University of Amsterdam
Amsterdam Business School
Master in International Finance
Master Thesis:
Growth of Index ETFs and its effects on
Underlying Securities
Yi Bin Dong
11763884
Supervisor: Dennis Jullens
Table of Contents
1. Introduction ... 3
2. Literature Review ... 5
3. Theoretical Framework ... 8
4. Hypothesis & Research Questions ... 12
5. Data ... 15
6. Regression Tests and Results ... 18
6.1 ETF Ownership and its effect on Volatility and Turnover ... 18
6.2 Identifying Non-Fundamental Volatility ... 21
6.3 Possible Extensions for Future Research ... 23
7. Conclusion ... 24
References... 25
1.
Introduction
Since the mid-1990s, exchange traded funds (ETFs) have become a popular investment vehicle, thanks to game changing feature such as continuous trading and very low
transaction costs. Shares of ETFs are traded on the major stock exchanges, and a significant majority of the funds aim to replicate the performance of an index. ETFs have shown spectacular growth, and the popularity of ETFs has skyrocketed in recent years. ETF daily trading volume exceeded 36% of overall stock market trading volume in the first half of 2016. By ownership measures, the largest ETF by asset size, State Street’s SPDR S&P 500 ETF (SPY) currently has net assets of $275 Billion USD, translating to approx. 1.3% of the total S&P 500 market capitalization only for one fund1, and the number of ETF listings has surpassed the number of individual stocks on the NASDAQ exchange. The non-trivial size and scale of the ETF market has to be taken into consideration when evaluating listed securities, and this paper aims to investigate further using the FTSE 100 and FTSE 250 universe.
Continuous trading, low fees are features that give an advantage to the ETF structure versus comparative assets in the mutual fund category, these factors could potentially improve price discovery and provide greater liquidity for the underlying securities. The additional layer of liquidity that ETFs provide could also serve as a double-edged sword. Specifically, ETFs attract investors with a short-term horizon, these investors’ trades may introduce noise to the price of the ETF. Since arbitrageurs continuously attempt to eliminate price discrepancies between the ETF and the underlying securities, investors’ demand shocks may be transmitted from the ETF to the underlying securities through the arbitrage process.
In this study, we aim to explore the relationship between ETF ownership of stocks, and the effect that ownership have on the quality of the underlying. On the level of underlying
securities, since arbitrageurs and APs (authorized participants, market maker for a given ETF) will act to ensure prices of the underlying securities are aligned with those of the ETF, we predict the underlying securities will display a greater co-movement with the index, leading
to an increase in asset prices, price volatility, and correlation with other indices. The focus of this paper is directed at exploring the liquidity shock propagation mechanisms related to ETF trading, and whether this trading causes increased volatility, and adds noise to the price of the underlying assets.
To quantify the relationship between ETF ownership and the quality of the underlying, we employ the trading data for ETFs and stocks traded on the London Stock Exchange (LSE). Similar studies regarding ETF ownership and volatility has been done in the context of the U.S. ETF market by Ben-David, Franzoni and Moussawi (2014), but no similar studies have been conducted in context of the UK market, perhaps due to the relatively smaller size of the market, we will establish a relevant theoretically framework and apply it in the context of the UK markets.
2.
Literature Review
Most ETFs are plain vanilla that track an index, the same way as passive index mutual funds. However, unlike index mutual funds, ETFs are listed on an exchange and continuously trade when the market is open. The effect that ETFs have on its underlying securities is not completely new to empirical research. Several papers have commented on the overall effects of ETF trading have on stock characteristics. For example, Hedge and McDermott (2004) found that after the introduction of index ETFs that track the DJIA and NASDAQ, liquidity on those exchanges improved significantly.
In the real trading environment, we can observe temporary differences between the price of an ETF share and its Net Asset Value (NAV), thus presenting an arbitrage opportunity. APs and arbitragers will seek to take advantage of arbitrage opportunity. The mechanism
through which ETFs’ activity may introduce noise into the underlying securities is explored in several studies. Malamud (2015) develops a model for ETFs in which APs create and redeem ETF shares. He shows that the creation/redemption mechanism propagates temporary liquidity shocks into the underlying securities. The model also shows that as the liquidity of the underlying securities increases, the degree of shock propagation increases. The topic of stock co-movement with indices have received significant attention from the academic research community. Kaul, Mehrotra, & Morck (2002) and Wurgler & Zhuravskaya (2002) all has found that adding a stock to an index affects its price. Also, Goetzmann & Massa (2003) and Barberis et al. (2005) found that correlation between the newly added stocks and other stocks in the index increases for stocks in the S&P 500 index. Da & Shive (2017) found that the higher the total ETF ownership of a stock, the more it co-moves with the market in subsequent month.
We present descriptive statistics for FTSE 100 and FTSE 250 Index ETFs in Table 1, to illustrate the growing importance of ETFs in the ownership of common stocks. Due to the growth of popularity of this asset class, ETF ownership of individual stocks has increased dramatically over the last decade. In this study, to avoid any discontinuity that could be caused by index addition and subtraction, we chose to focus on are the “plain vanilla” ETFs
basket, we chose all the plain vanilla FTSE 100 and FTSE 250 ETFs that are traded on the London Stock Exchange (LSE), and denominated in British Pound Sterling (GBP). We purposely chose to not include exotic ETFs, such as leveraged and inverse leveraged ETFs that use derivatives to replicate the performance of the asset they aim to track, because these funds do not physically hold the stocks, thus not useful for our purposes. For FTSE 100 stocks, the average fraction of a stock’s capitalization held by our sample ETFs has risen from 0.07% in 2004 to 0.5% in 2017.
