A Rare Move: The Effect of Switching from a Closing Call. Auction to a Continuous Trading

Transcription

1 A Rare Move: The Effect of Switching from a Closing Call Auction to a Continuous Traing Ya-Kai Chang Department of Finance College of Business Chung Yuan Christian University Robin K. Chou Department of Finance an Risk an Insurance Research Center College of Commerce National Chengchi University J. Jimmy Yang College of Business Oregon State University Aress corresponence to Ya-Kai Chang, Department of Finance, College of Business, Chung Yuan Christian University, 200 Chung Pei R., Jhongli, Taoyuan, Taiwan. Tel.: , ykchang@cycu.eu.tw. Robin K. Chou, Department of Finance, College of Commerce, National Chengchi University, 64 Zhinan Roa Sec. 2, Taipei, Taiwan. Tel.: , rchou@nccu.eu.tw; J. Jimmy Yang, College of Business, Oregon State University, Austin Hall 426, 2751 SW Jefferson Way, Corvallis, OR 97331, USA. Tel.: (541) , jimmy.yang@bus.oregonstate.eu. 1

2 A Rare Move: The Effect of Switching from a Closing Call Auction to a Continuous Traing Abstract This stuy investigates the effects of switching to a closing continuous traing (CCT) on market quality on the Taiwan futures market. We provie insight on the influence of the traing behavior of ifferent types of traers. We fin that investors become more patient in the time perio immeiately preceing (i.e., 5-minute before) the CCT phase (last 5-minute of traing), leaing to a reuction in the bi-ask sprea in that perio. Our empirical results also inicate an increase in the bi-ask sprea an volatility uring the CCT phase ue to iminishing investor patience. More importantly, after the introuction of CCT, the market liquiity an volatility as well as the closing price efficiency are significantly affecte by the traing activity of foreign institutional investors. JEL classification: G14, G15, G18 Keywors: Closing price; Closing auction mechanism; Market quality; Price iscovery; Traing activity 2

3 1. Introuction Exchanges aroun the worl continuously seek a better closing traing mechanism, but there are isagreements an variations in using either a closing call auction or continuous traing at the close. Thus, etermining the preferre closing metho is an important task. However, existing theoretical literature has yet to provie a clear answer on this issue. Empirical stuies on the closing mechanism are also lacking ue to the infrequent occurrence of closing mechanism changes. With a rare move to switch from a closing call to a closing continuous traing (CCT) mechanism on the Taiwan Futures Exchange (TAIFEX), we are able to provie the empirical evience on the changes in market quality before an after the switch, which shes light on the relative performance of closing mechanisms. In a call market, the buy an sell orers are cumulate over a specifie time interval for simultaneous execution in a batche trae at a single price. In contrast, in a continuous market a trae occurs at any time when a bi an offer match or cross each other. A number of theoretical moels incluing Garbae an Silber (1979), Menelson (1982, 1987), Mahavan (1992), Glosten an Milgrom (1985), Ho et al. (1985), an Buish et al. (2014, 2015) conclue that call markets can mitigate the problems of information asymmetry, but lose the avantages of the immeiacy of 3

4 traing an the lower cost of gathering market information provie by continuous markets. Empirically, the effects of changes in closing mechanisms are mixe. Some stuies fin that the introuction of a closing call proceure improves the price iscovery process an the market quality (Pagano an Schwartz, 2003; Comerton-Fore et al., 2007; Chang et al., 2008; Chelley-Steeley, 2008; Kanel et al., 2012; Ibikunle, 2015). 1 However, Sivanithy an Ng (2000), Comerton-Fore an Ryge (2006), an Chelley-Steeley (2009) o not fin evience in support of these benefits. 2 Similarly, there exist mixe views on the effects of introucing a CCT. Amihu et al. (1997), Lauterbach an Ungar (1997), Muscarella an Piwowar (2001) an Kalay et al. (2002) provie evience showing that a CCT is superior to a closing call auction traing. However, Lauterbach (2001) suggests that the introuction of a CCT may not improve market quality for illiqui stocks. 3 The effects of the changes of closing mechanism are far from conclusive. 1 Consistent with these finings, Schwartz (2001) points out that the closing call auction enhances price efficiency, lower price volatility an narrow bi-ask spreas. In aition, Kanel et al. (2012) fin a significant reuction in sprea an volatility, an an increase in aggressiveness of liquiity suppliers an price efficiency 10 minutes before the closing call auction. Furthermore, Chelley-Steeley (2008) an Ibikunle (2015) show that the least active securities at the Lonon Stock Exchange experience the greatest improvements to market quality following a closing call introuction. 2 Sivanithy an Ng (2000) an Comerton-Fore an Ryge (2006) emonstrate that the implementation of a closing call auction oes not completely remove market manipulation. Comerton-Fore an Ryge (2006) also inicate that matching criteria use to set auction prices can greatly influence the effectiveness of the closing calls. Finally, Chelley-Steeley (2009) shows that a closing call auction may not enhance market quality of the Lonon Stock Exchange for high liquiity securities. 3 Amihu et al. (1997), Lauterbach an Ungar (1997), an Kalay et al. (2002) inicate that CCT is superior to closing call auction in the Tel Aviv Stock Exchange. In aition, Muscarella an Piwowar 4

