Three essays on exchange traded funds

... focuses on the trading activity o f bond ETFs on the secondary market (trading on exchanges), ETFs can also trade on the primary market through the creation/redemption process Trades on the primary... The Bond ETF Market 28 I Introduction 29 II Bond ETFs and Data 31 III Results 34 IV Conclusions 41 Essay Three: D oes Exemption from the Uptick Rule Effectively Alleviate 43 Short-Sale Constraints?... reproduction prohibited w ithout perm ission I Introduction An exchange- traded fund (ETF) comprises o f a basket o f either stocks or bonds It has characteristics common to both closed- and open-end funds

THREE ESSAYS ON EXCHANGE-TRADED FUNDS Ramabhadran S. Thirumalai Submitted to the Faculty o f the University Graduate School in Partial Fulfillment o f the Requirements for the Degree Doctor o f Philosophy in the Kelley School o f Business Indiana University May, 2004 R ep ro d u ced with p erm ission o f th e copyright ow ner. Further reproduction prohibited w ithout perm ission. UMI Number: 3134050 Copyright 2004 by Thirumalai, Ramabhadran S. All rights reserved. INFORMATION TO USERS The quality of this reproduction is dependent upon the quality of the copy submitted. Broken or indistinct print, colored or poor quality illustrations and photographs, print bleed-through, substandard margins, and improper alignment can adversely affect reproduction. In the unlikely event that the author did not send a complete manuscript and there are missing pages, these will be noted. Also, if unauthorized copyright material had to be removed, a note will indicate the deletion. UMI UMI Microform 3134050 Copyright 2004 by ProQuest Information and Learning Company. All rights reserved. This microform edition is protected against unauthorized copying under Title 17, United States Code. ProQuest Information and Learning Company 300 North Zeeb Road P.O. Box 1346 Ann Arbor, Ml 48106-1346 R ep ro d u ced with p erm ission o f the copyright ow ner. Further reproduction prohibited w ithout p erm ission. ACCEPTANCE Accepted by the Graduate Faeulty, Indiana University, in partial fulfillment o f the requirements of the Degree of Doctor of Philosophy in Business. / V Craig W. Holden, Bo Chairman Robert H. Jennings Doctoral Committee (December 17, 2003) 11 R ep ro d u ced with p erm ission o f th e copyright ow ner. Further reproduction prohibited w ithout perm ission. © 2004 Ramabhadran S. Thirumalai ALL RIGHTS RESERVED 111 R ep ro d u ced with p erm ission o f th e copyright ow ner. Further reproduction prohibited w ithout perm ission. ACKNOWLEDGMENTS I would like to thank my committee members, Craig Holden, Robert Jennings, Heejoon Kang, and Charles Trzcinka, for their help and guidance towards the completion o f this dissertation. I would also like to thank Neal Galpin, Scott Smart, and Chad Zutter for helpful discussions and comments. I am grateful to the Finance faculty for their support over the years. I appreciate the encouragement from my fellow doctoral students. I would also like to thank m y friends for urging me on to finish this dissertation. I am thankful to my parents and my sister for their constant support and words o f encouragement. My deepest thanks go to my wife, Gayathri, for her patience, support, and understanding. IV R ep ro d u ced with p erm ission o f th e copyright ow ner. Further reproduction prohibited w ithout perm ission. ABSTRACT Essay one compares the pricing and liquidity o f passive exchange-traded funds (ETFs) to those o f active ETFs. Consistent with predictions, I find that average absolute price deviations from net asset value (N A V ) are larger for a non-arbitrageable basket security (active ETFs) than for an arbitrageable one (passive ETFs). Flowever, the mean price deviations are positive for passive ETFs and not different from zero for active ETFs. Further, contrary to predictions from theory, 1 find that the bid-ask spreads for active ETFs, which have one market maker each, who has an information advantage relative to the rest o f the market, are significantly narrower than those for passive ETFs, which have multiple market makers. Essay two examines the performance o f four bond ETFs relative to their underlying indices. Since all four ETFs hold only a representative sample o f bonds from their underlying indices, ETF performance may differ from index performance. I find that all four ETFs underperform their respective indices, which may be attributed to the ETFs holding only the most liquid bonds from their indices. I also examine the deviations o f market price from N A V . I find that the short­ term Treasury ETF has the smallest deviations and the corporate bond ETF has the largest deviations. This difference may be attributed to the difference in the underlying bonds’ liquidity. One puzzling result, however, is the positive premiums that these ETFs trade at on average. Essay three studies the effect o f ETF option listings on the underlying ETFs. Earlier studies have documented negative abnormal returns in equity around the introduction o f equity options, which have been attributed to the alleviation o f short-sale constraints on the underlying stocks. One short-sale constraint faced by investors is the uptick rule that prohibits investors from short selling on a downtick. Flowever, since ETFs are exempt from this rule, they may face lower short-sale constraints. This suggests that there may be no abnormal returns in ETFs around options listings. Consistent with this, I find no abnormal returns around options listings suggesting that ETFs are not short-sale constrained. I also find that there is no change in short interest around options listings. R ep ro d u ced with p erm ission o f th e copyright ow ner. Further reproduction prohibited w ithout perm ission. TABLE OF CONTENTS Essay One: Active vs. Passive ETFs 1 I. Introduction 2 II. Data and Methodology 10 III. Results and Discussion 11 IV. Conclusions 25 Essay Two: The Bond ETF Market 28 I. Introduction 29 II. Bond ETFs and Data 31 III. Results 34 IV. Conclusions 41 Essay Three: D oes Exemption from the Uptick Rule Effectively Alleviate 43 Short-Sale Constraints? I. Introduction 44 II. Data and M ethodology 47 III. Results 50 IV. Conclusions 56 Appendix 58 References 60 Tables 65 Figures 93 VI R ep ro d u ced with p erm ission o f th e copyright ow ner. Further reproduction prohibited w ithout perm ission. L IST OF T A B L E S A N D FIG U R E S Table I; Differences between passively and actively managed exchange-traded funds Table II: Summary statistics on ETFs Table III: Summary statistics on ETF premiums Table IV: Summary statistics on premiums o f ETFs matched on benchmark indices Table V: Summary statistics on premiums o f ETFs matched on geographic region o f benchmark indices Table VI: Linear regression o f premium measures on characteristics o f ETFs Table VII: Tests for persistence in ETF premiums Table VIII: Summary statistics for liquidity measures o f ETFs Table IX: Linear regression o f liquidity measures on trading characteristics o f ETFs Table X: Correlation coefficients Table XI: Daily turnover o f ETF shares Table XII: Creation and redemption o f bond ETF units Table XIII: Performance o f bond ETFs relative to their underlying indices Table XIV: Pricing o f bond ETFs relative to their net asset values Table XV: Linear regression o f premiums on lagged premiums Table XVI: Linear regression o f premium measures on characteristics o f ETFs Table XVII: N ew listings by options exchange and year Table XVIII: Cumulative abnormal returns around option listing date Table XIX: Cumulative abnormal returns for the 11-day window around options listings Table XX: Test for other short-sale constraints Table XXI: Change in short interest in ETFs around options listing Table XXII: Volume-return regressions Figure la: Time series plot o fppremium. Figure lb: Time series plot o f qpremium. vii R ep ro d u ced with p erm ission o f the copyright ow ner. Further reproduction prohibited w ithout perm ission. Figure 2: Histogram o f premiums for the four bond ETFs. Figure 3: Time-series plot o f bond ETF premiums. Figure 4: ETF versus equity DARs and CARs. V lll R ep ro d u ced with p erm ission o f th e copyright ow ner. Further reproduction prohibited w ithout perm ission. Essay One: Active vs. Passive ETFs R ep ro d u ced with p erm ission o f th e copyright ow ner. Further reproduction prohibited w ithout perm