Order Flow in the South: Anatomy of the Brazilian FX Market

Transcription

1 Order Flow in he Souh: Anaomy of he Brazilian FX Marke Thomas Y. Wu ;y Ocober 21, 2005 Absrac This paper conribues o he microsrucure approach o he exchange raes research by aking a closer look on he FX reail marke, he behavior of he cusomer ow and is impac on he exchange rae dynamics. We develop a model ha presens wo major disincions relaive o previous microsrucure models. Firs, cusomer ow is no simply he realizaion of a random variable. Insead, i reacs o changes in he exchange rae. Second, dealers may opimally decide o hold overnigh posiions in he FX marke. Using a Srucural VAR approach, we esimae he predicions of he model using a daase ha covers 100% of he Brazilian FX reail marke. We nd ha, in order o mee a US$ 1 billion cusomer order ow, dealers increase he domesic price of a dollar by approximaely 2.7%. We also nd ha a 1% depreciaion rae decreases he nancial cusomer ow by US$ 111 millions and he commercial ow by US$ 46 millions. I am sincerely graeful for he orienaion and suppor from my advisor Hélène Rey. I would like o hank Afonso Bevilaqua, Ilan Goldfajn, Marcelo Kfoury and Benny Parnes for heir e ors in making he daase available. This work bene ed from many discussions wih Sandro Andrade, Eduardo Casro and Flavio Fucs. I would also like o hank commens and suggesions from Dionísio Dias Carneiro, Gregory Chow, Ricardo Reis, Chrisopher Sims, Noah Williams, and seminar paricipans a he Suden Macro/Inernaional Workshop a Princeon Universiy, he Insiue of Economic Policy Research a he Casa das Garças and he Fifh Annual Trans-Alanic Docoral Conference. Finally, I would like o acknowledge nancial suppor from he Mellon Foundaion Gran in Inernaional Sudies. y Deparmen of Economics, Princeon Universiy. ywu@princeon.edu. 1

2 1 Inroducion Tradiional exchange rae models pu he focus of heir analysis on he rade in goods and asses beween residens and non-residens of a given economy. More speci cally, hey use as building blocks pariy condiions ha avoid large imbalances in hose rades: while he purchasing power pariy guaranees equilibrium in he marke of freely raded goods, he uncovered ineres pariy guaranees equilibrium in he markes of freely raded asses. However, empirical failure of such models has been widely documened, and he behavior of nominal exchange raes has been a major challenge o explain since Meese and Rogo (1983) evidence ha a naïve random walk model would ouperform a variey of models based on macroeconomic fundamenals in erms of ou-of-sample forecas. 1 A new line of research on exchange raes, known as he microsrucure approach, has shifed he aenion from he ow of goods and asses across naional borders o he order ows from he FX marke. Order ow is a measure of buying pressure. Evans and Lyons (2002), he bes represenaive of his new class of models, develop a model ha shows how inerdealer order ow, or buying pressure from he wholesale FX marke, aggregaes and reviews informaion abou cusomer order ow, or buying pressures from he reail FX marke. The empirical resul is quie impressive. They esimae heir model using daily deusche mark/dollar exchange raes and inerdealer order ows covering approximaely 54% of he marke and obain R 2 above 60 percen. They nd ha a US$ 1 billion of ne dollar purchases in he inerdealer marke increases he deusche mark price of a dollar by 0.5%. Mos of he previous sudies on he microsrucure approach are focused on he behavior of he inerdealer order ow. The wo excepions are Fan and Lyons (2003), who analyze ve years of monhly daa covering he euro/dollar and he yen/dollar FX markes, and Bjonnes, Rime and Solheim (2005), wih en years of daily daa on he Swedish krona FX marke. Despie of his research bias, as Fan and Lyons (2003) pu i, he inerdealer order ow is in a sense only a derivaive of he cusomer order ow, which means ha he rue caalys of he FX marke is he acion in he reail marke, and no from he wholesale marke. 2 This paper conribues o microsrucure approach o he exchange raes lieraure by aking a closer look on he FX reail marke, he behavior of he cusomer ow and is impac on he exchange rae dynamics. In order o do so, we use a daase ha is unique in a leas wo ways. Firs, i covers 100% of a counry s FX reail marke. Second, while oher 1 More recenly, Cheung, Chinn and Pascual (2003) reinforced his resul by esing a wider se of exchange rae models. 2 Words in quoes are used in Lyons and Fan (2003). 2

3 sudies are looking a he exchange rae beween currencies from wo developed counries, his is he rs ha covers an emerging economy. Our daase conains ransacion ows from he Brazilian FX reail marke beween dealers and hree ypes of counerparies: commercial cusomers (whose demand for FX is generaed by a rade in goods wih nonresidens), nancial cusomers (whose demand for FX is generaed by a rade in asses wih non-residens), and he cenral bank. The daa is aggregaed by each ype of counerpary on a daily basis, spanning a oal period of 4 years, from he 1s of July of 1999 unil he 30h of June of 2003, and covers 100% of he o cial Brazilian FX marke, one of he larges emerging economies. Moivaed by he observed behavior of he cusomer ow in our daase, we wrie a model wih wo major changes relaive o previous microsrucure models. Firs, cusomer ow is no simply he realizaion of a random variable. Insead, we move owards a general equilibrium model where cusomers demand for FX is in uenced by many macroeconomic fundamenals, including conemporaneous changes in he exchange rae. Second, dealers FX holdings a he end of each rading day are no exogenously se (o zero, for example). Insead, hey may opimally decide o hold overnigh posiions in he FX marke, depending on he expeced overnigh FX payo s: he ineres rae di erenial and he depreciaion rae. This second modi caion implies ha dealers from he Brazilian FX marke do no simply behave as inermediaries, maching buyers wih sellers wihin a rading day: if in a given day he volume ha cusomers need o buy do no mach he volume cusomers need o sell, he dealer may supply he exra liquidiy if he hinks ha i is opimal o do so. In oher words, he FX liquidiy supplied by dealers in our model is no limied o he inraday frequency: hey are also allowed o provide overnigh liquidiy. Our model describes a wo-way relaionship beween cusomer ow and he exchange rae. On he one hand, he need for FX liquidiy is decreasing on is price: he more appreciaed is he exchange rae, he cheaper are foreign goods and asses and higher is he demand for foreign currency. On he oher hand, here is a posiively sloped FX supply curve ha is explained by a porfolio balance e ec. Since dealers are risk averse, hey will charge a risk premium o supply he needed FX liquidiy and end he rading day wih an invenory level lower han iniially desired. This premium akes he form of a price change and he exchange rae depreciaes. In order o esimae his endogenous relaionship beween cusomer ow and exchange rae, we use a Srucural VAR approach, aking advanage of he informaion we have abou he ype of cusomer ha is rading wih in he dealer. The mos imporan idenifying 3

