Sniping to Manipulate Closing Prices in Call Auctions:

Evidence from the Hong Kong Stock Exchange*

Wing Suen
University of Hong Kong
(wsuen@econ.hku.hk)
and

Kam-Ming Wan
University of Hong Kong
(kmwan@econ.hku.hk)

Abstract. The Hong Kong Stock Exchange adopted call auctions

as the closing mechanism for a brief period. We find evidence of
abnormally large orders and large price changes during the last ten
seconds of the auction sessions. Such sniping attacks were
associated with the expiration of derivative products, which
provided an incentive for price manipulation. Prices had a
tendency to revert the day following a sniping attack. While prices
were, on average, more informative under a closing auction
procedure than under a random closing procedure, they were less
robust to manipulation attempts.

Last Revised: February 15, 2011

First Version: May 17, 2010

*
We are grateful to Matthias Buehlmaier, Y.K. Fu, Avanidhar Subrahmanyam, Harold Zhang, and seminar
participants at the University of Hong Kong for their comments. We are also thankful to Yangbo (Darcy) Song for
his excellent research assistance. All errors remain ours.

 I. Introduction
Call auctions are the dominant mechanism to determine closing prices of equities around the

world. Eight of the top ten stock market operators in the world currently adopt this mechanism.

The only exceptions are the Hong Kong Stock Exchange (henceforth HKEx) and Shanghai

Stock Exchange. Closing call auctions are also prevalent among top Asian stock exchanges,

including Tokyo Stock Exchange, Australian Stock Exchange, Korea Exchange, and Singapore

Stock Exchange. They are popular because extant studies (e.g., Pagano and Schwartz, 2003;

Schwartz, 2001; Cao, Ghysels, and Hatheway, 2000; Biais, Hillion, and Spatt, 1999; Madhavan,

1992) find that they improve price discovery and reduce price volatility. Other studies provide

anecdotal evidences to suggest that closing call auctions reduce price manipulation. Specifically,

they (e.g., Comerton-Forde, Lau, and McInish (2007) for the Singapore Stock Exchange; Hillion

and Souminen (2004) for the Madrid Stock Exchange; and Felixson and Pelli (1999) for the

Helsinki Stock Exchange) find that the closing price is subject to manipulation when it is

determined by other closing mechanisms.

The HKEx actually had adopted a closing auction mechanism for a brief period from

May 26, 2008 to March 22, 2009. In sharp contrast to the extant findings, the auction procedure

was eventually abandoned because of widespread suspicion of price manipulation. We take

advantage of this episode to examine the effects of closing auctions on price discovery and on

price manipulation. Consistent with conventional wisdom, we find that the closing auction

mechanism does, on average, improve the informativeness of the closing price. Yet, we also find

that a plain-vanilla closing call auction as the one adopted by the HKEx is susceptible to price

manipulation, particularly in the form of unexpectedly large (buy or sell) orders in the last few

seconds. We refer to such actions that cause abnormal patterns in prices and trading volume

during the last few seconds before the close as sniping.

 A case in point is the plunge in HSBC shares by 24.1 percent on March 9, 2009. The

plunge is the biggest one-day drop since the Black Monday in 1987, and its impact is

exceptionally large because HSBC is the largest HKEx-listed company with a worldwide market

capitalization of about US$200 billion by the end of 2009. On that day, HSBC shares dropped

from $43.5 at the previous close to $37.7 by the end of the continuous trading session at 4:00

p.m.1 The indicative stock price of HSBC fluctuated within a tight range of $37$38 during the

closing auction session (henceforth CAS) from 4:00 p.m. to 4:10 p.m. Just two seconds before

the closing auction was concluded, a trader sniped with an exceptionally large market sell order.2

HSBC shares immediately plunged from $37 and closed at $33, a drop of over ten percent in

merely two seconds. Coincidentally, when the market opened the next day, HSBC shares

bounced back to $37.25. This opening price was very close to the indicative market clearing

price three seconds before the close in the previous day. Panel (a) of Figure 1 presents the

indicative equilibrium price (IEP) and the primary buy and sell queue of HSBC shares during

that closing auction session. The sniping order was remarkably large, at over 4.7 million shares.

That order alone was more than fifty percent of the then prevailing indicative equilibrium

volume (IEV) and had a market value of approximately US$19.8 million.

[Insert Figure 1 here.]

We believe that the sniping incident on March 9, 2009 was well-planned because trading

of HSBC shares was also unusual during the CAS on the previous trading day (March 6, 2009).

Panel (b) of Figure 1 shows the trading activities during the CAS on that day. First, note that a

small but unusually aggressive limit sell order of $33 was submitted between 4:01 p.m. and

1
Stock prices are quoted in the Hong Kong dollar while values on transactions and firm market capitalization are
quoted in the U.S. dollar for ease of comparison. The Hong Kong dollar is pegged to the U.S. dollar at an exchange
rate of US$1:HK$7.8.
2
See the news report in Hong Kong Economic Times on March 10, 2009.

4:02p.m., causing the IEP to fall immediately to $33. Coincidentally, this IEP at $33 is identical

to the closing price of the following trading day. This limit sell order was aggressive because it

was submitted at a deep discount from the last transacted price of $43.15 at 4:00 p.m. Second,

this order was subsequently canceled even though that seller could have sold the shares at a price

significantly higher than $33. Third, an exceptionally large sell order of above 5.8 million shares

was canceled just one second prior to the cancellation deadline at 4:08p.m, which can be

considered as another form of sniping. It appears that some traders attempted to manipulate the

stock price of HSBC on March 6, 2009 but was not successful. Alternatively, they might have

tested the market reaction should HSBC shares drop to $33.3

Last-minute bidding (or sniping) is prevalent at eBay auctions and the practice is

extensively documented and studied by, among others, Roth and Ockenfels (2002; 2006), Bajari

and Hortacsu (2003), and Ely and Hossain (2009). Closing call auctions are different from eBay

and many on-line auctions in that both the buy side and the sell side submit bids and in that

multiple units are traded. The objectives of sniping in these markets are not the same. In eBay

auctions, sniping is a strategy to acquire an object at a low price. In financial markets, the

objective is to influence the market price directly, perhaps with a view to making money in

derivative markets or manipulating market sentiments. These differences notwithstanding, we

find evidence that sniping does occur in financial markets, especially when the call auction

mechanism does not contain safeguards against such manipulative behavior.

Our paper adds to the literature in several ways. First, in contrast to previous findings,

we find that a plain vanilla closing call auction is fragile and vulnerable to price manipulation.

3
There is anecdotal evidence for the practice of similar manipulative order placing activities during the CAS period.
The Hong Kong Securities and Future Commission had successfully prosecuted two individuals for such activities.
Details on these two cases are available from
http://www.sfc.hk/sfcPressRelease/EN/sfcOpenDocServlet?docno=09PR120 and
http://www.sfc.hk/sfcPressRelease/EN/sfcOpenDocServlet?docno=09PR122.

The issue of vulnerability of price manipulations is important to exchange operators because

such manipulations can hinder price discovery, which is a core quality of market viability (see

OHara, 2001; 2007) because it influences the risk of holding stocks. The fragility of the CAS is

partly due to the fact that allowable price changes at the auction procedure can be very

largefrom +900% to 89% of the prevailing market price. In contrast, the allowable price

change during the regular trading session is only 24 spreads, which is approximately 2.4% of

the prevailing market price. The possibility of large and sudden price movement during the CAS

increases the potential profits that manipulators can gain from sniping and reduces the scope for

other traders to react. The experience of closing price manipulation in the HKEx highlights the

importance of market design.

Second, many studies (e.g., Comerton-Forde and Putnins, 2010; Aggarwal and Wu,

2006) use prosecuted manipulation cases and find that manipulations happen typically in small

and illiquid stocks. The evidence presented in this paper suggests that closing price

manipulations can occur even for the most liquid stocks in one of the largest worldwide stock

exchanges when sufficient safeguards against manipulations are not in place. Our results also

shed light on the limitation of using prosecuted cases to draw inferences on the extent of price

manipulation. This is because the number of prosecuted cases is directly related to the capability

of regulators to accurately detect and successfully prosecute price manipulators.

Third, our findings are less susceptible to possible biases from confounding factors that

are correlated with time trend. Existing studies of other markets typically rely on a before-and-

after comparison, which may be difficult to disentangle from unobserved time trend. The unique

experience of the HKEx to adopt and then abandon closing auctions allows us to use both pre-

and post-CAS periods as two separate benchmarks to examine their effects. Unless one posits an

unobserved time trend that coincidentally reverts itself after the CAS period, we are more

confident that the effects so identified during the CAS period are attributable to the trading

mechanism rather than to other unobserved factors.

Fourth, we use market micro data at very fine time intervals (10-second and 10-minute

intervals) to measure price movements and trading activities. Focusing on such fine details

allows us to isolate the effects of market structure from the effects of broader market movements.

More importantly, in addition to measuring the effects of closing auctions, we also study

a motive to manipulate closing prices at the auctions. In particular, we link the performance of

the closing auctions in the equities market to the market for a derivative productcallable bull

and bear contracts (CBBCs). These contracts are similar to knock-out (barrier) options in that

they would be expired should the price of the underlying stock reach a pre-specified level. Yet,

unlike traditional knock-out options, CBBCs are typically not worthless on the expiration date.

The residual value of an expired CBBC can be a function of the closing price of the underlying

stock on the expiration day. Consistent with our expectations, we find that sniping is more likely

to occur on days when CBBCs are expired, particularly during the CAS period. We also

compare the informativeness of closing prices of the closing auction mechanism with that of the

random closing mechanism. Overall, we find that closing prices are more informative under the

closing auction mechanism, but they are less informative under the closing auction system on

days when the incentive to manipulate prices is high.

II. Closing Call Auctions vs. Random Closing at the HKEx

The HKEx is the seventh largest stock exchange ranked by domestic market capitalization and

the largest stock exchange ranked by money raised through initial public offerings in 2009 and

2010.4 Regular trading at the HKEx is conducted in two continuous trading sessions: (i) a

morning session from 10:00 a.m. to 12:30 p.m., and (ii) an afternoon session from 2:30 p.m. to

4:00 p.m. The following describes the closing procedures adopted by the HKEx during the CAS

and non-CAS periods.

Closing Auction Session

During a period of approximately ten months since May 26, 2008, the HKEx adopted call

auctions to determine closing prices. This mechanism was suspended indefinitely on March 23,

2009, soon after the plunge in HSBC shares incident. Under the closing auction mechanism,

only at-auction orders and at-auction limit orders are allowed. An at-auction order is an order

with a specified quantity but no specified price while an at-auction limit order is an order with a

specified price and quantity. At-auction orders are similar to regular market orders while at-

auction limit orders are to regular limit orders.5 The only exception is that the price specified by

the at-auction limit order is bounded by the 9-times restriction rule: the upper price limit of

submitted orders is 9 times (+900%) the nominal price at 4:00 p.m. and the lower price limit of

submitted orders is 1/9 times (89%) the price at 4:00 p.m.6 This cap is rarely a binding

constraint as the allowable price movement is huge. In contrast, limit orders submitted during

the continuous trading session are restricted to much narrower bands. Specifically, limit orders

must be inputted within 24 spreads of the prevailing nominal price (henceforth 24-spread rule).

One spread is identical to one tick and the tick size is a function of the stock price. For a $50

4
Statistics on the ranking of world stock exchanges are available from the World Federation of Exchanges.
Information on IPO ranking was reported in Hong Kong Seeking IPO 3-Peat in 2011, Wall Street Journal, January
2, 2011.
5
For the sake of simplicity, we refer at-auction orders as market orders whereas at-auction limit orders as limit
orders throughout this paper.
6
See Section 5.2.7 of the Market Operations and Trading of the HKEx at
http://www.hkex.com.hk/eng/global/faq/securities%20market.htm.

stock, 24 spreads correspond to $1.2 or 2.4%. The allowable price range of limit orders is

immensely wider during the closing auction session than during the continuous trading session.