ission. I. Introduction An exchange-traded fund (ETF) comprises o f a basket o f either stocks or bonds. It has characteristics common to both closed- and open-end funds. ETFs trade on an intraday basis on major exchanges like closed-end funds. Like open-end funds, the number o f ETF shares outstanding can change on a daily basis depending on whether there is a net creation or redemption o f shares in the ETF. ETFs are either passively or actively managed. Passive ETFs track an equity or bond index, whereas active ETFs attempt to outperform an equity or bond index. A comparison o f active and passive ETFs provides valuable insights about pricing efficiency because o f institutional features which either facilitate (in the case of passive ETFs) or limit (in the case o f active ETFs) arbitrage. Theory and evidence from option pricing literature (e.g. Black and Scholes (1973), Galai (1977, 1978), Phillips and Smith (1980)) and interest rate parity literature (e.g. Frenkel and Levich (1975), Rhee and Chang (1992), Fletcher and Taylor (1996)) suggest that when arbitrage, using financial securities, is possible, we should not observe large deviations from theoretical values. Moreover, evidence from purchasing power parity literature (e.g. Frenkel (1981), Adler and Lehmann (1983), Taylor (1988)) and closed-end fund literature (e.g. Pontiff (1996)) show that when there are limits to arbitrage using financial securities, large deviations R ep ro d u ced with p erm ission o f th e copyright ow ner. Further reproduction prohibited w ithout perm ission. from theoretical values can persist.'’^ Prior literature does not directly compare the mispricing o f arbitrageable and non-arbitrageable securities. Passive and active ETFs provide such an opportunity leading to the following hypothesis. Hypothesis 1: Mispricing fo r active ETFs, which are not arbitrageable, should be greater than mispricing fo r passive ETFs, which are arbitrageable. This study also exploits structural differences in active and passive ETF markets to test theories of liquidity. Active ETFs have only one market maker whereas passive ETFs have multiple market makers. Theory predicts that liquidity improves as the number o f market makers increases (Glosten (1989), Grossman and Miller (1988)). Empirical tests o f the effect o f the number o f market makers on bid-ask spreads (one measure of liquidity) find that hid-ask spreads decrease as the number o f market-makers increase (e.g. Wahal (1995), Fluang and Masulis (1999)). This literature indicates that the bid-ask spread should be wider for active ETFs than for passive ETFs. Further, only the market maker knows the active ETF’s net asset value (NAV) at all times, whereas the passive ETF’s NAV is public knowledge. If the active ETF market maker attempts to profit from his information advantage, liquidity will be lower for active ETFs than for passive ETFs (see Bhattacharya and Spiegel (1991)). This also suggests that hid-ask spreads should be wider for active ETFs than for passive ETFs. Empirical studies that examine the effect o f information asymmetry on market liquidity show that liquidity is inversely related to information asymmetry (e.g. Lee, Mucklow, and Ready (1993), Koski ' See Pakko and Pollard (1996) for list o f reasons for violation o f purchasing power parity. ^ Possible explanations for the closed-end fund discount include overvaluation o f illiquid or restricted assets in N A V calculation (Malkiel (1977)), managerial ability reflected in the market price but not in the N A V (Boudreaux (1973), Chay and Trzcinka (1999)), and/or tax liabilities on unrealized capital gains not reflected in the N A V (Roenfeldt and Tuttle (1973)). Alternatively, behavioral explanations o f the closedend fund puzzle attribute the discount to risk due to the presence o f noise traders (e.g. Z w eig (1973), De Long, Shleifer, Summers, and Waldman (1990) and Lee, Shleifer, and Thaler (1991)). R ep ro d u ced with p erm ission o f th e copyright ow ner. Further reproduction prohibited w ithout perm ission. and Michaely (2000)). These studies show that uninformed market makers reduce liquidity when exposed to informed traders. Contrary to this approach, this study tests how liquidity is affected when the market maker is informed relative to the rest o f the market. The second hypothesis is as follows. Hypothesis 2: Bid-ask spreads are greater fo r active ETFs than fo r passive ETFs. The insight gained from this study has significant policy implications. The Securities and Exchange Commission (SEC) is considering proposals to introduce active ETFs in the US (see DJNS (2001), SEC (2001)). Though it has been more than two years since the first o f these proposals were filed with the SEC and more than a year since the SEC requested public comments on its concept release, the SEC is yet to permit active ETFs in the US. The delay is due to the SEC’s concern about an appropriate structure for the active ETFs. An appropriate structure for an active ETF should, among other things, ensure that its shares are priced efficiently and have a liquid market (SEC (2001)). This paper examines such a structure for active ETFs by studying those listed on the Deutsche Borse. The first ETF was introduced on the Toronto Stock Exchange (TSE) in 1990 under the name TIPS (Toronto Index Participation units). It tracks the 35 largest stocks on the TSE. TIPS were introduced to attract small investors who stayed away from the market since the 1987 crash (Globe and Mail (1989)). The SPDR (the first ETF in the U.S.) was introduced in 1993 to attract retail investors who wanted to invest in the S&P 500 Index (DJNS (1993)).^ Since then ETFs have grown dramatically both in number and size. A study by Merrill Lynch in 2002 finds that there are 246 ETFs listed worldwide SPDR stands for Standard and Poor’s Depositary Receipts and is pronounced spider. R ep ro d u ced with p erm ission o f t h e copyright ow ner. Further reproduction prohibited w ithout perm ission. with nearly $130 billion in assets."^ All ETFs issued worldwide prior to 2000 were passively managed. The Deutsche Bdrse introduced eleven active ETFs in November 2000.^ One of the key reasons for introducing active ETFs was to give investors a security that could outperform an index since passive ETFs could only match the performance o f an index. In the time since, others followed and currently there are 24 active ETFs listed on the Deutsche B o r s e . ’ Table I lists the main characteristics o f passive and active ETFs on the Deutsche Borse. One o fth e key differences between passive and active ETFs is that the portfolio composition o f passive ETFs is released every day prior to start o f trading, whereas the composition o f active ETFs is never known to the market. This helps arbitrageurs to profit from mispricing in passive ETF shares since they know what shares to trade but rules out the same for active ETF shares. The exchange disseminates an indicative net asset value (INAV) all through the trading day that represents the true value o f a share in the passive ETF. The market makers post bid and offer prices based on the supply and demand for the passive ETF shares and the NAV. If the quoted prices deviate from the NAV o f the passive ETF, arbitrageurs can take advantage of the deviation using the creation/redemption process. The creation/redemption process allows traders to create shares in the passive ETF by putting together a portfolio consisting o f the individual securities of the passive ETF and exchanging this portfolio for shares in the passive ETF. The trust that manages the passive ETF issues the new shares to the traders. Similarly, The Merrill Lynch report quoted figures as o f the end o f May 2002, at which time the US had 102 ETFs with over $90 billion in assets. Investment Company Institute (2003a) reports that there are 118 ETFs in the US as o f April 2003 with over $110 billion in assets. ^ On the Deutsche BOrse, the passive ETFs are referred to as index funds and the active ETFs are called actively managed funds. ®N o active ETF has been closed down over my sample period. ’ Currently, the Australian Stock Exchange (A SX ) is the only other exchange to have active ETFs listed. On the A SX , active ETFs are called hybrid ETFs. Nine hybrid ETFs are currently listed on the A SX . R ep ro d u ced with p erm ission o f t h e copyright ow ner. Further reproduction prohibited w ithout perm ission. when traders want to redeem shares in the passive ETF, they surrender ETF shares to the trust and receive a portfolio consisting o f the underlying securities in return.