Vertical Integration, Competition, and Financial Exchanges: Is there Grain in the Silo?

Transcription

1 Vertcal Integraton, Competton, and Fnancal Exchanges: Is there Gran n the Slo? Steffen Juranek and Uwe Walz Ths verson: August 12, 2012 Abstract We nvestgate the ncentves for vertcal or horzontal ntegraton n the fnancal securtes servce ndustry. Our analyss shows that the ncentves for vertcal ntegraton depend on ndustry characterstcs revealng, for example, that ncentves ncrease n the demand for lqudty. Decentralzed decsons mght lead to an ndustry equlbrum wth lower aggregate profts for all partcpants. Ths coordnaton problem can be overcome by horzontal ntegraton. By lnkng our results to recent regulatory and nsttutonal developments we argue that multlateral tradng facltes and algorthmc tradng should lead to less vertcal ntegraton, whle the regulatory push towards brngng over-the-counter trades on organzed exchanges has the opposte effect. Keywords: vertcal ntegraton, horzontal ntegraton, competton, tradng, settlement JEL classfcaton: G15, L13, L22 We are grateful to Ka-Olver Maurer and Martn Reck (Deutsche Boerse) as well as ther teams for very helpful conversatons on the structure of the ndustry. Furthermore, we would lke to thank partcpants of the CFS round table on the economcs of exchanges for comments on an earler verson of the paper. The paper beneftted sgnfcantly from comments and suggeston by semnar partcpants at the CFS conference on the Industral Organzaton of Securtes Markets 2010, the EEA Meetng 2010, the Veren für Socalpoltk Annual Conference 2010, the EFMA 2011 Annual Meetng, the Campus for Fnance Research Conference 2012 and the House of Fnance Semnar Seres n Frankfurt. Goethe Unversty Frankfurt Address: Goethe Unversty Frankfurt, Grueneburgplatz1, 60323Frankfurt am Man, Germany. Tel: +49 (0) E-mal: juranek@ww.un-frankfurt.de Goethe Unversty Frankfurt and Center for Fnancal Studes. Address: Goethe Unversty Frankfurt, Grueneburgplatz 1, Frankfurt am Man, Germany. Tel: +49 (0) E-mal: uwe.walz@ww.un-frankfurt.de

2 1 Introducton The securtes exchanges and central securtes depostores (CSDs) that are at the center of modern captal markets are confronted wth what seems to be a permanent debate about ther organzatonal structure (see e.g. FESE (2010)) and wth attempts to change ths organzatonal structure. Promnent examples of such attempts are the recent horzontal merger of NYSE Euronext and Deutsche Boerse, whch faled because of anttrust concerns as well as the planned merger of Deutsche Boerse wth Euronext n The man reason the Deutsche Boerse/Euronext merger was not fnalzed n 2008 was strong doubts of Deutsche Boerse s nvestors (hedgefunds TCI and Attcus) concernng the vertcally ntegrated structure of Deutsche Boerse, whch led the nvestors to block the merger. They argued that t would be more valuable to splt up Deutsche Boerse nto one tradng and one post-tradng company and merge these frms on a horzontal level. Untl now, across contnents and markets, we observe very dfferent ndustry structures. On the one hand, we fnd a much more fragmented structure n Europe wth roughly 20 exchanges and CSDs (see FESE (2010)) compared to the U.S. market, whch has only a dozen exchanges and two CSDs. On the other hand, the degree of vertcal ntegraton dffers sgnfcantly among European countres and markets. For example, we see strong vertcal ntegraton n Germany and less vertcal ntegraton n other markets such as the Unted Kngdom. Aganst ths background, we provde nsghts on the dynamcs of the ndustral organzatons of the fnancal securtes servce ndustry, most notably on the nterrelaton of organzatonal desgn and market structure. Our man research questons focus on examnng under what crcumstances vertcal ntegraton s proftable, and when horzontal ntegraton outcompetes vertcal ntegraton from the vewpont of both frms and socety. We further ask for the mplcatons of current ndustry and regulatory trends on the organzatonal desgn of the fnancal securtes servce ndustry. We use these general research questons to nvestgate the drvers behnd vertcal and horzontal ntegraton n the fnancal securtes servce ndustry. We show that due to negatve externaltes on compettors, ndvdual ncentves to vertcally ntegrate may lead to a full vertcal ntegraton equlbrum, whch s assocated wth lower frm profts. We fnd that horzontal ntegraton domnates n such stuatons, leadng to horzontal rather than vertcal ntegraton n these cases. Ths result gves a further justfcaton to the TARGET2-Securtes ntatve of the Euroystem whch tres to mplement compatble systems among CSDs. On the bass of our general analyss we derve testable hypotheses that are drected towards major developments n the ndustry. Our analyss leads us to expect that the emergence of multlateral tradng facltes wll lower the ncentves to vertcally ntegrate. 2

3 We conjecture that the same s true wth respect to the growth n algorthmc tradng. Because algorthmc tradng provdes lqudty to exchanges and hence reduces the lqudty advantage of home markets, the ncentves to trade and settle n a vertcal slo dmnsh. Also, more over-the-counter (OTC) tradng s assocated wth fewer ncentves to buld up slos. To derve these results, we use a stylzed model that depcts the nterrelaton between the organzatonal desgn of fnancal securtes servce provders and the competton among them. The model ncorporates economes of scope and network effects at the dfferent levels of the value chan of the fnancal securtes servce ndustry. We assume that vertcal ntegraton makes t possble to mplement a hghly specfc relaton between an exchange and the assocated CSD. Although ths hghly specfc lnk makes trades routed through t less costly, t mposes addtonal costs to trades that are ether settled outsde the assocated CSD or traded on another exchange but settled n the assocated CSD. It also resembles the dea of vertcal ntegraton as a decson for a closed rather than an open standard that makes external lnkages partally ncompatble wth nternal processes. Hence, our nterpretaton of vertcal ntegraton s on the one hand n lne wth the nformaton we have receved from many ndustry experts n the course of a number of ntervews and dscussons, and on the other hand conforms wth the basc arguments from the theory of frm lterature n the tradton of Grossman and Hart (1986). To our knowledge we are the frst usng such an approach n dscussng the ncentves for vertcal ntegraton n network ndustres. Our paper s related to several strands of the lterature on the fnancal securtes servce ndustry. Frst, our paper touches on the topc of competton between exchanges. Ths topc s analyzed n dfferent manners by, e.g., Foucault and Parlour (2004), D Noa (2002), and Shy and Tarkka (2001) where Shy and Tarkka (2001) also nvolve a vertcal relaton between the brokers and stock exchanges. But ther paper centers on the role of allances between stock exchanges,.e., they examne cooperaton on a horzontal level. Second, we apply analoges to the queston of nterlnkng securtes settlement systems as s analyzed by Kauko (2004) and Kauko (2007). Thrd, our work s drectly related to the lterature on vertcal ntegraton n the fnancal securtes servce ndustry. Köppl and Monnet (2007) analyze the role of prvate nformaton n a merger between a stock exchange and a settlement provder and show that vertcal slos can prevent effcent consoldaton on a horzontal level. In contrast, Holthausen and Tapkng (2007) and Rochet (2005) model the vertcal relaton between custodan banks and a CSD. In Holthausen and Tapkng (2007) the CSD s both nput provder and compettor. These authors show that the CSD leverages ts monopoly power to compete for customers at the custodan level by rasng t rvals costs. Rochet (2005) asks f a CSD should be allowed to ntegrate 3

