Speculaor idenificaion: A microsrucure approach Ben Z. Schreiber* Augus 2011 Absrac This paper suggess a mehodology for idenifying speculaors in FX markes by examining boh he speculaive characerisics of all he players and he relaionships beween a player's ne purchases and posiions wih marke variables. A player is idenified as a speculaor only if his speculaive characerisics are exreme compared o oher players and his influence on exchange raes, implied volailiy, and bid-ask spread, is significan. Implemening he proposed mehodology on Israel's FX marke, which includes 366 large players, idenified 47 speculaors mos of hem foreign insiuions, local banks, and financial firms. Examining heir aciviy during 2008-2009 revealed ha several speculaors purchased FX before oulier depreciaion days, while before oulier appreciaion days hey sold FX. This means ha some speculaors joined or iniiaed he rend before oulier appreciaion or depreciaion days. Based on he behavior of hese speculaors, which were found during he sample period, i was possible o idenify heir speculaive behavior before oulier days during an ou-ofsample period. * Bank of Israel, Informaion and Saisics Deparmen Keywords: Foreign Exchange Markes, Microsrucure, Speculaion
1. Inroducion Speculaors usually provide FX markes wih liquidiy and sophisicaion. Many economies, however, suffer from speculaor aciviy such as "carry rade". Moreover, i is argued ha in small open economies speculaors cause sharp movemens in exchange raes and oher marke variables such as he Bid-Ask Spread (hereafer BAS) and Implied Volailiy (hereafer IV). Thus, i is in he ineress of he moneary auhoriies o idenify and rack a manageable amoun of speculaors. Tracking speculaors is of high imporance during 'oulier days', which are defined in his paper by exchange rae changes, IV, and BAS, using mulivariae oulier deecion echniques. Speculaors subsanially influence FX markes around he world however here is no consensus eiher in he lieraure or by praciioners as o he definiion of a speculaion. There is a wide range of definiions for he erm speculaion, some of which essenially include all day-o-day invesmen aciviy while ohers have a narrower scope. Moreover, he erm speculaion or speculaor in FX markes is more problemaic han in oher markes due he sophisicaion of he acive players, he cross-counry aciviy i.e. "carry rade", and he relaive scarciy of informaion regarding rading and players. 1 This paper suggess a mehodology o idenify speculaors by examining he following wo non-exclusive condiions: (1) a significan influence of he player's ne purchases on marke variables, especially on oulier days using TSLS regressions, and (2) exreme rading characerisics of a paricular player. Oulier days are defined based on he above marke variables (exchange rae, IV and BAS) and idenified using using he mulivariae oulier deecion echnique of Galeano, Pena, and Tsay (2006). A player is idenified as a speculaor by fulfilling boh condiions i.e., his speculaive characerisics are exreme compared o oher players and his influence on marke variables on oulier days, is significan. I was found ha he financial grouping (domesic banks, financial firms, and foreign insiuions) behavior was quie differen from ha of he commercial grouping (exporers, imporers, and insiuional invesors) paricularly on oulier days. Implemening he above mehodology during he sample period (2008-2009) idenified 76 large players whose ne purchases significanly influenced marke variables bu of which only 47 players whose rading characerisics were exreme compared o oher large players. Thus, here were 29 oher players who fulfilled condiion (1) 1 Alhough, speculaor and speculaion are no he same e.g., some speculaive aciviy is carried ou by exporers or imporers, his paper's focus is on speculaors who acively rade in he FX marke for financial profi hus, having a subsanial influence on marke variables. 2
