Closing Auction, Passive Investing, and Stock Prices
Yanbin Wu
Current draft: November 11, 2019
Over the last decade, the volume of market-on-close orders has increased to more than 10% of the
entire day’s trading volume. This paper investigates this rise and documents four stylized facts:
(i) passive investing leads to greater usage of market-on-close orders, consistent with passive
fund’s motivation for minimizing tracking error; (ii) the price impact from large market-on-close
order imbalances is economically large and transitory, leading to short-term price reversal; (iii)
a long/short trading strategy exploiting this reversal results in a significant risk-adjusted return
of 13.2 basis points per day, consistent with the hypothesis that investors are compensated by
providing liquidity to passive funds; and (iv) informed traders also use market-on-close orders,
consistent with Admati and Pfleiderer’s (1988) prediction that liquidity trades attract informed
trades. Overall, the set of findings demonstrates market-on-close order as an important trading
channel through which passive investing affects underlying stocks.
JEL Code: G12, G14
Keywords: Closing Auction, Passive Investing, Exchange Traded Funds, Return Predictability
I am extremely grateful to my committee members Narasimhan Jegadeesh, Jay Shanken, Jeff Busse, Clifton Green, and Ilia
Dichev for their invaluable guidance and support. I also thank Tetyana Balyuk, Francisco Barillas, Lawrence Benveniste, Tarun
Chordia, Rohan Ganduri, Christoph Herpfer, William Mann, Gonzalo Maturana, and seminar participants at Emory University for
helpful comments and suggestions. Address: Goizueta Business School at Emory University, 1300 Clifton Rd, GA 30322. E-mail:
1 Introduction
The remarkable growth of passive investing over the last decade has attracted attention among practition-
ers, policy makers, and academia.
There is substantial evidence showing that passive investing has impacted
different aspects of the financial markets, such as price discovery (Glosten, Nallareddy, and Zou (2016);
Israeli, Lee, and Sridharan (2017)), corporate governance (Schmidt and Fahlenbrach (2017)), and firms’ de-
cisions (Appel, Gormley, and Keim (2016); Appel, Gormley, and Keim (2018)). In particular, Ben-David,
Franzoni, and Moussawi (2018) show that passive funds, particularly exchange traded funds (ETFs), lead to
higher nonfundamental volatility of underlying stocks. They argue that the channel through which passive in-
vesting affects the constituent stocks is intraday arbitrage activity, since short-term arbitragers would quickly
exploit any discrepancy between the net asset value (NAV) of underlying portfolios and the market price of
passive funds.
Yet, Box, Davis, Evans, and Lynch (2019) find little evidence of this intraday arbitrage trading
when they directly examine the order flow between ETFs and underlying portfolios. In this paper, I explore
the channel through which passive funds affect underlying stocks.
Trading through closing auction is one possible but overlooked channel for passive investing to affect
stock prices. A closing auction is a batch auction that occurs at the end of the trading day for setting the
closing price through market-on-close orders. These orders were implemented in the early 2000s by the U.S.
exchanges, including NYSE and Nasdaq, to facilitate price discovery and increase liquidity at the end of
the day. Unlike normal market orders and limit orders, market-on-close orders are only crossed when the
market is closing. Over the past decade, the average trading volume for market-on-close orders has increased
dramatically, from 2% of the total daily trading volume in 2010 to nearly 10% in 2018 for S&P 500 stocks
(See Figure 1).
Put differently, the transaction volume at the closing crosses (which was around $10 billion
in 2018), on average, accounts for more than 10% of the transaction volume in the entire 6.5 trading hours
from 09:30 to 16:00.
Given this rise in market-on-close orders and the coincident popularity of passive funds over the past
According to 2018 Investment Company Institute Factbook, the total assets tracking passive indices surpassed $6.7 trillion in
asset under management (AUM) in the U.S. in 2017. Among the passive investing funds in 2017, index mutual funds accounted for
$3.4 trillion, and passive ETFs accounted for $3.3 trillion.
See also Israeli, Lee, and Sridharan (2017); Broman and Shum (2018); and Da and Shive (2018).
The market-on-close trades for each stock per day can be uniquely identified by their sale condition recorded in the Trade and
Quote (TAQ) database.
decade, I address the following research questions: First, do daily passive flows lead to the increased usage
of market-on-close orders? Second, what are the implications of this rise in the market-on-close orders for
underlying stocks and investors? Third, what are the implications of the clustering of trades through market-
on-close orders for market microstructure theories (Admati and Pfleiderer (1988); Admati and Pfleiderer
There are reasons to believe that passive fund flows lead to the increased usage of market-on-close orders.
First, the most important performance metric for a passive fund is the tracking error relative to the benchmark
it follows. A market-on-close order allows a passive fund to automatically minimize tracking errors, because
the transaction prices of constituents will be exactly the same closing prices that determine the benchmark
Furthermore, the increasingly large swing of daily flows into passive funds requires that funds deploy
their capital quickly and simultaneously into hundreds of underlying baskets, and market-on-close orders
could achieve this easily.
For index mutual funds, managers could directly buy or sell stocks using market-on-close orders in the
secondary market. For ETFs, the ETF sponsors interact with the secondary market only through authorized
participants (APs) to cater to the investors’ demand. To do so, the APs will create or redeem ETF shares in
exchange for the underlying basket at the end of the day. Similar to the argument by Pan and Zeng (2019)
that bond ETF APs may use the ETF creation/redemption process to manage their inventory risk, equity
ETF APs could minimize their inventory risk by using the market-on-close orders for the end-of-day creation
and redemption settlement. However, from the perspective of transaction cost, passive funds (as institutional
investors) do care about implementation shortfall (Anand, Irvine, Puckett, and Venkataraman (2012)). Placing
market-on-close orders without knowing the execution price or the price impact exposes the passive funds to
additional market risk and transaction costs.
I start the empirical analysis by showing that flows from passive funds are strongly related to increased
usage of market-on-close orders at the stock level and daily frequency by utilizing the daily ETF flows
Besides determining the benchmark level, the closing price is also a widely reference price for other financial assets. For mutual
funds, the net asset value (NAV) is calculated at the end of the day, based on the closing prices of the underlying securities. This NAV
is then used as the share price for purchase/redemption orders received on that day. For many derivative contracts, the settlement is
tied to the closing price of the underlying securities or indices on the day of expiration.
Because daily flows from index mutual funds are not available from major data vendors, my analysis focuses on ETF flows.
However, the results documented here could be generalized to index mutual funds.
contemporaneous market-on-close volume from TAQ data. The daily ETF flow for a stock is the aggregate of
the weighted daily flow from all ETFs that hold the same stock. Under various specifications, a one standard
deviation increase in ETF flows, on average, is associated with a 23% of a standard deviation increase in the
market-on-close volume.
To further address endogeneity concerns, I use the post-2007 annual Russell 1000 and Russell 2000 in-
dex reconstitutions as exogenous shocks to passive institutional holdings (Coles, Heath, and Ringgenberg
I conduct this test using two-stage least squares specifications in which I first instrument stock-level
ETF flows as exogenous variations around 1000/2000 threshold and then test the effects of instrumented ETF
flows on market-on-close volume. Consistent with the OLS results, I find that the exogenous increase in pas-
sive flows leads to a significant increase in market-on-close volume. Specifically, the first-stage regression
shows that the index assignment from the Russell 2000 to the Russell 1000, on average, leads to a decrease of
17% of a standard deviation in ETF flows. The second-stage regression estimates exceed the OLS estimates
in magnitude.
Next, I examine the asset pricing implication of this rise of market-on-close order for underlying stocks.
An important feature of the closing auction for NYSE exchange is that at 15:45, all market-on-close orders
will be aggregated and disseminated to the public and only market-on-close orders that offset the aggregate
order imbalances will be accepted from 15:45. This dissemination of order imbalances is somewhat like
the “sunshine” trading modeled by Admati and Pfleiderer (1991), since the passive funds preannounce their
trading needs. On the one hand, according to the theory suggested by Admati and Pfleiderer (1991), market-
on-close orders might have a minimal price impact at the aggregate level, due to the information dissemination
Specifically, the Russell 1000 and the Russell 2000 indices are constructed each year based on the end-of-May market capital-
ization ranks. There are only small differences in market value around the threshold and firms cannot precisely control their rankings,
such that being assigned to the left or right of the cutoff is as good as random. Before 2007, index assignment followed a simple
threshold: Stocks ranked from 1 to 1,000 were assigned to the Russell 1000 index, while stocks ranked from 1,001 to 3,000 were
assigned to the Russell 2000 index. Starting in 2007, the Russell implemented a new assignment procedure. After the initial market
capitalization breakpoints are determined, existing members are reviewed to determine whether they fall within the the accumulated
5% market cap range around the new breakpoints. This creates an upper band and lower band around the Russell 1000/2000 index
threshold (Coles, Heath, and Ringgenberg (2017)). Because of the value-weighted difference for the Russell 1000 and the Russell
2000 (Chang, Hong, and Liskovich (2015)), stocks that move to a different index will result in significant changes in the portfolio
weights, which in turn alters passive ownership and passive flows around the threshold.
The IV estimate captures the local average treatment effect (LATE), which only focuses on stocks that are potential index
switchers. The effect is larger than the OLS effect since these stocks change status drastically from receiving significant weight
(stocks that are at the top of the Russell 2000) when replicating passive indices to being neglected (stocks that are at the bottom of
the Russell 1000), while the OLS effect captures the effect for average stock.
and the information content of these sunshine trades. On the other hand, these orders may exhibit a price
impact because the aggregated market-on-close orders placed by all passive fund managers are so substantial
that providing immediacy and liquidity from the opposite side cannot be guaranteed, much like the price
impact of block trades examined by Chan and Lakonishok (1993) and Chan and Lakonishok (1995). However,
the price impact of large market-on-close orders is different from that of the block trades placed by other
institutional investors, because this price impact is quantified by the price movement during the “advertising”
period from 15:45 to 16:00, instead of the price movement after the transaction.
By leveraging proprietary NYSE closing auction data that provide the directions of order imbalances, I
investigate the magnitude of the price impact cross-sectionally for both buying and selling market-on-close
orders. This analysis suggests that the price impact (defined as the return in the last 15 minutes) on stocks in
the cross-sectional top quintile of market-on-close order volume and with a buying direction is significantly
positive: 10.1 basis points (bps) more than stocks with low market-on-close order volume. Similarly, the
price impact for stocks with high selling market-on-close order volume is significantly negative (10.8 bps)
compared to that of stocks with low market-on-close volume.
The counterfactual analysis based on the
various price impact estimates from the Breen, Hodrick, and Korajczyk (2002) specification, the Glosten and
Harris (1988) specification, effective spread, and quoted spread (Korajczyk and Sadka (2004)) demonstrates
that the price impact from the market-on-close orders is relatively large.
To further confirm that the price movement for the last 15 minutes is due to the dissemination of closing
auction information, I conduct two falsification tests: one using the 15-minute return from 15:15 to 15:30
and the other using the 15-minute return from 15:30 to 15:45 as dependent variables. The results show that
the price impact during these periods is marginally significant but economically smaller (0.5 bps) than that
during the last 15 minutes. The statistical significance is consistent with two economic interpretations. First,
it is possible that information about the order imbalances is leaked before 15:45, consistent with the practice
at NYSE that floor traders are notified of large market-on-close imbalanced stocks around 15:00. Second,
the market-on-close trades capture only part of the overall anticipated passive flows, and the trading activity
The subperiod analysis shows that this price impact is higher in the first half of the sample, suggesting that, over time, the market
incorporates the information from the closing auction efficiently. As an additional robustness check, I also use the market-on-close
trading volume, retrieved from TAQ data, and I show that stocks with high market-on-close trading volume move more during the
closing auction period.
during the day through market or limit orders may move the stock price in the same direction.
