Exchange Rate Uncertainty and International Portfolio Flows

Transcription

1 Economcs and Fnance Workng Paper Seres Department of Economcs and Fnance Workng Paper No Guglelmo Mara Caporale, Faek Menla Al and Ncola Spagnolo Exchange Rate Uncertanty and Internatonal Portfolo Flows May 13

2 Exchange Rate Uncertanty and Internatonal Portfolo Flows Guglelmo Mara Caporale a,b,c, Faek Menla Al a and Ncola Spagnolo a,d a Department of Economcs and Fnance, Brunel Unversty, London, UK b CESfo, Munch, Germany c DIW Berln, Germany d Centre for Appled Macroeconomc Analyss (CAMA), Canberra, Australa Aprl 13 Abstract Ths paper examnes the mpact of exchange rate uncertanty on dfferent components of portfolo flows, namely equty and bond flows, as well as the dynamc lnkages between exchange rate volatlty and the varablty of these two types of flows. Specfcally, a bvarate GARCH-BEKK-n-mean model s estmated usng blateral data for the US vs-à-vs Australa, the UK, Japan, Canada, the euro area, and Sweden over the perod 1988:1-11:1. The results ndcate that the effect of exchange rate uncertanty on equty flows s negatve n the euro area, the UK and Sweden, and postve n Australa, whlst t s negatve n all countres except Canada (where t s postve) n the case of bond flows. Under the assumpton of rsk averson, ths suggests that exchange rate uncertanty nduces a home bas and causes nvestors to reduce ther fnancng actvtes to maxmse returns and mnmse exposure to uncertanty. Furthermore, snce exchange rate volatlty and the varablty of flows are nterlnked, exchange rate or credt controls on these flows can be used to pursue economc and fnancal stablty. Keywords: Exchange rate uncertanty, Equty flows, Bond flows, Causalty-n-varance JEL Classfcaton: F31, F3, G15 Correspondng author. Emal: Guglelmo-Mara.Caporale@brunel.ac.uk

3 1. Introducton The macroeconomc effects of exchange rate uncertanty, especally on trade flows, have attracted consderable attenton snce the collapse of the Bretton Woods system n 1971 and the adopton of floatng exchange rates n March 1973, both n the theoretcal and emprcal lterature (see McKenze, 1999, for a comprehensve revew). By contrast, the mpact at the mcro level on equty and bond flows has yet to be nvestgated emprcally. In an nfluental study, Hau and Rey (6) develop an equlbrum framework n whch exchange rate returns, equty returns, and captal flows are jontly determned under ncomplete foregn exchange rsk tradng. Ther analyss s motvated by the recent mcrostructure approach to exchange rate determnaton whch has been shown to mprove remarkably the performance of exchange rate models, wth currency order flows explanng a substantal proporton of exchange rate changes (see, e.g., Evans and Lyons, ; 5; 8; Payne, 3; Rme et al., 1; Chnn and Moore, 11; and Duffuor et al., 1 among others). In addton, they argue that currency order flows and portfolo flows are ntmately related wthn the portfolo rebalancng framework snce they both reflect nvestors behavour. However, whle ther paper provdes a theoretcal framework for analysng the mplcatons of ncomplete foregn exchange rsk tradng for the correlaton structure of exchange rate changes and equty returns as well as net portfolo flows, t does 1

4 not nclude statstcal tests for the mpact of exchange rate uncertanty on portfolo flows across borders. The underlyng dea s that exchange rate volatlty ncreases transacton costs and reduces potental gans from nternatonal dversfcaton by makng the acquston of foregn securtes such as bonds and equtes more rsky, whch n turn affects negatvely portfolo flows across borders. Indeed, Eun and Resnck (1988) had prevously shown that exchange rate uncertanty s non-dversfable and has an adverse mpact on the performance of nternatonal portfolos. Ths fndng s also consstent wth the evdence presented n the study by Levch et al. (1999), who found, by surveyng 98 US nsttutonal nvestors, that foregn exchange rsk hedgng consttutes only 8% of total foregn equty nvestment. However, Eun and Resnck (1988) suggest that hedgng through forward exchange contracts and multcurrency dversfcaton are effectve ways to reduce exchange rate rsk. Glen and Joron (1993) and Eun and Resnck (1994) further provde evdence that hedgng n the forward exchange markets mproves the performance of dversfed portfolos of equtes and bonds. The present study makes a fourfold contrbuton to the exstng lterature. Frst, t analyses emprcally whether exchange rate uncertanty affects nternatonal portfolo flows and ther varablty. It s n fact the frst emprcal nvestgaton of ths knd, based on blateral monthly data for the US vs-à-vs sx developed economes, namely Australa,