4 vertcally wth custodan banks. Rochet (2005) concludes that the welfare effect of such a merger hnges on the trade-off between effcency gans and lower competton on the custodan level that are due to the merger. Ths trade-off s also at the center of attenton n our paper as well, but we focus on a dfferent aspect of the value chan, one whch nvolves very dfferent economc mechansms. The paper closest to our analyss s Tapkng and Yang (2006). These authors analyze dfferent ndustry settngs n the sense of vertcal or horzontal ntegraton n a two-country model. They conclude that from a socal perspectve horzontal ntegraton domnates vertcal ntegraton, whch tself s better than no consoldaton. Tapkng and Yang (2006) ask... under whch condtons a merger of CSDs s welfare mprovng and do not dscuss whether under these condtons, market forces would ndeed result n a merger (Tapkng and Yang (2006), p. 1772). In contrast, we nvestgate the ssue of vertcal and horzontal ntegraton from the vew pont of the exchange,.e., prvate frms and ther nvestors. The paper s organzed as follows. In secton 2 we shortly outlne the structure of the ndustry and present the basc model. Usng ths basc model, n secton 3 we dscuss the ncentves and consequences of vertcal ntegraton and derve further the equlbrum of the vertcal ntegraton decson. In secton 4, we compare vertcal wth horzontal ntegraton. In secton 5 we derve several testable hypotheses from our model, thereby mergng our analyss wth recent key ndustry developments. In secton 6 we conclude. 2 The basc model Before we outlne the basc model, we present the characterstcs of the fnancal securtes servce ndustry on whch we buld our model. We descrbe the relevant functons needed to process a trade and the man provders of these processes along the value chan. 2.1 Insttutonal background The securtes transacton process s bascally characterzed by three functons. The frst functon s the actual tradng process, e.g., the matchng of buyer and seller whch usually takes place on the exchanges, alternatve tradng platforms, or va over-the-counter oneto-one tradng. Snce traders want to mnmze the nfluence of ther orders on the prce, lqudty matters a lot at ths stage. The second functon s the clearng process. In ths process, the b-/multlateral oblgatons are calculated by the clearng house, whch n recent years has more frequently nvolved a Central Counterparty (CCP). Usually the clearng house s owned by the exchange. The thrd functon s the settlement process n whch transactons are completed and 4

5 the cash and securtes are transferred. Ths servce s offered for each partcular securty by monopolstc central securtes depostores (CSDs) that hold the securty and allow transactons by book entry. Snce systems can be used for dfferent securtes, and cash settlements mght be netted, economes of scope are present at ths stage. Besde CSDs also custodan banks offer these servces and take the role of an ntermedary. They usually have an account at the man CSDs that allows ther customers to trade securtes kept at dfferent CSDs (usually dfferent countres) va one account. Furthermore the CSDs offer safe-keepng for securtes, e.g., the dstrbuton of nformaton by the securty ssuer, dvdend flow, etc. The safe-keepng s needed to allow transactons but, unlke other processes, t s not necessarly nvolved n every transacton. If an entty owns the provders of all three transacton servces, then t s descrbed as a vertcally ntegrated exchange or a slo. Gven that we ncorporate ndustry-specfc characterstcs such as the nature of network effects or the monopoly poston of the CSDs, we vew our model as an analyss of the securtes tradng ndustry rather than as a general analyss of the pros and cons of vertcal and horzontal ntegraton n network ndustres n general. There are very few other papers dscussng (vertcal or horzontal) ntegraton decsons n network ndustres. Exceptons to ths are Church and Gandal (2000) and Dogan (2009) who focus on compatblty between networks on the demand sde whle we consder compatblty on the supply sde. 2.2 The model setup We consder a settng n whch three exchanges ( = A,B,C) compete wth each other. Besdes the three exchanges, there exst three CSDs (j = A,B,C). These CSDs may or may not be vertcally ntegrated wth the exchanges. We use a framework wth three exchanges/csds to allow for non-covered markets. Dong so enables us to analyze the effect of vertcal or horzontal ntegraton on the level of total lqudty of a securty, a basc ngredent of our model. Exchanges and CSDs: Clearng servces are provded by the exchanges and therefore are not consdered separately. The costs to trade one unt of a securty are dentcal across all three exchanges and are denoted by c T, the cost of settlement for a trade orgnated on exchange j and settled n CSD s c S j (,j = A,B,C). A securty that s lsted on a partcular exchange (for nstance, A) s kept n the respectve CSD (n ths case, A), mplyng that a gven securty can be traded on dfferent exchanges but settled n only one CSD. Ths fact gves CSDs monopoly power for a partcular securty. We assume perfect competton between custodan banks and therefore do not consder them n our analyss. 5