5 An efficient closing mechanism is important to market participants an regulators, because it assures the closing price to reflect funamental values of the asset. Many market participants, especially mutual funs, have strong incentives to trae at closing prices. Since the settlement with the eparting (or new) investors is base on the closing price, mutual funs that hanle investors reemption an injection of funs prefer to execute their traes as close to the closing price as possible to avoi the settlement risk. This potential eman for traing at the closing price results in liquiity shocks at the en of the traing ay an inuce large futures price movements, which cause closing prices to move away from funamental values (Cushing an Mahavan, 2000; Hagströmer an Norén, 2014). A closing mechanism that can promote price iscovery woul thus be esirable. An efficient price at the market close is critical because closing prices are wiely reference. Comerton-Fore an Ryge (2006), Kanel et al. (2012) an Hagströmer an Norén (2014) inicate that closing prices are use to evaluate portfolio returns an to measure mutual funs net asset value (NAV). Closing prices are also use to etermine the value of other erivatives contracts an to benchmark performance of portfolio managers. The closing price on futures markets is also important because it (2001) fin that market quality increases at the Paris Bourse for frequently trae stocks that are transferre from their call market to their continuous market. Despite the aforementione benefits, Lauterbach (2001) offers a ifferent insight an shows that not all stocks benefit from a continuous trae. For small an infrequently trae stocks, a switch from a call auction to a continuous traing may not be optimal. 5

6 is use as a basis for measuring whether investors margin requirements are met on futures markets. The mark-to-market system is one by computing investors aily profits or losses base on the aily closing price. If the current market value causes the margin account to fall below its require level, the investor will face a margin call. Although the closing traing mechanisms are important for futures markets, most stuies focus on stock markets. Existing literature examines primarily a switch from the CCT to the closing call auction.the empirical evience of transferring from a call auction to a CCT is scarce ue to the rarity of such events. To our knowlege, the Tel Aviv Stock Exchange (Amihu et al., 1997; Lauterbach an Ungar, 1997; Lauterbach, 2001; Kalay et al., 2002) an the Paris Bourse (Muscarella an Piwowar, 2001) are the only two markets that experience such a move. There has been no stuies on the futures markets that move from a closing call auction to a CCT. On October 8, 2007, the TAIFEX switche from a closing call auction to a CCT uring the last 5 minutes of traing, creating a goo opportunity to stuy the effects of the closing mechanisms of futures markets. The TAIFEX is an orer-riven electronic futures market within which there are no esignate market makers. Traing on the TAIFEX is carrie out from 8:45 to 13:45. 4 Accoring to the TAIFEX, the CCT mechanism is attache to the introuction of futures calenar sprea mechanism, 4 The TAIFEX closes 15 minutes after the spot market oes. 6

7 which was establishe to lower the position risk an price risk of traers executing calenar sprea traes. As a result, this event provies a unique natural experiment to examine the effects of switching to a CCT with no prior belief from the regulators in terms of the relative performance of both closing mechanisms. Market participants are likely to ajust their traing activities to aapt to the new closing environment. Unerstaning the ajustment of traing strategies is important, because it is likely to have implications for market quality. Traers may have precursor reactions to the closing mechanism changes, so the examination of only the last traing perio may provie an incomplete picture of the market quality changes. The changes in closing mechanisms coul affect the spreas (Pagano an Schwartz, 2003) an the volatility (Hillion an Souminen, 2004) in the preceing traing perio. We thus use a methoology similar to that of Kanel et al. (2012) to examine the time perio preceing the CCT phase to elineate a more complete picture of market quality changes associate with a switch of the closing mechanism. Another interesting issue is whether ifferent types of traers will have ifferent behavior ajustments in response to the closing mechanism changes. Previous stuies only show the average impact of the closing mechanism changes on the market. With the unique TAIFEX account-level ata, incluing foreign institutions, omestic institutions, futures proprietary firms, an iniviual traers, we provie a etaile 7

8 analysis on how ifferent types of traers are affecte by the change. Our results have important policy implications for regulators, who are especially concerne about the welfare gains an losses of ifferent types of market participants as a result of market structure changes. The results show that the introuction of the CCT has little effect on the intraay volume istribution throughout most of the ay, except for the last two 5-minute intervals before market close (i.e., 13:35-13:40 an 13:40-13:45). Following the aoption of this new traing mechanism, a significant portion of aily traing volume migrates from the perio 13:35-13:40 to the perio13:40-13:45 (the CCT phase). In aition, we fin a reuction in the bi-ask sprea an the volatility, yet not statistically significant for the latter, in the time interval preceing the CCT. Our empirical results also show that a significant increase in the bi-ask sprea an the volatility uring the CCT phase, as well as a reuction in the pricing efficiency of the closing price. Overall, our empirical results are consistent with Foucault et al. (2005) an Roşu (2009). Accoring to Foucault et al. (2005) an Roşu (2009), we suggest that a CCT introuction increases the patience of traers in the time perio preceing the CCT, because it offers investors an aitional opportunity to continue their traes for unwining the inventory on the same ay between 13:40 an 13:45 (the CCT phase). 8

9 Such an increase in the egree of patience encourages traers to submit more limit orers an fewer market orers, which causes a reuction in sprea an volatility. Furthermore, because traers rush to execute their traes near the en of the ay, they ten to submit more market orers, relative to limit orers, which in turn leas to a higher sprea an larger volatility uring the CCT phase. Moreover, we fin that the closing price becomes less informative after switching to a continuous closing mechanism because of the larger market noise over the last five minutes. Finally, for ifferent traer types, our empirical results show that the traing volume of foreign institutions shifts significantly from the interval 13:35-13:40 to the interval 13:40-13:45. These finings inicate that foreign institutions prefer to trae uner a continuous rather than a call auction environment uring the market close. Furthermore, we fin that the shift of foreign institutions traing to the CCT contributes negatively to the bi-ask sprea in the time perio preceing the CCT. However, changes in foreign institutions traing behavior contribute positively to the volatility uring the last five minutes an the pricing error of closing price. These results are not completely consistent with the preication of Amati an Pfleierer (1988). Accoring to Amati an Pfleierer (1988), with an increase in the traing volume of foreign institutions uring the CCT, the bi-ask sprea shoul 9

