Equity ETF Arbitrage and Daily Cash Flow Jon A. Fulkerson School of Business Administration University of Dayton 937-229-2404 jfulkerson1@udayton.edu Susan D. Jordan Gatton College of Business and Economics University of Kentucky 859-257-1626 sjordan@uky.edu Denver H. Travis* Haub School of Business Saint Joseph s University 610-660-1670 denver.travis@sju.edu January 8, 2018 * Contact author.
ABSTRACT Equity ETFs trading at a premium (discount) from NAV experience more creations (redemptions) than those with a price near NAV. We examine the relationship between creations/redemptions and ETF returns and find that creation activity serves to significantly reduce mispricing in all subsamples except for U.S. Indexes. Finally, we examine the factors that influence creations/redemptions. We find that creations are positively related to lagged P/NAV, share turnover, bid-ask spread, and age of the fund. Redemptions are also positively related to share turnover and age; but are negatively related (as expected) to lagged P/NAV, bid-ask spread, and five-day average return. INTRODUCTION Exchange Traded funds (ETFs) were first created 25 years ago and have exploded in popularity among institutional and retail investors. Net new shares issuance of domestic equity ETFs rose to $148 billion in 2016 up from $50 billion in 2015. In contrast, global and international equity ETFs had net share issuance of $20 billion in 2016, down from the $110 billion issuance in 2015. 1 ETFs are similar to mutual funds in that shares represent ownership in a basket of securities. Unlike mutual funds, ETFs trade on an exchange or other trading venues throughout the day at market-determined prices that can deviate from net asset value (NAV). Trading on the exchange is classified as a secondary market transaction. In addition to secondary market trading in outstanding shares, an authorized participant (AP) can create or redeem a block of ETF shares called a creation unit. When demand is strong, APs create additional shares and this activity helps to insure that the ETF s price does not deviate substantially from its NAV. If the ETF s 1 Data from the 2017 Investment Company Fact Book. 1
price falls below NAV, the AP can buy the ETF shares in the secondary market and redeem them in kind and then sell the underlying securities to earn a profit. This arbitrage mechanism is a key component in keeping the exchange price close to the NAV. As a consequence, the number of outstanding shares in an ETF will expand with creation activity and contract with redemption activity. An AP that wishes to use the arbitrage mechanism must submit a creation/redemption order (CR order) and, once submitted, the order has three business days to clear and settle. Thus, if an AP submits a create order on day T, the AP has until day T+3 to deliver the underlying basket of securities and the actual increase in the number of shares outstanding would then be reflected on day T+4. Domestic equity CR orders are normally processed through NSCC; while international equity CR orders are not processed through NSCC unless the ETF sponsor allows cash in lieu of the securities. 2 Petajisto (2017) examines the relationship between closing Price/NAV deviations and creations and redemptions. He finds that past premiums have a positive and significant impact on ETF share creations with APs acting on premiums within 1 to 10 days. In addition, he examines how premiums respond to share creation/redemption and finds the new shares impact the premium on the same day and for two subsequent days. He explains this behavior as: market makers offload their newly created ETF share in the secondary market immediately before and after the creation process, and thus the price pressure from the new shares arises contemporaneously within about one day of share creation Evans, Moussawi, Pagano, and Sedunov (2017) provide even more light on this process by examining short selling and failures to deliver (FTDs) in equity ETFs. They show how many APs also work as market makers, and use the market maker settlement rules to manage liquidity. 2 ICI Research Perspective, Volume 20, No. 5 September 2014 2
As opportunities to arbitrage deviations between the exchange price and NAV arrive, the AP can trade immediately and then delay the actual creation or redemption for several days. In their role as market maker, the ETF may not see the associated creation or redemption up to day T+6 or longer. This paper looks at daily flows (creations and redemptions) and returns for equity ETFs trading at a premium or discount to NAV. We examine the frequency and size of creations/redemptions following large price/nav deviations and find that both the frequency and the size of creations and redemptions are larger following days with large Price/NAV deviations. While consistent with arbitrage, we also investigate abnormal returns (or alpha) following days with large P/NAV deviations. Funds with creations and redemptions usually see the deviation from parity corrected, while funds without any AP activity have deviations that continue to persist. Our results are consistent with the quick correction of deviations in the presence or arbitrageurs, but longer-run persistence in their absence. We finally consider to what degree liquidity effects AP activity. Consistent with the bond ETFs considered by Fulkerson, Jordan, and Travis (2017), liquidity appears to be an important factor for APs in equity ETFs. Both secondary market liquidity and the liquidity of the underlying assets determine the likelihood of an AP arbitraging a premium or a discount. For example, ETFs with lower share turnover are less likely to have either a creation or redemption. BACKGROUND A. ETFS AND PREMIUMS 3
