Standard Call Auction and Closing Price Manipulation: Evidence from the Hong Kong Stock Exchange *

Transcription

1 Standard Call Auction and Closing Price Manipulation: Evidence from the Hong Kong Stock Exchange * Seongkyu Gilbert Park a, Wing Suen b and Kam-Ming Wan c Abstract. The Hong Kong Stock Exchange adopted a plain-vanilla closing call auction in 2008 and suspended its operation ten months later due to widespread suspicion of price manipulation. The Exchange relaunched the auction in 2016 with manipulation-deterrence enhancements. This unique setting allows us to apply the difference-in-differences methodology to examine the causal effect of a standard call auction on price manipulation. Our results indicate that a plainvanilla call auction mechanism is prone to closing price manipulation. Under this mechanism overnight price reversal is more pronounced on suspected manipulation days, e.g., days when sniping occurs and days when derivative options contracts expire. Last Revised: September 17, 2018 Keywords: Manipulation, Closing Call Auction, Overnight Price Reversal, Sniping, CBBC JEL Classification: D44, G14 * We are grateful to the HKEx for providing us data on callable bull and bear contracts and Matthias Buehlmaier, Steve Ching, Tarun Chordia, Carole Comerton-Forde, Y.K. Fu, Dave Mauer, Mark Seasholes, Avanidhar Subrahmanyam, Yexiao Xu, Harold Zhang, and seminar participants at the University of Hong Kong, the University of Texas at Dallas, 2012 FMA Annual Meeting, the 2012 China International Conference in Finance, and the 19th Conference on the Theories and Practices of Securities and Financial Markets for their comments. We are thankful to excellent research support from Ho-yin (Ernest) Ching, Chi-Chen (Fira) Ho, Kai-Jun (Jeremy) Lau, and Yangbo (Darcy) Song for their excellent research assistance. All errors remain ours. The former title of this paper is Sniping to Manipulate Closing Prices in Call Auctions: Evidence from the Hong Kong Stock Exchange. a Hong Kong Polytechnic University, School of Accounting and Finance, Li Ka Shing Tower, Hung Hom, Hong Kong, gilbert.park@polyu.edu.hk. b University of Hong Kong, Faculty of Business and Economics, K.K. Leung Building, Pokfulam Road, Hong Kong, wsuen@econ.hku.hk. c Hong Kong Polytechnic University, School of Accounting and Finance, Li Ka Shing Tower, Hung Hom, Hong Kong, kmwan@polyu.edu.hk. 1

2 1. Introduction The call auction mechanism is the dominant procedure to determine closing prices of listed equities. In 2018, eight of the world s top ten stock exchanges use this mechanism to set closing prices. The closing call auction mechanism is popular because it facilitates trades at the market close. It is also widely believed that call auctions improve price discovery and are robust to price manipulation. This is because they can absorb extreme liquidity shocks as order flows are consolidated at a single point in time (Barclay et al., 2008). Extant studies provide evidence that is broadly consistent with these beliefs (e.g., Chang, Rhee, Stone and Tang, 2008; Comerton- Forde, Lau, and McInish, 2007; and Hillion and Souminen, 2004; Pagano and Schwartz, 2003; Schwartz, 2001; Cao, Ghysels, and Hatheway, 2000; Biais, Hillion, and Spatt, 1999; Madhavan, 1992). However, the experience at the Hong Kong Stock Exchange (HKEx) challenges these beliefs, particularly the call auction s ability to deter price manipulation. To facilitate institutional investors to trade at the closing price, the HKEx introduced a standard closing call auction mechanism on May 26, The auction was conducted in a closing auction session ( CAS ) from 4:00-4:10 p.m. This mechanism is unique because it is devoid of any manipulation-deterrence features such as binding price limit, randomization of closing times, and volatility extensions. Just five days after the introduction, excessive price movements occurred during the closing session on the MSCI index rebalancing day. 2 This mechanism was eventually suspended due to extreme volatility and widespread suspicion of price manipulation: 1 Throughout the paper, standard and plain-vanilla are used interchangeably

3 Almost immediately, the system attracted criticism after closing prices in some shares began swinging wildly during the 10-minute auction. ( HKEx drops late auction, The Wall Street Journal Asia, March 13, 2009.) To reduce potential manipulation and extreme price volatility, the HKEx revamped the closing call auction mechanism ( revampcas ) and re-launched it on July 25, The revamped mechanism includes several new enhancements to deter manipulation, particularly random closing time and binding price limit. We use these two episodes to examine the relation of a plain-vanilla call auction mechanism and closing price manipulation. Our finding indicates that a plain-vanilla call auction mechanism is vulnerable to closing price manipulation, especially on days when the motives for manipulation are strong, for example, days when sniping occurs and days when callable bull and bear contracts (CBBCs) expire. A case in point was the plunge in HSBC shares on March 9, The plunge was the biggest one-day drop since Black Monday in 1987, and its impact was exceptionally large because HSBC was the largest company on the HKEx, having at that time a global market capitalization of approximately US$200 billion. On that day, the stock quote of HSBC fluctuated within a tight range of $37-$38 during most of the closing auction session. 3 Two seconds before the end of the closing session, a trader submitted an exceptionally large sell order. HSBC shares immediately plunged from $37 to close at $33, representing a drop of over ten percent in merely two seconds. When the market opened the next day, HSBC shares bounced back to the pre-sniping level of $ Panel A of Figure 1 presents the indicative equilibrium price and the primary buy and sell queue of HSBC shares during the closing session. The sell 3 Stock prices are quoted in Hong Kong dollars. The Hong Kong dollar is pegged to the U.S. dollar at an exchange rate of US$1:HK$ Throughout this paper, sniping is referred to unexpectedly large orders submitted in the last few seconds before the market close. 3