10 be larger in the perio preceing the CCT phase an the closing price shoul be less noisy. This paper contributes to the literature in several ways. First, we provie the first empirical evience on the effect of switching from a closing call auction to a CCT on the futures markets. Previous empirical evience on this topic is primarily base on stock markets. Secon, we consier the traing behavior of ifferent types of traers an their contribution to the market quality after the introuction of the CCT system. Thir, by investigating the traing interval preceing the CCT perio, we provie insights on the precursor reactions of traers in responing to a new closing traing mechanism. In aition, our empirical results provie important policy implications. The remainer of this article is organize as follows: Section 2 briefly reviews prior literature an proposes hypotheses. Section 3 escribes the ata an methoology. Section 4 presents the empirical results, an Section 5 conclues. 2. Literature review an hypothesis evelopment Given their traing strategy an objective, after a CCT introuction investors may ajust their traing behavior even before entering the CCT phase. Thus, besies the CCT phase (13:40-13:45), we also examine the time perio (5-minute) preceing the CCT phase. By oing so, we can provie a better picture on traer behavior 10

11 changes. 5 In what follows we provie the empirical preictions on the effects of the CCT introuction on market quality through the relate theoretical moels. We also summarize the preictions in the Appenix Preictions of Traing Volume Amati an Pfleierer (1988), base on Kyle (1985), evelop a theory in which uninforme an informe traers can choose the best time (i.e., the most liqui perio of the ay) to submit orers in orer to minimize their price impact per unit of aggregate volume, which inuces the emergence of enogenous spikes in traing volumes uring the ay. Extant literature notes that such spikes usually occur uring the CCT phase that gathers more liquiity an might be attribute to the better price control or ease of unwining inventory (e.g., Mclnish an Woo, 1992; Amihu et al., 1997; Lauterbach an Ungar, 1997; Muscarella an Piwowar, 2001; Schwartz, 2001; Steil, 2001; Kalay et al., 2002). Therefore, we preict that the traing volume woul shift from 13:35-13:40 to 13:40-13:45 after switching to a CCT, which leas to our first hypothesis. H1. The introuction of a CCT mechanism shifts the traing volume from the time perio immeiately preceing the CCT phase to the CCT phase. 5 In equity markets, a similar analysis is performe by Kanel et al. (2012). They stuy the effects of the introuction of a closing call auction by consiering the ajustment of investors traing behavior an focus on the preceing continuous perio rather than the closing call auction perio. 11

12 2.2. Preictions of Market Liquiity an Volatility Market liquiity an volatility can be affecte by the change in investors traing behavior. Accoring to Amati an Pfleierer (1991), there shoul be less price impact per unit of trae when investors move their traes to an environment which gathers more liquiity; that is, the price response to volume woul become less severe in the high-volume perio of the ay. They further point out that bi-ask sprea is lower, volatility is larger an price efficiency is better in the high-volume perio, compare with the low-volume perio, because market maker faces less information asymmetry. Therefore, after switching to a continuous closing mechanism, if the traing volume significantly moves from the time interval 13:35-13:40 to 13:40-13:45, then we preict an increase in the bi-ask sprea an a reuction in volatility between 13:35 an 13:40 ue to the reuce traing volume uring this perio. Moreover, we also preict that a ecrease in the bi-ask sprea an an increase in volatility between 13:40 an 13:45 because of the increase traing volume in this perio. However, an opposite preiction can be erive from Kaniel an Liu (2006). From the moel of Amati an Pfleierer (1988), informe traers can only submit market orers. Kaniel an Liu (2006) further allow informe traers to submit either 12

13 market orers or limit orers, an this ecision epens on the longevity of their own private information. They argue that as private information becomes longer live, the risk of unexecute orers that a traer bears from submitting a limit orer ecreases. In this case, informe traers may choose to submit limit orers instea of market orers. Strategically, traers with long-live private information woul prefer to use limit orers, while traers with short-live private information woul rather submit market orers. The CCT introuction provies an opportunity to exten the longevity of private information for those traers who want better price control. Informe traers woul thus prefer the CCT system, because their orers can be execute with a high egree of certainty at the market quotes (e.g., Lauterbach an Ungar, 1997; Schwartz, 2001; Steil, 2001). Knowing that they can trae continuously from 13:40 to 13:45, informe traers are expecte to become more patient an post fewer market orers an more limit orers between 13:35 an 13:40. Along the same lines, informe traers become less patient an post more market orers an less limit orers uring the last five minutes of the CCT phase (13:40-13:45), because it is their final traing opportunity before the market closes. Accoring to Kaniel an Liu (2006), a reuction (an increase) in the proportion of market orers that are submitte by informe traers implies a reuction (an increase) in the bi-ask sprea an volatility because the market maker 13