5 Canada, the euro area, Japan, Sweden, and the UK over the perod 1988:1-11:1. Second, unlke Hau and Rey (6) who assume that the supply of bonds s nfntely elastc, thereby smplfyng the dynamcs of bond acqustons n ther model, we examne the mpact of exchange rate uncertanty on bond and equty flows (as well as ther varablty) n turn. In ths way, we are able to evaluate the mpact of uncertanty on the ndvdual components of portfolo flows across borders. Accordng to Hau and Rey (6), exchange rate uncertanty should affect equty, but not bond flows; we provde some relevant emprcal evdence on ths ssue. Thrd, exstng emprcal studes on the relatonshp between exchange rate changes and portfolo flows nvestgate short-run dynamc nteractons only wth lnear dependence technques (.e., frst moment analyss). For example, Brooks et al. (4) and Hau and Rey (6) use smple correlatons and regresson analyss for the US vs-à-vs the euro area and Japan, and 17 OECD countres respectvely; Sourouns (4), Chaban (9), and Kodong and Ojah (1) estmate VAR models for four developed countres (the UK, Japan, Germany, and Swtzerland), three ol-exportng countres (Canada, Australa, and New Zealand), and four Afrcan countres (Egypt, Morocco, Ngera, and South Afrca) vs-à-vs the US. Ther results are charactersed by sgnfcant devatons from normalty and condtonal heteroscedastcty,.e. volatlty clusterng or the so-called ARCH effects (see Engle, 198) that are not captured by ther setup. By contrast, we model frst and second 3

6 moments smultaneously to analyse the dynamc nteractons between exchange rate changes and portfolo flows, n ths way avodng the potental ptfalls of earler studes. Fourth, snce volatlty s a measure of the nformaton flow (see Ross, 1989), t s of paramount mportance to understand how the stochastc nformaton arrvals n the form of smple portfolo nvestment shfts n bonds and equtes are transmtted to the foregn exchange market, and vceversa. Our analyss sheds lght on ths mechansm and thus provdes mportant nformaton to polcy-makers and regulators to formulate approprate polces based on mposng or relaxng credt controls on these flows dependng on the state of the economy, wth the am of achevng economc and fnancal stablty. The remander of the paper s organsed as follows. Secton descrbes the data and reports some descrptve statstcs. Secton 3 outlnes the econometrc model. Secton 4 dscusses the emprcal results, and fnally Secton 5 concludes.. The econometrc model We employ a bvarate VAR-GARCH (1, 1) n the BEKK specfcaton (Engle and Kroner, 1995) allowng for n-mean effects n order to examne the mpact of exchange rate uncertanty on equty and bond flows as well as the dynamc lnkages n the frst and second moments of these varables over the perod 1988:1-11:1. Varous lags of exchange rate 4

7 volatlty affectng the condtonal mean of equty and bond flows are ncluded n the specfcaton to avod the potental ptfalls of models allowng only for contemporaneous nteractons. The economc nterpretaton s that t mght take some tme for the nvestors response to exchange rate volatlty to be ncorporated nto ther strateges. Therefore the condtonal mean equaton s specfed as follows: y t p 1 y t p 1 h t t 1 ; ; ; t 1, t, t (1) where =[,, E t and EF t (BF t ) ndcate respectvely exchange rate changes and net equty (bond) flows. =[,,,, h 11,t and h,t represent the condtonal varances of exchange rate changes and net flows dependng on whether equtes or bonds respectvely are consdered. The parameters, measure the response of exchange rate changes and net flows to ther own lags, whlst, represent the mean spllovers from exchange rate changes to net flows, and vceversa. If the parameter s sgnfcantly dfferent from zero, ths mples that exchange rate uncertanty affects equty flows and/or bond flows. The nnovatons vector s assumed to be normally dstrbuted ~, wth ts correspondng varance-covarance matrx gven by ; s the nformaton set 5

8 avalable at tme t-1. Lags are ncluded sequentally n Equ. (1) untl seral correlaton s removed by employng the Hoskng (1981) multvarate Q-statstcs on the standardsed resduals / for = 1,. Note that contegraton tests between exchange rates and net flows have not been carred out as the former appear to be I (1) n most cases, whlst both equty and bond flows follow I () processes 1 (see Fg. 1). Hence, an error correcton term s not ncluded n Equ. (1). Havng specfed the condtonal mean equaton, we then estmate the multvarate GARCH model n ts BEKK representaton, ths beng a straghtforward generalsaton of the unvarate GARCH model of Bollerlslev (1986). The BEKK specfcaton has advantages compared to other multvarate GARCH specfcatons such as the VEC-GARCH model of Bollerslev et al. (1988) because of ts quadratc forms ensurng that the condtonal covarance matrces n the system are postve defnte. Unlke the Dynamc Condtonal Correlaton model of Engle (), whch estmates the tme-varyng correlatons drectly, the BEKK specfcaton allows for tme-varyng correlatons and also for nteractons between the varances n a lead-lag framework. Furthermore, the curse of dmensonalty hghlghted by Caporn and McAleer (1) s not a serous ssue n the present case wth only two varables. The model can be represented as follows: 1 Ths s confrmed by a battery of unt root tests; the results are avalable from the authors on request. For a survey on multvarate GARCH models, see Bauwens et al. (6). 6

