December 12, 2016, from

(27)

https://wrds-27

web.wharton.upenn.edu/wrds/ds/crsp/stock_a/dsf.cfm?navGroupHeader=Annual%20 Update&navGroup=Stock%20%2F%20Security%20Files

Xignite (2016). Xignite NAVs. Retrieved December 12, 2016, from http://www.xignite.com/product/mutual-fund-NAVs

Yahoo Finance (2016). FMCSX Historical Prices. Retrieved December 11, 2016, from https://finance.yahoo.com/quote/FMCSX/history?period1=1294009200&period2=145 1257200&interval=1wk&filter=history&frequency=1wk

Yahoo Finance (2016) ^GSPC Historical Prices. Retrieved December 11, 2016, from

https://finance.yahoo.com/quote/%5EGSPC/history?period1=1294009200&period2=1 451257200&interval=1wk&filter=history&frequency=1wk

Yahoo Finance (2016) ^MID Historical Prices. Retrieved December 11, 2016, from

https://finance.yahoo.com/quote/%5EMID/history?period1=1294009200&period2=14 51257200&interval=1wk&filter=history&frequency=1wk

Yahoo Finance (2016) ^NDX Historical Prices. Retrieved December 11, 2016, from

https://finance.yahoo.com/quote/%5ENDX/history?period1=1294009200&period2=14 51257200&interval=1wk&filter=history&frequency=1wk

Yahoo Finance (2016) ^RUA Historical Prices. Retrieved December 11, 2016, from

https://finance.yahoo.com/quote/%5ERUA/history?period1=1294009200&period2=14 51257200&interval=1wk&filter=history&frequency=1wk

Yahoo Finance (2016) ^RUT Historical Prices. Retrieved December 11, 2016, from

https://finance.yahoo.com/quote/%5ERUT/history?period1=1294009200&period2=14 51257200&interval=1wk&filter=history&frequency=1wk

Yahoo Finance (2016) SWSSX Historical Prices. Retrieved December 11, 2016, from https://finance.yahoo.com/quote/SWSSX/history?period1=1294009200&period2=145 1516400&interval=1wk&filter=history&frequency=1wk

Yahoo Finance (2016) SWTSX Historical Prices. Retrieved December 11, 2016, from https://finance.yahoo.com/quote/SWTSX/history?period1=1294009200&period2=145 1516400&interval=1wk&filter=history&frequency=1wk

Yahoo Finance (2016) UOPIX Historical Prices. Retrieved December 11, 2016, from

https://finance.yahoo.com/quote/UOPIX/history?period1=1294009200&period2=1451 516400&interval=1wk&filter=history&frequency=1wk

Yahoo Finance (2016). VFINX Historical Prices. Retrieved December 11, 2016, from

https://finance.yahoo.com/quote/VFINX/history?period1=1294009200&period2=1451 257200&interval=1wk&filter=history&frequency=1wk

The growth of ETFs and equity mutual fund investor activity