4 order was remarkably large, at over 4.7 million shares. That order alone represented fifty percent of the then-prevailing indicative equilibrium volume at the closing session. [Insert Figure 1 here.] We believe that the sniping incident on March 9, 2009 was well planned because the trading of HSBC shares was also unusual on the previous trading day (March 6). Panel B of Figure 1 shows the trading activities during the closing session on that day. First, a small but unusually aggressive limit sell order of $33 was submitted between 4:01 p.m. and 4:02 p.m., causing the indicative equilibrium price to fall immediately to $33. Coincidentally, this indicative equilibrium price at $33 was 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 over 5.8 million shares was canceled just one second prior to the cancellation deadline at 4:08 p.m. It appears that some traders attempted to manipulate the stock price of HSBC on March 6, 2009 but were not successful. Alternatively, they might have been testing the market reaction should HSBC shares drop to $33. In this paper, we systematically evaluate the performance of the closing call auction mechanism adopted by the HKEx. Our paper adds to the literature in several ways. First, contrary to conventional wisdom, we find that a standard call auction mechanism is prone to manipulation. In particular, a non-binding price limit increases profit from price manipulation (Kim and Park, 2010). Under the CAS regime, orders submitted during the closing auction ranged from +800% to -89% of the last traded price. In contrast, the price limit during the continuous trading session was only 24 ticks, which represents approximately ±2.4% of the 4

5 prevailing market price. Furthermore, sniping can be more successfully implemented under a fixed closing time because manipulators can surprise the market by submitting exceptionally large orders shortly before the end of the auction, leaving no time for other traders to react. Our findings are consistent with the auction design literature, which emphasizes that minor institutional details matter. Hasbrouck (2007, p.18) states that although auctions may appear simple, seemingly minor details of implementation can have profound effects. Consistent with this claim, current studies indicate that randomization of closing times and well-designed call auction matching algorithms can reduce manipulation (Hauser, Kamara, Shurki, 2012; Comerton-Forde and Rydge, 2006). Second, our paper contributes to the literature as it provides a clean identification which allows us to study the causal effect of adopting a plain-vanilla call auction on closing price manipulation. The unique experience of the HKEx to adopt, and then to suspend, and eventually to relaunch the closing auction mechanism allows us to apply the difference-in-differences methodology. Thus, our setting is less susceptible to possible biases from confounding factors that are correlated with time trends. Existing studies of other markets typically rely on a beforeand-after comparison, which may be difficult to disentangle from unobserved time trends. Third, we study a motive for manipulation in closing auctions. In particular, we link the performance of closing auctions of the underlying equity to the expiration day of a derivative product¾cbbc. CBBCs are also known as turbo warrants which are similar to knock-out options in that they expire if the price of the underlying stock reaches a pre-specified level, i.e., strike price. Importantly, on the expiration day CBBC investors are entitled to a residual value which can be determined by the closing price of the underlying stock. As CBBCs are actively traded among investors in Hong Kong, the incentive for manipulation is strong, particularly on 5

6 the expiration day. For example, in 2009 the total turnover value of CBBCs was approximately HK$1.7 trillion, constituting about 11 percent of total turnover on the HKEx. Consistent with our expectations, we find that during the CAS regime overnight price reversal is more likely to occur on the CBBC expiration day. However, under the revampcas regime we find no such price reversal on the expiration day. Our results corroborate evidence that abnormal trading strategies can be devised on the CBBC expiration day (Lei, 2014; Cheung et al., 2010). Fourth, we use market micro data at very fine time intervals (5-second and 10-minute intervals) to measure price movements. Focusing on such fine details allows us to isolate the effects of market structure from the effects of broader market movements. Doing so is particularly relevant, as the stock market was extremely volatile and on the decline during the CAS period. However, the extreme market condition should have no material effect on our results, as the relevant variables are measured at very fine intervals. For example, the absolute value of stock return in the final 5-second interval before the close was on average 4.6 basis points during the CAS period when the market condition was extremely turbulent. Last, many studies use prosecuted cases to examine price manipulation (Comerton-Forde and Putniņš, 2011&2013; Comerton-Forde and Rydge, 2006; Aggarwal and Wu, 2006). They find that price manipulation typically occurs among small and illiquid stocks. Because our sample comprises the largest companies of the HKEx, our evidence suggests that manipulation can occur even for large and liquid stocks. Further, this study sheds light on the limitation of using prosecuted cases to draw inferences on the extent of price manipulation, as the number of 6