bu no (2). The speculaors lis is almos cerainly no random since he probabiliy of randomly selecing he very same players by boh condiions is close o zero. The suggesed mehodology also makes i possible o disinguish beween players who are rading in he same direcion as he marke (appreciaion/depreciaion) and hose who are rading in he opposie direcion. Of he 47 speculaors, 11 raded wih he direcion of exchange rae changes i.e. sold (bough) USD in appreciaion (depreciaion) oulier days and anoher 14 who aced as conrarians i.e., sold (bough) USD in depreciaion (appreciaion) oulier days. Tracking speculaors during an ou-of-sample period (2010) revealed ha hey behaved differenly from oher large players near oulier days. The res of he paper is organized as follows. Secion 2 surveys he lieraure. Secion 3 presens basic saisics on he aciviy of he main secors in he local FX marke on normal and oulier days. The TSLS regression resuls are depiced and discussed in secion 4, secion 5 examines he behavior of speculaors during he sample period and an ou-of-sample period. Secion 6 summarizes. 2. Survey of he lieraure According o Graham and Dodd (1951) speculaion is any risky financial aciviy. Grossberg and Schreiber (2005) defined a speculaor in foreign currency markes as a player who is in general sophisicaed and who operaes simulaneously in more han one segmen (credi in one currency and deposis in oher currency) in order o exploi yield differenials (carry rade). Osler (2006) characerized a speculaor in foreign currency as a player who focuses on changes in exchange raes, in conras o a real player (such as an exporer or imporer) or an insiuional invesor who is ineresed in he level of exchange raes. Accordingly, a speculaor will prefer shorer periods of invesmen, will seek greaer leverage and will change his posiion from long o shor and vice versa wih a high frequency. His goal is o profi from shor-erm changes. As a resul, financial raders (paricularly speculaors) iniiae shor-erm movemens in exchange raes while commercial raders reac o such movemens. In an ineresing and exhausive survey of he foreign currency markes, Osler (2008) poined ou ha he proporion of speculaors in a ypical foreign currency marke accoun for up o 80 o 90 percen of oal aciviy hough he definiion she uses for a speculaor is only one of many alernaives. In order o deermine which variables should be used o define a speculaor, a small-scale survey was carried ou among professionals in he local banking and financial secors. According o he survey resuls and he relevan lieraure, speculaive aciviy is characerized according o he following parameers: (1) a relaively shor-erm invesmen horizons; (2) use of financial leverage (limied equiy relaive o rading volume); (3) rapid shifs from a long o a 3
shor posiion regardless of fundamenals; (4) he amouns of rade end o be round numbers; (5) he number and size of rades end o increase in a volaile marke; and (6) he players eiher influence or are influenced by marke variables. 3. Descripive saisics by secor: normal versus oulier days The daabase used in his paper is unique, as every single FX ransacion done hrough local banks mus be repored o he Bank of Israel. The analysis focuses on he large players who ac as speculaors and can affec marke variables. In order o deermine which secors influenced he exchange rae or were influenced by i (since ne purchases and he exchange rae are simulaneously deermined), his secion focuses on secors and financial insrumens ha influenced he ILS/USD exchange rae (or were influenced by i). The daabase includes he following secors: exporers, imporers, financial firms, foreign insiuions, insiuional invesors, and households and he following financial insrumens: spos (including same-day and TOM ransacions), forwards, call opions and pu opions. Afer omiing non-relevan observaions, he remaining approximaely 481 housand records were grouped by secor and rading day. 2 Then, 'oulier days' were defined based on he daily changes in he ILS/USD exchange rae, he implied volailiy (IV), and he Bid-Ask Ask Bid Spreads (BAS). The laer was calculaed as. Since here are hree relaed ( Ask Bid )/ 2 variables involved, a rading day was classified as an oulier using he