ETFs are similar to open-ended mutual funds, retail and institutional investors can trade ETF shares in the secondary market, and new ETF shares can be created and redeemed by APs. The price of ETF shares is determined by the demand and supply in the market, and it is possible for the price to diverge from the NAV of the underlying securities. The divergence of ETF share price from its NAV gives market participants an arbitrage opportunity to exploit, and in turn is an essential part of the price discovery for both the ETF shares, and the underlying securities. A transaction of ETF shares can take place in two main types of scenarios. First, an AP can create new ETF shares by transferring the securities underlying the ETF to the ETF sponsor. These transactions constitute the primary market for ETFs. Similarly, the AP can redeem ETF shares and receive the underlying securities in exchange. Second, once the ETF shares are created by the AP, these shares are freely traded on the secondary market, usually on the exchanges where they are listed.
Because arbitrage opportunities does exist in the ETF markets, market participants that are capable of executing these trades will attempt to take the opportunity to generate a profit. In order to illustrate the arbitrage process through the creation and redemption of ETF shares, we describe the two scenarios where an arbitrage opportunity could exist. (i) ETF premium (when the price of the ETF exceeds the NAV) and (ii) ETF discount (when the ETF price is below the NAV). In the case (i), APs can buy the cheaper underlying securities from the open market, submit them to the ETF sponsor, and ask for new ETF shares in exchange. Then the AP sells the new ETF shares on the open market for profit. This process puts downward pressure on the ETF price because of increase in supply of the ETF shares, reducing the premium. In the second case of an ETF discount, APs buy ETF shares in the open market and redeem them for the underlying securities from the ETF sponsor. Then the
APs can sell the securities for a profit. This generates positive pressure on the price of ETF shares and reduces the discount.
Arbitrage is also possible for market participants who are not APs, and without the creation and redemption of ETF shares. Since the underlying securities and ETFs are both traded on an exchange, investors can buy the cheaper asset and short sell the more expensive one. In the case of an ETF premium, traders buy the cheaper underlying securities and short sell the more expensive ETF. They hold the positions until prices converge, and then close down the positions to realize the arbitrage profit, the opposite process in the case of an ETF discount. Arbitrage activity is possible because ETF sponsors disseminating NAV values every 15 seconds throughout the trading day, and smooth functioning of arbitrage results in the low tracking error of these instruments. With trading costs lowering and market information being widely available and accessible, Marshall, Nguyen, and Visaltanachoti (2010) have found that arbitraging ETFs against the NAV has become a popular strategy with hedge funds and high-frequency traders.
3.
Theoretical Framework
In our study, we use Greenwood’s (2005) stock arbitrage model with risk-averse market maker to explain the method of shock transmission. The model yields the prediction that following a positive demand shock, the initial change in price is necessary so that
arbitrageurs who absorb the demand shock have positive returns following the event, and the prices will revert over time. Greenwood uses the redefinition of the Nikkei 225 Index for the event study, where a large group of securities received simultaneous demand side shock. This type of demand shock is similar to that of an ETF’s arbitrage process. If we have the ETF premium scenario, the arbitrageurs will buy the ETF’s underlying securities on the open market simultaneously, and bring the basket of securities to the AP to exchange for shares of that ETF, then sell the ETF shares on the open market for profit.
Applied to our investigation, the market makers in the model are APs and arbitrageurs in the ETF market. We then apply the model to two similar assets with the same
fundamentals, the ETF and the underlying securities that has the same NAV of the ETF. To graphically show the method of shock transmission, see Figure 1. Starting at Figure 1a, we create the situation where the ETF price and NAV are the same as the underlying securities, shown in Figure 1a. Then we introduce a non-fundamental shock, such as an exogenous increase in demand in the ETF market. This type of shock could be that a large institutional investor receives a large inflow, and they need to scale up their already existing ETF
allocations in their portfolio. Next, the arbitrageurs absorb the liquidity demand shock by shorting the ETF, which they require compensation for the negative inventory of ETFs they are taking on because the arbitrageurs are risk-averse, and consequently the price of the ETF has to rise, as shown in Figure 1b. At the same time, the arbitrageur needs to hedge their short ETF position, by taking a long position in the ETF’s underlying securities. To compensate the risk they are taking on in this position, the arbitrageur require
compensation, and the price of the basket of underlying securities will rise as a result, as shown in Figure 1c. In the end, when other sources of liquidity replenishes, prices for both the ETF and underlying securities revert back to fundamentals, as shown in Figure 1d.
Further studying Greenwood’s (2005) model, the long and short trades described in Figures 1b and 1c happen simultaneously. Since the two assets with identical payoffs will always end up having the same price, there cannot be discrepancy between the ETF price and NAV at any time. Greenwood’s model does describe the mechanism of the liquidity shock we are trying to study, but we need to create an extension to this model to capture the reality that ETF price and NAV can diverge for a short period of time.
Cespa and Foucault (2014) provides an useful extension we can apply to our existing framework in our study. The assume there are three types of traders in the market: (i) liquidity demanders who submit market orders in one of two markets they participate in, and two types of liquidity suppliers: (ii) market makers who specialize in one asset class, and (iii) cross-market arbitrageurs who trade in both markets. Arbitrageurs are the traders that respond to the mispricing of the assets in the two markets. To apply to the framework establish previously in our study, we establish a scenario which trades occur sequentially. The sequence begins with a liquidity shock to the ETF, and then the market makers respond and absorb the shock by way of a price adjustment. Next is followed by the market makers of the underlying securities observing the price realization of the ETF and adjusting their
own price. At the same time, cross-market arbitrage trades occur and price converges between the two assets as a result.
The mechanism above gives the possibility of an alternative scenario where price discovery happens after a shock to the assets’ fundamentals. To illustrate this scenario of a
fundamental shock, we propose that price discovery occurs in the ETF market first because it is more liquid. When new fundamental information hits the market, as illustrated in Figure 2b, ETF prices will adjust accordingly first, shown in Figure 2c, and the underlying security prices adjustment follows with a delay, as in Figure 2d.
