The Impact of Crossng on Market Qualty: an Emprcal Study on the UK Market Carole Gresse Unversté de Rems Champagne-Ardenne CEREG, Unversté de Pars-Dauphne e-mal: carolegresse@hotmal.com JEL classfcaton: Keywords: G9 crossng network, alternatve tradng system, lqudty, transacton costs, adverse selecton, fragmentaton Prelmnary draft Unfnshed work Ths research was launched wth the help of ITG Europe, whch provded data from ts crossng network, techncal assstance and hgh-sklled knowledge. It has benefted from the expert collaboraton of John Mnderdes, head of portfolo tradng and research, Davd Karat, research strategst, and George McClntock, head of data systems at ITG Europe. Besdes, the LSE market data were provded by the CEREG (Pars-Dauphne Unversty). Fnally, the results, comments and conclusons n the paper have been acheved n total ndependence from any publc or prvate nsttuton and wholly reflect the opnon of the author. All errors are mne. Suggestons are welcome. Any correspondence should be sent to Carole Gresse 0, rue du Commandant Lamy 750 Pars.
The Impact of Crossng on Market Qualty: an Emprcal Study on the UK Market Abstract Snce 998, a few alternatve tradng systems have been operatng on the UK stock market. As a result, nvestors as well as dealers can not only trade on the central market of the LSE but may also submt orders to anonymous crossng networks, among whch the most actve has been POSIT, the matchng system run by the agency stockbroker ITG. By comparng market data from the LSE and nternal data from the POSIT crossng network over a 6-months' perod, ths paper tests the mpact of crossng on market qualty and, n partcular, on transactons costs, adverse selecton and volatlty.
The Impact of Crossng on Market Qualty: an Emprcal Study on the UK Market. Introducton Durng the last decades, competton has ntensfed between major market centres as well as between establshed exchanges and new tradng systems. Among other catalysts, progress n automaton and technologcal nnovaton have reduced the cost of establshng new propretary tradng systems. As a result, new electronc tradng systems have emerged all around the world and mult-system tradng has ncreased. European stock markets have made no excepton to the rule. New electronc systems, when not regstered as exchanges, have been regulated by the SEC and desgnated as Alternatve Tradng Systems (ATS). They nclude Electronc Communcaton Networks (ECN) and Crossng Networks (CN). The former are defned by the SEC as electronc systems that wdely dssemnate to thrd partes orders entered by an exchange market maker or OTC market maker, and permt such orders to be executed aganst n whole or n part. Though, the term ECN does not nclude any system that crosses multple orders at specfed tmes at a sngle prce. Conversely, the latter are defned by the SEC as systems that allows partcpants to enter unprced orders to buy and sell securtes and that crosses orders at specfed tmes at a prce derved from another market. The competton comng from these new tradng facltes has changed the structure of fnancal markets, and probably also the role of ntermedares on these markets. The mplcatons for lqudty are of much nterest for academcs, regulators and nvestors. In order to address, at least partally, ths ssue, ths paper focuses on the consequences of tradng through a crossng network, by testng market and CN prvate data. Concentratng on the mpact of crossng on European stock markets s relevant because, f ECNs have tremendously developed n the US, t s stll not the case n Europe where only CNs have emerged. ATS are defned by the SEC as automated systems that centralse, dsplay, match, cross or otherwse execute tradng nterest but that are not regstered wth the Commsson as natonal securtes exchanges or separated by a regstered securtes assocaton. Accordng to Barclay, Hendershott and McCormck (00), ECNs are nvolved n more than a thrd of total NASDAQ tradng volume and are now attemptng to buld market share n NYSE-lsted ssues.
.. The development of CNs CNs generally promse anonymty and lower transacton costs, but do not guarantee executon. In such, they address the needs of a certan type of traders. As a matter of fact, nsttutonal nvestors have long expressed the need for tradng systems that provde lowcost executon whle sacrfcng mmedacy and executon guarantees. Ths led to the development of the Reuters Instnet Crossng Network, ITG s POSIT and the New York Stock Exchange s Crossng Network n the US, the largest of these CNs beng POSIT. In Europe, two London-based crossng networks 3 are currently actve on European stock markets. POSIT, the crossng system of ITG Europe, was the frst one to open for European ordnary shares n 998. It has been followed by E-crossnet n 999, but stll has the bggest market share. These crossng networks match buy and sell orders perodcally, at specfed tmes of the tradng day. At the hour of a match, buy and sell orders are matched n order to maxmse the tradng volume but wthout calculatng any transacton prce. Executed orders are crossed at the central market md-quote. As a result, snce 998, nsttutonal nvestors and broker dealers have several venues to trade on the UK stock market: they can ether submt an order to the central market of the London Stock Exchange (LSE) or submt t to an CN. In the former case, they ncur the bd-ask spread but get hgher executon guarantee. In the latter case, ther probablty of executon s lttle but they are provded anonymty, they ncur no adverse selecton cost as ther orders are not vsble from the rest of the market and f executed, they trade at the md-quote wth no mplct transacton costs... The theoretcal debate The development of CNs n Europe rases several questons around the trade-off between the benefts of competton and the potental costs of order flow fragmentaton. The debate began n 979, wth Hamlton, who ponted out the two opposte effects of mult-market tradng and the devaton of a part of the order flow from the central market. Ether multsystem tradng ncreases competton among lqudty provders and thus reduces bd-ask spreads, or, conversely, the fragmentaton of the order flow between several locatons lowers economes of scale and probabltes of executon, resultng n hgher volatlty and spreads. 3 Let us note that an other brokerage frm, Garban, has also began to run some crosses on the UK market.