mulivariae oulier deecion echnique proposed by Galeano, Pena, and Tsay (2006) (see Appendix 1). Overall, according o his echnique, here were 38 oulier days during he sample period (2008-2009) and an addiional 10 days during he ou-of-sample period (2010). The oal number of oulier days represens 6.5 percen of he oal rading days. The following rading characerisics were seleced o idenify speculaors: 1. Number of rades on oulier days over normal days - NUM: A large number of rades may be an indicaor of speculaive aciviy (day rading, for example). Furhermore, a large number/volume of rades on days wih higher volailiy (such as oulier days) is addiional indirec evidence of speculaive aciviy. 2. Number of days unil expiraion DTE (weighed average of days for spo, forward and opion rades): I is conjecured (see for example, Osler (2008)) ha speculaors prefer shorerm liquid financial insrumens in order o reduce BAS coss. 2 Of abou 2500 players in he marke, he seleced 366 large players accouned for approximaely 90% of oal volume. 4
3. Heerogeneiy HET, is calculaed as (buy - sell)/(buy + sell), such ha is values ranges from +1 o -1, respecively. Values close o 0 characerize a speculaor. 4. The proporion of rades in round numbers - RND: Speculaors ofen rade in round numbers, in conras o exporers and imporers who are seeking proecion for expor and impor ransacions ha do no necessarily involve round numbers. An index closer o one indicaes ha here was a large proporion of rades involving round numbers while an index closer o zero indicaes he opposie. 5. Changes in he direcion of he posiion - CHG: Frequen changes in he direcion of he ne posiion - POS i.e. from a long posiion on he shekel o a shor posiion or vice versa, migh be an indicaor of aciviy ha is no for purposes of hedging or long-erm invesmen. 6. Relaive posiion - REL: Toal ne purchases relaive o he overall posiion (in absolue erms) is mean o reflec he level of leverage in financial aciviy. Speculaors are usually characerized by large relaive posiions. The perceniles of he above parameers were calculaed for each player and he simple mean of hese perceniles was deermined a speculaive score from 1 o 366. The players hen were sored from high o low according o hese scores (excep DTE and HET, in absolue erms, which were sored from low o high). Thus, a higher score indicaes ha he player is a speculaor. Table 1 presens hese parameers by secor. [Ener Table 1 here] The differences beween secors are quie noiceable; of he 366 large acive players, 45% were foreign insiuions, while financial firms, exporers, and imporers accouned for 40%. The differences in rading characerisics beween secors were even more pronounced. For example, local banks and financial firms were acive on more days relaive o heir proporions by number of players (13% versus 7% and 17% versus 14%, respecively) han were expor and impor firms (10% versus 14% and 8% versus 12%, respecively). Also, he average size of a ransacion and he posiion of he financial grouping i.e. local banks, financial firms, and foreign insiuions (mainly inernaional banks) differ from hose of he commercial grouping i.e. exporers, imporers, and insiuional invesors. In his regard, he negaive posiion of exporers and insiuional invesors (hedging fuure FX incomes) and he posiive posiion of imporers (hedging fuure FX expendiure) are much larger han he posiions of he financial grouping (in absolue erms). The differences beween he financial and commercial groupings can also be seen in DTE (panel b), HET (panel d), CHG (panel f), and REL (panel g). The players of financial grouping can be characerized as follows: hey end o rade in shor-erm insrumens (DTE < 10 days), heir heerogeneiy is close o zero, hey frequenly change 5