Next, I investigate whether this price impact is short lived, since the ETF flows are mostly demand-driven
and unrelated to fundamental shocks for a given stock. In addition, the fact that large order imbalances require
immediacy at the end of the day suggest that liquidity providers should be compensated for providing liquidity
to passive funds (See Nagel (2012) and Duffie (2010)). I use Fama and MacBeth (1973) regressions at a
daily frequency to analyze the cross-sectional relation between return reversals and market-on-close trading
volume. The cross-sectional results show that stocks with high market-on-close trading volume experience
significant reversals on the following day. Results that decompose close-to-close returns into overnight returns
and intraday returns further suggest that this reversal is not only persistent during the overnight period but also
continuous during the following open-to-close trading period. The results are robust across different stock size
quintiles and different exchanges. Furthermore, the results are not subsumed by order imbalance; Chordia and
Subrahmanyam (2004) find that daily order imbalance predicts next day returns. This suggests that the daily
reversal pattern observed from market-on-close trades is not just a proxy for the entire day trading activities.
To rule out the possibility that my results are driven by market microstructure bias,
such as a bid-ask
spread bounce, I use the midpoint of the quoted bid and ask prices to calculate the daily return. My results
still hold. Using NYSE order imbalance data, I find that the price reversals for high market-on-close trading
volume are symmetric for buying orders and selling orders. This further confirms that the price impact is
temporary and unrelated to fundamental information. The findings provide additional evidence on how the
price dynamics are influenced by institutional trades at the intraday and overnight level, consistent with the
recent studies (Heston, Korajczyk, and Sadka (2010); Bogousslavsky (2018); Gao, Han, Li, and Zhou (2018);
and Lou, Polk, and Skouras (2018)).
Given such a strong short-term reversal, a long/short trading strategy yields significant profits both eco-
nomically and statistically. During the whole sample period, a daily rebalanced portfolio that (a) buys stocks
Nonsynchronous trading bias (Scholes and Williams (1977)) cannot be the cause of return reversals here since the last trade for
a given day is the market-on-close trade, which is the same across all stocks.
My findings, however, differ from these studies in several ways. First, the analysis focuses on the reversals of individual stocks
instead of return continuation at the portfolio level. Heston, Korajczyk, and Sadka (2010) examine the return continuation at half-hour
intervals that are exact multiples of a trading day and Gao, Han, Li, and Zhou (2018) focus on the market intraday momentum, in
which the first half-hour market return predicts the last half-hour market return. Second, the interpretation of my findings is consistent
with the price pressure from passive demands. Lou, Polk, and Skouras (2018) find that the overnight and intraday patterns at a monthly
frequency are consistent with the tug-of-war/client
ele effect, while Bogousslavsky (2018) examines intraday return patterns across
different anomalies.
with high selling closing order imbalances and (b) sells stocks with high buying closing imbalances results
in a risk-adjusted return of 13.2 bps per day, or a 33.26% annualized return with a Sharpe ratio of 3.13. Fur-
thermore, both the long and short legs of the trading strategy yield a similar magnitude of abnormal returns,
consistent with the cross-sectional regression findings.
Taken together, this reversal pattern is consistent with the empirical finding that stocks with higher ETF
ownership exhibit a higher volatility and more deviation from a random walk (Coles, Heath, and Ringgenberg
(2017) and Ben-David, Franzoni, and Moussawi (2018)). To illustrate the contribution of market-on-close
orders on the volatility, I repeat the analysis using S&P 500 stocks by regressing monthly volatility calculated
from daily return on ETF ownership and the results are consistent with Ben-David, Franzoni, and Moussawi
(2018) that higher ETF ownership leads to higher volatility. Then, instead of using close-to-close return to
calculate the daily return and monthly volatility, I use open to 15:45 daily return to purge the impact of the
closing auction. The results suggest that impact of ETF ownership on the volatility of stocks shrinks by 60%
compared to the original coefficient estimates, shedding light on the importance of the market-on-close orders
for stocks with different ETF ownership.
Last, besides passive funds, informed traders might use market-on-close orders to take advantage of
market-depth and camouflage their trading intention. Early market microstructure models (Admati and Pflei-
derer (1988); Admati and Pfleiderer (1991)) predict that uninformed traders who have discretion over the
timing of their trades will optimally cluster their trades to minimize their price impact. These early models
also predict that informed traders will herd with uninformed traders to camouflage their trades. Nevertheless,
herding is difficult to be observed. The growth of market-on-close orders allows me to empirically test these
predictions directly.
Specifically, I conduct two analyses. My first analysis examines the daily return reversal for a subset
of stocks with a high probability of informed market-on-close trades, based on the contradictory trading
directions from both ETF flows and market-on-close imbalance sides. A weaker and insignificant return
reversal among these stocks provides supportive evidence that the informed trades might use market-on-close
orders because the price impact is permanent due to private information (Nagel (2012)). The second analysis
The results are robust for different trading strategies and for different subsample periods. However, the risk-adjusted return for
the second half period is lower than the risk-adjusted return for the first half period, suggesting that investors may become aware of
this pattern and exploit it.
focuses on the trading pattern around earnings announcements. I find abnormal market-on-close trading
volume that does not derive from ETF flows. I also find that informed investors might use market-on-close
orders as a way of exiting positions, consistent with the buy the rumor and sell the news” pattern documented
in Kaniel, Liu, Saar, and Titman (2012) and Kadan, Michaely, and Moulton (2018). The overall evidence here
suggests that informed traders could also use market-on-close orders as an alternative venue for trading.
This paper contributes to the literature in several ways. First, a growing literature explores the impact of
passive investing on asset prices. In particular, recent studies of ETF ownership highlight the unintended con-
sequences for underlying securities, such as the increase of nonfundamental volatility (Ben-David, Franzoni,
and Moussawi (2018)), increased co-movement in returns (Da and Shive (2018)), increased commonality in
liquidity (Agarwal, Hanouna, Moussawi, and Stahel (2019)), more deviation from the random walk (Coles,
Heath, and Ringgenberg (2017); Ben-David, Franzoni, and Moussawi (2018)), and reduction in information
efficiency (Israeli, Lee, and Sridharan (2017)). Brown, Davies, and Ringgenberg (2018) show that ETF flows
at a monthly frequency can predict future stock returns, especially for leveraged ETFs.
However, most of these studies provide indirect evidence that how ETFs impact underlying assets is
through creation and redemption throughout the day or through arbitrage activities exploited by hedge funds
and high-frequency traders.
Box, Davis, Evans, and Lynch (2019) examine the intraday arbitrage opportu-
nities between ETFs and their underlying portfolios and find little evidence that mispricing events alter the
direction of constituent order flows. To my knowledge, this paper is the first to show that, aside from arbi-
trage activities, the price impact of large market-on-close orders is one important but overlooked mechanism
through which ETFs influence underlying assets. I demonstrate that this price impact is economically and
statistically significant and gives rise to a new arbitrage opportunity that allows for the exploitation of the
short-term reversal. This short-term reversal is also consistent with the hypothesis that investors are compen-
sated by providing the liquidity to passive investors (Nagel (2012) and Duffie (2010)).
Second, this paper contributes to the burgeoning literature on the price impact of institutional flow-driven
trades. For instance, both Coval and Stafford (2007) and Frazzini and Lamont (2008) analyze the long-
term return reversal pattern subsequent to mutual fund flow-induced trading while Lou (2012) focuses on
As Ben-David, Franzoni, and Moussawi (2018) state, “The evidence on the role of arbitrage trades is admittedly indirect. Further
research should delve deeper into the channels linking ETF ownership to stock volatility.
the long-term return continuation pattern due to price pressure. Ben-Rephael, Kandel, and Wohl (2011) use
data from aggregate daily mutual fund flows in Israel, and find that these flows create a temporary price
pressure that is reversed within 10 trading days. Besides return predictability, studies (e.g., Greenwood and
Thesmar (2011); Ant
on and Polk (2014); Akbas, Armstrong, Sorescu, and Subrahmanyam (2015)) examine
other consequences of institutional flow-driven trades, such as co-movement and mispricing. More recently,
Etula, Rinne, Suominen, and Vaittinen (2019) show that month-end price pressure originated from monthly
payment cycle leads to price reversal at the beginning of month. Different from previous papers that focus on
flows from actively managed mutual funds or hedge funds at the quarterly or monthly frequency, this paper
analyzes the impact of flows that originate from passive investing at the daily frequency and shows that the
impact of “dumb money” from passive investing is economically large and plays an increasingly important
role in the asset price dynamics observed in recent years.
Third, my paper complements research on price manipulation at close and the effect of closing auction
on end-of-day price discovery and liquidity. Price manipulation at close has been examined by previous
studies such as Hillion and Suominen (2004), Comerton-Forde and Putnin¸
s (2011), and Ben-David, Franzoni,
Landier, and Moussawi (2013). The adoption of closing auction is an attempt to alleviate such manipulation.
Yet, relatively few studies examine the closing auction after its adoption in the U.S. stock market.
McCormick, and Spatt (2009) utilize floor trading data from the first quarter of 2005 to study the role of the
specialists during the closing auction after its adoption by the NYSE. My paper demonstrates the increasing
importance of closing auctions on financial markets, especially due to passive investing. Additionally, my
findings shed light on the trade-off between minimizing tracking errors and better execution even for passive
funds and have policy implications for the design of closing auctions.
The remainder of this paper is organized as follows. Section 2 reviews the institutional background of
closing auctions and passive investing. Section 3 describes the data and the construction of the variables.
Section 4 provides empirical analysis. Section 5 concludes.
Before its adoption in the U.S stock market, market-on-close orders are used for the settlement of option expirations. For instance,
Cushing and Madhavan (2000) investigate the stock returns following market-on-close order imbalance, especially on the days when
options expire.
2 Institutional Background
2.1 Closing auction and close order types
The closing auction is the last event of the trading day across all major exchanges and is designed to
determine the closing price for each stock. The closing price is crucial, as it is the most widely published
reference price for all equity-linked products, such as mutual funds, exchange trade products (ETPs). During
the trading day, a stock can be traded on any exchange. But at the close, it reverts to the exchange where it is
listed. This means that stocks listed on the NYSE will have dramatic closing auction volume on the NYSE at
closing, and the process is similar for Nasdaq-listed stocks.
Three order types are allowed in this process: market-on-close (MOC) orders, limit-on-close (LOC) orders
and imbalance-only (IO) orders (also referred to as closing offset (CO) orders in the Nasdaq market). A MOC
order represents interest that must trade in the closing auction, irrespective of the price. Like a usual limit
order, an LOC order is an order to buy or sell a stock if the closing price is at or better than the bidder’s limit
price. Finally, an IO order is a limit-order type that offsets daily order imbalances at the market close. At
the NYSE, MOC and LOC orders can be entered from 7:00 until 15:45 and cannot be canceled except in the
case of an error.
Starting from 15:45, the NYSE electronically publishes a rundown of open interest on each
stock, including information such as imbalance quantity, paired quantity, imbalance side, and indicated match
At 16:00, auctions are run for all orders, and closing prices are published shortly thereafter.