On the one hand, mult-market tradng models predct that the fragmentaton of the order flow wll reduce lqudty. Accordng to Mendelson (987), the dspersal of orders between several markets lowers the probablty of executon and lowers lqudty. Chowdry and Nanda (99) show that nformed trade volumes get hgher wth the number of markets : market makers ncur hgher adverse selecton costs but trade prces are more effcent. Hendershott and Mendelson (000) model a dealershp market wth competton for order flow comng from a CN. They show that the devaton of some orders from the exchange on to the CN and unexecuted orders comng back to the market from the CN make the market rsker, so that market makers wden ther spreads. The CN makes long lved nformatonal advantages more proftable and ncrease adverse selecton..3. Objectve and general organsaton of the paper CNs allow nformed traders to submt orders anonymously wth no publc dsclosure. Ths new opportunty for nformed traders may well ncrease the cost of nformaton asymmetry. Besdes, by fragmentng the order flow, CNs may also lower the nformatonal content of trades. These negatve effects wll be referred to as the fragmentaton effect. On the other hand, gvng the opportunty to trade at the md-quote, CNs such as POSIT contrbute to reduce the average cost of tradng, ncrease the competton between lqudty provders, as market makers and lmt order traders, resultng n lower bd-ask spreads, and fnally, f they brng nformed traders to trade hgher volumes, the hgher proporton of nformed transactons could well enhance effcency. Such postve effects wll be referred to as the competton effect. Ths paper emprcally nvestgates whch of these theoretcal effects s actually domnant on the UK stock market, by analysng the mpact on the market of orders submtted and crossed nto POSIT, over a sx months perod. It s organsed as follows. Secton provdes nformaton on the organsaton of the UK stock market. Secton 3 presents the workngs of the POSIT crossng network. Secton 4 descrbes the data and lays out the tested hypotheses. Methodology and results are developed n secton 5.. The organsaton and the workngs of the UK stock market The London Stock Exchange (LSE) admnsters three stock markets: the Alternatve Investment Market (AIM) for domestc small and growng companes, the Domestc Equty Market (DEM) for ordnary shares of UK, Channel Islands and Isle of Man companes and other companes wth prmary UK lstng (prncpally Irsh companes), and the Internatonal Equty Market (IEM) for non-uk stocks. 3
Three tradng platforms are operatng for the DEM and the AIM: SETS, SEAQ and SEATS PLUS. SETS s the tradng platform for most lqud stocks of the DEM. SEAQ s the man tradng platform for non SETS domestc equty market securtes and SEATS PLUS s the tradng platform for less lqud domestc stocks (SEATS securtes) and AIM securtes... The Stock Exchange Electronc Tradng Servce (SETS) SETS s the electronc order book that replaced the quote-drven competng market maker system for most lqud domestc equty market securtes. These securtes are ether consttuents of the FTSE 00 Share Index, or UK consttuents of the Eurotop 300 Index, or securtes that have ndvdual stock optons traded on LIFFE. The tradng day for SETS securtes runs from 8:00 am to 4.30 pm subject to a random openng and closng tme adjustment. The tradng day begns wth an openng aucton, then goes on wth a contnuous aucton, and ends wth a closng aucton. Auctons may also take place durng the tradng day trggered by substantal prce movements. Auctons are preceded by an aucton call perod durng whch member frms are permtted to enter and delete only two types of orders on the order book: lmt orders and market orders. 4 At the openng, the aucton call perod lasts from 7:50 am to a random tme between 8:00:00 am and 8:00:30 am ( the random start perod ). At the end of ths random start perod, the order book s frozen and an aucton matchng algorthm s run. 5 Once the aucton matchng process s complete, contnuous tradng begns. Four types of order can be nput durng contnuous tradng: lmt orders, at best orders, 6 execute and elmnate orders 7 and fll or kll orders. 8 Automatc executon suspensons may occur f the executon of an order other than fll or kll causes a prce movement exceedng 5% aganst the last transacton prce. In that case, tradng s suspended durng 5 mnutes then an aucton call perod takes place. 4 Market orders have a specfed sze but are entered wthout a prce. They can only be nput durng an aucton call perod. 5 Ths calculates the prce at whch the maxmum volume of shares n each securty can be traded. All orders that can be executed at ths prce wll be executed automatcally, subject to the prce and tme prortes. If the aucton matchng process results n a prce whch s 0% above or below the prce of the last automatcally executed trade of the prevous busness day, then aucton matchng s temporarly halted. The aucton call perod s then extended for mnutes. At the end of ths extenson, the aucton matchng process s run agan and there wll be no further prce checks. 6 No lmt prce s specfed on an at best order, whch s executed at as many dfferent prces as necessary untl the order s completed n full. As much of the order as possble wll be executed mmedately and any remander wll be cancelled. 7 Execute and elmnate orders allow partcpants to fll as much of ther order as s avalable on the order book up or down to a specfed lmt prce. So, they are entered wth a quantty and a lmt prce. They are executed mmedately for as much as possble and any unexecuted porton wll be cancelled. 4