* The portfolio composition o f active ETFs is not public knowledge and they do not have the creation/redemption feature.^ However, the fund manager makes public the top ten holdings o f the active ETF at the end of each month. She decides on which stocks to select and how to weight them. This decision is based on the fund’s investment objectives and the fund manager’s private information about the future performance o f securities. For investors to be involved in a creation/redemption process for active ETFs, they will have to know the active ETF’s portfolio composition. The fund manager will not share her portfolio composition with other investors since they may attempt to free ride on her information. Consequently, arbitrageurs cannot take advantage o f arbitrage opportunities and bring prices back to p a r i t y . " The onus o f maintaining parity between the market Under the creation/redemption feature, arbitrageurs face negligible holding costs since their positions may last a few seconds or minutes. If transaction costs to replicate the entire portfolio o f a passive ETF are high, it is possible that arbitrageurs trade a representative sample o f securities in the index, in which case creation or redemption is not possible. In this case, arbitrageurs may face significant holding costs and mispricing may not disappear completely. ®The fund manager o f the active ETF reveals the portfolio composition to the market maker and the exchange daily. Additionally, she informs the other two parties if she makes any change to the portfolio during the trading day. In addition to being able to trade active ETF shares on the exchange, investors can also buy or sell shares directly from the fund manager at the closing N A V at the end o f the trading day (similar to open-end funds). This suggests that one possible way to take advantage o f mispricing in active ETF shares is to buy (sell) shares on the exchange and sell (buy) shares directly to (from) the fund manager if the market price is at a discount (premium) relative to the N A V (shares can be purchased on the exchange and redeemed from the fund manager and vice versa). In this scenario, investors do not have to know the true com position o f the active ETFs portfolio. To profit from this strategy, investors must be able to buy and sell simultaneously in the two markets, which is not possible due to restrictions. The two markets are the primary and secondary markets. The primary market is the one in which investors trade directly with the fund manager. The secondary market is the exchange where trades take place between investors. To trade with the manager, investors are required to place their orders prior to the opening o f markets. This suggests that investors w ill place orders based on the previous day’s closing market price and closing N A V . But their orders w ill be executed by the fund manager at the follow ing day’s closing N A V , by which tim e the market price would have also changed. Hence investors who attempt to profit from possible arbitrage opportunities w ill base their strategies on stale prices and may end up making loses instead o f profits. '' Passive ETFs are arbitrageable in the pure riskless sense whereas active ETFs are not. Since a part o f the com position o f the active ETF is made public every month, som e investors may attempt to profit from mispricing using their partial knowledge o f the composition o f the active ETF. Other investors may try to R ep ro d u ced with p erm ission o f t h e copyright ow ner. Further reproduction prohibited w ithout perm ission. price and the NAV falls on the market maker. The market maker posts bid and offer prices based on the supply and demand for the active ETF shares and his private information about the active BTF’s portfolio composition. The Deutsche Borse requires all ETFs to declare a benchmark index. In the case of passive ETFs, each one explicitly tracks its benchmark index. Investors expect the passive ETF’s performance to match its benchmark’s performance. On the other hand, the exchange requires the active ETF’s fund manager to select a benchmark index, subject to its approval, such that investors can compare the active ETF’s composition and performance to its benchmark. This does not imply that the fund must only hold securities that are a subset of the securities in the index. For example, the fund may hold securities that are in the same sector or geographic region as some securities in the benchmark but not necessarily the same seeurities. The Deutsche Dorse’s stated purpose for requiring each active ETF to select a benchmark index is to have a yardstick against which to compare the active ETF’s performance. Since there are only a limited number of indices available, it is possible that more than one ETF has the same benchmark index. Prior studies have investigated the pricing and liquidity o f only passive ETFs. Ackert and Tian (2000) find that the degree o f mispricing on the SPDR, which has low arbitrage costs, is insignificant suggesting that speculators successfully arbitrage any discrepancies. Contrary to their finding for the SPDR, they find that the level o f infer the com position through a style analysis by regressing past active ETF returns on various benchmarks (see Sharpe (1992)). These attempts by investors may still fall short o f being able to replicate the active ETF exactly. A t best, these w ill be speculative arbitrages. The fund manager o f the active ETF reveals the portfolio composition to the market maker and the exchange daily. Additionally, she informs the other two parties if she makes any change to the portfolio during the trading day. The market maker and/or the fund manager could potentially profit from deviations between N A V and market price o f active ETF shares i.e. they could act as arbitrageurs since they know the portfolio com position o f the fund. Rules and regulations on the Deutsche Borse preclude the market maker and the fund manager from acting as arbitrageurs. The rules and regulations also prohibit the market maker from disclosing the portfolio com position to third parties. R ep ro d u ced with p erm ission o f t h e copyright ow ner. Further reproduction prohibited w ithout perm ission. mispricing for the Midcap SPDR (ETF in the US that tracks the S&P Midcap 400 Index), which has higher arbitrage costs, is significant indicating that pricing errors are insufficient to cover trading costs. Engle and Sarkar (2002) study a large sample o f US passive ETFs (passive ETFs that track domestic US indices) and find that on average they are priced efficiently, the only exception being short-lived and minor deviations from the NAV during the trading day. Contrary to their finding for US passive ETFs, they find large and persistent deviations for a sample of international ETFs (passive ETFs that track foreign indices) listed in the US. Engle and Sarkar attribute these differences to the larger transaction costs associated with trading shares listed in other countries. Hegde and McDermott (2003) examine the liquidity o f Diamonds and Qubes,'^ which are both passive ETFs, and their underlying stocks. Consistent with Subrahmanyam (1991), they find that adverse selection is lower for the ETF shares than for the underlying individual stocks. They also find that the introduction o f ETFs improves the liquidity o f the underlying stocks and increases the volume o f trading and open interest in the related futures markets. This is the first study that examines the pricing efficiency and liquidity o f both active and passive ETFs. In particular it compares the pricing efficiency and liquidity of active ETFs to that o f passive ETFs listed on the Deutsche Borse. The sample includes ETFs that trade on the Xetra system of the Deutsche Borse. Xetra is a hybrid trading system where liquidity is provided by market maker(s) and limit orders. ETF market makers are obliged to maintain fair prices and liquid markets. Additionally, market makers are required to post quotes for at least 90 percent o f the effective trading time.''