[Figure 1 here]
To illustrate the growing importance of closing auction trades, I present descriptive statistics for the S&P
500 and all stocks traded on the NYSE in Figure 1. For S&P 500 stocks, the daily average fraction of a
stock’s closing auction trades over total shares traded has risen almost threefold, from 3.5% in 2010 to 10%
Different exchanges have different rules about their cutoff times: 15:50 (Nasdaq) or 15:59 (NYSE Arca). There have been
several rules changes that relate to the closing auction process, with the most recent change on March 1, 2010. With this change, the
time of dissemination changed then from 15:40 to 15:45, and exchange systems automatically and electronically tracked MOC and
LOC interest, as opposed to the designated market maker (DMM). After this change, the frequency of order imbalance dissemination
reduced from 15 seconds to 5 seconds.
See NYSE rule 123C for details about the closing auction procedure:
SelectedNode=chp 1 3&manual=/nyse/rules/nyse-rules/.
in 2018. This represents, on average, 203 million shares and $14.9 billion traded daily through the closing
auction (based on the first six months of 2018). The statistics for all stocks traded on the NYSE show a similar
2.2 Passive investing: Index mutual funds and ETFs
Passive investing is designed to track the performance of a market index. To do this, the fund manager
purchases all or a representative sample of the securities in the index through the form of index mutual funds
and ETFs.
While index mutual funds were introduced in the 1970s, the first index ETFs were offered in
1993 as an SPDR investment vehicle, tracking S&P 500. Passive investing has grown remarkably over recent
years in terms of total net assets under management as well as the number of ETFs being offered. According
to the 2018 Investment Company Fact Book, the total assets of these funds (both index mutual funds and
ETFs) reached $6.7 trillion by the end of 2017, and these funds comprised 35% of the total net assets in long-
term funds, compared to 15% in 2007. In particular, the total net assets for index mutual funds accounted for
$3.4 trillion, up from $619 billion in 2005, while index ETFs accounted for the remaining $3.3 trillion.
Even though both index mutual funds and ETFs are designed for passive investing, there are two major
differences between these two types of funds. First, ETFs, like stocks, are listed on an exchange, and investors
can buy and sell them throughout the day at market-determined prices. However, index mutual funds can be
traded only once per day at the closing price.
Another unique feature of ETFs is the creation–redemption mechanism that operates between the fund
and authorized participants (APs). An ETF does not interact directly with the secondary market. Instead,
APs play a key role in the primary market for ETF shares because APs are the only investors allowed to
interact directly with the fund. An AP can create (redeem) ETFs shares by transferring to (receiving from)
the ETF the underlying securities. Creation and redemption follows a predefined procedure specified in the
authorized participant agreement or the authorized participant handbook. This procedure determines how
large predefined blocks (often 50,000 ETF shares or creation units) are exchanged at designated times (usually
Most ETFs are standalone and differ from mutual funds in terms of structure. Vanguard ETFs are not standalone ETFs, and they
are merely another share class of the Vanguard open-end index mutual fund
end-of-day) and at designated prices (typically closing prices) between APs and funds.
Creation and redemption at the end of the trading day can be viewed as two components that originate
from the two different roles of the APs. First, creations and redemptions are partly the results of arbitrage
activity. Specifically, if an ETF trades at a discount relative to its basket of stocks, APs have an incentive to
buy ETF units in the market and short-sell the underlying basket during intraday trading. Near the end of the
trading day, the APs then deliver the ETF shares they bought to the ETF sponsor, and the APs receive the
underlying basket to cover the short position. In the case of a premium, the APs short-sell the ETF unit and
buy the underlying basket during intraday trading. At the end of the trading day, APs deliver the underlying
basket, and receive ETF shares from the ETF sponsor to maintain zero inventory risk. Figure A.2 illustrates
examples of these two scenarios. The profit of arbitrage activity is locked in during the intraday trading
session, since the end-of-day creation or redemption is based on the closing NAV of the ETF and on the
closing prices of underlying securities. This means there is no arbitrage for this end-of-day process.
At the end of the trading day, the value of the creation (or redemption) basket equals the value of the
creation unit based on the ETF’s NAV at the end of the day on which the transaction was initiated. Under both
scenarios, APs might participate in the closing auction for the underlying securities to minimize mismatching
risk or inventory risk. Therefore, similar to index mutual funds, inflows and outflows captured by the ETF
shares outstanding are also closely related to the closing auction.
[Figure 1 here]
Another part of creation and redemption derives from an investor’s demand for exposure; that is, an
AP’s activity can be viewed as a technology for adjusting the shares outstanding of the ETF in response to
the demand for the exposure provided. For instance, if an institutional investor (or an aggregation of retail
investors) seeks a large block of a particular ETF’s shares, the investor may turn to an AP to facilitate a
creation. The buyer delivers either cash or securities to the AP, who in turn delivers the basket of securities to
the ETF sponsor. The ETF sponsor then issues ETF shares to the AP (i.e., a creation) to give to the buyer.
See appendix A.1 for one example of an Authorized Participants agreement. There could be a cash adjustment if an AP creates
or redeems in cash.
3 Data
For ETF and daily flow data, I first use the Bloomberg terminal to identify all ETFs that are traded on
the major U.S. exchanges. Following previous studies on ETFs, leveraged and inverse-leveraged ETFs are
The information on ETFs from Bloomberg contains the total shares outstanding at the end of the
day, the ticker symbol, CUSIP, and the calculated daily flows in dollars. I then match the ETFs with the Center
for Research in Security Prices (CRSP) database for all securities that have a share code of 73 by symbol and
by CUSIP to obtain daily trading information. My final ETF sample contains 365 distinct ETFs for 2010
to 2018. For ETF holdings data, I use the Thomson-Reuters Mutual Fund Ownership database by following
procedures. First, I identify the fundno of the ETFs by matching the historical ticker-date with the s12type8
dataset in Thomson-Reuters. Then, I retrieve the last available detailed quarterly holding information using
table s12type3 dataset.
At the end of each day, for each stock i in the CRSP stock file universe, I construct the daily ETF flow as
ET F f low
i, j,t
where j is the set of ETFs that hold stock i; w
i, j,t
is computed as the most recently available fund portfolio
percentage weight of ETF j in stock i (i.e., the dollar ownership by the fund portfolio in stock, divided by total
equity assets in the fund portfolio, based on the most recent quarterly report); abs($Flows)
is the absolute
value for the daily dollar flow for each fund j at time t, where the daily dollar flow is measured based on the
change of total shares outstanding at day-end from Bloomberg for each ETF; $Volumes
is the dollar trading
volume over the same period t. The motivation to use the sum of the absolute value of flows is based on the
premise that different funds cannot trade directly to cancel out the opposite trading need.
For market-on-close data, I use TAQ second (pre-2014) and millisecond (post-2014) data to identify each
stock’s market-on-close volume based on its primary listing exchange for each day with the sale condition of
“6” (closing print) mainly for NYSE-listed securities. For Nasdaq-listed securities, trades from the closing
auction can also be identified by the sale condition of “M” (market center close price). However, in most
AUM from leveraged and inverse-leveraged ETFs account for 2% of the whole sector according to BlackRock (December 2014).
cases, these trades have identical price and volume information with the sale conditions of “6” and will be
discarded to avoid double accounting.
My sample period for closing auction volume spans March 2010 to June 2018 because the NYSE (also
NYSE Arca) did not report the closing auction until July 2008 and because the regulation on the dissemination
of the closing order imbalance changed in July 2008 and in March 2010 (as described in section 2). I cross-
match with CRSP, and I retain only common stocks (shrcd in 10 or 11), which means that I exclude securities
such as warrants, preferred shares, American Depositary Receipts, closed-end funds, and REITs.
I also
exclude stocks with zero daily trading volume or zero market-on-close volume. Stocks with prices of less than
$5 are also excluded to ensure market micro-structure issues do not affect the analysis. I define the market-
on-close volume measure as the total shares of market-on-close volume divided by the total trading volumes
on each date. For control variable order imbalance, I use the modified Lee and Ready (1991) algorithm to
identify buy or sell orders following Blume, Easley, and O’Hara (1994) and Holden and Jacobsen (2014). For
the portion of my analysis that uses intraday prices or returns, I use the last available valid trade price from
the TAQ data as the end-of-period stock price, and these returns are winsorized at the 1st and 99th percentile
to avoid extreme outliers. Additionally, I use the NYSE close imbalance feeds data from July 2008 to June
2018 to identify the direction of the market-on-close order imbalances. The closing auction imbalance feeds
data provide information for each stock on imbalance quantity, paired quantity, imbalance side, and indicated
match prices every five seconds from 15:45 until 16:00.
For the analysis using the Russell reconstitution, I first collect the constituents of the Russell indices (RTY
index, RIY index, and RAY index) from the last trading day in May of each year before reconstitution and
the last trading day in June each year after reconstitution. Because my sample mainly starts from 2010, after
Russell Inc. implemented a new banding policy around the 1000 cutoff to mitigate index turnover,
I follow
the procedure from Coles, Heath, and Ringgenberg (2017) to construct new upper and lower bounds in order
to construct new thresholds for identifying potential index switchers.
Under the new policy, after the initial market capitalization breakpoints are determined, existing members
As suggested by Boehmer, Jones, and Zhang (2008), ticker symbols are sometimes reused, and ticker symbols in CRSP do not
always match the ticker symbols in the NYSE data, especially for firms with multiple share classes. Tickers and CUSIPs are used to
ensure accurate matching.
Russell reconstitution guide from Russell Inc.:
are reviewed to determine whether they fall within an accumulative 5% of the market cap range around the
new breakpoints. This creates an upper band and an lower band around the Russell 1000/2000 index threshold.
The width of each band equals 2.5% of the total market cap of the Russell 3000E (which consists of 4,000
stocks or the whole stock universe if the total number of eligible stocks is less than 4,000). Stocks within the
bands do not switch their index assignment from previous year.
To construct the post-banding sample, I perform the following procedure. First, all candidate stocks
from the CRSP are ranked based on their end-of-May market cap, and the threshold for rank 1,000 is then
imputed. Then, I calculate the cumulative market cap from stocks ranked 1 to 1000 (CMC
) and the
cumulative market cap from stocks ranked 1 to 4000 (CMC
). The upper band is then determined based
on the largest n, such that the cumulative market cap from stocks ranked 1 to n (CMC
) is smaller than
0.025 ×CMC
. The lower band is determined in a similar way.
Finally, information regarding earnings announcements and analyst forecasts is extracted from the Insti-
tutional Brokers’ Estimate System (I/B/E/S).
[Table I here]
Table I presents summary statistics for the whole sample. The sample consists of 6,663,021 stock-day obser-
vations over the whole period from July 1, 2008, to June 30, 2018. Over this period, the average volume for
the market-on-close orders across all stocks and all days is about 62,000 shares, which on average accounts
for 6.4% of total daily volume. There is $1.57 million in absolute flows per stock-day (i.e., 14.1% of daily
volume). Variable definitions are available in the Appendix A.I.