9 H CC A A BH B () t t 1 t 1 t 1 In matrx form, t can be specfed as: H H 11, t 1, t H H c C c 1, t 11 1, t u CC A u, t c a, A a , t 1 1 a a u 1 1, t 1 u, B 1, t 1 u u, t 1, t 1 b b 11 1 H A B H b b 1 11, t 1 1, t 1 H H 1, t 1, t 1 B (3) where C s constraned to be a lower trangular matrx and A and B are respectvely ARCH and GARCH parameter matrces. Equ. (3) shows that n the BEKK model each condtonal varance and covarance n H t s modelled as a functon of lagged condtonal varances and covarances, lagged squared nnovatons and the cross-product of the nnovatons. Volatlty s transmtted between exchange rate changes and net equty/bond flows through two channels represented by the off-dagonal parameters n the ARCH and GARCH matrces: a symmetrc shock, and the condtonal varance,. Volatlty transmsson from exchange rate changes to net equty/bond flows can be analysed by testng the null hypothess, and n the opposte drecton. Such causalty-n-varance tests wthn the multvarate GARCH-BEKK models have superor power to the cross correlaton functon 7

10 (CCF) two-step approach of Cheung and Ng (1996) (see Hafner and Hewartz, 8). Causalty-n-varance s tested usng the followng lkelhood rato test statstc: LR = (L r L ur ) x df (4) where L r and L ur ndcate the restrcted and unrestrcted log-lkelhood test statstc; LR follows the ch-squared dstrbuton wth degrees of freedom equal to the number of the restrcted coeffcents (df). Gven that, as stated earler, the nnovatons are assumed to be normally dstrbuted, the log lkelhood functon for such a model s gven by: Tn 1 L( ) ln( ) t t1 (lnh t H t 1 t ) t (5) where n s the number of equatons, two n our case; T s the number of observatons, whch s 87; and s a vector of unknown parameters to be computed. More specfcally, we use the Quas-Maxmum Lkelhood (QML) method of Bollerslev and Woolbrdge (199) to calculate the standard errors that are robust to devatons from normalty. 3 As a fnal check of the 3 We use the SIMPLEX free-dervatve method, whch s useful to mprove the ntal values, and then the BFGS standard algorthm to obtan the standard errors (see Engle and Kroner, 1995; Kearney and Patton, among others). Ths procedure was mplemented wth a convergence crteron of.1. 8

11 adequacy of the estmated model we employ the Hoskng (1981) multvarate Q-statstc for the standardsed squared resduals to evaluate whether or not the ARCH and GARCH dynamcs have been approprately captured n the condtonal varance equaton, Equ. (3). 3. Data descrpton We examne the mpact of exchange rate uncertanty on dfferent components of portfolo flows, namely equty and bond flows, as well as the dynamc lnkages between these flows and exchange rate changes for the US vs-à-vs the UK, Japan, Canada, Australa, Sweden, and the euro area. Throughout, the US s consdered the domestc or home economy. Snce the data on portfolo nvestment flows, obtaned from the US Treasury Internatonal Captal (TIC) System, 4 are sampled at a monthly frequency, we employ monthly data from 1988:1 to 11:1 for all seres. The reason for selectng ths start date s that portfolo flows for the perod precedng 1988 are known to be nsgnfcant (see Brooks et al., 4). Net equty (bond) flows are calculated as equty (bond) nflows mnus outflows. Whle nflows are measured as net purchases and sales of domestc (US) assets (equtes and bonds) by foregn resdents, outflows are measured as net purchases and sales of foregn assets (equtes and bonds) by domestc resdents (US). Wth regard to the euro area, we aggregate the data for the 4 They are retreved from the US Treasury Department webste 9

12 ndvdual EMU countres (Austra, Belgum-Luxemburg, Fnland, France, Germany, Ireland, Italy, the Netherlands, Portugal, Span) to extract cross-border bond and equty flows between the US and ths regon. Postve numbers mply net equty and bond nflows (n mllons of US dollars) towards the US or outflows from the counterpart countres. Followng Brennan and Cao (1997), Hau and Rey (6), and Chaban (9) among others, we normalse these flows usng the average of ther absolute values over the prevous 1 months, snce wthout scalng model convergence s dffcult to acheve. The exchange rates are end of perod data, defned as US dollars per unt of foregn currency; the source s the IMF s Internatonal Fnancal Statstcs (IFS). Exchange rate changes are calculated as 1, /, where P E,t represents the log of the exchange rate at tme t. For the perod precedng the ncepton of the euro,.e. before 1999, we use US dollar per ECU as the euro area s exchange rate. Descrptve statstcs are dsplayed n Table 1. The mean of monthly exchange rate changes s postve (US dollar deprecaton) for Japan and Canada, and negatve (US dollar apprecaton) for the rest of the countres. On the other hand, the monthly mean of net equty flows s postve for Sweden and Canada and negatve for the remanng countres, ndcatng equty nflows from Sweden and Canada towards the US and outflows from the US towards the other countres. The monthly mean of net bond flows s negatve for Australa and postve 1