^ Diamonds track the D ow Jones Industrial Average and Qubes track the Nasdaq 100. The effective trading time is the time o f continuous trading only and excludes auctions. R ep ro d u ced with p erm ission o f t h e copyright ow ner. Further reproduction prohibited w ithout perm ission. They also face restrictions on the maximum quoted spread and minimum quoted size. The limit order book is open but trader identity remains anonymous. The Xetra exists side-by-side with a trading floor where trades execute using a fixed number o f intraday call auctions. Further, trading on Xetra begins and ends with a call auction. There may be additional auctions during the day on Xetra during which the exchange halts continuous trading.'^ By comparing the pricing efficiency and liquidity o f active ETFs to those of passive ETFs listed on the same exchange, I am able to control for the effects o f market structure (other than the number o f market makers) and other market-wide factors on pricing and liquidity. Consequently, resulting differences in pricing efficiency and liquidity are attributable to differences in the structures o f active and passive ETFs. This study’s results permit conclusions with direct policy implications regarding the merits of different ETF structures. Consistent with my hypothesis, I find that absolute deviations in market price from NAV are larger for active ETFs than for passive ETFs. However, I find that the mean deviations are positive and significant for passive ETFs but insignificant for active ETFs. I also find persistence in these deviations for passive ETFs but no persistence for active ETFs. Contrary to expectations, I find that the bid-ask spreads for active ETFs are smaller than those for passive ETFs. This result persists even after controlling for differing characteristics o f active and passive ETFs. This suggests that a single informed market maker is better able to maintain liquid markets than multiple uninformed market makers. For comparisons o f the trading floor and the computerized trading systems o f the Deutsche Borse, see Theissen (2002). Hau (2001) has a brief description o f the Xetra. For a detailed description o f the Xetra, see DB (2002). R ep ro d u ced with p erm ission o f t h e copyright ow ner. Further reproduction prohibited w ithout perm ission. The remainder o f the paper proceeds as follows. In the next section, I describe my data and methodology. In Section III, I present empirical results and discussions. Conclusions and implications follow in Section IV. II. Data and Methodology Data is collected from Bloomberg. It consists o f daily closing market prices, daily closing NAVs, daily closing bid and offer prices, and daily volumes for all active and passive ETFs starting from the listing date on Deutsche Borse to the end o f August 2002. Additionally, data on the number o f market makers for each ETF is collected from the Deutsche Bdrse Web site. The final sample consists o f 57 passive ETFs and 15 active ETFs. The oldest passive ETF started in April 2000 and latest passive ETF in July 2002. Similarly, the oldest active ETF in the sample started in November 2000 and the latest in May 2002. The Appendix lists all the ETFs used in this study along with their listing date. I define two measures o f premium for ETF i at close o f trading on day t as follows: ppremium^ = In and (1) qpremium^i = In (2) where In is the natural logarithm, Pu is the closing per-share market price o f ETF i on day t, Qit is the midpoint o f the closing bid and offer prices o f ETF i on day t, and NA Vu is the closing per-share net asset value o f ETF i on day t. ppreminmu is a market price-based measure o f ETF premium, whereas qpremiurriit is a quote-based measure o f ETF 10 R ep ro d u ced with p erm ission o f t h e copyright ow ner. Further reproduction prohibited w ithout perm ission. premium. The logarithmic premium measure is commonly used in closed-end fund and ETF literature to measure mispricing (e.g. Chay and Trzcinka (1999), Engle and Sarkar (2002)).'^ For both measures, a negative premium indicates a discount. I use the percentage quoted and effective spreads as measures o f liquidity, which are defined for ETF i on day t as follows: p g s p r e a d „ ^ ^ S 3 id ^ _ (3) Qii pespread 2x\ Pn - Q u \ (4) Q il where offern is the closing offer price for ETF i on day t, bidu is the closing bid price for ETF i on day t and Pu and Qu are the same as defined earlier. These measures are commonly used in microstructure literature to measure liquidity (e.g. Stoll (2000), Hegde and McDermott (2003)). A wider spread implies lower liquidity since they measure investors’ transaction costs on a round-trip transaction. III. Results and Discussion Table II presents summary statistics on the ETFs. Table II shows that relative to passive ETFs, active ETFs are larger and have share prices that are twice as large. The higher price o f active ETFs suggests that using a percentage spread is better than using a euro spread because it adjusts for price clustering effects (see Harris (1991)). One explanation for the larger size of active ETFs is the minimum size requirement o f €50 million for active ETFs at start-up but no such requirement for passive ETFs (see Table If premium is measured as the difference between price and N A V expressed as a percentage o f N A V , this results in a skewed distribution for the premium measure, where premium takes values between -1 and infinity. The logarithmic measure has a symmetric distribution that takes values between negative and positive infinity. This is more conducive to using standard statistical tests. 11 R ep ro d u ced with p erm ission o f t h e copyright ow ner. Further reproduction prohibited w ithout perm ission. I). Expense ratios are higher for active ETFs than for passive ETFs. A reason for the higher expense ratio o f active ETFs is compensation for the manager’s expertise in picking securities. Expense ratios are lower for passive ETFs since manager’s expertise is not required to pick stocks because passive ETFs track indices. Table III presents summary statistics on ETF premiums. In Panel A, each premium measure is averaged across all days for each passive ETF (active ETF) and the summary statistics o f the resulting series are shown. Absolute deviations from NAV are larger for active ETFs than for passive ETFs. There is also evidence o f passive ETFs trading at significant premiums relative to their NAVs, whereas active ETFs do not trade at a premium. However, there is little evidence o f any difference between premiums for passive and active ETFs. Evidence from option pricing, interest rate parity, and purchasing power parity literature suggests that the magnitude o f mispricing should be larger for non-arbitrageable securities than for arbitrageable securities but does not predict a direction for the mispricing (discount or premium). Examining the volatility o f the premium measures will provide insights about the magnitude o f active and passive ETFs’ mispricing. A comparison o f the volatility o f ppremium and qpremium for passive ETFs and active ETFs is presented in Table III. In the cross-section (Panel A o f Table III), the standard deviations for ppremium {qpremium) are 88 (28) basis points for passive ETFs and 54 (59) basis points for active ETFs.'^ The measures are significantly different for passive ETFs and active ETFs, though ppremium suggests that volatility is higher for passive ETFs whereas qpremium suggests that volatility is higher for active ETFs. ppremium may suffer from stale closing prices for less liquid ETFs and hence These measures are higher than the 15 basis points that Engle and Sarkar (2002) report for passive ETFs in the US. 12 R ep ro d u ced with p erm ission o f t h e copyright ow ner. Further reproduction prohibited w ithout perm ission. interpretations based on qpremium are more appropriate. There is less variation in the premiums o f passive ETFs than o f active ETFs. This result is consistent with the hypothesis that non-arbitrageable securities have larger mispricing than arbitrageable securities. All ETFs do not have the same length o f time-series. As a result some ETFs with shorter time-series o f data receive larger weights in Panel A. If mispricing is related to the age o f the ETF, it will introduce a bias in the results. To check for such biases, I determine time-series statistics o f the premium measures. Equally-weighted portfolios of passive ETFs (active ETFs) are formed on every calendar date and the mean premium measures for these portfolios are calculated for each day. The daily mean premiums are plotted against