So far we have established two scenarios of liquidity shock propagation mechanism. However, we need to determine whether or not the liquidity shock propagation actually takes place. The main difference in the two scenario of liquidity shock propagation is that if the shock is a non-fundamental one, illustrated by Figure 1, will be followed by a reversal in asset prices, while the shock in alternative scenario is a fundamental one, illustrated by Figure 2, and is followed by price discovery. In the second part of our regression analysis, we will test for price reversal after arbitrage activity.
One potential challenge for our study is an alternative hypothesis that is, if ETFs were not available for the investor, they would simply trade the basket of underlying securities, and
ETFs are only a vehicle through which the same investor would trade these assets. However, Amihud and Mendelsons (1986) and Atkins and Dyl’s (1997) studies regarding bid-ask
spread and Constantinides’ (1986) study regarding transaction costs yielded some important conclusions. The liquidity of an asset is translated to its bid-ask spread, where more liquid assets have smaller bid-ask spreads and vice versa. Amihud and Mendelson’s model predicts that higher spread assets yield higher expected returns, and there is a clientele effect where investors with longer holding periods select assets with higher spreads. Conversely,
investors with shorter holding periods will self-select into the assets with the lowest trading costs. Moreover, Atkins and Dyl’s model predicted that investors’ average holding periods are an increasing function of the bid-ask spread, confirmed the prediction using common stocks listed on the NYSE and Nasdaq, and concluded that securities with lower bid-ask spreads have higher trading volume.
Building upon these previous studies, we can conceive an assumption that ETFs do attract a new type of investors, who has shorter holding periods, and would not trade the underlying securities because of trading costs and illiquidity. Operating under the assumption, we can reasonably predict that higher ETF ownership is related to higher turnover of the underlying securities, and we will attempt to test that hypothesis in our regression analysis.
4.
Hypothesis & Research Questions
Based on the theoretical framework, we establish a main testable hypothesis, and use three research questions to design quantitative tests towards testing the main hypothesis.
Main Hypothesis: The arbitrage between ETFs and their underlying securities propagates liquidity shocks from the ETF side to the prices of the underlying securities, therefore, non-fundamental volatility of the underlying securities increases due to ETF ownership.
To test the main hypothesis, the three main research questions are as follows: i. To what extent does ETF ownership increase volatility of the underlying? ii. To what extent does ETF ownership increase turnover of the underlying? iii. To what extent does ETF trading add noise to the price of the underlying?
In order to sufficiently address the research questions, we need to collect data on the ETFs of interest, and trading information for those ETF’s underlying securities. The context of the study is geographically limited to the UK market, and plain vanilla ETFs that track the FTSE 100 and FTSE 250 Indices. We use London Stock Exchange (LSE) data to first identify all ETFs traded on the exchange. Next, we exclude leveraged and short ETFs, and filter the list down to only equity ETFs. From the list of equity ETFs, we select the funds that track the FTSE 100 and FTSE 250 Indices using physical replication only. We gather time-series data for equity and ETF trading including market cap, volume by value, number of shares outstanding, price, and we use Thomson Reuters DataStream as our primary data source for both ETF data and equity data.
To construct a quantitative test for research question (i) and (ii), we construct two separate OLS regressions, using the same set of explanatory variables. The first test is to explore whether there are any relationships between ETF ownership of a stock index (defined as the percentage of the market capitalization held by ETFs), and turnover of the stock index. While the second test will be conducted to test for the relationship between ETF ownership and volatility of the stock index.
ETF ownership of FTSE 100 stocks at time t is defined as the sum of the dollar value of all ETFs tracking this index divided by the stock index’s capitalization. In formula form:
𝐸𝐸𝐸𝐸𝐸𝐸 𝑂𝑂𝑂𝑂𝑂𝑂𝑂𝑂𝑂𝑂𝑂𝑂ℎ𝑖𝑖𝑖𝑖𝐹𝐹𝐹𝐹𝐹𝐹𝐹𝐹100,𝑡𝑡 = ∑𝑗𝑗=1 𝐽𝐽 𝐴𝐴𝐴𝐴𝐴𝐴
𝑗𝑗,𝑡𝑡
𝐴𝐴𝑀𝑀𝑀𝑀𝑀𝑀𝑀𝑀𝑡𝑡𝑀𝑀𝑀𝑀𝑀𝑀𝐹𝐹𝐹𝐹𝐹𝐹𝐹𝐹100,𝑡𝑡 (1)
Where j is number of ETFs replicating the FTSE 100 Index, and AUM is asset under management of ETF j. We will do the same for the FTSE 250 Index to construct the ETF ownership measure.
In our regressions, to ensure that our results are driven by exogeneous variation in ETF ownership, the ownership measure is constructed by only using the physical replication ETFs so the weighting of the ETF and Indices are as close as possible, while controlling for
unobservable characteristics that are potentially correlated with our dependent variables. One endogeneity problem could potentially exist within the ETF ownership variable in our OLS regressions. Because smaller market cap stocks tend to have higher volatility, which is one of the main dependent variables in our regression analysis, the negative relation could produce a positive relation by design between ownership and volatility. To avoid this problem, we include controls for stocks total market cap in all of our regressions.
To construct a quantitative test for research question (iii), we again will use an OLS
regression, and focus on exploring whether there is mean reverting component of the stock price after a demand shock from the ETF side. To quantify this ETF demand shock, the best measure is the fund flows of whole basket of similar ETFs. Information on fund flows is not directly available on DataStream, therefore we need to compute this measure using the available data.