Fnally, the closng aucton call perod begns at 4:30 pm and ends at a random tme between 4:35:00 pm and 4:35:30 pm ( random close tme ). At the random close prce, the auctonmatchng algorthm s run, n the same way as the openng aucton. Moreover, f the total volume that would execute s less than a pre-determned multple of the Normal Market Sze (NMS) 9 of the securty, then aucton matchng wll not occur. Besdes, partcularly large trades or trades wth non-standard condtons can be negotated away from the order book, enablng member brokerage frms to commt rsk captal to large trades... SEAQ SEAQ s the screen based compettve market makng segment of the Exchange tradng system for non order book domestc equty securtes. A SEAQ securty s a domestc equty market securty for whch a mnmum of two market makers regster wth the Exchange. Each market maker s oblged to dsplay frm two-way prces on SEAQ n the NMS, 0 or reduced NMS n the case of reduced sze market makers, durng the Mandatory Quote Perod (MQP), whch lasts from 8:00 am to 4:30 am. From 7:30 am to 8:00 am, quotes may be opened but prces are regarded as beng ndcatve only. From 4:30 pm to 5:5 pm, market makers may contnue to dsplay frm quotes but are not oblged to do so and the tradng system remans open for tradng reportng. Durng the tradng day, the best bd and best offer prces quoted by market makers on SEAQ are commonly referred to as the yellow strp. In the event that quotatons by or more market makers are dentcal n terms of prce, the best quote wll be the one that was entered frst. Besdes, three auctons, where only lmt orders can be submtted, are run durng the tradng day for SEAQ securtes that are part of the FTSE 50 Share Index. The SEAQ aucton tmes are :00 am, 3:00 pm and 4:45 pm. These auctons were turned nto crosses n Aprl 00, but nearly no transactons are effectvely executed through these batch algorthms. 8 A fll or kll order must be entered wth a quantty and may be defned wth a lmt prce. It wll be executed ether n full or not at all and non-executed fll or kll orders do not st on the book. 9 For a SETS securty, the NMS s a quantty specfed by the Exchange accordng to the average sze of trades over the last months. The pre-determned multples for the closng volume check are 0,5 for securtes wth a NMS of 5000 or more, or the aggregate executable volume of 500 for securtes wth a NMS of below 5000. 0 The NMS classfcaton of SEAQ securtes are revewed quarterly usng the followng formula: (value of customer turnover n prevous months n )/(closng md-prce on last day of quarter 0000). NMS s are then rounded up or down to one of the followng bands: 00, 00, 500, 000, 000, 3000, 5000, 0000, 5000, 5000, 50000, 75000, 00000, 50000, 00000. Some market makers are granted specal permsson to dsplay prces n smaller quanttes than NMS. The reduced NMS s half the NMS rounded down to the nearest NMS band. 5
.3. SEATS PLUS SEATS PLUS s the Stock Exchange Alternatve Tradng Servce for the tradng SEATS and AIM securtes. It s a mxed system whch supports the entry of both market makers quotes and orders. NMS for SEATS securtes s set at 000 shares and NMS for AIM securtes falls under the same regme as used for SEAQ securtes. Three types of orders are avalable for entry on SEATS PLUS: frm exposure orders (lmt orders), ndcatve exposure orders and ht orders (order submtted to execute automatcally aganst a frm exposure order). The entry of exposure orders s allowed throughout the trade reportng perod (7:5 am-5:5 pm). Yet, the entry of ht orders s restrcted to the MQP. 3. The POSIT crossng network Run by the agency stockbroker ITG, POSIT s an ntra-day electronc tradng system, 3 whch matches buy and sell orders at predetermned tmes n the day. The system s totally anonymous and reduces transacton costs by usng md-market prcng for executon. It was created n 987, as a jont venture between ITG Inc. and BARRA Inc., the Calforna based quanttatve house, n response to the dealng neffcences n the US equty market, prmarly wth respect to smaller, less traded, stocks and n partcular n response to the ssues of market mpact. As a matter of fact, "openng" to the market a decson to buy or sell a less lqud stock would often result n a major shft n the share prce of that stock, wth lttle or no turnover havng taken place. Ths led to reluctance from nsttutonal nvestors to even pass orders to the market n such stocks. ITG crossng technology ams at addressng these neffcences. Already operatng n the US and the Australan markets, POSIT was launched n Europe n 998, and s now workng n ten European countres (UK, Germany, Swtzerland, France, Belgum, Netherlands, Italy, Span, Sweden, Fnland). 3.. The crossng technology and the matchng tmes Orders can be submtted to POSIT contnuously, at any tme of the tradng day. Anonymty s protected and order detals are never dvulged externally or dsclosed to the market. Submssons are free of charge. An ndcatve exposure order ndcates a prce at whch the person on whose behalf the order s dsplayed may be prepared to deal. 3 POSIT stands for Portfolo System for Insttutonal Traders. 6
The propretary matchng algorthm wthn POSIT s run at desgnated tmes each day. In order not to allow gamng and manpulatng strateges, at the desgnated tme of a match, a random executon tme wthn a seven mnute wndow s generated from the POSIT computer so that no one, nether clents nor the tradng desk, s aware of the exact match tme. Any order receved before the desgnated match tme wll be ncluded n the match pool, but any order receved after the start of the match wndow wll be taken on a best endeavour bass up to the tme the match s run. Any order subsequently receved would be for the next scheduled match. The POSIT algorthm compares all submtted orders confdentally and s set to maxmse the total value of shares traded, gven the constrants 4 assocated wth submtted orders. Matchng orders are crossed at the rulng md-prce taken from the lead market quote for each stock, 5 and reported to the relevant authorty after executon. Only executed orders are charges a 0 bass ponts brokerage commsson. The match tmetable (n UK tme) conssts of sx ntra-day matchng tmes as follows: 9:00 am, 0:00 am, :00 am, :00 am, :00 pm, 3:00 pm, and accordng to tradng actvty, some days, a seventh match s also run at 4:00 pm. Table POSIT match tmes (UK tme) Perod Match tmes 8 nov 998 9 sep 999 :00 am, 5:00 pm 0 sep 999 9 jan 000 9:30 am, :00 am, 3:00 pm 0 jan 000 7 mar 000 9:30 am, :00 am, :00 am, 3:00 pm 8 mar 000 9 nov 000 9:30 am, :00 am, :00 am, 3:00 pm, (4:00 pm) 30 nov 000 5 jan 00 8:30 am, 9:30 am, :00 am, :00 am, 3:00 pm, (4:00 pm) 6 jan 00 8 mar 00 8:45 am, 9:30 am, :00 am, :00 am, 3:00 pm, (4:00 pm) From 9 mar 000 9:00 am, 0:00 am, :00 am, :00 am, :00 am, 3:00 pm, (4:00 pm) Lnes n grey correspond to the observaton perod. 4 Clents can assocate dfferent types of constrants on the orders they submt to POSIT, so as to avod unfavourable match executons. These constrants are detaled n Appendx. 5 POSIT technology also offers clents the ablty to generate trades that requre market prnts (e.g. nternal crosses across dfferent underlyng clents) by means of "drected crosses". These bespoke matches may take place at any tme durng the tradng day, outsde of the normal scheduled match tmes and may use the standard POSIT md-pont prcng or some other benchmark prcng, e.g. VWAP. These drected crosses are excluded from our dataset. 7