calendar time in Figure 1. Figure la is for ppremium and Figure lb is for qpremium. There are no discernable patterns in the deviations o f market price from NAV for both active and passive ETFs. ppremium appears to be similar for both active and passive ETFs. qpremium appears to be more volatile for active ETFs than for passive ETFs. This is consistent with the hypothesis that non-arbitrageable securities have larger deviations from NAY. The summary statistics o f the daily mean premiums o f the equallyweighted portfolios are presented in Panel B o f Table III. The time-series data show results consistent with those from Panel A. This indicates that absolute deviations from NAV are smaller for passive ETFs than for active ETFs. This is consistent with the hypothesis that arbitrageable securities have lower mispricing than non-arbitrageable securities. The volatility o f premium measures does not show if either type o f ETF trades more often at a premium or discount. If an ETF trades consistently at a premium or 13 R ep ro d u ced with p erm ission o f t h e copyright ow ner. Further reproduction prohibited w ithout perm ission. discount, the mean and median should be significantly different from zero. Evidence from option pricing and interest rate parity literature suggests that passive ETFs should trade at insignificant premiums. Results from Brickley and Schallheim (1985) suggest that active ETFs should also trade at insignificant premiums. They find that the closedend fund discount disappears between when a closed-end fund announces that it will reorganize as an open-ended fund and the actual reorganization.'* They also find a large reduction in the discount around the announcement o f the reorganization. They attribute part of the reduction in discount over time to the resolution o f the uncertainty about fund reorganization. This evidence suggests that investors’ ability to trade with the fund manager at the NAV eliminates the closed-end fund discount. Though investors can trade active ETFs intraday, investors can also trade directly with the fund manager at the NAV at the close o f trading. Active ETFs are, in effect, open-ended funds that trade intraday. The ability o f investors to trade active ETF shares with the fund manager at the NAV suggests that active ETF shares will not trade at significant deviations from NAV. In the cross-section (Panel A o f Table III), the mean ppremium and qpremium for passive ETFs are 32 and 9 basis points, respectively, and 14 and 22 basis points, respectively, for active ETFs. Both the premium measures for passive ETFs are significantly different from zero, though the mean qpremium is not economically significant. The premium measures for active ETFs are not significant. The premium measures for passive and active ETFs are not significantly different from each other. Further, the median ppremium and qpremium for passive ETFs are 22 and 5 basis points, respectively, and -10 and -6 basis points, respectively, for active ETFs. A nonparametric After reorganization, in most cases, the closed-end fund ceases to trade intraday. Investors will have to trade directly with the fund manager once a day at the close o f trading and all trades are executed at the NAV. 14 R ep ro d u ced with p erm ission o f t h e copyright ow ner. Further reproduction prohibited w ithout perm ission. test o f medians suggests that both premium measures are significantly different from zero for passive ETFs, whereas they are not significant for active ETFs. The median qpremium for passive ETFs is not economically significant. The median ppremium of passive ETFs is statistically different from that o f active ETFs whereas the median qpremiums are not different from each other. Contrary to expectations, there is evidence that premiums are more common than discounts for passive ETFs. On the other hand, as expected, premiums and discounts occur randomly for active ETFs. The results from the time-series data in Panel B o f Table III are similar. The average mispricing o f active ETFs is not different from zero implying that the active ETF structure on the Deutsche Borse maintains efficient pricing o f active ETF shares. The liquidity o f stocks in the benchmark indices may affect the pricing o f ETFs. If an ETF holds a set o f relatively illiquid securities, then the prices used to calculate its NAV may be stale resulting in evidence o f large mispricing. To control for this illiquidity effect, I select active and passive ETFs that have the same benchmark index. This results in a sample of 9 active and 12 passive ETFs.’^ The summary statistics o f the premium measures o f this sample are in Table IV. Panel A contains the cross-sectional statistics and Panel B contains the time-series statistics. Consistent with results from Table III, I find that the volatility o f premium measures is larger for active than for passive ETFs. This implies that deviations from NAV are larger for active ETFs than for passive ETFs even after controlling for some liquidity effects. The mean and median measures for The number o f active and passive ETFs are not the same since in some cases there is more than one passive ETF for a given benchmark index. For five active ETFs, I find passive ETFs with exactly the same benchmark index. For three other active ETFs, the match is not exact. I find passive ETFs w hose benchmarks are in the same sector or industry as the active ETF’s benchmark. For the last active ETF in the subsample (benchmark is the D ow Jones STOXX 600), there is no passive ETF that has a similar benchmark. But there are 18 passive ETFs that track the different sectors o f the D ow Jones STOXX 600. 1 determine the equally-weighted premium measures on a daily basis for these 18 passive ETFs and use this measure as a match for the corresponding active ETF. 15 R ep ro d u ced with p erm ission o f t h e copyright ow ner. Further reproduction prohibited w ithout perm ission. passive ETFs are statistically significant but not economically significant whereas they are insignificant for active ETFs. The premium measures for passive ETFs are significantly different from those of active ETFs. This is also consistent with findings from Table III suggesting that the positive mean premiums for passive ETFs are not due to illiquidity effects. The different trading hours o f the underlying securities and the ETFs may he another source of stale prices. The ETFs on the Deutsche Borse track indices that hold securities from all over the world. The ETFs may use stale prices o f securities that trade on markets which are closed when the NAV is calculated. This could lead to larger deviations in market price from NAV. To control for this effect, I divide my sample o f active and passive ETFs into three suhsamples: ETFs based on European indices, ETFs based on US indices, and ETFs based on global indices.^^ 1 then compare the mispricing of active and passive ETFs within each o f the three subsamples. Results o f this analysis are in Table V. For all three subsamples, the results are similar to those in Table III. Deviations in market price from NAV are larger for active than for passive ETFs. Passive ETFs trade more often at a premium than at a discount. These results hold even after controlling for different sources o f stale prices. The results in Tables III, IV, and V are based on univariate tests. The results may differ in a multivariate setting. To this end, I run the following linear regression: PM„ = CTo + a^crl + where InF;., + ajG,,. + , (5) is the absolute value o f the premium measure ppremium or qpremium for ETF i on day t, c>l is the square o f NAV returns for ETF i on day t (a measure of There is only one ETF based on an Asian index and so I drop this ETF from the analysis. 16 R ep ro d u ced with p erm ission o f t h e copyright ow ner. Further reproduction prohibited w ithout perm ission. volatility), In Vu is the natural logarithm o f the volume for ETF i on day t, Gu is a dummy variable that takes value 1 if ETF /’s benchmark index is an index o f US securities and 0 otherwise, is a dummy variable that takes value 1 if ETF /’s benchmark index is an index o f global securities and 0 otherwise, Gy, is a dummy variable that takes value 1 if ETF z’s benchmark index is an index o f Asian securities and 0 otherwise, and Z), is a dummy variable that takes value 1 if ETF i is an active ETF and 0 if it is a passive ETF.