Here we define Fund Flows of FTSE 100 ETFs at time t as the change in shares outstanding of the ETF from the previous trading day, multiplied by the daily closing price of the ETF. To construct the explanatory variable, we sum the fund flows from each ETF, and divide by the total AUM of those ETFs. In formula form:
𝐸𝐸𝐸𝐸𝐸𝐸 𝐸𝐸𝐹𝐹𝐹𝐹𝑂𝑂𝑂𝑂𝐹𝐹𝐹𝐹𝐹𝐹𝐹𝐹100,𝑡𝑡 = Σ𝑗𝑗=1 𝐽𝐽 �𝑁𝑁𝑁𝑁𝐹𝐹𝑁𝑁
𝑗𝑗,𝑡𝑡−𝑁𝑁𝑁𝑁𝐹𝐹𝑁𝑁𝑗𝑗,𝑡𝑡−1�∗𝑃𝑃𝑗𝑗,𝑡𝑡
Where j is number of ETFs replicating the FTSE 100 Index, NOSH is the number of shares outstanding of the ETF j, and P is the price of the ETF j, and AUM is asset under
management of ETF j. We purposely do the summation separately in both the numerator and denominator because we want to look at the overall fund flows with respect to all of the ETFs covering the same index. We will do the same to construct the ETF Flows measure for funds covering the FTSE 250 index.
5.
Data
Using trading data collected from Thompson Reuters DataStream, our sample contains 3392 daily observations for a total of 9 ETFs between 2005/1/1 to 2017/12/31. Table 1 presents descriptive statistics for ETF ownership of the stocks in the index, and the AUM of the ETFs. The table shows that ETF ownership of the stocks have increased dramatically from 0.06% of the entire market cap of the FTSE 100 index, to 0.45% by the end of 2017.
Year Average number of ETF Average ETF AUM (millions GBP) Average ETF ownership (%)
FTSE 100 FTSE 250 FTSE 100 FTSE 250 FTSE 100 FTSE 250 2005 1 1 836.31 87.67 0.0658 0.0426 2006 1 1 1237.1 189.11 0.0849 0.0754 2007 1.33 2 1609.36 317.09 0.1047 0.1187 2008 2 3 1780.9 246.89 0.1441 0.1357 2009 2.67 3.33 3095.6 240.56 0.2696 0.1555 2010 4.33 4.33 3487.46 384.79 0.2472 0.1842 2011 5 5 3636.6 520.40 0.2462 0.2323 2012 6 5 3686.84 487.38 0.2508 0.2165 2013 8 5 4827.59 913.45 0.2914 0.3266 2014 8 5.67 6087.35 1206.02 0.3597 0.3920 2015 8 6 6737.83 1249.21 0.4008 0.3646 2016 8 6 6928.04 1140.66 0.4168 0.3388 2017 9.13 6 8397.61 1580.98 0.4457 0.4092 Average 4.96 4.10 4026.81 658.79 0.2560 0.2302 Table 1. ETF ownership statistics for the FTSE 100 and FTSE 250 indices.
While processing our raw data, we use the number of shares outstanding of ETFs, which is reported daily, to calculate the value of each ETF, then aggregate them to form a total measure of asset under management (AUM) for the basket of ETFs in our sample. We restrict our sample to the basket of plain vanilla ETFs that only invest in UK stocks, because the sample will not encounter a stale pricing issue if the funds have global equity, so the price of the ETF shares will be the same or very close to the NAV. Then, for every ETF in our sample, we sum the AUM to create a time-series of ETF ownership measure on a daily frequency.
In the first of our two tests, the focus is to test to what extent does ETF ownership have on stock volatility and turnover. We directly have daily turnover data that we can match with our constructed ETF ownership measure. Then, we compute the stock volatility measure by using the daily returns of the FTSE 100 index for the same time period. We want to test the volatility effect of ETF ownership in the short-term and long term. Since returns data is posted in a daily frequency, we build a rolling series of volatility over the next 3 days (t+3) to test the volatility effect in the short term. As for the long term, we will use the average ownership measured within the month, and the index volatility of that month to measure the long term effect. Repeat to build the same variables for the FTSE 250 universe.
The second test focuses on price reversal, and to identify price reversal, we need to first compute the daily returns of the stock index. Once we have identified the first price/return co-movement, we extend the time horizon to identify the mean reverting portion of the initial price movement.
Based on the variable definitions established in section 4, we present descriptive statistics for all of the variables we will use in our regression analysis, Table 2 reports summary statistics for variables used to identify the volatility and turnover effects, and Table 3 reports summary statistics for variables used to identify price reversal in our sample ETF basket.
FTSE 100 N Mean STDEV Minimum Median Maximum
Volatility T+3 3388 0.87% 0.76% 0.00% 0.67% 8.88% Daily ETF Turnover % 3390 0.96 0.83 0.03 0.88 19.63 Daily ETF Ownership 3390 0.2560% 0.1239% 0.0628% 0.2555% 0.4870% Monthly Volatility 156 0.9789% 0.6009% 0.2682% 0.8543% 4.8537% Monthly Avg. Ownership 156 0.2559% 0.1243% 0.0631% 0.2546% 0.4744%
Table 2a. Descriptive statistics for variables used in FTSE 100 volatility and turnover effect regressions.
FTSE 250 N Mean STDEV Minimum Median Maximum
Volatility T+3 3388 0.84% 0.69% 0.00% 0.65% 6.27% Daily ETF Turnover % 3390 0.87 1.01 0.01 0.60 14.38 ETF Ownership 3390 0.2297% 0.1234% 0.0364% 0.2159% 0.4570% Monthly Volatility 156 1.1118% 1.8178% 0.3144% 0.7716% 22.5924% Monthly Avg. Ownership 156 0.2301% 0.1237% 0.0369% 0.2168% 0.4523%
FTSE 100 N Mean STDEV Minimum Median Maximum
ETF Flows % 3370 0.01 0.15 -4.07 0.00 4.07 Index Returns t 3370 0.0124% 1.1404% -9.2656% 0.0097% 9.3843% Index Returns t+20 3370 0.2619% 4.3114% -32.0329% 0.7413% 15.5801%
Table 3a. Descriptive statistics for variables used in FTSE 100 regressions for identifying price reversal.