6 Introducng them nto our model and allowng them to provde lnks to dfferent CSDs would not alter our qualtatve analyss. We normalze the total number of securtes to one. We allow for a symmetrc setup, settng the number of securtes lsted on ether exchange to one thrd. Traders are unformly dstrbuted on the permeter of a crcle wth a length equal to one and a densty equal to one. All traders demand nelastcally one unt of each securty lsted on any of the three exchanges. The reservaton prce for all traders for tradng and settlement servces s denoted by V. Ths reservaton prce ncludes the prce of the securty traded. Because we are only nterested n the overall number of trades rather than the blateral relaton between seller and buyer, we assume that ths reservaton prce s dentcal for all traders. The three exchanges and the correspondng CSDs are symmetrcally located on the permeter of the crcle at zero, one thrd and two thrds (see Fgure 1). Our analyss s based on the dea that the market s not fully covered. Hence, we leave room for market coverage effects n order to allow vertcal ntegraton to affect tradng volume. We also avod havng CSDs face a prce-nelastc demand wth all the specal features of such a specfc demand curve. Our no-full-coverage assumpton s n lne wth the clearly observed home bas (see, e.g., Tesar and Werner (2008)) under whch nvestors focus more heavly on local securtes, for example by concentratng on the (perceved) costs of prce dsperson and the (perceved) nformatonal advantages from buyng local assets. In the absence of fully covered markets, nvestors are more lkely to trade local securtes compared to securtes lsted on other exchanges. Although CSDs can prce-dscrmnate between trades orgnated at dfferent exchanges, exchanges can only charge one prce for securtes kept at dfferent CSDs. Ths assumpton s not only justfable n relaton to current practce but also because regulatons make t very dffcult for exchanges to dscrmnate aganst nvestors on the bass of ther settlement provder. In contrast, CSDs can observe trades orgnated on dfferent exchanges and hnder arbtrage from occurrng: hence, they are able to prce-dscrmnate. Although the no-dscrmnaton assumpton for exchanges s not decsve for our analyss, the ablty of CSDs to prce-dscrmnate s mportant for our argument for vertcal ntegraton. But ths assumpton s even more natural, snce t only requres that CSDs can charge dfferent prces vs-a-vs ther respectve exchanges as compared to arm s length transacton wth the competng exchange. We denote the prce of CSD for trades takng place on exchange j by p S j, pt stands for the prce of exchange. Traders: The further away a trader s located from the exchange on whch he s actually tradng, the hgher s the dsutlty he realzes from the trade. Suppose a trader s located at x and trades on exchange. We defne the closest dstance between the 6

7 x A 3 x A 4 B C x A 1 x A 2 0;1 A Fgure 1: Industry structure for a securty lsted on A trader and the exchange as gx. Then, the trader realzes a dsutlty of tg x where t denotes the degree of dfferentaton across the exchanges. Ths dsutlty term reflects the dea that there are dfferences across exchanges that stem from locatonal dfferences, such as language, regulaton, and the lke. The more pronounced these dfferences are, the larger s t. We nterpret ths parameter t as the degree of fnancal market ntegraton. We note that these dfferences usually take the form of dscrete steps. Our contnuous set-up reflects the fact that these features have dfferent levels of mportance for dfferent knds of traders (nsttutonal, prvate, hgh-frequency). These varyng perceptons of the dfferences are taken nto account by a contnuous functon, whch keeps the model also tractable. We assume that traders have to pay both prces and expect the exchange on whch the securty s lsted as beng the more lqud one. The latter assumpton ncreases the utlty of traders tradng on the home exchange by k. Ths assumpton s n lne wth the observaton of Hallng, Pagano, Randl, and Zechner (2008) that the lqudty of a stock s usually concentrated on the stock exchange where the company has ts prmary lstng. In the followng, we refer to k as the lqudty parameter. We are aware that for a full understandng of the competton between exchanges the role of the lqudty should be endogenous. But snce we are nterested n the organzaton decson we beleve that ths short-cut s suffcent. Therefore, we can state the utlty of a trader beng located at pont x on the crcle who consders buyng one unt of a securty that s lsted on exchange A as: { Ux A = V p T A ps AA +k tga x f tradng takes place on exchange A (1) V p T j p S Aj tgj x f tradng takes place on exchange j (j = B,C) 7

8 If securtes are lsted on exchanges B or C the correspondng utlty functons apply. Demand: Therefore, we can derve the total demand of the exchange on whch the securty s lsted (say, A) as the sum of the two margnal traders (x A 1 and 1 xa 2, see fgure 1) beng located between ths exchange, and the two exchanges wth whch t competes (B and C). We derve total demand for the two remanng exchanges from the sum of the respectve demand accrung to these exchanges when competng wth exchange A ( 1 3 xa 1 for exchange B and x A 2 2 for exchange C) and the respectve demand that arses from 3 the margnal traders on exchanges B and C, who are ndfferent between buyng or not buyng at all (x A for exchange B and 2 3 xa 4 backyard). for exchange C, to whch we refer as the Dervng the margnal traders yelds the followng demand functons: d A AB = x A 1 = p T B pt A +ps AB ps AA +k t 2t d A AC = 1 xa 2 = pt C pt A +ps AC ps AA +k t 2t d A BA = 1 3 xa 1 = pt A pt B +ps AA ps AB k t 2t d A CA = x A = pt A pt C +ps AA ps AC k t 2t d A BB = xa = V pt B ps AB t d A CC = 2 3 xa 4 = V pt C ps AC, (7) t wth d A j denotng the demand for trades on exchange of securtes lsted on exchange A when competng wth exchange j. The total demand for trades on each exchange for a securty lsted on exchange A emerges as d A A = da AB + da AC, da B = da BA + da BB, and da C = da CA + da CC. If tradng for a securty takes place on exchange B or C, then we can derve demand functons by replacng A wth the respectve exchange on whch the securty s lsted. Hence, we can state the proft functon of the exchanges as: (2) (3) (4) (5) (6) π T = 1 3 (pt c T ) j d j, (8) and that of the CSDs as: π S = 1 (p S j 3 cs j )d j. (9) j We mpose an assumpton that ensures that the market s not fully covered whch avods havng aggregate demand be completely prce-nelastc. Two further assumptons on our 8