Assets managed by ETFs rose from $608 billion in 2007 to $2.5 trillion in 2017. 3 ETFs are largely indexed mutual funds listed on US exchanges, and compete against traditional index mutual funds (Agapova, 2011). Investors may prefer ETFs over mutual funds for several reasons, including lower expense ratios on average and intraday liquidity. ETFs have very high volume, with median equity ETF turnover of almost 100% per week (see Table 1). A key feature of ETFs is the intrinsic arbitrage mechanism. ETFs trade on exchanges just like stocks, and the price moves as investors buy and sell. However, ETFs also allow shares to be created and redeemed daily by authorized participants (APs) using in-kind transactions. In other words, an AP must deliver (receive) the underlying positions of the ETF in order to create (redeem) new shares, effectively trading at the NAV of the portfolio. When the exchange price differs significantly from the NAV, the ETF structure puts APs in a position to earn arbitrage profits by transacting simultaneously with the ETF and on the exchange. This is a core feature of ETFs, as the built-in arbitrage mechanism should keep the exchange price close to the NAV of the underlying portfolio. 4 However, the mechanism does not always mean that exchange prices match NAVs. Several studies show that ETFs regularly trade at premiums and discounts to NAV (for example, see Engle and Sarkar, 2006; Fulkerson and Jordan, 2013; Broman, 2016; and Petajisto, 2017). Madhavan and Sobczyk (2014) develop a model of ETF price dynamics that suggests that deviations are related to price discovery and transitory liquidity effects. In the first case, the deviations occur because of information about the underlying portfolio that is not reflected simultaneously in the exchange price. In the second case, short term liquidity constraints lead to deviations. 3 Data from the Investment Company Institute s 2017 factbook: www.icifactbook.org 4 By comparison, closed-end funds do not allow for the creation or redemption of shares, and therefore often trade at a discount to the funds NAVs. 4
Piccotti (2017) offers a third explanation that suggests that some premiums can be explained by liquidity segmentation. Specifically, in some cases investors are willing to pay a premium for an ETF that has high liquidity relative to its underlying assets. Broman and Shum (2017) find supporting evidence suggesting that investors have a greater demand for ETFs with much higher liquidity than the underlying assets. Against this background and given the frequency of these deviations, several studies consider the role of arbitrage activity in the pricing of ETFs. B. ARBITRAGE ACTIVITY OF ETFS When the exchange price is lower (higher) than the NAV, the arbitrage strategy for an AP is to buy (sell) on the exchange and redeem (create) at NAV through the ETF. In this sense, creation and redemption activity should increase when prices deviate from NAV. Clifford, Fulkerson, and Jordan (2014) show that inflows (outflows) increase (decrease) in the month after premiums (discounts) occur consistent with arbitrage activity. Petajisto (2017) also finds that daily deviations predict future flows over the next few days. The deviations usually quickly disappear; equity ETF deviations often correct by the next day (Fulkerson and Jordan, 2013; Broman, 2016), while bond ETFs may persist more than a week (Fulkerson, Jordan, and Riley, 2014). The relative short-term life of deviations suggests that arbitrage is occurring, but only recently have studies begun to connect deviations more directly to AP activity. Using equity ETFs, Brown, Davies, and Ringgenberg (2016) and Petajisto (2017) find that creations (redemptions) predict lower (higher) returns in equity ETFs over several days, consistent with arbitrage by the APs. Fulkerson, Jordan, and Travis (2017) find that bond ETF deviations also 5
correct more quickly when there is creation and redemption activity among high premium and discount ETFs. Evans, Moussawi, Pagano, and Sedunov (2017) highlight one complication of using cash flows as a measure of arbitrage activity. Most APs are also market makers, and in that role, they have more leeway as to when they actually create and redeem shares. For example, market makers have six days to deliver on transactions, compared to only three days for most investors. Evans et al., show how an AP can trade new shares of an ETF during day 0, but not actually create or redeem any shares for several days, an activity they refer to as operational shorting. The practice improves liquidity, but also predicts systemic and counterparty risk. In the context of investigating the impact of arbitrage activity on premiums and discounts, it suggests that a researcher observing a premium on day t might see a creation or redemption any time in the period t to t+6. These studies get closer to the question of AP arbitrage activity, but the most direct connection of creation and redemption activity to arbitrage would be to examine the full trading activity of the APs. Unfortunately, this level of data is not readily available. A final question arises as to the effectiveness of the ETF structure when there are barriers to arbitrage. Prior studies find that premiums and discounts are larger and more volatile when the underlying assets have traits associated with higher transaction costs. For example, premiums are more volatile for ETFs investing in bonds and real estate compared to equities (Petajisto, 2017). Likewise, when ETFs hold international assets that close at a time different than US markets, premiums and discounts are larger (Engle and Sarkar, 2006; Delcoure and Zhong, 2007; Petajisto, 2017). Sudden drops in the liquidity of ETFs have led to less arbitrage activity and greater deviations between the exchange price and NAV (Borkovec, Domowitz, Serbin, and Yegerman, 2010; Ben-David, Franzoni, and Moussawi, 2014; Cespa and Foucault, 2014). In a 6
more general model, Madhaven and Sobczyk (2015) find that moving from highly liquid to less liquid assets increases the time for arbitrage to correct prices. DATA AND METHODS To construct our sample, following Evans et al., we collect daily shares outstanding data for equity ETFs from Bloomberg for the period of January 1, 2001 to December 31, 2015. We obtain fund characteristics from the CRSP Survivor-Bias-Free Mutual Fund Database, and daily prices and returns from the CRSP Daily Stock Database. These observations are matched by ticker and date. Following Fulkerson, Jordan, and Travis (2017), we apply standard filters to the sample. In an attempt to eliminate the unusual behavior of newly issued funds, we remove observations where the fund is less than six months old and where the total net assets is under $20 million. In order to be categorized as an equity ETF, we require that a fund contains at least 80% equities at some point during the sample. Abnormally large daily changes in return-adjusted cash flow are excluded when greater the 200% or less than -50%. A final evaluation by hand is performed to confirm that the fund has an equity based strategy. For this study, only passively managed equity ETFs are included in the sample. The final sample includes 798 passively managed equity ETFs with 1,133,726 daily observations for the January 1, 2001 to December 31, 2015 period. Exhibit 1 provides summary statistics. The average fund in the sample is composed of 95% equities, has a total net asset value of $1.7 billion, is six years old, has an expense ratio of 49 basis points and a portfolio turnover ratio of 35% per year, and comes from a fairly large fund family. On average, these funds experience high daily trading volume and trade at very narrow bid-ask spreads. The price-to-net asset value ratio for these funds is usually very close to one, with a mean of 1.00024, but does 7