FTSE 250 N Mean STDEV Minimum Median Maximum
ETF Flows % 3370 0.01 0.15 -1.68 0.00 5.09 Index Returns t 3370 0.0331% 1.1271% -7.4565% 0.0871% 7.4620% Index Returns t+20 3370 0.6499% 4.8858% -35.6424% 1.0887% 17.6788%
6.
Regression Tests and Results
6.1 ETF Ownership and its effect on Volatility and Turnover
The focus of our tests is whether ETF ownership leads to an increase in the volatility of the underlying securities. The first source of identification is the variation in ETF ownership of FTSE 100 and FTSE 250 Indices over time.
We can interpret from Equation (1) that, the variation of ETF ownership in stocks over time primarily comes from the ETFs’ asset under management. Since we have chosen to analyze plain vanilla ETFs that physically replicate the FTSE 100 and 250 Indices, there should not be any variations in weighting schemes within our basket of sample ETFs with respect to the index they are tracking. Also, we are doing this analysis under the assumption that the variation in ETF ownership is exogeneous with respect to our dependent variables of stock volatility and turnover. It is possible to argue that investor demand for ETFs, which in turn translates to AUM, could be related to fundamental information, which also would affect volatility and turnover. However, the basket of ETFs in our analysis are passive index ETFs, not actively managed funds. The weighting scheme of each fund is the same as the index, thus the way these AUM translate into demand for the index constituent stocks is
exogenous. Given the above considerations, we believe our assumptions are well-based.
Turnover
We begin by looking at whether ETF ownership has any impact on intraday turnover, which is the highest frequency of the data we are able to gather. From what we established in the theoretical framework, the hypothesis of liquidity shock propagation also implies that the basket of ETF securities will attract high-turnover clientele. To test this prediction, we first build our dependent variable of turnover and control it for size, dividing the total turnover of FTSE 100 stocks by the total market cap on the same day. Then, we regress our
dependent variable on ETF ownership, as well as adding stock-level controls for size and liquidity. We use the inverse of price as control for liquidity, and the logarithm of total stock market cap as control for size.
𝐼𝐼𝑂𝑂𝐼𝐼𝑂𝑂𝐼𝐼𝐼𝐼𝐼𝐼𝐼𝐼 𝐸𝐸𝑇𝑇𝑂𝑂𝑂𝑂𝐹𝐹𝑇𝑇𝑂𝑂𝑂𝑂 %𝑡𝑡= α +𝛽𝛽1𝐸𝐸𝐸𝐸𝐸𝐸 𝑂𝑂𝑂𝑂𝑂𝑂𝑂𝑂𝑂𝑂𝑂𝑂ℎ𝑖𝑖𝑖𝑖 %𝑡𝑡+ 𝛽𝛽2log(𝑚𝑚𝐼𝐼𝑂𝑂𝑚𝑚𝑂𝑂𝐼𝐼𝑚𝑚𝐼𝐼𝑖𝑖𝐹𝐹𝐹𝐹𝐹𝐹𝐹𝐹100) + 𝜀𝜀 (3)
We run the regression in the form of Equation (3), and the results are presented in the appendix Table 4. The regression yields a positive and significant relation between ETF ownership percentage and turnover.
𝛽𝛽1= 3.039 (0.187) 3.039 ∗ 0.1239
0.83 = 0.453 ≈ 45%
The regression coefficient shows that with a one standard deviation increase in ETF ownership of 0.124%, turnover increases by 45% of a standard deviation, which translates to an increase of 0.38% of total daily turnover of the underlying securities. We get the same result of a significant and positive relationship between ETF ownership and daily turnover when we test in the FTSE 250 universe, where one standard deviation increase in ETF ownership translate to 51% 1of one standard deviation increase in daily turnover of the
underlying. By the regression estimate, the effect appears to be economically significant, and does support the view that ETFs attract high-turnover clientele, and the trading of underlying securities increases as ETF ownership increases.
Volatility
Next, we look at whether ETF ownership has any impact on volatility of the underlying, and test our main hypothesis that increased ETF ownership leads to increased volatility of its underlying securities. We collected daily price data for the FTSE 100 Index and from that we computed the daily returns for the index. To explore the short term effect on volatility of the underlying, we test the relationship between daily ETF ownership and the volatility of the underlying over the next three trading days. From the daily returns data, we build a rolling time series for volatility of the FTSE 100 Index over the next three days as the dependent variable in this regression test, and use the same controls for size and liquidity.
𝑉𝑉𝐹𝐹𝐹𝐹𝐼𝐼𝐼𝐼𝑖𝑖𝐹𝐹𝑖𝑖𝐼𝐼𝐼𝐼𝑡𝑡+3= α + 𝛽𝛽1𝐸𝐸𝐸𝐸𝐸𝐸 𝑂𝑂𝑂𝑂𝑂𝑂𝑂𝑂𝑂𝑂𝑂𝑂ℎ𝑖𝑖𝑖𝑖%𝑡𝑡+ 𝛽𝛽2log (𝑚𝑚𝐼𝐼𝑂𝑂𝑚𝑚𝑂𝑂𝐼𝐼𝑚𝑚𝐼𝐼𝑖𝑖)𝑡𝑡+ 𝜀𝜀 (4)
The results of the regression shows a positive and significant relationship between ETF ownership and short term volatility of the underlying securities. The results of the regression from Equation (3) is presented in appendix Table 4.