14 faces less (more) averse selection problem. Thus, we hypothesize that the CCT introuction woul lea to a ecrease in bi-ask sprea an volatility uring the 13:35-13:40 perio, because informe traers become more patient in this time interval; meanwhile, we preict that the CCT system woul lea to an increase in bi-ask sprea an volatility uring the last five minutes as informe traers become less patient in this perio. Another plausible argument makes the same preiction on market quality through the liquiity-base moels propose by Foucault et al. (2005) an Roşu (2009). They show that all traers in the limit orer book market share the same information with no information asymmetry, an their traing strategies are etermine by the relative patience of traers. They point out that patient traers who play the role of liquiity suppliers ten to submit limit orers, whereas impatient traers who play the role of liquiity emaners ten to submit market orers. After switching to a CCT system, traers gain an aitional opportunity to unwin their inventory on the same traing ay. Liquiity traers like a CCT mechanism because it provies them an ability to take a position an unwin it uring the same ay, so as to avoi the risk of carrying unwante inventory (Amihu et al., 1997; Steil, 2001). As a result, when liquiity traers know that they can continue their traes between 13:40 an 13:45, they become more patient an place fewer market orers an more limit 14

15 orers over the 13:35-13:40 time interval. However, they woul become less patient an place more market orers an less limit orers between 13:40 an 13:45, because they know this is their final opportunity for unwining their inventory before the market close. In aition, Foucault et al. (2005) an Roşu (2009) inicate that the market bi-ask sprea epens on the proportion of patient traers. That is, when the proportion of patient traers increases, competition between patient traers causes the bi-ask sprea to become narrower. On the other han, when the proportion of patient traers ecreases, patient traers o not nee to place more aggressive limit orers to achieve orer execution, leaing to a wier bi-ask sprea. Accoring to Foucault et al. (2005) an Roşu (2009), an increase (a reuction) in the proportion of patient traers implies a reuction (an increase) in the bi-ask sprea an volatility because of higher (lower) competition between patient traers. Thus, we expect that after the introuction of a CCT system, the bi-ask sprea an volatility woul reuce between 13:35 an 13:40, ue to the increase proportion of patient traers uring this perio. We also preict that the bi-ask sprea an volatility woul increase uring the CCT phase because of the increase proportion of impatient traers in this perio. These preictions lea to the following hypotheses: 15

16 H2.1. The aoption of a CCT mechanism increases spreas in the time perio preceing the CCT (Amati an Pfleierer, 1988). H2.2. The aoption of a CCT mechanism reuces spreas in the time perio preceing the CCT (Foucault et al., 2005; Kaniel an Liu, 2006; Roşu, 2009). H3.1. The aoption of a CCT mechanism reuces spreas uring the CCT phase (Amati an Pfleierer, 1988). H3.2. The aoption of a CCT mechanism increases spreas uring the CCT phase (Foucault et al., 2005; Kaniel an Liu, 2006; Roşu, 2009). H4. The aoption of a CCT mechanism lowers volatility in the time perio preceing the CCT (Amati an Pfleierer, 1988; Foucault et al., 2005; Kaniel an Liu, 2006; Roşu, 2009). H5. The aoption of a CCT mechanism increases volatility uring the CCT phase (Amati an Pfleierer, 1988; Foucault et al., 2005; Kaniel an Liu, 2006; Roşu, 2009) Preictions of Closing Price Efficiency Switching from a call auction to a continuous closing mechanism provies informe traers an aitional opportunity to consummate their traes with better control on the price. In this setting, a fraction of aily orer flow of informe traers woul migrate to the last 5 minutes (13:40-13:45). Accoring to Amati an Pfleierer (1988), prices become more informative uring that perio with an increase in the traing volume of informe traers. We thus expect the price iscovery process woul be improve after a CCT is aopte. 16

17 The same preiction can be mae from a ifferent perspective. When facing a final traing opportunity to take avantage of their private information at the en of the ay, informe traers are less patient an are likely to place more market orers uring the last 5 minutes. Accoring to Kaniel an Liu (2006), the closing price reveals more information if traers submit market orers instea of limit orers. Thus, we expect the CCT introuction woul enhance the closing price efficiency. However, an opposite preiction can be mae from the argument of Foucault et al. (2005) an Roşu (2009). Specifically, traers who nee to unloa their positions to avoi an exposure to the overnight risk become more impatient an rush to execute their traes as the en of the ay approaches. This leas to higher price changes an higher volatility, which in turn result in higher market noise uring the last 5 minutes, which results in a less efficient closing price after the introuction of a CCT mechanism. Hence, how the price efficiency changes is an empirical issue an the above iscussions lea to the following hypotheses: H6.1. The aoption of a CCT mechanism leas to improve closing price efficiency (Amati an Pfleierer, 1988; Kaniel an Liu, 2006). H6.2. The aoption of a CCT mechanism leas to eteriorate closing price efficiency (Foucault et al., 2005; Roşu, 2009). 17

18 3. Data an Methoology 3.1. Data an Samples The sample of our stuy is the Taiwan Stock Exchange Capitalization Weighte Stock Inex futures (hereafter, TAIEX futures) contracts trae on the TAIFEX. The traing unit of the TAIEX futures is the inex value of the TAIEX 200 New Taiwan Dollars (NT$).The nearest contracts are use in orer to focus our analyses on the most liqui futures contracts. 6 The ataset we employe is obtaine from the TAIFEX intraay futures atabase an consists of comprehensive information on all transactions at the account level for the TAIEX futures between January 1, 2007 an August 1, The transaction ata contain the etaile history of orer flows, orer book, an the ientity of the traers. For each orer, the ata set reports the time stamp of arrival of the orer, its sign (buy or sell), the quantity emane or offere, an particularly for the purpose of our stuy the ientification of traers. The traer coe enables us to categorize four types of traers: foreign institutions, omestic institutions, futures proprietary firms an iniviual traers. 6 When the traing volume of the first eferre contract is greater than the traing volume of the nearest contract uring the maturity month, we roll over to the first eferre contract an these rollovers often occur towar the secon half of the time to maturity. 18