Ths current tmetable results from several changes summarsed n table. When ITG Europe launched POSIT for UK equtes n November 998, only two daly matches were run at :00 am and at 3:00 pm. A thrd match was ntroduced at 9:30 am, n September 999 and a further one, at mdday, was added n January 000. Then, n March 000, the unoffcal 4:00 pm match was ntroduced. A new 8:30 am was added n November 000 and moved to 8:45 am n January 00. Fnally, the match tmes were moved to the current hourly tmetable n March 00. 4. Data and testable hypotheses The data used n the emprcal nvestgatons consst of LSE hgh frequency market data for all UK domestc stocks, over a sx months perod from July to December 000, and of POSIT order data, for UK stocks, on the same perod of tme. 4.. Market data Tck by tck market data from the London stock market nclude transacton data and best prces data. Best prces correspond to the best bd and offer market makers' quotes for SEAQ and SEATS stocks and to the best lmt prces from the order book for SETS stocks. Quanttes assocated to best prces are not avalable so that the NMS s used as a proxy. 4.. POSIT data We have been provded wth POSIT data over the same observaton perod. These data conssted of two SQL tables. One table ncluded the characterstcs of the orders submtted n the CN, such as the ITG code dentfyng the stock, the sze of the order n number of shares, the type of the ntator, that s "nsttutonal nvestor" or "broker-dealer", the constrants assocated wth order and the date and tme of the match to whch the order s beng submtted. The second table ncluded the characterstcs of the orders executed n the CN: the stock ITG code, the executed quantty, the type of ntator, the md-prce used for executon and the date and tme of the correspondng match. All these characterstcs are used for the emprcal tests, except the constrants assocated to the orders (see Appendx ) as they convey no relevant nformaton for our purpose. Before runnng any emprcal tests, these raw data have been rearranged for the purpose of the research n a few ways. Frst, the submsson table was merged wth the executon table, so as to allow to exhbt for each submsson whether t s totally or partally executed, or not executed at all. Then, we establshed the correspondence between the stock ITG codes and the ISIN codes used to dentfy stocks n the LSE market database. Fnally, a procedure was 8
set up to determne whether a submsson to POSIT was made for the frst tme or whether t was an order resubmtted after remanng unexecuted n the prevous match. In the end, a sngle table was bult up. It contans, for each submsson to POSIT: the stock ITG code, the ISIN code, the date and tme of the match that s submtted to, the type of the ntator, the submtted quantty, the executed quantty and the prce of executon f any. 4.3. The sample The study s frst restrcted to SEAQ stocks, as POSIT has been partcularly actve on these stocks snce ts launch, and wll further be extended to SETS equtes. Durng the observaton perod, that s 7 tradng days from July to December 000, SEAQ market-makers quoted prces for 657 domestc stocks, among whch 647 were prced n GBP (ether pennes or pounds) and 0 were prced n USD. Over the 647 GBP quoted stocks, 643 were effectvely traded whle no transacton took place for 4 of them. The total amount n GBP traded over these 643 stocks was 80 450 mllons, that s 44 35 mllons n number of shares. Over these volumes, 955 mllons of GBP (379 mllons of shares) were traded through POSIT, whch equals,% of the whole volume traded on the market over the 6-months' observaton perod. Detals on ntra-day tradng actvty for these 643 stocks are gven n tables and 3. Then, most llqud stocks were elmnated from ths sample, to avod strange effects due to extreme values. Lookng at the dstrbuton of average quoted spreads across the sample, t appears that the average quoted spread s under 30% for 98,% of the 643 stocks, whch represents 99,99% of the total traded volume over the observaton perod for these stocks. Henceforth, excludng the stocks 6 for whch the average quoted spread exceeds 30% does not substantally dstort the sample. Furthermore, we requre that every stock ncluded n the sample was effectvely quoted by SEAQ market makers durng at least 00 days of the observaton perod, n order to get smlar numbers of daly observatons (.e. between 00 and 7) for all the stocks and thus to make emprcal varables comparable on an nter-stock bass. Followng these crtera, the sample s then reduced to 45 stocks, from whch one s excluded because effectve spreads appled on transactons for ths stock can not be computed. 7 6 These equtes are generally very low-prced stocks, wth very lttle tradng volumes, whch explans ther ncredbly hgh spreads. 7 To determne the sde of a trade, we use two condtons: the sde offcally reported by the market maker who declared the trade and the dfference between the transacton prce and the current md- 9