𝛽𝛽1= 2.595 (0.179)
2.595 ∗ 0.1239
0.766 = 0.421 ≈ 42%
The regression coefficient shows that with one standard deviation increase in ETF ownership, the short term volatility of the underlying securities increases by 42% of one standard deviation, which translates to a 32 bps increase in short term volatility of the underlying. The same positive and significant relationship appears in the test for the FTSE 250 universe, but the magnitude of the effect is smaller. In the FTSE 250 universe, one standard deviation increase in ETF ownership is accompanied with 15% of one standard deviation increase and 10 bps increase in short term volatility. The relationship appears to be economically significant, and aligns with our main hypothesis as well.
Now that we showed a significant relationship between ETF ownership and short term volatility. It is possible that the effect on short term volatility caused by liquidity shock propagation mechanism, and ETF arbitrage activity will be washed out over a longer time horizon. To test this possibility, we study whether the same effect still exist at frequencies that are relevant to longer term investors. From our we computed daily returns of the FTSE 100 index, we construct a volatility measure at the monthly frequency. Since we are using monthly volatility as the dependent variable, we build our ETF ownership measure as the monthly average to match the time horizon, and also build the control variables in the same manner.
𝑀𝑀𝐹𝐹𝑂𝑂𝐼𝐼ℎ𝐹𝐹𝐼𝐼 𝑉𝑉𝐹𝐹𝐹𝐹𝐼𝐼𝐼𝐼𝑖𝑖𝐹𝐹𝑖𝑖𝐼𝐼𝐼𝐼 = α + 𝛽𝛽1 𝐴𝐴𝑇𝑇𝑂𝑂𝑂𝑂𝐼𝐼𝐴𝐴𝑂𝑂 𝐸𝐸𝐸𝐸𝐸𝐸 𝑂𝑂𝑂𝑂𝑂𝑂𝑂𝑂𝑂𝑂𝑂𝑂ℎ𝑖𝑖𝑖𝑖% + 𝛽𝛽2log(𝑚𝑚𝐼𝐼𝑂𝑂𝑚𝑚𝑂𝑂𝐼𝐼𝑚𝑚𝐼𝐼𝑖𝑖) + 𝜀𝜀 (5)
We run the regression test for volatility in the form of Equation (5), and the results are presented in appendix Table 5. The test reveals a positive and significant relationship between ETF ownership and volatility of the underlying securities on a monthly frequency.
2.936 ∗ 0.124%
0.60% = 0.607 ≈ 61%
Extrapolate from the regression coefficient shows that with one standard deviation increase in ETF ownership, the volatility of the underlying increases by approx. 61% of one standard deviation. The result provides supporting evidence that the effect on volatility by ETF ownership does persist and is not washed out over a longer time horizon.
However, when we test in the FTSE 250 universe, no significant relationship can be
extracted from the data. One possible explanation could be that arbitrageurs are less likely to rely on smaller stocks to replicate ETFs. Therefore, prices and in turn, volatility of small stocks are less impacted by ETF ownership in the long run.
6.2 Identifying Non-Fundamental Volatility
The result from the previous finding that higher ETF ownership is related to increased volatility does not directly mean that ETFs increase the noise in the prices of the underlying. Amihud and Mendelson (1987) has conducted a study on trading mechanisms and stock returns, and provides a model to explain that trading prices is positively related to the speed of prices adjustment to fundamentals. This model generates the prediction that if ETF arbitrage activity makes prices adjust faster to fundamentals, then the fundamental volatility of the underlying security increases. The hypothesis we are testing in this paper focuses on non-fundamental volatility, which is different from the prediction generated by Amihud and Mendelson’s model.
Here we want to test whether ETFs add noise to the prices of the underlying securities, and we construct a test that focuses price reversals to explore this further. Previously
established in theoretically frameworks, we argued that if the initial price movement caused by the shock is reversed, then the trigger is a non-fundamental shock, as depicted in Figure 1. Otherwise, if prices stays at the new level after the initial shock, or revert only partly, then the shock contains new fundamental information, as depicted in Figure 2. We will try to distinguish these two scenarios, and test whether ETF arbitrage causes a mean-reverting component in underlying stock prices.
For the price reversal test, we follow a similar test conducted by Ben-David, Franzoni and Moussawi (2014), and use ETF flows at the daily frequency to attempt to identify price reversals in the underlying. ETF flows is defined as the creation and redemption of ETF shares, and are the results of APs’ arbitrage activity, which is essential to keep tracking error of the ETFs to a minimum. For our basket of ETFs, we collected the number of shares
outstanding, and price data at the daily frequency. To construct ETF flows as a conditioning variable, we define ETF flows as a fraction of the previous day asset under management. On the day which the flow occurs, we expect a price move that is in the same direction as the flows. If there is a large inflow, we expect the price to move higher and vice versa,
regardless of the reason for the trade is fundamental or non-fundamental. If at least part of the shock is non-fundamental, we should see a price reversal in the next coming days. In order to capture this phenomenon, we regress the stock index returns at different time horizons on the ETF flows, and include the same controls for size.
𝑅𝑅𝑂𝑂𝐼𝐼𝑇𝑇𝑂𝑂𝑂𝑂𝑂𝑂𝑡𝑡= α + 𝛽𝛽1 𝐸𝐸𝐸𝐸𝐸𝐸 𝐸𝐸𝐹𝐹𝐹𝐹𝑂𝑂𝑂𝑂𝑡𝑡+ 𝛽𝛽2log(𝑚𝑚𝐼𝐼𝑂𝑂𝑚𝑚𝑂𝑂𝐼𝐼𝑚𝑚𝐼𝐼𝑖𝑖) + 𝜀𝜀 (6)
The results of the regression of Equation (6) does not yield any significant relationship between ETF flows and same day returns, the 𝛽𝛽1regression coefficient is very close to 0 and
not statistically significant. In fact, when we run regressions using returns over the next 20 trading days to test for indications of price reversal, the test result implies that ETF flows does not have any impact on price movement or reversal. The result is not consistent with the prediction produced by our hypothesis that ETF flows will cause a price movement in the same direction.