^' Days o f high volatility correspond to when there is a higher level o f information flow into the market. Information may be incorporated faster into the ETFs than the individual stocks, which results in larger deviations of market price from NAV.^^ As a result, the coefficient o f volatility should be p o s i t i v e . Vo l u me affects the premium negatively. When volume is high, liquidity costs are small. In this scenario, there is better synchronization between ETFs and the individual stocks leading to smaller deviations between market price and NAV.^'^ The three geographic region dummies should be positive because their trading hours are different from the Deutsche Borse’s trading hours. ETFs based on non-European indices will suffer from stale prices in NAY calculations leading to larger deviations between market prices and NAY. Finally, the dummy should be positive because active ETFs are non-arbitrageable whereas passive ETFs are arbitrageable. Since som e ETFs are not traded everyday, closing prices are m issing on som e days. To avoid losing data for these days, I calculate return variance as the square o f the daily N A Y return rather than returns based on closing prices. Hasbrouck (2003) shows that ETFs lead index futures and other research (e.g. Chan (1992)) show that index futures lead the underlying individual stocks in price discovery. This suggests that ETFs lead individual stocks in price discovery. For example, see Flarris (1989). He finds large absolute futures-cash basis during the 1987 stock market crash, which he attributes to nonsynchronous trading problems between the futures and the stock markets. See Long and Officer (1997). 17 R ep ro d u ced with p erm ission o f th e copyright ow ner. Further reproduction prohibited w ithout perm ission. Results for regression (5) are in Table VI. The first regression has pprem ium as the dependent variable and the second has qpremium as the dependent variable. The main result is that the coefficient o f the dummy is positive and significant as expected. This shows that active ETF prices have larger deviations from NAY than passive ETF prices. The coefficient o f volatility is positive and significant as expected. Consistent with prior research, this shows that mispricing increases with volatility. The ETFs lead the underlying securities in price discovery. On the other hand, the coefficient o f volume is positive but significant in only one regression. This suggests that mispricing may increase with volume. An explanation for this could be the high inflow o f information on high volume days (see, for example, Karpoff (1987)). This results In ETFs leading individual securities in price discovery resulting in large deviations. Two o f the geographic region dummies are positive and significant as expected whereas one is negative and significant. The dummies for ETFs based on US and global indices are positive suggesting that mispricing is larger for these ETFs than for those based on European indices. This is consistent with the fact that mispricing is larger when trading hours o f the ETF and individual security markets are not synchronized. On the other hand, the result for the dummy for ETFs based on Asian indices is contrary to expectations. These ETFs have smaller deviations between market price and NAY than ETFs based on European indices. The ETF dummy is positive and significant as expected suggesting that active ETFs have larger deviations between market price and NAYs. Consistent with my hypothesis, this result is due to the arbitrageability of passive ETFs and non-arbitrageability of active ETFs. 18 R ep ro d u ced with p erm ission o f th e copyright ow ner. Further reproduction prohibited w ithout perm ission. The passive ETFs’ positive mean for the premium measures in Table III suggests that premiums occur more often than discounts. As a further test, I conduct a proportions test on the time-series data o f the whole sample. I classify positive premiums as ones and non-positive premiums are zeroes. If the premiums are randomly positive and negative, the proportion o f ones (or zeroes) should be 50 percent. The results are presented in Table VII. Ones occur 67 percent of the time for passive ETFs and 50 percent o f the time for active ETFs. A test of significance shows that the proportion o f ones for passive ETFs is significantly different from 50 percent whereas that for active ETFs is not different from 50 percent. This suggests that premiums occur more often than discounts for passive ETFs whereas premiums and discounts are equally likely for active ETFs. These results are consistent with those in Table III. Evidence o f positive premiums for passive ETFs does not show if the premiums persist. Option pricing and interest parity literature suggest that the premiums should not persist. On the other hand, evidence from purchasing power parity and closed-end fund literature suggest that active ETF premiums should persist. I use a runs test to check for persistence of the premiums. A run is a continuous sequence o f zeroes or ones (zeroes and ones as defined earlier) o f length at least one. For example, the sequence 0100111 consists o f 4 runs - 1 zero, 1 one, 2 zeroes, and 3 ones. I determine the number o f runs in the sample. For a given sample o f zeroes and ones, the expected number o f runs in the sample is given by ( 6) = and the variance o f the number of runs is given by 19 R ep ro d u ced with p erm ission o f th e copyright ow ner. Further reproduction prohibited w ithout perm ission. (7) where R is the observed number o f runs from the sample, T is the total number o f observations in the sample, To is the number o f zeroes in the sample, and T/ is the number o f ones in the sample. The test statistic is given by - 4 # ^ . ^V ar(R ) (8) which is distributed asymptotically as standard normal. The results for the runs test are in Table VII. The number o f runs for ppremium and qpremium for passive ETFs are 246 and 228, respectively, both o f which are significantly different from their respective expected number o f runs. The premium measures for active ETFs have 213 and 222 runs, respectively, both o f which are not significantly different from their respective expected number o f runs. This suggests that deviations from NAY for passive ETFs persist over time whereas deviations for active ETFs do not persist. Even though passive ETFs have the creation/redemption feature that enables arbitrageurs to profit from deviations in market price from NAY, transaction costs limit the extent to which arbitrageurs find the mispricing profitable. On the other hand, active ETFs are priced efficiently by the market maker who knows the NAY at all times. The active ETF structure on the Deutsche Bdrse ensures efficient pricing o f its shares. The SEC should consider a similar structure for active ETFs in the US. I compare the liquidity o f the two types o f ETFs next. Table YIII presents summary statistics on the two liquidity measures. Univariate tests show that liquidity is higher for active ETFs than for passive ETFs. The mean (median)pqspread is 1.83 (1.86 ) percent for passive ETFs and 0.80 (0.77) percent for active ETFs. A test o f the hypothesis 20 R ep ro d u ced with p erm ission o f th e copyright ow ner. Further reproduction prohibited w ithout perm ission. that the bid-ask spread is wider for active ETFs than for passive ETFs is rejected for both measures. The results are similar for the mean and median pespread. This suggests that despite active ETFs having only one market maker, who has an information advantage over the rest o f the market, as opposed to passive ETFs that have multiple market makers with no information advantage, active ETFs have more liquid markets than passive ETFs. To disentangle the effects o f other variables on the liquidity measures, I run a regression similar to Stoll (2000). The regression is follows: LMj, = « ()+ « , In + Ujcrl + 1.5 Lehman 1-3 year Treasury bond ETF (SH Y ) Frequency Percentage o f observations 0 0 0 0 0 0 0 0 4.08 13 17 5.33 86.83 277 12 3.76 0 0 0 0 0 0 0 0 0 0 0 0 Lehman 7-10 year Treasury bond ETF (ILF) Frequency Percentage o f observations 0 0 1 0.31 4 1.25 15 4.69 64 20.00 15 4.69 151 47.19 60 18.75 7 2.19 2 0.63 1 0.31 0 0 0 0 0 0 C/) C/) 82 Lehman 20+ year Treasury bond ETF (TLT) Frequency Percentage o f observations 1 0.31 6 1.88 9 2 .82 31 9.72 69 21.63 9 2 .82 92 28.84 59 18.50 34 10.66 4 1.25 2 0.63 0 .94 3 0 0 0 0 Goldman