The closing auction session is conducted between 4:00 p.m. and 4:10 p.m. It can be

divided into two periods: (i) order input period (4:004:08 p.m.); and (ii) pre-order matching

period (4:084:10 p.m.). Traders can submit either market or limit orders during the order input

period. However, only market orders are permitted during the pre-order matching period. Any

unfilled limit orders during the continuous trading session will be automatically carried forward

to the order input period of the CAS. Throughout the CAS, an IEP and IEV are displayed to the

market participants. The IEP is the price obtained from a single-price call auction which yields

the maximum number of shares transacted (i.e., IEV).7 After 4:10 p.m., orders are paired up and

transactions are executed at the IEP at 4:10 p.m., and this will be the closing price of the day.8

Appendix 1 describes the algorithm to compute IEP and IEV during the closing auction session.

Orders cancellation and amendment are allowed during the order input period but prohibited

during the pre-order matching period. In addition, short selling is prohibited during the CAS.

The Random Closing Procedure

The random closing procedure is used to determine the closing price during the pre- and post-

CAS periods. The closing price of a stock is determined by taking the median of five nominal

prices in the last minute of the continuous trading session.9 The system takes five snapshots on

7
The IEP must be a price between the highest limit bid and the lowest limit ask that maximizes the matched shares.
If there is a tie in IEP, three tie-breaker rules would apply. The first rule selects the price with the lowest order
imbalance. If this fails to break the tie, the second rule would apply and pick the price which is closest to the
nominal price at 4:00 p.m. Should these two rules also fail to break the tie, the third rule would apply and pick the
highest price as the IEP.
8
If there is no IEP available after the CAS (i.e., there is no match at all), then the median price of five nominal price
snapshots taken in the last minute of the continuous trading session is used as the closing price. Snapshots are taken
at 15-second intervals during 3:594:00 p.m.
9
Nominal prices are not necessarily the last transacted prices. Specifically, the nominal price is the last transacted
price only when the last transacted price falls within the current quoted bid-ask spread. Otherwise, the nominal

the nominal prices at 15-second intervals starting from 3:59 p.m. The median of these prices will

then be taken as the closing price of the day. Choosing the median of five snapshot nominal

prices ensures that the closing price will not be biased by one single trade. Note that the 24-

spread rule applies to orders submitted during this closing procedure because it is part of the

continuous trading session.

III. Data and Sample

Our sample spans a period of 34 months. It covers a one-year period pre-CAS

(5/26/20075/25/2008), a one-year period post-CAS (3/23/20093/22/2010), and a ten-month

period during CAS (5/26/20083/22/2009). This study focuses on the constituent stocks of the

Hang Seng Index, because they are the largest and most liquid stocks on the HKEx. Our sample

companies are large in size and comparable to S&P500 companies in terms of market

capitalization. The average market capitalization of our sample companies was US$34.5 billion

with a range from US$2.9 billion (COSCO Pacific) to US$200.5 billion (HSBC) in 2009, while

that of S&P500 companies was US$18.67 billion with a range from US$0.81 billion to

US$329.73 billion in 2009.10 Our sample stocks constitute 60.9 percent of the market

capitalization of all stocks listed on the HKEx at the end of 2009.11 Because their closing prices

are used to price many derivative products, investigating these stocks allows us to study the

incentives for manipulating their prices.

There were 47 constituent stocks in the Hang Seng Index at various points of the sample

period. We exclude eight stocks because they have less than 60 daily observations in either one

price refers to the bid price if the bid price is greater than the last transacted price and the ask price if the ask price is
smaller than the last transacted price.
10
Descriptive statistics on S&P 500 companies are obtain from the Standard and Poors website at
http://www2.standardandpoors.com/spf/pdf/index/SP_500_Factsheet.pdf.
11
See HKEx Fact Book 2009.

of the three periods. We construct our dataset from four data files published by the HKEx.

Specifically, data on intra-day bid and ask quotes, indicative equilibrium prices, indicative

equilibrium volume, and primary buy and sell queues are collected from the Bid and Ask Record

files; data on transaction prices and volume are from the Trade Record files; data on day-high

and day-low are from the Day-end Closing Data files; and data on the expiration dates of CBBCs

are from various years of HKEx Fact Book and the HKEx website.12 Transaction prices and

volume are recorded to the nearest second. Data on bid-ask quotes, IEP, IEV, and queues are

recorded to the nearest one-thousandth of a second since January 1, 2008, but only at 30-second

intervals prior to that date.

IV. Sniping

Roth and Ockenfels (2002) find that last-minute bidding (or sniping) is prevalent in eBay and

Amazon auctions. They propose a variety of strategic and behavioral reasons for sniping. For

example, rational informed bidders can use sniping to protect their private information. We

propose a novel explanation for sniping that is particularly relevant to stock markets: uninformed

traders use sniping to influence closing prices. We argue that sniping is a rational strategy for

manipulators to influence closing prices. Implementation of a successful price manipulation is

more certain and more likely to be achieved under the closing call auction mechanism than under

the random closing mechanism. This is because closing prices are concluded at a fixed deadline

under the closing call mechanism whilst they are randomly chosen at five snapshots under the

random closing procedure. The fixed deadline allows manipulators to surprise the market by

submitting an exceptionally large market order in the final seconds. Sniping is a rational strategy

because it gives no time for other traders to react, particularly to nullify the price impact from

12
The HKEx website provides data on expired CBBCs for the recent 12 months at
http://www.hkex.com.hk/eng/cbbc/download/dnfile.asp.

10

sniping. Manipulators can profit from sniping when closing prices of equities are used as

settlement prices for derivatives, e.g., CBBCs, stock warrants, stock index futures, and stock

options (see Kumar and Seppi, 1992; Stoll and Whaley, 1991; Chamberlain, Cheung, and Kwan,

1989). More generally, because closing prices are recorded and widely used as a summary

measure of a stocks performance, manipulating closing prices can be an instrument to influence

market sentiments, particularly relevant for pump-and-dump manipulations.

What constitutes sniping or a successful snipe attack in financial markets is open to

different interpretations. Conceptually, we are most interested in sudden changes in prices just

before the call auction closes, because manipulators can profit from sniping only when this

strategy can successfully influence closing prices. However, we also examine sniping in trade

volume (sudden surge in at-auction orders right before the auction ends) and sniping in both

price and trade volume for completeness. We reason that sniping in trade volume is a necessary

but not sufficient condition to influence the closing price. This is because a successful sniping

attack requires the manipulator to submit a surprising large order in final seconds, allowing no

time for the market to react. Yet, sniping attempts in trade volume may fail to influence closing

prices as manipulators can miscalculate. One possibility for this is that the impact of a sniping

order can be nullified or substantially weakened by an unexpectedly large incoming order on the

opposite side. Imagine that two manipulators simultaneously submit a large but opposing-

direction order in the final seconds. These opposing-direction orders nullify the intended price

effect of each other. Such manipulative attempts will not be reflected in a sudden change in

price, but will show up as a sudden increase in trade volume.

We measure sniping as follows: snipe(x) is a binary variable and takes the value of

one if the absolute change in x for a stock during the final ten seconds before the close is strictly

11

greater than the absolute change in x for the stock in any of the remaining five 10-second

intervals of the closing minute, where x is either p (price), v (trade volume), or pv (both price

and trade volume).13 We compute the absolute change in x by dividing the closing minute into

six 10-second intervals. The closing minute is taken from 4:094:10 p.m. during the closing

auction session of the CAS period, and from 3:594:00 p.m. during the continuous trading

session of the pre- and post-CAS periods. We use the beginning and ending values of x to

compute the 10-second absolute change in x of each interval. We measure price by IEP and

trade volume by IEV in the CAS period. Conversely, we measure price by actual transaction

price and trade volume by actual transaction volume in the pre- and post-CAS periods. The

variable snipe(pv) is equal to one if both snipe(p) and snipe(v) are equal to one.

Table 1 presents descriptive statistics on the sniping variables we have constructed. Our

results indicate that sniping in trade volume as well as sniping in both price and trade volume are

significantly more prevalent during the CAS period than during the non-CAS periods.

Specifically, the likelihood of sniping in trade volume is 6.1%, 25.0%, and 2.6% in the pre-CAS,

CAS, and post-CAS periods, respectively. This means that, in a quarter of our firm-day

observations in the CAS period, the most intense trading activities of the closing minute occur

during the last 10-second interval. Similarly, the likelihood of sniping in both price and trade

volume is 1.1%, 7.4%, and 0.6% in the pre-CAS, CAS, and post-CAS periods, respectively.

However, evidence for sniping in price during the CAS period is less overwhelming.

Specifically, the likelihood of sniping in price is 9.95%, 11.8%, and 11.9% in the pre-CAS, CAS,

and post-CAS periods, respectively.

[Insert Table 1 here.]

13
Our results are qualitatively similar but weaker in statistical significance if we either shorten the sniping interval
to five seconds or lengthen it to fifteen seconds.

12

 Table 2 presents estimates from a probit regression model on the relation between sniping

in price and sniping in trade volume after controlling for firm fixed-effects. Consistent with the

descriptive statistics in Table 1, our results in column (1) of Table 2 indicate that the likelihood

of sniping in price in the CAS period is significantly more prevalent than that in the pre-CAS

period but statistically indistinguishable from that in the post-CAS period. In the next

regression, we examine the impact of sniping in trade volume on the likelihood of sniping in

price by adding a dummy variable for sniping in trade volume and interacting this variable with a

dummy variable for the pre-CAS period and a dummy variable for the post-CAS period. Our

results in column (2) show that sniping in trade volume increases the likelihood of a sudden

change in price in the final ten seconds. More importantly, the effect of snipe(v) on

snipe(p) is significantly stronger (at the 5% level) during the CAS period than during the

non-CAS periods.

[Insert Table 2 here.]

Similarly, the results in column (3) show that sniping in trade volume is substantially

more likely in the CAS period than in the non-CAS periods, and that these differences are

statistically significant at the 5% level. Column (4) shows that sniping in price is positively

associated with sniping in trade volume, and that the relationship between the two is stronger

during the CAS period than during the non-CAS periods.

Price manipulation at the close can cause a large change in stock price during the closing

period. Thus, we use a large change in stock price during the closing period as a proxy for price

manipulation and examine if it is related to sniping. We first identify incidents of a large price

change during the closing period. We then examine if those incidents are caused by sniping,

particularly during the CAS period. Specifically, we say that a stock experiences a large price

13

change during the closing period if the absolute price change during this period is greater than

one percent of the last transacted price before the closing period. The closing period is

4:004:10 p.m. in the CAS period and 3:594:00 p.m. in the non-CAS periods. Table 3 presents

estimates from probit regression models on the above relationship, controlling for the firm fixed-

effects. We find that large price changes in the closing period are significantly associated with

sniping. More importantly, this relation is stronger during the CAS period than during the non-

CAS periods, and the difference is statistically significant at the 5% level.

[Insert Table 3 here.]

To successfully influence the closing price, a trader should submit a single large order in

the final seconds. His intentions could be leaked out if he attempted to break up his orders,

which could attract order flows from the opposing direction to nullify the intended price impact.

Further, breaking orders up in the final seconds entails an execution risk as late orders may miss

the fixed deadline and therefore be unfilled. In sum, we expect that trade volume should be more

concentrated in the CAS period than in the non-CAS period as sniping should be more frequently

used to manipulate the closing price in the CAS period.

Unfortunately, data on order arrivals and the limit order book are unavailable to us. To

investigate whether trade volume is more concentrated in the CAS period, we use transaction

records to measure concentration ratio of trade volume. Specifically, we use the difference-in-

differences methodology to examine the concentration ratio of trade volume measured during the

closing period relative to that during a benchmark period. The closing period is 4:004:10 p.m.

in the CAS period and 3:594:00 p.m. in the non-CAS periods. Our benchmark periods are the

10-minute interval of 3:504:00 p.m. in the CAS period and the 1-minute interval of 3:583:59

14

p.m. in the non-CAS periods. The concentration ratio of trade volume is measured by the

number of shares traded by the largest transaction to the total trade volume during the period.