4 assumpion is ha he nancial cusomer ow and he commercial cusomer ow do no a ec each oher conemporaneously. This does no mean ha he commercial cusomer ow and he nancial cusomer ow are no correlaed a he daily frequency. Macroeconomic variables such as he domesic and foreign ineres raes, he counry risk premium and, specially, he exchange rae may a ec boh ypes of ows simulaneously. However, once we conrol for hese poenial sources of common shocks, our model ells us ha he decision o rade goods wih non-residens is independen of he decision o rade asses wih a non-residen. In he real world, hese decisions are made by di eren ypes of agens, wih di eren objecives: on one hand, he rade in asses is based on heir expeced payo s (dividends and price changes); on he oher hand, he rade in goods is based on he uiliy provided by he good, if i is a nal good, or is conribuion o he producion of anoher good, if i is an inermediary good. Given his ideni caion sraegy, we esimae ha in order o mee a US$ 1 billion cusomer order ow, dealers increase he domesic price of a dollar by approximaely 2.7%. We also nd ha a 1% depreciaion rae decreases he nancial cusomer ow by US$ 111 millions and he commercial ow by US$ 46 millions. The magniude of he e ec of cusomer ow on he real/dollar exchange rae is abou 5 imes he e ec of inerdealer ow on he deusche mark/dollar price esimaed by Evans and Lyons (2002). Alhough boh esimaes are no perfecly comparable, since we use cusomer ow and Evans and Lyons use inerdealer order ow, here are sill reasons o expec a larger e ec in our daase. Firs, he real/dollar marke is a much smaller FX marke han he dollar/deusche mark marke, boh in erms of volume or liquidiy, and i is naural for he price impac of a rade o be larger in he less liquid marke where i is harder o ener or exi a posiion. Second, he exchange rae in an emerging economy is much more volaile han in a developed economy, which means ha i is a riskier asse, and herefore a larger price change is required for a risk averse agen o hold i. The res of his paper is organized as follows. Secion 2 describes he model. Secion 3 presens he daase and some of is characerisics. Secion 4 esimaes he wo-way e ecs beween exchange raes and cusomer ows. Secion 5 discusses he role of he FX derivaives marke. Secion 6 concludes. 4

5 2 Model 2.1 The Cusomer Flow In his secion we will model he currency ows in one small emerging economy. The world will be composed by one large developed counry and J small emerging economies. We assume ha each small emerging economy only rade goods and asses wih he developed counry, while he developed counry rades goods and asses wih all J emerging economies. This assumpion simpli es calculaions and can be easily relaxed. Each counry, developed or emerging, issues one nancial asse and produces wo ypes of specialized goods, one nal good and one inermediary good. These goods are specialized in he sense ha hey are no subsiues across counries. The nal goods canno be raded beween counries (which means ha each counry consumes is own producion). The rade in inermediary goods beween he developed and he emerging counries deermines he commercial ow in each counry. Trade in asses beween he small emerging economy and he developed counry deermines he nancial ow The Small Emerging Economy The jh small emerging economy is populaed by an overlapping generaion of a coninuum of idenical represenaive households wih mass equal o 1. Each represenaive household lives for wo periods, consuming in boh periods, bu receiving income only when young. There are also wo coninuums of idenical represenaive rms, boh wih mass 1. rs se of represenaive rms produces he nal consumpion good and he second se of represenaive rms produces he inermediary good. The We assume ha for each young represenaive household here is one nal good represenaive rm and one inermediary good represenaive rm. Represenaive Households The domesic represenaive household from he jh small emerging economy needs FX o buy foreign asses so ha he can smooh his consumpion across boh periods. Assume ha he represenaive household born in period receives income only when he is young. Available sources of income for he young represenaive agen are: he wage W j received for supplying inelasically one uni of labor, a lump-sum ransfer of pro s j from he rms and a lump-sum ransfer T j (ha can be eiher posiive or negaive) from he governmen. In order o save for consumpion when old, he young 5

6 represenaive agen invess in wo ypes of asses: he domesic risk free bond and he large developed counry asse (ha is going o be called foreign bond). For each uni of domesic currency ha is invesed in he domesic bond in period, he young represenaive household receives 1 + r j unis of domesic currency in period + 1. For each uni of domesic currency ha is invesed in he foreign bond in period, he young represenaive household receives 1 + r + s j +1 s j in period + 1, where S j is he amoun of domesic currency necessary o purchase one uni of foreign currency in period and s j d log S j = ds j =S j. We will assume by now ha given all informaion available in period he (log linearized) exchange rae s j +1 is normally disribue wih mean E s j +1 and condiional variance V ar s j +1. We will show a he end of he nex secion ha his hypohesis is in fac rue when we calculae he nal soluion for he exchange rae. Each represenaive household chooses how much o consume in each period, C j y; and Co;+1, j and how much o hold of he foreign asse, B j;f, in order o maximize he expeced presen value of his consumpion level in boh periods, given a period uiliy funcion wih consan absolue risk aversion, u (C) = exp ( j C), subjec o is budge consrain: subjec o C j o;+1 = 1 + r j max u Cy; j + E u C j C j o;+1 y; ;Cj o;+1 W j + j + T j C j y; + B j;f s j +1 s j + r r j (1) (2) The opimal demand of he represenaive household from he jh small emerging economy for he foreign bond is: Represenaive Firms = E s j +1 s j + r r j j V ar s j (3) +1 B j;f There are wo ypes of represenaive rms in he jh emerging economy. One ype produces he domesic inermediary good. This inermediary good is used as inpu in he nal goods secor in boh he domesic counry and he large developed counry. The amoun of he domesic inermediary good ha is expored o he developed counry depends on he developed counry nal goods rm s problem. The represenaive rm of he nal good secor in he jh emerging economy produces he consumpion good using as inpus boh he foreign and he domesic inermediary goods. Le Y j be he oal amoun of he nal good produced in counry j, P f be he domesic prices of he foreign and domesic inermediary goods, and K j;f and P j and 6