experience some daily deviations from one, with a standard deviation of 0.00668. Similar to Fulkerson, Jordan, and Travis [2017], we measure daily fund flow with a return-adjusted change in daily total net assets. Daily total net asset value is estimated as the product of daily shares outstanding and daily NAV by the following: FFFFFFFF tt = SSSS tt NNNNNN tt SSSS tt 1 NNNNNN tt 1 (1+rr tt ) SSSS tt 1 NNNNNN tt 1 (1) where SO represents daily shares outstanding from Bloomberg, NAV represents daily net asset value from CRSP, and r represents daily return from CRSP. Creations flow represents flow from increases in shares outstanding while Redemptions flow represents flow from decreases in shares outstanding. Since decreases in shares outstanding produce negative flow values, for illustrations and analysis, Redemptions flow is measured in absolute value. The majority of observations are no flow days as seen by the median creation and redemption flow values of 0% each. [Insert Exhibit 1 about here.] We use the Lipper asset and class codes in CRSP to categorize the funds and form subsamples for analysis. Exhibit 2 provides a summary of subsample sizes by percentage of observations and number of funds. We categorize a fund as U.S. Index if it is composed of a broad index of U.S. based equities. The International Index category contains funds of broad indexes of either International or Global equities. The Sector Index category represents domestic or international funds that are limited to specific sectors of the economy such as utilities or consumer discretionary. These first three categories make up over 95% of the sample and are of somewhat similar proportion. The fourth and smallest category, about 4.1% of the sample, is the Leveraged fund category. We label a fund as leveraged if it has a name indicating leveraged or shorting such as the 130/30, 2x, 3x, 2x, 3x, and inverse funds. 8
[Insert Exhibit 2 about here.] Trading in the ETF market can be seen in both the primary market, as measured by flow, and the secondary market on the exchange. In order to compare the relative size of the primary and secondary market activity for the equity ETF sample, Exhibit 3 displays summary statistics for the dollar flow and dollar volume where dollar flow is estimated by the return-adjusted change in total net asset value daily and dollar volume is measured by the product of daily trading volume and the closing price of the fund. As noted previously, the majority of days are no flow days. Dollar volume is almost seven times larger than dollar flow for the average fund. Thus, the large majority of ETF trading is in the secondary market. [Insert Exhibit 3 about here.] For more perspective on the frequency of primary market activity in the sample, Exhibit 4 displays data on the percent of observations that contain creation or redemption activity. In full sample, creations occur 13.5% of the time while redemptions occur 7.7% of the time. The higher frequency of creations over redemptions can likely be attributed to the overall growth of the ETF industry during the sample period. As shown, U.S. Index funds have the highest relative frequency of creations while leveraged funds experience the highest relative frequency of redemptions. For all subsamples, the majority of observations are no flow days. [Insert Exhibit 4 about here.] FLOW AND THE PRICE-TO-NAV RATIO When an ETF s market price deviates from its net asset value, arbitrage opportunities arise for the authorized participants (APs). Exhibit 5 displays frequency of primary market activity on the five days prior (time: t-5 to t-1), the day of (time: t), and the five days following 9
(time: t+1 to t+5) a price-to-nav ratio deviation of various sizes. Panel A provides creation frequency surrounding a price-to-nav premium event while Panel B provides redemption frequency surrounding a price-to-nav discount event. In Panel A, when the price-to-nav ratio of the average fund reaches the 99 th percentile or more, creation frequency is shown to decrease to a level below the full sample reference level on that same day. In the days following this event, creation frequency spikes at over 21% and remains abnormally high for up to five days following the event. For a 95 th to 99 th percentile price-to-nav ratio deviation day, a similar yet smaller in magnitude trend is observed. For a 75 th to 95 th percentile price-to-nav ratio deviation day, a small increase in creation frequency is observed in the first two or three days following the event. In Panel B, when the price-to-nav ratio of the average fund reaches the 1 st percentile or less, redemption flow is shown to increase slightly leading up to the event, decrease on the day of the event, and increase sharply after the event for the next 5 days. For a 1 st to 5 th percentile day, the redemption flow is observed to increase slightly before and on the event day followed by a large increase the day after and remaining high for up to day t+5. For a 5 th to 25 th percentile day, redemption flow is shown to increase up to day t+1 and slowly decrease in the following days. [Insert Exhibit 5 about here.] Along with flow frequency, average flow size surrounding a large deviation in price-to- NAV can increase since an AP may have an incentive to create (redeem) a larger quantity of creation units. Exhibit 6 displays average daily size of primary market flow activity on the five days prior (time: t-5 to t-1), the day of (time: t), and the five days following a price-to-nav ratio (time: t+1 to t+5) deviation of various sizes. Panel A provides average creation flow size surrounding a price-to-nav premium event while Panel B provides average redemption flow 10