7 prosecuted cases is very small. 5 Further, they are limited by the capability of regulators to accurately detect and successfully prosecute price manipulators. 2. Trading Mechanisms Trading in the HKEx is conducted in two continuous auction sessions. The morning session is from 9:30 a.m.-12:00 p.m. and the afternoon session is from 1:00-4:00 p.m. The opening price is determined by a call auction mechanism with a fixed opening time in the pre-opening session from 9:00-9:30 a.m. 6 During our sample period, the HKEx used three different mechanisms to determine the closing price Closing Auction Session: CAS regime For approximately ten months from May 26, 2008 to March 22, 2009, the HKEx introduced a closing auction session to determine the closing price. This session lasted ten minutes from 4:00-4:10 p.m. The closing price was concluded by a call auction mechanism with a fixed closing time at 4:10 p.m. During the closing session, buy and sell orders set to determine the closing price would accumulate. Traders could place, modify and cancel market and limit orders during the first eight minutes of the session. 7 In the final two minutes of the session, new market 5 Despite our statistical evidence, which suggests systematic attempts of closing price manipulation when the HKEx first adopted the closing call auction mechanism from May 26, 2008 to March 22, 2009, we find only three prosecuted cases of closing price manipulation of listed equities during that period. 6 During our sample period, changes in HKEx trading hours were made to increase overlap the trading periods with the China s stock market trading hours. Until March 6, 2011, pre-opening auctions were held from 9:30 a.m. to 10:00 a.m., morning session from 10:00 a.m. to 12:30 p.m., and afternoon session from 2:00 p.m. to 4:00 p.m. From March 7, 2011 to March 4, 2012 (phase one), trading hours were revised so that pre-opening session holds from 9:00 a.m. to 9:30 a.m., morning session from 9:30 a.m., to noon, and afternoon session from 1:30 p.m. to 4:00 p.m. Current trading hours (phase two) began from March 5 th of See Market/News/Market-Consultations/2006-to-2010/September-2010-Consultation-Paper-on-Proposed- Changes/Conclusions/cp cc.pdf for more details about the trading hours changes around our sample period. 7 The official names for market and limit orders allowed during the closing auction session are at-auction orders and at-auction limit orders, respectively. 7

8 orders were allowed to place but modification and cancellation of existing orders were disallowed. Throughout the closing session, market participants could observe the resulting indicative equilibrium price (IEP) and indicative trade volume (IEV), without any trades actually taking place. 8 At 4:10 p.m., orders were matched, and transactions were executed at the indicative equilibrium price at 4:10 p.m., which was the closing price of the day. 9 The indicative equilibrium price is the price obtained from a single-price call auction that yields the maximum number of shares transacted. In addition, short selling was prohibited during the closing session. Appendix A describes the algorithm used to compute IEP and IEV during the closing session. During the closing auction session, the price specified by a limit order was bounded by the 9-times restriction rule : the upper price limit of submitted orders is 9 times (+800%) the nominal price at 4:00 p.m., and the lower price limit of submitted orders is 1/9 of (-89%) the nominal price at 4:00 p.m. In contrast, limit orders submitted during the continuous trading session were restricted to much narrower bands. Specifically, they must be input within 24 ticks of the prevailing market price (henceforth the 24-tick rule ). For example, 24 ticks correspond to a price range of ±2.4% for a stock trading at $ Median Pricing Procedure: Non-CAS regime Starting in 1993, the HKEx used the median pricing procedure to compute the closing price. The closing price was determined by the median of five snapshot nominal prices in the final minute of the afternoon session. These nominal prices are taken at five specific times, which are 15 8 According to the call auction matching algorithm of the HKEx, 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 the 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. 9 If there is no IEP available after the closing session (i.e., there is no match at all), the closing price is determined by the median of five nominal prices in the final minute of the afternoon session. 8

9 seconds apart; from 3:59-4:00 p.m. Choosing the median of these prices ensure that the closing price would not be unduly influenced by one single trade. Note that the 24-tick rule applied to orders submitted during this procedure because it is part of the continuous trading session. This mechanism was abandoned when the closing auction session was first introduced on May 26, However, it was reinstated on March 23, 2009 after the closing auction session was suspended and remained effective until the HKEx introduced the revamped closing call auction mechanism on July 24, Revamped Closing Auction Session: RevampCAS regime On July 24, 2016, the HKEx reintroduced the closing auction session with two major enhancements in call auction design to deter manipulation. First, the closing time is randomized between 4:08-4:10 p.m. Second, the price limit is significantly reduced to only five percent (±5%) of the reference price set at 4:00 p.m. The reference price is determined by the median of five nominal prices taken in the final minute of the continuous trading session rather than a single nominal price taken at 4:00 p.m. used in the CAS period. Other minor enhancements include no modification and cancellation of existing orders during the final four minutes of the closing session. However, limit orders are allowed throughout the entire session. Other design features are identical to those used in the CAS period. Currently, the revamped mechanism is applied to mostly large and liquid stocks, including constituent stocks of Hang Seng Composite LargeCap and MidCap Indexes, H shares which have A shares listed on a Mainland Exchange, and all exchange traded funds. 3. Data and Sample 9

