Master Thesis
A study on the relation between past performance and fund flows for Equity based
Exchange Traded Funds.
Lukas Muijs - 10901817
Master International Finance (Part-Time, Class 2015 - 2017)
Supervisor: Florian Peters
1 Table of Contents
ABSTRACT 2
1 INTRODUCTION 3
2 HYPOTHESIS & RESEARCH QUESTIONS 6
3 EXCHANGE TRADED FUNDS EXPLAINED 7
3.1.1 ETF FUNDAMENTALS 7
3.1.2 ETFHISTORY:FROM CONCEPT TO PRODUCT 9
3.1.3 ETFGLOBAL MARKET OVERVIEW 10
3.1.4 COMPETITION BETWEEN INDEX FUNDS AND EXCHANGE TRADED FUNDS 12
4 LITERATURE REVIEW 14
4.1 RELATED RESEARCH TOPICS 14
4.1.1 CASH FLOW –PAST PERFORMANCE RELATIONSHIP 14
4.2 ETFFUND FLOWS &PERFORMANCE MEASUREMENTS 16
4.2.1 FUND FLOWS 17
4.2.2 TRACKING ERROR MEASUREMENTS AND THEIR DETERMINANTS 18
4.2.3 BID-ASK SPREAD &LIQUIDITY 21
4.2.4 INFORMATION RATIO 21
5 SAMPLE & VARIABLE DESCRIPTION 23
5.1 DATA COLLECTION &VARIABLE DEFINITION 23
5.1.1 SAMPLE CONSTRUCTION 23
5.1.2 VARIABLE COMPUTATION. 24
5.2 DESCRIPTIVE STATISTICS AND PRELIMINARY ANALYSES 27
5.2.1 PRESENTATION OF THE DATA SAMPLE 27
5.2.2 DESCRIPTIVE STATISTICS 29
6 EMPIRICAL RESULTS ETF PERFORMANCE AND FUND FLOWS 33
6.1 EMPIRICAL RESEARCH MODEL 33
6.2 BASELINE RESULTS 35
6.3 DIFFERENCES ETFCATEGORIES 37
6.3.1 SYNTHETIC VERSUS PHYSICAL ETFS 37
6.3.2 DIFFERENCES BY INDEX TYPE 38
6.4 DISCUSSION 41
6.4.1 FUND FLOWS VS.PASSIVE RETURNS 41
6.4.2 FUND AGE VS.FUND FLOWS. 42
6.4.3 ETFFUND CHARACTERISTICS AND FUND FLOWS 42
6.4.4 BEST PERFORMING FUNDS ACCORDING TO THE REGRESSION OUTCOMES 44 6.5 ROBUSTNESS DISCUSSION &RECOMMENDATIONS FOR FURTHER RESEARCH 44
7 CONCLUSIONS 46
8 REFERENCES 48
Abstract
In this paper I investigate how ETF fund flows are influenced by passive returns and other typical ETF performance measurements such as tracking error and liquidity. These relationships are tested on a wide sample of physical and synthetic equity based ETFs and compared with research on actively managed mutual funds. The main finding with regard to past returns is that investors in ETFs cannot be seen as a homogenous group because they seem to value different ETF characteristics however there are some commonalities. High tracking error is overall an undesirable feature, high bid ask spreads seem to attract investors who pursue taking benefits from arbitrage opportunities. For passive returns I identify a positive effect for fund flows to physical ETFs, however synthetic fund flows are negatively correlated with passive returns. These contradicting results confirm that ETFs cannot be seen as simple investment products but are complex securities that remain to have high growth potential. Furthermore I also conclude that the variance of fund flows that is explained by past performance is very small for ETFs relative to the more conventional actively managed mutual funds.
1 Introduction
Within the fund management industry, actively managed mutual funds have been dominating the sector for decades. This landscape is changing quickly however due to the introduction of the passive fund management concept and the growth of its popularity embodied by the dramatic increase of assets that are invested in Exchange Traded Funds (ETFs). Although the mutual fund industry is very established, the ETFs market is still evolving. In the period 2000 to 2016 ETF assets under management (AUM) exploded from $79 billion predominantly in the US to $3.5 trillion globally. The attractiveness of ETFs is often attributed to their unique characteristics that distinguish them from more conventional mutual funds. While ETFs are a relative new type of fund, every investor considering an investment in an ETF will have to go through the same traditional selection process whereby selecting the right funds to construct a suitable investment portfolio is the main concern. Although past performance does not guarantee future returns, consistent with the normal paradigm of financial economics it can be expected that investors would invest mainly in the investment funds with the best performance track record and economic outlook. This should mean that the best performing ETFs must have the highest inflows of new assets under management (AUM). Using passive returns and specific performance measurements for ETFs that are described in the paper of Roncalli and Hassine (2013) this paper will investigate whether and how ETF cash flows respond to past performance.
When looking to the “classical” relationship between funds flows and past performance for actively managed mutual funds Berk & Green (2004) found that cash flows are positively correlated to past performance. This indicates that investors seem to have a preference for funds with the best track record. When testing this relationship for passively managed index mutual funds that are often compared with Exchange Traded Funds (ETFs), Elton et al (2002) found a similar relationship, however they stated that investments also seem to flow to funds that have less positive performance. While the ETF fund popularity will certainly continue to grow in the next years, little is written on how ETF cash flows behave in relation to past performance.
The papers of Berk & Green (2004) and Elton et Al (2002) are focused on relating mutual fund cash flows with the most commonly used performance measurement, “alpha”, which is excess return generated by the fund holdings relatively to their benchmark that represents the market portfolio. But how are cash flows influenced for ETFs whose main strategy is to replicate their benchmark exactly and therefore do not pursue the generation of excess returns? More cumbersome is measuring performance differences for ETFs tracking the same index because, considering the theory
on the law of one price, these funds should generate the same return. Following this principle it would mean that there are no serious grounds to have a preference for one ETF over another when they follow the same index. Considering that ETF markets are not perfect due to market imperfections like transactions cost, liquidity and taxes, intuitively it can be expected that performance differences between funds and sponsors can be identified. Hassine and Roncalli (2013) published in their paper a methodology using liquidity, investment costs and tracking error volatility that provides index-tracking ETFs with an efficiency rating. Based on their efficiency measurements they concluded that some funds and issuers in their sample indeed performed consistently better than others. Using the performance variables described by Roncalli and Hassine as proxies for ETF performance, coherent with the finding of Berk & Green this paper researches the hypothesis that passive returns, low tracking error ratios and high liquidity should have a positive effect on inflow of new assets under management (AUM). Getting insights how ETF funds flows move is valuable information for new issuers as currently the ETF markets currently seems to be consolidated around a limited number of issuers and therefore obtaining new markets share seem difficult. However with so much growth potential there still might be plenty of opportunities for new issuers to enter the ETF market.
The dataset for this thesis covers the period 2007 and early 2017 and includes monthly data on net flows and prices of 2121 synthetic and physical equity funds. Based on the results it can be concluded that the variance of ETF flows that is explained by passive returns and other performance metric wields significantly less explanatory power than the effect of excess returns on mutual fund flows. Furthermore significant differences between the behaviour of fund flows to synthetic and physical funds are identified. These differences indicate that investors in synthetic funds seem to value different ETF characteristics than shareholders in physical funds. The main finding with regard to past returns is that investors in synthetic funds appear not to chase positive past returns and are not affected by liquidity. The results of this research do show some general consistencies across all ETF types. On aggregated level tracking error volatility seem to be regarded by al ETF investors as an undesirable characteristic, and generally speaking ETF trade activity seem to drive fund flows. When considering more basic characteristics, fund age and cost are decreasing factors for ETFs while fund size increases fund flows.