size surrounding a price-to-nav discount event. As shown in Panel A, average creation flow size is above the full sample reference average before, during, and after a deviation anywhere above the 75 th percentile. On the day after the event, average size does increase sharply for the 99 th percentile and above event; this is followed by a relative decrease in average creation size. The effect is much less pronounced for the 95 th to 99 th percentile event. For the 75 th to 95 th percentile event, timing appears to have much less impact on average creation size. In Panel B, when the price-to-nav ratio drops to the 1 st percentile or less, average redemption flow size remains above the full sample reference point before the event, decreases on the day of the event, and sharply increases on the day after the event; by day t+5, average redemption size is still higher than it was prior to the event. For a 1 st to 5 th percentile event, a similar but much more muted effect is observed. For a 5 th to 25 th percentile discount event, the timing of the event does not appear to effect average redemption size significantly. [Insert Exhibit 6 about here.] ABNORMAL RETURNS FOLLOWING A PRICE-TO-NAV RATIO PREMIUM OR DISCOUNT We next consider the impact of returns following a price-to-nav ratio premium or discount and whether those returns are affected by creation or redemption flow. For each day during the sample, we rank funds by the price-to-nav ratio. Equal-weighted portfolios are formed for the top and bottom deciles. Portfolio returns for the five days following the premium event are regressed on the Fama-French-Carhart four-factor model factors. 5 Resulting abnormal returns, or alphas, are displayed for top decile portfolios in Exhibit 7 and for bottom decile portfolios in Exhibit 8. 5 Fama-French-Carhart factor data are obtained from Ken French s website. 11
[Insert Exhibit 7 about here.] In Panel A of Exhibit 7, results are displayed for going short the ETF price and going long the ETF NAV (i.e., buy the equities and create or sell the ETF), which would be the naturally expected primary market trades following a price-to-nav ratio premium event. In column i, results are displayed for the entire top decile regardless of forthcoming flow. For the full sample and the largest three subsamples (U.S. Index, International Index, and Sector Funds), the short price alpha is large and significant while the long NAV alpha is negative and significant; the relative size of the short price alpha appears to be larger than the long NAV alpha for these groups. Alpha signs are reversed for the Leveraged subsample. Column ii displays results for funds that experienced neither creations nor redemptions during the following five days after the price-to-nav premium event. For the full sample and the three largest subsamples, column ii results are similar to column i results but with overall smaller alphas. The Leveraged subsample resulted in insignificant alpha values in column ii. Column iii provides results for funds that experienced net creation flow during the five days following the price-to- NAV ratio premium event. For U.S and International Indexes in column iii, alpha values are somewhat larger than the no flow alphas in column ii. Sector fund alphas are lower in column iii than reported in column ii. Leveraged subsamples alphas are insignificant in column iii. Column iv displays results for only funds that experienced net redemptions during the five days following the price-to-nav premium event. For the three largest subsamples, alpha coefficients are smaller in column iv compared to column iii. For the Leveraged subsample in column iv, alpha coefficients are relatively large and significant. Panel B of Exhibit 7 displays the combined arbitrage alphas from Panel A (i.e., the result of combining the short price and long NAV returns). Column v displays arbitrage alphas for only 12
those funds that experienced no creations or redemptions in the five days following the premium event. For full sample, the arbitrage alpha is 2.31% per year and significant at the 1% level. For the U.S. Index subsample, the arbitrage alpha is 0.54% and significant at the 10% level. For the International Index subsample, the arbitrage alpha is 5.51% and significant at the 1% level. For the Leveraged fund subsample, the arbitrage alpha is negative and significant at the 10% level. In column vi, alphas are displayed for only those funds that experienced net creation flow during the five days following the premium event. The full sample arbitrage alpha is 5.97% per year and significant at the 1% level. For all subsamples, the arbitrage alpha value in column vi is positive and significant. Column vii provides alphas for only those funds that experience net redemptions following the premium event. None of the alphas in column vii are significant. Column viii displays the difference in alphas between columns vi and v (i.e., the difference between net creations arbitrage alpha and no flow arbitrage alpha). For full sample and all but the U.S. Index subsample, the difference in alphas is significant at the 1% level. Column ix provides the difference in alphas between columns vii and v (i.e., the difference between net redemptions arbitrage alpha and no flow arbitrage alpha). Only the Leveraged subsample difference in alphas in column ix is significant. It may be concluded from column viii that following a price-to-nav premium event, creation activity in the primary market serves to significantly reduce mispricing in all subsamples except for U.S. Indexes. For U.S. Indexes, even though there is significant creation activity, the secondary market appears to be just as effective as the primary market at reducing mispricing between the ETF price and NAV. In Exhibit 8, the corresponding results for the daily bottom decile of price-to-nav (discount) ratio values are displayed. For a price-to-nav ratio discount, we examine the portfolio returns for going long the ETF price and going short the NAV of the fund in Panel A. 13