Sachs $ InvesTop corporate bond ETF (LQD) Frequency Percentage o f observations 0 0 1 0.31 2 0.63 6 1.88 11 3.44 4 1.25 23 7.19 40 12.50 43 13.44 33 10.31 31 9.69 108 33.75 17 5.31 1 0.31 CD ■D O Q . C o CD Q . ■CDD C/) (/) Table XIV (continued) Panel B. Descriptive statistics fo r premium o o ■D cq ' O’ Q CD ■D O Q . C a Mean (%) Median (%) Standard Deviation (%) Minimum (%) Maximum (%) N Lehman 1-3 year Treasury bond ETF (SHY) 0.0467’' 0.0486’' 0.0316 -0.0487 0.1349 318 Lehman 7-10 year Treasury bond ETF (IFF) 0.0399’' 0.0468’ 0.0928 -0.3357 0.4229 319 Significantly different from zero at the 1 percent level. o ■o o CD Q . ■CDD (/) (/) 83 Lehman 20+ year Treasury bond ETF (TLT) 0.0403’ 0.0341’ 0.1521 -0.4677 0.6477 318 Goldman Sachs $ InvesTop corporate bom d ETF (LQD) 0.4284’ 0.3902’ 0.3331 -0.3876 1.5620 319 CD ■D O Q . C o C D Q . ■C D D C (/)/) o o ■D o 3 Table XV Linear regression of premiums on lagged premiums This table shows the estimates o f regressing daily premiums on lagged premiums, premium is the natural logarithm o f the daily closing price over the daily closing net asset value (NAV). A negative premium denotes a discount. N is the number o f observations used in each regression, p-value appears in parentheses below the coefficient estimates. Dependent variable: premium CD "n 3? CD ■D O Q. C a o 3 ■D O Intercept premiumt-i Adj. Lehman 1-3 year Treasury bond ETF (SHY) 0.0398 (0 .0001 ) 0.1569 (0.0047) 0.0252 0.0221 N 315 Lehman 7-10 year Treasury bond ETF (lEF) 0.03803 ( 0 .0 0 0 1 ) 0.06008 (0.2818) 0.0037 0.0035 317 C D Q . ■C D D (/) (/) 84 Lehman 20+ year Treasury bond ETF (TLT) 0.0401 (0 .0 0 0 1 ) -0.0049 (0.9307) 0.0000 -0.0032 315 Goldman Sachs $ InvesTop corporate bond ETF (LQD) 0.1089 (0 .0001 ) 0.7423 (0 .0001 ) 0.5551 0.5537 317 Table XVI Linear regression of premium measures on characteristics of ETFs The dependent variable is the absolute value o f the premium, which is the natural logarithm o f the ratio o f the daily closing price to the daily closing net asset value (NA V ). is the difference between the intraday high and low prices divided by the daily closing price, In V is the natural logarithm o f the daily volume, DI is a dummy variable that takes value 1 for lEF and zero otherwise, D2 is a dummy variable that takes value 1 for TLT and zero otherwise, and D3 is a dummy variable that takes value 1 for LQD and zero otherwise. N is the number o f observations used in the regression. Dependent variable: premium Intercept ....................... in V Dl D2 D2 N R" Adj. Estimate 0.2256 7.0512 -0.0173 -0.0165 0.0028 0.4349 p-value 0.0419 0.0565 0.0764 0.5372 0.9336 0.0001 353 0.6110 0.0654 85 R ep ro d u ced with p erm ission o f th e copyright ow ner. Further reproduction prohibited w ithout perm ission. Table XVII New listings by options exchange and year This table shows the distribution o f new ETF option listings across the different options exchanges in the U.S. and across time. Though an ETF option may be listed on more than one exchange, a listing is classified as a new listing only if it is being listed for the first time. Exchanges on which ETF options are listed are the Chicago Board Options Exchange (CBOE), American Stock Exchange (AM EX), International Securities Exchange (ISE), Pacific Exchange (PacificEx), and Philadelphia Stock Exchange (PHLX). Year 1998 1999 CBOE 2000 2001 2002 1 2003 - 15 5 AMEX ISE PacificEx PHLX 1 - - - - - - - 5 - - - - - 1 8 7 - - - 1 - - - 86 R ep ro d u ced with p erm ission o f th e copyright ow ner. Further reproduction prohibited w ithout perm ission. CD ■D O Q . C o C D Q . ■C D D C (/)/) O O ■D cq' c O’ Q CD ■D O Q. C a o ■o o O’ CJ Q Table XVIII Cumulative abnormal returns around option listing date This table shows the underlying ETF’s cumulative abnormal returns (CARs) that are estimated using the Brown and Warner (1985) m ethodology, with an estimation period over [L - 80, L - 6] (L is the day o f option listing). Returns are computed over the estimation period using the market m odel and the constant mean returns model. The Wilshire 5000 index returns are used as the index for the market model. The constant mean returns m odel assumes that the ETF has a constant daily return. The left h alf o f the table shows the cross-sectional mean and median CARs, assuming that each event is independent o f the other events. The right half o f the table groups events into portfolios and treats them as single securities. Tw o criteria are used to form these portfolios: day-of-listing, w hich groups securities with identical listing dates, and window-overlap, which groups securities with overlapping 11-day event window s. The significance o f the mean CARs are tested using the usual t-statistic. The significance o f the median CARs are tested using the W ilcoxon signed-rank statistic. The number o f observations (N ) corresponding to each method and estimation model is shown after the mean and median measures. Market model Constant mean returns model Individual listings Event window = 11 days Event window = 3 days Median Mean Median Mean (% ) N N (% ) (% ) (% ) - 0.22 43 0.65 0.55 43 -0.16 0.58 0.67 43 0.13 -0.13 43 T3 CD (/) (/) 87 Portfolio grouping (event window =11 days) Window-overlap groups Day-of-listing groups Median Mean Median Mean N N (% ) (% ) (% ) (% ) 0.64 16 22 0.39 0.28 0.81 0.04 0.45 22 -0.40 -0.14 16 CD ■D O Q . C o CD Q . ■CDD C/) Table XIX Cumulative abnormal returns for the 11-day window around options listings (/) O O ■D cq ' This table shows the daily abnormal returns (D A R s) and the cumulative abnormal returns (CARs) over an 11-day event window [L - 5, L + 5] (L is the day o f option listing). The market model with the W ilshire 5000 index is estimated over [L - 80, L - 6]. Abnormal returns are cumulated over the event w indow starting from L - 5. Mean and median measures under Individual listings are computed cross-sectionally, assuming each event is independent o f the others. In the other two categories, securities are grouped together into portfolios and these portfolios are treated as single securities. Tw o criteria are used to form these portfolios: day-of-listing, which groups securities with identical listing dates, and window-overlap, which groups securities w ith overlapping 11-day event w indow s. The significance o f the mean CARs are tested using the usual t-statistic. The significance o f the median CARs are tested using the W ilcoxon signed-rank statistic. The number o f observations (N) corresponding to each method is shown at the top o f each column. O’ Q CD ■D O Q . C a Mean o ■o o CD Q . ■CDD Days around listing date L-5 L-4 L-3 L-2 L-1 L L+1 L+2 L+3 L+4 L+5 Individ ual listings (N = 43) OAR (%) CAR (%) 0.14 0.14 0.35 0.49 0.64’ 0.15 0.14 0.78’ -0.19 0.59 0.02 0.61 0.01 0.62 0.04 0.66 0.11 0.77 0.08 0.85 -0.20 0.65 Day-of-listing groups (N = 22) CAR (%) DAR (%) 0.38 0.38 0.69 0.31 0.79 0.10 1.05 0.26 0.84 - 0.21 0.82 - 0.02 0.76 -0.06 0.90 0.14 0.98 0.08 1.08 0.09 0.81 -0.27 Window-overlap groups (N =16) CAR (%) DAR (%) 0.56 0.56 0.95 0.39 0.96 0.01 0.54 1.50 1.20 -0.30 -0.13 1.08 0.71 -0.37 0.91 0.20 0.10 1.01 0.08 -0.45 1.09 0.64 -0.12 0.05 0.29 0.06 -0.17 0.15 -0.17 0.21 0.11 0.11 -0.29 -0.03 0.39 0.44 0.70 0.57 (/) (/) Median L-5 L-4 L-3 L-2 L-1 L -0.07 0.18’ 0.11 0.03 -0.32" 0.08 -0.07 0.15 0.40 0.30 -0.03 0.19 - 0.12 0.20 0.15 0.08 -0.38’ 0.16 88 0.13 0.12 CD ■D O Q . C o CD Q . ■CDD C/) Table XIX (continued) (/) O O ■D c q ' O’ Median Days around listing date L+1 L+2 L+3 L+4 L+5 Individ ual listings (N = 43) DAR (%) CAR (%) 0.34 0.05 0.22 -0.08" -0.01 0.39 0.48 -0.01 -0.02 0.55 Day-of-listing groups (N = 22) CAR (%) DAR (%) 0.44 0.07 0.33 -0.06 0.02 0.37 0.45 -0.05 0.28 -0.06 indicate statistical difference from zero at the 5 and 10 percent levels, respectively. CD CD ■D O Q . C a o ■o o CD Q . ■CDD (/) (/) 89 Window-overlap groups (N = 16) CAR (%) DAR (%) -0.10 0.45 0.49 0.07 0.69 0.08 0.04 0.45 -0.17 0.39 Table XX Test for other short-sale constraints This table shows tests for short-sale constraints other than the uptick rule by running a cross-sectional OLS regression. The dependent variable is the 11-day cumulative abnormal return (CAR) over the event window [L - 5, L + 5] (L is the day o f option listing) computed using the the Brown and Warner (1985) m ethodology. The first regression uses the market model to compute CARs. The market model uses the Wilshire 5000 index as the benchmark index. The second regression