The remainder of the paper is arranged as follows. Essential information on exchange traded funds and related topics will be analysed and summarized first. The literature review will explain the calculations behind the ETF performance measurements such as tracking error and liquidity. Next, a description of the data and an overview of the sample will be provided. The paper will proceed by
applying the methodologies to the sample and offer a thorough analysis of the exchange-traded funds in the sample using multiple statistical methods. We compare the results across funds that fully replicate their benchmark, optimize their holding by selecting stocks with similar correlations or use derivatives. In the next step the results are tested across different ETF types as the data sample includes broad market, small, mid and large cap funds. A part will be presented that focuses on potential limitations of this study and provides recommendations for future research. The paper is finalised by a discussion of its implications and a conclusion summarising the main findings.
2 Hypothesis & research questions
This thesis will try to reject the hypothesis that ETFs AUM cash flows are positively influenced by performance measurements. ETFs however, are offered in different formats, as they are listed on different exchanges, track different indices and have distinct replication structures (physical replication vs. synthetic). Because of these differences it could be possible that the effect of the predictor variables on the dependent variable can vary between different types of trackers. This allows extra opportunity for further exploration as it creates the possibility to identify new predictors variables
1. Is the variance of ETF cash flows that is explained by historical performance significant? Do funds flow to the ETFs with the best past returns.
2. Berk & Green (2004) and Evans (2010) Identified that flows to younger mutual funds respond more dramatically to performance than more mature funds. Is fund age a mitigating factor in ETFs AUM cash flows?
3. Is the variance of cash flows of index tracking ETFs that is explained by the performance measurements (liquidity, tracking error and passive returns) of Roncalli and Hassine (2013) significant?
4. ETFs follow different replication techniques and can either be synthetic, or try to fully reconstruct the index. Rompotis (2008) showed in his paper that the structure of the ETF has an effect on track error volatility, difference etc. Is replication strategy a mitigating factor in ETFs AUM cash flows?
After calculating the betas for each independent variable we will apply the binomial test with a 95% confidence level to assess the appropriateness of each of the models.
3 Exchange Traded Funds Explained
Buying and selling ETFs shares is very easy to understand for the average private investor as they trade like normal stocks. Distinguishing ETFs from other funds types however can be difficult as their differences are sometimes very subtle and therefore easily overlooked. Considering that ETFs are the focus point of this thesis it is essential to provide a clear understanding how to distinguish ETFs from other fund types.
3.1.1 ETF fundamentals
According to the definition used by the US financial supervisor, the Securities and Exchange Commission (SEC), ETFs are open-ended collective investment schemes that are unlike traditional mutual funds listed on an exchange and traded throughout the day like a stock on the secondary market. Most existing ETFs invest primarily in equity however there are funds that offer exposure to other asset classes like fixed income, commodities, currencies and futures but these funds usually have different structures and are therefore subject to different regulations. Similar to index mutual funds, ETFs are regarded as passively managed funds because generally, ETFs track the performance of a target benchmark, often a stock index, by replicating its underlying portfolio of shares.
Primary market transactions that lead to the subscription or redemption of ETF shares are executed by “ Authorized Participants” (AP) such as brokers. Depending on the quantity of sell and purchase orders, the AP creates ETF shares by constructing and exchanging a basket of stocks that matches the benchmark index for an equal amount in value for an X number of stocks in the ETF. The redemption of ETF stocks takes place in reverse manner as the authorized participant in this case exchanges an X quantity of ETFs stocks for a redemption basket of shares that is similar in value. ETFs offer several advantages over more conventional mutual funds. Comparable to listed companies, ETFs are more transparent as they register redemptions and subscriptions of new shares on a continuous basis. ETFs also trade like stocks and can be purchased and sold throughout the day without any direct front or backload fees that are typical for mutual funds. Moreover like normal shares, ETFs can be sold short or purchased on margin and it is also possible for investors to buy put and call options on most ETFs. The Important benefit of these characteristics is that the price of the ETF trades usually for a price that is very close to the NAV (Net Asset Value) of the underlying holdings. While mutual funds are “forward priced” meaning that all buy and sell order placed throughout the day will receive the same NAV the next time it is calculated1. Mutual funds trade usually at a discount relative to the NAV of their underlying holdings.
Engle and Sarkar (2006) studied ETF price deviations and they mention that these market frictions enable investors to benefit from arbitrage opportunities. Liquidity issues that make it difficult to execute the share creation and redemption process properly are mostly the root causes of these deviations, but there are multiple other reasons. It is also common to see that very popular ETFs trade at a premium over their underlying net asset value of their holdings. Other markets distortions that are described are the situations whereby the shares of an ETF can be traded continuously because of listings in multiple locations while the shares of the underlying index which is followed haven’t started its trading day yet. Not surprisingly the actual delivery of underlying shares can be cumbersome especially due to the delays when clearing these international shares.
ETFs also offer a potential tax advantage over mutual funds as holding ETF shares is from a fiscal perspective very different. The investor is not taxed on the underlying shares that are held by the ETF but on the shares he or she owns on the ETF itself. A practical example to explain the difference is the share redemption process. When a mutual fund investor redeems shares, the fund must sell securities to meet the redemption. Depending on the share position, this can trigger capital gains taxes that are to be paid by the remaining shareholders. In contrast when retail investors wish to redeem their ETF shares, no underlying shares usually have to be sold as they simply sell them to other investors in the secondary market. Exchanging ETF shares for the underlying portfolio is also not regarded as a taxable event. Although it is uncommon for the average retail investor to swap ETF shares for the underlying shares, large institutional investors frequently make use of this option. Lastly ETFs are usually cheaper than mutual funds because shares are offered to investors through brokers and not directly from the issuer who saves on marketing (12b-1) expenses resulting in lower management costs. However investors should take the brokerage fees into account.
New ETFs are issued by a financial institution, and is often is referred to as the fund “sponsor” or issuer. The issuer decides what investment strategy the ETF will follow. For index trackers, the issuer selects an index and a tracking method. Early ETFs tracked Broad Market indices (S&P 500, Dow Jones, NASDAQ), and their portfolios holdings were based on market capitalization of the underlying stocks. New index-based ETFs not only follow more “exotic” benchmarks but also use different portfolio construction techniques that are based on sales or book value. While still focused primarily on a single index numerous sponsors nowadays select their portfolio holding depending on a selection of characteristics, such value or growth funds and then weight the selected securities equally or by market capitalization. Other customized ETF approaches include the selection of stocks with a low volatility, maximize diversification, or achieve a high or low degree of correlation with the market. Over the last years ETFs that incorporate the usage of leverage or portfolio replication using
derivatives is also used more often. Using derivatives is especially popular in Europe. Some issuers such as Power Share, Lyxor and Amundi are specialized in leveraged ETFs or use mainly derivatives to replicate their benchmark portfolio.