9 parameter range mply that standard Salop-type competton occurs between the exchanges. For a securty lsted on exchange the frst assumpton requres that x 3 < x 4 whch s stated n the frst part of Assumpton 1. The frst regularty condton for Salop-type competton requres that the competton between exchanges takes place for the margnal trader who s located n between these exchanges,.e., x 1 < 1/3 and 2/3 < x A 2 < 1 (part 1 of Assumpton 2). Furthermore, our second regulatory condton requres that as usual n Salop-models the surplus generated by a trade needs to be postve for the margnal trader beng located between the exchanges (see parts 2 of Assumptons 1 and 2). We mpose (see for the detaled dervaton Appendx A.3): Assumpton 1 and Assumpton t > v > 9 8 t 5 8 k, t > k > 9 5 t 8 5 v. In Assumpton 1 and 2, v = V c S c T represents the net socal reservaton prce or the gans from the trade. Snce n what follows we consder only very small devatons from our symmetrc cost settng, we need not consder dfferent v j. 3 Vertcal ntegraton n the tradng ndustry We start our analyss by frst ntroducng our noton of vertcal ntegraton. We defne vertcal ntegraton as a process that allows specfc adjustments between the respectve tradng and settlement processes (e.g., establshng more effcent straght-through processng), and faster coordnaton n the vertcally ntegrated organzaton as compared to arm s length transactons. Vertcal ntegraton allows for specfc nvestments between tradng and settlement, most notably n the area of software and IT processes. In the absence of vertcal ntegraton, such specfc nvestments mght lead to severe hold-up problems between the two partes nvolved. The specfc nvestments te together exchange A and CSD A. However, ths closer lnk between the two comes at a cost: t makes the nteracton of exchange A wth CSD B and C as well as the nteracton of CSD A wth exchange B and C more costly, because t becomes more dffcult for them to route trades of securtes not lsted on exchange A. Hence, vertcal ntegraton resembles a closed standard wth bascally a partally ncomplete technology. The effcency of the standard ncreases but the nteracton wth agents 9

10 outsde the standard becomes more dffcult (see, e.g., Shy (2001)). We depct ths concept as follows: wth the vertcal ntegraton of settlement and tradng n A, trades on A are settled at lower costs n A (c S AA = cs y) but all cross-routngs become more costly (c S Aj = cs ja = cs + y). In both expressons y denotes the effcency parameter assocated wth vertcal ntegraton. In Fgure 2, we llustrate ths entre process of vertcal ntegraton, whch creates a more effcent lnk between settlement n A and tradng n A but hgher costs for the cross-lnks. S B c S c S S C c S +y c S T B T C c S c S +y T A c S y c S +y c S +y S A Fgure 2: Vertcal Integraton of A only We analyze these changes n effcency n the nteracton between exchanges and settlement organzatons. Settlement and tradng prces n the vertcally ntegrated organzaton are set separately by the exchanges and CSDs. We delberately do not examne the beneft of vertcal ntegraton va a combned prce-settng n our set-up. We do not take nto account the nternalzaton of the external effect of the prcng decson of the tradng entty and the CSD, the double margnalzaton effect, as well as the possblty to use prces as a strategc factors aganst compettors. From our pont of vew ths lacuna s justfed by three arguments. Frst, because the mplementaton of an ntegrated decson process requres a proper transfer prcng system, whch s often qute cumbersome. Second, because monopolsts such as the CSDs usually attract the attenton of regulators and are therefore not able to capture the profts assocated wth the strategc effects va the 10

11 prce-settng mechansms ncludng foreclosure. Thrd, and most mportant, the effects of the nternalzaton process are obvous and very well nvestgated (see, e.g., Trole (1988)). The effects clearly favor vertcal ntegraton. Thus, by gnorng ths effect we bas aganst vertcal ntegraton. We note that readers should keep ths fact n mnd when nterpretng our results. The sequence of moves comprses two stages. In the frst stage, the CSDs smultaneously decde whether to vertcally ntegrate wth ther respectve exchange or stay unntegrated. In the second stage, gven ther ntegraton decson CSDs and exchanges decde on ther proft-maxmzng prces and traders decde where to trade. An alternatve nterpretaton of our frst stage, whch also leads to the same results, s that only one CSD decdes whle facng compettors that are already ntegrated to a dfferent extent. Ths nterpretaton reflects dfferent hstorcal precondtons compared to expectatons drvng potentally our results n the man analyss. We wll come back to ths alternatve, sequental nterpretaton later. In the followng, we solve for subgame perfect Nash equlbra. t 1 CSDs decde upon vertcal ntegraton t 2 CSDs and exchanges set prces t Fgure 3: Tmng of the game 3.1 Prcng decsons We derve the proft-maxmzng tradng and settlement prces for exchanges and CSDs from Eqs. (8) and (9). Dong so gves us the subsequent reacton functons for,j,l = {A,B,C} and j l : p T = 1 2 ct t+ 1 4 V (pt j +p T l ) 1 8 ps 3 16 (ps j +p S l) (ps j +p S l +p S jj +p S ll), (10) p S = 1 2 cs k t 1 2 pt (pt j +p T l )+ 1 2 (ps j +p S l) 1 4 (cs j +c S l), (11) p S j = 1 2 cs j 1 6 k t+ 1 6 pt 1 2 pt j ps 1 6 cs + 1 V. (12) 3 We note three dfferent mechansms n these reacton functons. Frst, prces of the correspondng up- or downstream actvtes are strategc substtutes (see Bulow, Geanakoplos, and Klemperer (1985)),.e. p T / ps j < 0 j and ps j / pt j < 0 j. Hence, prce ncreases by the settlement provder (for trades orgnated at ether exchange) nduce the exchanges 11