uses the constant mean returns model to compute CARs. This model assumes that the ETF has a constant daily return. Abnormal returns are cumulated starting L - 5 . The explanatory variables are as follow s. D, is a dummy variable that takes a value o f one if the index underlying the ETF is a narrow index and zero otherwise. ln(K,) is the natural logarithm o f the average daily volume o f the ETF over the estimation period [L - 80, L - 6]. Index classification as narrow or not is based on the American Stock Exchange’s classification o f an ETF as a broad-based or sector ETF. Sector ETFs track narrow indices whereas broad-based do not track narrow indices. R-squared (R^), adjusted R-squared (Adj. R^), and the number o f observations used in each regression (N ) are shown at the bottom o f the table. The p-values o f the coefficient estimates are in parentheses below the estimates. Dependent variable: CAR Market model Intercept 1.6283 (0.5324) -0.0659 (0.9225) -0.1573 (0.4890) 0.0121 -0.0373 43 Di ln(FO R" Adj. R" N Constant mean returns model 3.6606 (0.4547) -0.7137 (0.5749) -0.2881 (0.4991) 0.0177 -0.0314 43 90 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. Table XXI Change in short interest in ETFs around options listing This table presents descriptive statistics on the change in short interest in the ETFs around options listing. The change in short interest (ASI) is the measured as the difference between the ratio o f the short interest in month t + 1 to the shares outstanding in month t + 1 and the ratio o f the short interest in month t - 1 to the shares outstanding in month t - 1, where t is the month o f ETF option listing. N is the number o f observations used in computing the descriptive statistics. Significance o f mean is tested using the usual tstatistic. Significance o f median is tested using the W ilcoxon signed-rank test. A SI Mean (%) Median (%) -1.57 0.98 Standard deviation (%) 26.19 Minimum (%) Maximum (%) N -129.24 63.94 43 91 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. Table XXII Volume-return regressions This table presents results from the volume-returns regressions for ETFs. For each sample ETF, the daily volume is regressed on the daily return. The regressions for the pre-listing period are over [L - 80, L - 6] and for the post-listing period are over [L + 6, L + 80] (L is the day o f option listing). The number o f regressions in which the daily returns coefficient estimate is positively (negatively) significant is tabulated in the second and third columns. Significance is measured at the 5 percent level. The mean and median o f the t-statistic o f the daily returns coefficient estimate is presented in the last two columns. Significance o f the mean and median t-statistics are tested using the usual t-statistic. Number of regressions Pre-listing Post-listing 43 43 Number of negatively significant «i 1 4 Number of positively significant ai 1 1 Mean tstatistic Median tstatistic -0.0125 0.1208 0.0241 0.3300 92 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. CD ■D O Q . C o CD Q . ■CDD C/) W o' o 8 . 00 % Passive ETFs o Active ETFs 6 . 00 % o o ■D cq' 4.00% 2 . 00 % 0 . 00 % o CD ■D O Q . C a o o ■O S -2 .0 0 % -4.00% o - 6 . 00 % CD Q . - 8 . 00 % ■CDD - 10 . 00 % (/) (/) - 12 . 00 % 4/11/2000 7/20/2000 10/28/2000 2/5/2001 5/16/2001 8/24/2001 12/ 2/2001 3/12/2002 6 /20/2002 Date Figure la . Tim e series plot o f pprem ium . This figure presents a time-series o f plot o f the price-based premium measure ppremium for passive and active ETFs. On each calendar date, ppremium is averaged across all available passive (active) ETFs. This average is then plotted against calendar date. 93 CD ■D O Q . C o CD Q . ■CDD C/) (/) 15.00% Passive ETFs Active ETFs O O ■D cq 10 . 00 % ' 5.00% O’ Q CD ■D O Q . C a I ■g 0 .0 0 % o ■o o -5.00% CD Q . - ■CDD 10 . 0 0 % - (/) (/) -15.00% 4/11/2000 7/20/2000 10/28/2000 2/5/2001 5/16/2001 8/24/2001 12/2/2001 3/12/2002 6 / 20/2002 Date Figure lb . Tim e series plot o f qpremium. This figure presents a tim e-series o f plot o f the quote-based premium measure qpremium for passive and active ETFs. On each calendar date, qpremium is averaged across all available passive (active) ETFs. This average is then plotted against calendar date. 94 CD ■D O Q . C o CD Q . ■CDD C/) (/) 350 □ SHY H IE F TLT O O ■D cq 300 Q LQ D ' 250 O’ Q 200 c CD ■D O Q . C a [...]...I Introduction An exchange- traded fund (ETF) comprises o f a basket o f either stocks or bonds It has characteristics common to both closed- and open-end funds ETFs trade on an intraday basis on major exchanges like closed-end funds Like open-end funds, the number o f ETF shares outstanding can change on a daily basis depending on whether there is a net creation or redemption o f shares in the... execute using a fixed number o f intraday call auctions Further, trading on Xetra begins and ends with a call auction There may be additional auctions during the day on Xetra during which the exchange halts continuous trading.'^ By comparing the pricing efficiency and liquidity o f active ETFs to those of passive ETFs listed on the same exchange, I am able to control for the effects o f market structure (other... first regression in Panel A The only other exception is the coefficient o f volatility, which is now positive and significant as expected This suggests that the effective spread increases with volatility One explanation for some o f the coefficients in the above regressions having signs contrary to expectation is the correlation between some o f the explanatory variables A check o f the condition numbers... Jones Industrial Average and Qubes track the Nasdaq 100 The effective trading time is the time o f continuous trading only and excludes auctions R ep ro d u ced with p erm ission o f t h e copyright ow ner Further reproduction prohibited w ithout perm ission They also face restrictions on the maximum quoted spread and minimum quoted size The limit order book is open but trader identity remains anonymous... SEC first received applications for the introduction o f active ETFs in early 2001 (DJNS (2001)) Following this, the SEC issued a concept release seeking comment from investment professionals and the public (see SEC (2001)) This concept release seeks to answer questions about the structure o f active ETFs One of the main concerns is the importance o f the creation/redemption feature that all passive... market crash, which he attributes to nonsynchronous trading problems between the futures and the stock markets See Long and Officer (1997) 17 R ep ro d u ced with p erm ission o f th e copyright ow ner Further reproduction prohibited w ithout perm ission Results for regression (5) are in Table VI The first regression has pprem ium as the dependent variable and the second has qpremium as the dependent... these ETFs than for those based on European indices This is consistent with the fact that mispricing is larger when trading hours o f the ETF and individual security markets are not synchronized On the other hand, the result for the dummy for ETFs based on Asian indices is contrary to expectations These ETFs have smaller deviations between market price and NAY than ETFs based on European indices The ETF... Section III, I present empirical results and discussions Conclusions and implications follow in Section IV II Data and Methodology Data is collected from Bloomberg It consists o f daily closing market prices, daily closing NAVs, daily closing bid and offer prices, and daily volumes for all active and passive ETFs starting from the listing date on Deutsche Borse to the end o f August 2002 Additionally,... passive exchange traded funds Passive ETFs are arbitrageable securities whereas active ETFs are non-arbitrageable I find that absolute deviations from NAV are larger for active ETFs than for passive ETFs Further, I find that active ETFs, on average, trade at insignificant deviations from their NAVs On the other hand, there is evidence o f passive ETFs trading at statistically significant but economically... there is better synchronization between ETFs and the individual stocks leading to smaller deviations between market price and NAV.^'^ The three geographic region dummies should be positive because their trading hours are different from the Deutsche Borse’s trading hours ETFs based on non-European indices will suffer from stale prices in NAY calculations leading to larger deviations between market prices