3.1.2 ETF History: From concept to product
Two studies dated from the sixties that are regularly quoted in literature (Elton et Al 2004, Roncalli, 2013) as fundamental to the idea of passive investment management were Eugene Fama (1963, 1970) his efficient markets theory and the research of Micheal Jensen (1968) on mutual fund manager’s performance throughout the period 1945 – 1964. Fama suggested that markets are efficient and securities prices reflect all public and insider information and consequently past returns cannot be used to predict future performance. Fama claimed hereby that investors, on average, could only generate the same return as the market and therefore should simply apply a buy-the-market and hold strategy. Jensen applied this idea in a more applied study and concluded from his sample that on average mutual funds were indeed not able to outperform this buy-the-market and hold policy. Although it can be argued that his paper is by now relatively out-dated, Gary Gastineau (2001) chronologically describes what the first securities were that enabled investors to hold and trade an entire market portfolio in a single transaction. Moreover he explains in a very comprehensive manner how the passive investment concept eventually led to the first true ETF product and evolved into a fast growing industry.
In the 1970 and 80s modest progressions in electronic order entry technology and the availability of Fund Characteristics
Trading
TABLE 1
Differrences Exchange Traded Funds and Mutual Funds This table summarized the most fundamental differences between mutual funds and ETFs.
Exchange Traded Funds Mutual Fund
Automatic Investing Minimum Investment
Intraday pricing: ETFS are listed on exchanges and trade like stocks. Prices move like stocks. Like regular stocks options are often traded on ETF
Forward pricing: NAV is priced once a day after closing of the market.
Daily disclosure of holdings. Full disclosure of NAV
during trading hours Months or quarterly disclosure of holdings. Daily disclosure of NAV. Purchased usually via brokerage account. Dependent on distribution agreements with
intermediaries. Bid-ask spread utilized for buy / selling order. Like
trading stocks brokerage commissions are
Sometimes broker commissions. Front end and back load commissions are utilized by funds.
Disclosure Accessibility Transaction Costs Expense Ratios Tax Efficiency Trading on Margin
Single shares in ETFs can be bought. Price will fluctuate like stocks depending on demand.
Limited ability to manage tax liability. Not allowed.
Yes, may carry trade commissions.
Often minimum investment is required depending on manager.
Typically Lower expenses because promotion costs paid by brokers.
Higher expenses included 12b-1 (marketing or distribution costs).
In-kind redemptions and secondary market sales reduces capital gains taxes.
Allowed
“program” transactions. The introduction of the S&P 500 index future contracts made it possible in a single trade called an Exchange of Futures for Physicals (EFP) to swap an entire market portfolio position into a futures position. Since futures contracts are standardized, very large in size and the payments of margin requirements are problematic for smaller investors, market makers recognized the need for a security that could be traded in smaller amounts. The first exchange listed product that satisfied this need in 1989 was the Index Participation Shares (IPS) that tracked the S&P 500. The product showed significant public interest but was short-lived because it was considered to be a futures contract by the federal court of Chicago and should have been traded on a futures exchange. Shortly after, in 1990 another temporary product called the Toronto Stock Exchange Index Participation (TIPS) was issued which mirrored the TSE-35 and TSE-100 index. Being also relative successful due to their low expense ratio they proved to be in turn very cost-inefficient for the exchange and its members itself, which led quickly to their liquidation. After the removal of the IPS at the American exchanges, Standard & Poor’s Depositary Receipts (SPDRS) were introduced which are simple unit trusts holding an S&P 500 portfolio that can be changed in case the composition of the index changes. The selection for the unit trust structure overcame the cost inefficiency problem because the conventional construction of typical mutual funds requires the expensive installation of a board of directors. The “Spider” fund, which is currently still one of the most popular ETFs in the world, only experienced an exponential growth of asset under management (AUM) in the late 1990’s as it took several years before investors became accustomed to the innovative share creation-redemption process. The successful introduction of the Spiders quickly gave rise to similar products offering exposure to other major US indices such as the “Diamonds” (Dow Jones) and “Cubes” (NASDAQ 100). The last important introduction that had considerable impact was the introduction of World Equity Benchmark Shares (WEBS). WEBS were important for two reasons; they hold stocks not issued by US incorporated firms and they are organized via a mutual fund structure instead of a unit trust structure. Although until 2008, most ETFs were required to track specified indices, the range of products now includes funds that track industry indexes and follow smart beta strategies. One of the major ETF houses Power Shares even created a first actively managed ETF in 2008.
3.1.3 ETF Global Market Overview
The ETF market is growing at a rapid pace and due to its popularity there is an abundance of public information available on the ETF market. Due to the quick developments it is very essential to find very recent information. This thesis considers two reports. The US based Investment Company Institute (ICI), which can be regarded as relatively independent source, publishes each year a very detailed report on US holdings in investment funds. Considering that most of the ETFs are still US originated it covers a large part of the ETF market. Although most content is still focused on mutual
funds it latest 2017 report contains a good overview of the latest development in the US ETF market. Not only does it describes the growth of the market but also researches what type of retail investors favour ETFs. Apart from the ICI, Blackrock, which is by far the largest ETF issuer, also publishes a yearly report on latest ETF funds developments. Their reports should be regarded as a complementary source to ICI yearly overview as the Blackrock reports have a global scope. Although very comprehensive it can be argued that this source is less independent. In the following paragraphs the most relevant issuers and markets characteristics are described.
3.1.3.1 Size of ETF Market
Although the $36 trillion AUM in actively managed mutual funds still dwarfs the $4.5 trillion that is currently invested worldwide in ETFs, the popularity of passively managed index mutual funds and ETFs has increased exponentially (ICI Source) since 2000. Following the figures published by Blackrock in December 2016. The US market (73%) remains very dominant as it has by far the highest concentration of ETFs AUM followed from a large distance by Europe (16% and Asia (9%). When looking to the popularity of asset classes, ETFs AUM are mostly concentrated in equity based ETFs (78%) followed by fixed income (17%) and commodities (5%). While equity remains overall the dominant asset class on all continents it can be argued that some regional markets have slight preference for certain asset classes. In Europe 25% of the ETFs AUM are fixed income based, a third higher than the global average, whilst investors in the Asian ETF market seem to have an appetite for more risky asset classes as the AUM are even more concentrated on equity (88%) and remarkably the Commodity based ETFs (9%) outrank the fixed based funds in popularity (3%). When looking to relative growth the BlackRock report writes that over the last year’s fixed income based ETF enjoyed larger growth than equity based ETFs. In the years to come it will be interesting to see whether the composition of the ETF landscape can become fairly different in a couple of years.
3.1.3.2 ETF Sponsors
Worldwide the ETF landscape remains dominated by Blackrock. Through it iShares subsidiary it has almost 37% of the AUM worldwide under its wings. The only region where Blackrock does not lead the AUM ranking is Asia where other more local names such as Nomura, Daiwa and Nikko have a head length advance over IShares. Although the market is still growing it can be said that the market is currently very concentrated around the top ten issuers who manage around 84% of the total AUM worldwide. The level of concentration however seem to differentiate per geographical region, the top ten issuers in the US and Europe hold respectively 95 and 92.5% of the market while in Asia this seems to be more dispersed as 22.7% of AUM are owned by other providers. Having mentioned the all-round dominance of iShares worldwide earlier, apart from State Street and Vanguard, most top ten managers only seem to owe their place in the overall top ranking due to their large market share in their “Home Market”. Power Shares for example has a well-established fourth position on the global ranking but has only significant presence in the US. The same accounts for other global top names such as Nomura with only a lead position in the Asian region and Deutsche Asset Management in Europe. Apart from the top three names it can be said that other issuers seem to have a specialization in their home market and although on a global level the ETF market looks very consolidated growth opportunities in less developed ETF markets such as the EU and Asia and other asset classes might threaten the head positions of the main ETF houses.