19 Because of the availability of etaile transaction ata on each type of traer, we can stuy the behaviors of ifferent traer types when examining the effects of the CCT aoption. In the empirical analysis, we ivie the stuy perios into two sub-perios surrouning the aoption of the CCT on Oct. 8, 2007, i.e., from January 1, 2007 through October 1, 2007 (Pre-perio) an from November 1, 2007 through August 1, 2008 (Post-perio). The escriptive statistics of the aily traing volume an the ollar traing volume for the four types of traers are reporte in Panels A an B of Table 1. Panel A shows that the largest proportion of transactions across the two sample perios is attributable to iniviual traers. For example, total traing volume of iniviual traers accounts for 68.26% an 68.43% of all traes for the Pre-perio an the Post-perio, respectively. Futures proprietary firms rank secon (Pre-perio: 18.61% an Post-perio: 18.76%) an foreign institutions rank thir (Pre-perio: 10.30% an Post-perio: 10.35%) in terms of the traing volume. Domestic institutions account for only 2.83% of the traing volume in the Pre-perio an 2.46% in the Post-perio. Panel B presents similar results base on the traing value. [Insert Table 1 here] The overall istribution of traing volume among the four types of traers remains at the similar level from the Pre-perio to the Post-perio. The introuction of 19

20 the CCT oes not seem to affect the egree of participation by any type of traers in this market. Although the traing volume ecreases from the Pre-perio to the Post-perio, the olloar traing volume actually increases. There is thus little concern of reuction in traing activity after the introuction of the CCT Variables We consier various relevant measures, incluing quote bi-ask sprea, realize volatility, an the pricing error to examine the market quality of the TAIFEX after switching to a new closing mechanism. Each of these measures is calculate in 5-minute intervals. We first estimate quote bi-ask spreas using intraay bi-ask quotes to proxy for market liquiity. Quote bi-ask sprea uring the interval t is efine as: Askt Bit Quote Bi Ask Spreat, (1) M t where Askt an Bi t are the lowest selling an highest buying prices at the en of the 5-minute interval, respectively. M t enotes the mipoint of Ask t an Bi t, which is consiere as a proxy for the true value of the contract, which minimize the influence of the bi-ask bounces. 20

21 We next use the realize volatility propose by Anersen et al. (2001) to measure market volatility. Our intraay realize volatility for the interval t is calculate as follows: n 1 2 t, n Realize Volatility 100* R N, (2) t N where R n is the 1-minute intraay return, an N is an inicator for the 1-minute intraay returns uring that 5-minute interval t. 7 Rt, n Mt, n Mt, n 1 ln( ) ln( ), where M tn, is the bi-ask mipoint at the en of the 1-minute interval. Finally, we use the cost-of-carry moel to estimate the pricing errors cause by traing noises, with the theoretical futures price * F at time t formulate as: F S e * ( r iv)( D ) t t where S t is the spot inex at time t; r is the risk-free interest rate; an iv is the annualize ivien yiel for holing the unerlying inex over the perio from the current ay to the futures expiration ay D ; an ( D ) is annualize by iviing the number of ays by 365. The pricing error, PE t, is efine as the absolute eviation in the theoretical price form its observe futures price ivie by the observe futures price shown in Equation (3): 8 7 We sample the ata at the 1-minute frequency. Therefore, the realize volatility for each 5-minute interval is the sum of the five 1-minute square returns. 8 As a robustness check, in alternative tests, we also take into account the transaction cost c in the 21

22 PE t F F F * t t t, (3) where F t is the observe futures price. The theoretical price was viewe as the intrinsic value of the futures contract. The pricing error etermines how closely the observe price tracks the intrinsic value. A lower pricing error implies better price efficiency Empirical Moels We examine the impact of switching to the CCT on market liquiity, volatility an closing price efficiency, conitioning on the traing volume of ifferent traer types transferre to the CCT phase. The three empirical moels are specifie as follows: Sprea D D * Volume (4) p 12 p i, 1 Volatility D D * Volume, (5) p 22 p i, 2, an calculation of the pricing error, PE t * t F F t F t c, an obtaine similar results. 22

23 CPE D D * Volume (6) p 32 p i, 3 When analyzing en-of-ay market microstructure effects relate to the CCT introuction, it is essential to control for the ifferences between the two perios before an after aoption that are not likely to be event-relate. In this stuy, we control for the perio-specific market conitions across the two sample perios through normalizing all variables of interest by their average values calculate over the one hour interval from 11:00 to 12:00 in the same ay. 9 This normalization approach is consistent with Kanel et al. (2012), an we follow the same proceure in all the subsequent regression analyses. In these three equations, the normalize sprea on ay is efine as Sprea BAS BAS 13:35 13:40 13:40 13:45 (or 11:00 12:00 BAS BAS 11:00 12:00 BAS ( BAS 13:40 13:45 ). 13:35 13:40 ) is the Quote Bi-Ask Sprea uring the 13:35-13:40 (13:40-13:45) traing interval, an BAS11:00 12:00 is the average Quote Bi-Ask Sprea in the time interval 11:00-12: RV13:35 13:40 RV13:40 13:45 Volatility (or RV RV 11:00 12:00 11:00 12:00 ) enotes the normalize volatility on ay, where RV13:35 13:40 ( RV 13:40 13:45 ) is the realize volatility uring 9 The state of the limit orer book between 11:00 am-12:00 pm is assume to be inepenent of the market closing traing mechanism. 10 The average Quote Bi-Ask Sprea uring the time interval 11:00-12:00 is the average of the twelve 5-minute Quote bi-ask Spreas. We compute the other normalize variables in the same way. 23