Column i values are from all funds in the bottom decile regardless of flow. For all but the Leveraged subsample, alphas are negative and significant for long price and positive and significant for short NAV; signs are reversed for the Leveraged subsample. Column ii provides values for only those funds that experience no creations or redemptions in the five days following the price-to-nav discount event. Alpha values for this no flow group are smaller than in column i. Column iii contains alpha values for the funds that experience net creation flow following the discount event; in full sample and for the subcategories except the leveraged funds, the alpha values are smaller than in columns i or ii. Column iv provides alpha values for those funds that experience net redemptions over the five day period following the discount event. Except for the leveraged subsample, the alpha values in column iv are smaller than the no flow column. [Insert Exhibit 8 about here.] In Panel B of Exhibit 8, the combined arbitrage alphas are displayed from the long price and short NAV combined strategy. In column v, the no flow funds exhibited positive and significant alpha values for all but the leveraged subsample. In full sample, the alpha for discount arbitrage is 3.89% per year and significant at the 1% level. The International Index subsample exhibited the largest alpha value at 6.27%. Column vi provides alpha values for the funds with forthcoming net creations. Alpha values are insignificant except for the Leveraged subsample, which has a negative alpha value. In column vii, arbitrage alpha values are positive and significant for all but the Leveraged subsample. In full sample, the alpha value of 6.05% per year is significant at the 1% level. The International Index subsample has the largest discount arbitrage alpha value at 8.90% per year. When comparing the difference between alphas, in column ix the difference between the net redemption alpha and the no flow alpha is positive and 14
significant in full sample, the International Index subsample, and the Sector fund subsample. In full sample, the net redemptions arbitrage alpha is 2.84% per year larger than the no flow arbitrage alpha. For the International Index and Sector fund subsamples, it appears that redemption flow has a significant effect on reducing the price-to-nav ratio discount during this sample period. For U.S. Index funds, as in the price-to-nav premium scenario from Exhibit 7, the secondary market appears to be just as effective as the primary market at reducing price-to- NAV deviations. DRIVERS BEHIND CREATION AND REDEMPTION We provide multivariate regression analysis in Exhibit 9 to explore the driving forces behind equity ETF creation and redemption. We use a two-way fixed effects panel regression to model the creation and redemption activity over a five-day window following independent variables that include fund characteristics, exchange activity, and prior returns. Fund characteristics include the age of a fund at observation, fund size, fund family size, expense ratio, and turnover ratio. Exchange variables include prior five-day trading volume average and standard deviation, prior five-day bid-ask spread average and standard deviation, prior one-day price-to-nav ratio along with two lags of this variable, prior five-day price-to-nav ratio standard deviation, and prior five-day share turnover. Return variables include prior five-day return average and standard deviation along with the square of five-day prior average return to potentially capture any return-chasing convexity. Fixed effects are used to control for individual funds and for the month and year time periods. We model creations and redemptions separately. Panel A provides results for creations and Panel B for redemptions. All independent variables are converted to z-scores for ease of coefficient interpretation. 15
[Insert Exhibit 9 about here.] In Panel A of Exhibit 9, the creation model regression results are displayed for the full sample and subsamples of U.S. Index, International Index, Sector Funds, and Leveraged Funds. In full sample, the fund characteristics control variables are mostly significant. For a one standard deviation increase in fund age, total creations over the five following days increase by8.918%. Fund size, fund family size, and expense ratio are inversely related to creations. When looking at exchange variables, creations increase with a decrease in average trading volume and an increase in trading volume volatility. As average bid-ask spreads widen, creations increase. As expected, across all subsamples, creations increase when the prior day price-to- NAV ratio increases; one-day lag and two-day lag of the prior price-to-nav ratio is positive and significant for all but the U.S. Index subsample. In full sample, a one standard deviation increase in the prior one-day price-to-nav ratio of a fund leads to an average increase in creations of 0.238%. This coefficient value appears to be surprisingly low but we must recall from Exhibit 4 that over 80% of observations have no creation activity. The 2-day and 3-day lagged price-to- NAV ratios are generally positively related to creations for the subsamples, which the exception of the US Index for which the relationship is not significant. For full sample and the International Index subsample, an increase in price-to-nav volatility leads to an increase in creations. Share turnover is positively related to creations for all subsamples. Prior average five-day return is positively related to creations for U.S. Index funds. For Leveraged funds, a decrease in prior average return leads to an increase in creations. In all but U.S. Indexes, positive return convexity is observed. For all but Sector Funds, a decrease in return volatility leads to an increase in creations. In summary, creations are more common with increases in bid-ask spread, price-to- 16