3.1.4 Competition between Index Funds and Exchange Traded Funds
With respect to competition, index funds are often viewed as the close cousins of ETFs because both fund types follow passive investment strategies and therefore it is often assumed that they are fishing in the same pool of potential investors. Several interesting papers Rompotis (2009), Agapova (2009, 2010) and Svetina (2010), researched their differences in detail but overall they question whether both fund types are rightfully regarded as substitutes. On an aggregated level, in terms of performance ETFs, all authors find that both fund types substantially produce quite similar results relative to the indices they are tracking. Seemingly, performance wise, there would be no grounds to prefer an index fund over an ETF and vice versa. The results of Svetina’s (2010) research paper deviates from this generic observation because she concludes that the ETFs from her sample seem to deliver slightly better performance than retail index funds.
Regarding the competition between the two fund types, Agapova (2009) also studied the effect of the introduction of new ETF or Index funds on the shares outstanding for existing funds of both types. She found in her sample that the average effect of new entrants permanently reduces the number of outstanding shares of comparable existing fund with five percent after the first share
issuance. Since both fund types seem to be affected, in this regard it can be argued that ETFs and index funds are indeed competing for the same group of investors. Although these results strengthen the opinion that both funds are contending, when looking to investor profiles Rompotis (2009) states that ETF and index fund attracts different types of investors. Investors with a higher risk appetite are described to have a preference for ETFs, while the ordinary, more risk adverse, retail investor would favour index funds. In addition Rompotis argues that ETFs are not only used as investment vehicles but also function as useful hedging tools for large institutional investors, because organizations are not allowed by regulatory restraints to use derivative products to hedge their portfolio against risks. The ICI also dedicates an interesting special page in their yearly fund report to investor profiles. They observe that the average US investor uses mutual funds mainly for the 501 (k) pension plans while more wealthy investors having larger investment portfolios are more likely to invest in ETFs.
Although both Svetina (2009) and Agapova (2010) found reasons to support the competition perception they also bring nuance to their conclusions as their research results leave room to argue that both funds types are not perfect substitutes. Svetina based her statement on the fact that only 17% of her sample of 584 US equity ETFs followed the same benchmarks as any existing index fund, thereby indicating that ETFs offer not only another but also wider range of investment opportunities than index funds. Agapova researched the competition between the two funds in another paper using a very specific example as she investigated exclusively whether the existing portfolio of index funds which are managed by Vanguard are being cannibalized by its newly issued ETFs product range. The results on her research are in particular interesting for fund managers as she states that combining the two fund types in their product portfolio enables the sponsor in fact to benefit from spill over effects and synergies such as tax efficiencies. Both fund types therefore seem to leverage on each other’s success and issuers could have serious benefits from issuing both types of investment products.
4 Literature Review
4.1 Related Research Topics
Even though the turbulent growth figures of ETF assets under management that were seen in the early 2000’s have normalized somewhat in the past years there is little reason to doubt ETF popularity will continue to grow, thereby leaving seemingly plenty of possibilities for new ETFs sponsors to expand. Although ETFs are increasingly popular with investors very little is written on how cash flows of these funds behave in relation to past performance and moreover whether they are also influenced by other factors. Most academic papers on ETFs that do exist describe more technical aspects of ETFs such as the influence of liquidity on the ability of an ETF to track its index and tracking error. This chapter will discuss the most important topics that led eventually to creation of this paper.
4.1.1 Cash flow – Past performance relationship
The two most important research papers that form the basis of this thesis as well as other research papers on cash flows are the Berk & Green (2002, 2004) and Elton et al. (2002) publications. Berk & Green (2004) focus their research on mutual funds and use “alpha” as their unit of measurement for past performance. Alpha is the excess return generated by the fund holdings when benchmarked to the returns of an index that represents the market portfolio. They do find a positive correlation between fund flows and alpha, which indicates that investors indeed chase past returns and seem to invest in the mutual funds with the best performance. Berk & Green (2004) write also in their paper that several other factors such as the age of the fund appear to have a moderating effect as they observe that relatively cash flows appear to respond more heavily to performance for young funds than older funds. Evans (2010) describes the influence of age on mutual funds flows in more detail. Evans explains that after its introduction, a new fund enters a so-called “incubation” phase for about a year. Throughout this incubation period fund flows more respond more heavily to performance than they would at a later stage when they reach maturity.
Together with Jules van Binsbergen (2013) Jonathan Berk published a somewhat revised version of his previous 2004 paper whereby they did not use alpha, but considered absolute return as a more reliable proxy for performance. Green & Binsbergen deem absolute returns as a more reliable alternative to alpha because several researches like Stein (2002), Chen et al (2004) indicated that relative performance ratios such as alpha are wiped out when mutual funds increase in size. Consistent with his earlier paper, investors in mutual funds appear to invest in the mutual funds that generated constant absolute returns. Although it can be argued that large funds are more cost
effective because they benefit from economies of scale Tufano and Sevick (1997) hypothesize why relative performance such as alpha is wiped out when funds grow larger. When the fund increases in size the fund manager looses flexibility because it becomes harder to create or liquidate positions as a result of the increasing size of the transactions. Chen et Al. (2004) were among the first to research this relationship and state that liquidity is a very contributing factor in this case as they observe that relative performance especially erodes for funds that invest in small cap stocks. Xuemin Yan (2008) found evidence for this observation of Josheph Chen regarding liquidity. Yan also raises an interesting question whether liquidity interacts with mutual fund flows. Other small important relations from Yan are that expenses seem to be negatively correlated with fund age, thereby indicating that funds become cheaper when they grow older. Not surprisingly, another observation of Xuemin is that fund size seems to be positively correlated with the age of the fund.
While the earlier mentioned papers focus on actively managed mutual funds, the paper of Elton et al (2002) is focused on S&P 500 tracking index funds. Elton recognizes that alpha is a less relevant performance measurement for passively managed funds and describes the usage of additional variables such as investment cost (management fees and other costs), tracking error volatility and structuration risks as proxies to measure performance. Intuitively this model might be more difficult for the average private investor to comprehend than alpha, in theory however, in line with the findings of Berk & Green (2004) these factors should function as the alpha for passively management funds. Elton concluded not only that cash flows to index funds respond to past returns in the same way as flows to mutual funds, they also observed that new assets seem to flow into index funds that have lower performance ratings. Elton et Al attributes this “irrational” investment behaviour to marketing activities of brokers and other intermediaries. In the fund management industry it is not uncommon that fund issuers often sign distribution agreements with brokers and investment advisors who in turn influence investors to choose funds out of their fund portfolio that might not necessarily have the best performance track record. This matter is also described in the latest asset management study report of the FCA that was published recently. In the report the supervisor is in particularly concerned about the vertical integration of advice and fund management firms. Another important aspect from the Elton paper that is also discussed later in this paper considers the description of the fund family effect. They write that regardless of positive past performance, the reputation of an established fund manager seem to be enough to influence investors in their fund selection process. Considering the current consolidation of the ETF market around a limit number of major players that is described in paragraph 2.1.3.2 there is reason to assume that ETF cash flows are also influenced by the family effect.