Table 4 presents descriptive statistics on the concentration ratio of trade volume in each

of the three periods. During the CAS period, the mean concentration ratio is 2.6 times larger

than that during the benchmark period. In contrast, the mean concentration ratios during the

closing period and the benchmark period are roughly the same when call auctions were not in

use.

[Insert Table 4 here.]

The ultimate yardstick for a successful sniping attack is the ability to influence the

closing price. We first examine the absolute price change in the final ten seconds before the

close. In Table 5, we see that the absolute price change in the final 10-second interval is

significantly smaller in the CAS period (0.07%) than those in the pre-CAS (0.12%) and post-

CAS (0.14%) periods. Our findings are consistent with extant studies in that call auctions

consolidate order flows and reduce price volatility. While it is not shown here, the average trade

volume during the closing period is 1.8, 4.2, 1.6 million shares in the pre-CAS, CAS, and post-

CAS periods, respectively. This suggests that order flows are substantially thicker under the

closing auction mechanism than under the random closing mechanism.

[Insert Table 5 here.]

Nevertheless, the closing auction mechanism is vulnerable to price manipulation in the

form of sniping. Although the mean absolute change in price in the last ten seconds is smaller in

the CAS period, we find that this variable exhibits a greater variability in the tails under the

closing auction system. Conditional on price changes being large and economically significant

(larger than one or two percent), the average absolute price change is significantly larger in the

15

CAS period than those in the non-CAS periods. The average absolute price change in the final

ten seconds conditional on it being greater than two percent is 2.29%, 3.69%, and 2.33% in the

pre-CAS, CAS, and post-CAS periods, respectively. The differences in means between the CAS

period and each of the non-CAS period are large and have economic and statistical significance.

V. Motives for Sniping: Expiration of Callable Bull and Bear Contracts

The closing price can be used for a variety of purposes, including the pricing of derivative

products and providing a yardstick for the accounting of mutual fund performance. Because

closing prices are widely reported in the media, manipulating closing prices can also be used as a

tool to influence market sentiments. Since we do not know the long and short positions of

individual investors, it is generally difficult to uncover the identities of traders who may have an

incentive to manipulate closing prices. In this section, we take advantage of information about

one particular derivative productcallable bull and bear contractsto explore one possible

financial motive for manipulating closing prices. We emphasize that this is not the only possible

motive for price manipulation, but it is one that we can have sufficient information to study.

In Hong Kong, a financial motive to influence closing prices of equities is that

manipulators can profit from the CBBC market. The residual value of a CBBC can depend on

the closing price of the underlying stock on the expiration date. Specifically, the residual value

of an expired bull CBBC is determined by the settlement price less the strike price (a pre-

specified price) if this amount is positive, and is zero otherwise. The settlement price is the

minimum price of the underlying stock from the expiration time to the next trading session of the

day (typically till the market closes). Similarly, for a bear CBBC, the residual value is

determined by the difference between the strike price and the maximum price of the underlying

16

stock from the expiration time to the next trading session of the day. In addition, the incentive

and profit from manipulating closing prices can be large as the allowable price movement during

the CAS is much larger thanks to the 9-times restriction rule.14

Table 6 presents descriptive statistics for expired CBBCs and the incidence of day-high

and day-low at the close in each period. A day-high event is said to occur if the closing price is

the highest price of the trading day, and a day-low event occurs when the closing price is the

lowest price of the day. The number of expired CBBCs is quite small in the pre-CAS period

because they are a new derivative product. Indeed, CBBCs were introduced by the HKEx in

2006, but trading of CBBCs has increased rapidly since 2008. The turnover value of CBBCs

was US$9.2 billion in 2007, reaching US$133.96 in 2008 and US$215.99 billion in 2009.15

Table 6 shows that for our sample stocks the number of expired CBBCs is 55, 830, and 2719 in

the pre-CAS, CAS, and post-CAS periods, respectively. Overall, expired bull and bear CBBCs

are roughly comparable in number. Yet, the likelihood of either day-high or day-low at the close

is substantially higher in the CAS period than in the non-CAS periods. For example, the

probability of day-high at the close is 7.2% in the CAS period, and is about 3.7% in the pre-CAS

and post-CAS periods.

[Insert Table 6 here.]

To have a meaningful comparison between periods, we combine observations in the pre-

CAS and post-CAS periods into a single period, and call it the non-CAS period. This is

necessary as the number of expired CBBCs is quite small in the pre-CAS period, particularly for

expired bear contracts. We use a probit regression model with firm fixed-effects to examine the

impact of the number of expired CBBCs on the likelihood of sniping. Table 7 presents

14
Note that while the closing price can be used as the settlement price for a CBBC, indicative equilibrium prices
(IEP) computed during the CAS cannot be used as the settlement price.
15
See HKEx Fact Book 2009.

17

estimates for our three measures of sniping, snipe(p), snipe(v), and snipe(pv). The

variable ncbbc measures the number of expired CBBC for a stock on a given day. Consistent

with our expectations, the estimate for ncbbc in column (1) is positive and statistically

significant at the 5% level. This suggests that sniping in price is particularly more likely on days

when the number of expired CBBCs is large during the CAS period. Yet, this relation is

substantially weaker during the non-CAS period. The estimate for ncbbcnonCAS is negative

and statistically significant at the 5% level. However, results in columns (2) and (3) suggest that

the number of expired CBBCs is not significantly correlated with sniping in trade volume or

sniping in both price and trade volume.

[Insert Table 7 here.]

In addition, we divide expired CBBCs according to their types (bull or bear) to examine

how they affect the likelihood of observing day-high or day-low at the close. Bull CBBCs are

similar to call options and expire when the underlying stock price is lower than a pre-specified

price, i.e., the call price. The residual value of an expired bull CBBC is computed by the

positive amount of the minimum price of the underlying stock less the strike price during the

period when the CBBC expires and up to the next trading session of the day. In other words,

closing prices could be used to determine the residual value of bull contracts if they are the

minimum price of the underlying stocks. We hypothesize that the probability of day-low at the

close should be particularly large on days when the number of expired bull CBBCs is large.

Conversely, the probability of day-high events should be higher when the number of expired

bear contracts is large. In addition, this relation should be stronger in the CAS period than in the

non-CAS period.

18

 Table 8 presents estimates from the probit regression models of the effects of expired bull

and expired bear CBBCs on the probability of day-high at the close, the probability of day-low at

the close, and returns in the final 10-second interval. Congruent with our expectations, the

probability of day-low (day-high) at the close is significantly higher on days when the number of

expired bull (bear) CBBCs is large. The estimates for nbull and nbear are positive and

statistically significant at the 1% level. In addition, this relation is noticeably stronger in the

CAS period than in the non-CAS period. The estimates for nbullnonCAS and

nbearnonCAS are negative and statistically significant at the 1% level. Column (3) of Table

8 also shows that the number of expired bull CBBCs is significantly correlated with the stock

returns in the final ten seconds and that this relation is stronger in the CAS period than in the

non-CAS period. These two estimates are marginally significant at the 10% level. The stock

return in the last ten seconds is 0.07% lower on the day when one bull CBBC is expired during

the CAS period while it is almost 0% (= 0.0716% 0.0722%) in the non-CAS period.

[Insert Table 8 here.]

VI. Price Informativeness: Closing Call Auctions vs. Random Closing Mechanism

This section investigates the impact of the introduction of closing call auctions on price

informativeness. Specifically, we use goodness-of-fit (i.e., adjusted R-square) of the standard

market model to measure informativeness of closing prices (see, for example, Roll, 1988). We

reason that if closing prices fully reflect all publicly available information under the call auction

mechanism, we would expect that a greater co-movement between an individual stock return and

the market return in the period when closing call auctions are used than that in the period when

such auctions are not used.

19

 We use two measures of stock returns. The first one is the close-to-close return measured

over a one-day interval from the closing price in the previous day to the closing price in the

current day. The second one is the last-ten-minute return measured over the final ten-minute

interval before the close. Specifically, it is measured by the last transacted price during the

continuous trading session at 4:00 p.m. to the closing price at 4:10 p.m. of the current day during

the CAS period. During the non-CAS period, it is measured by the last transacted price at 3:50

p.m. to the last transacted price at 4:00 p.m. of the current day. The close-to-close return is the

conventional measure for price informativeness and is commonly used to examine precision of

closing prices (e.g., Comerton-Forde, Lau, and McInish, 2007; Pagano and Schwartz, 2003).

The last-ten-minute return captures the extent of price informativeness during the closing period

in the CAS period. The latter measure should be less vulnerable to biases from confounding

factors because it measures co-movement between an individual stock return and the market

return within a narrow window of time when closing prices are determined. As randomness in

setting closing prices can reduce the goodness-of-fit, we use last transacted prices of the day

instead of closing prices to compute the two returns in the pre-CAS and post-CAS periods. This

adjustment is warranted because using closing prices to measure returns in these periods can bias

our results in favor of finding that closing prices are more informative during the CAS period.16

Table 9 uses a firm fixed-effects model to examine the impact of adopting call auctions

on price informativeness. Results using close-to-close returns and last-ten-minute returns are

reported in columns (1) and (2), respectively. Consistent with previous studies, our findings

show that closing prices are more informative under the auction mechanism than under the

random closing mechanism. The goodness-of-fit estimate from the close-to-close return is

16
Our results are qualitatively identical but quantitatively stronger if the closing price is used to compute returns in
the pre-CAS and post-CAS periods.

20

64.3% during the CAS period. However, this estimate drops by 8.0 and 15.6 percentage points

in the pre-CAS and post-CAS periods, respectively. These differences have economic and

statistical significance. The results using the goodness-of-fit for the last-ten-minute returns are

qualitatively similar.

[Insert Table 9 here.]

Our evidence from the previous sections indicates that the closing auction mechanism is

vulnerable to price manipulation, particularly in the form of sniping. We further investigate

whether the informativeness of closing prices is poorer on days when price manipulation is likely

to occur. To precisely measure the impact of this possibility, we drop the goodness-of-fit

methodology for two reasons. First, the power of the test from this approach is weak in

measuring price informativeness precisely on the exact day when stock price is manipulated.

Second, the incidence of price manipulation should be infrequent (may be several days in a year)

to ensure the market would not develop counter-sniping strategies to nullify the price impact

from price manipulation.

Operationally, we use the extent of overnight price reversal to measure informativeness

of the closing price. We reason that if the closing price is precisely measured to reflect all

publicly available information, price reversal would be unlikely. Moreover, price reversal allows

us to distinguish whether sniping is used by informed or uninformed traders. If a sniping attack

is initiated by informed (uninformed) traders, price momentum should continue (reverse) on the

following open. This is because trades initiated by insiders have informational value and can

predict future return, while those from manipulators distort the fundamental value of the stock.

In his seminal paper, Kyle (1985) finds that in a model with multiple round of trading, insider

transactions are positively correlated over time and their transactions cause prices to be

21

increasingly informative over time. Consistent with this hypothesis, Brochet (2010) and

Lakonishok and Lee (2001) find that the market reacts quickly and positively (negatively) to

announcements of insider purchases (sales).

We expect that price reversal should be stronger on days when price manipulation is

likely to occur and that this relation should be particularly strong during the CAS period than

during the pre-CAS and post-CAS periods. We follow Pagano and Schwartz (2003) to use the

close-to-open return to measure the extent of price reversal. We use three methods to identify

days that price manipulation is likely to occur: (i) days when large price changes during the final

10-minute interval occur; (ii) days when CBBCs expire; and (iii) days when sniping attacks

occur.

Although the results in Table 9 suggest that closing auctions produce more informative

prices on average, they may be vulnerable to infrequent attempts at price manipulation. To

measure such infrequent attempts, it is useful to focus on outliers, as successful price

manipulation tends to produce large and sudden changes in stock prices. For each stock in each

trading day, we calculate the percentage price change during the final ten-minute interval. In

column (1) of Table 10, we select the ten observations with the largest absolute percentage price

changes for each of the pre-CAS, CAS, and post-CAS periods. The total sample size is therefore

30. Column (2) selects the observations with the largest 100 absolute percentage price changes

during the last ten-minute interval for each period. In columns (3) to (5), the sample is

progressively expanded to include observations with the largest 200, 500, and 1500 largest

absolute percentage price changes during the final ten minutes before the close. The full sample

of observations is used in column (6).