24 the 13:35-13:40 (13:40-13:45) interval, an RV11:00 12:00 is the average realize volatility uring the time interval 11:00-12: Similarly, CPE PE is PE 11:00 12:00 the normalize pricing error for futures closing price on ay. PE is the pricing error of the closing price, an PE11:00 12:00 is the average pricing error uring the time interval 11:00-12: Our main explanatory variable is Volume i, TV TV i,13:40 13:45 i,13:35 13:40 TV i,11:00 12:00, which is the normalize volume migrate from the 13:35-13:40 interval to the 13:40-13:45 interval for type i traers on ay, where i = foreign, omestic, proprietary an iniviual, enoting foreign institutions, omestic institutions, futures proprietary firms an iniviual traers, respectively. TVi,13:40 13:45 refers to type i traers traing volume accumulate over the interval 13:40-13:45, an TVi,11:00 12:00 is the average type i traers volume over the 11:00-12:00 interval. D p is a ummy variable that equals one uring the Post-perio an zero otherwise. The interaction term between Volume i, an D p is inclue in our regression to investigate the relationship between the change in the traing volume of ifferent traer types an the market quality measures following the introuction of the CCT system. 11 As a robustness check, we also employ the high-low volatility estimator propose by Parkinson (1980) for measuring the market volatility an our conclusions remain unchange. 12 We also use the volume weighte average price within the last one minute as the closing prices for the Post-perio an fin similar results. 24

25 Finally, to aress the potential omitte variable problem, we moify Equations (4), (5) an (6) by incluing Volume foreign, Volume omestic, Volume proprietary, an Volume iniviual as regressors in the same regression. We specify our alternative regression moels as follows. Sprea D D * Volume D * Volume p 42 p foreign, 43 p omestic, D * Volume D * Volume 44 p proprietary, 45 p iniviual, 4 (7) Volatility D D * Volume D * Volume p 52 p foreign, 53 p omestic, D * Volume D * Volume 54 p proprietary, 55 p iniviual, 5 (8) an CPE D D * Volume D * Volume p 62 p foreign, 63 p omestic, D * Volume D * Volume 64 p proprietary, 65 p iniviual, 6 (9) For all regression moels, we employ the Newey an West s (1987) robust heteroskeasticity, autocorrelation-consistent stanar errors. 4. Empirical Results 4.1. Univariate Tests 25

26 We first present the allocation of the traing volume over the course of the whole ay for the sample perios before an after the switch to the CCT in Table 2. The result inicates that the volume is significantly reallocate uring the last two time intervals, relative to other traing intervals after switching to the CCT. The ratio of volume allocation uring the time interval 13:30 to 13:40 ecreases from 2.50% in the Pre-perio to 1.88% in the Post-perio, while uring the time internal 13:40 to 13:45, it increases from 1.96% in the Pre-perio to 2.41% in the Post-perio. These finings suggest that investors may elay their traing activity to the last 5-minute of the CCT phase, consistent with the empirical finings that investors prefer the continuous traing system as the closing mechanism (e.g., Amihu et al., 1997; Lauterbach an Ungar, 1997; Muscarella an Piwowar, 2001; Kalay et al., 2002). It shoul be note that the intraay istributions of traing volume, other than those of the last 10 minutes, are similar between the two sub-perios. However, after the introuction of the CCT, the traing volume reallocates in the last two 5-minute intervals. As a result, our empirical analyses focus on the market quality uring these two 5-minute intervals. [Insert Table 2 here] Table 3 shows the averages of the normalize traing volume for ifferent traer types, bi-ask sprea, volatility, an closing price efficiency for the two sub-perios, 26

27 as well as the ifferences in the averages of variables. 13 All variables are normalize as being ivie by their respective averages from 11:00 am an 12:00 pm. To examine the effects of ifferent traer types, we segregate the total volume into those by four traer types. Table 3 shows that the mean normalize volumes for foreign institutions an futures proprietary firms uring the 13:35-13:40 interval in the Post-perio are an 1.328, respectively, which are significantly lower than those in the Pre-perio. However, the mean normalize volumes for foreign institutions an futures proprietary firms uring the 13:40-13:45 interval in the Post-perio are significantly higher than those of the Pre-perio. Results from Table 3 inicate that after switching to a continuous closing traing mechanism, a significant proportion of traing volume, especially for foreign institutions an futures proprietary firms, shifts from the 13:35-13:40 interval to the 13:40-13:45 interval, consistent with H1, which hypothesizes that investors migrate their traing from the time interval immeiately preceing the CCT phase to that uring the CCT phase. Interestingly, for iniviual traers, the traing volume uring the CCT phase is similar between the Pre-perio an Post-perio. The change in the futures closing mechanism oes not seem to impact their traing behavior. 13 As a robustness check, we also use the Wilcoxon test to examine our intereste variables for the ifferences in meians between the two sample perios, an our results remain unchange. 27

28 For the overall market, traing volume significantly shifts from the time perio immeiately preceing the closing phase to that uring the CCT phase, possibly ue to an increase in investors patience as iscusse earlier. If investors prefer a continuous traing mechanism to close the market, they may choose to elay their traing activity to the last 5 minutes. Our finings support the argument that investors prefer to trae uner a continuous traing environment because informe traers may obtain a better price control (e.g., Lauterbach an Ungar, 1997; Schwartz, 2001; Steil, 2001) an liquiity traers are likely to experience lower risk of carrying unwante inventory (e.g., Amihu et al., 1997; Steil, 2001). Table 3 also presents the univariate test of liquiity, volatility an closing pricing errors. We fin that for the time interval 13:35-13:40, the mean bi-ask sprea in the Post-perio is significantly lower than that in the Pre-perio, inicating a higher liquiity in the perio immeiately prior to the CCT phase after the introuction of the CCT. However, for the time interval 13:40-13:45, the mean normalize bi-ask sprea in the Post-perio is significantly higher than that in the Pre-perio, inicating a lower liquiity uring the last five minutes after the introuction of the CCT. In aition, the results show that the mean volatility of the Post-perio is lower than that of Pre-perio uring the 13:35-13:40 interval, whereas the volatility in the Post-perio is higher than that of the Pre-perio uring the 13:40-13:45 interval. 28