While there is an abundance of literature available on mutual fund flows one research paper was found that investigates the relationship between ETFs fund flows and performance. Serdar Kalaycioglu (2004) examines this relationship by using daily, weekly and monthly data frequencies of five major ETF index trackers from the period may 2000 until December 2003. To measure fund flows Serdar uses two methods. Firstly he measures the fund flows per ETF using the change in outstanding shares between the periods. Secondly he computes an average fund flow variable using a market capitalization weighted average of the fund flows of each single ETF. In his paper he documents a surprisingly significant negative correlation between the individual and aggregated flows and the market returns and their lags. His explanation of this negative relationship is twofold. Firstly Serdar writes that ETFs are often used for cash management purposes and as hedging instruments. Second explanation is that fund flows create pricing pressure on the market returns of the ETFs. Having written earlier that in the last years the ETF market has exploded it can be considered that the relative small sample of Kalaycioglu, which covered only the time period of March 2000 to 2003, is somewhat out-dated and not representative for the todays ETF market. A more recent but non-academic survey held in 2014 representing 87% of the ETF industries global assets at the time, consultancy firm Ernst and Young (2014) identified the most important selection criteria for investors when choosing an ETF. The bid-ask spreads (13%), management fee (17%), tracking error (20%) and liquidity/size (23%) were stated to be relevant factors however the fund manager reputation (30%) was specified by investors as the most important reason to select a ETF. Triggered by the large percentage of respondents selecting reputation, as most important reason to invest in a certain ETF there is also room to question whether the ETF market is indeed free from significant market disturbances.
4.2 ETF Fund Flows & Performance Measurements
Other than Mutual funds, ETFs should provide investors with the same returns as their underlying index and therefore typically performance measurements such as alpha, absolute or adjusted returns are not useful as proxies for ETF performance. A unique research paper published by Hassine and Roncalli (2013) is dedicated entirely to a methodology to distinguish badly from good performing ETFs. Through the usage of historical data on data tracking error volatility, liquidity and information ratio as proxies for performance they conclude that some fund managers consistently perform better than others. This should mean that funds with low tracking error volatility, high liquidity and information ratios, should experience the highest inflow of new assets under management. This paper is however not focused on distinguishing good from bad performing fund managers but we will test whether investors respond to these performance indicators that are used as variables to measure the efficiency ratio that is developed by Roncalli and Hassine. Before testing the effect of
these indicators on fund flows this paragraph will discuss the most important literature that has been published on the ETF performance indicators.
4.2.1 Fund Flows
Literature on Mutual Fund performance and fund flows describes different ways to measure fund flows. Serdar Kalaycioglu (2004) uses in his paper a combination of two methods. Firstly he uses the changes in total shares outstanding while in his second method he considers the changes in total net assets (TNA). The growth rate of total net assets (TNA) has been used in some of the most influential papers on mutual fund flows. TNA data is available in CRSP survivorship free mutual fund database. For instance, in Sirri and Tufano (1998) the fund flow (𝐹𝐹𝑖𝑖,𝑡𝑡 ) is measured by:
(1) 𝐹𝐹𝑖𝑖,𝑡𝑡= TNA𝑖𝑖,𝑡𝑡+1 − TNATNA 𝑖𝑖,𝑡𝑡 (1 + r𝑖𝑖,𝑡𝑡) 𝑖𝑖,𝑡𝑡
Where (𝐹𝐹𝑖𝑖,𝑡𝑡 ) is the net flow into fund i at time t, and (𝑅𝑅𝑖𝑖,𝑡𝑡 ) is the respective return of the fund. Most
studies to date considered only net flows, as the data is readily available while the data of inflows and outflows is more cumbersome to collect. For US based ETF these flows need to be collected manually from SEC filings through the usage of N-SAR reports. As net flows are the result of offsetting gross inflows with gross redemptions it can be expected that both flows be also affected by performance in very different manners. Clifford et al. (2011) and Edelen (1999) found that the relationship of fund inflows to performance is much stronger than that of outflow. Ivkovic and Weisbenner (2009), however, found both inflow and outflow are sensitive to past performance, and that inflow is more responsive to relative performance while outflow is affected by absolute performance. Important restraint of the N-SAR fillings that can be obtained from the SEC website is that it only relates to US based ETFs so in case info on out flows on non-US ETFs is required other sources need to be utilized. No research describes any databases that include the in- and outflows on non-US based funds.
Another measurement that is utilized considers the growth total market capitalization of the fund. This variable however is seemingly impractical as the total market capitalization aggregates both the inflows of new investments as well as the increase of its NAV. While NAV fluctuates daily some funds don’t experience creation or redemption of shares on a daily basis therefore uses growth in market capitalization therefore it can be argued that using this variable to measure fund flows can lead to biased results.
4.2.2 Tracking Error measurements and their determinants
Markowitz (1952) defines two criteria that should be considered by investors when selecting their investment portfolio. Expected return is regarded, as a desirable thing while variance of returns is considered undesirable. Following this principle, the efficient portfolio generates the maximum amount of return for a given level of risk (risk corresponds to the variance of returns). With regard to ETFs it is undesirable that the tracker performance deviates from the returns that are generated by its index. Despite the aim to replicate the same returns and risk exposure of its underlying index not all ETFs full their objective with the same level of accuracy. The performance differences are caused by the fact that replication strategies cannot track their benchmark exactly at all times. The terminology that is commonly used to refer to these deviations is the tracking error. If a sponsor is not able to track its benchmark accurately, investors may interpret this as that the fund is failing to meet his investment objective. Needless to say that any ETF sponsor should try to minimize its tracking error ratio as much as possible.
In the efficiency measurements model of Hassine and Roncalli they replace the volatility of the “market portfolio” by the volatility of the tracking error as a proxy for risk, however they are not the first to consider tracking error as performance measurement for ETFs. Results on tracking performance have not been conclusive, as there seem to be substantial differences between ETFs Types and exchanges. Most research has been done on ETFs listed in the United States and a few have discussed particular countries in Europe or Asia. Roll (1992), Poper and Yadav (1994) were amongst the first to discuss the utilization of tracking error to assess the performance of passively managed index funds whilst Frino an Gallagher (2001,2002) were the first to practically apply this ratio as a way to measure the performance of S&P 500 index funds. Frino and Gallagher describe several factors that influence tracking error. Firstly they hypothesize that tracking error is caused by delays in receiving stock dividends and changes in the S&P 500. Dividend yields are related to tracking error as the index immediately assumes that when listed stocks that are part of an index pay dividends, these are reinvested immediately. In reality, investors as well as the ETF sponsor, face delays in receiving dividends payment. Because of these delays the ETF ability to track the index are eroded because their fund managers have to wait before the cash obtained from the dividends can be reinvested. Extra costs are made because of the trading activities that are required to reinvest the dividend payment occur additional transaction costs. Frino and Gallagher also relate tracking error with ETF expenses, lower expenses result in lower tracking error ratios. Expenses are the associated costs when operating in financial markets, which includes brokerage fees and stamp duties. Economies of scale are also related to tracking error. Because of size relative trading costs go down which leads to lower levels of tracking error. Lastly the level of risk in the market is an important
factor. When the market faces a lot of volatility it becomes harder for the index to track their index leading to higher tracking error ratios. One of the main examples is the flash crash of 2010 that occurred as a result of the volatility in the stock market that was caused by the Greek debt-crisis2. Although tracking error is described in academic literature as one performance variable there are several different methodologies to compute tracking error. Aroskar and Ogden (2012) summarized five methods in order to calculate the tracking error for Exchange Traded Notes (ETN). Despite the fact that Aroskar and Ogden used a sample of ETNs instead of ETFs, the same tracking error formulas are also applicable for ETFs. Their paper describes five different methods to compute tracking error however in academic papers only three measurements seem to be used. In line with academic practice this paper will consider the only the first three methodologies however for the sake of completeness all five methods are discussed and explained in this section.