13 for the other countres. Ths ndcates the exstence of bond nflows from all countres except Australa (for whch there s evdence of bond outflows) vs-à-vs the US. Exchange rate changes are found to exhbt hgher volatlty than the two flows. Furthermore, equty flows appear to be charactersed by hgher volatlty than bond flows (although ther volume s very small). As for the thrd and fourth moments, exchange rate changes, net equty flows, and net bond flows all exhbt skewness and excess kurtoss n most cases. The Jarque-Bera (JB) test statstcs mply a rejecton at the 1% level of the null hypothess that exchange rate changes and the two flows are normally dstrbuted n all countres n queston. [Insert Table 1 about here] Fg. 1 shows monthly exchange rate changes, net equty flows and net bond flows for all countres over the perod under nvestgaton. Volatlty clusterng s clearly present n all cases, suggestng that an ARCH model mght be requred to capture t. The seres also appear to be covarance statonary. [Insert Fg. 1 about here] 4. Emprcal results 11

14 The mcrostructure approach s partcularly suted to analysng the relatonshp between bond and equty flows and exchange rate changes. The objectve of our analyss s to establsh whether exchange rate uncertanty affects equty and bond flows across borders, and also whether there s a volatlty transmsson (hence nformaton flows) between these flows and exchange rate changes and, f so, n what drecton causalty runs. The QML estmates of the bvarate GARCH (1, 1) BEKK parameters as well as the assocated multvarate Q-statstcs (Hoskng, 1981) are dsplayed n Tables 7 for Australa, Canada, the euro area, Japan, Sweden, and the UK, respectvely. Panel A and B n each Table concern the bvarate regresson of exchange rate changes aganst equty and bond flows respectvely. The Hoskng multvarate Q-statstcs of order (6) and (1) for the standardsed resduals n the exchange rate changes-equty flows equaton ndcate the exstence of no seral correlaton at the 5% level, when the condtonal mean equatons are specfed wth p=1 for Japan, p= for Sweden and p=3 for the other countres (the nsgnfcant parameters n the mean equatons have been dropped). Wth regard to the exchange rate changes-bond flows relatonshp, whlst no dynamc terms appear to be necessary for Sweden, settng p=1 for the UK, p= for the euro area, p=3 for Australa and Canada and p=5 for Japan s requred to capture adequately the dynamc structure n these cases. [Insert Tables -7 about here] 1

15 As can be seen from the Tables, the dynamc nteractons between exchange rate changes and net equty and bond flows, captured by and, suggest that there exst lmted dynamc lnkages between the frst moments compared to the second ones. The results n the mean equaton ndcate the exstence of mean spllovers between exchange rate changes and net bond flows n Japan, from bond flows to exchange rate changes n Canada and the UK, and from equty flows to exchange rate changes n the euro area. Wth regard to the mpact of exchange rate uncertanty on equty flows, the results suggest that exchange rate volatlty affects equty flows negatvely n the euro area, Sweden, and the UK, and postvely n Australa, and has no effect n Canada and Japan. Its mpact on bond flows, on the other hand, appears to be negatve n all countres except Canada for whch t s postve. The observed negatve mpact on equty as well as bond flows has mportant mplcatons. Frst, t ndcates that rsk averse market partcpants respond to exchange rate uncertanty by reducng ther fnancng actvtes, hence favourng domestc rather than foregn securtes n ther portfolos to reduce ther exposure to exchange rate volatlty. Second, n contrast to Hau and Rey (6) who assume that bonds are usually hedged nstruments not affected by exchange rate uncertanty, t appears that uncertanty n fact affects bond as well as equty flows, and the former more wdely, snce a negatve mpact s found n fve of the sx countres consdered. Ths s consstent wth the results of Fdora et al. 13

16 (7), who found n a wde set of ndustralsed and emergng economes that exchange rate volatlty s an mportant factor for blateral portfolo home bas, ths beng hgher for bonds than for equtes. Ther ratonalsaton of the hgher home bas for bonds compared to equtes s that t s consstent wth Markowtz-type nternatonal CAPM specfcatons n whch less volatle fnancal assets should show larger home bas. The estmates of the condtonal varance equatons ndcate that exchange rate changes (net equty/bond flows) exhbt condtonal heteroscedastcty: the dagonal elements of the ARCH matrces are sgnfcant at the 1% level n all cases except for equty flows n Australa and bond flows n Australa, Sweden, and the UK. Furthermore, the condtonal varances exhbt persstence n all cases except for equty flows n Canada. Whle the persstence of the condtonal varance of exchange rate changes ranges from.54 (Japan) to.98 (euro area), the persstence of the correspondng flows ranges from.38 (Sweden) to.91 (euro area) for net equty flows and from.43 (Japan) to.98 (Canada) for net bond flows. The ARCH, 11, and GARCH, 11, estmates for exchange rate changes n the bvarate GARCH BEKK models are rather smlar, regardless of whether the relatonshp wth bond or equty flows s consdered (see Panels A and B respectvely n all Tables). More specfcally, the change n 11 s less than 1% and ths also apples to 11, except for Japan where the change s around 6%. Furthermore, the off-dagonal elements of the ARCH and 14