12 to lower ther prces strategcally. Second, the prces of the compettors on the tradng level are strategc complements,.e., p T / pt j > 0; j. Increases n the prces of the compettor lead to ncreases n the prces of the other exchanges. Ths pattern shows the conventonal feature of prce competton wth mperfect substtutes. Thrd, we note that because exchanges cannot prce dscrmnate, they react strategcally to settlement prce changes even f ther traders do not have to ncur these settlement prces. We get p T / p S j > 0, for example, exchange A ncreases ts prce f CSDs ncrease ther prces for settlements orgnated onexchange B or C. On the settlement level, the CSDs do not compete wth each other at all but the prces set to the traders on the dfferent exchanges do nteract wth each other. All these nteractons are decsve n our analyss of the vertcal ntegraton process. We solve the above system of prces smultaneously (see Appendx A.1 for detals) and fnd for vertcal ntegraton of exchange and CSD A only (see Appendx A.1.1): Lemma 1 Vertcal ntegraton leads to a decrease n all tradng prces. Ths effect s less pronounced n the ntegrated exchange A compared to the unntegrated exchanges B and C. Wth settlement, only the servces provded va the drect, more effcent lnk become cheaper. All other settlement servces become more expensve. The somewhat surprsng result of the effect of vertcal ntegraton on tradng prces stems from the fact that drect tradng and settlement prces are strategc substtutes (see Eq.(10)): hgher settlement prces lead exchanges to reduce ther tradng prces. Hence, exchanges B and C, whch face hgher settlement prces for securtes lsted on A, have an ncentve to reduce prces. Gven that tradng prces are strategc complements, ths reducton trggers a decrease n A s tradng prce. Ths effect s even renforced by the ncrease n the other settlement prces that leads to a decrease n exchange A s tradng prce as well. A further channel through whch vertcal ntegraton affects the payoffs of all agents s the mpact of vertcal ntegraton on traders behavor and market coverage. Compared to non-ntegraton, we fnd for ntegraton of exchange and CSD A only (see Appendx A.2): Lemma 2 () Market coverage of securtes lsted on all exchanges decrease. () The vertcally ntegrated exchange A wns trades vs-a-vs exchanges B and C n securtes lsted on A whle losng trades for securtes lsted on B and C. The decreased coverage of the market noted n the frst part of Lemma 2 results from an ncrease n the sum of tradng and settlement prces, whch traders located n the 12

13 backyard of B have to pay. The result of the second part of Lemma 2 s due to the fact that va vertcal ntegraton, cross-platform lnks become more costly, thereby makng the respectve home exchange more compettve. 3.2 Decsons to ntegrate vertcally We now turn to the analyss of the ntegraton decsons. In order to reach an equlbrum, we analyze frst the ncentves of a sngle CSD to ntegrate vertcally wth ts respectve exchange gven that the other CSDs are expected not to ntegrate. In a second step we address the ncentves to vertcally ntegrate f the other CSDs are expected to ntegrate vertcally; ether one or both of them. At last, we derve on the bass of the results of the frst two steps the resultng equlbra of the vertcal ntegraton stage The stand-alone case Gven that CSD B and C stay unntegrated we use our fndngs on prces and quanttes to derve the proft dfference of the sum of profts n tradng and settlement n A between ntegraton and no ntegraton. We defne the resultng dfference as y π(0)(see Appendx A.5.2): (π T,A A +πs,a A ) (πt A +πs A ) = y198900k+8476t v y t Vertcal ntegraton f other CSDs are expected to ntegrate y π(0). (13) We examne frst the ncentves for CSD A to vertcally ntegrate f only one of the other CSDs (say B) s also expected to do so. We derve the proft dfference π(1) as (see Appendx A.5.3): (π T,AB A +π S,AB A ) (π T,B A +πs,b A ) = y36075k+6097t 17784v+47160y t y π(1). (14) If both competng CSDs are expected to ntegrate, then we fnd as proft dfference π(2) (see Appendx A.5.4): (π T,ABC A +π S,ABC A ) (π T,AB A +π S,AB A ) = 3588k +1248v+1612t+4143y y y π(2). (15) 24336t We note that the parameter v has an opposte effect n Eqs. (13) and (15). The negatve mpact of v on π(0) reflects the fact expressed n Lemma 2 that vertcal ntegraton leads to a decrease n market coverage. Ths proft-decreasng effect s most pronounced f the market coverage s large,.e., f v s large. In the step towards full ntegraton (.e., antcpatng a B- and a C-slo), vertcal ntegraton only leads to lower costs, and hence, to lower prces. These lower prces allow the exchange to reach more traders n the backyard f v s large, hence ncreasng the proft gans π(2) of the vertcally ntegratng CSD A. 13

14 We fnd (see Appendx A.6 for detals): Lemma 3 π(0) < π(1) < π(2) > 0. Vertcal ntegraton s more attractve the more other CSDs are expected to vertcally ntegrate compared to a stuaton n whch the other CSDs are expected to stay unntegrated. The explanaton of Lemma 3 s that the ntegraton of one CSD (say, B) mples that the costs for A that come from ts own ntegraton decrease. After the ntegraton of B, A faces a compettor wth an effcent lnk; thus makng t more attractve to establsh an effcent lnk on ts own as well. After B s ntegraton, there s already an neffcent lnk between A and B, thereby elmnatng these addtonal costs for A s ntegraton. Takng these two effects together makes A s ntegraton more attractve after B s ntegraton, compared to the stuaton n whch A goes for a head start n vertcally ntegratng. Ths effect s even more pronounced f not only B, but both B and C are already ntegrated The vertcal ntegraton equlbra We use Lemma 3 to deduce the resultng equlbra of our ntegraton decson stage. Because π(2) s always postve, the n-between case wth only two CSDs optng for vertcal ntegraton can never be an equlbrum. The same s true for a settng n whch only one CSD decdes for vertcal ntegraton, snce dong so would requre that π(0) > 0 and π(1) < 0, an obvous contradcton to Lemma 3 above. Because of our symmetrc settng, only two equlbrum canddates reman: ether all CSDs stay unntegrated or all of them ntegrate vertcally. Although ntegraton of all CSDs always consttutes an equlbrum n the ntegraton-decson stage (snce π(2) > 0), non-ntegraton s only an equlbrum f π(0) < 0. The remanng queston s whether ths equlbrum s ndeed feasble. Fgure 4 addresses ths queston. Fgure 4 dsplays Eq. (13) and Assumptons 1 and 2 on the parameters (the grey area s not compatble wth these assumptons), n k v space. The fgure shows that vertcal t t ntegraton can ndeed pay off. The whte range n the fgure dsplays the parameter combnatons that are not only feasble, but that also ncreases the sum of profts of A wth vertcal ntegraton even f the margnal effcency effect s evaluated at y = 0,.e., π(0) > 0. In the black regon, ntegraton does not pay off. However, t turns out, that our full ntegraton equlbrum mght lead to lower payoffs for all CSDs,.e., we are ether facng a prsoner s dlemma outcome n a settng n whch the full ntegraton equlbrum s unque, or a potental coordnaton problem wth two equlbra. In both cases, CSDs may end up beng fully ntegrated but wth lower profts compared to the non-ntegraton case. These problems become obvous when we derve the dfference n profts between a stuaton n whch no CSD s ntegrated to a stuaton n whch all CSDs are ntegrated 14