The first and most simplistic method to calculate the tracking error is utilized by Wong and Shum (2010). Firstly they compute the continuous (logarithmic) daily returns of the ETF (𝑅𝑅𝑖𝑖,𝑡𝑡 ) and its index
(𝑅𝑅𝑏𝑏,𝑡𝑡 ) using the same identical formula:
(2) 𝑅𝑅𝑖𝑖,𝑡𝑡= ln �𝑅𝑅𝑅𝑅𝑖𝑖,𝑡𝑡
𝑖𝑖,𝑡𝑡−1� and (3) 𝑅𝑅𝑏𝑏,𝑡𝑡= ln �
𝑅𝑅𝑏𝑏,𝑡𝑡
𝑅𝑅𝑏𝑏,𝑡𝑡−1�
Wong and Shum calculated the tracking error (𝑇𝑇𝑇𝑇1) simply by subtracting the fund’s return from the
return of the index. For instance, an ETF tracking the CAC40 had a 5% return, whereas the CAC40 index had a 6% return. The tracking error is 1%.
(4) 𝑇𝑇𝑇𝑇1 = 𝑅𝑅𝑖𝑖,𝑡𝑡− 𝑅𝑅𝑏𝑏,𝑡𝑡
This simple method to compute tracking error does distinguish between a positive and a negative tracking error. In the case that the ETF is not consistent with having either a positive or a negative tracking error, the final value for the tracking error using this method may not be accurate and lead to an underestimation of the true tracking error. Although this simple method has its limitation it is a fast and easy way to identify what the degree of tracking error is of a particular fund. The practical usage of this tracking error indicator would not be interesting for investors with a long position as they will surely not complain if the ETF will perform on a certain day slightly better of worse than its index. Most investors would be more interested to see on average how accurate the ETF tracks its index. This is our second method to compute tracking error.
2 When stock markets opened in the US on May 6, 2010 the Dow Jones index was down. This decrease was related to the concerns around the debt crisis in Greece. In the afternoon with the Dow Jones down more than 300 points for the day, the equity market began to fall rapidly, losing an additional 600 points in
This second method to determine the tracking error (𝑇𝑇𝑇𝑇2) is first described in a study by Gallagher
and Segara (2005) and can be regarded as a slightly more sophisticated version of the first method that was described earlier and should thus be more informative than the first method. The second method, the daily average absolute tracking error, is calculated as follows:
(5) 𝑇𝑇𝑇𝑇2= ∑ �𝑒𝑒𝑖𝑖,𝑡𝑡� 𝑛𝑛 𝑡𝑡=1
𝑛𝑛 and (6) 𝑒𝑒 = 𝑅𝑅𝑖𝑖,𝑡𝑡− 𝑅𝑅𝑏𝑏,𝑡𝑡
This method uses the sum of the absolute difference between the returns instead of the simple difference. Where n is the amount of days and the simple difference (𝑒𝑒) is defined as the differentiation between the return of the ETF and the index.
The third way and much more explanatory way to compute tracking error (𝑇𝑇𝑇𝑇3) uses the standard
deviation of the difference in the fund and index returns over time. This method is referred to in the paper of Roncalli and Hassine as the tracking error volatility. Described again first by Gallagher and Segara the standard deviation of the difference is calculated using the following equation:
(6) 𝑇𝑇𝑇𝑇3 = �𝑛𝑛 − 1 ��𝑒𝑒1 𝑖𝑖,𝑡𝑡− 𝑒𝑒��𝚤𝚤 2 𝑛𝑛
𝑡𝑡=1
By using the standard deviation calculation, investors obtain a better understanding how well the ETF is tracking its benchmark over time. A low deviation is positive as it indicates that the ETFs replicates the index very closely whereby a high standard deviation informs the investor that the ETF is not very accurate in tracking its benchmark. Important to note is the fact that if the ETF consistently underperforms or outperforms its benchmark index by some stable amount every single day, the standard deviation (and thus the tracking error) will be zero and the tracking error using this third method will understate the actual tracking difference. Nevertheless, this third method is thus one to use with some caution. Despite this, it is a method widely used in academic literature for calculating the tracking error but it has the previously mentioned pitfall. Similar to the first method, it is best to combine this measure of tracking error with some other indicators.
The last two methodologies consider a linear regression approach and use R-squared to indicate how well an ETF tracks its underlying benchmark. Aroskar et al. (2012) argues that the R-squared of the a normal regression equation is another indicator for the tracking ability because it is a statistic that is very intuitive and easy to interpret. It shows how much of the variation in the ETF price is explained by variation in the price of the underlying index. Chu (2011) also uses the outcome of a normal regression equation but uses the standard error as a method to determine the tracking error. The
standard error of a regression can be seen as the average distance of all of the observed data points and the estimated regression line. It is important to note however that this method only gives a good approximation to the tracking error in case the β-coefficient is equal to 1. If the β-coefficient is not equal to 1, the tracking error may be overstated according to Pope and Yadav (1994).
4.2.3 Bid-Ask Spread & Liquidity
Other than typical fund costs which are related to trading costs and management fees there are other hidden costs such as the bid-ask spread that effects the returns for investors. The bid-ask spread resembles the difference between the “bid” price which indicates at which price investors can buy the security and the ask price which reflects at which price the ETF can be sold. The second performance indicator described by Rocalli and Hassine considers the bid-ask spread that is a measurement that indicates the liquidity and efficiency of the fund. An ETF usually trades as closely to its or NAV, as possible. ETF Liquidity is a popular research field because of the complex creation and redemption process of ETF baskets which can lead to large deviations from the NAV. Early research (Elton, Gruber, Comer and Li) on ETFs show that most funds underperformed their index due to high transaction costs and liquidity issues. They write that several developments in the market led to significant improvement in the liquidity problem. Intuitively most easy development to understand was the maturity of the market. As more sponsors started to issue new ETFs, due to competition funds costs started to go down while liquidity increased. Calamia, Deville and Riva (2013) describe a very specific example whereby the introductions of two new fund tracking the CAC 40 broker the four year monopoly of Lyxor and resulted in a significant drop in the spread between the bis ask price. Other more fundamental changes are the conversion from a system using fractional tick sized to more practical usage of decimal sizes. While Nguyen, van Ness (2007) showed that multimarket trading of ETFs also expanded liquidity, especially for the larger funds. However, for some low volume ETFs, bid-ask spreads may exist and widen. Trading ETFs with large spreads eats away at potential returns since they affect the ETF purchase and sales prices.