posiions (CHG > 0.06), and heir relaive ne flow is larger han ha of oher secors (REL > 0.74). In conras, he RND of financial firms was unexpecedly low. The evidence in Table 1 shows differen rading characerisics bu do no reveal how, if any, he players' behavior changed during oulier days. Figure 1 presens he oulier days during he period 2008-2010 and he ILS/USD exchange rae changes divided by IV. [Ener Figure 1 here] There were 48 oulier days (wih red markers) and 690 normal days. Oulier days were defined by he above 3 marke variables (ILS/USD exchange rae changes, BAS, and IV) using mulivariae oulier deecion echniques suggesed by Galeano, Pena, and Tsay (2006). This echnique, which is described in Appendix 1, akes several relaed variables (mulivariae ime series) and projecs hem ono a single series wihou requiring pre-deermined specificaions of he mulivariae model or he ARMA process. The oulier days were derived from ha single series using LR ess. I can be seen ha mos oulier days are concenraed in clusers paricularly in 2008 (26 days). During 2009 and 2010 here were only 12 and 10 oulier days, respecively. As can be seen from he figure, here are spikes in he exchange rae changes divided by IV above he oulier days (red markers). 4. The relaionship beween marke variables and players' ne purchases and posiions As menioned above, speculaive aciviy has a number of unique rading characerisics however, i can also be correlaed wih marke variables (exchange rae changes, IV, and BAS). In oher words, alhough i mainly influences marke variables by iniiaing a move oward an appreciaion or depreciaion, speculaive aciviy can also be influenced by marke variables. In order o examine he relaionships beween marke variables and he players' ne purchases and posiions, a wo-equaion sysem was esimaed for each of he 366 players, using TSLS: (1a) (1b) X 1 1NFL 1App NFL 1Dep NFL 1POS 1 1 2 2X 2POS 1 NFL 2 Where, X is he projeced series derived from ILS/USD exchange rae changes, IV, and BAS using Galeano, Pena, and Tsay (2006)'s mulivariae oulier deecion echnique, NFL is he ne purchase, App and Dep are dummy variables for appreciaion and depreciaion oulier days, respecively, which ake on he value 1 for an oulier and 0 oherwise, POS is he FX posiion, and is he noise erm. Table 2 presens he regression resuls broken down by speculaor ype. [Ener Table 2 here] 6
Panel (a) presens some basic saisics for speculaor ype and panel (b) presens he TSLS regression resuls. The four ypes of speculaors are defined as follows: (1) Pro (wih he rend) players ha sell USD on oulier appreciaion days and buy on oulier depreciaion days; (2) Con (conrarian) players ha buy USD on oulier appreciaion days and sell on oulier depreciaion days; (3) Sell players ha sell USD on boh appreciaion and depreciaion oulier days; and (4) Buy players ha buy USD on boh appreciaion and depreciaion oulier days. The upper par of panel (a) presens he number of players in each caegory. Of he 366 regressions (one for each player), only 212 yielded reasonable resuls (posiive Adj. R 2 and number of observaions > 50), of which 44 were for Pro players and 43 were for Con players. The oher wo ypes of speculaors, i.e., Buy and Sell players are less imporan since heir behavior is consisen across he sample period regardless of wheher he oulier day was an appreciaion or depreciaion day. According o he mehodology used, a player is classified as a speculaor only if he has boh speculaive characerisics and a srong relaionship wih marke variables, especially near oulier days. Thus, among he 212 poenial speculaors by he regression resuls,162 had also an exreme speculaive rading characerisics. Thus, 76% of he players fulfilled boh condiions. The lower par of panel (a) presens he same informaion as he upper par excep wih he consrain ha he significance level for he main exogenous variables, i.e. app*nfl and dep*nfl, exceeded 0.95. In his case, only 76 regressions yielded reasonable resuls wih he above wo consrains binding; of hese, 19 were Pro (rend) players and 27 Con players. The combined lis of speculaors consised of 47 players; of hese, 11 Pro and 14 Con players (58% and 52% respecively of he reasonable regressions under he consrains). This is no a random resul as he probabiliy of randomly selecing he very same players ha fulfill boh condiions is close o zero. 3 An addiional advanage of using his lis is he combinaion of wo independen sources of daa: he exchange rades sysem for players' rading characerisics and he exchange rae sysem for marke variables. Alhough all he seleced speculaors fulfilled boh condiions, i.e. hey had he rading characerisics of speculaors and hey influenced marke variables, moneary auhoriies usually focus on Pro players since hey are he ones ha iniiae rends and influence marke variables. 3 The probabiliy o randomly selecing he very same 47 players from a given group of 76 players regardless 1 1 of he inernal order is: 0. 76 76! 47 47!*29! 7