[Insert Table 10 here.]

22

 To measure the degree of price manipulation in the closing period in the CAS period, we

include returns measured over the final 10-minute interval before the close (R10m) and interacted

this variable with a binary variable for the pre-CAS period and a binary variable for the post-

CAS period. To obtain a comparable measure of returns in the pre-CAS and post-CAS period,

we measure returns over the final 10-minute interval from 3:504:00 p.m.

We find that during the CAS period price reversal is particularly large and noticeable on

days when very large price changes occur during the final 10-minute interval. For instance, the

estimate for R10m in column (1) is negative and suggests that when the price changes by one

percent during the CAS, 87.4% of this change would revert on the following open. In contrast,

the estimates for R10mpre-CAS and R10mpost-CAS in column (1) are positive. This implies

that only 56.5% (= 0.3091 0.8740) and 38.6% (= 0.4876 0.8740) of the same price change in

the pre-CAS and post-CAS periods, respectively, would revert on the following open. These

estimates are large and have economic and statistical significance. The only exception is that the

estimate for R10mpre-CAS is not statistically significantly at the 10% level. While it is not

shown here, the average absolute percentage price change in the final 10-minute interval before

the close of the ten largest observations is 5.8%, 8.4%, and 4.2% in the pre-CAS, CAS and post-

CAS periods, respectively.

Price reversals during the CAS period are weaker when we expand our sample by

including observations with smaller price changes. The results in columns (2)(6) show that

while the price reversal estimates (i.e., R10m) during the CAS period remain negative across all

regressions, their absolute magnitudes are negatively correlated with the sample size.

Consistent with our expectation that the incidence of price manipulation should be

infrequent, our results indicate that, relative to the non-CAS periods, price reversal during the

23

CAS period is pronounced only on days when price changes in the final 10-minute interval are

very large. Specifically, the absolute magnitudes of the estimates for R10mpre-CAS and

R10mpost-CAS in columns (2)(3) are smaller than those in column (1). In addition, they

remain positive and statistically significant at the 10% level, suggesting the price reversal

remains more likely in the CAS period than in the non-CAS period. Nevertheless, the estimates

for R10mpre-CAS and R10mpost-CAS in columns (4)(6) become negative. This suggests that,

on average, price reversals during the CAS period are less likely than non-CAS period on days

when price changes in the final 10-minute interval are small.

The results from the full sample group in column (6) imply that when the price change by

one percent in the final 10-minute interval during the CAS period, only 23.8% of this change

would revert on the following open. In contrast, 29.4% in the pre-CAS period and 33.9% in the

post-CAS period of the same price change would revert on the following open. The larger price

reversals in the CAS period relative to the non-CAS periods are statistically significant at the 5%

level.

Table 11 presents firm fixed-effects regression results of the impact of expired CBBCs on

price reversal. Again, because of the small number of expired CBBCs in the pre-CAS period, we

combine observations in the pre-CAS and post-CAS periods into a single period, and call it the

non-CAS period. The estimate for ncbbcR10m indicates that price reversal is strong and

statistically significant at the 5% level on days when CBBCs expired during the CAS period. If a

CBBC expires on a certain day and the underlying stock price drops by 1% in the final 10-

minute interval, the stock price would bounce back by 0.082% on the following open. Yet, if the

same event happens in the non-CAS period, price momentum would persist in the following

open on days with expired CBBCs. The estimate for ncbbcnonCASR10m is positive and

24

statistically significant at the 1% level. Specifically, during the non-CAS period, the same event

would cause the stock price to drop further by 0.005% (= 0.0868 0.0817) when the market

opens the next trading day.

[Insert Table 11 here.]

Table 12 presents firm fixed-effects regression results on the impact of the extent of

sniping on price reversal during the CAS period. To measure the degree of sniping, we include

returns measured over the ten-second interval before the close and interact this variable with our

measures of sniping and a binary variable for the non-CAS period. Consistent with our

expectations, the estimates for snipe(p)R10s in column (1) and snipe(pv)R10s in

column (3) imply that the degree of price reversal is large on days when the degree of either

sniping in price or sniping in both price and trade volume is large during the CAS period. These

estimates are statistically significant at the conventional levels. However, such a price reversal is

not observed on days when only sniping in trade volume is observed in the CAS period.

According to the estimates in column (1), if sniping in price occurs on the day for a stock and the

price of this stock drops by 1% in the final 10-second interval before the close, the stock price

would bounce back by 0.73% when the market opens the next trading day. Nevertheless, the

estimate for snipe(p)nonCASR10s implies that the degree of price reversal for the same

scenario is significantly smaller in the non-CAS period. In particular, the stock price would

bounce back by only 0.06% (= 0.73% + 0.67%) if the same incident occurs during the non-CAS

period.

[Insert Table 12 here.]

VII. Other Evidence on Closing Price Manipulation during the CAS Period

25

There are a variety of motives and strategies for price manipulation. Hillion and Suominen

(2004) and Felixson and Pelli (1999) argue that brokers manipulate the closing price to improve

their execution qualities. Fund managers can manipulate the closing price to window-dress

their performances.17 Specifically, Goetzmann, Ingersoll, Spiegel, and Welch (2007) point out

that traditional measures of fund performance can be gamed and vulnerable to price

manipulations. Manipulators may use pump-and-dump scheme to influence closing prices to

sway market sentiment, as closing prices are widely used by the media, investors, and investment

managers to determine asset values (see Kyle and Viswanathan, 2008). Closing prices are also

likely to be manipulated on days especially when these prices are being used to determine index

prices, particularly for major market indices which add or delete constituent stocks from their

portfolios. Anecdotal evidence suggests that price manipulations are particularly rampant on

days when the Morgan Stanley Capital International index (MSCI) re-balances their major

indices during the CAS period.18 As costs of closing price manipulation are lower under the

auction mechanism, the financial incentive to manipulate the closing price on those MSCI index

re-balancing days should be larger during the CAS period than during the non-CAS periods.

17
Some anecdotal evidences in Hong Kong are consistent with the hypothesis that window dressing was being used
during the CAS period. Mutual funds typically use October 31 to compute their (quarterly or semi-annually)
performances. A case in point is the unusual price movements for some stocks during the closing auction session on
October 31, 2008. For all stocks listed on the main board of the HKEx, we compute their returns during the 10-
minute interval of the closing auction session on that day. We rank changes in stock returns by their absolute
values. We find that the top ten changes in returns are all positive and these increases are large with a minimum
increase in stock return of 14%. The top two changes in returns are exceptionally large. During the 10-minute
interval, stock price of Sinotrans rose from $1.02 to $1.68 (by 64.7%) while that of China Insurance International
Holdings rose from $11.98 to $17.80 (by 48.6%). However, these price increases were not permanent as they
subsequently reversed. Stock price of Sinotrans dropped from $1.68 to $1.30 when the market opened on the next
trading day and then closed at $1.15 the following day. Similarly, shares of China Insurance International Holdings
dropped from $17.8 to $13.46 when the market opened on next trading day and then closed at $11.70 two trading
days later. In contrast, the Hang Seng Index increased from 13,968 to 14,436 (by 3.35%) when the market opened
on next trading day and then closed at 14,384 two trading days later.
18
Suspected price manipulation incidents on the MSCI index rebalancing days are not uncommon. For example,
Comerton-Forde, Lau, and McInish (2007) examine the Singapore Exchange and find that suspected price
manipulations had happened on the days when MSCI re-balances its index. Yet, these suspected cases occurred
before the Singapore Exchange adopted a call auction mechanism to determine the closing price.

26

 To examine the impact and extent of closing price manipulations on days when MSCI re-

balances their major indices, we expand our sample from 39 stocks to all stocks listed on the

main board of the HKEx. The expanded sample includes small- and mid-cap stocks in addition

to large-cap stocks. The inclusion of mid- and small-cap stocks is necessary as they constitute

literally all the stocks that are added to or deleted from major MSCI indices. Table 13 presents

the top ten last-ten-minute return in percentage terms by their absolute values for stocks listed on

the main board of the HKEx in each of the three CAS period.19 The last-ten-minute return is

measured over a ten-minute interval from the last transacted price at 4:00 p.m. (3:50 p.m.) to the

closing price at 4:10 p.m. (4:00 p.m.) in the CAS (non-CAS) period. Specifically, Panels (a) and

(b) of Table 13 present those changes in the CAS period (i.e., May 30, 2008 in panel (a) and

November 25, 2008 in panel (b)) while panels (c) and (d) list those changes in the pre-CAS (i.e.,

November 30, 2007) and post-CAS period (i.e., May 29, 2009) , respectively. Data on close-to-

open return for these stocks are also provided in Table 13.

[Insert Table 13 here.]

Results in panels (a) and (b) indicate that the mean last-ten-minute return (R10m) by

absolute value for the twenty observations during the CAS period is 11.99%. This figure is large

and has economic significance. In 19 of these 20 stocks, we observe a large and meaningful

price reversal on the following open. Our findings strongly suggest that these closing prices are

not informative and likely to be manipulated on the two MSCI index re-balancing days during

the CAS period.

19
Bid-ask spread is typically large for penny stocks. A simple round-trip transaction at the current bid and ask can
produce a large price change even if this change is not triggered by price manipulation. Imagine the last transacted
price of a penny stock is $0.1 and the prevailing best bid and offer is $0.09 and $0.10. Thus, an arrival of a small
sell order executed at the best bid represents a change in stock price of 10%. To prevent a simple (uninformed) bid-
ask bouncing effect from potentially biasing our results, we exclude stocks with prices less than $1 from our sample
tabulated in Table 13. Yet, our results are qualitatively and quantitatively similar if these stocks are included.

27

 In contrast, the mean last-ten-minute return by absolute value for the twenty observations

is 9.49% and 5.28% in the pre- and post-CAS periods, respectively. These two estimates are

significantly smaller than that of the CAS period. In addition, price reversal is also less

frequently observed. In only 8 out of a total of 20 stocks do we observe a large and meaningful

price reversal on the following open.

VIII. 24-Spread Rule and Price Manipulation during the CAS Period

The relaxation of the 24-spread rule during the CAS period should be a key factor that

contributes to price manipulation. During the continuous trading session, a binding constraint

for the allowable price change at any instant of time is the 24-spread rule. The HKEx removed

the 24-spread rule for orders submitted the closing auction session, greatly increasing the

allowable price movement. If the last transacted price of a stock is $50 at 4:00 p.m., the stock

price can deviate by only 2.4% (or $1.2) under the 24-spread rule. Yet, stock price can increase

by 900% (or $450) or decrease by 89% ($44.5) under the 9-times restriction rule during the CAS.

As larger price changes are feasible under the CAS, the relaxation of the 24-spread rule should

make price manipulation more profitable. We therefore expect to observe more price

manipulation during the CAS period than during the non-CAS periods.

We test this hypothesis by examining stocks that are likely to be bound by the 24-spread

rule by examining informativeness of closing prices on days when the 24-spread rule is violated.

As data on order arrivals and limit order books are unavailable to us, we cannot precisely

identify stocks that have standing limit orders that are 24 spreads away from their nominal prices

at 4:00 p.m. As a compromise, we use IEP as a surrogate to identify such a violation. As limit

orders are firm (i.e., cannot be canceled or modified) during the final two-minute interval before

28

the close in the CAS period, we use IEP during this 2-minute interval to infer whether firm limit

orders beyond 24-spreads have been submitted during the CAS. We define a violation of the 24-

spread rule at the buy-side (i.e., buy24spr) if the maximum IEP during the last two minutes is

greater than 24 spreads of the nominal price at 4:00 p.m. Similarly, we define a violation of the

24-spread rule at the sell-side (i.e., sell24spr) if the minimum IEP during the last two

minutes is less than 24-spread of the nominal price at 4:00 p.m. We further construct a variable

to capture the violation of the 24-spread rule at both the buy-side and the sell-side. The variable

24spr takes the value of one if buy24spr is one, minus one if sell24spr is one, and zero

if otherwise. We note that our measure does not capture all instances when the 24-spread rule is

violated, as the IEP may not reach limit orders that are very high or very low. We follow similar

procedures to measure deviations from 24 spreads in the non-CAS period: buy24spr is equal

to one if the maximum price in 3:584:00 p.m. is more than 24 spreads higher than the price at

3:50 p.m., and sell24spr is equal to one if the minimum price in 3:584:00 p.m. is more than

24 spreads lower than the price at 3:50 p.m.