29 These show that the CCT introuction lea to a significant reuction in the volatility uring the interval immeiately preceing the CCT phase, an result in a significant increase in the volatility uring the CCT phase. Furthermore, the empirical results also inicate that the closing price error of the Post-perio are higher than that of Pre-perio over the time interval 13:40-13:45, which may imply an increase in the closing pricing errors after the CCT introuction. From all above finings, we clearly see the impact of the introuction of a CCT on traing volume, market liquiity, volatility, an price efficiency. Our regression analyses in the following sections will provie more insights on the effect of the introuction of the CCT on market quality. [Insert Table 3 here] 4.2. Market Liquiity The empirical estimates of the quote bi-ask sprea Equations (4) an (7) are reporte in Table 4. Panel A reports results for the 5-minute perio preceing the CCT phase an Panel B reports results for the CCT phase. Moels (1) to (4) in both panels report the results by traer types. The coefficients of all perio ummy variables ( D ) p are significantly negative as shown from Moels (1) to (4) in Panel A, while the coefficients are all significantly positive from Moels (1) to (4) in Panel B. Thses show that after the introuction of the CCT, the market liquiity in the time perio 29

30 immeiately preceing the CCT phase increases, but it ecreases uring the CCT phase. These finings support both H2.2 an H3.2 erive from the preictions by Foucault et al. (2005), Kaniel an Liu (2006), an Roşu (2009). In aition, our empirical results show that the coefficient of D is significantly negative in Moel (1) of Panel A. This inicates p * Volume foreign, that after the implementation of the CCT, the more the foreign institution traing volume shifts to the CCT, the lower the quote bi-ask sprea uring the interval 13:35-13:40 will be. These finings are not consistent with Amati an Pfleierer (1988) as they suggest that the bi-ask sprea shoul be higher in the low-volume perio. Moel (5) of Panels A an B as the interaction term between the perio ummy variable, D p, an the change in the traing volume for each traer type as explanatory variables in the same regression. The empirical results show similar patterns. [Insert Table 4 here] 4.3. Market Volatility Table 5 presents the results for volatility. Panel A reports results for the interval immeiately preceing the CCT phase an Panel B reports results for the CCT phase. Moels (1) to (4) of Panel A show that the coefficients of D p are negative, while 30

31 Panel B shows that the coefficients of D p are positive. These show that the introuction of the CCT leas to a reuce market volatility (yet not significant) in the time interval immeiately prior to the CCT phase an a higher market volatility uring the CCT phase. As iscusse earlier, after switching to a continuous closing mechanism, investors become more patient an submit more limit orers an fewer market orers in the time interval preceing the CCT. This change in traing behavior likely lowers market volatility. On the contrary, investors become less patient an submit more market orers than limit orers uring the CCT phase, resulting in an increase in market volatility. These empirical results seem to confirm H4 an H5 as preicte by Amati an Pfleierer (1988), Foucault et al. (2005), Kaniel an Liu (2006) an Roşu (2009). Regaring the changes in the traing volume for ifferent traer types on market volatility, Moel (1) of Panel B show that the coefficient of D Volume, is p * foreign significant positive. This inicates that after the aoption of the CCT system, the increases in foreign institutions traing volume uring the CCT phase are positively relate to the market volatility. This result is consistent with Amati an Pfleierer (1988), who suggest that uring a high volume perio, volatility will increase. However, the changes in traing volume for other traer types are not foun to be affecting volatility. We re-estimate the regressions by controlling for the changes in 31

32 traing volume of other traer types in Moel (5) in Table 5 an obtain similar results for the effect of foreign institutions traing volume on volatility. [Insert Table 5 here] 4.4. Closing Price Efficiency Table 6 presents the effect of CCT on the pricing errors for the closing price. The coefficients of D p are positive, but not significant, in Moels (1) to (4), which shows that the CCT inuces a higher pricing errors for the closing price (i.e., lower closing price efficiency). This is likely because traers become less patient an rush to execute their traes at the en of the ay an this high liquiity eman woul inuce higher price changes an higher volatility, which generates more market noises. Thus, the introuction of the CCT mechanism seems to eteriorate the closing price efficiency. These finings are consistent with H6.2 suggeste by Foucault et al. (2005) an Roşu (2009). We further examine the effect of the traing volume shift to the CCT phase by ifferent traer types on the closing price efficiency after the introuction of the CCT. From Moels (1) an (2) in Table 6, we fin that the perio ummy variable, D p, interactions with the changes in the traing volumes of foreign institutions ( p * Volume foreign, D ) an omestic institutions ( D * Volume omestic, ) are p 32