17 GARCH matrces ndcate that shocks to exchange rate changes (net equty flows) affect the condtonal varance of net equty flows (exchange rate changes) at the 1% level n the euro area and Japan. The results also show that shocks to exchange rate changes (net bond flows) affect the condtonal varance of net bond flows (exchange rate changes) at the 1% level n all cases except Japan. More specfcally, the causalty-n-varance (.e., the nformaton flow) tests based on lkelhood rato test statstcs provde evdence of strong causalty-n-varance from equty flows to exchange rate changes n the case of the euro area and bdrectonal causalty-nvarance n the case of Japan. There s also causalty-n-varance from bond flows to exchange rate changes n Australa, the euro area, and Sweden, as well as bdrectonal causalty n Canada and the UK. A possble explanaton for the exstence of stronger dynamc lnkages between exchange rate changes and bond flows rather than equty flows s that foregn exchange dealers usually follow bond yelds n ther tradng behavour, wth such yelds, n turn, drvng cross-border bond acqustons, whch results n volatle exchange rates. Spllovers from the exchange rates may also be due to the fact that nvestors adjust ther portfolos on the bass of ther volatlty. Also, the lmted lnkage between exchange rate changes and bond flows n Japan can be explaned by the fact that a hgh percentage of Japanese debt s fnanced nternally, prmarly by Japanese penson funds, hence blateral 15

18 bond flows between the US and Japan have no mpact on exchange rate volatlty, and vceversa. Fnally, the Hoskng multvarate Q-statstcs of order (6) and (1) for the squared standardsed resduals suggest that the multvarate GARCH (1, 1) structure s suffcent to capture the volatlty n the seres. 5. Conclusons In ths paper, we have analysed the mpact of exchange rate uncertanty on bond and equty flows, as well as the dynamc lnkages between exchange rate volatlty and the varablty of these flows, usng data for the US vs-à-vs sx advanced economes, namely Australa, the UK, Canada, Japan, Sweden, and the euro area over the perod 1988:1-11:1. Estmatng bvarate GARCH BEKK n mean models, we fnd evdence that exchange rate volatlty mpacts on equty flows negatvely n the euro area, Sweden, and the UK and postvely n Australa. Furthermore, n contrast to Hau and Rey (6), t also affects bond flows negatvely n all countres except Canada where the effect s postve. The general concluson that can be drawn from these results s that exchange rate volatlty nduces rsk averse nvestors to reduce ther fnancng actvtes and to favour domestc to foregn assets n ther portfolos n order to mnmse ther exposure to volatlty. 16

19 The causalty-n-varance analyss suggests the exstence of strong spllovers from equty flows to exchange rate changes n the euro area and bdrectonal causalty-n-varance n Japan. As for the lnkages between exchange rate changes and bond flows, causalty-nvarance from bond flows to exchange rate changes s found for Australa, the euro area, and Sweden, and bdrectonal causalty for Canada and the UK. These fndngs have mportant polcy mplcatons, snce they suggest that polcy-makers and economc and fnancal regulators could use exchange rate or credt controls on equty as well as bond flows as nstruments to acheve economc and fnancal stablty. 17

20 References, R.F. and K.F. Kroner Bauwens, L., Laurent, S., Rombouts, J.V.K., 6. Multvarate GARCH models: a survey. Journal of Appled Econometrcs 1, Bollerslev, T., Generalzed autoregressve condtonal heteroskedastcty. Journal of Econometrcs 31, Bollerslev, T.P, Engle, R.F., Wooldrdge, J.M., A captal asset prcng model wth tme varyng covarances. Journal of Poltcal Economy 96, Bollerslev, T.P., Wooldrdge, J.M., 199. Quas-maxmum lkelhood estmaton and nference n dynamc models wth tme-varyng covarances. Econometrc Revew 11, Brennan, M.J., Cao, H.H., Internatonal portfolo nvestment flows. Journal of Fnance 5, Brooks, R., Edson, H., Kumar, M.S., Sløk, T., 4. Exchange rates and captal flows. European Fnancal Management 1, Caporn, M., McAleer, M., 1. Do we really need both BEKK and DCC? A tale of two multvarate GARCH models. Journal of Economc Surveys 6, Chaban, M., 9. Commodty currences and equty flows. Journal of Internatonal Money and Fnance 8,

21 Cheung,Y.W., Ng, L.K., A causalty n varance test and ts applcaton to fnancal market prces. Journal of Econometrcs 7, Chnn, M.D., Moore, M., 11. Order flow and the monetary model of exchange rates: evdence from a novel data set. Journal of Money, Credt and Bankng 43, Duffuor, K., Marsh, I.W., Phylakts, K., 1. Order flow and exchange rate dynamcs: an applcaton to emergng markets. Internatonal Journal of Fnance and Economcs 17, Engle, R.F., 198. Autoregressve condtonal heteroskedastcty wth estmates of the varance of U.K. nflaton. Econometrca 5, Engle, R.F.,. Dynamc Condtonal Correlaton- a smple class of multvarate GARCH models. Journal of Busness and Economc Statstcs, Engle, R.F., Kroner, K.F., Multvarate smultaneous generalzed ARCH. Econometrc Theory 11, Eun, C.S., Resnck, B.G., Exchange rate uncertanty, forward contracts, and nternatonal portfolo selecton. The Journal of Fnance 43, Eun, C.S., Resnck, B.G., Internatonal dversfcaton of portfolo nvestment: U.S. and Japanese perspectves. Management Scence 4, Evans, M., Lyons, R.K.,. Order flow and exchange rate dynamcs. Journal of Poltcal Economy 11,