4.2.4 Information Ratio
The information ratio indicates whether a fund manager managed to beat its benchmark in a specific period. The ratio serves as a tool that investors use when selecting exchange-traded funds (ETFs) or mutual funds based on investor risk profiles. Although the funds being compared may be different in nature, the information ratio standardizes the returns by dividing the difference between index and ETF return with the standard deviation of the tracking error.
with a low ratio by taking on additional risk. Additional risk can be achieved through leveraging. Information ratio is calculated using the following formula.
(7) 𝐼𝐼𝑅𝑅𝑖𝑖,𝑡𝑡 = 𝑅𝑅𝑏𝑏,𝑡𝑡− 𝑅𝑅𝑖𝑖,𝑡𝑡
� 1𝑛𝑛 − 1 ∑ �𝑒𝑒𝑛𝑛𝑡𝑡=1 𝑖𝑖,𝑡𝑡− 𝑒𝑒��𝚤𝚤 2
For example, assume fund manager A has an annualized return of 14% and a tracking error of 9%, while fund manager B has an annualized return of 9% and tracking error of 4%, and the index has an annualized return of -2%. Therefore, manager A's IR is 1.333, or 14 - (-2) / 9 and manager B's IR is 1.75, or 9 - (-2) / 4. Although manager B had lower returns, the portfolio has a better IR. This problem concerns benchmarked funds with low tracking-error volatility in particular. In this case, the information ratio cannot be used to select funds. This is especially true for trackers that aim to have the lowest tracking-error volatility. Moreover, the information ratio is not appropriate for measuring the relative performance of the tracker that presents the lowest tracking error. Indeed, it is generally a fund that has a small but negative excess return performance. Because its information ratio is negative, this tracker is not efficient from the information ratio perspective. It means that no investors are interested in using it. This theoretical point of view is in contradiction with practice, because passive investors like to consider funds with the smallest tracking error volatility.
The Sharpe ratio is similar to the IR and is used to measure risk-adjusted returns. However, the Sharpe ratio measures the difference between an asset's return and the risk-free rate of return divided by the standard deviation of the asset's returns. The difference between the Sharpe ratio and the IR is that the IR aims to measure the risk-adjusted return in relation to a benchmark, such as the Standard & Poor's 500 Index (S&P 500). Additionally, the IR measures the consistency of an investment's performance, while the Sharpe ratio measures how much an investment portfolio outperformed the risk-free rate of return on a risk-adjusted basis. Combating both the dimension of risk as well as performance, the Sharpe Ratio and Information Ratio are regularly used to directly compare the performance of identical funds. Although the Sharpe Ratio adds little value to measure ETF performance because ETFs do not focus on outperforming their benchmark, the Information Ratio mentioned by Grinold & Khan (2000) is regarded as a better measurement as it captures how much the returns of the ETF deviate from the benchmark returns and puts it into perspective with the risk associated with this deviation. There are however some drawbacks when using the information model. Firstly that it assumes that the benchmark can be fully replicated. Secondly it ignores the magnitude of tracking error volatility, which may lead to ranking a fund with higher tracking error volatility above a fund with marginal volatility.
5 Sample & variable description
Parallel to the growth in ETF Assets under Management, the ETF fund market experiences an increase of new sorts of ETFs. While the first ETF mainly used equities as underlying assets class and tried to fully replicate the returns of a well-known broad market index such as the Dow Jones, nowadays a wide range of different ETF products are now available to investors. Apart from the focus on different asset classes and the increasing usage of leverage, there is also an growing number of funds that look for certain niches in the market, the Horizons Medical Marijuana Life Sciences ETF (HMMJ) that invest mostly in Canadian medical companies that grow medicinal Marihuana being the most colourful outlier. To be clear what kind of fund relationships are being tested in this thesis, the following chapter described the data sample along with its basic characteristics.
5.1 Data collection & Variable definition
This study is limited to non-leveraged equity based ETF’s leaving other asset classes for future research. While most researches on ETFs are limited to US based securities this data sample will also include European and Asian ETFs. For the collection of the fund data two databases, Bloomberg and the CRSP Survivor-Bias-Free Mutual fund database (that also contains ETF data nowadays) were considered. Although the CRSP survivorship-bias free database is the most commonly used source in academic studies due to the free access for academic institutes, the information utilized in this study is mainly obtained from Bloomberg. Bloomberg also includes information from active and funds that have been liquidated therefore the database eliminate survivorship bias.
5.1.1 Sample construction
First step in the construction of the ETF sample was to find equity based ETFs that were actively traded at the time of this study. Lists with ETFs that are currently traded can easily be downloaded from the public websites of most ETFs sponsors, however the amount of fund information that is published varies per company. After the completion of the ETF list a practical impediment was encountered when trying to download information from Bloomberg. Although Bloomberg has a user-friendly way to download information, to do so it is required to utilize the unique Bloomberg ticker for each ETF. While looking up tickers in Bloomberg is not difficult it is important to note that a single fund can be listed on various exchanges and therefore has for each place of listing a unique ticker. To avoid the inclusion of the same fund in the sample twice, any duplications were identified through the usage of the unique ETF ISIN3 and CUSIP4 codes of each fund. After downloading initially
3 International Securities Identification Number (ISIN) is a unique code twelve-character code that serves for the uniform identification of a security. ISINs are issued for instruments such as bonds, commercial paper, stocks and warrants. The ISIN codes is not specified per location of listing, and therefore often another identifier such as the exchange code, are mentioned in addition to the ISIN.
information of 8000 Bloomberg tickers the decision was made to include only the data belonging to the ticker for the principal exchange where the fund is listed. It is worth mentioning this as the bid, ask and closing price of the same fund can vary due to the place of listing. In the fund literature this attributed by the fact that some stock markets are more liquid than others. After matching the Bloomberg tickers with the Bloomberg database monthly data from 2007 to 2017 on the Net Fund Flow, NAV prices, Market Value, Bid-Ask prices, closing prices, traded volume in secondary market and the numbers of outstanding shares were retrieved.
The second step was to focus the sample by removing incomplete information. When examining the completeness of the data 2168 ETFs are encountered with at least some complete information on all variables per monthly data point. When conducting random normality checks on the main variables several funds were identified with consistent extreme values. To increase normality of the sample and reduce the impact of these outliers on the regression models described in chapter 4, all extreme observations that fell inside the lowest 1st and highest 99th percentile were removed. These filters, as well as removing the data points whereby we have incomplete data leads to a sample of 2121 ETFs with 153544 of monthly observations throughout the period of 2007 – 2017. Going forward we will refer to this sample as the “Raw Data Sample”.
5.1.2 Variable computation.
After the completion of the Raw Data Sample some data computations have to be made to generate the additional variables that are needed for the execution of the regression models in Chapter 4. This paragraph will outline what computations were made to the variables in the Raw Data sample and how the final data set was generated. The final data is referred to as the “Data Sample”. The composition of the sample and its characteristics is described in paragraph 3.2.