7 K j;j be he oal amouns of he foreign and domesic inermediary goods demanded by he represenaive rm of he inermediary secor in counry j. Given a Cobb-Douglas producion funcion for he represenaive rm of he nal good secor: Y j = A j K j;f j K j;j 1 j (4) where 0 < j < 1 and A j is a produciviy shock erm, he opimal amoun of he foreign inermediary good o be impored will be posiively relaed o he nal good domesic producion and negaively relaed o he relaive price in local currency: K j;f The Large Developed Counry = j Y j P f (5) The large developed counry demand for he jh small emerging economy asse will be modeled as coming from a represenaive Global Invesmen Fund (GIF) specialized in invesing in he J emerging markes. The large developed counry demand for he jh small emerging counry good will be modeled as coming from a represenaive rm ha produces a consumpion good using as inpus inermediary goods from all J + 1 counries. Demand for Asses in he Developed Counry The GIF manager of generaion receives an endowmen of W unis of he large developed counry currency (dollars) and decides how much o inves in he J + 1 asses available: he large developed counry asse and he J small emerging counries asses. For each dollar invesed in he large developed counry risk free bond in period, he GIF receives in period + 1 a gross reurn of 1 + r. For each dollar invesed in he jh emerging economy risky asse, he GIF receives in period a gross reurn of 1 + r j + s j s j +1. h i 0, The GIF manager of generaion chooses B f = B f;1 ; B f;2 ; :::; B f;j where B f;j is he amoun invesed in he jh small emerging economy asse (measured in unis of he developed counry currency), in order o maximize his expeced uiliy wrien on his period + 1 wealh, W +1, subjec o is budge consrain. If we specialize he GIF manager uiliy funcion o have consan absolue risk aversion, we nd in he soluion ha he GIF holdings of he jh emerging economy asse, B f;j, is posiively correlaed o ha counry s asse expeced excess reurn and negaively correlaed wih ha counry s asse condiional 7

8 variance. Also, B f;j will be posiively correlaed wih he excess reurn in a counry j risky asse if he condiional covariance beween he one period ahead exchange raes is posiive. We can collec all oher excess reurn erms excep for he jh small emerging economy excess reurn in he variable j and wrie: B f;j = j s j E s j +1 + r j r + j (6) Demand for Goods in he Developed Counry We will model he developed counry s impor for he jh emerging economy good as being generaed by a represenaive rm ha produces a consumpion good using as inpus inermediary goods from all J + 1 counries. Le Y f be he oal producion of he nal good in he developed counry, K f;f be he amoun of he inermediary good produced and used by he large developed counry, and K f;j be he oal amoun of he jh emerging economy inermediary good used by he large developed counry. Also, le P f; and P j; be he prices of he developed and jh emerging inermediary goods boh measured in foreign currency (in oher words, he inernaional price of he inermediary goods) and assume ha he producion funcion of he represenaive rm is given by: Y f = A f K f;f JY j=1 K f;j j (7) where 0 < < 1, 0 < j < 1, P J j=1 j = 1 and A f is a produciviy shock erm. Toal impors in he large developed counry for jh emerging inermediary goods is a posiive funcion of heir oal producion of nal goods and a negaive funcion of real price of he inermediary good in foreign currency: K f;j = j Y f P j; (8) Cusomer Flow in he Small Emerging Economy The commercial ow is he excess demand for FX from cusomers ha is generaed by he rade in goods beween a counry and he res of he world. For he small emerging economy j in our model, his is he value of he impors of inermediary goods from he large developed counry minus he value of he expors of inermediary goods o he large developed counry, 8

9 measured in unis of he foreign currency: X C;j = P f; K j;f P j; K f;j (9) The nancial ow is he excess demand for FX ha is generaed by he rade in asses beween a counry and he res of he world. For he small emerging economy j in our model, i is composed by four erms: X F;j = B j;f (1 + r 1 ) B j;f 1 B f;j r j 1 s j + s j 1 B f;j 1 (10) The rs wo erms are he nancial ow ha is generaed by domesic agens, where B j;f represens he amoun young households are invesing in he developed counry and (1 + r 1 ) B j;f 1 represens he amoun ha he old households are bring back from he developed counry. The las wo erms are he nancial ow ha is generaed by he GIF, where B f;j is he amoun ha he young GIF manager is invesing in he emerging economy and 1 + r j 1 s j + s j 1 B f;j 1 is he amoun ha he old GIF manager is bring back from he emerging economy. Toal cusomer ow in he small emerging economy j is simply he sum of he commercial ow, given by equaion (9), and he nancial ow, given by equaion (10): X j = X C;j + X F;j (11) In order o simplify he expressions for oal cusomer ow X j described in equaion (11) we can log linearize i wih respec o he curren exchange rae s j, and all oher macroeconomic variables such as he domesic and inernaional ineres raes, he domesic and inernaional GDP, and he J small emerging economies excess reurn erms (for furher simpli caion we will summarize all oher domesic and inernaional macroeconomic variables ino one vecor F j ): X j = j 0 + j 1s j + j 2F j (12) The imporan hing o noice in equaion (12) is ha he oal cusomer ow is a negaive funcion of he exchange j = " j Y j S j 2 # " 1 j V ar s j +1 + r r j + j + B f;j 1 # (13) 9