22 Evans, M., Lyons, R.K., 5. Meese-Rogoff redux: mcro-based exchange rate forecastng. Amercan Economc Revew Papers and Proceedngs, Evans, M., Lyons, R.K., 8. How s macro news transmtted to exchange rates? Journal of Fnancal Economcs 88, 6-5. Fdora, M., Fratzscher, M., Thmann, C., 7. Home bas n global bond and equty markets: the role of real exchange rate volatlty. Journal of Internatonal Money and Fnance 6, Glen, J., Joron, P., Currency hedgng for nternatonal portfolos. The Journal of Fnance 48, Hafner, C.M., Herwartz, H., 8. Testng for causalty n varance usng multvarate GARCH models. Annals of Economcs and Statstcs 89, Hau, H., Rey, H., 6. Exchange rates, equty prces, and captal flows. The Revew of Fnancal Studes 19, Hoskng, J.R.M., Equvalent forms of the multvarate portmanteau statstc. Journal of the Royal Statstcal Socety 43, Kearney, C., Patton, A.J.,. Multvarate GARCH modelng of exchange rate volatlty transmsson n the European Monetary System. Fnancal Revew 35, 9-48.

23 Kodongo, O., Ojah, K., 1. The dynamc relaton between foregn exchange rates and nternatonal portfolo flows: evdence from Afrca s captal markets. Internatonal Revew of Economcs and Fnance 4, Levch, R.M., Hayt, G.S., Rpston, B.A., survey of dervatve and rsk management practces by U.S. nsttutonal nvestors, Survey conducted by the NYU Salomon Center, CIBC World Markets, and KPMG, avalable at McKenze, M.D., The mpact of exchange rate volatlty on nternatonal trade flows. Journal of Economc Surveys 13, Payne, R., 3. Informed trade n spot foregn exchange markets: an emprcal nvestgaton. Journal of Internatonal Economcs 61, Rme, D., Sarno, L., Sojl, E., 1. Exchange rate forecastng, order flow and macroeconomc nformaton. Journal of Internatonal Economcs 8, Ross, S.A., Informaton and volatlty: the no-arbtrage martngale approach to tmng and resoluton rrelevancy. Journal of Fnance 44, Sourouns, G., 4. Captal flows and exchange rates: an emprcal analyss. Avalable at: 1

24 E-Australa 1 E-Canada 1 E- Euro area BF-Australa 5 BF-Canada 5. BF-Euro area EF-Australa EF-Canada 5 EF-Euro area E-Japan 1 E-Sweden 1 E-UK BF-Japan BF-Sweden 5 BF-UK EF-Japan.5 EF-Sweden EF-UK Fg. 1. Tme seres of exchange rate changes (E), net bond flows (BF), and net equty flows (EF) of the sx advanced economes over the perod 1988:1 11:1.

25 Table 1 Summary of descrptve statstcs for the normalzed net portfolo flows and exchange rate changes. Statstcs Varable Australa Canada Euro area Japan Sweden UK Mean E t EF t BF t St. Dev E t EF t BF t Skewness E t EF t BF t Ex. kurtoss E t EF t BF t JB E t EF t BF t Notes: E t, EF t, and BF t ndcate exchange rate changes, net equty flows, and net bond flows, respectvely; JB s the Jarque-Bera test for normalty. ndcate sgnfcance at the 1 % level. 3

26 Table The estmated bvarate GARCH BEKK n mean model for Australa. Panel A: Exchange rates (E t ) and equty flows (EF t ) Panel B: Exchange rates (E t ) and bond flows (BF t ) E t (=1) EF t (=) E t (=1) BF t (=) Condtonal Mean Equaton (.159), t (.79) 1, t3, t5 (.1).18 (.157).158 (.19).157, t3.19 (.49).11, t.6 (.6).14 (.7) Condtonal Varance Equaton c 1 c 1 b 1 b.496 (.496) c (.545) c.4 (1.76).363 (.87).133 (.41).9 (.37).6 (.785).7 1 (.58).5 (.311).14 b 1 (.39).47 b (.56) (.1).13 (.81).753 (.19).54 (.46).38 (.76).949 (.1).33 (.71).8 (1.148).11 (.3).76 (.15).1 (.4).849 Loglk Loglk Q (6) 7.654[.74] Q (6) 9.83[.981] Q (6) 1.73 [.979] Q (6) 6.41 [.4] Q (1) 49.47[.414] Q (1) 3.46[.95] Q (1) [.966] Q (1) [.319] Tests of No Volatlty Transmsson: () Bdrectonal between E t and EF t Tests of No Volatlty Transmsson: () Bdrectonal between E t and BF t H : 1 1 b 1 b 1 LR=1.748[.781] H : 1 1 b 1 b 1 LR=11.66 [.] () From E t to EF t () From E t to BF t H : 1 b 1 LR=.15[.939] H : 1 b 1 LR=.135 [.934] () From EF t to E t () From BF t to E t H : 1 b 1 LR=1.639[.44] H : 1 b 1 LR=1.37 [.5] Note: E t, EF t, and BF t ndcate exchange rate changes, net equty flows, and net bond flows, respectvely; whle LR ndcates lkelhood rato test statstcs. Heteroscedastcty-consstent standard errors are n parentheses (.), whereas p- values are reported n [.]. Q (p) and Q (p) are multvarate Hoskng (1981) tests for p th order seral correlaton on the standardzed resduals z t and ther squares z, respectvely where = 1 (for exchange rate changes (E t t )), (for net equty flows (EF t ) and net bond flows (BF t )). The covarance statonarty condton s satsfed by all the estmated models, all the egenvalues of (A 11 A 11 + B 11 B 11 ) beng less than one n modulus. ndcates statstcal sgnfcance at the 1% level. ndcates statstcal sgnfcance at the 5% level. ndcates statstcal sgnfcance at the 1% level. (.65) 4