10 Our sample begins with all constituent stocks in the Hang Seng Index between May 26, 2008 and January 22, Our final sample has 36 stocks after excluding stocks with less than 180 daily observations in each of the four sub-periods (to be described later). Appendix B presents a list of stocks used in the final sample. These stocks are large and comparable in size to S&P500 component stocks, with market capitalizations ranging from US$2.9 billion (COSCO Pacific) to US$200.5 billion (HSBC) in Our sample stocks were actively traded and constituted 61 percent of the market capitalization of all stocks listed on the HKEx at the end of We choose these stocks for two reasons. First, they allow us to evaluate the vulnerability of a standard call auction to closing price manipulation for large and liquid stocks. Second, closing prices of these stocks are commonly used to price derivative products. This allows us to study the incentive of manipulation under different closing mechanisms. Our study combines two sample periods covering a total of 38 months. The first sample period is from May 22, 2008 through January 22, 2010 which overlaps with the subprime mortgage crisis ( Crisis period ). We further divide this period into two 10-month sub-periods covering two different closing mechanisms. The first sub-period is 5/26/2008-3/22/2009 under the closing auction session ( CAS ) regime. The second sub-period is 3/23/2009-1/22/2010 under the median pricing procedure ( non-cas ) regime. The second sub-period is used as a control group to compute the first difference in the main variable of interest during the crisis period. The second sample period is from May 25, 2015 through April 24, 2017 which is after the subprime mortgage crisis ( non-crisis period ). Similarly, we split it into two 9-month subperiods covering two different closing mechanisms. The first sub-period is 7/25/2016-4/24/2017 under the revamped closing auction session ( revampcas ) regime. The second sub-period is 10

11 10/25/2015-7/24/2016 under the median pricing procedure ( non-cas ) regime. This second sub-period is used as a control group to compute the second difference of the main variable of interest during the non-crisis period. We construct our dataset from six data files published by the HKEx. 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, Historical Full Book Market Securities Market files, and Historical Order Book and Statistics Update - Securities Market files; data on transaction prices and volume are from the Trade Record files; data on day-high and day-low events are taken from the Day-end Closing Data files; and data on expiration dates of CBBCs are taken from Stock Static Data (Daily) files, HKEx Fact Book, and the HKEx website. 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. 4. Empirical Strategy: Difference-in-differences methodology To examine the effect of standard closing price auction on closing price manipulation, we apply the difference-in-differences (DD) methodology. Our baseline regression model is as follows: (1) yi,t = b1 CASt CrisisPeriodt + b2 CASt + b3 CrisisPeriodt + controls + b0 + vi,t where yi,t is a proxy for the outcome of closing price manipulation; CASt is a binary variable equal to one when the closing price is determined by a call auction mechanism (i.e., under the CAS and revampcas regimes) and zero otherwise; CrisisPeriodt is a binary variable equal to one during the subprime mortgage crisis period ( ) and zero otherwise. This variable measures the increased demand for closing price manipulation because the stock market condition was extremely volatile during the subprime mortgage crisis period. The estimate b1 11

12 measures the standard call auction effect on closing price manipulation. In particular, this is a difference-in-differences estimate. The first difference measures the response of closing price manipulation to the introduction of the plain-vanilla closing auction during the crisis period. The second difference is the response of closing price manipulation to the introduction of the revamped closing auction during the non-crisis period. 5. Empirical Analysis 5.1. Sniping The most important price of the day is the closing price because it is used for many purposes. For example, they are used to price derivative products, determine mutual funds value and performance, and ascertain brokers execution quality (Kumar and Seppi, 1992; Stoll and Whaley, 1991; Chamberlain, Cheung, and Kwan, 1989). Therefore, the closing price is prone to manipulation. To influence the closing price, manipulators may submit unexpectedly large orders shortly before the market close to create order imbalances, leaving no time for the market to react (Allen and Gale, 1992; Putninš, 2012). Sniping creates order-imbalance at the market close, thus pushing the closing price away from the fundamental value. If sniping is motivated by manipulation reasons, the distorted closing price should be corrected and revert toward the fundamental value when new orders arrive at the market open on the following day. Therefore, we expect overnight price reversal to be more pronounced on days when manipulative sniping occurs. More specifically, if manipulative sniping is more likely to occur in a standard call auction than other closing procedures, we expect that overnight price reversal to be more pronounced under the CAS regime, compared with under other regimes. We follow Pagano and Schwartz (2003) and use the close-to-open return to measure the extent of overnight price 12