We use the close-to-open return to examine informativeness of closing prices on days

when 24-spread rule is violated. Table 14 presents firm fixed-effects regression results on the

impact of the 24-spread rule on price reversal during the CAS period. Consistent with our

expectations, the estimate for 24spr in column (1) is positive while that for 24sprnonCAS is

negative. Both estimates are large and have economic and statistical significance. The close-to-

open return is 2.04% for orders that violated the 24-spread rule during the CAS period. In

contrast, the close-to-open return is +0.28% (= 2.32% 2.04%) for orders that violated the 24-

spread rule during the non-CAS period.

[Insert Table 14 here.]

29

 If we separate violations by buyer- or seller-initiated orders, our results indicate that price

reversal is stronger for limit sell orders that are 24-spread away from the prevailing nominal

price than comparable limit buys orders. This relation is particularly strong during the CAS

period than during the non-CAS period. The estimate for sell24spr in column (2) is positive

while that for sell24sprnonCAS is negative. Both estimates are large and have economic

and statistical significance. These estimates imply that the close-to-open return is +2.2% for

limit sell orders that violated the 24-spread rule during the CAS period while that for the non-

CAS period is 1.1% (= 3.29% + 2.19%). The effect is weaker for limit buy orders that

violated the 24-spread rule during the CAS period. The estimate for buy24spr in column (2) is

negative while that for buy24sprnonCAS is positive. Both estimates are large in economic

terms. The first estimate is statistically significant at the 1% level, but the latter is only

marginally significant at the 10% level.

IV. Conclusions

Our empirical findings provide convincing evidence that a plain vanilla call auction to determine

the closing price is susceptible to price manipulation, particularly in the form of sniping.

According to the ten-month experience at the HKEx, we find that the closing price is more

informative or accurately priced under the closing auction mechanism. Yet, closing prices can

be uninformative or distortive on days when price manipulation is likely to occur, e.g., on days

when large price changes in the final ten-minute interval occur, when CBBCs expire, when

sniping attacks occur, and when closing prices are used as benchmark prices for re-balancing

MSCI major indices.

30

 This study has important policy implications given call auctions are widely used to

determine opening and closing prices for equities around the world. In the wake of the extreme

price volatility for some stocks when MSCI re-balanced their major indices on May 30, 2008, the

HKEx had decided to introduce a price control mechanism during the closing auction session to

reduce price manipulation. This enhancement is scheduled to be effective on June 22, 2009.

Yet, the enhancement has never been implemented because the closing auction session was

indefinitely suspended since March 23, 2009. The proposed price control mechanism allows

closing prices to deviate from the nominal price at 4:00 p.m. by only 2%. We believe that this

enhancement would weaken the incentive to manipulate the closing price by making the act less

profitable, because the maximum price deviation from the prevailing nominal price at 4:00 p.m.

would be drastically reduced from +900% or 89% to 2% during the closing auction session.

However, this enhancement cannot eradicate the incentive to manipulate prices in the

form of sniping because the deadline remains fixed rather than random. The proposed price

control mechanism changes the focal time for price manipulation from 4:10 p.m. to 4:00 p.m. As

the allowable movement of closing prices is directly determined by the nominal price at 4:00

p.m., the nominal price at 4:00 p.m. would become the target for price manipulation. There are a

variety of methods to manipulate this nominal price; one possibility is sniping successively at

4:00 p.m. Specifically, manipulators may submit a sequence of surprisingly large market orders

in the final minute before 4:00 p.m., with each order being just fractions of a second away from

one another. Next, a sequence of small but aggressive limit sell orders (e.g., 24 spreads from the

prevailing nominal price) are submitted concurrently such that their buy orders can exactly

crossed with those incoming limit sell orders to generate valid transaction prices.20

20
Our conjecture should be warranted because this closing price manipulation tactic is not uncommon. In fact,
similar tactics have been detected and successfully prosecuted by the Hong Kong Securities and Futures

31

 By simultaneously submitting a sequence of surprisingly large market buy (sell) orders

and a sequence of small but aggressive limit sell (buy) orders in the final seconds of the

continuous trading session, nominal prices at 4:00 p.m. can be marked up (down) substantially.

Imagine that a trader submits ten individual and yet surprisingly large market buy orders, each

executed one second apart in the final ten seconds of the continuous trading session.

Concurrently, the same trader submits ten small limit sell orders with each specified at a price

2.4% lower than the prevailing nominal price and again each executed one second apart in the

final ten seconds of the continuous trading session. If successful, this strategy can boost stock

price by a maximum of 26.77%.21 Yet, this manipulation strategy can be costly to implement as

it should give time for the market to react.

We believe a fixed deadline together with a large allowable price movement contribute to

the vulnerability of the closing call auction to price manipulation in the HKEx. In fact, many

worldwide stock exchanges have additional refinements to reduce extreme price movement

concluded from the closing auction mechanism. These refinements include: (i) a daily price

limit, (ii) a random (rather than fixed) deadline to determine the closing price, and (iii) a special

price stabilizing mechanism to reduce volatility of the closing price. Table 15 presents

refinements to the closing auction mechanism to reduce extreme price movement in ten major

stock exchanges around the world. The first refinement is to impose a daily price limit. This

refinement is very common to major stock exchanges in Asia, e.g., Tokyo Stock Exchange,

Korea Exchange, Taiwan Stock Exchange, and Shenzhen Stock Exchange. Another refinement

increases costs of manipulation by adopting a random deadline to determine the closing price.

Commission when the HKEx used the random closing procedure to determine the closing price. Details of such
cases are available from http://www.sfc.hk/sfcPressRelease/EN/sfcOpenDocServlet?docno=08PR94.
21
If each of such buy order can boost stock price by 2.4%, 10 such orders can boost stock price by a total of 26.77%
( = (1+0.024)10 1).

32

Examples of this refinement include Australian Securities Exchange, London Stock Exchange,

and Deutsche Brse. The third refinement includes an explicit order-balancing mechanism to

stabilize the closing price. For example, New York Stock Exchange and Nasdaq widely

disseminate the net order imbalance to market participants for orders designated to determine the

closing price. In addition, in the final 15 (20) minutes before the close, the NYSE (Nasdaq)

accepts only orders that are on the stabilizing side of the market. Another form of the price-

stabilizing mechanism is to trigger a volatility auction/interruption when the closing price

deviates significantly from the last transacted price or a certain pre-specified price limit. This

mechanism is adopted by Euronext and Deutsche Brse.

[Insert Table 15 here.]

To weaken the incentive of price manipulation in the form of sniping, call auctions with a

random deadline to close should be less susceptible to sniping than call auctions with a fixed

deadline. In fact, call auctions with a random deadline to close is commonly practiced in major

stocks exchanges, namely the London Stock Exchange, Deutsche Brse, and Australian Stock

Exchange. To our knowledge, notable price manipulation is rare under such a random closing

mechanism. The variety of safeguards adopted in other stock exchanges and the experience of

the price manipulations under the plain-vanilla call auction system in HKEx provides a lesson

that seemingly minor details can matter when it comes to the design of trading mechanisms.

33

 References

Aggarwal, R., and G. Wu. 2006. Stock Market Manipulations. Journal of Business 79:
19151953.

Bajari, P., and A. Hortacsu. 2003. The Winners Curse Reserve Prices and Endogenous Entry:
Empirical Insights from eBay Auctions. RAND Journal of Economics 34(2): 329355.

Biais, B., P. Hillion, and C. Spatt. 1999. Price Discovery and Learning During the Pre-Opening
Period in the Paris Bourse. Journal of Political Economy 107: 12181248.

Brochet, F. 2010. Information Content of Insider Trades before and after the Sarbanes-Oxley
Act. The Accounting Review 85(2): 419446.

Cao, C., E. Ghysels, and F. Hatheway. 2000. Price Discovery without Trading: Evidence from
the Nasdaq Pre-Opening. Journal of Finance 55: 13391365.

Chamberlain, T., C. Cheung, and C. Kwan. 1989. Expiration-day Effects of Index Futures and
Options: Some Canadian Evidence. Financial Analyst Journal 45(5): 6771.

Comerton-Forde, C., S. T. Lau, and T. McInish. 2007. Opening and Closing Behavior Following
the Introduction of Call Auctions in Singapore. Pacific Basin Finance Journal 15: 1835.

Comerton-Forde, C., and T. Putnins. Forthcoming. Measuring Closing Price Manipulation.

Journal of Financial Intermediation.

Ely, J., and T. Hossain. 2009. Sniping and Squatting in Auction Markets. American Economic
JournalMicroeconomics 1(2): 68-94.

Felixson, K., and A. Pelli. 1999. Day End ReturnsStock Price Manipulation. Journal of
Multinational Financial Management 9: 95127.

Goetzmann, W. N., J. Ingersoll, M. Spiegel, and I. Welch, 2007. Portfolio Performance

Manipulation and Manipulation-proof Performance Measures. Review of Financial Studies
20(5): 15031546.

Hillion, P., and M. Suominen. 2004. The Manipulation of Closing Prices. Journal of Financial
Markets 7: 351375.

Kumar, P., and D. J. Seppi. 1992. Futures Manipulation with Cash Settlement. Journal of
Finance 47: 14851502.

Kyle, A. 1985. Continuous Auctions and Insider Trading, Econometrica 53: 13151335.

34

Kyle, A., and S. Viswanathan. 2008. How to Define Illegal Price Manipulation, American
Economic Review: Papers & Proceedings 98: 274279.

Lakonishok, J., and I. Lee. 2001. Are Insider Trade Informative? Review of Financial Studies
14(1): 79111.

Madhavan, A. 1992. Trading Mechanisms in Securities Markets. Journal of Finance 47(2): 607
642.

OHara, M. 2007. Optimal Microstructures. European Financial Management 13(5): 825832.

OHara, M. 2001. Designing Markets for Developing Countries. International Review of Finance
2(4): 205215.

Pagano, M., and R. Schwartz. 2003. A Closing Call's Impact on Market Quality at Euronext
Paris. Journal of Financial Economics 68: 439484.

Roll, R. 1988. R2. Journal of Finance 43(3): 541566.

Roth, A., and A. Ockenfels. 2002. Last-Minute Bidding and the Rules for Ending Second-Price
Auctions: Evidence from eBay and Amazon Auctions on the Internet, American Economic
Review 92(4): 10931103.

Roth, A., and A. Ockenfels. 2006. Late and Multiple Bidding in Internet Second Price Auctions:
Theory and Evidence Concerning Different Rules for Ending and Auction, Games and Economic
Behavior 55: 297320.

Schwartz, R.A. 2001. The Call Auction Alternative. In: Schwartz, R.A. (Ed.), The Electronic
Call Auction: Market Mechanism and Trading, Building a Better Stock Market. Kluwer
Academic Publishers, Boston.

Stoll, H., and R. Whaley. 1991. Expiration-day Effects: What Has Changed? Financial Analyst
Journal 47(1): 5872.

35

Figure 1. Indicative Equilibrium Price and Primary Buy and Sell Queue for HSBC Shares
during the Closing Auction Sessions
The solid line represents the indicative equilibrium price (IEP); the dashed line represents the primary sell queue;
and the dotted line represents the primary buy queue. Vertical line marks the beginning of the pre-order matching
period, when order cancellation and amendment are prohibited. Circle represents the IEP at $33. The IEP must be a
price at and between the highest limit bid and the lowest limit ask and maximizes the matched shares, i.e., the
indicative equilibrium volume (IEV). The primary sell (buy) queue is the queue of at-auction sell (buy) orders and
at-auction limit sell (buy) orders with a specified price at or more competitive than the IEP.