27 Table 3 The estmated bvarate GARCH BEKK n mean model for Canada. Panel A: Exchange rates (E t ) and equty flows (EF t ) Panel B: Exchange rates (E t ) and bond flows (BF t ) E t (=1) EF t (=) E t (=1) BF t (=) Condtonal Mean Equaton (.97 ) (.81).16 (.99 ), t1 (.61), t3 (.53) (.84).49 1, t.136 (.67).143, t3.11,t Condtonal Varance Equaton c 1.6 c 1 c 1 b 1 b (.164) 1.7 (.4).38 (.5).1 (.97 ).91 (.34).74 (.158).1 (3.6 ).17 (.61) c 1.6 (.13).3 (.18).5 (.63 ).314 (.47).6 (.131) (.38 ) b (.18).97 (.13).4 b (.63)..4 (.13 ) (.7).7 (.1 ).7 (.31).19 (.36).17 (.8).989 Loglk Loglk Q (6) 16.1 [.88] Q (6) [.897] Q (6) [.96] Q (6) [.99] Q (1) 9.31 [.984] Q (1) 37.1 [.788] Q (1) [.968] Q (1) 3.7 [3.7] Tests of No Volatlty Transmsson: () Bdrectonal between E t and EF t Tests of No Volatlty Transmsson: () Bdrectonal between E t and BF t H : 1 1 b 1 b 1 LR=.11[.733] H : 1 1 b 1 b 1 LR=8.697 [.69] () From E t to EF t () From E t to BF t H : 1 b 1 LR=1.38[.538] H : 1 b 1 LR=8.116 [.17] () From EF t to E t () From BF t to E t H : 1 b 1 LR=.798[.67] H : 1 b 1 LR=7.77 [.] Note: See notes to Table. (.6) 5

28 Table 4 The estmated bvarate GARCH BEKK n mean model for the euro area. Panel A: Exchange rates (E t ) and equty flows (EF t ) Panel B: Exchange rates (E t ) and bond flows (BF t ) E t (=1) EF t (=) E t (=1) BF t (=) Condtonal Mean Equaton , t1 (.178 ) (.916).9, t1 (.11).3 (.194) (.74).14, t.314, t.171 (.58) (.59), t3.19, t.49 (.57) (.7),t. (.15) Condtonal Varance Equaton c 1 c 1 b 1 b.48 c 1 (.113) c.819 (.69).115 (.3).1 (.74 ).98 (.7).38 (.7 ).1 (.181).1 (.7 ) 1.38 (.73).3 (.7 ) b 1.91 b (.3).94 (.5 ).4 (.96).174 (.66).313 (.1).968 (.).134 (.49 ) (.58).5 (.56 ).1 (.7 ).159 (.67).18 (.8).936 Loglk Loglk Q (6).615 [.661] Q (6) [.64] Q (6) 18.9 [.788] Q (6) [.95] Q (1) [.645] Q (1) [.656] Q (1) 4.47 [.771] Q (1) [.66] Tests of No Volatlty Transmsson: () Bdrectonal between E t and EF t Tests of No Volatlty Transmsson: () Bdrectonal between E t and BF t H : 1 1 b 1 b 1 LR=9.35[.5] H : 1 1 b 1 b 1 LR=1.87 [.11] () From E t to EF t () From E t to BF t H : 1 b 1 LR=1.83[.41] H : 1 b 1 LR=3.86 [.13] () From EF t to E t () From BF t to E t H : 1 b 1 LR=7.86[.19] H : 1 b 1 LR=1.88 [.1] Note: See notes to Table. (.1) 6