13 reversal. The close-to-open return is defined as (Po,t+1 Pc,t)/Pc,t, where Pc,t is the closing price on the day and P0,t+1 is the opening price of the following day. We have two measures of sniping: sniping in price and sniping in trade volume. Sniping in price refers to a sudden change in price just before the market close, whereas sniping in trade volume refers to a sudden surge in orders right before the auction ends. Current studies show that these measures are relevant attributes to identify the manipulator s intent (Comerton-Forde and Rydge (2006) and Comerton-Forde and Putninš (2011). 10 We construct the sniping variables as follows: snipe(x) is a binary variable that takes the value of one if the absolute change in x for a stock during the sniping measurement window is strictly greater than the absolute change in x for the stock in all benchmark intervals, where x is either p (price) or v (trade volume). We use the time when the closing price is set to define our sniping measurement window. The sniping measurement window is the 5-second interval before the market close under the CAS and revampcas regimes. Under the CAS regime, the sniping measurement window is 4:09:55 4:10:00 p.m. because the closing time is fixed at 4:10 p.m. Under the revampcas regime, the sniping measurement window varies daily because the closing time is randomized between 4:08 4:10 p.m. For example, in the revampcas regime if the market closes at 4:08:35 p.m., the sniping measurement window will be 4:08:30 4:08:35 p.m. During the non-cas regime, it is the 5-second interval prior to the snapshot nominal price being taken as the closing price. For example, if the median snapshot price occurs at 3:59:15 p.m., the sniping measurement window will be 3:59:10 3:59:15 p.m. To successfully influence the closing price in these periods, manipulators have to influence the median snapshot 10 Extant studies also find that a large increase in bid-ask spread is an important attribute for closing price manipulations (see Comerton-Forde and Rydge, 2006; Comerton-Forde and Putninš, 2011). However, we are unable to examine this attribute because data on bid-ask spreads are unavailable in the CAS and revampedcas periods, as only a single indicative equilibrium price is reported under the closing auction session (CAS). 13

14 price. To do that, they must manipulate at least three of the five snapshot nominal prices. Therefore, prices taken at these five snapshots might be manipulated and thus unsuitable as benchmark prices. Consequently, in these two periods, we choose four benchmark intervals in the final minute before the close: (i) 3:59:05 3:59:10 p.m., (ii) 3:59:20 p.m. 3:59:25 p.m., (iii) 3:59:35 p.m. 3:59:40 p.m., and (iv) 3:59:50 p.m. 3:59:55 p.m. Similarly, four comparable benchmark intervals are created during the closing minute under the CAS and revampcas regimes. 11 Under the CAS and revampcas regimes, we measure price and trade volume by the indicative equilibrium price (IEP) and the indicative equilibrium trade volume (IEV), respectively. Under the non-cas regime, they are measured by the actual transaction price and the actual transaction volume, respectively. Table 2 reports the mean likelihood of sniping by periods and the difference-indifferences estimates of these variables. We present data by CAS in columns (1) and (2) and the differences in mean likelihood of sniping between CAS and non-cas. Rows 3 and 6 of the table present the changes in mean likelihood of sniping between the crisis period and the non-crisis period. Rows 1 3 report the estimates for sniping in price whereas those for sniping in trade volume are reported in rows 4 6. The results in rows 1 and 4 indicate that during the crisis period it is inconclusive whether sniping is more likely to occur in a standard closing auction mechanism. Contradictory to our expectations, the mean likelihood of sniping in price is significantly less likely to occur in the standard closing auction than in the median pricing procedure during the crisis period For example, in the CAS period the four comparable benchmark intervals are fixed as follows: (i) 4:09:05 4:09:10 p.m., (ii) 4:09:20 4:09:25 p.m., (iii) 4:09:35 4:09:40 p.m., and (iv) 4:09:50 4:09:55 p.m. 12 Conversations with the HKEx officials indicated that non-manipulative sniping may be likely under the median pricing procedure. This is because many institutional investors (e.g., passively-managed mutual funds) are obligated to fill their customer orders at the closing price. Therefore, to reduce the tracking error they may submit orders shortly before the five snapshot times, hoping that the average transacted price is close to the actual closing price. Our results on overnight price reversal (to be reported in Table 3) are consistent with this claim. 14

15 [Insert Table 2 here] In contrast, the results in rows 2 and 4 show that, during the non-crisis period, sniping is meaningfully less likely to occur in the revamped closing auction than in the median pricing procedure. More precisely, sniping rarely occurs in the closing auction with manipulationdeterrence features. For example, the mean sniping in price is only 0.7 percentage point under the revamped CAS regime. It is immediately obvious that the difference-in-differences (DD) estimates are large and statistically distinguishable from zero. These estimates indicate that the relative increase in sniping is significantly large in the plain-vanilla auction mechanism. For example, the relative increase in mean sniping in price is 4.5 percentage point in the plain-vanilla auction mechanism. Overall, our results indicate that sniping is more likely to occur when a standard call auction is used to determine closing prices. However, sniping can be motivated by informational reasons. Thus, we use overnight price reversal to study this possibility. If sniping is motivated by private information, the closing price should agree with the fundamental value, resulting in no overnight price reversal. If sniping is motivated by manipulation, the closing price should deviate from the fundamental value, resulting in overnight price reversal. To measure the intensity of sniping, we interact the sniping variable with the stock return measured over the 5-second interval during the sniping measurement window (R5s). Our control variables include the stock return during the sniping measurement window (R5s) and the overnight return for the overall market (Rm). The overnight return for the overall market is included to account for the arrival of new overnight information which can affect the overnight price reversal. Therefore, we use a triple differences methodology to formulate our baseline model as follows: (2) Rcoi,t = g1 manipulatei,t CrisisPeriodt CASt + g2 manipulatei,t CrisisPeriodt 15