(a) Monday, March 9, 2009

The image cannot be display ed. Your computer may not hav e enough memory to open the image, or the image may hav e been corrupted. Restart y our computer, and then open the file again. If the red x still appears, y ou may hav e to delete the image and then insert it again.

15
39

Primary Buy and Sell Queue (million shares)

38
Indicative Equilibrium Price ($)

10
35 36 37

5
34
33

0
1600 1601 1602 1603 1604 1605 1606 1607 1608 1609 1610
Time (hhmm)


(b) Friday, March 6, 2009
The image cannot be display ed. Your computer may not hav e enough memory to open the image, or the image may hav e been corrupted. Restart y our computer, and then open the file again. If the red x still appears, y ou may hav e to delete the image and then insert it again.
44

50
43

Primary Buy and Sell Queue (million shares)

36 37 38 39 40 41 42

40
Indicative Equilibrium Price ($)

10 20 30
35
34
33

1600 1601 1602 1603 1604 1605 1606 1607 1608 1609 1610
Time (hhmm)

36

Table 1. Descriptive Statistics Sniping

The binary variable snipe(x) takes the value of one if the absolute change in x for a stock during the final 10-
second interval before the close is strictly greater than any absolute change in x for the stock in any of the remaining
five 10-second intervals of the closing minute, where x is either p (price), v (trade volume), or pv (both price and
trade volume). We compute the absolute change in x by dividing the closing minute into six 10-second intervals.
The closing minute to be 4:094:10 p.m. during the closing auction session of the CAS period, and 3:594:00 p.m.
during the continuous trading session of the pre- and post-CAS periods. We use the beginning and ending values of
x to compute the 10-second absolute change in x of each interval. We measure price by IEP and trade volume by
IEV in the CAS period. We measure price by actual transaction price and trade volume by actual transaction
volume in the pre- and post-CAS periods. The corresponding standard errors are reported in parenthesis.

Pre-CAS CAS Post-CAS

snipe(p) 0.0995 0.118 0.119
(0.299) (0.322) (0.324)

snipe(v) 0.0609 0.250 0.0262

(0.239) (0.433) (0.160)

snipe(pv) 0.0106 0.0734 0.00616

(0.102) (0.261) (0.0783)

N 9,547 7,856 9,736

37

Table 2. Relation between Sniping in Price and Sniping in Trade Volume

We use a probit regression model to examine the relation between sniping in price and sniping in trade volume after
controlling for firm fixed-effects. The dependent variable is our measure of sniping in either price or trade volume.
The variable snipe(p) [snipe(v)] takes the value of one if the absolute change in price [trade volume] during
the final 10-second interval for the stock is strictly greater than any absolute change in price [trade volume] for the
stock in any of the remaining five 10-second intervals of the closing minute, and zero otherwise. The variable
preCAS (postCAS) takes the value of one if the observation is taken in the pre-CAS (post-CAS) period, and zero
otherwise. The corresponding robust standard errors clustered at the firm-period level are reported in parenthesis.
Statistical significance is marked at the 1% (***), 5% (**) and 10% (*) levels.

(1) (2) (3) (4)

snipe(p) snipe(p) snipe(v) snipe(v)
snipe(v)preCAS -0.6450***
(0.0696)

snipe(v)postCAS -0.5639***
(0.1010)

snipe(v) 1.0209***
(0.0465)

snipe(p)preCAS -0.7952***
(0.0722)

snipe(p)postCAS -0.7774***
(0.0864)

snipe(p) 1.1466***
(0.0523)

preCAS -0.0972*** 0.2509*** -0.8825*** -0.7606***

(0.0317) (0.0343) (0.0325) (0.0318)

postCAS 0.0102 0.3714*** -1.2840*** -1.1772***

(0.0297) (0.0326) (0.0381) (0.0393)

constant -1.2229*** -1.6143*** -0.5684*** -0.7269***

(0.0318) (0.0350) (0.0364) (0.0370)

Firm Fixed-effect Yes Yes Yes Yes

N 27,139 27,139 27,139 27,139
Pseudo R2 0.0046 0.0426 0.1459 0.1855

38

Table 3. Sniping and Large Price Change in the Closing Period

We use a probit regression model to examine the relation between sniping and the incidence of large price changes
during the closing period after controlling for firm fixed-effects. The dependent variable is our measure of sniping
in price (i.e., snipe(p)), trade volume (i.e., snipe(v)), or both price and trade volume (i.e., snipe(pv)). The
snipe(p), snipe(v), and snipe(pv) takes the value of one if the absolute change in price, trade volume, and
both price and trade volume during the final 10-second interval before the close for the stock is strictly greater than
the absolute change in price, trade volume, and both price and trade volume, respectively, for the stock of any of the
remaining five 10-second intervals of the closing minute, and zero if otherwise. The variable Price1% takes the
value of one if the absolute price change during the closing period is greater than one percent of the last transacted
price before the closing period. The closing period is 4:004:10 p.m. in the CAS period and 3:594:00 p.m. in the
non-CAS periods. The variable preCAS (postCAS) takes the value of one if the observation is taken in the pre-
CAS (post-CAS) period, and zero if otherwise. The corresponding robust standard errors clustered at the firm-
period level are reported in parenthesis. Statistical significance is marked at the 1% (***), 5% (**) and 10% (*)
levels.

(1) (2) (3)

snipe(p) snipe(v) snipe(pv)
Price1%preCAS -0.5532*** -0.4223* -0.8352**
(0.1865) (0.2326) (0.3561)

Price1%postCAS -0.4129** -0.3751** -0.6544***

(0.1747) (0.1847) (0.2226)

Price1% 0.6257*** 0.3264** 0.7461***

(0.1425) (0.1306) (0.1393)

preCAS -0.0736** -0.8663*** -0.8151***

(0.0302) (0.0325) (0.0447)

postCAS 0.0421 -1.2676*** -1.0147***

(0.0275) (0.0382) (0.0555)

constant -1.2566*** -0.5826*** -1.4615***

(0.0313) (0.0366) (0.0438)

Firm Fixed-effect Yes Yes Yes

N 27,139 27,139 27,139
Pseudo R2 0.0081 0.1470 0.1407

39

Table 4. Concentration Ratio of Trade Volume

This table presents the concentration ratio of trade volume in the closing period in the pre-CAS, CAS and post-CAS
periods, respectively. CR1c/CR1b is the concentration ratio in the closing period divided by the concentration ratio
in the benchmark period. CR1c is the fraction of shares attributable to the largest transaction during the closing
period, where the closing period is the 10-minute interval of 4:004:10 p.m. in the CAS period and the 1-minute
interval of 3:594:00 p.m. in the non-CAS periods. CR1b is the concentration ratio of trade volume measured
during the benchmark period. Specifically, it is the 10-minute interval of 3:504:00 p.m. in the CAS period and the
1-minute interval of 3:583:59 p.m. in the non-CAS periods. The corresponding standard errors are reported in
parenthesis.

Pre-CAS CAS Post-CAS

CR1c/CR1b 116.1 264.5 95.26
(80.62) (170.9) (67.44)

CR1c 0.221 0.157 0.135

(0.145) (0.0905) (0.0832)

CR1b 0.224 0.0750 0.171

(0.135) (0.0554) (0.100)

N 9,392 7,775 9,658

Table 5. Last Ten-Second Changes in Price

This table presents the mean absolute price change in the final 10-second of trading in the Pre-CAS, CAS, and Post-
CAS periods, respectively. The variable Abs(p10s) is the mean absolute price change in percent in the final 10-
second interval of the closing minute of a stock; and Abs(p10s) | Abs(p10s) >1% [Abs(p10s) | Abs(p10s) >2%] is
the mean absolute price change in the final 10-second interval of the closing minute of a stock on conditional that
the absolute price change in the final 10-second interval of the closing minute of the stock exceeds 1 percent [2
percent]. The corresponding standard errors are reported in parenthesis.

(1) (2) (3) t-statistics t-statistics

Pre-CAS CAS Post-CAS (1)(2) (3)(2)
Abs(p10s) 0.123 0.0731 0.140 0.0496*** 0.0666***
(0.209) (0.286) (0.217) (0.0108) (0.0111)

Abs(p10s) | Abs(p10s) >1% 1.320 1.954 1.350 -0.6339*** -0.6040***

(0.341) (1.576) (0.366) (0.1092) (0.1111)

Abs(p10s) | Abs(p10s) >2% 2.285 3.689 2.333 -1.4046** -1.3560**

(0.270) (2.411) (0.266) (0.5378) (0.5389)

N 9,547 7,856 9,736

40

Table 6. Descriptive Statistics Expired CBBCs and Day-High/Day-Low Events

This table presents the incidence of expired callable bull and bear contracts (CBBCs) in pre-CAS, CAS, and post-
CAS periods, respectively. A CBBC is expired if it is either expires naturally on the maturity date or expires early.
An early expiration is known as the mandatory call event (MCE) and occurs when the underlying assets price
reaches the call price at any time prior to the maturity date. We include category R CBBCs only in our sample. A
category R CBBC refers to a CBBC where its call price is different from the strike price and the holder may receive
a small amount of cash payment (henceforth residual value) upon the occurrence of the MCE. Once the MCE is
triggered, the residual value of the CBBC is determined primarily by the positive amount of the settlement price less
the strike price. The settlement price is typically the minimum (maximum) price of the underlying asset for a bull
(bear) CBBC after the MCE and up to the next trading session. Day-high at the close is the fraction of observations
that the closing price (last transacted price) equals the highest transacted price of the day in the CAS (non-CAS)
period; and day-low at the close is the fraction of observations that the closing price (last transacted price) equals the
lowest transacted price of the day in the CAS (non-CAS) period. Figures in percentage terms are reported in
parenthesis and the corresponding standard errors are reported in bracket.

Pre-CAS CAS Post-CAS

Number of expired CBBCs 55 830 2719
(0.58%) (10.6%) (27.9%)

Number of expired bull CBBCs 50 495 1344

(0.52%) (6.30%) (13.8%)

Number of expired bear CBBCs 5 335 1375

(0.05%) (4.26%) (14.1%)

Fraction of day-high at the close 0.0372 0.0719 0.0373

[0.189] [0.258] [0.189]

Fraction of day-low at the close 0.0271 0.0728 0.0282

[0.162] [0.260] [0.166]

N 9,547 7,856 9,736

41

Table 7. Sniping and Expired CBBCs

We use a probit regression model to examine the relation between sniping and the number of expired CBBCs after
controlling for firm fixed-effects. The dependent variable is our measure of sniping in price (snipe(p)), trade
volume (snipe(v)) or both price and trade volume (snipe(pv)). The variables snipe(p), snipe(v), and
snipe(pv) take the value of one if the absolute change in price, trade volume, and both price and trade volume
during the final 10-second interval before the close for the stock is strictly greater than the absolute change in price,
trade volume, and both price and trade volume, respectively, for the stock of any of the remaining five 10-second
intervals of the closing minute, and zero if otherwise. The variable ncbbc is the number of expired CBBCs on the
day for the stock, and zero if otherwise. The variable nonCAS takes the value of one if the observation is taken in
the pre-CAS and post-CAS periods, and zero if otherwise. The corresponding robust standard errors clustered at the
firm-period level are reported in parenthesis. Statistical significance is marked at the 1% (***), 5% (**) and 10%
(*) levels.