33 significantly positively relate to the pricing error of closing price, CPE. That is, after the introuction of the CCT, the closing price efficiency eterioriates with the increase in the traing volume of foreign institutions an omestic institutions migrating to the CCT phase. These results are not consistent with the preication of Amati an Pfleierer (1988), who suggest that the prices become more informative in the high-volume perio. Finally, an increasing boy of literature argues that iniviual traers are following the market trens (i.e., the traing behavior of foreign institutions) (e.g., Shleifer an Summers, 1990; Nofsinger an Sias, 1999; Kaniel et al., 2008; Barber an Oean, 2008). 14 However, our empirical results show that the perio ummy variable, D p, interactions with the traing volumes of iniviual traers moving to the CCT phase ( D Volume, ) is not significant. p * iniviual One plausible explanation is that iniviual traers o not seem to follow foreign institutions, as we fin that the correlation coefficient of the traing volume shift to the last five minutes between foreign institutions an iniviual traers is only 0.38 in the Post-perio, whereas the correlation coefficient is 0.62 in the Pre-perio. The CCT mechanism is likely to attenuate the tenency of iniviual traers following foreign institutions. Iniviual traers may not want to trae at the en of the ay 14 Nofsinger an Sias (1999) propose that iniviual traers are tren-chasers because they are likely to be affecte by fa an fashion. Shleifer an Summers (1990) show that iniviual traers woul her if they follow the same signals from the avices of brokers an financial gurus. Kaniel et al. (2008) an Barber an Oean (2008) further fin that iniviual traers rely more on public information when they make their investment ecisions. 33

34 because of higher volatility. Thus, the CCT system makes iniviual traers submerge uring the last five minutes an lower the effect on the pricing error of closing price. The results in Moel (5) are consistent with our main finings after aressing the potential problem of omitte variables. [Insert Table 6 here] 4.5. Traer Patience Accoring to Foucault et al. (2005), Kaniel an Liu (2006) an Roşu (2009), we conjecture that after the introuction of the CCT, investors become more patient uring the perio immeiately preceing the CCT phase. In this section, we provie evience by analyzing orer aggressiveness to test this conjecture. To examine orer aggressiveness, we use a classification scheme similar to Bias et al. (1995) to classify orers into four categories: Most aggressive (MostAgg), More aggressive (MoreAgg), Less aggressive (LessAgg), an Least aggressive (LeastAgg). The most aggressive orers are market orers or marketable limit orers, which are limit orers that buy at or above the prevailing ask or sell at or below the prevailing bi. More aggressive orers inclue limit orers that are submitte within the prevailing bi an ask. Less aggressive orers are limit orers at prices equal to the prevailing bi an ask on the 34

35 same sie of the orer book. The least aggressive orers inclue limit orers that buy at below the prevailing bi or sell at above the prevailing ask price. Table 7 reports the results on orer aggressiveness uring the interval immeiately prior to the CCT phase (13:35-13:40.) After the introuction of the CCT, the most aggressive orers ecrease by 14.3%, 21.4%, an 20.6%, while the least aggressive orers increase by 15.5%, 22.4%, an 5.3% for foreign institutions, futures proprietary firms an iniviual traers, respectively. The empirical results in Table 7 inicate that following the introuction of the CCT, foreign institutions, futures proprietary firms an iniviual traers become more patient in the perio immeiately preceing the CCT phase. Furthermore, the proportion of aggressive orers becomes significantly lower in the Post-perio in the time perio preceing the CCT, inicating that traers become more patient on their trae execution after the introuction of the CCT. We thus fin that after the introuction of the CCT, investors strategically elay their traes until the last continuous traing phase. These empirical results again are consistent with the theoretical preictions of Foucault et al. (2005), Kaniel an Liu (2006) an Roşu (2009). [Insert Table 7 here] 35

36 5. Conclusions With a etaile account-level ataset, we examine a unique event of switching to the CCT on the TAIFEX to access its impact on the market liquiity, volatility, an the price iscovery process at the market closing. More importantly, we provie evience on how the migration of traing volume near the aily market closing by ifferent traer types affect the market quality. Several interesting results emerge. First, we show that the effects of the introuction of the CCT affect the market significantly uring the last 10 minutes of traing an the traing volume is reallocate uring this perio. We observe that after the introuction of the CCT, investors elay their traes to the last 5 minutes as traing volume is foun to be shifte from the 13:35-13:40 interval to the 13:40-13:45 interval. In aition, we fin that the CCT mechanism makes investors more patient, which leas to a reuction in the bi-ask sprea an volatility in the interval immeiately prior to the CCT phase. We also fin an increase in the bi-ask sprea an volatility right before the market closing, because investors become less patient uring the last 5 minutes of traing. Finally, we fin that the introuction of the CCT makes the closing price less informative, especially when foreign institutions shift more of their traing to the CCT phase. 36

37 Overall, we fin strong empirical support for the theoretical moels of Foucault et al. (2005) an Roşu (2009). Our finings might be attribute to the fact that TAIEX futures represents a basket of assets an thus averse selection costs may be iversifie away, leaing a lower information asymmetries problem (see Subrahmanyam, 1991; Berkman et al, 2005). Given that investors are subject to less information asymmetry on the market, our empirical results provie more support for the liquiity-base moels of Foucault et al. (2005) an Roşu (2009). Accoring to Foucault et al. (2005) an Roşu (2009), the introuction of the CCT makes investors more patient in the interval right before the CCT phase an thus increases (ecreases) their tenency to submit limit (market) orers, in the interval, which in turn inuces a reuction in bi-ask sprea an volatility in this perio. Contrarily, the CCT reuces investors patience in the CCT phase, an thus ecreases (increases) their tenency to submit limit (market) orers uring the CCT phase, leaing to an increase in the bi-ask sprea an volatility as well as the pricing error before market closes. Secon, by analyzing the changes in traing behavior of ifferent traer types, we fin that the traing volume of foreign institutions are migrate significantly to the CCT phase. This fining shows hat foreign institutions prefer to trae uner a continuous closing environment. In aition, we also fin that the fact that the 37