10 2.2 The FX Marke Descripion In his secion we will describe he ineracion beween cusomers and dealers in he FX marke of a small emerging economy. Since we will be looking a jus one small counry, he superscrip j ha indexed each of he J emerging economies in he previous secion will no longer be necessary. The rading model ha we use has some key feaures of Lyons (1997): here are only wo rading rounds, wih he cusomer-dealer ransacions being limied o he rs round; rading wih muliple parners is feasible; and i is a simulaneous rade model, which means ha all demand schedules have o be submied simulaneously. However, our model depars from Lyons (1997) and oher FX microsrucure models such as Evans and Lyons (2002) in a couple of ways. We have a compeiive raional expecaions equilibrium model: dealers can condiion heir orders on marke clearing prices. Also, dealers receive an addiional signal prior o he second round, afer he rs round is concluded. This signal ries o capure one characerisic of he iner-dealer rade in FX marke: he readjusmen of heir desired FX holdings before he marke closes using all new informaion ha was colleced during he day while markes were already open. Our model also presens wo imporan improvemens relaive o previous microsrucure models. Firs, cusomer ow is no simply he realizaion of a random variable. In he previous subsecion, we modeled his ype of ow as a funcion of he exchange rae and oher macroeconomic fundamenals. The second imporan improvemen is ha dealers FX holdings a he end of each rading day are no exogenously se (o zero, for example): dealers may opimally decide o hold overnigh posiions in he FX marke. The desired amoun will be deermined by heir pro funcions and he expeced overnigh FX payo s: he ineres rae di erenial and he depreciaion rae from one period o anoher. This implies ha he dealers in our model play wo roles in he FX marke: he rs as inermediaries, maching buyers and sellers wihin a rading day, and he second as invesors. Since he dealers pro s as invesors in he FX marke comes from he overnigh payo of heir FX holdings, hey behave in a manner very similar o he households from he previous subsecion, which smoohed consumpion using foreign asses generaing he nancial ow. However, i is imporan o noice ha dealers di er from he nancial invesors since hey may also pro from inra day ransacions, ha is, from poenial movemens of he exchange rae from he rs rading round o he second. This will become clear in he nex subsecion when we presen more speci c deails abou he dealers problem. 10

11 2.2.2 Soluion There is an overlapping generaion of a coninuum of idenical dealers in he FX marke. Each dealer wihin his generaion is indexed by i 2 [0; 1] and lives for wo periods. The young dealers are responsible for supplying he excess demand of he cusomer in a given period. Le Q be he marke wide level of FX invenory held by he dealers in period and le X be he cusomer ow of he same period. Since a posiive cusomer ow means ha here is a posiive excess demand for FX, we can wrie: Q = Q 1 X (14) In each period, rading occurs in wo rounds. In he rs round, each young dealer i rades wih he old dealers and wih is cusomers a he rs round equilibrium price s ;1. In he second and nal rading round, each dealer i readjuss is invenory of FX by rading wih oher dealers. In he rs round of he following period, + 1, he young dealer of generaion will sell all his posiions, collec his pro s, and exi he FX marke. Le Q i ;1 and Q i ;2 be he FX holdings of dealer i afer he rs and he second rading rounds respecively. The dealer s pro funcion can be wrien as: i; = Q i ;1 (s ;2 s ;1 ) + Q i ;2 s +1;1 + V s ;2 (15) The rs par of he pro funcion of dealer i from generaion is given by is rs round holdings (he amoun purchased from he old dealers and from his cusomers) imes he change in he equilibrium exchange rae from he rs o he second rading round, s ;2 s ;1. The second par is he payo ha he receives from is overnigh FX holdings. This payo is composed by he change in he exchange rae equilibrium price from he second round of period unil he rs round of period + 1, s +1;1 s ;2 (when he is going o sell his posiions and exi he marke), plus he consan overnigh dividend V paid by he FX from period o + 1. To simplify he calculaions, we will assume from now on ha V = 0. All dealers maximize idenical negaive exponenial uiliy de ned over is pro s, wih consan coe cien of risky aversion. Marke clearing condiions imposes ha he oal amoun of FX held by he young dealers a he end of he rs round equals he oal amoun of FX sold by he old dealers exiing 11

12 he marke less he oal amoun demanded by cusomers: Z 1 0 Q i ;1di = Q 1 X (16) Since dealers only rade wih oher dealers in he second and nal round, he oal amoun of FX held by he dealers overnigh has o be equal he oal amoun of FX a he end of he rs round: Z 1 Q i ;2di = Q (17) 0 The informaion srucure is he following. The realizaion of he macroeconomic fundamenals F N F ; 2 F is known before markes are opened. Cusomers observe i and decided how much FX hey need o rade in he rs round. Afer he rs rading round and before he second rading round, each bank receives new informaion, in he form of a common signal. The signal, F c = F +1 +" i, wih " N (0; 2 "), is correlaed o he following period macroeconomic fundamenals and will a ec he following period s cusomer ow. Le E ;j (:) and V ar ;j (:) be respecively he expecaion and he variance of a random variable condiional on all informaion available in round j of period. Given ha, condiional on all informaion available a he rs round, s ;2 is normally disribued, and condiional on all informaion available a he second round, s +1;1 is also normally disribued, dealer i opimal FX holdings in each rading round will be given by: Q i ;1 = E ;1 (s ;2 ) s ;1 V ar ;1 (s ;2 ) Q i ;2 = E ;2 (s +1;1 ) + V s ;2 V ar ;2 (s +1;1 ) (18) (19) Proposiion 1 There are raional expecaions equilibrium exchange raes for rading round 1 and rading round 2 wihin he class of funcions of he form: s ;1 = Q F (20) s ;2 = Q + 2 F c (21) 12