This unexpected result could be due to a few reasons. First of all, the basket of ETFs in our sample are passive index funds, the flows in and out of passive funds may not reflect fundamental information. Moreover, the relative size of the UK ETF market is much smaller compared to the U.S. market, and the size of the flows are not large enough to create significant momentum in price of the underlying, and further exploration is recommended.
6.3 Possible Extensions for Future Research
Although this analysis found a significant relationship between ETF ownership and short term volatility of the underlying, it may be possible to expand the scope of the research and gain additional explanatory power by extending the existing structure.
First of all, we could build the ETF ownership measure from the single stock level, and extend the universe to all stocks traded on the LSE. The goal of this method is to capture all ETF holdings of a stock, and construct the ETF ownership measure from the stock side. The basket of ETFs in our study represents approx. 82% of total AUM held by ETFs in the UK market. Because stocks are often held by more than one index (eg. FTSE 100 Index and FTSE All-Shares Index), we could be potentially missing part of the explanatory power by omitting the ETFs tracking other indices.
Moreover, we can employ a regression discontinuity design as a different identification strategy based on research done by Chang, Hong and Liskovich (2013). This identification strategy exploits a mechanical rule that allocates stocks to the FTSE 100 and FTSE 250 Indices. Since both of these indices are capitalization weighted, the top stocks in the FTSE 250 index will receive much higher amount of passive indexing flows than the smallest stocks in the FTSE 100 index. We can conduct an event study and explore when a stock switches to either index, there will be large amount of exogeneous variation in ETF ownership of that particular stock.
7.
Conclusion
In the current trend of cost saving in the investment industry, ETFs have been rapidly
gaining momentum among both institutional and retail investors, as ETFs provide a low cost way to build a well-diversified portfolio. Moreover, this study does point out an important relationships between ETF ownership and the volatility of FTSE 100 constituent stocks.
Our results clearly show that arbitrage activity between ETFs and their underlying securities leads to an increase in volatility. More specifically in the case of FTSE 100 Index stocks, one standard deviation increase in ETF ownership (0.124%), is associated with approx. 32 bps of increased volatility of the underlying. Liquidity shock from the ETF market due to the ETF arbitrage process does appear to propagate to its underlying securities in the short term, causing an increase in underlying volatility. This increase in volatility appears to be very economically significant, and should be taken into consideration when analyzing the quality of these securities.
However, we were not able to detect any meaningful relationship between ETF fund flows and index returns. This could be because the UK ETF market is relatively smaller in terms of the world ETF market, and funds traded in other places could also include UK stocks. For example, HSBC.L and RDSA.L, two of the largest stocks trading on the LSE, is also held by large ETFs trading in the U.S. market. The resulting fund flows from only the UK market may not be enough to cause any return co-movement to the magnitude that generates a clear and testable pattern. The main hypothesis of this study is only partly confirmed, further extensions described in section 6.3 can be a useful starting to identifying non-fundamental volatility caused by ETF arbitrage trading in more granular detail.
References
Amihud, Y., & Mendelson, H. (1986). Asset pricing and the bid-ask spread, Journal of
Financial Economics, 17, 223–249.
Amihud, Y., & Mendelson, H. (1987). Trading mechanisms and stock returns: An empirical investigation, Journal of Finance 42(3), 533–553.
Atkins, A. & Dyl, E. (1997). Transactions Costs and Holding Periods for Common Stocks,
Journal of Finance 52 (1), 309–325.
Bae, K. H., Kang, J. K., & Wang, J. (2013). The costs of ETF membership: The valuation effect of ETFs on underlying firms. Working Paper.
Barberis, N., Shleifer, A., & Wurgler, J. (2005). Comovement. Journal of Financial Economics, 75, 283–317.
Ben-David, I., Franzoni, F. A., & Moussawi, R. (2014). Do ETFs Increase Volatility?. Working Paper, National Bureau of Economic Research
Ben-David, I., Franzoni, F. A., & Moussawi, R. (2016). Exchange Traded Funds (ETFs). Working Paper.
Cespa, G. & Foucault, T. (2012). Illiquidity shock propagation and liquidity crashes, Working paper, HEC Paris.
Chang, Y., Hong H, & Liskovich I., (2013). Regression Discontinuity and the Price Effects of Stock Market Indexing, NBER working paper 19290.
Constantinides, M. (1986). Capital Market Equilibrium with Transaction Costs, Journal of
Political Economy 94(4), 842–862.
Da, Z., & Shive, S. (2017). Exchange traded funds and asset return correlations. European
Financial Management, 24, 136-168
Greenwood, R. (2005) Short- and long-term demand curves for stocks: Theory and evidence on the dynamics of arbitrage. Journal of Financial Economics 75(3), 607–649.
Goetzmann, W., & Massa, M. (2003). Index funds and stock market growth. Journal of
Business, 76, 1–28.
Hegde, Shantaram P., McDermott, John B. (2004). The market liquidity of DIAMONDS, Q’s and their underlying stocks. Journal of Banking and Finance 28, 1043-1067.
Israeli, Doron, Charles M. C. Lee, & Suhas Sridharan (2015). Is there a Dark Side to Exchange Traded Funds (ETFs)? An Information Perspective, Working Paper, Stanford University. Kaul, A., Mehrotra, V., & Morck, R. (2002). Demand curves for stocks do slope down: New
Lee, C. M. C., and E. So. 2015. Alphanomics: The Informational Underpinnings of Market Efficiency. Foundations and Trends in Accounting 9 (No. 2-3): 59-258.