15 Fgure 4: Effect of vertcal ntegraton on proft of A (see Appendx A.5.4): (π T,ABC +π S,ABC ) (π T +π S ) = y75k+t 72v+111y 225t yγ = A,B,C. (16) In Appendx A.7 we show that the parameter set for whch π(0) < 0 holds s a subset of parameter set for whch Γ < 0 prevals,.e., whenever π(0) < 0 we get Γ < 0. Gven the results, we can state: Proposton 1 In equlbrum ether all CSDs decde to ntegrate, or none of them does so. Full ntegraton s always an equlbrum, but non-ntegraton only consttutes an equlbrum f and only f π(0) < 0. Hence, we fnd three potental settngs dependng on parameter settng 1. Wth π(0) > 0 and Γ > 0 full ntegraton s the only equlbrum. CSDs profts ncrease wth ntegraton. 2. Wth π(0) > 0 and Γ < 0 full ntegraton s the only equlbrum. CSDs profts decrease wth ntegraton (prsoner s dlemma case). 3. Wth π(0 < 0 (and henceγ < 0) full ntegraton as well as non-ntegraton are both equlbra. CSDs profts decrease wth ntegraton (coordnaton problem case). 15

16 Fgure 5 llustrates the dfferent equlbra settngs. Above the black area π(0) > 0, and below the whte one Γ < 0 prevals. Hence, n the whte area case 1 prevals. In the grey area we end up wth case 2, the prsoner s dlemma. In the black area case 3 wth a multplcty of equlbra emerges wth the possblty of a coordnaton falure leadng to the bad full ntegraton equlbrum. Cases 2 and 3 result from pronounced negatve externaltes that lead to the possblty that ndvdual actons produce outcomes that lower the profts of all players nvolved. Case 3 becomes most lkely f ether exchange has a large backyard. Pronounced gans from trade,.e., large v, mply that an overproportonal part of trades take place va the neffcent lnk, makng unlateral devaton from the non-ntegraton settng not attractve. Ths case s especally true f the securtes lsted on the dfferent exchanges are mperfect substtutes, makng them attractve for foregn traders, e.g., for dversfcaton purposes. If we nterpret our settng n a sequental manner, n whch one after the other CSD decdes, then hstory matters. Gven that hstory matters wth case 3, ntal startng ponts matter, makng t dffcult to make a clear-cut predcton, ndependent of the startng pont. Ths result has an mportant mplcaton for the organzaton structure among European countres. Snce the fnancal securtes ndustry was orgnally organzed along natonal lnes, exchanges and CSDs n the ndvdual countres have often ntegrated ther systems. Hence, wth the creaton of a sngle European market, because of the coordnaton problem the ndustry may be stuck wth vertcal ntegraton, even though separaton would be more proftable. Fgure 5: Effect of vertcal ntegraton of all fnancal securtes servce provder on profts 16

17 What the CSDs see as a potentally excessve degree of vertcal ntegraton happens because vertcal ntegraton comes wth negatve externaltes mposed on the other CSDs and exchanges. We outlne ths mechansm n more detal below. 3.3 Comparatve statcs From our analyss so far we know that the decsve factor n whether non-ntegraton may result n equlbrum s that π(0) < 0. Usng Eq. (13), we compute comparatve statc effects. We fnd: Proposton 2 Full vertcal ntegraton s more lkely to be the only equlbrum f demand for lqudty s hgh ( ( π(0))/ k > 0), and the gans from trade are low ( ( π(0))/ v < 0). The effect of more ntegrated fnancal markets s ambguous: f the lqudty effect k s relatvely large compared to v, then a hgher degree of ntegraton ncreases the proftablty of vertcal ntegraton and vce versa for a relatvely small k and y. The thnkng behnd these fndngs s that the more lqudty matters, the hgher s the share of trades on exchange A of securtes kept safe n CSD A usng the effcent lnk. Hence, the cost advantage of vertcal ntegraton s most pronounced. Hgher gans from trade (larger v) lead to more trades on exchange B of securtes kept safe n A and vce versa. These trades are settled through the neffcent lnk after vertcal ntegraton. Thus, snce these trades ncrease n absolute and relatve terms wth hgher gans from trade, hgher gans from trade make vertcal ntegraton less attractve. The effect of less ntegrated fnancal markets (hgher t) s ambguous and depends on k and v. If the lqudty effect k s relatvely large compared to the gans from trade, then a hgher t decreases the proftablty of vertcal ntegraton ( ( π(0))/ t > 0 ) and vce versa. The thnkng behnd ths mechansm s that the demand va the neffcent lnks, ether from the ntegrated CSD A to unntegrated exchange B/C or from exchange A to CSD B/C, ncreases n v because more trades are ntated from the backyard. The contrary s true for k, snce fewer trades are ntated on the home market. In contrast, the demand va the effcent lnk s ndependent of v but ncreases n k. An ncreasng t reduces the nfluence of both k and v. Therefore, f k s relatvely large compared to v, an ncreasng t ncreases the demand va the neffcent lnks because t protects exchanges from the lqudty dsadvantage. The demand va the neffcent lnk decreases n t f k s relatvely small because then the loss of traders n the backyards domnates. In contrast, the demand va the effcent lnk always decreases n t. Summng up these effects, relatvely 17