13 These prices are given by: 0 = 0 = 0 = 1 (22) 1 = F F (23) 2 = F F (24) 1 = " (25) 2 F F 2 = F ( 2 F + 2 ") (26) Proof. See appendix 2. Corollary 2 The change in he exchange rae from he end of period period can be wrien as: wih 1 < 0 and 2 > 0 if he following condiion is veri ed: 1 o he end of s ;2 = 1 X + 2 F c (27) j 1 j < F (28) Proof. Equaion (27) follows direcly from equaion (21) and equaion (14). Also, since 1 < 0 and 2 > 0, hen from equaion (26) we can see ha he only way 2 > 0 is if condiion (28) is sais ed. Finally, if 2 > 0 hen from equaion (25) i is necessarily he case ha 1 < 0. Keeping in mind ha he exchange rae is de ned as he amoun of domesic currency necessary o purchase one uni of he foreign currency, 1 < 0 and 2 > 0 means ha he exchange rae depreciaes if here is pressure from cusomers o purchase FX from dealers (X > 0) or if he signal received by dealers indicaes an increase in fuure cusomer demand (F c > 0 could be inerpreed, for example, as dealers becoming more pessimisic abou he counry risk premium in he following period). One nal commen should be made abou he condiion ha limis he sensiiviy of he cusomer ow o exchange rae changes. Imagine an exogenous shock ha increases he cusomer ow by X. Given equaion (27), his excess demand will depreciae he exchange rae. If 1 is negaive, he exchange rae depreciaion will feedback ino he cusomer ow by reducing i. However, if he magniude of he feedback e ec is large enough, he decrease in he cusomer ow can be so srong ha i will generae 13

14 a negaive cusomer ow larger han he iniial shock X, which will generae a subsequen appreciaion of he exchange rae even larger in magniude han he original depreciaion. 3 Daase 3.1 Descripion The daase used in his paper conains ransacion ows beween dealers and hree ypes of counerparies: commercial cusomers (whose demand for FX is generaed by a rade in goods wih non-residens), nancial cusomers (whose demand for FX is generaed by a rade in asses wih non-residens), and he cenral bank. The daa is aggregaed by each ype of counerpary on a daily basis, spanning a oal period of 4 years, from he 1s of July of 1999 unil he 30h of June of 2003, and covers 100% of he o cial Brazilian FX marke, one of he larges emerging economies. If compared o oher daases presened in previous sudies, his is unique in a leas wo ways. Firs, i is he only one ha covers 100% of a counry s FX marke. Second, while oher sudies are looking a he exchange rae beween currencies from wo developed counries, his is he rs ha covers an emerging economy. The daase was obained from he SISBACEN, an elecronic sysem of collecion, sorage and exchange of informaion ha connecs he Brazilian cenral bank and all oher agens operaing in a Brazilian nancial marke, including he FX marke. The cenral bank closely moniors all aciviies involving capial ows. For insance, i requires all dealers operaing a he Brazilian FX marke o inpu ino he SISBACEN on daily basis informaion abou he characerisics of each of heir ransacions in ha marke. These characerisics include he price, he volume, he ype of counerpary anoher dealer, he cenral bank or a cusomer and, if he counerpary is a cusomer, he naure of he underlying economic ransacion ha generaed he demand for exchanging FX. The deailed informaion inpued by he FX dealers ino he SISBACEN is only observed by he cenral bank. Afer some considerable delay, some summary daa aggregaed in a lower ime frequency is released o he dealers hrough he SISBACEN. The Brazilian FX marke is a decenralized muliple dealer marke. The rading beween dealers and commercial and nancial cusomers occurs in he reail FX marke. This marke is also called he primary marke, since is ne ransacions a ec he counry s aggregae invenory of foreign currency. Trading beween dealers and he cenral bank does no a ec he counry s invenory, since boh are domesic agens, bu i does a ec he dealers marke 14

15 wide level FX invenory. Inerdealer rading does no a ec he counry nor he marke wide invenory. When rading wih cusomers, dealers do no behave as marke makers, ha is, hey are no required o quoe a rm bid and ask price a which hey are ready o buy or sell FX a any ime while he marke is open. They may condiion heir quoe on wheher he cusomer wans o buy or sell and also on he size of he ransacion. Cusomers, in order o rade FX wih dealers, need o have proper jusi caion, ha is, hey have o show documenaion wih respec o he underlying economic ransacion wih a non-residen ha is generaing he need o exchange foreign currency. Since he cusomers (and heir underlying economic aciviies) are he agens who iniiae he ransacion wih he dealers, we will sign each ransacion from heir poin of view. This means ha if cusomer ow is posiive, i represens pressure from cusomers o buy FX or, equivalenly, pressure on dealers o sell FX. The same procedure will be aken wih he ransacions beween dealers and he cenral bank. I is he cenral bank who iniiaes he ransacions wih dealers; herefore a posiive inervenion ow represens pressure from he cenral bank o buy FX from dealers or, equivalenly, pressure from dealers o sell FX. All FX ows will be measured in US$ billions. The daase also includes he foreign ineres rae, he domesic ineres rae, and a measure of he Brazilian counry-risk premium, all hree in rs di erence. The foreign ineres rae is he daily annualized Fed Funds rae, he domesic ineres rae is he daily annualized Selic rae, and he risk premium is measured as he spread of he C-Bond (he mos liquid Brazilian Brady bond in he sample period) over he Treasury, measured in annualized raes, so a 1% risk premium is equivalen o a 100 basis-poins spread of he yield of he C-bond in % a.a. over he yield of a Treasury bill wih equivalen mauriy, also in % a.a. Table 1 presens summary saisics. 3.2 Firs Look a he Daa Figure 1 shows he relaionship beween he Brazilian exchange rae, de ned as he domesic price for one US dollar, and he cumulaive cusomer order ow in Brazil, where he cumulaive cusomer order ow in a dae is de ned as he sum of all cusomer ows beween dae 0 (July 1, 1999) and dae. Two feaures are noeworhy in his graph. Firs, we can see ha he correlaion beween he cumulaive cusomer ow and he exchange rae is negaive (and signi can a he 1% signi cance level). Second, marke wide cusomer ow each day does no ne o zero, which means ha dealers of he Brazilian FX marke do no simply behave 15