16 + g3 manipulatei,t CASt + g4 CrisisPeriodt CASt + g5 manipulatei,t + g6 CrisisPeriodt + g7 CASt + controls + firm FE + g0 + vi,t where Rcoi,t is the close-to-open return, and manipulatei,t = snipe(x) R5s captures the intensity of sniping. Table 3 presents firm fixed-effects regression results for the effect of the standard closing auction on overnight price reversal on days when sniping occurs. Columns (1) and (2) report the difference-in-differences estimates of this relation during the crisis period and the non-crisis period, respectively. The triple differences estimates are reported in column (3). The results based on sniping in price are reported panel A whereas those based on sniping in trade volume are reported in Panel B. [Insert Table 2 here] We find that overnight price reversal is pronounced subsequent to sniping under the plain-vanilla auction mechanism during the crisis period. For example, the results in column (1) of panel A indicate that during the crisis period if a sniping occurs and depresses the closing price by one percentage point, the stock price will bounce back by 1.2 percentage points on the following open under the plain-vanilla auction mechanism, compared with a further price decline of 0.22 (= ( )) percentage point on the following open under the median pricing procedure. Quite the contrary, the results in column (2) indicate that during the non-crisis period there is no meaningful overnight price reversal subsequent to sniping under the revamped closing auction mechanism. It is clear that the triple differences estimates are large and statistically distinguishable from zero. They indicate that the relative increase in overnight price reversal 16

17 subsequent to sniping is meaningfully large under the standard closing auction mechanism. Overall, our results point to the same conclusion that closing price manipulation via sniping is more likely to occur in a plain-vanilla call auction mechanism Expired CBBCs We use a particular type of derivative product callable bull and bear contracts to study a motive for closing price manipulation. We emphasize that this is not the only possible motive for price manipulation, but it is the only derivative product in the HKEx that uses the closing price of the underlying equity to determine the settlement price on the expiration day. 13 On the expiration day, CBBC buyers are entitled to a residual value which is determined by the settlement price less the strike price if this amount is positive and zero otherwise. For an expired bull (bear) CBBC contract, the settlement price is the minimum (maximum) price of the underlying stock from the expiration time to the next trading session of the day (typically including the market close). 14 To reduce the residual value, i.e., settlement expenses paid to CBBC buyers, CBBC issuers may depress the closing price of the underlying equity on the expiration day of bull 13 In Hong Kong, the settlement price for an index future/option contract is the average based on five-minute quotation and the closing index on the expiration day, whereas that for derivative warrants is the average closing price computed over a five-day period immediately before the expiration day. 14 A typical bull contract is issued with an expiration day, allowing investors to hold a bullish position on the underlying stock. Further, it pre-specifies a call price and a strike price. The call price sets a threshold to determine whether the contract will expire prematurely, while the strike price is used to determine the residual value of the contract on the expiration day. Imagine that a bull contract is issued with an expiration period of one year and the stock price of the underlying asset is $100 on the issuing date. The contract also specifies a call price at $90 and a strike price at $80. In the next day, if the stock price drops to $90 at 11 a.m. during the morning session, the bull contract will expire immediately. The settlement price will be determined by the minimum of all transacted prices of the underlying equity from 11 a.m. until the market close (inclusive of the closing price). If the minimum price of the underlying equity during that period is $82, the residual value of the bull contract will be $2 [= settlement price strike price = $82 $80 = $2]. 17