18 more trades are processed over the neffcent lnk and the gans of ntegraton decrease whle the dfferentaton ncreases f k s relatvely large. 3.4 Welfare consequences of vertcal ntegraton Next, we nvestgate the welfare consequences of vertcal ntegraton. We do so by comparng welfare n the non-ntegraton case wth that n the ntegraton case. We thereby employ a broad welfare measure embracng surplus of all traders (TS) plus the profts of all exchanges and CSDs. In equaton (17), we express the welfare effect (see Appendx A.5.4): (TS ABC TS)+3[(π T,ABC +π S,ABC ) (π T +π S )] = y 225k+11t 192v+321y 150t y W. To determne whether prvate ncentves towards vertcal ntegraton accord wth the socal motvatons to acheve full ntegraton, we compare Eqs. (13) and (17). Comparng these two expressons shows that, compared to π(0), welfare ncreases for a larger set of parameters (see Appendx A.8). For a parameter regon vertcal ntegraton s advantageous for socety even f t s the pareto-domnated equlbrum from the vewpont of the exchanges. We get: Proposton 3 1. Wth π(0) > 0, decentralzed decson makng leads to the welfare optmal full ntegraton soluton. 2. Wth π(0) < 0, we can dstngush two subcases: (a) Wth W > 0, due to the coordnaton problem, decentralzed decson makng ether leads to welfare optmal full ntegraton or to too lttle vertcal ntegraton. (b) Wth W < 0, there s potentally too much vertcal ntegraton. There are two facts that explan why wth π(0) < 0 decentralzed decson makng of CSDs may lead potentally to suboptmal decsons. On the one hand, ntegratng unlaterally leads to negatve externaltes on other CSDs and exchanges. Too much vertcal ntegraton mght preval. On the other hand, vertcal ntegraton leads to postve external effects for consumers. Ths s revealed by computng the dfferences n trader surplus between the full ntegraton and the non-ntegraton case (see Appendx A.5.4): (17) TS = y 25k+3t 16v+33y. (18) 50t Snce these external effects are not taken nto account by ndvdual CSDs, too lttle vertcal ntegraton mght occur. 18

19 4 Horzontal ntegraton Gven network economes as well as economes of scale and scope, horzontal ntegraton s an alternatve to vertcal ntegraton that s ntensvely and hotly dscussed n the fnancal press (see, e.g., Economst (2008)). Therefore, we compare horzontal ntegraton wth vertcal ntegraton. We consder horzontal ntegraton only at the settlement level. At ths level, horzontal ntegraton leaves the market structure unchanged, and the settlement provders stay n ther monopolstc poston. Such a stuaton s comparable to the ndustry organzaton n the U.S. wth very few actors on the settlement level. In our framework, the horzontal ntegraton of exchanges for explotng lqudty and network effects leads to a jont exchange that elmnates the lmted access of certan traders to certan stocks, and thus also removes the coverage gap n our base model. Snce the horzontal ntegraton of exchanges changes the market structure at the tradng level, t also changes the entre model structure, whch makes a comparson wth vertcal ntegraton nfeasble. Hence, we consder the case n whch CSDs decde on horzontal ntegraton. To capture the concept of network effects and of economes of scope at the clearng and settlement level, we model horzontal ntegraton as cost savngs of all settlement provders denoted by b y (b > 0). So, whle exchanges costs amount to c T, the costs of all settlement provder are c S by. Gven that the monopolstc settlement provders all experence dentcal cost savngs, ther profts always ncrease. Snce there are only gans and no costs to horzontal ntegraton, ths domnance over the non-ntegraton case s straghtforward. Nonetheless, t s not clear whether ths domnance s true for the comparson wth vertcal ntegraton whch offers hgher cost savngs for the effcent lnk but cost-ncreases for the neffcent ones. We ntroduce the decson about horzontal ntegraton as a stage that occurs before the vertcal ntegraton decson. In ths earler stage CSDs decde about the ntegraton of all CSDs. If they declne to ntegrate horzontally, they may choose smultaneously to ntegrate vertcally n the followng stage and then compete n prces. If there s horzontal ntegraton, then the prce-settng stage follows drectly. t 1 t 2 t 3 CSDs decde upon horzontal ntegraton CSDs and exchanges set prces CSDs decde upon vertcal ntegraton t Fgure 6: Tmng of the game 19

20 4.1 The ntegraton equlbra The CSDs choose horzontal ntegraton whenever t s ndvdually (and because of symmetry, jontly) proftable for them to do so. Hence, CSDs decde to ntegrate horzontally whenever the payoff s larger than the equlbrum payoff of the second stage subgame, whch s ether non-ntegraton or full vertcal ntegraton. By pluggng the proft-maxmzng prces together wth the demand functons n the proft functons, we can derve the proft dfference of horzontal ntegraton compared to non-ntegraton for all settlement provders. Ths proft dfference shows the sum of settlement provders profts after horzontal ntegraton mnus the one before ntegraton (see Appendx A.9): (πa s +3by +πs B +πs C ) = 8by2t+6v 75t > 0. (19) Horzontal ntegraton decreases the costs and s therefore always proftable compared to non-ntegraton. The proft dfference, whch s decsve when we compare horzontal wth vertcal ntegraton, becomes even larger whenever the gans from trade are larger because t ncreases the market sze. The same s true for more ntegrated fnancal markets. More ntegrated fnancal markets (smaller t) mply a larger reach of each securty n foregn markets and hence a larger market sze that can be exploted at lower costs. CSDs wll opt for horzontal ntegraton whenever non-ntegraton s an equlbrum n thesecondstage,.e., π(0) < 0(andhence Γ < 0).Inthatcase, thenon-ntegratonequlbrum s pareto-domnant to the full vertcal ntegraton equlbrum. Gven our above fndngs, horzontal ntegraton always domnates full vertcal ntegraton n that parameter regon. Ths fndng mples that the avalablty of horzontal ntegraton overcomes the coordnaton problem of the vertcal ntegraton case. Furthermore, horzontal ntegraton s also the equlbrum for the parameter regon of the prsoners dlemma,.e., π(0) > 0 and Γ < 0, because the profts are lower wth full ntegraton than wth non-ntegraton. CSDs antcpate ths problem of the second stage game and therefore coordnate on horzontal ntegraton. At last, for π(0) > 0 and Γ > 0, vertcal ntegraton s stll the only equlbrum n the followng subgame and pareto-domnates non-ntegraton. Hence, the equlbrum of the whole game depends on the sze of the cost savngs parameter b. Substractng Eq. (19) from the threefold of (16) yelds: Σ Profts = y 75k+t 72v +111y 8b(2t+6v +3by). (20) 75t We nterpret ths expresson as the ncentves for CSDs to choose horzontal rather than vertcal ntegraton. 20