NAV ratio, and share turnover; creations are less common with increases in trading volume and return volatility. In Panel B of Exhibit 9, redemption model regression results are displayed. Similar to creations, redemptions are more common with older and smaller funds with lower expense ratios. Only U.S. Index fund redemptions are shown to be affected by average trading volume, as displayed with a positive relationship. A decrease in the bid-ask spread is followed by an increase in redemption for the Sector Index subsample; the opposite is observed for Leveraged Funds. The expected result of a negative relationship between the prior day price-to-nav ratio and redemptions is observed; the two-day lag and three-day lag variables are also negative and significant for all subsamples except U.S. Indexes; U.S. Indexes displayed a positive relationship between redemptions and the price-to-nav ratio two and three days prior. The volatility of the price-to-nav ratio is positively related to redemptions for only the International Index fund subsample. Share turnover is positively related to redemptions for all but U.S. Index funds. For the three largest subsamples, a decrease in prior average returns leads to increased redemptions; redemptions for leveraged funds increase when prior returns have increased. Across all subsamples, positive return convexity is linked to increases in redemptions. In summary, U.S. Index fund redemptions are related to older funds with increases in trading volume and decreases in the prior price-to-nav ratio and prior returns. For International Index funds and Sector funds, redemptions are related to larger family sizes, lower expense ratios, high turnover ratios, decreases in the price-to-nav ratio, increases in share turnover, and deceases in returns. For Leveraged funds, redemptions increase for older funds, increases in bid-ask spreads, decreases in the price-to-nav ratio, and increases in share turnover and prior returns. 17
CONCLUSION This paper contributes to the existing literature on ETF fund flows by examining flow activity around premiums and discounts for equity ETFs. In general, premium ETFs with creation activity see a subsequent decrease in the size of the premium; while discount ETFs with redemption activity see a subsequent decrease in the size of the discount. Despite this impact, our sample of equity ETFs experience creation activity in only about 13 percent of our observations and redemptions in only about 7 percent of our observations. We use a multivariate two-way fixed effects regression model to examine the factors that may affect the likelihood of a creation or redemption. As expected, we find that premiums (discounts) are positively related to creations (redemptions). In addition, we find that ETFs with high share turnover have more creations/redemptions and that the ETF size is inversely related to creations/redemptions. We also find that the five-day return is generally inversely related to redemptions. Presumably, the lack of demand from lower returns causes the AP to redeem shares. 18
REFERENCES Agapova, Anna. Conventional Mutual Index Funds Versus Exchange Traded Funds. Journal of Financial Markets, 14 (2011), pp. 323-343. Antoniewicz, Rochelle, and Jane Heinrichs. 2014. Understanding Exchange-Traded Funds: How ETFs Work. ICI Research Perspective 20, no. 5 (September). Available at www.ici.org/pdf/per20-05.pdf. Ben-David, Itzhak, Francesco Franzoni, and Rabih Moussawi. Do ETFs Increase Volatility? NBER working paper No. 20071. Borkovec, Milan, Ian Domowitz, Vitaly Serbin, and Henry Yegerman. Liquidity and Price Discovery in Exchange-Traded Funds: One of Several Possible Lessons from the Flash Crash. Journal of Index Investing 1 (2010), pp. 24-42. Broman, Markus. Liquidity, Style Investing, and Excess Comovement of Exchange-Traded Fund Returns. Journal of Financial Markets 30 (2016), pp. 27-53. Broman, Markus, and Pauline Shum. Relative Liquidity, Fund Flows and Short-term Demand: Evidence from Exchange-Traded Funds. Syracuse University working paper (2017). Cespa, Giovanni, and Thierry Foucault. Illiquidity Contagion and Liquidity Crashes. Review of Financial Studies 27 (2014), pp. 1615-1660. Clifford, C., J. Fulkerson, and B. Jordan. What Drives ETF Flows? Financial Review, 49 (2014), pp. 619-642. Delcoure, Natalya, and Maosen Zhong. On the Premiums of ishares. Journal of Empirical Finance 14 (2007), pp. 168-195. Engle, R., and D. Sarkar. Premiums-discounts and Exchange Traded Funds. Journal of Derivatives, 13 (2006), pp. 27-45. Evans, Richard B. and Moussawi, Rabih and Pagano, Michael S. and Sedunov, John, ETF Short Interest and Failures-to-Deliver: Naked Short-Selling or Operational Shorting? Darden Business School Working Paper No. 2961954 (2017). Fulkerson, J., and B. Jordan. Reading Tomorrow s Newspaper: Predictability in ETF Returns. Journal of Index Investing, 4 (2013), pp. 23-31. Fulkerson, J., S. Jordan, and D. Travis. Bond ETF Arbitrage Strategies and Daily Cash Flow Journal of Fixed Income, 27 (2017), pp. 49-65. Madhavan, Ananth, and Aleksander Sobczyk. Price Dynamics and Liquidity of Exchange- Traded Funds. Journal of Investment Management 14 (2016), pp. 1-17. 19
Petajisto, A. Inefficiencies in the Pricing of Exchange-traded Funds. Financial Analyst Journal, 73 (2017), pp. 24-54. Stratmann, T., Welborn, J.W., 2016. Informed short selling, fails-to-deliver, and abnormal returns, Journal of Empirical Finance 38, 81 102. 20
Exhibit 1 Summary Statistics for Equity ETF Sample, January 1, 2001 December 31, 2015 Summary statistics are provided for the equity ETF sample for the period January 1, 2001 to December 31, 2015. The sample includes 798 equity ETFs with 1,133,726 daily observations. Shares outstanding data is collected from Bloomberg. All other data is from CRSP. Observations are included only when data is complete. Percentage in equities is the common stock holding percentage reported. TNA is the total net assets reported in millions of dollars. Age for an observation, in years, is the difference between the observation date and the fund first offer date. Expense ratio is the total annual management fee. Portfolio turnover ratio is the portion of the fund reported bought or sold in the prior year. Family fund count is the number of ETFs in our sample that is managed by the same firm, matched by the family fund code in CRSP. Family size is the total TNA, shown in millions, managed by the same firm. Average 1-day return is the average return over the prior one-day period, annualized. Average 5-day return is the average return over the prior five-day period, annualized. Std. dev. 5-day return is the standard deviation of daily returns over the prior five-day period, annualized. Shares outstanding is shown in billions. Volume 5-day average is the average daily volume over the prior 5-day period, shown in hundreds. Volume 5-day std. dev. is the standard deviation of daily volume over the prior 5-day period, shown in hundreds. Share turnover 5-day is the total volume over the prior five days divided by shares outstanding from the beginning of that five day period. Spread 5-day average and