4 Results
4.1 ETF flows and market-on-close volume
My first hypothesis is to test whether ETFs influence the underlying assets through the closing auction at
the end of the trading day. In this section, I begin the analysis by first showing that ETF flows are correlated
with market-on-close volume at the individual stock level. APs are motivated to create or redeem either
because they are actively involved in arbitrage activities or because they are trasmitting demand from the
tracking indices of institutional or retail investors. In both cases, since the orders are based on the closing
prices of the underlying securities and are cleared only once per day at 16:00, APs have the incentive to place
a market-on-close order. To formally test this hypothesis, I estimate a panel regression of daily market-on-
close volume on ETF flows as follows:
MOC Volume
= α + βETF flow
+ γX
+ Fixed Effect + ε
where MOC Volume
is the market-on-close trading volume for stock i from the TAQ data, scaled by total
trading volume in day t; the key variable of interest is ETF flow
, the aggregated absolute flows at time t
for stock i from ETFs holding the stock, as defined in Equation (1); and X
denotes a vector of time-varying
controls for stock characteristics.
To address potential omitted variables issue, I include stock and day fixed effects to account for both time-
invariant stock effects and time-varying effects. I also control for a number of stock characteristics that might
be related to closing auction trading. The variable log(market cap) is the natural log of market capitalization of
the stock. Amihud ratio measures illiquidity and is computed as the absolute daily return divided by the total
daily volume in millions of dollars, following Amihud (2002). Return
and Return
represent stock returns
over the same day and over the previous day, respectively. Finally, I include Turnover and Order imbalance as
explanatory variables to capture the aggregate-level trading, since APs could use alternative trading strategies,
such as submitting a market order during the day or spreading orders throughout the normal trading session.
Order imbalance is calculated using the modified Lee and Ready (1991) algorithm, following Blume, Easley,
and O’Hara (1994) and Holden and Jacobsen (2014). It is defined as the total dollar purchase minus total
dollar sell, normalized by the total dollar trading volume. Turnover is the total daily trading volume divided
by market capitalization. To ease interpretation, for each day, I standardize both the dependent variable and
the main explanatory variables to have a mean of zero and a unit of standard deviation. Standard errors are
clustered at the stock and daily levels to account for error correlation within stocks and days as suggested in
Petersen (2009).
The results are presented in Table II. Column 1 reports the estimate of a simple benchmark regression
of market-on-close trading volume on stock-level ETF flows after including both stock and day fixed effects.
Consistent with the hypothesis, ETF flows are positively associated with market-on-close trading volume. In
particular, a one standard deviation increase in ETF flow is associated with a 27% of one standard devia-
tion increase in market-on-close trading volume. The coefficient is statistically significant at the 1% level.
Columns 2 to 4 add stock-characteristics controls and various fixed effects. Column 2 adds only firm fixed
effects in addition to all stock characteristics. The estimated coefficient shrinks to 25.5% but retains its sta-
tistical significance. The estimate of β from Column 3 is still significant after controlling for the day fixed
effect. The larger coefficient of 30.2% suggests that certain time-invariant characteristics are important in
determining the market-on-close trading volume. The coefficient in Column 4 shrinks to 25.4%, but remains
significant. The estimated impact is economically sizable: A one-standard-deviation increase in ETF flows
leads to a increase of 2.01% (0.255 × 7.9%) of market-on-close trading volume, which is a 31.4% increase
relative to the mean market-on-close volume of 6.4%.
[Table II here]
The independent variable ETF flow
used in Panel A is measured by the sum of absolute stock-level
The assumption behind using this measure is that APs, who are responsible for creation and redemp-
tion from various ETFs, will not cross-trade stocks that are commonly held by multiple ETFs.
For example,
APs, who are responsible for both the creation of the iShares S&P 500 ETF and the redemption of the State
Street’s S&P 500 ETF will submit separate orders through a secondary market or through market makers.
However, this assumption might not always be true because APs could submit netted orders. In addition, the
clearing process for creation or redemption requires that deposit securities must be delivered to a Depository
Trust Company (DTC) through the National Securities Clearing Corporation (“NSCC”). Therefore, the total
trading for each stock could be netted at a clearing center. To take cross-trading or netting into consideration,
an alternative measure of ETF flows, which nets the flows at the stock level, is defined as
ET F f low
i, j,t
Ben-David, Franzoni, and Moussawi (2018) argues that this component of flows as the sum of absolute stock-level flows captures
part of non-fundamental explanations since investors might reshuffle money across the ETFs holding a given stock.
For mutual funds, within the same fund family, different funds may cross-trade or do inter-fund lending.
Panel B of II presents the results that repeat the same analysis as in Panel A but uses the alternative
definition of ETF flows described above. Across all four specifications, the coefficients of the estimate are
still significant at the 1% level, suggesting market-on-close trading volume is driven mainly by net inflows
of money into the ETFs. For instance, the specification in Column 4 that adds all stock controls and both
stock and day fixed effects shows that a one standard deviation increase in ETF flow is associated with a
22.5% of a standard deviation increase in market-on-close trading volume. The small difference between
the magnitude of the coefficient in Panel A and the coefficient in Panel B suggests that this new ETF flow
measure is noisier and that netting or cross-trading might be a less important factor when determining the
market-on-close trading volume.
Table II, Panel C shows that the effect has increased over time and is robust across different exchanges.
The fact that the effect intensifies is consistent with the significant increase in passive funds over the last 10
years. For the sake of brevity, in each of the subsample analyses, I only show the coefficient on ETF flow
When using all controls, the coefficient on ETF flow
changes from 0.248 for the sample period of July
2008 to June 2014 to 0.250 for the sample period of July 2014 to June 2018. The analysis using stocks from
different exchanges shows that this effect is pronounced across all stocks. The effect is stronger for Nasdaq
stocks (0.276) compared to NYSE stocks (0.185). To summarize, the main conclusion from this analysis is
that ETF flows predict statistically and economically significant variation in market-on-close trading volume.
4.2 A quasi-natural experiment: Russell reconstitution
The OLS results discussed in the previous section suggest that an increase in market-on-close trading
volume is strongly correlated with passive fund flows. Even though I include fixed effects and control for a
number of time-varying stock characteristics, the causal inference is still questionable if certain time-varying
variables that co-determine market-on-close trading volume and ETF flows are omitted. In this section, I
conduct analyses using the annual Russell index reconstitution experiment as a quasi-natural experiment,
which allows me to exploit mechanical changes in passive flows.
The Russell 1000 and 2000 indices are constructed based on the end-of-May market capitalization ranks
See Appel, Gormley, and Keim (2016); Chang, Hong, and Liskovich (2015); Ben-David, Franzoni, and Moussawi (2018); and
Coles, Heath, and Ringgenberg (2017).
each year. Russell Inc. reconstitutes the indices on the last Friday in June of each every year based on end-
of-May stock capitalization; the composition remains constant for the rest of the year. Before 2007, index
assignment followed a simple threshold: Stocks ranked from 1 to 1,000 were assigned to the Russell 1000,
while stocks ranked from 1,001 to 3,000 were assigned to the Russell 2000. Such a simple threshold creates
a quasi-natural experiment to exploit mechanical changes in passive index.
Starting in 2007, Russell implemented a new assignment banding procedure that replaced the original
rule. After the initial market capitalization breakpoints are determined, existing members are reviewed to
determine whether they fall within in accumulative 5% market cap range around the new breakpoints. This
creates an upper band and a lower band around the Russell 1000/Russell 2000 index threshold. The width of
each band equals 2.5% of the total May cap of the Russell 3000E (which is 4000 stocks or the whole stock
universe if the total number of eligible stocks is less than 4000). Stocks within the bands do not switch their
index assignment from the previous year. After the 2007 banding, an existing member will not be replaced
if its market cap is within 5% of the cutoff range. This significantly decreases the addition and deletion of
stocks due to Russell reconstitution.
However, even under the new regime, firms still cannot control their rankings precisely, thus being as-
signed to the left or right of the cutoff (i.e., either upper band or lower band) is quasi-random. This upper
band/lower band division actually creates the new thresholds for index switching, and the assignment is es-
sentially quasi-random around the band threshold (Coles, Heath, and Ringgenberg (2017)).
Because the Russell indices are value-weighted, this random assignment leads to significant differences
in the portfolio weights and further differences in passive ownership around the threshold. The weights of the
top stocks in the Russell 2000 are about 10 times larger than those of the bottom stocks in the Russell 1000
(Chang, Hong, and Liskovich (2015)). Consequently, a significantly larger amount of passive money tracks
the top Russell 2000 stocks than the bottom Russell 1000 stocks. Stocks that were at the top of the Russell
2000 but then switched to the Russell 1000 would receive less institutional ownership and would expect lower
passive fund flows.
I conduct a two-stage least squares estimation. For my main analysis, the sample is composed of stocks
that in May, before index reconstitution, are potential switchers in the Russell 2000 around the upper band
threshold. The sample composition remains constant for all the months between June, the first month after
index reconstitution, and April of the following year. In the first stage, I conduct a regression of ETF flows on
an indicator variable of whether the stock switches index membership in June:
ETF flows
= α + β 1(Switcher
) + γX
+ Fixed Effect + ε
In the second stage, I regress market-on-close trading volume on the fitted value of ETF flows from the
first stage. The regression is
MOC Volume
= α + β
ETF flow
+ γX
+ Fixed Effect + ε
In addition to the control for the rank of market capitalization, which is the assignment variable, I include
the same set of controls as in the OLS regressions. I include only time fixed effects since the identification
strategy relies on the cross-sectional variation by comparing switchers versus nonswitchers. Standard errors
are double clustered at the stock and day level. Similar to the OLS regression, all variables, except return
variables, are standardized cross-sectionally to have a mean of zero and a unit of standard deviation for ease
of interpretation. I use first-degree linear polynomial specification for the ranking variable.
Table III, Panel A shows the first-stage regressions. Columns 1 to 2 report the baseline regression with
different bandwidths (±100 and ±200). The results of this test show that switching indices has a strong and
significant impact on the ETF flows. The coefficient on the switch indicator in Column 1 suggests that the
average ETF flow in the 12 months after reconstitution decreases by about 17.2% of a standard deviation for
stocks that switched to the Russell 1000. The magnitude of the estimate is similar, at -0.155 in Column 2,
where the bandwidth increases to 200. To mitigate concerns that there might be some anticipated flow-driven
trading during the migration period of June, the sample in Columns 3 and 4 omits the day observations from
the month of June and the analysis is repeated. The conclusion remains unchanged.
[Table III here]
Table III, Panel B reports the second-stage coefficient estimates of the effect of ETF flows on market-on-
close trading volume. The effect of ETF flows on market-on-close trading volume is significant across all
samples and bandwidths. The coefficients range from 0.341 to 0.526 and are statistically significant at the 1%
level. The economic magnitude from the IV estimates is larger than the OLS estimates. This is not surprising
because, first of all, the OLS sample includes all common stocks that have low ETF ownership, while the
IV sample focuses on the stocks included in the Russell indices. Another reason is that IV estimates reflect
the local average treatment effect, which is the effect of the treatment group compared to the control group
around the Russell threshold. The drastic status change for index membership results in a greater impact than
for average stock.
Overall, the results using Russell reconstitution provide support for the causal interpretation of the positive
relation between ETF flows and market-on-close trading volume.
4.3 Closing auction and cross-sectional return predictability
4.3.1 Price impact during auction period
Given the evidence that ETF flows lead to more market-on-close trading volume, in this section I investi-
gate whether the process of the closing auction affects asset price dynamics. The trading activity in a security
provides valuable data about the information structure and subsequent price moves in the security.
is strong empirical evidence that demand and supply shocks can affect individual stock prices. For instance,
event studies that focus on compositional changes in the S&P 500 index show that additions increase share
prices while delisting decreases prices.
Also, the literature on block trades generally finds evidence of tem-
porary price pressure on individual securities.