21 We check the sgn of Eq. (20) and solve for the crtcal b. Dong so just leads to ndfference between vertcal and horzontal ntegraton and yelds (at y = 0): b = 75k +t 72v. (21) 16(t+3v) Hence, b salwayssmallerthanone.wthb = 1,thentegratedCSDfaceseffcent lnksto all exchanges wth horzontal ntegraton. Wth vertcal ntegraton, the ntegratng CSD must trade off the drect effcent lnk for the ndrect neffcent lnk. The cost dsadvantage of the neffcent lnk outweghs the compettve advantage that the other settlement provders compettve advantage and deters them from relyng on the neffcent lnks as well. Thus, horzontal ntegraton domnates vertcal ntegraton for b = 1. The drect followng for the equlbrum decson on horzontal ntegraton s that CSDs decde to ntegrate horzontally whenever the actual b s larger than b, and vce versa. We summarze our fndngs n Proposton 4. Proposton 4 In equlbrum, CSDs wll decde ether to fully ntegrate vertcally or horzontally. Hence, we fnd three potental equlbra dependng on parameter settng: 1. Wth Γ < 0 CSDs choose to ntegrate horzontally, thereby solvng the prsoner s dlemma and coordnaton problem. 2. Wth Γ > 0 and b > b, CSDs choose to ntegrate horzontally. 3. Wth Γ > 0 and b < b, CSDs choose to ntegrate vertcally. Ths result sheds new lght on the TARGET2-Securtes ntatve. The objectve of the Eurosystem s TARGET2-Securtes ntatve s to establsh an IT platform that offers centralzed settlement servces. By connectng all nvolved CSDs, the man am s to reduce the barrers and costs of cross-border trades. In our model, TARGET2-Securtes ft the horzontal ntegraton case. All systems become compatble to each other, and thus economes of scale and scope decrease the costs of settlng trades. In a Memorandum of Understandng (TARGET2-Securtes (2009)) all major European CSDs confrmed ther ntenton to use TARGET2-Securtes once t s n operaton. Hence, the ntatve may be seen as a mechansm for CSDs to coordnate on horzontal ntegraton. Even though t was ntroduced by the Eurosystem, the CSDs have taken up the ntatve n order to avod adverse compettve effects of ther organzaton structures. Therefore, our analyss shows that not only the regulator may have an ncentve to lower the costs of cross-border trades va horzontal ntegraton but that t may be n the nterest of the CSDs as well. 21

22 4.2 Comparatve statcs By usng horzontal ntegraton CSDs can overcome the prsoner s dlemma and the coordnaton problem that mght occur wth vertcal ntegraton. The ntegraton decson crucally depends on the cost savngs parameter b. The comparatve statcs of b are n lne wth the ones on π(0) and we can summarze: Proposton 5 Vertcal ntegraton becomes more attractve for CSDs relatve to horzontal ntegraton f the lqudty effect s more pronounced (larger k), and the gans from trade become smaller (smaller v). The effect of more closely ntegrated fnancal markets (larger t) s ambguous. The attractveness of vertcal ntegraton ncreases n t whenever v > k, and vce versa. The thnkng here s rather smlar to that n Proposton 2. Wth vertcal ntegraton, a more pronounced lqudty effect shfts more trades nto the effcent lnk. Ths shft has a postve effect on the profts of the ntegrated CSD, makng vertcal ntegraton relatve to horzontal ntegraton more attractve. In contrast, wth horzontal ntegraton, all settlement lnks become cheaper to the same extent. Hence, a change n the market shares of the dfferent exchanges does not affect profts n the case of horzontal ntegraton and leaves no room for an effect of k on the ncentve to horzontally ntegrate. Gans from trade mply that relatvely more trades are settled va the neffcent lnks makng vertcal ntegraton less and at the same tme horzontal ntegraton more attractve. The effect of the fnancal market ntegraton parameter t s ambguous. A lower degree of ntegraton (larger t) decreases the proftablty of horzontal ntegraton, because fewer addtonal trader are captured. In contrast, Proposton 2 shows, that f k s relatvely large, then an ncreasng t decreases the proftablty of vertcal ntegraton. In sum, we stll have the ambguty. 4.3 Welfare comparson Here, we evaluate from a welfare perspectve the dfferent ndustry equlbra n ether full horzontal or vertcal ntegraton. We examne whether there s a potental coordnaton effect, askng f t s socally feasble to have fullvertcal ntegraton when prvate ncentves stand n the way such that ntal horzontal ntegraton s preferred or vce-versa. Snce we have already nvestgated the prvate ncentves to reach ether of the two ndustry equlbra n Proposton 4, we evaluate full vertcal ntegraton and horzontal ntegraton from a welfare pont of vew. Computng the relatve gans n welfare accrung 22

23 from full vertcal ntegraton (see Eq. (17)) compared to horzontal ntegraton gves us (see Appendx A.9): Σ Welfare = y 225k+11t 192v+321y 32b(2t+6v+3by). (22) 150t Evaluated at y = 0 ths expresson yelds the crtcal socal b s at whch socety s just ndfferent between full horzontal and full vertcal ntegraton. b s = 225k +11t 192v. (23) 64(t+3v) Next, we compare the prvate ncentves for reachng ether equlbrum wth the relatve advantages of ether equlbrum from a socal pont of vew. When we compare b wth b s we fnd sgn (b b 75k 7t 96v s) = sgn ( ). (24) 64(t+3v) Fgure 7 llustrates ths dfference and dstngushes between the two cases n whch these dfferences are postve (upper part) or negatve (lower part). Excessve vertcal ntegraton {}}{ 0 b s b 1 b 0 b b s 1 }{{} Excessve horzontal ntegraton b Fgure 7: Prvate vs socal ncentves Hence, we fnd: Proposton 6 Socal compared to prvate perspectve: Wth strong (weak) lqudty effects and weak (strong) socal net reservaton prces, the ndustry mght end up n a bad equlbrum n whch settlement provders are vertcally (horzontally) ntegrated. To better explan the economc reasonng behnd ths result, we note that both vertcal and horzontal ntegraton mpose a postve externalty on traders. The externalty of 23