Spread 5-day std. dev. are calculated from the prior five day period daily bid-ask spread scaled by the midpoint of the bid-ask spread. P-NAV ratio 1-day average, and P-NAV ratio 5-day std. dev. are calculated from prior daily P-NAV ratio data, where P-NAV ratio is calculated by the bid-ask midpoint divided by the NAV ratio. Flow is calculated as the return adjusted daily percent change in TNA, as shown in Equation 1. Creations flow is the flow from increases in shares outstanding. Redemptions flow is the flow from decreases in shares outstanding. Flow from redemptions naturally creates negative flow values; Redemptions flow is displayed here in absolute value. Category Variable Mean Median Std. Dev. Fund Characteristics Percentage in common stock 95.23 99.48 13.72 TNA 1731 263 6783 Age 6.06 5.36 3.93 Expense ratio 0.0049 0.0050 0.0023 Portfolio turnover ratio 0.3502 0.2100 0.6032 Family fund count 77.01 62.00 53.82 Family size 168,082 74,639 187,177 Returns Average 1-day return 0.00036 0.00076 0.01680 Average 5-day return 0.00037 0.00069 0.00712 Std. dev. 5-day return 0.01296 0.01011 0.01102 Market variables Shares outstanding 32.30 6.70 91.50 Volume 5-day average 1,627,673 68,660 9,777,299 Volume 5-day std. dev. 486,688 35,419 2,677,584 Share turnover 5-day 3.5727 0.9641 9.3616 Spread 5-day average 0.0020 0.0012 0.0030 Spread 5-day std. dev. 0.0008 0.0004 0.0023 P/NAV ratio 1-day 1.00024 1.00006 0.00668 P/NAV ratio 5-day std. dev. 0.00230 0.00071 0.00377 Daily flow Creations flow 0.00357 0.00000 0.02514 Redemptions flow 0.00212 0.00000 0.01399 21
Exhibit 2 Equity ETF Styles by Sample We categorize the equity ETF sample by equity investment or holdings style as labeled by Lipper class and asset codes. We label a fund as U.S. Index if it is composed of a broad index of U.S. based equities. The International Index category contains funds of broad indexes of either International or Global equities. The Sector Index category represents domestic or international funds that are limited to specific sectors of the economy such as utilities or consumer discretionary. These first three categories make up over 95% of the sample and are of somewhat similar proportion. The fourth and smallest category, about 4.1% of the sample, is the Leveraged fund category. We label a fund as leveraged if it is managed leveraged or short such as the 130/30, 2x, 3x, 2x, 3x, and inverse funds. Style Categories % of Observations No. of Funds U.S. Index 29.4% 236 International Index 28.9% 264 Sector Index 37.6% 281 Leveraged 4.1% 46 22
Exhibit 3 Dollar Volume and Dollar Flow for the Equity ETF Sample In this table, we present summary statistics on the dollar volume and dollar flow of the bond ETF sample. Dollar volume, i.e. secondary market volume, is calculated daily by multiplying the volume of shares traded times closing price. Dollar flow, i.e. primary market volume, is calculated by the return adjusted change in total net assets (TNA), where TNA is estimated daily by multiplying shares outstanding times NAV. Total volume is the sum of dollar volume and dollar flow. Variable Mean Median Std. Dev. Dollar Volume $95,800,000 $2,300,000 $995,000,000 Dollar Flow $14,000,000 $540,000 $125,000,000 Total Volume $111,000,000 $3,700,000 $1,070,000,000 Dollar / Total Volume 71.1% 79.7% 26.4% Total Volume / TNA 12.6% 1.2% 287% 23
Exhibit 4 Percent of Observations with Creations or Redemptions 18.00% 16.00% Create Redeem 14.00% 12.00% 10.00% 8.00% 6.00% 4.00% 2.00% 0.00% Full Sample US Index International Sector Leveraged 24
Exhibit 5 Panel A: Frequency of Creations around a Price-to-NAV Premium Event 22.0% 20.0% 18.0% >99th percentile 95th to 99th percentile 75th to 95th percentile full sample reference 16.0% 14.0% 12.0% 10.0% -5-4 -3-2 -1 0 1 2 3 4 5 Days surrounding the P/NAV premium event 25
Exhibit 5 Panel B: Frequency of Redemptions around a Price-to-NAV Discount Event 18.0% 16.0% 14.0% <1st percentile 1st to 5th percentile 5th to 25th percentile full sample reference 12.0% 10.0% 8.0% 6.0% -5-4 -3-2 -1 0 1 2 3 4 5 Days surrounding the P/NAV discount event 26
Exhibit 6 Panel A: Size of Creations around a Price-to-NAV Premium Event 5.50% 5.00% 4.50% >99th percentile 95th to 99th percentile 75th to 95th percentile full sample reference 4.00% 3.50% 3.00% 2.50% 2.00% -5-4 -3-2 -1 0 1 2 3 4 5 Days surrounding the P/NAV premium event 27
Exhibit 6 Panel B: Size of Redemptions around a Price-to-NAV Discount Event 5.50% 5.00% 4.50% <1st percentile 1st to 5th percentile 5th to 25th percentile full sample reference 4.00% 3.50% 3.00% 2.50% 2.00% -5-4 -3-2 -1 0 1 2 3 4 5 Days surrounding the P/NAV discount event 28
Exhibit 7 Abnormal Returns Following a P/NAV Premium Panel A: Short Price and Long NAV Each of the sample, funds are ranked by their P/NAV ratios. The results in this exhibit are obtained following those funds that rank in the top decile. For each day, equally-weighted portfolios of those premium funds are formed for the funds prices and the funds NAV values over the next five days. In order to test for abnormal returns following a premium P/NAV ratio day, 5-day returns are calculated from going short the funds prices and from going long the funds NAV values. The return risk premiums of an equally-weighted portfolio of these ETFs are regressed on a Fama-French-Carhart four factor model. The annualized excess return, or Alpha, is displayed along with its respective p-value. The number of observation days of available funds is provided below each scenario. Column i is for all funds in the group regardless of flow. Column ii is for only those funds that experienced no flow in the following five days. Column iii is for only those funds that experienced a net increase in flow, or net creations, in the following five days. Column iv is for only those funds that experienced a net decrease in flow, or net redemptions, in the following five days. [i] [ii] [iii] [iv] 1-day P/NAV Percentile Range Short Price: Apha for the next 5 days Long NAV: Alpha for the next 5 days No Flow over the trading period Short Price: Apha for the next 5 days Long NAV: Alpha for the next 5 days Net Creations over the trading period Short Price: Apha for the next 5 days Long NAV: Alpha for the next 5 days Net Redemptions over the trading period Short Price: Apha for the next 5 days Long NAV: Alpha for the next 5 days Full Sample U.S. Index International Index Sector Leveraged