GBP volumes traded throughout the tradng day 8 Table Tradng volumes for UK GBP-prced SEAQ stocks Total over the perod Average per day On the market 78 39 47 70 67 53 3 In POSIT 954 583 303 7 56 404 % n POSIT,%,% Number of trades throughout the tradng day Table 3 Number of trades for UK GBP-prced SEAQ stocks Total over the perod Average per day Average sze of trade n GBP On the market 6 930 7 80 34 657 In POSIT 7 685 6 4 04 % n POSIT 0,34% 0,34% --- Consequently, the fnal sample conssts of 450 SEAQ stocks. Orders were submtted nto POSIT for 6 of them, out of whch 56 were traded at least once n the CN. The POSIT share n total traded volume exceeded % for 75 of these stocks and exceeded 5% for. 73 78 mllons of GBP were traded on the stocks of the sample from July to December 000, over whch,7%, that s 935 mllons, were transacted through POSIT. These 73 78 mllons of GBP represented,7 mllons of transactons, on whch 0,35% (.e. 7 534 trades) were POSIT-executed orders. Such s the case because the average sze of a trade n POSIT s 3,7 tmes the average sze of a trade on the market. These fgures are set out n Tables 4 and 5. GBP volumes traded throughout the tradng day Table 4 Tradng volumes for the stocks of the sample Total over the perod Average per day On the market 73 77 66 48 580 453 749 In POSIT 934 7 54 7 360 00 % n POSIT,7% --- quote at the tme of the trade. Followng Lee and Ready (99), a postve dfference s supposed to ndcate a purchase whle a negatve dfference would ndcate a sale. In case of contradcton between both condtons, we consder that the sde of the trade s unknown and that the effectve spread appled on the transacton ( trade prce mdquote mdquote ) can not be computed. 8 Overnght transactons are excluded from the analyss as POSIT s only workng durng the day. The tradng ntra-day perod we consder lays from 8:00 am to 5:00 pm, as we notced that tradng volumes keep hgh tll 5:00 pm, even f the MQP closes at 4:30 pm. 0
Number of trades throughout the tradng day On the market - buyng orders - sellng orders Table 5 Number of trades for the stocks of the sample Total over the perod 70 504 99 458 97 040 Average per day 7 09 9 445 7 646 Average sze of trade n GBP 33 963 3 544 34 48 In POSIT 7 534 59 4 067 % n POSIT 0,35% --- --- Throughout the observaton perod, the market was rather bearsh. The cross-sectonal average close-to-close return rose to 0,0659% for the sample. Ths average s obtaned as follows. Frst, for each stock of the sample, the equally weghted mean of daly returns s calculated n logarthm on closng md quotes. Second, a mean of the ndvdual average returns s computed, weghtng each stock by the total volume traded on the stock over the perod. The cross-sectonal average volatlty of close-to-close returns, computed n the same manner, equals,807% (see Table 6). Cross-sectonal analyss on stock-by-stock average returns Weghted mean Table 6 Daly returns Standard devaton Mnmum Maxmum Number of stocks Close-to-close average returns -0,0659% 0,336% -,8335% 0,7894% 450 Close-to-close returns volatlty,807%,8036% 0,4% 450 Concernng transacton costs, the cross-sectonal mean of tme-weghted average quoted spreads s,373% whle the volume-weghted average effectve spread appled on transactons only equals,77%, the average effectve spread on sales (,0%) beng substantally superor to the average effectve spreads on purchases (,39%). As a measure for depth, the average NMS n GBP equals 98,55 over the perod. Furthermore, the level of lqudty for the sample s estmated usng the Kyle depth coeffcent, that s the devaton n prce from md-quote to accept to be able to trade one more unt of share. We wll calculate ths varable n percentage of md prce and wll refer to t as the unt margnal cost (UMC). Over our sample, the cross-sectonal mean of the average quoted UMC equals 0,000359% of md quote. It goes down to 0,0007% when calculated wth effectve spreads appled on trades (see Table 8). Yet, the sample s heterogeneous n terms of lqudty, as for the most lqud stock, the average quoted UMC equals 0,000007%
whle t rses to 0,0830% for the least one. Dfferences n lqudty across the sample are even more strkng, lookng at the average effectve UMC, the lowest beng nearly null (0,37.0-8 ) and the hghest one reachng 0,0990%. Tables 7 and 8 provde statstcs on transacton costs, depth and lqudty. Cross-sectonal analyss on stock-by-stock average spreads Table 7 Transacton costs Weghted mean Standard devaton Mnmum Maxmum Number of stocks Tme-weghted average quoted spread,373%,4875% 0,4847% 8,6074% 450 Volume-weghted average effectve spread,7700%,7780% 0,0338% 50,5649% 450 V-W average effectve spread on purchases,39%,3398% 0,0000% 5,6957% 446 V-W average effectve spread on sales,0%,4578% 0,0000% 78,8388% 449 Cross-sectonal analyss on stock-by-stock average UMC Table 8 Unt margnal cost (UMC) n % of md-prce Weghted mean Standard devaton Mnmum Maxmum Number of stocks Tme-weghted average quoted UMC 0,000359 0,000705 0,000007 0,0830 450 Volume-weghted average effectve UMC 0,0007 0,0007 0,00000037 0,0990 450 V-W average effectve UMC on purchases 0,0004 0,0068 0,000000,8585 446 V-W average effectve UMC on sales 0,0009 0,00039 0,000000 0,00397 449 Quotes on the stocks of our sample are not very frequently revsed: the average number of quote revsons per day and per stock s,6 (see Table 9). Cross-sectonal analyss on stock-by-stock averages Table 9 Number of quote revsons Weghted mean Standard devaton Mnmum Maxmum Number of stocks Average daly number of quote revsons,6 8,0 56 450 Tradng actvty wthn POSIT was substantal for the stocks of the sample. The amount of submtted orders reached 43 6 mllons of GBP over the perod, that s 58,55% of the total traded volume on the market. Includng resubmssons, ths amount goes up to 7 846 mllons of GBP. Insttutonal nvestors ntated 5,8% of these orders whle market makers submtted 48,%. Sellng orders (59,86%) exceed buyng orders (40,4%), ths mbalance beng hgher for orders placed by market makers. Market makers placed nearly twce (,9 tmes) more sell orders than buy orders n the CN.
If the major part of submtted orders came from nsttutonal nvestors, most executed orders were market maker-ntated (67,69% of the volume executed n POSIT over the perod). Table 0 dsplays more detals about submssons and executons n the CN over the observaton perod. Table 0 Tradng actvty n POSIT Total over the observaton perod From nsttutonal nvestors From brokerdealers Total submtted volume* n GBP 43 6 465 83 356 870 573 0 805 595 59 - n % of total submtted volume* 00,00% 5,80% 48,0% Total number of submtted orders* 3 9 35 663 87 58 Average sze of an submtted order* 350 370 66 893 37 70 Total submtted buy volume* n GBP 7 36 8 8 0 058 683 880 7 68 38 338 - n % total submtted volume* 40,4% 3,30% 6,84% Total number of submtted buy orders* 4 480 4 484 7 996 Average sze of an submtted buy order* 407 88 694 469 59 63 Total submtted sell volume* n GBP 5 835 643 65 98 86 694 3 537 456 9 - n % of total submtted volume* 59,86% 8,49% 3,36% Total number of submtted sell orders* 80 7 79 59 53 Average sze of an submtted sell order* 30 0 580 678 7 398 Total executed volume n GBP 934 7 54 30 005 599 63 76 95 - n % total executed volume 00,00% 3,3% 67,69% - n % of total market traded volume,7% 0,4% 0,86% Total executed volume over total submtted volume,7%,35% 3,04% Total number of executed orders 7 534 848 5 686 Average sze of an executed order 4 067 63 43 76 * Includng new submssons only / excludng resubmssons of unexecuted orders Before nvestgatng the mplcatons for the market of the CN actvty, the testable hypotheses ensung from theoretcal models are lsted n next subsecton. 4.. Testable hypotheses From theoretcal models on fragmentaton, we derve a seres of testable hypotheses on the mplcaton of tradng through a CN for market qualty. 3