Greenwood (2008) examines transitory price effects upon a
weighting change to the Nikkei 225. Furthermore, fire sales from distressed mutual funds also distort stock
prices over a long horizon (Coval and Stafford (2007)).
In a closing auction, an auction order can be submitted during normal trading hours until 15:45, but the ag-
gregated market-on-close order imbalance information will not be disseminated for NYSE-listed stocks until
15:45 to members of the public who subscribe to the feed service. Similar to the price pressure hypothesis of
See Blume, Easley, and O’Hara (1994), Campbell, Grossman, and Wang (1993), Conrad, Hameed, and Niden (1994).
See Goetzmann and Massa (2003); Harris (1986); Shleifer (1986); Beneish and Whaley (1996); Lynch and Mendenhall (1997).
Kraus and Stoll (1972), Lakonishok, Shleifer, and Vishny (1992), Chan and Lakonishok (1993), Chan and Lakonishok (1995).
Scholes (1972), where stock prices can diverge from their information-efficient values because of uninformed
shocks to excess demand to compensate those who provide liquidity, the price impact should be the strongest
for stocks that have a high market-on-close trading volume imbalance when the information disseminates.
Furthermore, since this excess demand is not driven by fundamental information about firms, the price impact
for stocks that have high market-on-close orders should be symmetrical in terms of buying and selling. To
verify this conjecture, I rely on the NYSE proprietary closing auction order imbalance feed data, which pro-
vide the direction of the order imbalance, allowing for a comparison of the price impact on both buying and
selling orders. The sample in this analysis consists of all NYSE stocks that have at least a $5 share price from
March 1, 2010 (when the NYSE changed its dissemination time rule), to June 30, 2018. The price impact is
computed as the stock return for the period starting from 15:45 to 16:00, based on the trading prices from the
TAQ data.
I then run a Fama-Macbeth regression of this return measure on the measure for market-on-close
trading volume cross-sectionally. Standard errors are corrected for heteroskedasticity and autocorrelation up
to five lags.
[Table IV here]
Table IV shows the Fama-Macbeth regression results, regressing the last 15-minute return on highmoc buy
and highmoc sell as the main independent variables, where highmoc buy equals to 1 if the stock is in the top
quintile of market-on-close trading volume across all stocks listed on the NYSE and the imbalance side of the
closing auction at the beginning of dissemination is ‘buy’; highmoc sell is defined in a similar way. Control
variables include the turnover ratio and order imbalance.
The results show that, on average, the 15-minute return of stocks with high buying market-on-close vol-
ume is significantly higher (10.1 bps) than that of stocks with low market-on-close volume. In contrast, the
15-minute return for stocks with high selling market-on-close volume is significantly lower (-10.8 bps) than
that of stocks with low market-on-close volume. These results suggest a price impact during the last 15 min-
utes for stocks with high market-on-close order imbalances, and the price movement is economically large,
regardless of whether the imbalance side is buy or sell.
The price at 15:45 is based on the first observed transaction price that occurred in the period starting at 15:45:00, after intertwining
with quota data to exclude outliers, as in Holden and Jacobsen (2014) and Bogousslavsky (2018).
Results are virtually the same when I use various lags.
To address the concern that the price movement might be caused by information other than the closing
auction feeds, Panel B of Table IV shows the results of tests that use the 15-minute return for the period from
15:15 to 15:30 and the return for the period from 15:30 to 15:45 as dependent variables. Interestingly, the
analysis shows that the price moves in a statistically significant way even before order imbalance disseminates,
suggesting a possible information leak.
However, from the economic magnitude perspective, the price
impacts during both periods are small: the price impact during 15:15 to 15:30 is about 0.5 bps (-0.2 bps)
for stocks with high buying (selling) market-on-close order imbalances, and the price impact during 15:30 to
15:45 is about 1.2 bps (-1.4 bps) for stocks with high buying (selling) market-on-close order imbalances.
As a robustness check, I also use the market-on-close trading volume from the TAQ database to repeat
the analysis. I use the absolute value of the stock return from 15:45 to 16:00 to capture the price movement
because the TAQ database does not reveal the market-on-close order imbalance direction (i.e., the market-on-
close order can either buy or sell).
Table A.II, Panel A reports the Fama-Macbeth regression coefficients. The result in Column 1 confirms
that the price impact is larger when the market-on-close volume is high, with market-on-close volume again
defined as market-on-close trading volume divided by the total trading volume for the day. A 1% increase in
market-on-close trading volume leads to 0.28 bps change in price movement for the 15-minute period, and
the coefficient is significant at 1%.
To address the concern that the market-on-close trading volume might be skewed, I use the measure,
highmoc, which is an indicator of whether the stock is in the top quintile of the market-on-close volume for
the day. Column 2 reports the coefficient estimates using this measure. Again, the result is consistent with
my hypothesis. Cross-sectionally, stocks in the high quintile of market-on-close volume move 2.17 bps more
than stocks that are not in the top quintile. Panel B repeats the same placebo tests that use absolute returns
for the period from 15:15 to 15:30 and the absolute return for the period from 15:30 to 15:45 as dependent
variables. Contrary to the results in Table IV, the coefficient estimates on the highmoc from both columns are
insignificant at the 5% level. This supports the argument that the price impact during the last 15 minutes is
due to the closing auction dissemination.
The U.S. Securities Exchange Commission fined the NYSE for improper distribution of market data (including closing auction
imbalance data) in 2012 and 2014. Mayhew, McCormick, and Spatt (2009) show that specialists could front-run the closing auction
before such information is disseminated to floor traders.
The analysis so far shows the absolute level of the price impact due to the high market-on-close order
imbalance. For investors who submit market-on-close orders, the alternative trading strategy could be simply
placing market orders during normal trading hours. Given the large quantity of trading needs, the price impact
of such block trades could also be large. To quantify the economic magnitude of the transaction costs from
using market-on-close orders relative to counterfactual situations, I conduct a market impact cost analysis
following Korajczyk and Sadka (2004).
Specifically, I consider four alternative price impact measures: the Breen, Hodrick, and Korajczyk (2002)
(henceforth BHK) measure; the Glosten and Harris (1988) (henceforth GH) measure; and the effective and
quoted spread. Korajczyk and Sadka (2004) use intraday transaction data from January 1993 to May 1997 to
estimate market impact costs. These costs are estimated each month for each stock using the TAQ data. Using
cross-sectional relationships between these market impact costs and firm characteristics, the out-of-sample
price impact costs can then be estimated.
[Table V here]
Table V presents the counterfactual price impact estimates together with the actual price impact for stocks
with high market-on-close order imbalance. The results show that the price impact would ranges from 0.12
bps to 3.39 bps among all alternative price impacts estimates.
The estimates for a high buying market-on-
close and a high selling market-on-close are similar. However, all the estimates from these four specifications
are less than the actual price impact (11.24 bps for buying and 11.46 bps for selling).
Taken together, these results suggest that the price impact due to the high market-on-close order imbalance
volume is large during the last 15 minutes of the trading day and it is symmetrical in terms of buying and
selling orders.
See the Appendix for a detailed estimation of each measure.
The estimate is based on the same market-on-close trading volume and the time interval is based on 30 minutes, following
Korajczyk and Sadka (2004). The effective spread and the quoted spread are proportional to the trading volume, while the BHK
measure and the GH measure are nonproportional to the trading volume.
4.3.2 Closing auction and price reversal
Having provided evidence consistent with a price impact for stocks that have high market-on-close trading
volume during the last 15 minutes of the trading day, I now test whether the closing auction process impacts
the price movement beyond this 15-minute interval. Since the demand shocks originate from nonfundamental
shocks, such a price impact should be transitory and should lead to price reversal. To test this hypothesis,
I use Fama-Macbeth regressions at a daily frequency to analyze the cross-sectional relation between return
reversals and market-on-close trading volume as follows.
= α + β
+ β
× highmoc
+ γX
+ ε
The sample consists of all common NYSE stocks from March 2010 to June 2018 with market-on-close trading
volume determined from the TAQ data, as in the previous section. The dependent variable in our baseline re-
gression is the stock’s daily return (close-to-close return). To capture the price reversal, I interact the indicator
variable highmoc with the previous day’s return, so that a negative and significant coefficient β
would suggest
that, cross-sectionally, a stock with highmoc exhibits a negative serial correlation. I control for a number of
stock characteristics that might relate to the prediction of short-term returns: log(market cap), Amihud ratio,
order imbalance, and turnover. Standard errors are corrected for heteroskedasticity and autocorrelation up to
ve lags (Newey and West (1987)).
[Table VI here]
Table VI reports the coefficient estimates from the Fama-MacBeth regression. The results in Column 1,
Panel A show two findings. First, there is weak evidence of negative first-order serial correlation at a daily
frequency for stocks that do not have a high market-on-close order imbalance in our sample period. The
coefficient on Return
is negative (
0.004) but insignificant.
Second, the β
coefficient implies that stocks
with high market-on-close trading volume exhibit strong negative autocorrelation, which is consistent with my
hypothesis that the price pressure is temporary.
The estimate of the coefficient is
0.022 with the t-statistic
Chordia et al. (2005, Table 1) find that these autocorrelations are smaller in more recent subperiods.
A large literature studying stock return serial correlation shows that at a monthly or annual frequency, there is significant negative
7.06. The magnitude of the reversal for stocks that have high market-on-close trading volume is more
than five times larger than the reversals for stocks that have lower trading volume. Column 2 shows this result
after controlling for stock characteristics that relate to short-term return predictability. The β
coefficient is
0.019 and remains significant at the 1% level.
The evidence so far suggests that stocks in the top quintile of high market-on-close trading volume have a
stronger reversal using close-to-close returns. To closely investigate when such a reversal occurs and whether
it is persistent, I decompose the close-to-close returns into overnight returns and intraday returns, following
Lou, Polk, and Skouras (2018) and Bogousslavsky (2018).
1, (7)
1 + r
1 + r
Columns 3 and 4 of Table VI, Panel A report the cross-sectional regression results, using these two returns
as dependent variables, respectively. The results show that the reversal is quite persistent and significant
both during the next open-to-close period (
0.011) and during the overnight period (
0.007). Interestingly,
,the coefficient on Return
is also significant and negative when the dependent variable is the overnight
return. This result comports with the finding in Lou, Polk, and Skouras (2018) that stocks with relatively high
overnight returns over the previous month have, on average, relatively low intraday returns in the subsequent
The evidence using the TAQ data thus far suggests that stock returns for high market-on-close trading
volume will reverse regardless of whether the market-on-close order imbalance is high buying or high selling.
As a robustness check of whether the reversal pattern might differ for buying versus selling orders, I conducted
an additional subsample analysis using NYSE proprietary closing auction order imbalance data focusing on
serial correlation (Jegadeesh (1990)); however, at a daily frequency, the results are weak (Fama and French (1988)). Collin-Dufresne
and Daniel (2016) show that the reversal effect for residual return is remarkably strong in even the largest 100 most liquid stocks.
The decomposition assumes that dividend adjustments, share splits, and other corporate events that could mechanically move
prices take place overnight, and this assumption is reasonable, as argued in Lou, Polk, and Skouras (2018).
See also Hendershott, Livdan, and R
osch (2018) and Bogousslavsky (2018) about the return difference over the trading day and
NYSE-listed stocks. Specifically, I run the similar Fama-MacBeth regression as follows.