Top Decile Top Decile Top Decile Top Decile Top Decile 139.13% -135.34% 138.03% -135.71% 133.81% -127.85% 88.92% -87.87% [0.000] [0.000] [0.000] [0.000] [0.000] [0.000] [0.000] [0.000] obs. days 3768 obs. days 3693 obs. days 3617 obs. days 2640 141.21% -140.73% 122.16% -121.77% 134.22% -133.58% 126.99% -126.84% [0.000] [0.000] [0.000] [0.000] [0.000] [0.000] [0.000] [0.000] obs. days 3766 obs. days 3312 obs. days 3396 obs. days 2312 141.86% -135.04% 118.71% -113.21% 130.97% -121.60% 105.95% -106.00% [0.000] [0.000] [0.000] [0.000] [0.000] [0.000] [0.000] [0.000] obs. days 3758 obs. days 3231 obs. days 3395 obs. days 1410 138.65% -137.31% 134.86% -134.47% 122.78% -119.14% 99.56% -98.71% [0.000] [0.000] [0.000] [0.000] [0.000] [0.000] [0.000] [0.000] obs. days 3715 obs. days 3489 obs. days 3203 obs. days 1843-85.44% 86.58% 11.41% -12.76% -100.48% 103.69% -326.26% 326.92% [0.010] [0.009] [0.267] [0.215] [0.171] [0.158] [0.001] [0.001] obs. days 1947 obs. days 1345 obs. days 1129 obs. days 985 29
Exhibit 7 Abnormal Returns Following a P/NAV Premium Panel B: Premium Arbitrage Each of the sample, funds are ranked by their P/NAV ratios. The results in this exhibit are obtained following those funds that rank in the top decile. For each day, equally-weighted portfolios of those premium funds are formed for the funds prices and the funds NAV values over the next five days. In order to test for abnormal returns following a premium P/NAV ratio day, 5-day returns are calculated from going short the funds prices and from going long the funds NAV values. The return risk premiums of an equally-weighted portfolio of these ETFs are regressed on a Fama-French-Carhart four factor model. The annualized excess return, or Alpha, is displayed along with its respective p-value. The number of observation weeks of available funds is provided below each scenario. Column v is the combination of going short price and long NAV for only those funds that experienced no flow in the following five days. Column vi is for only those funds that experienced a net increase in flow, or net creations, in the following five days. Column vii is for only those funds that experienced a net decrease in flow, or net redemptions, in the following five days. Column viii is the resulting difference in alphas between net creations funds and no flow funds. Column ix is the resulting difference in alphas between net redemptions funds and no flow funds. [v] [vi] [vii] [viii] [ix] No Flow over the trading period Net Creations over the trading period Net Redemptions over the trading period Net Creations No Flow Net Redemptions No Flow 1-day P/NAV Percentile Range Alpha for Premium Arbitrage Alpha for Premium Arbitrage Alpha for Premium Arbitrage Difference in Alphas Difference in Alphas Full Sample U.S. Index International Index Sector Leveraged Top Decile Top Decile Top Decile Top Decile Top Decile 2.31% 5.97% 1.01% 3.47% 0.09% [0.001] [0.000] [0.372] [0.000] [0.918] obs. days 3693 obs. days 3617 obs. days 2640 obs. days 3553 obs. days 2617 0.54% 0.65% 0.04% 0.37% -0.04% [0.084] [0.001] [0.908] [0.308] [0.931] obs. days 3312 obs. days 3396 obs. days 2312 obs. days 2996 obs. days 2062 5.51% 9.38% -0.14% 3.16% -1.24% [0.000] [0.000] [0.950] [0.001] [0.505] obs. days 3231 obs. days 3395 obs. days 1410 obs. days 2868 obs. days 1301 0.64% 3.46% 0.61% 3.14% 1.19% [0.209] [0.000] [0.550] [0.000] [0.276] obs. days 3489 obs. days 3203 obs. days 1843 obs. days 2997 obs. days 1768-1.25% 3.15% 0.66% 4.23% 2.85% [0.063] [0.003] [0.607] [0.003] [0.094] obs. days 1345 obs. days 1129 obs. days 985 obs. days 740 obs. days 632 30
Exhibit 8 Abnormal Returns Following a P/NAV Discount Panel A: Long Price and Short NAV Each of the sample, funds are ranked by their P/NAV ratios. The results in this exhibit are obtained following those funds that rank in the bottom decile. For each day, equally-weighted portfolios of those discount funds are formed for the funds prices and the funds NAV values over the next five days. In order to test for abnormal returns following a discount P/NAV ratio day, 5-day returns are calculated from going long the funds prices and from going short the funds NAV values. The return risk premiums of an equally-weighted portfolio of these ETFs are regressed on a Fama-French-Carhart four factor model. The annualized excess return, or Alpha, is displayed along with its respective p-value. The number of observation weeks of available funds is provided below each scenario. Column i is for all funds in the group regardless of flow. Column ii is for only those funds that experienced no flow in the following five days. Column iii is for only those funds that experienced a net increase in flow, or net creations, in the following five days. Column iv is for only those funds that experienced a net decrease in flow, or net redemptions, in the following five days. [i] [ii] [iii] [iv] 1-day P/NAV Percentile Range Long Price: Alpha for the next 5 days Short NAV: Alpha for the next 5 days No Flow over the trading period Long Price: Alpha for the next 5 days Short NAV: Alpha for the next 5 days Net Creations over the trading period Long Price: Alpha for the next 5 days Short NAV: Alpha for the next 5 days Net Redemptions over the trading period Long Price: Alpha for the next 5 days Short NAV: Alpha for the next 5 days Full Sample U.S. Index International Index Sector Leveraged Bottom Decile Bottom Decile Bottom Decile Bottom Decile Bottom Decile -139.87% 144.63% -136.96% 141.12% -121.98% 123.04% -128.67% 134.69% [0.000] [0.000] [0.000] [0.000] [0.000] [0.000] [0.000] [0.000] obs. days 3754 obs. days 3664 obs. days 3307 obs. days 3331-137.46% 138.03% -130.31% 131.65% -120.37% 120.48% -122.54% 123.02% [0.000] [0.000] [0.000] [0.000] [0.000] [0.000] [0.000] [0.000] obs. days 3715 obs. days 3487 obs. days 2498 obs. days 2933-137.51% 145.03% -120.14% 126.46% -98.63% 101.10% -117.03% 125.92% [0.000] [0.000] [0.000] [0.000] [0.000] [0.000] [0.000] [0.000] obs. days 3727 obs. days 3332 obs. days 2100 obs. days 2854-134.33% 136.18% -126.14% 127.64% -126.20% 127.16% -113.69% 116.95% [0.000] [0.000] [0.000] [0.000] [0.000] [0.000] [0.000] [0.000] obs. days 3708 obs. days 3417 obs. days 2466 obs. days 2900 44.53% -44.93% 29.60% -29.39% -54.92% 49.15% 176.31% -174.97% [0.036] [0.034] [0.212] [0.213] [0.564] [0.608] [0.002] [0.002] obs. days 1940 obs. days 1331 obs. days 602 obs. days 1304 31