ETF Type 213 0.1% 28665 18.7% 4610 3.0% 7076 4.6% 291 0.2% 39138 25.5% 9096 5.9% 18313 11.9% 3180 2.1% 898 0.6% 1031 0.7% 6114 4.0% 2536 1.7% 1066 0.7% 173 0.1% 13416 8.7% 2158 1.4% 15570 10.1% Broad Market Large-cap Mid-cap Small-cap Nonindex Table 3
Representation of ETF styles in Raw Data Sample
This table provide the summary statistics from 2007 to 2017 for the raw data sample of equity ETFs. This table shows the distributions of the monthly observations along the different ETF types. The sample is mutually exclusive meaning that the total of all percentages sums up to 100%. An observation is included in the raw data sample when complete data was available in Bloomberg for all variables. The ETF type categorization is determined by Bloomberg classifications. The categorization methodology is based on fund name, investment objective and summarized holding allocation.
Replication Strategy
Full Optimized Derivative
5.1.2.1 Monthly fund flows
While fund management literature focuses on TNA, this thesis makes use of the monthly fund flow variable that is available in the Bloomberg database. This variable represents the net value of the shares that are created or redeemed by the authorized participants as a result of increasing or decreasing demand in the ETF. When looking to the skewness of the fund flows data, the variable is negatively skewed due to the outliers that are caused by the inflows of the larger funds in the sample. To mitigate the impact of these extreme values Joseph Chen () decides in their paper on fund size and performance to take the logarithm of each observation. When taking the logarithm from the net funds flow leads to a variable that follows a smooth normal distribution. By taking the fund flows of new assets under management the fund flows deviate from the TNA results that are taken from a combination of the closing price is multiplied by the number of shares outstanding.
5.1.2.2 Fund Returns and index returns
The Monthly Fund Returns for each individual fund and the Monthly Index Return of its underlying index are the first two variables that are computed. To generate the returns for each index, monthly closing prices for each index are retrieved from Bloomberg in addition to the ETF return data that was obtained earlier. It is essential to mention this as other papers used the NAV of the ETF holdings in order to calculate the returns of the fund. This study uses the closing prices of each fund and not the NAV. Closing price is important because it resembles the price for which the investor can buy or sell the fund. The tracking error that will be discussed later in this chapter will be based on the closing price instead of the NAV. Even if the NAV was used instead of the closing price, it is not expected that it would generate very different results because the share creation and redemption process generally keeps ETF prices and their NAV very close.
To cover for the sample of ETF, monthly price data for 1470 individual indices is to be collected. The high number of indices that is included in the sample is not surprising as it was found earlier in the literature (see chapter 2.1.1) that the number of new indices that are raised by ETF sponsors have been exploded in the past years. As can be seen in table three and four, a large part of the sample data sample does not follow a particular mainstream index. Continuous returns for both the funds and indices are computed using the following formulas (2) and (3) that are described in chapter 2.2.3.
(2) 𝑅𝑅𝑖𝑖,𝑡𝑡= ln �𝑅𝑅𝑅𝑅𝑖𝑖,𝑡𝑡
𝑖𝑖,𝑡𝑡−1� and (3) 𝑅𝑅𝑏𝑏,𝑡𝑡= ln �
𝑅𝑅𝑏𝑏,𝑡𝑡
𝑅𝑅𝑏𝑏,𝑡𝑡−1�
extra data points are lost because the Bloomberg database does not include the monthly returns of all underlying indices. Only funds whereby the corresponding index returns were found are included in the final data sample. The returns of both the ETF and each index had be retrieved otherwise the tracking error volatility could not be calculated. This will be explained in more detail in the next paragraph.
5.1.2.3 Tracking Error volatility and bid-ask spread
The most important dependent variable for this thesis is the tracking error volatility. Similar to the volatility of a stock the tracking error is an indication of risk because it indicates to what extend the ETF returns deviate from their benchmark. The Bloomberg and CRSP database contain cross sectional data regarding the tracking errors and the information ratio however both databases do not store monthly time-series data on these variables. Bloomberg being the main source for this thesis computes an average of tracking error for the past 1, 3, 6 and 12 months as well as the past 3 and 5 years but this information is only useful for regressions using cross sectional data but not for panel regressions. Since tracking error and in particular tracking error volatility are the most important variables for this thesis it was decided to manually compute using three tracking error coefficients. Using the monthly returns data which was discuss in the earlier paragraph the tracking error was calculated in three ways according the formulas (4), (5) and (6) that are described in the ETF. To compute the average tracking error and volatility for each monthly observation proved to be cumbersome because in order to do so daily returns for the 153544 monthly observations had to be calculated. In order to compute the monthly tracking error average and volatility for the entire sample a decision was made to improvise. To compute the tracking error volatility the mean and standard deviations for each data point was calculated using the eleven previous data points. While calculating these monthly means and standard deviations we lose one additional data point on top of the observations we lost while computing the monthly returns based on closing prices.
5.1.2.4 Other variables
Apart from the main three variables fund flows, tracking error and bid – ask spread we compute several additional variables. In this short paragraph we briefly describe these variables and how they are computed. In the paper of Elton et Al (2004), a note was made regarding the family effect. The family effect entails that the reputation of the fund manager has mitigating effect on the fund flows. As a proxy for reputation we check which funds in our sample belong to the same fund manager and compute the size of fund family by aggregating their market value. Similar to net fund flows the distribution of the fund family variable is negative skewed and has a very large tail because of the
inclusion of some very large fund families such as Ishares and State Street. To deal with these extreme variables similar to fund flows the logs are taken from variable family fund leading again to a better normally distributed sample. Another computation concerns the fund age variable. The fund age is calculated when first looking up the inception date of each fund in our sample. After collecting all inception days, the age of the fund at the time of the observation was calculated subtracting the date of the observation with the inception date. After the computation some of the ETFs appear to have a negative age. When examining these negative outliers we find that this concern funds that were either merged with another fund or were taken over by other fund managers. In order not skew the results of our fund family results we decide to leave out the observations with a negative age. After removing these outliers the distribution of the age category proved also to be negatively skewed. To make the fund age variable better suited to a linear regression model the natural logarithm are taken again leading to a normally distributed sample.
After the completion of the fund flows the bid-ask spread is calculated for each data point by taking the difference between the bid and ask prices. The bid and ask spread is based on the monthly closing prices. Similar to the tracking error the average bid ask spread and volatility is calculated using the earlier eleven monthly data points. Lastly having calculated the tracking error volatility on a monthly basis it is possible to compute the information ratio use the formula (7) for each monthly observation.
5.2 Descriptive Statistics and Preliminary Analyses
As can be seen in table three and four the data sample consists of different types of ETFs. Table (4) presents the size of our final data sample after removing all the observations that are lost as a result of the data computations. To focus the data sample the decision was made to remove the few funds that follow a blend portfolio replication strategy as the number of observations was too little to generate results on which any general inferences could be based.
5.2.1 Presentation of the data sample
The largest part of the sample that can be seen in table four consists of ETFs that follow broad and Large-cap market indices. Broad Market indices such as the Dow Jones or Willshire 5000 reflect the movement of an entire market. Popular Large cap indices are the S&P 500 and Nasdaq. Large cap is an abbreviation of the term “large market capitalization” and refers to indices that track the performance of companies with a market capitalization exceeding ten billion dollar according to Bloomberg.