18 CBBCs while inflating it on the expiration day of bear CBBCs. 15 Therefore, we expect that the probability of a day-low (day-high) event at the close should be particularly high on the expiration day of bull (bear) CBBCs when the standard closing auction is used than when this mechanism is not used. We use a probit regression model with firm fixed-effects to examine the likelihood of day-low or day-high at the close on the expiration day of bull or bear CBBCs under the standard closing auction mechanism. The dependent variable is a dummy variable to capture whether the closing price is the highest or the lowest price of the day: day-low is a dummy variable equal to one if the closing price is the lowest transacted price of the day for the stock, and day-high is defined analogously. As the incentive to inflate or depress the closing price vary with the type of expiring CBBC contracts, we create two binary variables to capture the type of expired CBBC contracts. The variable ExpireBull is a binary variable equal to one for the underlying stock on the expiration day of a bull CBBC contract; the variable ExpireBear is defined analogously. Panel A of Table 4 presents marginal effects of observing a day-low event at the close on the expiration day of bull CBBCs while those of observing a day-high event at the close on the expiration day of bear CBBCs are reported in panel B. Columns (1) and (2) report the difference-in-differences estimates of the likelihood of observing a day-low or a day-high event to the introduction of closing auctions during the crisis period and the non-crisis period, respectively. The triple differences estimates of this relation are reported in column (3). Again, 15 In principle, both issuers and buyers of CBBCs have incentive to manipulate the closing price of the underlying equity on the expiration day. However, the incentive to manipulate should be stronger for CBBC issuers than for CBBC buyers because the issuer market is very concentrated. In terms of the dollar amount, the top three CBBC writers issued an aggregate of 75.5% of all CBBCs in Hong Kong from June 12, 2006 to May 31, In contrast, many CBBC buyers were small retail investors. Consistent with this claim, Li et al. (2018) find that gross profits of CBBC issuers are substantial. 18

19 our results imply that the standard closing auction mechanism is prone to manipulation on days when CBBCs expire. Specifically, the closing price is meaningfully more likely to be the lowest (highest) transacted price of the day on the expiration day of a bull (bear) CBBC contract under the standard auction mechanism than under the median pricing procedure. However, the estimated effect is stronger on days when bull contracts expire than on days when bear contracts expire. Numerically, the probability of observing a day-low incident on the expiration day of a bull contract is 5 percentage point more likely under the standard closing auction mechanism than under the median pricing procedure [column (1), panel A], compared with only 3.1 percentage point more likely on the expiration day of a bear contract. This result is likely due to the significant decline in stock prices during the period when the plain-vanilla auction mechanism was implemented. Therefore, many bull CBBC contracts were knocked out, thereby providing incentives for manipulators to depress the closing price on the expiration day. [Insert Table 4 here] As is expected, we find no such relations under the revamped closing auction mechanism. This result shows that the revamped closing auction is effective to deter manipulation on days when CBBCs expire. Similarly, the triple differences estimates also point to the same conclusion that day-low at the close incident is significantly more likely to occur on days when bull CBBC contracts expire under the standard closing auction mechanism. However, we find no such relation about the day-high incident on days when bear CBBC contracts expire based on the triple differences estimates in column (3), panel B. It is possible that expiration of CBBC contracts may reflect the underlying market condition which is unrelated to manipulation attempts. Thus, we use the overnight price reversal to examine this possibility. Based on our discussion earlier, we predict that overnight price 19

20 reversal should be stronger on the CBBC expiration day under the standard closing auction mechanism than under other alternative closing procedures. To measure the intensity of closing price manipulation on the CBBC expiration day, we use the number of expired CBBC contracts on the day (ncbbc) interacted with the stock return measured over the closing auction session (R10m). This stock return variable is justifiable because extant studies show that a large change in price before the close is a key attribute of closing price manipulation. For example, Cushing and Madhavan (2000) find that the last five minutes of trading account for a large fraction of daily return variability and others find that an unexpectedly large price change before the close is a key indicator in prosecuted cases of closing price manipulation (Comerton-Forde and Putninš, 2011&2013; Comerton-Forde and Rydge, 2006). As closing auctions were not used in the non- CAS period, we measure R10m by the stock return in the final ten-minute interval before the market close (R10m), i.e., 3:50-4:00 p.m. We use the same baseline model in equation (2) to estimate the standard closing auction effect on overnight price reversal on days when CBBCs expire. Table 5 presents firm fixedeffects regression results for the effect of adopting the standard closing auction on overnight price reversal on days when CBBCs expire. Our results in Table 5 corroborate evidence in Table 4 in that closing price manipulation is more likely to occur on the CBBC expiration day under the standard closing auction mechanism than under other closing procedures. In column (1), the coefficient on ncbbc R10m CAS is and statistically significant from zero. This indicates that if the underlying stock price drops by one percentage point in the final ten-minute interval before the close on the expiration of a CBBC contract, the stock price will bounce back by an additional 0.07 percentage point on the following open under the plain-vanilla auction mechanism than under the median pricing procedure. Consistent with our previous findings, we 20

21 find no such overnight price reversal under the revamped closing auction mechanism. The triple differences estimate also points to the same conclusion that significantly greater overnight price reversal occurs on the expiration day of CBBC contracts under the standard closing auction mechanism. [Insert Table 5 here] 6. Closing Price Informativeness Lastly, we examine the relation of a standard closing auction and closing price informativeness. Price changes are due to two major factors. The first one is the arrival of new information which induces a change in the fundamental value of the asset. The second one is trading frictions which induce noises to the observed stock price (Stoll, 2000). Closing price manipulation can be viewed as a source of frictions as it does not involve changes to the fundamental value of the asset. Therefore, closing price manipulation increases the volatility of close-to-close return. To measure the incremental volatility due to manipulation, we use the volatility of open-to-open return as a control to exclude price changes due to the arrival of new information. This is because any new information reflected in the close-to-close return should be equally reflected in the open-to-open return. We are confident that the open-to-open return is a fair benchmark for two reasons. First, the opening procedure at the HKEx remains unchanged throughout the sample period. Second, the opening price is less prone to manipulation because the opening price is not used to price any derivative products in the HKEx. We follow the literature to compute these returns and a variance ratio between the opento-open return and close-to-close return (French and Roll, 1986; Amihud and Mendelson, 1987). Specifically, the open-to-open return is calculated as Ro = (Po,t+1 Po,t)/Po,t, where Po,t is the opening price on the day and Po,t+1 is the opening price of the following day for the same stock. 21