Marshall, B., Nhut H. Nguyen & Visaltanachoti, N. (2010) ETF arbitrage, Working paper, Massey University.
Madura, J., & Ngo, T. (2008). Impact of ETF inception on the valuation and trading of component stocks. Applied Financial Economics, 18(12), 995-1007.
Malamud, Semyon, (2015). A Dynamic Equilibrium Model of ETFs, Swiss Finance Institute Working Paper.
Pedersen, L. H. (2015). Efficiently Inefficient: How Smart Money Invests & Market Prices Are
Determined. Princeton University Press: Princeton NJ.
Wurgler, J., & Zhuravskaya, E. (2002). Does arbitrage flatten demand curves for stocks?
Appendix
Table 4. ETF Ownership, Intraday Stock Turnover, and Short-term Volatility
This table reports the OLS regression estimates for Equations (3) and (4), as described in Section 6.1. Standard errors are reported within the parentheses, *, **, *** represent statistical significance at the 10%, 5%, 1% level.
Dependent Variable: Dependent Variable: 𝐼𝐼𝑂𝑂𝐼𝐼𝑂𝑂𝐼𝐼𝐼𝐼𝐼𝐼𝐼𝐼 𝐸𝐸𝑇𝑇𝑂𝑂𝑂𝑂𝐹𝐹𝑇𝑇𝑂𝑂𝑂𝑂𝑡𝑡 𝑉𝑉𝐹𝐹𝐹𝐹𝐼𝐼𝐼𝐼𝑖𝑖𝐹𝐹𝑖𝑖𝐼𝐼𝐼𝐼𝑡𝑡+3
(i) (ii) (iii) (iv) FTSE 100 FTSE 250 FTSE 100 FTSE 250 𝐸𝐸𝐸𝐸𝐸𝐸 𝑂𝑂𝑂𝑂𝑂𝑂𝑂𝑂𝑂𝑂𝑂𝑂ℎ𝑖𝑖𝑖𝑖 %𝑡𝑡 3.039*** (0.187) 4.188*** (0.179) 2.595*** (0.1793) 0.837*** (0.125)
𝐹𝐹𝐹𝐹𝐴𝐴(𝑚𝑚𝐼𝐼𝑂𝑂𝑚𝑚𝑂𝑂𝐼𝐼𝑚𝑚𝐼𝐼𝑖𝑖)𝑡𝑡 0.063*** (0.016) -0.008*** (0.002) -0.209*** (0.015) -0.028*** (0.001)
N 3390 3390 3388 3Í388 Adjusted R2 0.268 0.201 0.209 0.163
Table 5. ETF Ownership and Monthly Stock Volatility
This table reports the OLS regression estimates for Equation (4), as described in Section 5.1. Standard errors are reported within the parentheses, *, **, *** represent statistical
significance at the 10%, 5%, 1% level.
Dependent Variable: 𝑀𝑀𝐹𝐹𝑂𝑂𝐼𝐼ℎ𝐹𝐹𝐼𝐼 𝑉𝑉𝐹𝐹𝐹𝐹𝐼𝐼𝐼𝐼𝑖𝑖𝐹𝐹𝑖𝑖𝐼𝐼𝐼𝐼𝑡𝑡 (i) (ii) FTSE 100 FTSE 250 𝐴𝐴𝑇𝑇𝑂𝑂𝑂𝑂𝐼𝐼𝐴𝐴𝑂𝑂 𝐸𝐸𝐸𝐸𝐸𝐸 𝑂𝑂𝑂𝑂𝑂𝑂𝑂𝑂𝑂𝑂𝑂𝑂ℎ𝑖𝑖𝑖𝑖 %𝑡𝑡 1.435*** (0.384) (1.529) 0.849 𝐹𝐹𝐹𝐹𝐴𝐴(𝑚𝑚𝐼𝐼𝑂𝑂𝑚𝑚𝑂𝑂𝐼𝐼𝑚𝑚𝐼𝐼𝑖𝑖)𝑡𝑡 -0.065*** (0.074) (0.010) -0.002 N 156 156 Adjusted R2 0.346 0.002
Table 6. Price Reversal Test
This table reports the OLS regression estimates for price reversal tests, as described in Section 6.2. Standard errors are reported within the parentheses, *, **, *** represent statistical significance at the 10%, 5%, 1% level.
Dependent Variable: Dependent Variable: 𝑅𝑅𝑂𝑂𝐼𝐼𝑇𝑇𝑂𝑂𝑂𝑂𝑂𝑂𝑡𝑡 𝑅𝑅𝑂𝑂𝐼𝐼𝑇𝑇𝑂𝑂𝑂𝑂𝑂𝑂𝑡𝑡+20
(i) (ii) (iii) (iv) FTSE 100 FTSE 250 FTSE 100 FTSE 250 𝐸𝐸𝐸𝐸𝐸𝐸 𝐸𝐸𝐹𝐹𝐹𝐹𝑂𝑂 %𝑡𝑡 (0.0001) 0.0001 (0.1299) -0.0020 (0.0005) -0.0004 (0.5618) 0.2972
𝐹𝐹𝐹𝐹𝐴𝐴(𝑚𝑚𝐼𝐼𝑂𝑂𝑚𝑚𝑂𝑂𝐼𝐼𝑚𝑚𝐼𝐼𝑖𝑖)𝑡𝑡 0.0077** (0.0030) (0.0017) 0.0022 -0.0772*** (0.0114) -0.02833*** (0.0071)
N 3370 3370 3370 3370 Adjusted R2 0.0021 0.0006 0.0134 0.0049