29 Table 5 The estmated bvarate GARCH BEKK n mean model for Japan. Panel A: Exchange rates (E t ) and equty flows (EF t ) Panel B: Exchange rates (E t ) and bond flows (BF t ) E t (=1) EF t (=) E t (=1) BF t (=) Condtonal Mean Equaton , t1 (.19 ) (.8).1 1, t1.6), t1 (.46) Condtonal Varance Equaton c 1 c 1 b 1 b.53, 1 t, t 1, t3, t3 1, t4 1, t5,t.19 c 1 (.3).1. c (.66 ) (.156 ).356 (.98).357 (.315 ).54 (.13).64 (.349).31 (.3 ) 1.37 (.133).31 (.31) b (.81) b.86 (.11 ).77 (.6).65 (.).14 (.48).37 (.1) (.46).39 (.19).16 (.73).84 (.46).98 (.59).15 (.55).91 (.5) 1.6 (.14).43 (.196 ).65 (.73).59 (.343).799 (.3).41 (.65 ).37 (.1).11 (.49).743 (.89).47 (.3 ).58 (.9).14 (.38).439 Loglk Loglk Q (6) [.136] Q (6) [.776] Q (6) 3.66 [.484] Q (6) 1.51 [.94] Q (1) [.57] Q (1) [.97] Q (1) [.161] Q (1) [.971] Tests of No Volatlty Transmsson: () Bdrectonal between E t and EF t Tests of No Volatlty Transmsson: () Bdrectonal between E t and BF t H : 1 1 b 1 b 1 LR=16.95 [.1] H : 1 1 b 1 b 1 LR=5.8 [.1] () From E t to EF t () From E t to BF t H : 1 b 1 LR=1.55 [.5] H : 1 b 1 LR=1.457 [.48] () From EF t to E t () From BF t to E t H : 1 b 1 LR=9.661 [.7] H : 1 b 1 LR=4.14 [.16] Note: See notes to Table. (.84) 7

30 Table 6 The estmated bvarate GARCH BEKK n mean model for Sweden. Panel A: Exchange rates (E t ) and equty flows (EF t ) Panel B: Exchange rates (E t ) and bond flows (BF t ) E t (=1) EF t (=) E t (=1) BF t (=) Condtonal Mean Equaton , t1 (.179).45 (.196 ), t.137 (.69), t5.13 (.8) Condtonal Varance Equaton c 1 c 1 b 1 b.75, t (.59) 1.18 c (.81 ) (.757 ).5 (.94).47 (.55).74 (.79).68 (.38) c (.41).3 1 (.47).56 (.51).13 b 1 (.3).38 b (.185).66 (.165).8 (.4) (.38).881 (.17).4 (.93).433 (.97).79 (.83).445 (.13) (.13).1 (.38).17 (.41 ).116 (.16 ). (.3).88 Loglk Loglk Q (6) [.84] Q (6) 1.66 [.968] Q (6) 4.57 [.43] Q (6) [.76] Q (1) [.9] Q (1) 3.93 [.945] Q (1) [.797] Q (1) [.764] Tests of No Volatlty Transmsson: () Bdrectonal between E t and EF t Tests of No Volatlty Transmsson: () Bdrectonal between E t and BF t H : 1 1 b 1 b 1 LR=5.611 [.3] H : 1 1 b 1 b 1 LR= [.9] () From E t to EF t () From E t to BF t H : 1 b 1 LR=.6 [.73] H : 1 b 1 LR=.369 [.831] () From EF t to E t () From BF t to E t H : 1 b 1 LR=4.9 [.1] H : 1 b 1 LR=1.913 [.1] Note: See notes to Table. (.61) 8

31 Table 7 The estmated bvarate GARCH BEKK n mean model for the UK. Panel A: Exchange rates (E t ) and equty flows (EF t ) Panel B: Exchange rates (E t ) and bond flows (BF t ) E t (=1) EF t (=) E t (=1) BF t (=) Condtonal Mean Equaton (. ), t1 (.139), t (.51), t3.156 (.48), t3.8 (.17) Condtonal Varance Equaton c 1 c (.5) 1 (.7) b 1 b.186 1, t1 (.54).96,.659 c 1 (.146). c (.7 ) (.154).899 (.7).5 (.56).3 1 (.4) (.194) t.6 (.97) b 1.3 (.4).468 b (.3) (.185).34.5 (.5).9 (.16).173 (.41).65 (.139).39 (.67 ).968 (.38).34 (.88) (.19). (.63).7 (.4).1 (.36).66 (.9).9 Loglk Loglk Q (6) [.85] Q (6) [.989] Q (6) 1. [.637] Q (6) 7.45 [.157] Q (1) [.776] Q (1) 4.9 [.993] Q (1) [.837] Q (1) [.698] Tests of No Volatlty Transmsson: () Bdrectonal between E t and EF t Tests of No Volatlty Transmsson: () Bdrectonal between E t and BF t H : 1 1 b 1 b 1 LR=4.181 [.381] H : 1 1 b 1 b 1 LR=.154 [.] () From E t to EF t () From E t to BF t H : 1 b 1 LR=1.161 [.559] H : 1 b 1 LR= [.] () From EF t to E t () From BF t to E t H : 1 b 1 LR=.866 [.38] H : 1 b 1 LR=6.743 [.34] Note: See notes to Table. (.) 9