= α + β
highmoc buy
× Return
highmoc sell
× Return
+ γX
+ ε
The independent variable and dependent variable are the daily returns, and the dummy variable highmoc buy
is an indicator of whether (a) the stock is in the top quintile of market-on-close trading volume across all
stocks listed on the NYSE and (b) the imbalance side of the closing auction at the beginning of dissemination
is buy. The indicator highmoc sell is defined similarly. The main variables of interest are β
and β
, which
capture the return reversal for high buying market-on-close orders and high selling market-on-close orders,
[Table VII here]
Table VII shows the Fama-Macbeth coefficient results. Consistent with the previous finding, the coeffi-
cients β
and β
load significantly negatively across all regressions. This suggests a strong return reversal for
both high buying market-on-close orders and high selling market-on-close orders. Furthermore, for stocks
that have high buying market-on-close orders, this reversal is slightly stronger than the reversal pattern for
stocks with high selling market-on-close orders. Columns 3 and 4 show the results using subsequent intra-
day returns and subsequent overnight returns as their respective dependent variables. The reversal pattern is
persistent and stronger during the overnight period.
To address the concern the my return analysis used is based on the close-to-close return, which might be
affected by the price pressure from previous day’s closing auction, Table A.III, Panel A shows the repeated
analysis using the same day open-to-close return as the independent variable and also in the interaction term.
The coefficient of interest β
0.015 (t-stat=
5.09) when the dependent variable is the close-to-close return
0.008 (t-stat=
2.81) when the dependent variable is the subsequent intraday return. As an additional
robustness check, Table A.III, Panel B uses the return from 15:45 to 16:00 as the independent variable and in
the interaction term. The conclusions remain unchanged.
Studies show that market microstructure bias and small cap stocks could contribute to the short-term
reversal. To alleviate the concern that the short-term reversal documented here is not driven by these biases, I
conduct two additional tests, and results are presented in the Appendix.
The first test controls for size explicitly by dividing stocks into different groups. In particular, at the end of
each month, stocks are sorted into ve groups based on their market capitalization. Then, I run Fama-Macbeth
regressions for those stocks classified into different groups. The results in Table A.IV show that, across all
size groups, the coefficient β
is negative and significant. (Even for stocks from the large size group, the
estimate is
0.0197 with t-stat of
The second test reported in A.V uses the average of the bid and ask quotes at closing to calculate the
close-to-close returns, and the results suggest that the price reversal cannot be explained by bid-ask bounce
bias. Nonsynchronous trading bias is another market microstructure bias that potentially leads to negative
serial correlation. However, since all stocks in the sample involve the closing auction, which occurs at same
time across all stocks at 16:00, it is unlikely that the reversal pattern can be explained by nonsynchronous
Taken together, stocks with high market-on-close trading volume experience a larger price impact during
the last 15 minutes of trading. Since flow-driven trading is due to nonfundamental shocks, the returns reverse
overnight and in the following day. Combining the evidence that ETF flow leads more market-on-close orders,
the results lend supportive evidence to the recent finding that ETF ownership leads to increasing deviation
from a random walk for underlying assets (Ben-David, Franzoni, and Moussawi (2018); Coles, Heath, and
Ringgenberg (2017)).
4.3.3 Long–short trading strategies
Given the significant serial correlation documented here for stocks that have high market-on-close trading
volume, one might wonder whether a closing auction imbalance can be used as a signal to form profitable
trading strategies. In this section, I consider two trading strategies. The first is to form daily equal-weighted,
long–short portfolios based on the direction of the closing auction imbalance side on the previous day and the
closing auction total imbalance quantity from the NYSE proprietary data. Specifically, at the end of day t,
stocks that are (a) in the highest quintile of the closing auction imbalance quantity and (b) on the imbalance
side of selling based on the first disseminated closing auction information at 15:45 will be bought. Meanwhile,
the stocks will be sold short if they are (a) in the highest quintile of the closing auction total imbalance quantity
but (b) on the imbalance side of buying. The sample consists of all nonzero closing auction volume stocks
available from the NYSE order imbalance feed data from July 2008 to June 2018.
[Table VIII here]
Table VIII, Panel A reports the alphas and factor loadings for such a long–short portfolio. The results
show that the alpha for the long–short portfolio is statistically and economically large: 13.29 bps for the
CAPM model (with a t-stat of 10.11) and 13.22 bps for the Fama–French three-factor model (with a t-stat of
9.81) and 13.20 bps for a four-factor model (with a t-stat of 9.91). This is equivalent to a 33.26% risk-adjusted
annualized return. The factor loadings on SMB, HML, and UMD are insignificant, suggesting the return from
this long–short trading strategy is not driven by these common factors. The return decomposition from the
long and short legs of the portfolio shows that both legs yield significant alphas (6.07 bps for the long leg and
7.23 for the short leg).
The second trading strategy is to form daily equal-weighted, long–short portfolios based on the direction
of returns in the previous day and the market-on-close trading volume, using only the TAQ data.This trading
strategy does not rely on the proprietary real time closing auction order imbalance data. Nagel (2012) also
uses the lagged return as a noisy proxy for unobserved market-maker inventory imbalance when investigating
the short-term reversal strategies. Specifically, at the end of day t, stocks that are (a) in the highest quintile of
market-on-close trading volume and (b) also have a negative return will be bought, and stocks that are (a) in
the highest quintile of market-on-close trading volume but (b) have a positive return will be sold short. The
sample consists of all common stocks with nonzero market-on-close volume from July 2008 to June 2018,
and the holding horizon starts from the close of day t to the close of the following day.
Table VIII, Panel B reports the alphas and factor loadings from this long–short portfolio. The zero–
investment portfolio has, on average, a market-adjusted return of 7.44 bps per day and it is statistically sig-
nificant. After adjusting for the Fama–French three factors and Fama–French plus momentum, the alphas
maintain almost the same magnitude (7.36 bps and 7.37 bps, respectively) and are still significant at 1%.
Similar to the results from the first trading strategy, both the long and short legs yield significant alphas (2.69
bps for the long leg and -4.70 for the short leg).
A caveat with this long/short strategy is that it assumes the return direction (positive or negative) is the
same as the order imbalance direction, which might not be true if the price movement during the day were
information-driven. Another assumption is that stocks with a high market-on-close order imbalance can be
quickly identified, using the record of the TAQ data, which may not be available until after the market closes.
This strategy assumes that orders can be placed at the closing price, which may not be realistic. Given the
fact that the price reversal is persistent during the overnight period and also the subsequent intra day, modified
trading strategies that hold stock during the overnight period and the next intraday period are also considered.
Table VIII, Panel C reports the different holding period returns from these two trading strategies. The
holding horizon starts either from the same day’s close to the next morning’s open (i.e., the overnight return
in Columns 1 and 3) or from next morning’s open to the next day’s close (i.e., the open-to-close return in
Columns 2 and 4). The results show that the return for the long–short portfolio comes not only from the
overnight period (5.55 bps with a t-stat of 13.83) but also from the open-to-close period (7.84 bps with a t-stat
of 7.38).
As a robustness check, Table A.VI shows the results of the subperiod analyses. The results are robust
across different sample periods. The fact that the alphas from the second period are lower than the first period
is consistent with the conjecture that the market becomes more efficient regarding liquidity provision during
the closing auction. Value-weighted portfolios are also formed under both trading strategies, and the results
presented in Table A.VII show consistent but weaker abnormal returns, suggesting some of the abnormal
returns can be attributed to the liquidity constraints from the small cap stocks.
While the size and significance of such a long–short strategy indicates the return predictability based on
closing auction order imbalances, transaction costs for individual investors could erode these profits. Never-
theless, it is possible for professional investors (e.g., high-frequency traders) with low trading costs to exploit
such a trading strategy. Another concern for such a trading strategy is information availability, especially the
closing auction volume. These trading strategies suggest that providing liquidity to passive funds by assessing
the closing auction order imbalances is more profitable than using the public TAQ data.
Major exchanges provide the closing auction feeds that provide information about closing auction volume at high frequency
during the closing auction. The subscription fee to the stock-exchange market-data feed is not cheap, and it has increased recently,
Overall, the takeaway from this section is that the price reversal, due to the closing auction order im-
balance, caused by ETF flows, can be exploited. And, by providing liquidity to passive funds, the trading
strategies result in significant abnormal returns.
4.4 Closing auction, ETF ownership, and volatility
Ben-David, Franzoni, and Moussawi (2018) find that stocks with higher ETF ownership display signifi-
cantly higher volatility because the liquidity shocks propagate to the underlying securities through the arbi-
trage channel. Besides the continuous arbitrage activities throughout the day, closing auction orders placed
by passive funds and authorized participants could cause additional volatility to the underlying securities due
to the price impact and the price reversals documented here. To verify this conjecture, I first replicate the
main analysis in Ben-David, Franzoni, and Moussawi (2018) using S&P 500 stocks from 2010 to 2018 by
regressing volatility on ETF ownership and controls. The ETF ownership is defined as follows.
ET F ownership
i, j,t
Following their procedure, the analysis is conducted at the monthly frequency, where the daily volatility is
computed using all daily returns within the month. All controls are from the end of the prior month. Standard
errors are double clustered at the stock and month level. Volatility and ETF ownership are standardized by
subtracting the sample mean and then dividing by the sample standard deviation in the entire sample. To
purge the possible effect of the closing auction, I calculate the daily return using the price from opening
and the transaction price at 15:45. Then the monthly alternative volatility is computed based on this return
measure. The correlation between these two volatility measures is around 0.77.
[Table IX here]
Table IX presents the results. Columns 1 and 2 show that, consistent with the results from Ben-David, Fran-
zoni, and Moussawi (2014), stocks with higher ETF ownership have higher volatility, which is computed
based on the close-to-close return. The magnitude of the coefficient resembles the one they document: a
though the SEC recently overturned the approval of increasing higher fees for certain feed data. See
one-standard-deviation increase in ETF ownership is associated with a 10.5% of a standard deviation change
in daily volatility. Columns 3 and 4 show the same panel regression results with the volatility based on an al-
ternative return measure. The magnitude of the relationship between ETF ownership and volatility decreases
by 42.6% (60.4%) to 0.060 (0.026) in Column 3 (Column 4). These results provide suggestive evidence that
the closing auction might contribute to the higher volatility of stocks with higher ETF ownership.
4.5 Do informed traders use market-on-close orders?
The results so far have focused on the impact of passive investing on stock price dynamics through the
channel of market-on-close orders, based on the premise that passive funds and authorized participants are
the main users of market-on-close orders. However, it remains unclear whether market-on-close orders could
be used by other market participants. Models, such as that of Admati and Pfleiderer (1988), shows that the
increasing level of liquidity trading induces more informed trading, consistent with the empirical finding that
the average volume of shares traded is U-shaped and clustered within one day. Given the increasing intensity
of market-on-close volume over the past decade, are informed traders induced to place market-on-close orders
to camouflage their trading intentions? In this section, I provide some suggestive evidence supporting this
To begin with, I construct the following measure to quantify market-on-close trades that do not originate
with passive funds:
Volume residual
ET F f lows,i,t
total tradingi,t
where $Volume
represents the total dollar trading volume of the market-on-close orders, and $Volume
ET F f lows,i,t
represents the total aggregate ETF flows in dollar volume. The difference between these two is then normal-
ized by the total daily trading volume.