Panel (b) presens means of coefficiens and T-saisics of he TSLS regression of he four speculaor ypes. By consrucion, he main exogenous variables in equaion (1a) (given he consrains of a significance level greaer han 0.95) i.e. app*nfl and dep*nfl, are significan. For example, on appreciaion oulier days, he mean coefficien of Pro players is 3.38 (Prob. < 0.001) while ha of Con players is 4.00 (Prob. < 0.001). Oher variables in equaion (1a) were insignifican, including he NFL and POS -1. As he focus of he regressions is on speculaive aciviy, especially on oulier days, hese resuls indicae ha speculaors' behave differenly on oulier days han on normal days. In general, he coefficiens in equaion (1b) were insignifican, implying ha players' ne purchases were no influenced by marke variables, on average. Thus, he significance level of equaion (1b) is much lower han ha of (1a). 5. Analysis of he Pro players Having divided he players from he combined lis ino four ypes of speculaors, i was possible o examine he influence of each ype on marke variables, especially on oulier days, and he abiliy o forecas oulier days by racking heir behavior. This was done by esimaing he following Fixed Effec pooled regressions (which have cross-secion weighs and Whie's Heeroscedasiciy consisen sandard errors and co-variances) for each player ype: (2) X i NFL i i POS i i App NFL i i Dep NFL 3 Dep i POSi i X i 1 Ci i 3 1 where, C is he fixed effec parameer. Table 3 presens he resuls. [Ener Table 3 here] As can be seen, all he ypes behaved as expeced; he coefficiens of Pro (Con) players were significanly and negaive (posiive) on oulier appreciaion days () and significanly and posiive (negaive) on oulier depreciaion days (). The oher wo groups also behaved as expeced; he coefficiens of Sell (Buy) players are significanly negaive (posiive) on boh appreciaion and depreciaion oulier days. An indicaion of he abiliy o forecas speculaors' behavior can be obained from he lagged variables. The coefficiens of he Pro player posiions ( for appreciaion and for depreciaion) during he hree days prior o an oulier appreciaion (depreciaion) day are significan and negaive (posiive) a he 0.95 significance level and he respecive coefficiens of he Con players were also found o be significan, wih he opposie sign. Ineresingly, he coefficiens of he lagged variables for Sell and Buy players were i i App i 3 3 1 POS i 8
insignifican. The overall significance level of he pooled Fixed Effec regression for Pro players was much higher han ha of he oher ypes while he coefficiens for he paricular player in all caegories found insignifican. In oher words, he four ypes were quie homogenous. Thus, moneary auhoriies can choose o rack only Pro players, in order o forecas oulier days. Moreover, he resuls indicae ha i is Pro players ha iniiae a marke rend, which ends wih an oulier day, while oher groups reac o he move, paricularly, Con players whose behavior is conrarian. Figure 2 presens he behavior (i.e. mean posiions) of 11 Pro, 14 Con, 14 Sell, and 8 Buy players near 10 oulier days during he ou-of-sample period (he year 2010). During ha period, here were 10 oulier days derived by he same echnique as he oulier days during he in-sample period. The 10 oulier days were evenly divided beween appreciaion and depreciaion days. [Ener Figure 2 here] The upper par of he figure presens mean posiions of he various player ypes around oulier appreciaion days while he lower par presens mean posiions around oulier depreciaion days. As