16 as inermediaries maching buyers and sellers during he day. If, for example, a he end of a given day here are more buyers han sellers, he dealers may supply he exra liquidiy overnigh. The negaive correlaion beween cumulaive cusomer order ow and he exchange rae give us a ase of he role played by endogeneiy: on he one hand, exchange rae movemens a ec he price of foreign goods and asses and herefore a ec he cusomer order ow; on he oher hand, demand pressures for FX a ec he exchange rae. This is he opposie resul compared o wo previous sudies. Evans and Lyons (2002) nd a srong posiive correlaion a he daily frequency beween he exchange rae and inerdealer order ow for he deusche mark/dollar and yen/dollar markes and Fan and Lyons (2003) nd a posiive correlaion a he monhly frequency beween he exchange rae and cusomer order ow for he euro/dollar and yen/dollar markes. These sudies inerpre he posiive correlaion from he FX liquidiy suppliers poin of view: pressure o buy FX increases he price charged by dealers. In our case, he negaive correlaion found in he sample period is more likely o be explained by he poin of view of he agens who are demanding he FX liquidiy: as he exchange rae slowly depreciaes, cusomers demand for foreign currency slowly decreases, as a resul for example of smaller impors and higher expors. Figure 2 shows he behavior of he commercial cusomer ow during he sample period and is negaive correlaion wih he exchange rae con rms our inuiion. However, we can also noice ha in periods where sharp increases in cusomer order ow are observed here are spikes in he exchange rae (second half of 2001 and second half of 2002, when lower con dence in he Brazilian economy generaed large porfolio ou ows), as a resul of dealers seing a higher price o supply he excess demand for FX coming from he nancial cusomers. This posiive correlaion beween nancial cusomer ow and he exchange rae can be seen in gure 3. In order o ge a more formal sense on he presence of endogeneiy in our daase, we presen wo ses of empirical evidence. Firs, we run bivariae Granger causaliy ess beween exchange rae movemens and each ype of order ow separaely (commercial, nancial and inervenions). The lag lengh of each bivariae VAR is based on he Schwarz informaion crierion. The es resuls ell us ha we can rejec he null hypohesis ha exchange rae movemens do no Granger cause he commercial nor he nancial ows a he 1% signi cance level. Also, we can rejec he null hypohesis ha exchange rae movemens do no Granger cause inervenion ows a he 10% signi cance level (see able 2 for deailed resuls). This resul di ers from Killeen, Lyons and Moore (2005), which nd no evidence of Granger causaliy running from he french franc/deusch mark exchange rae o he iner- 16

17 dealer order ow of he same marke. However, we mus bear in mind ha hese resuls are only an indicaion of he presence of endogeneiy since Granger causaliy measures precedence and informaional conen raher han conemporaneous reverse causaliy in he sense of feedback rading. The second se of evidence is provided by running single equaion OLS regressions using he rs di erence of he log of he exchange rae as he dependen variable and he oal cusomer ow from he reail marke as a regressor. Noe ha his regression is comparable o Evans and Lyons (2002), even hough heir dependen variable is inerdealer order ow and no cusomer order ow. In heir model, inerdealer order ow should be a muliple of he cusomer iniiaed ransacions. Since in our daase he ransacions beween dealers and cusomers were iniiaed by he cusomers, we should expec a similar resul from heir paper: a posiive and signi can coe cien indicaing ha pressure o buy increases he price of FX. However, our esimaed coe cien is negaive, and i is signi can a he 1% signi cance level if we do no include any conrols, and no signi can a he 5% signi cance level if we also include he rs di erences of he domesic ineres rae, he foreign ineres rae and he measure of he Brazilian counry risk premium (see able 3 for deailed resuls). Wha does his resul mean? Tha pressure o buy in he reail FX marke does no increase he FX price or ha he demand for FX in he reail marke decreases as a resul of an increase in he price of FX? Almos cerainly, boh inerpreaions are wrong. The coe cien in he OLS regression is mos likely biased owards zero because of he presence of endogeneiy, and herefore i does no have any economic meaning. We have also noiced from gure 1 ha cusomer ows from he reail marke does no ne o zero a he marke wide level a he daily frequency. This means ha dealers are providing liquidiy overnigh, by absorbing he change in he counry s FX invenory ha is generaed by he cusomers underlying economic ransacions (in goods and asses) wih non-residens. Since he balance of paymens summarizes in a sysemaic way all ransacions beween residens and non residens, a good way o check he qualiy of he daase is o compare our cusomer ow daa wih is balance of paymens equivalen. In order o do his comparison, we mus nd he balance of paymens measure of change in he counry s invenory of FX. While he cusomer ow may di er from zero in any given period, he ne balance of all enries in he balance of paymens is always equal o zero due o he double enry sysem. The cusomer ow represens he amoun of paymens ha exceeds revenues in he ransacions beween non-dealer residens wih non residens. Therefore, i should equal o he sum of he curren accoun, he capial accoun and he nancial accoun as 17