[Figure 2 here]
Figure 2 provides the time-series trend of this measure, averaged cross-sectionally among S&P 500 stocks
and all common stocks. The pattern shows that over time, the market-on-close trading volume that does not
originate from the ETF flows have increased. This evidence is consistent with the hypothesis that other market
participants might also use market-on-close orders. However, caution is called for regarding this noisy proxy:
First of all, due to the limits of data availability, I can capture only the daily flows from ETFs, not the daily
flows from index mutual funds.
The increasing trend could be due to the unmeasured flows from index
mutual funds and other index-linked trading strategies. Second, the measure assumes ETF flows are being
executed 100% by the closing auction, which might not be true. Therefore, I conducted two additional tests.
4.5.1 Closing auction, informed trading, and price reversals
The main analysis of price reversals shows that stocks with high closing auction order imbalances will
reverse because their prices are driven by passive flows that are not related to fundamental. However, if the
price impact from the high closing auction is due to private information, the price impact will be permanent
and will not induce negative serial correlation (Glosten and Milgrom (1985) and Nagel (2012)). Among stocks
in the high quintile of closing auction volume residual, defined in Equation (10) cross-sectionally, market-on-
close orders with a direction of sell (buy) but with positive (negative) aggregate ETF inflows are more likely
to be information-driven orders. In other words, a stock that is sold (bought) by aggregate ETFs (inferred
based on the direction of flows) but has an aggregate buying (selling) market-on-close order imbalance might
be a stock traded by informed traders. To formally test this, I repeat the same Fama-Macbeth regression based
on the specification in Equation (8) across different subsamples.
[Table X here]
Table X presents the results. The first column is the same as in Table VII, including all NYSE stocks with
standard filters. Both the coefficients on the interaction terms are significant, and the estimate in Column 1 are
slightly different from those in Table VII because the sample analysis here requires that the aggregated ETF
flows at the stock level cannot be missing. The second column repeats the analysis for those stocks that are less
likely to have information-driven closing trades, and the last column shows the coefficient estimates for those
A number of studies that focus on daily mutual fund flows, such as those by Chalmers, Edelen, and Kadlec (2001); Edelen and
Warner (2001); Greene and Hodges (2002); Rakowski (2010); Kaniel and Parham (2017); and Agarwal, Jiang, and Wen (2018) use
TrimTab data, which rely on voluntary disclosure from fund managers and therefore have limited coverage of the funds. For instance,
TrimTab data do not cover passive funds from Vanguard and Blackrock, which account for more than 50% of the market share of
passive funds, according to Morningstar.
stocks that are more likely to have information-driven closing trades. Consistent with the conjecture, the price
reversal for stocks that are more likely to have information-driven closing trades is weaker and insignificant.
4.5.2 Closing auction and informed trading around earnings announcement
In this subsection, I utilize earnings announcements as a laboratory setting to provide further evidence
that market-on-close orders could be used by informed traders. The goal of earnings announcements, as
major company information events, is to release information to the market, and informed investors should be
especially active at these times.
For instance, Jegadeesh and Titman (1993) estimate that the three-day returns around earnings announce-
ments represent approximately 25% of momentum profits. La Porta, Lakonishok, Shleifer, and Vishny (1997)
report that about 25% of the returns to various value strategies considered by Lakonishok, Shleifer, and Vishny
(1994) are concentrated on the three days around earnings announcements. More recently, Engelberg, Mclean,
and Pontiff (2018) investigate 97 stock return anomalies and find that anomaly returns are six times higher on
earnings announcement days.
This evidence suggests that the short window around earnings announcements is a period in which in-
formed trades occur. Market-on-close orders could be used by informed traders as a way to enter or exit the
positions since trading along with passive funds using market-on-close trades can hide their intentions and
increase the profits they gain from private information.(e.g., Admati and Pfleiderer (1988)).
To begin with, I follow the literature (Dellavigna and Pollet (2009); Kaniel, Liu, Saar, and Titman (2012);
and Hirshleifer, Lim, and Teoh (2009)) by constructing the normalized earnings surprise and abnormal re-
turns during the earnings announcement as proxies for the surprise. Specifically, I define the standardized
unexpected earnings (SUE) as actual earnings minus the earnings forecast then divided by the price on the
forecast date. The earnings forecast is the mean analyst forecast one month before the earnings announcement.
Another earnings surprise proxy is the abnormal return at the time of the earnings announcement, which is
defined as CAR[-1,1], the cumulative buy-and-hold return from one day before the earnings announcement to
one day after, adjusted for the characteristics benchmarks described in Daniel, Grinblatt, Titman, and Wermers
(1997) (i.e., size, book-to-market, and momentum).
To measure the abnormal market-on-close trading volume that does not derive from passive ETF flows, I
follow Dellavigna and Pollet (2009) and define
Volume residual
/(H h + 1)
Volume residual
/10 (11)
where Volume residual
is calculated in Equation (10) as the market-on-close trading residual on day u and
τ is the date of the earnings announcement in quarter t for company i. The measure V
is the percentage
increase in volume around the announcement date at horizon (h,H). In particular, I investigate three abnormal
volume during the three horizons: V
, V
, and V
In each calendar quarter from 2010 to 2018, quarterly earnings announcement observations are sorted
in that quarter based on the absolute value of the earnings surprise (CAR[-1,1]). I then calculate the mean
abnormal market-on-close residual trading volumes before the announcement, around the announcement, and
after the announcement. The abnormal trading volume difference between earnings surprise quintiles 5 and 1
captures the trading volume difference that responds to different magnitudes of earnings surprises.
[Table XI here]
Table XI, Panel A reports the results. During the announcement period [-1, 1] and the post-announcement
period [2, 4], there are significant increases in abnormal trades that use the market-on-close orders. Further-
more, stocks in the top quintile of earning surprises experience significantly higher abnormal market-on-close
trading volume than stocks in the bottom quintile of earnings surprises across three periods. There is little
evidence of increasing abnormal market-on-close trades in the pre-announcement period [-4, -2] for stocks in
the bottom quintile of earnings surprises. An analysis using the cumulative abnormal return as a proxy for
earnings surprise shows a similar pattern.
Table XI, Panel B takes a step further to analyze the trading imbalance pattern during the earnings an-
nouncement periods. By leveraging the direction of both ETF flows and order imbalances, I construct the
following signed measure of market-on-close trades:
Trade Imb residual
Signed $Volume
closing auction,i,t
Signed $Volume
ET F f lows,i,t
total tradingi,t
As mentioned earlier, this measure has two limitations that can undermine the statistical power of empirical
tests. First, it captures only the flows from ETFs, not from index mutual funds. Second, it assumes that
all ETF flows are executed entirely through the closing auction. Nevertheless, I attempt to provide the best
possible evidence using this measure. The abnormal trading imbalances are constructed in a way similar to
abnormal trading volume in Equation (11), with a negative sign meaning abnormal selling and a positive sign
meaning abnormal buying.
One distinct pattern in Panel B is the substantial buying for stocks with negative earnings surprises and
selling for stocks with positive earnings surprises in the post-earnings announcement period [2, 4]. This
suggests that investors use market-on-close orders to exit the position they accumulated during or before the
earnings announcement. This result is consistent with the pattern of “buying the rumor and selling the news”
in the context of individual investors (Kaniel, Liu, Saar, and Titman (2012)) as well as institutional investors
(Kadan, Michaely, and Moulton (2018)).
Taken together, the results in this section provide several lines of supportive evidence that the market-
on-close orders can be used by informed traders. This is consistent with the theoretical implication that high
liquidity trades induce more informed trades.
5 Conclusion
This paper investigates the popularity of increasing closing auction trades. First, I show that due to the
increase in passive investing, market-on-close trading volumes are increasing dramatically. Stocks with more
passive flows from ETFs have higher market-on-close trading volume than otherwise similar securities. Using
a quasi-natural experiment based on the reconstitution of the Russell indices, I provide a causal interpretation
of this result.
Next, I investigate the impact of increasing market-on-close trading volume on stock price dynamics. I
show that stocks with high market-on-close trading volume have larger price movement during the end of
day auction period. Furthermore, since the demand shock from ETFs is nonfundamental, stocks that have
high market-on-close trading volume cross-sectionally show a strong return reversal pattern that is consistent
with the temporary price pressure hypothesis, and investors are compensated by providing liquidity to passive
investors. A long–short trading strategy that exploits this temporary price pressure results in a significant
risk-adjusted 13.2 bps per day, or 33.26% annualized return.
Besides passive funds, market-on-close orders might also be used by informed traders, consistent with the
theoretical implications of the clustering of informed trades and liquidity trades.
Overall, these findings provide direct evidence that the closing auction is an important channel through
which demand shocks from ETFs can propagate into underlying securities. The significant price impact
for stocks that have high market-on-close trading volume suggests that the closing price for some stocks
may be distorted due to the increasing volume of market-on-close orders, which originate from aggregate
passive flows. My findings have important implications for investor welfare. If the popularity of market-on-
close orders causes price to deviate from their fundamental value, then some passive funds may benefit from
executing their orders at other than closing prices at the expense of larger tracking errors.
Admati, Anat R., and Paul Pfleiderer, 1988, A Theory of Intraday Patterns: Volume and Price Variability, The
Review of Financial Studies 1, 3–40.
Admati, Anat R., and Paul Pfleiderer, 1991, Sunshine Trading and Financial Market Equilibrium, The Review
of Financial Studies 4, 443–481.
Agarwal, Vikas, Paul Hanouna, Rabih Moussawi, and Christof Stahel, 2019, Do ETFs Increase the Common-
ality in Liquidity of Underlying Stocks?, Working Paper .
Agarwal, Vikas, Lei Jiang, and Quan Wen, 2018, Why Do Mutual Funds Hold Lottery Stocks?, Working
paper .
Akbas, Ferhat, Will J. Armstrong, Sorin Sorescu, and Avanidhar Subrahmanyam, 2015, Smart money, dumb
money, and capital market anomalies, Journal of Financial Economics 118, 355–382.
Amihud, Yakov, 2002, Illiquidity and stock returns: cross-section and time-series effects, Journal of Financial
Markets 5, 31–56.
Anand, Amber, Paul Irvine, Andy Puckett, and Kumar Venkataraman, 2012, Performance of institutional
trading desks: An analysis of persistence in trading costs, Review of Financial Studies 25, 557–598.
on, Miguel, and Christopher Polk, 2014, Connected Stocks, Journal of Finance 69, 1099–1127.
Appel, Ian R., Todd A. Gormley, and Donald B. Keim, 2016, Passive investors, not passive owners, Journal
of Financial Economics 121, 111–141.
Appel, Ian R, Todd A Gormley, and Donald B Keim, 2018, Standing on the shoulders of giants: The effect of
passive investors on activism, The Review of Financial Studies 32, 2720–2774.
Ben-David, ITZHAK, Francesco Franzoni, Augustin Landier, and Rabih Moussawi, 2013, Do hedge funds
manipulate stock prices?, The Journal of Finance 68, 2383–2434.
Ben-David, Itzhak, Francesco Franzoni, and Rabih Moussawi, 2014, Do ETFs Increase Volatility?, Technical
Report w20071, National Bureau of Economic Research, Cambridge, MA.
Ben-Rephael, Azi, Shmuel Kandel, and Avi Wohl, 2011, The Price Pressure of Aggregate Mutual Fund Flows,
The Journal of Financial and Quantitative Analysis 46, 585–603.
Ben-David, Itzhak, Francesco Franzoni, and Rabih Moussawi, 2018, Do ETFs Increase Volatility?, Journal
of Finance, forthcoming .