. Fragmentaton effect vs competton effect H. The competton effect domnates the fragmentaton effect. If H s true, the effectve spread should be negatvely related to the share of traded volume executed n the CN (postvely otherwse).. Adverse selecton H. The fragmentaton of the order flow between the central market and the CN creates addtonal adverse selecton costs. Under H, spreads would ncrease wth the share of order flow submtted to POSIT. 3. Inventory costs H3. The CN gves an opportunty to market makers to reallocate ther postons wth no mplct tradng cost and thus lowers nventory costs. Under H4, quoted spreads should be negatvely related to the share of volume traded by market makers through the CN. 4. Unexecuted CN order flow and mplct transacton sots H4. Unexecuted order flow comng back from the CN to the central market for executon, creates temporary tenson on lqudty, ether because t ncreases adverse selecton, as demonstrated n Hendershott and Mendelson (00), or because t suddenly generates abnormal nventory costs for market makers. Provded H3, spreads would wden wth the amount of unexecuted CN order flow. 5. Market mpact H5. Crossng reduces global market mpact of trades. Gven the realsaton of H5, the return volatlty per unt of traded volume would decrease wth the share of volume transacted through the CN. H6. Crossng reduces temporary market mpact. Provded H6, ntra-day volatlty around VWAP would be negatvely related to the share of traded volume executed n the CN. H7. Crossng reduces short-term volatlty around fundamental value. Under H7, the rato hourly return volatlty over daly return volatlty would decrease wth the share of traded volume executed n the CN. 5. Informed tradng and effcency H8. Informed traders submt more orders to the CN than to the central market. Returns could then be explaned not only by buy and sell trades on the market but also by buy and sell orders submtted to the CN. 4
H9. As for the mplcatons of H8 for effcency, the followng alternatve wll be tested: ether nformed tradng through the CN harms effcency and slows down the dscovery of prces (H9a), or, by ncreasng the total proporton of nformed tradng on the market, crosses n the CN accelerate the dscovery of prces (H9b). If H9a (H9b) holds, then the number of quote revsons per day would decrease (ncrease) wth the share of order flow (ether submtted or executed) gong to the CN. H0. Fnally, f crossng accelerates quote revsons, does t effectvely speed up the dscovery of prces (H0a), or does t just make the process of dscovery more complex (H0b)? Under H0a (H0b), the lower the share of order flow (ether submtted or executed) gong to the CN, the later (the earler) prce movements would take place wthn the tradng day. 5. Methodology and results The methodology conssts of a cross-sectonal analyss on stock-by-stock average measures, as a preparatory work to a stock-by-stock temporal analyss. We frst defne the varables used n the cross-sectonal regressons and then gve the results. 5.. Varables and notatons Dependent varables Our analyss uses as dependent varables average spreads, measures of volatlty, average daly returns and the average number of quote changes per day. s the average quoted touch or market spread (.e. the dfference between the best offer and the best bd quoted on the market reported to the md-quote), for stock, calculated by weghtng each quoted spread wth ts tme of valdty. ES s the average effectve spread weghted by trade volumes for stock. σ DTV s the emprcal standard devaton of close-to-close returns on the observaton perod for stock dvded by the average daly traded volume n GBP. Ths varable measures the mpact of tradng volumes on prces. σ ( VWAP) s the average standard devaton of trade prces from VWAP durng the tradng day. VWAP s the volume-weghted average prce of the stock on a gven tradng day and s used by operators as a benchmark ether to prce transactons or to valuate tradng performance. σ( VWAP) measures short-term volatlty around the mean level of prces due to mplct transacton costs and market mpact of trades. 5
( ) ( ) σ h σ oc s consdered as a proxy for short-term volatlty around fundamental value. It reports the sum of hourly return varances to σ ( ) ( oc), the varance of open-to-close returns. Let us note σ k the varance of md-quote returns on the kth hour of the 8,5 tradng day for stock : σ ( h) = σ ( k). k= R s the mean of daly returns calculated n logarthm on closng md-quotes for stock. NQ s the average number of quote revsons per day. As a dependent varable, t s a proxy for the speed of prce adjustment to nformaton. SP s computed to measure, for each stock, the speed of prce adjustment throughout the tradng day. It s the mean of the dfferent hours (,, 3, 4, 5, 6, 7, 8, 8.5) of the tradng day, each hour k beng weghted by σ 8,5 k= σ ( k) ( k) σ = σ ( k) ( h). Thus, 8,5 σ SP = k k= σ ( k) ( h). The hgher SP, the later prce movements take place wthn the MQP. Control varables For each dependent varables, control varables are determned by runnng stepwse lnear regressons of the dependent varable over a range of possbly explanng varables and keepng the most powerful model. For each dependant varable, the results from the stepwse regressons confrm theoretcal and ntutve predctons. Table descrbes the selected control varables and Table exhbts, for each dependent varable, the correspondng control varables Explanng varables Our analyss conssts of examnng the effects of varables measurng the tradng actvty n the CN, on the dependent varables we have defned. These CN-actvty related varables are: X, whch s the total volume n GBP traded through POSIT for stock over the perod, n percentage of the total volume (n GBP) of stock traded on the market, NS, whch s the total amount n GBP of orders submtted to POSIT on stock reported to the total market traded volume, ncludng new submssons only and excludng resubmssons of unexecuted orders, 6