can be seen from he figure, Pro players iniial posiion was around zero compared o oher ypes especially Con players. In addiion, Pro players acually changed heir posiion before an oulier day in he direcion of he move while Con players changed heir posiion in he opposie direcion, paricularly on appreciaion oulier days. Moreover, only Pro players and o some exen Con players reurned o heir iniial posiion following an oulier appreciaion days. This ends o confirm heir saus as iniiaors of rends in he FX marke. The behavior of oher ypes of players was less volaile and did no exhibi his ype of behavior near oulier days. 9
6. Summary Alhough he presence of speculaors usually increases he liquidiy and sophisicaion of an FX marke, speculaive aciviy, such as carry rade, is ofen no o he benefi of an economy, paricularly small open ones. I is argued ha in small open economies speculaors cause sharp movemens in exchange raes and oher marke variables such as he Bid-Ask Spread (BAS) and Implied Volailiy (IV). Thus, i is in he ineress of he moneary auhoriies o idenify and rack a manageable amoun of speculaors. Tracking speculaors is of high imporance during 'oulier days', which are defined in his paper by exchange rae changes, IV, and BAS, using mulivariae oulier deecion echniques. The paper suggess a mehodology for idenifying speculaors in FX markes by examining boh he speculaive characerisics of players and he relaionship beween marke variables and a player's ne purchases and posiions. A player is idenified as a speculaor only if his speculaive characerisics are exreme in comparison o oher players and his influence on he above marke variables is significan. The daa consised of daily observaions for he years 2008-9, which were obained from wo sources: he exchange rades sysem for players' rading characerisics and daa from he foreign currency marke, including changes in he ILS/USD rae, he BAS, and he IV. The findings indicae ha he financial groping i.e. domesic banks, foreign insiuions, and financial firms, differ significanly from oher secors in a number of imporan characerisics ha were used o idenify speculaive aciviy. The applicaion of he proposed mehodology o Israel's FX marke idenified 47 speculaors who fulfilled boh condiions: heir speculaive characerisics were exreme in comparison o oher large players and heir influence on he marke variables, were significan. An analysis of heir aciviy during 2008-9 revealed ha 11 of hem purchased FX prior o oulier depreciaion days and sold FX prior o oulying appreciaion days. This indicaes ha here were speculaors who iniiaed a marke move, or a leas joined in, prior o an oulier appreciaion or depreciaion day. Based on he classificaion of speculaors ino ypes during he sample period (2008-2009), i was possible o idenify heir speculaive behavior prior o oulying days during an ou-of-sample period i.e. he year 2010. 01
References Galeano, P., D. Pena, and R.S. Tzay (2006), Oulier deecion in mulivariae ime series by projecion pursui, Journal of he American Saisical Associaion Vol. 101, 654-669. Graham, B. and D. Dodd (1951), Securiy analysis: Principles and Technique, McGraw-Hill. Grossberg, A. and B. Schreiber (2005), Speculaive aciviy in foreign currency, credi and deposis by residens of Israel, Israel Economic Review, pp. 95 130. Lemmon, M. and S. Ni (2008), The effecs of invesor senimen on speculaive rading and prices of sock and index opions. Unpublished Working Paper, Universiy of Uah. Osler, C.L. (2006), Macro lessons from microsrucure, Inernaional Journal of Finance and Economics 11, 55-80. Osler, C.L. (2008), Foreign exchange microsrucure: A survey of he empirical lieraure, Encyclopedia of Complexiy and Sysem Science, Springer. 00