Beneish, Messod D., and Robert E. Whaley, 1996, An Anatomy of the ”S & P Game”: The Effects of
Changing the Rules, Journal of Finance 51, 1909–1930.
Blume, Lawrence, David Easley, and Maureen O’Hara, 1994, Market Statistics and Technical Analysis: The
Role of Volume, Journal of Finance 49, 153–181.
Boehmer, Ekkehart, Charles M. Jones, and Xiaoyan Zhang, 2008, Which Shorts Are Informed?, Journal of
Finance 63, 491–527.
Bogousslavsky, Vincent, 2018, The Cross-Section of Intraday and Overnight Returns, Working Paper .
Box, Travis, Ryan Davis, Richard Evans, and Andrew Lynch, 2019, Intraday Arbitrage Between ETFs and
Their Underlying Portfolios, Working paper 73.
Breen, William J, Laurie Simon Hodrick, and Robert A Korajczyk, 2002, Predicting Equity Liquidity, Man-
agement Science 48, 15.
Broman, Markus S, and Pauline Shum, 2018, Relative liquidity, fund flows and short-term demand: Evidence
from exchange-traded funds, Financial Review 53, 87–115.
Brown, David C., Shaun Davies, and Matthew Ringgenberg, 2018, ETF Arbitrage and Return Predictability,
SSRN Scholarly Paper ID 2872414, Social Science Research Network, Rochester, NY.
Campbell, John Y., Sanford J. Grossman, and Jiang Wang, 1993, Trading Volume and Serial Correlation in
Stock Returns, The Quarterly Journal of Economics 108, 905–939.
Chalmers, John M. R., Roger M. Edelen, and Gregory B. Kadlec, 2001, On the Perils of Financial Intermedi-
aries Setting Security Prices: The Mutual Fund Wild Card Option, Journal of Finance 56, 2209–2236.
Chan, Louis K. C., and Josef Lakonishok, 1993, Institutional trades and intraday stock price behavior, Journal
of Financial Economics 33, 173–199.
Chan, Louis K. C., and Josef Lakonishok, 1995, The Behavior of Stock Prices Around Institutional Trades,
Journal of Finance 50, 1147–1174.
Chang, Yen-Cheng, Harrison Hong, and Inessa Liskovich, 2015, Regression Discontinuity and the Price
Effects of Stock Market Indexing, The Review of Financial Studies 28, 212–246.
Chordia, Tarun, and Avanidhar Subrahmanyam, 2004, Order imbalance and individual stock returns: Theory
and evidence, Journal of Financial Economics 72, 485–518.
Coles, Jeffrey L., Davidson Heath, and Matthew Ringgenberg, 2017, On Index Investing, Working paper .
Collin-Dufresne, Pierre, and Kent Daniel, 2016, Liquidity and Return Reversals, Working paper 42.
Comerton-Forde, Carole, and T
alis J Putnin¸
s, 2011, Measuring closing price manipulation, Journal of Finan-
cial Intermediation 20, 135–158.
Conrad, Jennifer S., Allaudeen Hameed, and Cathy Niden, 1994, Volume and Autocovariances in Short-
Horizon Individual Security Returns, Journal of Finance 49, 1305–1329.
Coval, Joshua, and Erik Stafford, 2007, Asset fire sales (and purchases) in equity markets, Journal of Financial
Economics 86, 479–512.
Cushing, David, and Ananth Madhavan, 2000, Stock returns and trading at the close, Journal of Financial
Markets 3, 45–67.
Da, Zhi, and Sophie Shive, 2018, Exchange traded funds and asset return correlations, European Financial
Management 24, 136–168.
Daniel, Kent, Mark Grinblatt, Sheridan Titman, and Russ Wermers, 1997, Measuring Mutual Fund Perfor-
mance with Characteristic-Based Benchmarks, Journal of Finance 52, 1035–1058.
Dellavigna, Stefano, and Joshua M. Pollet, 2009, Investor Inattention and Friday Earnings Announcements,
Journal of Finance 64, 709–749.
Duffie, Darrell, 2010, Presidential address: Asset price dynamics with slow-moving capital, The Journal of
finance 65, 1237–1267.
Edelen, Roger M, and Jerold B Warner, 2001, Aggregate price effects of institutional trading: a study of
mutual fund flow and market returns, Journal of Financial Economics 59, 195–220.
Engelberg, Joseph, R. David Mclean, and Jeffrey Pontiff, 2018, Anomalies and News, Journal of Finance 73,
Etula, Erkko, Kalle Rinne, Matti Suominen, and Lauri Vaittinen, 2019, Dash for Cash: Monthly Market
Impact of Institutional Liquidity Needs, The Review of Financial Studies hhz054.
Fama, Eugene F., and Kenneth R. French, 1988, Permanent and Temporary Components of Stock Prices,
Journal of Political Economy 96, 246–273.
Fama, Eugene F., and James D. MacBeth, 1973, Risk, Return, and Equilibrium: Empirical Tests, Journal of
Political Economy 81, 607–636.
Frazzini, Andrea, and Owen A. Lamont, 2008, Dumb money: Mutual fund flows and the cross-section of
stock returns, Journal of Financial Economics 88, 299–322.
Gao, Lei, Yufeng Han, Sophia Zhengzi Li, and Guofu Zhou, 2018, Market intraday momentum, Journal of
Financial Economics 129, 394–414.
Glosten, Lawrence R, and Lawrence E Harris, 1988, Estimating the components of the bid/ask spread, Journal
of Financial Economics 21, 123–142.
Glosten, Lawrence R, and Paul R Milgrom, 1985, Bid, ask and transaction prices in a specialist market with
heterogeneously informed traders, Journal of Financial Economics 14, 71–100.
Glosten, Lawrence R, Suresh Nallareddy, and Yuan Zou, 2016, Etf activity and informational efficiency of
underlying securities, Columbia Business School Research Paper .
Goetzmann, William N., and Massimo Massa, 2003, Index Funds and Stock Market Growth, Journal of
Business 76, 1–28.
Greene, Jason T, and Charles W Hodges, 2002, The dilution impact of daily fund flows on open-end mutual
funds, Journal of Financial Economics 65, 131–158.
Greenwood, Robin, 2008, Excess Comovement of Stock Returns: Evidence from Cross-Sectional Variation
in Nikkei 225 Weights, Review of Financial Studies 21, 1153–1186.
Greenwood, Robin, and David Thesmar, 2011, Stock price fragility, Journal of Financial Economics 102,
Harris, Lawrence, 1986, A transaction data study of weekly and intradaily patterns in stock returns, Journal
of Financial Economics 16, 99–117.
Hendershott, Terrence, Dmitry Livdan, and Dominik R
osch, 2018, Asset Pricing: A Tale of Night and Day,
Working paper .
Heston, Steven L, Robert A Korajczyk, and Ronnie Sadka, 2010, Intraday patterns in the cross-section of
stock returns, Journal of Finance 65, 1369–1407.
Hillion, Pierre, and Matti Suominen, 2004, The manipulation of closing prices, Journal of Financial Markets
7, 351–375.
Hirshleifer, David, Sonya Seongyeon Lim, and Siew Hong Teoh, 2009, Driven to Distraction: Extraneous
Events and Underreaction to Earnings News, Journal of Finance 64, 2289–2325.
Holden, Craig W., and Stacey Jacobsen, 2014, Liquidity Measurement Problems in Fast, Competitive Mar-
kets: Expensive and Cheap Solutions, Journal of Finance 69, 1747–1785.
Israeli, Doron, Charles M. C. Lee, and Suhas A. Sridharan, 2017, Is there a dark side to exchange traded
funds? An information perspective, Review of Accounting Studies 22, 1048–1083.
Jegadeesh, Narasimhan, 1990, Evidence of Predictable Behavior of Security Returns, Journal of Finance 45,
Jegadeesh, Narasimhan, and Sheridan Titman, 1993, Returns to Buying Winners and Selling Losers: Impli-
cations for Stock Market Efficiency, Journal of Finance 48, 65–91.
Kadan, Ohad, Roni Michaely, and Pamela C. Moulton, 2018, Trading in the Presence of Short-Lived Private
Information: Evidence from Analyst Recommendation Changes, Journal of Financial and Quantitative
Analysis 53, 1509–1546.
Kaniel, Ron, Shuming Liu, Gideon Saar, and Sheridan Titman, 2012, Individual Investor Trading and Return
Patterns around Earnings Announcements, Journal of Finance 67, 639–680.
Kaniel, Ron, and Robert Parham, 2017, WSJ Category Kings The impact of media attention on consumer
and mutual fund investment decisions, Journal of Financial Economics 123, 337–356.
Korajczyk, Robert A., and Ronnie Sadka, 2004, Are Momentum Profits Robust to Trading Costs?, Journal of
Finance 59, 1039–1082.
Kraus, Alan, and Hans R. Stoll, 1972, Price Impacts of Block Trading on the New York Stock Exchange,
Journal of Finance 27, 569–588.
La Porta, Rafael, Josef Lakonishok, Andrei Shleifer, and Robert Vishny, 1997, Good News for Value Stocks:
Further Evidence on Market Efficiency, Journal of Finance 52, 859–874.
Lakonishok, Josef, Andrei Shleifer, and Robert W. Vishny, 1992, The impact of institutional trading on stock
prices, Journal of Financial Economics 32, 23–43.
Lakonishok, Josef, Andrei Shleifer, and Robert W. Vishny, 1994, Contrarian Investment, Extrapolation, and
Risk, Journal of Finance 49, 1541–1578.
Lee, Charles M. C., and Mark J. Ready, 1991, Inferring Trade Direction from Intraday Data, Journal of
Finance 46, 733–746.
Lou, Dong, 2012, A Flow-Based Explanation for Return Predictability, Review of Financial Studies 25, 3457–
Lou, Dong, Christopher Polk, and Spyros Skouras, 2018, A Tug of War: Overnight Versus Intraday Expected
Returns, Journal of Financial Economics, forthcoming 51.
Lynch, Anthony W., and Richard R. Mendenhall, 1997, New Evidence on Stock Price Effects Associated with
Changes in the S&P 500 Index, The Journal of Business 70, 351–383.
Mayhew, Stewart, Tim McCormick, and Chester S. Spatt, 2009, The Information Content of Market-on-Close
Imbalances, the Specialist and NYSE Equity Prices, Working paper .
Nagel, Stefan, 2012, Evaporating Liquidity, The Review of Financial Studies 25, 2005–2039.
Newey, Whitney K., and Kenneth D. West, 1987, A Simple, Positive Semi-Definite, Heteroskedasticity and
Autocorrelation Consistent Covariance Matrix, Econometrica 55, 703–708.
Pan, Kevin, and Yao Zeng, 2019, Etf arbitrage under liquidity mismatch, Working Paper .
Petersen, Mitchell A., 2009, Estimating Standard Errors in Finance Panel Data Sets: Comparing Approaches,
Review of Financial Studies 22, 435–480.
Rakowski, David, 2010, Fund Flow Volatility and Performance, The Journal of Financial and Quantitative
Analysis 45, 223–237.
Schmidt, Cornelius, and R
udiger Fahlenbrach, 2017, Do exogenous changes in passive institutional ownership
affect corporate governance and firm value?, Journal of Financial Economics 124, 285–306.
Scholes, Myron, and Joseph Williams, 1977, Estimating betas from nonsynchronous data, Journal of Finan-
cial Economics 5, 309–327.
Scholes, Myron S., 1972, The Market for Securities: Substitution Versus Price