Table Selected control varables σ Control varable TV BTV STV NMS ES EUMC NQ NT Sgnfcaton Standard devaton of daly returns for stock Logarthm of the total volume n GBP traded on stock over the perod Logarthm of the total amount n GBP of buyng trades on stock Logarthm of the total amount n GBP of sellng trades on stock Logarthm of the average NMS (n GBP) of stock Average quoted market spread for stock Average effectve spread weghted by trade volumes for stock Average effectve margnal cost of one share for stock Average number of quote revsons per day for stock Average number of trades per day for stock Table Correspondence between dependent varables and control varables Dependant varable Control varables σ,tv, NQ ES σ,tv, NMS σ DTV ES, EUMC σ ( VWAP), σ,tv, NT σ ( h) σ ( oc) NT,TV,, NQ R BTV, STV QN SP NT, NMS,TV, σ TV,,, NQ NT S, whch s the total amount n GBP of all orders submtted to POSIT on stock reported to the total market traded volume, ncludng resubmssons of unexecuted orders, U, whch s the total amount n GBP of unexecuted POSIT orders reported to the total market traded volume for stock (U = NS X ), XM, whch s the total volume n GBP of stock traded by market makers through POSIT, n percentage of the total volume (n GBP) of stock traded on the market, XI, whch s the total volume n GBP of stock traded by nsttutonal nvestors through POSIT, n percentage of the total volume (n GBP) of stock traded on the market, 7
SBI, whch s the total amount n GBP of buy orders submtted to POSIT by nsttutonal nvestors on stock reported to the total market traded volume, SSI, whch s the total amount n GBP of sell orders submtted to POSIT by nsttutonal nvestors on stock reported to the total market traded volume, SBM, whch s the total amount n GBP of buy orders submtted to POSIT by market makers on stock reported to the total market traded volume, SSM, whch s the total amount n GBP of sell orders submtted to POSIT by market makers on stock reported to the total market traded volume. 5.. Fragmentaton effect vs competton effect Frst, to test H, we regress, as shown n equaton (), the average effectve spread on X, the share of the CN n the total traded volume for stock, controllng for σ, TV and NMS. ES = a + bσ + ctv + dnms + ex + ε () By runnng a lnear OLS-regresson on our 450 observatons, we obtaned ES = 9, 799 + ( 5,874 ) 0,866 σ 0,9 TV 0,438 NMS,8.0 X + ε ( 0,0 ) ( 0,554 ) ( 3,645 ) ( 0,6 ) (), wth R²=48,6%. The e coeffcent s negatve but not sgnfcantly dfferent from zero. Runnng the regresson on the sub-sample of the 56 stocks effectvely traded through the CN over the perod, the results become: ES = 8, 77 + 0,509σ 0,55 TV 0,343 NMS 0,56 X + ε ( 5,459 ) ( 8,98 ) (,65 ) (,988 ) (,878 ) (3), wth R²=9,7%. e s sgnfcantly negatve at a % level n equaton (3). Accordng to these results, the fragmentaton s not domnant, and the domnaton of the competton effect s slght. 5.3. Adverse selecton Regressons (4) and (5) tests H: = a + b σ + c TV + d NQ + e NS + ε (4), ES = a + bσ + ctv + d NMS + e NS + ε (5). The estmates gven n equatons (6) and (7) show that the relatve amount of orders submtted to the CN has a negatve mpact on quoted spreads and no sgnfcant effect on 8
effectve spreads, whch allow us to conclude that t does not create addtonal adverse selecton costs. ES = 0,659+ ( 3,03 ) = 9, 78 + ( 5,957 ) 0,94 σ,3 TV 6,87.0 NQ,0.0 NS + ε ( 7,8 ) ( 0,033 ) ( 5,3 ) ( 0,44 ) ( R = 6,5% ) ( 3,08 ) ( 3, 730 ) 4 ( 3,08 ) 0,865 σ 0,897 TV 0,443 NMS +,936.0 NS + ε ( R = 48,6% ) ( 0,37 ) 5 (6). (7). 5.4. Inventory costs To test H3, we run the regresson (8), frst on the total sample, secondly on a sub-sample consstng of the 56 stocks for whch trades have been executed through the CN. = a + bσ + ctv + cnq + dxm + ε (8) Results for the total sample (450 stocks) show that quoted spreads are negatvely related to the relatve amount of transactons traded by market makers through POSIT but the negatve coeffcent s not sgnfcantly dfferent from zero (see equaton 9). = 0,64+ ( 30,5 ) 0,907 σ,079 TV 6, 74.0 NQ 0,08 XM + ε ( 6,37 ) ( 4,7 ) ( R = 6,% ) (,958 ) (,34 ) (9). Focusng on the sub-sample (56 stocks), ths negatve effect becomes sgnfcant at a % threshold (see equaton 0), meanng that tradng n the CN reduces nventory costs for market makers and allow them to tghten quotes. = 8,5+ ( 8,996 ) 0,635 σ 0,36 TV 8, 76.0 NQ 0,36 XM + ε ( 7,84 ) ( 6,503 ) ( R = % ) ( 6,30 ) ( 4,604 ) (0). 5.5. Unexecuted order flow and mplct transacton costs H4 s tested through regressons () and (): = a + b σ + c TV + d NQ + e NS + ε (), ES = a + bσ + ctv + d NMS + e NS + ε (), frst on the total sample, second on the sub-sample of the stocks actually traded n the CN. The results dsplayed n equatons (3), (4), (5) and (6) show that the unexecuted order flow n the CN does not sgnfcantly harm lqudty. On the whole sample: 9
ES = 0,659+ ( 3,03 ) = 9, 78 + ( 5,956 ) 0,94 σ,3 TV 6,88.0 NQ,0.0 U + ε ( 7,8 ) ( 0,03 ) ( 5,3 ) ( R = 6,5% ) ( 0,44 ) ( 3,083 ) ( 3, 73 ) 4 ( 4,39 ) 0,865 σ 0,897 TV 0,443 NMS +,937.0 U + ε ( R = 48,6% ) ( 0,37 ) 5 (3), (4). On the sub-sample (56 stocks): = 7,45 + (7,34 ) 0,683 σ 0,304 TV 9, 74.0 NQ 4,887.0 U + ε ( 9,656 ) ( 5,308 ) ( R = 6% ) ( 6,968 ) (,47 ) 4 (5), ES = 7,56 + 0,538 σ 0,67 TV 0,47 NMS +,9.0 U + ε ( 4,507 ) ( 9,579 ) (,05 ) (,43 ) ( R = 8,5% ) 4 ( 0,538 ) (6). 5.6. Market mpact Testng H5, we fnd no sgnfcant mpact of X on σ DTV : crosses through POSIT have no sgnfcant effect on the global market mpact of trades. Yet, regresson (7) ndcates that they reduce temporary market mpact and valdates H6. σ ( VWAP) =, 78 + ( 8,98 ) 0,6 (,876 ) + 0,4 TV ( 9,685 ) σ + 0, ( 6,504 ) 3,85.0 (,98 ) X + 4,7.0 + ε ( 4, 76 ) 3 ( R NT = 43,4%) However, ths negatve effect s manly due to market makers crosses, as shown n regresson (7). (8). σ ( VWAP) =, 784+ ( 8,94 ) 0,5 σ (,806 ) 6,8.0 ( 3,08 ) + 9,987.0 XM ( 6,46 ),.0 ( 0,565 ) + 4,.0 XI ( 4, 794 ) + ε 3 NT + 0,07 TV ( R ( 9, 747 ) = 43,5%) (8). H7, tested n regresson (9), s rather smlar to H6, except that t focuses on short-term volatlty around fundamental value, as the dependent varable reports hourly return volatlty to open-to-close return volatlty. Accordng to the estmates, the excess of shortterm volatlty over long-term volatlty decreases wth the share of order flow executed n the CN. 0