18 long as we subrac from he nancial accoun all he cash holdings of he dealers from he FX marke and also he loans made by he cenral bank wih inernaional organizaions (since he cenral bank is no a cusomer from he FX reail marke). According o he fh ediion of he balance of paymens Manual issued by he Inernaional Moneary Fund (BPM5), he change in cash balances held by FX dealers is regisered in he Financial Accoun in he enry Oher Invesmens - Currency and Deposis. This accoun regisers he second enry of a ransacion beween non-dealer residens and nonresidens ha was paid wih foreign currency ha was bough or sold wih dealers from he FX marke. The rs enry will depend on he naure of he ransacion, wheher i was in he goods marke (e.g.: expor, impor... ) or in he asses marke (e.g.: direc invesmen, porfolio invesmen... ). Sill according o he BPM5, when he cenral bank purchases foreign currency from a dealer, here is a decrease in dealers cash holdings, and a corresponding increase in he cenral bank s cash balances, ha will be accouned in Reserve Asses. However, no all changes in Reserve Asses re ec a cenral bank s inervenion. If he cenral bank receives a loan of US$ 10 billions from he Inernaional Moneary Fund, here will be an increase in reserve asses even hough no inervenion in he FX marke occurred. This means ha here will be no enry in Oher Invesmens - Currency and Deposis associaed wih he increase in he Reserve Asses. The second enry of his operaion in he balance of paymens will be in Oher Invesmens - Loans - Moneary Auhoriy. Once we eliminae all enries associaed wih loans received by he cenral bank from is reserve asses accoun, we are lef wih he changes in is balances due o inervenions. Figure 4 compares he cusomer ow from he FX reail marke wih he measure of paymens imbalances calculaed using he balance of paymens. Figure 5 compares he cenral bank inervenion ows from he FX marke wih is increase in reserve asses due o inervenions in he FX marke measured by he balance of paymens. In boh graphs, he daa on FX marke ows was aggregae ino a monhly series in order o mach he highes frequency a which daa on he balance of paymens is available. Also, we show he 3 monhs moving average of each series so ha we could faciliae he visual comparison (monhly behavior is oo volaile). While hese gures poin ou o a close relaionship beween our daase and he balance of paymens, formal ess canno rejec he equaliy beween each FX marke ow and is balance of paymens equivalen (see appendix 3). A nal issue sill relaed o he fac ha dealers supply liquidiy overnigh insead of neing ou heir invenories o zero is wheher hey are making pro s or losses from hese posiions and from which counerpary are hese pro s or losses coming from. I is imporan 18

19 o noice ha hese are no he only source of he dealers pro s. One componen of he dealers pro s in he FX marke comes from is role as an inermediaion, by maching a buyer wih a seller and pro ing from he bid-ask spread. Anoher imporan source of pro s comes from he dividends paid by he foreign currency (he ineres rae di erenial). We are ineresed a he FX dealers speculaion pro s, ha is, he pro s ha arises only from he change in he price of he asse he is rading. We wan o have an esimae of wheher, when dealers hold more FX han he usual in a given day, does he price of FX in average goes up (pro s) or down (losses) in he nex day. We calculae hese speculaive pro s using he mehodology described in Hau (2005). Le X i be he daily aggregae demand of counerpary i (commercial cusomers, nancial cusomers or he cenral bank). Now imagine ha here is a represenaive counerpary i hrough he sample period and de ne Q i = P T =1 Xi as his invenory of FX. Le Q i = 1 T P T =1 Qi be he long run average invenory and e Q i = Q i Q i be he daily deviaion of he represenaive counerpary i from is long run average. Finally, le s be he exchange rae. The oal pro of he represenaive counerpary i is given by: i = TX n=1 eq i s (29) Since he dealer is he counerpary of hese agens, hen if represenaive counerpary i is making pro s, i is he case ha he dealers are having losses by rading wih hem. Therefore, he dealers aggregae pro is he sum of he losses of he represenaive commercial cusomer, nancial cusomer and cenral bank. Also noe ha if he dealers ended every day wih he same average invenory, heir aggregae speculaive pro s would necessarily be zero. The resuls ell us ha he commercial cusomers are having losses, and ha he nancial cusomers and he cenral bank are having pro s. However, he magniude of commercial cusomers losses is greaer han he sum of he nancial cusomers and he cenral bank pro s. Therefore, dealers are also having pro s. In order o have an idea of he relaive size of he pro s dealers make, we divide heir overall pro by he average absolue size of he daily aggregae ow. The resul is ha for each dollar ha he dealers supplied overnigh, hey made a daily pro of 0.035%, or a 9.3% annualized pro rae. 19

20 4 Empirical Evidence 4.1 Esimaion Sraegy Le s be he daily change in he spo exchange rae, X be he cusomer order ow and Z be a se of oher macroeconomic variables o be speci ed laer. The main objecive of his secion is o esimae he posiive e ec of cusomer ow on exchange raes, prediced by he model developed in his paper, incorporaed by he coe cien in he following equaion: s = X + Z + " (30) However, in order o obain a consisen esimae of we have o ake ino accoun he simulaneous deerminaion beween cusomer ow and exchange rae: " # " s X # = Z + The ideni caion sraegy applied in his paper is o include in he vecor Z all possible sources of common shocks ha may a ec simulaneously boh variables and hen o rely on he independency of he srucural innovaions. " " 1; " 2; # (31) Since he vecor of common shocks also includes as regressors lags of he endogenous variables, we esimae a Srucural VAR, inroduced by Sims (1986). However, independency of he srucural innovaions does no add enough resricions for us o idenify he sysem. The second par of he esimaion sraegy is o disaggregae X, he cusomer order ow, in X F, he nancial cusomer ow, and X C, he commercial cusomer ow. A a rs glance, his move does no seem o be e ecive since i adds more unknowns o be esimaed ino he sysem: s X F X C 7 5 = Z " 1; " 2; " 3; (32) Wih he assumpion ha innovaions " i;, i = 1; 2; 3, are independen we can idenify only 3 ou of he 6 endogenous coe ciens in he sysem. Bu we can use non-sample informaion o impose furher resricions on hese coe ciens. Based on he model developed in his paper, wo addiional resricions o be made are: a. The conemporaneous e ecs of he nancial cusomer ow and he commercial cus- 20