Appendix 1: Deecing Ouliers in Mulivariae Time Series The following mehod deecs ouliers in mulivariae cross-dependen and auocorrelaed ime series. The basic idea, which was developed by Galeano, Pena, and Tsay (2006), is o projec mulivariae ime series ono a single univariae series, which makes i possible o use hen any sandard oulier deecion echnique. This echnique was used o projec hree relaed marke variables (ILS/USD exchange rae changes, BAS and IV) ono a single projeced series X, which made i possible o deec ouliers using a Likelihood Raio Tes (LRT). Galeano, Pena, and Tsay (2006) show ha he projecion, which maximizes or minimizes he kurosis coefficien of he projeced series, yields he direcion ha maximizes he relaive size of he oulier, i.e. he raio beween he oulier size and he variance of he projeced observaions. 2k-1 orhogonal projecions were added o deal wih any idenificaion problems ha migh arise from he possible occurrence of muliple ouliers. The following algorihm formally describes his mehod: Suppose ha Y Y Y,..., Y ' 1 T series. 1, 2 k,..., is he observed k dimensional vecor ( m ) 1. Le m = 1 and Z Y. 2. Le ( M ) Z 1 T T 1 Z ( m ) ( m ) Z Z ( m ) ' ( m ) Z and find V m such ha v m 1 arg max ' ( m) v v 1T m m Z T 1 ' v m Z ( m) ( m) Z 4 ( m1 ) ( m ) 3. If m = k hen sop; oherwise define Z I v v m ' m ( m ) Z Z, ha is, ( m1 ) Z is he projecion of he observaions in an orhogonal direcion ono v m. Le m = m + 1 and go o sep 2. 4. Repea he same procedure o minimize he objecive funcion in sep 2, in order o obain anoher se of k direcions - v,..., k 1 v2k. The nex sage is o search for ouliers in he univariae projeced series. This can be done using any sandard procedure, such as he Likelihood Raio Tes (LRT), as recommended by he auhors. 02
Table 1 Basic saisics of he secors' rading aciviy in he Israeli FX OTC marke: 2008-2009 local banks financial grouping financial firms foreign residens households commercial grouping expor firms insiuional invesors impor firms ohers oal # players 24 52 163 10 8 50 43 16 366 % of oal 7% 14% 45% 3% 2% 14% 12% 4% 100% # acive days 8,711 10,904 29,405 1,304 1,382 6,381 5,555 2,004 65,646 % of oal 13% 17% 45% 2% 2% 10% 8% 3% 100% a) Ne flow (NFL), USD millions Mean -0.52 0.13-0.13 0.07-1.03-2.83 4.53 0.56 0.00 Max 257 151 285 20 133 125 376 58 376 Min -281-239 -631-28 -150-105 -175-100 -631 Sd. 20 11 13 3 19 8 13 6 13 b) Posiion (POS), USD millions Mean -3.11-1.76 2.57 3.10-85.81-22.60 24.90 2.09-1.17 Max 300 227 646 68 351 194 427 153 646 Min -296-506 -584-18 -652-1031 -103-147 -1031 Sd. 43 26 26 9 156 48 41 22 44 c) Days To Expiraion (DTE) Mean 9.22 4.84 5.90 18.77 20.95 20.82 24.89 18.25 10.17 Max 369 301 859 253 2002 1529 584 278 2002 Min 0 0 0 0 0 0 0 0 0 Sd. 22 17 21 32 70 51 48 31 31 d) Heerogeneiy (HET) Mean -0.05 0.00-0.03 0.00 0.07-0.64 0.57 0.06-0.03 Max 1 1 1 1 1 1 1 1 1 Min -1-1 -1-1 -1-1 -1-1 -1 Sd. 0.61 0.81 0.86 0.81 0.91 0.72 0.74 0.93 0.84 e) Round ransacions (RND) Mean 0.69 0.26 0.81 0.41 0.37 0.45 0.35 0.40 0.60 Max 1 1 1 1 1 1 1 1 1 Min 0 0 0 0 0 0 0 0 0 Sd. 0.35 0.40 0.36 0.45 0.46 0.48 0.46 0.47 0.45 f) Changing Posiion (CHG) Mean 0.12 0.08 0.07 0.06 0.02 0.01 0.01 0.03 0.06 Max 1 1 1 1 1 1 1 1 1 Min 0 0 0 0 0 0 0 0 0 Sd. 0.32 0.27 0.26 0.23 0.14 0.08 0.12 0.17 0.23 g) Relaive Ne Flow (REL) Mean 1.26 0.76 0.75 0.45 0.16 0.14 0.20 0.24 0.56 Max 700 373 2400 200 21 500 401 100 2400 Min 0 0 0 0 0 0 0 0 0 Sd. 11.70 5.50 14.31 5.02 0.87 3.71 3.89 1.84 9.87 This able presens some aciviies of he main 8 secors. Ne flow (panel a) calculaes as all buy flows - sell flows. In (panel b), Posiion includes all insrumens (spo, TOM, same day, and fuures and opions in dela values) excep swaps. Heereogeneiy is calculaed (panel d) as (buy - sell)/(buy + sell) so, he maximal and minimal values are +1 and -1, respecively. Round ransacions represens he raio of ransacions of a muliple of one million USD ou of all ransacions. Changing posiion (panel f) means ha a player changed he posiion from long o shor or vice versa (change = 1, no change = 0). Relaive Ne Flow (panel g) is calculaed as Ne flow/posiion in absolue erms. 03