(1) (2) (3)

snipe(p) snipe(v) snipe(pv)
ncbbcnonCAS -0.0828** -0.0168 -0.0306
(0.0394) (0.0309) (0.0335)

ncbbc 0.0871** 0.0198 0.0493

(0.0392) (0.0302) (0.0331)

nonCAS -0.0313 -1.0489*** -0.9488***

(0.0266) (0.0277) (0.0388)

constant -1.2343*** -0.5698*** -1.4231***

(0.0301) (0.0381) (0.0432)

Firm fixed-effect Yes Yes Yes

N 27,139 27,139 27,139
Pseudo R2 0.0039 0.1378 0.1289

42

Table 8. Expired CBBCs and Likelihood of Day-high and Day-low at the Close
We use a probit regression model to examine the effect of expired CBBCs on the probability of day-high and day-
low at the close. The dependent variable is either a binary variable for day-high at the close, a binary variable for
day-low at the close, or stock return in percent in the final 10 seconds interval before the close. Specifically, day-
high takes the value of one if the closing (last transacted) price of a stock is the highest transacted price of the day
for the stock in the CAS (non-CAS) period, and zero otherwise; and day-low takes the value of one if the closing
(last transacted) price of a stock is the lowest transacted price of the day for the stock in the CAS (non-CAS) period,
and zero otherwise. The variable p10s measures the stock return in the last ten seconds interval before the close.
The variable nbear (nbull) is the number of expired bear (bull) CBBCs for a firm in the day, and zero if
otherwise. The variable nonCAS takes the value of one if the observation is taken in the pre-CAS and post-CAS
periods, and zero if otherwise. The corresponding robust standard errors clustered at the firm-period level are
reported in parenthesis. Statistical significance is marked at the 1% (***), 5% (**) and 10% (*) levels.

(1) (2) (3)

day-low day-high p10s
nbullnonCAS -0.1093*** 0.0716*
(0.0372) (0.0402)

nbull 0.1125*** -0.0722*

(0.0282) (0.0402)

nbearnonCAS -0.1642*** -0.0028

(0.0444) (0.0057)

nbear 0.1977*** 0.0054

(0.0419) (0.0049)

nonCAS -0.4655*** -0.3223*** -0.0132*

(0.0333) (0.0288) (0.0072)

constant -1.5280*** -1.5066*** 0.0149**

(0.0473) (0.0431) (0.0066)

Firm fixed-effect Yes Yes Yes

N 27,139 27,139 27,139
Pseudo R2 0.0430 0.0299 0.0035

43

Table 9. Informativeness of Closing Price across Periods

We use a firm fixed-effects model to examine informativeness of the closing price in the pre-CAS, CAS, and post-
CAS periods. We use goodness-of-fit (i.e., adjusted R-square) of the market model as a proxy for price
informativeness. We use two stock return intervals to measure goodness-of-fit of the market model: (i) close-to-
close return; and (ii) last-ten-minute return. Specifically, for each stock in each period, we regress the close-to-
close return for a stock on the market close-to-close return. The close-to-close return is the simple percentage return
of the closing price (last transacted price) between day t and day t+1 in the CAS (non-CAS) period. For each stock
in each period, we obtain an estimated R-square. To precisely measure price informativeness during the closing
period, we use the last-ten-minute return. The last-ten-minute return is measured over the final 10-minute interval
before the close. Specifically, it is measured by the last transacted price during the continuous trading session at
4:00 p.m. to the closing price at 4:10 p.m. of the current day during the CAS period. During the non-CAS period, it
is measured by the last transacted price at 3:50 p.m. to the last transacted price at 4:00 p.m. of the current day. The
variable preCAS (postCAS) takes the value of one if the observation is taken in the pre-CAS (post-CAS) period,
and zero if otherwise. The corresponding robust standard errors are reported in parenthesis. Statistical significance
is marked at the 1% (***), 5% (**) and 10% (*) levels.

(1) (2)
R2 (close-to-close) R2 (last-ten-minute)
preCAS -0.0804*** -0.0376***
(0.0221) (0.0072)

postCAS -0.1561*** -0.0387***

(0.0177) (0.0074)

constant 0.6426*** 0.0441***

(0.0114) (0.0048)
N 117 117
adj. R2 0.4324 0.4090

44

Table 10. Large Price Change and Informativeness of Closing Price

We use a firm fixed-effects model to examine the effect of large price changes during the final 10-minute before the
close on price informativeness. For each period, we rank observations according to their absolute percentage price
changes during the final 10-minute interval. Next, for each period a fixed number of observations with the largest
absolute price changes (e.g., top 10, top 100, and top 200) is chosen and pools together in each regression. We use
the close-to-open return to proxy for informativeness of the closing price. The close-to-open return is the simple
percentage return between the closing price (last transacted price) of day t and the opening price of day t+1 in the
CAS (non-CAS) period. The variable R10m is the simple stock return in the final 10-minute interval before the close,
i.e., 4:004:10 p.m. in the CAS period and 3:504:00 p.m. in the non-CAS periods. The variable preCAS
(postCAS) takes the value of one if the observation is taken in the pre-CAS (post-CAS) period, and zero if
otherwise, and Rm is the close-to-open return for the Hang Seng Index of the day. The corresponding robust standard
errors clustered at the period level are reported in parenthesis. Statistical significance is marked at the 1% (***), 5%
(**) and 10% (*) levels.

Dependent Variable: Close-to-open Return

(1) (2) (3) (4) (5) (6)
Top 10 Top 100 Top 200 Top 500 Top 1500 Full Sample
R10mpreCAS 0.3091 0.2906** 0.1580* -0.0002 -0.0320* -0.0560**
(0.5962) (0.0497) (0.0379) (0.0179) (0.0109) (0.0068)

R10mpostCAS 0.4876*** 0.1329** 0.0761* -0.0442** -0.0835** -0.1012***

(0.0382) (0.0268) (0.0199) (0.0060) (0.0129) (0.0066)

R10m -0.8740*** -0.4762*** -0.3631*** -0.2798*** -0.2533*** -0.2379***

(0.0532) (0.0366) (0.0330) (0.0092) (0.0083) (0.0048)

preCAS 7.3294 -0.5835* -0.4065** -0.1249 -0.0544 0.0483**

(5.5893) (0.1362) (0.0675) (0.0603) (0.0236) (0.0063)

postCAS 6.4051* -0.1608 -0.0343 -0.0016 -0.0382 0.0186

(1.6934) (0.1006) (0.1193) (0.0377) (0.0158) (0.0068)

Rm 0.0884 0.9037 1.0112*** 0.9971*** 0.9985*** 0.9614***

(1.4915) (0.0418) (0.0075) (0.0343) (0.0199) (0.0199)

constant -3.3796* 0.6122** 0.2586* 0.1929** 0.1314*** 0.0382***

(0.9707) (0.1284) (0.0742) (0.0239) (0.0105) (0.0030)

Firm Fixed-effect Yes Yes Yes Yes Yes Yes

N 30 300 600 1,500 4,500 27,055
Pseudo R2 0.5447 0.4446 0.5356 0.5576 0.5899 0.5794

45

Table 11. Expired CBBCs and Informativeness of Closing Price

We use a firm fixed-effects model to examine the effect of the number of expired CBBCs on price informativeness.
We use the close-to-open return as a proxy for informativeness of the closing price. The close-to-open return is the
simple percentage return between the closing price (last transacted price) of day t and the opening price of day t+1 in
the CAS (non-CAS) period. The variable R10m is the simple stock return in the final 10-minute interval before the
close, i.e., 4:004:10 p.m. in the CAS period and 3:504:00 p.m. in the non-CAS periods. The variable ncbbc is
the number of expired CBBC for a firm on the day, and zero if otherwise; nonCAS takes the value of one if the
observation is taken in the pre-CAS and post-CAS periods, and zero if otherwise; and Rm is the close-to-open return
of the Hang Seng Index of the day. The corresponding robust standard errors clustered at the firm-period level are
reported in parenthesis. Statistical significance is marked at the 1% (***), 5% (**) and 10% (*) levels.

(1)
close-to-open return
ncbbcnonCASR10m 0.0868***
(0.0247)

ncbbcR10m -0.0817***
(0.0238)

ncbbcnonCAS 0.0294
(0.0491)

ncbbc -0.0306
(0.0491)

nonCASR10m -0.0907
(0.0560)

R10m -0.2194***
(0.0526)

nonCAS 0.0313
(0.0221)

Rm 0.9612***
(0.0300)

constant 0.0401**
(0.0194)

Firm fixed-effect Yes

N 27,055
adj. R2 0.5796

46

Table 12. Sniping and Informativeness of Closing Price

We use the firm fixed-effects model to examine the effect of sniping on price informativeness. We use the close-to-
open return as a proxy for informativeness of the closing price. The close-to-open return is the simple percentage
return between the closing price (last transacted price) of day t and the opening price of day t+1 in the CAS (non-
CAS) period. The binary variable snipe(x) takes the value of one if the absolute change in x for a stock during
the final 10-second interval before the close is strictly greater than any absolute change in x for the stock in any of
the remaining five 10-second intervals of the closing minute, where x is either p (price), v (trade volume), or pv
(both price and trade volume). The variable R10s is the simple stock return in the final 10-second interval before the
close for the stock. The variable nonCAS takes the value of one if the observation is taken in the non-CAS, and
zero otherwise, and Rm is the close-to-open return for the Hang Seng Index of the day. The corresponding robust
standard errors clustered at the firm-period level are reported in parenthesis. Statistical significance is marked at the
1% (***), 5% (**) and 10% (*) levels.

Dependent Variable: Close-to-open Return

(1) (2) (3)
snipe(p) snipe(v) snipe(pv)
snipe(x)nonCASR10s 0.6731*** 0.4597 0.7397**
(0.2508) (0.3150) (0.3387)
snipe(x)R10s -0.7326*** -0.5780** -0.7475***
(0.2352) (0.2395) (0.2233)
snipe(x)nonCAS -0.1190 0.0612 0.0517
(0.0769) (0.0612) (0.1312)
snipe(x) 0.0757 -0.0066 0.1090
(0.0680) (0.0482) (0.0822)
NONCASR10s -0.7388*** -0.6504*** -0.7251***
(0.1760) (0.1425) (0.1399)
R10s 0.2689* 0.1559 0.2235*
(0.1608) (0.1343) (0.1317)
nonCAS 0.0398 0.0234 0.0336
(0.0243) (0.0271) (0.0231)
Rm 0.9585*** 0.9583*** 0.9581***
(0.0300) (0.0301) (0.0301)
constant 0.0187 0.0280 0.0190
(0.0207) (0.0241) (0.0199)
Firm fixed-effects Yes Yes Yes

N 27,054 27,054 27,054

adj. R2 0.5739 0.5736 0.5739

47

Table 13. Top Ten Price Changes on Days when MSCI Performs Index Re-balancing

This table presents the top ten largest absolute percentage change in prices in the final 10-minute interval before the
close for all stocks listed on the main board of the HKEx with price of above $1 on the day when MSCI performs its
semi-annual index re-balancing activity in the CAS (in panels (a) and (b)), pre-CAS (in panel (c)) and post-CAS
periods (in panel (d)), respectively. The variable R10m is the simple stock return in the final 10-minute interval
before the close for a stock, i.e., 4:004:10 p.m. in the CAS period and 3:504:00 p.m. in the non-CAS periods; Rco
denotes the the close-to-open return and is the simple percentage return between the closing price of day t and the
opening price of day t+1; P10m,t is the nominal price 10-minute before the close in day t; Pc,t is the closing price in
day t; and Po,t+1 is the opening price in day t+1. Bold-faced indicates that a stock experiences price reversal on the
following open.