21 omer ow on he exchange rae are no di eren. In oher words: 12 = 13. b. The nancial cusomer ow and he commercial cusomer ow do no a ec each oher conemporaneously. In oher words: 23 = 0 and 32 = 0; Resricion (a) is jusi ed by he fac ha wha generaes he price e ec in our model is he change in he dealers aggregae invenory a he marke wide level; and i does no maer where he change comes from. Resricion (b) does no mean ha he commercial cusomer ow and he nancial cusomer ow are no correlaed a he daily frequency. Indeed, common sources of shocks may a ec boh ypes of ows simulaneously, such as changes in he exchange rae, in he domesic and foreign ineres raes, in he counry risk premium and also pas shocks o cusomer ows. However, once we conrol for hese poenial sources of common shocks, our model ells us ha he decision o rade goods wih non-residens is independen of he decision o rade asses wih a non-residen. In he real world, hese decisions are made by di eren ypes of agens, wih di eren objecives. Given hese wo addiional non-sample resricions based on he heoreical model developed in his paper, our model becomes jus-ideni ed: Esimaion Oupu s X F X C 7 5 = Z " 1; " 2; " 3; (33) The se of exogenous variables included as regressors are he rs di erences of he domesic ineres rae, r, he rs di erence of he foreign ineres rae, r, and he rs di erence of he Brazilian risk premium (inroduced in he previous secion), rp, and wo oher ses of deerminisic variables, weekday dummy variables and monh dummy variables: A 0 y = px px A j y j + B j Z j + dummies + " (34) j=1 j=0 where he vecors of endogenous and exogenous variables are given by (able 4 shows ha we can rejec he presence of a uni roo in all hese variables): h i 0 y = s X F X C (35) h i 0 Z = r r rp (36) 21

22 and he marices associaed o he exogenous variables are: A 1 = i 11 i 12 i 21 i (37) B j = i 31 i 32 b j 11 b j 12 b j 13 b j 21 b j 22 b j (38) and he marix of conemporaneous relaions is: A 0 = b j 31 b j 32 b j (39) Before we sar our esimaion procedure we should commen he fac ha we rea he domesic ineres rae and he counry risk premium as exogenous variables. This means ha we do no believe ha daily ucuaions on exchange raes or in he cusomer ows cause any e ec on any of hese wo variables in he same day. Firs, he shor erm ineres rae does no reac direcly o he movemens of he exchange rae or he cusomer ows because he moneary policy regime in Brazil during he whole sample period was (and sill is) an In aion Targeing regime. However, one could sill argue ha he shor erm ineres rae reacs indirecly o he exchange rae because of he e ecs of he laer on he in aion rae hrough he price of he impored nal and inermediary goods. Alhough his pressure may indeed exis a he quarerly or monhly frequency, i does no exis in he daily frequency. The reason is because he Brazilian Cenral Bank conducs is moneary policy in a very similar way o he Federal Reserve. I has a moneary policy commiee ha resembles he FOMC, which mees on average once every monh, when hey decide wha should be he level of shor erm ineres rae during he nex 30 days. Afer ha, hey manage o keep shor raes close o he argeed level by consanly adjusing he supply of money. The dashed line in gure 6 con rms he very lile daily variabiliy in he behavior of he domesic ineres rae. In he same gure, we also presen he counry risk premium (solid line), which indeed presens a lo of daily variabiliy. However, his variable is perceived as a measure of he markes assessmen of he probabiliy ha a counry migh defaul on is deb obligaions. Therefore, we follow Blanchard (2004) and Favero and Giavazzi (2004) 22

23 heoreical and empirical argumens ha he mos imporan deerminan of he counry risk premium is he agens percepion abou he scal policy, which could be summarized by he deb over GDP raio, or he primary surplus. The rs sep of he esimaion procedure is o choose he mos appropriae value for p in equaion (34), which gives us he lag lengh of he endogenous variables in he VAR. For simpliciy, we also se he lag lengh of he exogenous variables equal o p. This choice is made by esimaing 5 di eren versions of equaion (34), each wih a di eren value for p, varying from 1 o 5. Then, for each choice of p we compare he Akaike and Schwarz informaion crieria, we perform lag lengh exclusion ess on he endogenous variables and we es for serial correlaion in he residual. All resuls indicae ha he mos appropriae choice is a value of p equal o 2 (see able 5 for deailed informaion crieria and ess saisics). The second sep is o esimae he reduced form VAR (esimaed coe ciens and sandard errors presened in able 6). Finally, using he variance-covariance marix of he reduced form residuals and he resricion discussed in he esimaion sraegy descripion, we can esimae he marix A 0 in equaion (34). The following equaions presen he endogenous relaionships implied by he esimaed coe ciens. All coe ciens are signi can a he 1% signi cance level (see sandard errors in able 7): s = 0:027 X F + X C + e1 (40) X F = 11:06s + e 2 (41) X C = 4:61s + e 3 (42) The esimaed slope 0 12 associaed wih he supply curve is 2.7%. This means ha in order o mee a US$ 10 millions cusomer order ow, he dealers of he FX marke increase he FX price by approximaely 0.03%. The coe cien 0 21 refers o he response of he nancial cusomer ow o exchange rae movemens: a 1% appreciaion in he exchange rae increases cusomers demand for foreign exchange associaed wih he rade of asses by US$ 111 millions. Finally, he coe cien 0 31 refers o he response of he commercial cusomer ow o exchange rae movemens: a 1% appreciaion in he exchange rae increases he cusomers demand for foreign exchange associaed wih he rade of goods by US$ 46 millions. These esimaes are robus o he choice of he VAR lag lengh. In order o show ha, able 7 also presens he esimaed marix of endogenous relaionships A 0 for 5 di eren VAR s, wih lag lenghs varying from 1 o 5. As we change he lag lengh of he VAR, he esimaed value for each coe cien only changes marginally (di erence is less han 1 sandard 23