Table 2 TSLS regression resuls of four ypes of speculaors, 2008-2009 Panel A Pro players Con players Sell players Buy players All players (buy in depreciaion days & sell in appreciaion days) (sell in depreciaion days & buy in appreciaion days) (sell in all oulying days) (buy in all oulying days) # all coefficiens 58 60 53 41 212 # all coefficiens & speculaive characerisics 44 43 40 35 162 % speculaors (based on he combined lis) 76 72 75 85 76 # robus coefficiens (0.95) 19 27 18 12 76 # robus coefficiens & speculaive characerisics 11 14 14 8 47 % speculaors (based on he combined lis) 58 52 78 67 62 Panel B (1a) Endogenous variable: X 1 0.07-0.02-0.15-0.07-0.03 (0.63) (0.06) (-0.67) (-0.15) (0.04) 1-3.38 4.00-4.24 3.13-0.20 (-4.74) (4.03) (-3.99) (3.44) (-0.16) 1 13.35-4.34-3.81 4.34 0.58 (3.56) (-3.24) (-3.03) (2.69) (-0.20) 1 0.10-0.03-0.04-0.04 0.00 (0.45) (-0.35) (0.00) (0.12) (-0.03) 1-0.87-0.05-0.50-0.29-0.45 (-0.92) (-0.14) (-0.55) (-0.04) (-0.53) Adj. R 2 0.30 0.28 0.35 0.29 0.19 D.W. 1.80 1.79 1.79 1.85 1.80 (1b) Endogenous variable: NFL 2 0.08-0.01-0.17 0.19 0.02 (0.35) (0.04) (-1.56) (0.96) (-0.08) 2 0.01-0.01 0.03-0.02 0.00 (0.17) (-0.19) (0.21) (-0.23) (0.23) 2 5.66-3.38-0.49-0.41 0.44 (2.91) (-3.78) (-0.51) (-0.73) (-0.35) Adj. R 2 0.10 0.03 0.09 0.06 0.05 D.W. 1.93 2.00 1.94 1.92 1.93 This able presens he resuls of he following wo equaions sysem: (1a) 1NFL 1App NFL 1Dep NFL 1POS (1b) NFL 2 2 X 2 POS 1 2 X 1 1 1 Where, X is he projeced series derived from exchange rae changes, IV, and BAS using he Galeano e al. (2006) echnique. NFL is ne purchases, App and Dep are dummy variables for appreciaion and depreciaion oulying days, respecively, POS is he FX posiion, BAS is he Bid-Ask Spread, and e is he noise erm. By he TSLS regressions only 76 ou of 366 players were found significan a he 0.05 level in he main variables i.e. App*NFL and Dep*NFL. Of he 76 players 11 were defined as Pro (rend) players and 14 as Con (conrarians) by he combined crieria i.e. robus coefficiens & speculaive characerisics. 04
Table 3 Pooled (Fixed effec) regression resuls of four ypes of players, 2008-2009 Pro players Con players Sell players Buy players All players (buy in depreciaion days & (sell in depreciaion days & (sell in all oulying (buy in all oulying sell in appreciaion days) buy in appreciaion days) days) days) # robus coefficiens (0.95) & speculaive characerisics 11 14 14 8 47 Endogenous variable: X i 0.02 0.01 0.00 0.02 0.00 (2.51) (0.46) (0.05) (1.62) (1.95) i 0.00-0.01 0.00 0.01 0.00 (0.50) (-1.85) (-0.82) (1.19) (0.94) i -0.49 0.11-0.68 0.34-0.01 (-3.09) (1.34) (-4.21) (3.79) (-0.47) i 0.11-0.70-1.08 0.80 0.04 (2.29) (-2.00) (-4.57) (3.50) (1.59) i -0.05 0.02-0.01-0.08-0.05 (-5.78) (2.07) (-0.33) (-1.50) (-5.13) i 0.07-0.01 0.02-0.01 0.02 (5.63) (-2.76) (1.58) (-0.27) (2.16) i -0.77 0.23-0.20-0.20 0.24 (5.63) (6.21) (7.51) (3.83) (23.25) Adj. R 2 0.16 0.06 0.09 0.03 0.03 D.W. 1.95 1.94 1.88 1.96 1.91 This able presens he resuls of a pooled fixed effec esimaon: X i App infl i ipos i i App NFLi idepnfl i i 3 3 3 POS i i POS i i Xi 1 1 3 1 C Where, X is he projeced series derived from exchange rae changes, IV, and BAS using he Galeano e al. (2006) echnique. NFL is he purchases, App and Dep are dummy variables for appreciaion and depreciaion oulying days, respecively, POS is he FX posiion, C is he fixed effec coefficien, and e is he noise erm. The player groups are derived from Table 2 i.e. found significan a he 0.95 level in he main variables (App*NFL and Dep*NFL) and had speculaive characerisics (he combined crieria). Dep i i 05
Figure 1 Oulier days and he daily changes in ILS/USD exchange rae divided by (IV) 25 20 15 10 5 0-5 -10-15 -20-25 Oulier days are represened by red squared markers. 06
Figure 2 Mean posiions (USD, millions) of four player ypes around oulier days, 2010 20 Appreciaion days 10 0 Pro Con Sell Buy All -10-20 -30-40 day -4 day -3 day -2 day -1 Oulying Day day -1 day -2 day -3 Depreciaion days 25 15 5 Pro Con Sell Buy All -5-15 -25-35 -45 day -4 day -3 day -2 day -1 Oulying Day day -1 day -2 day -3 The figure presens posiions of he four player ypes saring four days before he oulying days and ending hree days afer. The examined period is 2010 which is an ou-of-sample period. The posiions of Pro players are more volaile compared o oher ypes boh in appreciaion and depreciaion oulier days. 07