σ σ ( h) ( oc) =,59 3,9.0 ( 9,674 ),7.0 ( 8,38 ) ( 5,036 ) NQ TV + 3,363.0,9.0 ( 8, 705 ) (,834 ) 3 X NT + ε +,03.0 ( 6,347 ) ( R = 6,4% ) (9). 5.7. Informed tradng and effcency Another ssue of nterest s the nformatonal content of orders submtted to the CN. One may farly wonder whether orders anonymously submtted to POSIT convey more nformaton than orders executed on the market. In order to test ths hypothess (H8), as shown n regresson (0), average daly returns are regressed on market buy and sell trade volumes and on the relatve amounts of buy and sell orders submtted to POSIT by nsttutonal nvestors on the one hand, by market makers on the other hand. R = a + bbtv + cstv + dsbi + essi + fsbm + gssm + ε (0). The results of the regresson dsplayed n equaton () are somehow dsappontng. The R² coeffcent s very low, whch means that buyng and sellng trade volumes are not effcent control varables for R. R = 7,9.0 (,386 ) + 7,08.0 ( 4,9 ) 4 +,78.0 (,84 ) BTV SBI 7,.0 ( 4,4 ) 7 7,63.0 ( 0,58 ) STV SSI +,44.0 + ε ( 3,06 ) 3 SBM ( R,5.0 (,334 ) =,4%) 4 SSM (). Yet, average returns are sgnfcantly related to the POSIT order flow comng from market makers: they are postvely (negatvely) related to the relatve amount of buy (sell) orders submtted by market makers nto the CN. However, we can not conclude to a predctve power on returns of these varables, as the regresson s run on average or aggregated measures over the whole perod. For further nterpretatons, a analyss on stock by stock temporal seres wll be carred out. Fnally, n order to test H9,.e. the effect of the CN actvty on prce dscovery, the coeffcents n regressons () and (3) are estmated: NQ = a + b NT + c TV + d NMS + e σ + f + g S + ε (), NQ = a + b NT + c TV + d NMS + e σ + f + g X + ε (3). Equatons (4) and (5) lay out the results: NQ = 5,7+ () 0, NT + 0,69 TV ( 35,43 ) 9,44.0 ( 4,87 ) + 0,5 NMS ( 4,48 ) ( 6,57 ) 6 + 7,039.0 S + ( 0,67 ) ε + 0,4 σ (4), ( 3,503 ) ( R = 7,4% )
NQ = 3,39+ ( 4,66 ) 0,3 NT + 0,57 TV ( 35,45 ) 9,5.0 ( 4,378 ) ( 4,8 ) + 0,83 X ( 3,853 ) + 0,465 NMS ( 5,88 ) + ε + 0,35 σ (5). ( R = 77, %) ( 3,857 ) They show that nvsble submssons to POSIT do not slow down the dscovery of prces and conversely, executons n POSIT accelerate quote revsons, whch valdates H9b. Yet, we may stll wonder f the acceleraton of quote revsons actually make prce adjustments faster (H0). Testng the mpact of S and X on SP n regressons (6) and (7), we confrm that submssons to POSIT do not slow down the dscovery of prces, but we can not valdate H0a: though the effect of X on SP s negatve, t s not sgnfcant. SP = 5,34 ( 6,49 ) 0,53 ( 7,73 ) TV + 3, 74.0 ( 0,805 ) 6 +,74.0 S ( 6,06 ) + ε NT + 4, 768.0 ( 5,87 ) ( R = 0,3% ) 4,08.0 ( 3,5 ) NQ (6), SP = 5,97 ( 6,4 ) 0,5 3, 74.0 TV ( 7,595 ) (,766 ) +,39.0 X + ε ( 5,85 ) NT + 4,646.0 ( 5,678 ) ( R = 0,4% ) 3,8.0 ( 3,6 ) NQ (7). Yet, t s worth denotng that, splttng X n ts two components XM and XI, there s a sgnfcantly negatve effect of XM upon SP (regresson 8): we may then assess that market makers' tradng n the CN accelerate the dscovery of prces. SP = 5,306 ( 6,4 ) 0,5 6, 7.0 TV ( 7,595 ) (,99 ) +,03.0 XM ( 5,85 ) 6,43.0 ( 0,86 ) NT + 4,585.0 3 XI ( 5,678 ) + ε 3,49.0 NQ ( 3,6 ) ( R = 0,5% ) (8). 6. Concluson We do not fnd that tradng through a CN create addtonal adverse selecton costs, conversely to what predcts nformaton-based theoretcal models on mult-system tradng. Our results assess that, when a CN operates on a dealershp market, the competton effect domnates the fragmentaton effect. Moreover, the CN s used as a lqudty-provdng system reducng nventory costs. Fnally, executons through the CN accelerate quote revsons, and facltates the dscovery of prces for market makers.
Appendx : Order constrants avalable n POSIT To control over unpredctable match outcomes, a range of constrants can be appled to ether ndvdual stocks, pars of stocks or to sngle or dual drecton lsts. Prce constrants Prce lmts may be attached to an order to protect aganst adverse prce movements n the market between the tme the order s sent and the match tme. The constrant smply ndcates whether your order s avalable for the match pool, but does not generate any external nformaton. The only prce at whch a match can occur n POSIT s at the md-pont of the current bd/offer spread. For example, f a prce lmt on a buy order s 450 and the rulng md prce s 45, then the order does not partcpate n the match. If the md prce was 448, then the order would be ncluded n the match and executed at 448, but not at 450. Ths constrant s at ndvdual stock levels and may only be appled n the currency of the prmary prce quote. Mnmum shares The ablty to set a mnmum number of shares to trade out of a total order sze per stock or for all stocks n a portfolo or lst s avalable n POSIT. Clents may wsh to receve no flls of less than 5,000 across a lst for example. For a partcularly large order, of say mllon shares, a mnmum fll of 50,000 n that lne alone may be requred. Mnmum value Lke the mnmum share constrant above, t may be approprate, partcularly for a lst of stock wth varyng prces, to set a mnmum value, of say 50,000 per stock. Ths constrant can be set n any currency that s held on the system, such as Brtsh Pounds, US Dollars, Euros, etc. Cash mbalance For a portfolo, lst or par of orders, a cash mbalance constrant s avalable relatng to the maxmum amount that buy orders can exceed sell orders or vce versa, both n absolute value terms and by shares, f approprate. Ths constrant s partcularly useful for (a) generally n managng the results of unpredctable match outcomes, (b) ensurng that avalable cash for nvestment as a result of, say, a restructurng s not compromsed but overbuyng or (c) ensurng that any necessary cash s rased, where requred. For example, the constrant may be that buy orders on a lst cannot come nto operaton untl 5m of sales has been acheved. 3