22 The close-to-close return (Rc) is measured analogously. The variance ratio is computed as Var(Ro)/Var(Rc), where Var(Ro)is the variance of the open-to-open return for a stock during the period, and Var(Rc)is computed analogously. Next, we identify days which are prone to closing price manipulation. Based on our evidence in the previous section, our suspected manipulation days include days when CBBCs expire and days when sniping occurs. In particular, we have three measures of closing price manipulation as follows: (i) CBBC expiration days; (ii) days when sniping in price occurs; (ii) days when sniping in trade volume occurs. Therefore, for each measure of manipulation we divide our sample in two subsamples: manipulation sample and non-manipulation sample to conduct empirical tests. As manipulation introduces noises to the closing price and increases the volatility of close-to-close return, the variance ratio should be small in a period when closing price manipulation is rampant. Thus, we hypothesize that if the introduction of a plain-vanilla auction mechanism enhances closing price manipulation on CBBC expiration days and sniping days, the variance ratio on such days should be significantly smaller than that on non-manipulation days under the period when the standard closing auction mechanism is used than under the period when this mechanism is not used. Table 5 reports the mean variance ratio on manipulation days in columns (1)-(3), non-manipulation days in columns (4)-(6), the difference-in-differences estimates in column (7). We also present the mean variance ratio of CAS, i.e., when a standard closing call auction is used in columns (1) and (4); non-cas, i.e., when the median pricing procedure is used in columns (2) and (5); and their differences by manipulation days in columns (3) and (6). In each panel, the first row reports the estimates during the crisis period and those for the non-crisis period are reported in the second row. The last row of each panel presents the 22

23 changes in mean variance ratio between the crisis period and the non-crisis period. In each panel, the triple differences estimate is reported in the last cell of the last row. [Insert Table 6 here] Our results indicate that trading frictions are greater at the opening than at the closing. This result is robust and regardless of the closing procedure. All the mean variance ratios are greater than one and comparable to those in the literature (e.g., Amihud and Mendelson, 1987; Cheung et al., 1994). It is clear that there is a noticeable difference in closing price informativeness between manipulation days and non-manipulation days, and between the standard call auction mechanism and the revamped call auction mechanism. In particular, during the crisis period the variance of the close-to-close return is meaningfully smaller on nonmanipulation days in the CAS period than in the non-cas period. This indicates that investors who trade at the close are exposed to a smaller volatility in the period when the standard closing auction mechanism is used than in the period when the median pricing procedure is used. However, the results in the first rows of each panel also indicate that the accuracy of the closing price on manipulation days is similar between the CAS period and the non-cas period during the crisis period. Our findings are robust to all manipulation measures. The difference-indifferences estimates in column (7) of the first row of each panel indicate that the accuracy of the closing price is similar between the standard closing auction mechanism and the median pricing procedure. Unlike the results from the standard closing auction, we find that the revamped closing auction mechanism is robust to closing price manipulation. The results in the second rows of each panel indicate that during the non-crisis period the variance ratio is meaningfully greater on manipulation days during the revampcas period than during the non-cas period. This implies 23

24 that investors who trade at the close on suspected manipulation days are exposed to a smaller volatility under the revamped closing auction mechanism than under the median pricing procedure. However, on non-manipulation days, the accuracy of the closing price appears to be fairly similar between the revampcas period and the non-cas period. The results in column (6) of the second row indicate that the variance ratio is slightly smaller in the revampcas period than in the non-cas period. However, two of the three estimates are statistically indistinguishable from zero. The difference-in-differences estimates in column (7) of the second rows of each panel indicate that the closing price is more accurately priced under the revamped closing auction mechanism than under the median pricing procedure. The only exception is that the difference-in-differences estimate for the sniping in price is not statistically significant at the conventional level. This result is not surprising due to the fact that sniping in price rarely occurs in the revampcas period. Last, the triple differences estimates indicate that closing prices are meaningfully less accurate when they are determined by a standard closing auction mechanism than other closing mechanisms. On the flip side, our results suggest that closing prices are more informative under a closing call auction mechanism with manipulation-deterrence features than other closing procedures. Our results in Table 6 are consistent with our findings in the previous sections in that a standard call auction is vulnerable to manipulation, particularly on days when the motive of manipulation is strong. Additionally, on manipulation days closing prices formed under the standard call auction mechanism are also less informative than those formed based on other closing mechanisms. 7. Conclusions 24