(a) MSCI Index Re-balance on May 30, 2008 CAS period

Stock Code Company Name R10m Rco P10m,t Pc,t Po,t+1
1224 C C Land 24.92 -15.09 6.42 8.02 6.81
0276 Mongolia Energy 18.16 -8.93 14.98 17.70 16.12
1880 Belle Intl. 13.72 -9.20 8.60 9.78 8.88
0069 Shangri-La Asia 13.36 -6.84 23.20 26.30 24.50
1186 China Railway Construction 12.18 -9.29 12.48 14.00 12.70
2626 Hunan Nonferrous Metals -11.82 7.28 2.96 2.61 2.80
1393 Hidili Industry Intl. Development 10.95 -8.55 13.70 15.20 13.90
3377 Sino-Ocean Land 10.32 -5.61 6.30 6.95 6.56
2689 Nine Dragons Paper 8.70 -5.88 7.82 8.50 8.00
0002 CLP 7.88 -6.38 65.35 70.50 66.00

(b) MSCI Index Re-balance on November 25, 2008 CAS period

Stock Code Company Name R10m Rco P10m,t Pc,t Po,t+1
0276 Mongolia Energy -29.12 13.51 2.61 1.85 2.10
1068 China Yurun Food Group 12.99 -3.00 8.85 10.00 9.70
2314 Lee & Man Paper Manufacturing -10.45 -5.00 2.68 2.40 2.28
0242 Shun Tak Hold. -9.86 5.47 1.42 1.28 1.35
0754 Hopson Development -8.88 1.03 2.14 1.95 1.97
0020 Wheelock and Co. 8.33 -5.38 12.00 13.00 12.30
1000 Beijing Media Corp. -7.95 9.09 2.39 2.20 2.40
2689 Nine Dragons Paper -7.26 2.61 1.24 1.15 1.18
0682 Chaoda Modern Agriculture 6.47 -4.90 4.79 5.10 4.85
0573 Tao Heung -6.45 6.03 1.24 1.16 1.23

48

(c) MSCI Index Re-balance on November 30, 2007 Pre-CAS period

Stock Code Company Name R10m Rco P10m,t Pc,t Po,t+1
0309 Lo's Enviro-Pro 14.75 0.00 1.22 1.40 1.40
0110 Fortune Telecom 12.40 0.69 1.29 1.45 1.46
1004 Rising Development 10.56 0.00 1.80 1.99 1.99
0508 Chevalier Pacific 10.44 -0.50 3.64 4.02 4.00
3933 United Laboratories Intl. 9.24 1.92 4.76 5.20 5.30
1033 Sinopec Yizheng Chemical Fibre 8.61 4.14 2.67 2.90 3.02
0163 Emperor Intl. 7.74 -1.15 3.23 3.48 3.44
0735 China Power New Energy Dev. 7.38 3.05 1.22 1.31 1.35
0577 Paul Y. Engineering Group 6.90 -0.65 1.45 1.55 1.54
3368 Parkson Retail Group -6.83 1.21 88.55 82.50 83.50

(d) MSCI Index Re-balance on May 29, 2009 Post-CAS period

Stock Code Company Name R10m Rco P10m,t Pc,t Po,t+1
0236 San Miguel Brewery Hong Kong 9.91 2.46 1.11 1.22 1.25
1109 China Resources Land 7.91 -2.89 16.68 18.00 17.48
0732 Truly Intl. Hold. -5.02 5.66 5.58 5.30 5.60
1882 Haitian Intl. Hold. 4.90 0.00 2.04 2.14 2.14
1813 KWG Property Holding -4.74 5.41 4.85 4.62 4.87
0583 SCMP Group 4.35 1.39 1.38 1.44 1.46
0927 Fujikon Industrial Hold. 4.08 0.00 1.47 1.53 1.53
0966 China Insurance Intl. Hold. 4.02 0.00 13.94 14.50 14.50
2398 Good Friend Intl. Hold. 4.00 -5.38 1.25 1.30 1.23
1766 CSR 3.89 3.16 4.88 5.07 5.23

49

Table 14. Restriction on 24-spread on Informativeness of Closing Price

We use a firm fixed-effects model with the close-to-open return as the dependent variable. The variable R10m is the
simple stock return between 4:004:10 p.m. in the CAS period and between 3:504:00 p.m. in the non-CAS periods.
The variable buy24spr (sell24spr) equals one if the largest (smallest) indicative equilibrium price taken
within the final 2-minute interval before the close deviates more than 24 spreads from the transacted price taken at
the final 10-minute interval before the close, i.e., 4:00 p.m. in the CAS period and 3:50 p.m. in the non-CAS periods,
and equals zero otherwise. The variable 24spr takes the value of one if buy24spr is one, minus one if
sell24spr is one, and zero if otherwise. The variable nonCAS equals one if the observation is taken in the pre-
CAS and post-CAS periods, and equals zero otherwise; and Rm is the close-to-open return of the Hang Seng Index of
the day. The corresponding robust standard errors clustered at the firm-period level are reported in parenthesis.
Statistical significance is marked at the 1% (***), 5% (**) and 10% (*) levels.

(1) (2)
close-to-open return close-to-open return
buy24sprnonCAS 1.7125*
(1.0124)

buy24spr -1.9753***
(0.7159)

sell24sprnonCAS -3.2880***
(0.8003)

sell24spr 2.1938***
(0.5658)

24sprnonCAS 2.3178***
(0.7598)

24spr -2.0425***
(0.5908)

nonCASR10m -0.1069* -0.1080*

(0.0546) (0.0547)

R10m -0.2048*** -0.2049***

(0.0511) (0.0511)

nonCAS 0.0330 0.0337

(0.0223) (0.0222)

Rm 0.9609*** 0.9612***
(0.0301) (0.0301)

constant 0.0384* 0.0382*

(0.0198) (0.0196)

Firm fixed-effect Yes Yes

N 27,055 27,055
adj. R2 0.5800 0.5800

50

Table 15. Refinements to Closing Auction Mechanism to Reduce Extreme Price Movement
around Ten Major Worldwide Stock Exchangesa

This table presents refinements to the closing auction mechanism to reduce extreme price movement in ten major
stock exchanges around the World.

Stock Exchanges Daily Price Limit Deadline Other refinements on the closing auctions
Tokyo Stock Exchange Yes Fixed No
(sliding scale with
respect of the
previous closing
price)
Korea Exchange Yes Fixed No
(15% of the
previous closing
price)
Taiwan Stock Exchange Yes Fixed No
(7% of the
previous closing
price)
Shenzhen Stock Yes Fixed No
Exchange (10% of the
previous closing
price)
Australian Securities No Random No
Exchange
London Stock Exchange No Random No
New York Stock No Fixed Yes
Exchangeb (accepts on-close orders on the stabilizing
side of the market in the final 15 minutes)
Nasdaqb No Fixed Yes
(accepts imbalance-only orders on the
stabilizing side of the market in the final 20
minutes)
Deutsche Brse No Random Yes
(triggers a volatility auction when the closing
price deviates significantly from the last
transacted price)
Euronext No Fixed Yes
(triggers a volatility interruption when the
closing price reaches certain price limits)
a
Information on the refinements on the closing auction mechanisms among worldwide stock exchanges are
obtained from the two HKEx consultation papers: (i) Introduction of a price control mechanism during the closing
auction session in the securities market and (ii) The introduction of a closing auction session available from
http://www.hkex.com.hk/eng/newsconsul/mktconsul/marketconsultation.htm.

b
Information on other refinements on the closing auction mechanism for the NYSE and Nasdaq are obtained from
their corporate websites.

51

Appendix 1. Algorithm to Compute IEP and IEV during the Closing Auction Session

The indicative equilibrium price (IEP) must be a price at and between the highest limit bid and
the lowest limit ask and maximizes the matched shares, i.e., the indicative equilibrium volume
(IEV). If there is a tie in IEP, three tie-breaker rules apply. The first rule selects the price with
the lowest order imbalance as the IEP. If this rule fails to break the tie, the second rule would
apply and pick the price which is closest to the nominal price at 4 p.m. as the IEP. If these two
rules fail, the third rule would apply and pick the highest price as the IEP.

The following example illustrates the algorithm to compute IEP, IEV, and the primary queue for
buy and sell orders during the CAS. The primary queue is the queue of at-auction orders and at-
auction limit orders with a specified price at or more competitive than the IEP. Let us assume
that the best bid and offer at 4pm is $37 and $38, respectively and the limit order book at
4:07:59pm is presented in (i) as follows:

(I) Benchmark Case

4:07:59 p.m. Price Acc. Acc. Matched Order
Bid (Buy Orders) Ask (Sell Orders) Buy Sell Order Imbalance
Price Quantity Price Quantity $39 2,000 13,500 2,000 11,500
At-auction 1,000 At-auction 2,000 $38 3,000 3,500 3,000 500
$39 1,000 $37 1,000 $37 4,000 3,000 3,000 1,000
$38 1,000 $38 500
$37 1,000 $39 10,000 Primary Queue IEP IEV
Buy 3,000 $38 3,000
Sell 3,500

(IA) Sniping at the Sell-side: A large at-auction sell-order of 18,000 arrives at 4:09:58 p.m.
4:09:58 p.m. Price Acc. Acc. Matched Order
Bid (Buy Orders) Ask (Sell Orders) Buy Sell Order Imbalance
Price Quantity Price Quantity $39 2,000 31,500 2,000 29,500
At-auction 1,000 At-auction 20,000 $38 3,000 21,500 3,000 18,500
$39 1,000 $37 1,000 $37 4,000 21,000 4,000 17,000
$38 1,000 $38 500
$37 1,000 $39 10,000 Primary Queue IEP IEV
Buy 4,000 $37 4,000
Sell 21,000

(IB) Sniping at the Buy-side: A large at-auction buy-order of 18,000 arrives at 4:09:58 p.m.
4:09:58 p.m. Price Acc. Acc. Matched Order
Bid (Buy Orders) Ask (Sell Orders) Buy Sell Order Imbalance
Price Quantity Price Quantity $39 20,000 13,500 13,500 6,500
At-auction 19,000 At-auction 2,000 $38 21,000 3,500 3,500 17,500
$39 1,000 $37 1,000 $37 22,000 3,000 3,000 19,000
$38 1,000 $38 500
$37 1,000 $39 10,000 Primary Queue IEP IEV
Buy 20,000 $39 13,500
Sell 13,500

52

(II) Benchmark Case with a small but aggressive limit sell-order at $33
4:07:59 p.m. Price Acc. Acc. Matched Order
Bid (Buy Orders) Ask (Sell Orders) Buy Sell Order Imbalance
Price Quantity Price Quantity $39 2,000 13,600 2,000 11,600
At-auction 1,000 At-auction 2,000 $38 3,000 3,600 3,000 600
$39 1,000 $33 100 $37 4,000 3,100 3,100 900
$38 1,000 $37 1,000 $33 4,000 2,100 2,100 1,900
$37 1,000 $38 500
$39 10,000 Primary Queue IEP IEV
Buy 4,000 $37 3,100
Sell 3,100

(IIA) Sniping at the Sell-side: A large at-auction sell-order of 18,000 arrives at 4:09:58 p.m.
4:09:58 p.m. Price Acc. Acc. Matched Order
Bid (Buy Orders) Ask (Sell Orders) Buy Sell Order Imbalance
Price Quantity Price Quantity $39 2,000 31,600 2,000 29,600
At-auction 1,000 At-auction 20,000 $38 3,000 21,600 3,000 18,600
$39 1,000 $33 100 $37 4,000 21,100 4,000 17,100
$38 1,000 $37 1,000 $33 4,000 20,100 4,000 16,100
$37 1,000 $38 500
$39 10,000 Primary Queue IEP IEV
Buy 4,000 $33 4,000
Sell 20,100

(IIB) Sniping at the Buy-side: A large at-auction buy-order of 18,000 arrives at 4:09:58 p.m.
4:09:58 p.m. Price Acc. Acc. Matched Order
Bid (Buy Orders) Ask (Sell Orders) Buy Sell Order Imbalance
Price Quantity Price Quantity $39 20,000 13,600 13,600 6,400
At-auction 19,000 At-auction 2,000 $38 21,000 3,600 3,600 17,500
$39 1,000 $33 100 $37 22,000 3,100 3,100 18,900
$38 1,000 $37 1,000 $33 22,000 2,100 2,100 19,900
$37 1,000 $38 500
$39 10,000 Primary Queue IEP IEV
Buy 20,000 $39 13,600
Sell 13,600

(IIC) Order Cancellation: The limit sell-order at $33 is canceled prior to 4:08:00 p.m.
4:08:00 p.m. Price Acc. Acc. Matched Order
Bid (Buy Orders) Ask (Sell Orders) Buy Sell Order Imbalance
Price Quantity Price Quantity $39 2,000 13,500 2,000 11,500
At-auction 1,000 At-auction 2,000 $38 3,000 3,500 3,000 500
$39 1,000 $33 100 $37 4,000 3,000 3,000 1000
$38 1,000 $37 1,000
$37 1,000 $38 500
$39 10,000 Primary Queue IEP IEV
Buy 3,000 $38 3,000
Sell 3,500

53