Return and Volatility Spillovers from Developed to Emerging Capital Markets: The Case of South Asia

Transcription

1 Reurn and Volailiy Spillovers from Developed o Emerging Capial Markes: The Case of Souh Asia Yun Wang #, Abeyrana Gunasekarage ## and David M. Power ### # Performance Group, Commerce Commission, Level 6, Terrace, PO Box 2351, Wellingon, New Zealand. Phone , Fax , E- mail:yunyun99nz@homail.com. ## Deparmen of Finance, Waikao Managemen School, Universiy of Waikao, Privae Bag 3105, Hamilon, New Zealand. Phone: ; Fax: ; agubo@waikao.ac.nz. ### Deparmen of Accounancy and Business Finance, Universiy of Dundee, Dundee DD1 4HN, Scoland, UK. Phone: ; Fax: ; E- mail: D.M.Power@dundee.ac.uk. Absrac This sudy examines reurn and volailiy spillovers from he US and Japanese sock markes o hree Souh Asian capial markes (i) he Bombay Sock Exchange, (ii) he Karachi Sock Exchange, and (iii) he Colombo Sock Exchange. We consruc a univariae EGARCH spillover model which allows he unexpeced reurn of any paricular Souh Asian marke o be driven by a local shock, a regional shock from Japan, and a global shock from he USA. The sudy discovers reurn spillovers in all hree markes, and volailiy spillovers from he US o he Indian and Sri Lankan markes, and from he Japanese o he Pakisani marke. Regional facors seem o exer an influence on hese hree markes before he Asian financial crisis bu he global facor becomes more imporan in he pos-crisis period. JEL Classificaion: G14; G15 Key Words: Inegraion; Spillover; EGARCH Model; Asymmeric Effec.

2 I. Inroducion The heme of inegraion among inernaional capial markes and he mechanism whereby informaion is ransmied among differen sock exchanges has been exensively researched in he modern finance lieraure. This opic has araced he aenion of financial economiss as he urmoils which occur in some capial markes have far reaching consequences on securiy prices on heir counerpars in oher counries. For example, he Ocober 1987 crash no only eliminaed more han 20 per cen of he marke value of US equiies bu also sen shock waves around he world. The Asian financial crisis had a similar impac on many oher emerging markes in Lain America as well as in Easern Europe. The liberalizaion of capial movemens ogeher wih advances in compuer echnology and he improved worldwide processing of news have improved he possibiliies for naional financial markes o reac rapidly o new informaion from inernaional sock exchanges. Early invesigaions in his area analysed he inerrelaedness among developed capial markes using correlaions of sock reurns. For example, Hilliard (1979) examined indices for en inernaional equiy markes (Amserdam, Frankfur, London, Milan, New York, Paris, Sydney, Tokyo, Torono, and Zurich) during he world-wide financial crisis creaed by he OPEC embargo in he period of and found ha he mos inra-coninenal prices move simulaneously. More recenly, Eun and Shim (1989) applied vecor auoregression (VAR) mehodology o sudy daily index daa for nine of he larges sock exchanges in he world (Ausralia, Canada, France, Germany, Hong Kong, Japan, Swizerland, he UK, and he US) and discovered a subsanial amoun of muli-laeral ineracion among hese markes; he US sock marke was he mos influenial and none of he oher markes explained any movemens in US reurns. Joen and von Fursenberg (1990) arrived a a similar view; 2

3 hey highlighed evidence of growing inernaional inegraion among he four major equiy markes of Germany, Japan, he UK and he US in he 1980s. Becker e al. (1990) concluded ha he informaion from he US marke could be used o rade profiably in he Japanese marke as here was a high correlaion beween he open-oclose reurns of US shares in he previous rading day and he reurns of Japanese equiies in he curren period. Koch and Koch (1991), who used a dynamic simulaneous equaions model o invesigae he conemporaneous and lead-lag relaionships among eigh naional sock exchanges (Ausralia, Germany, Hong Kong, Japan, Singapore, Swizerland, he UK, and he US), discovered a growing level of marke inerdependence wihin he same geographical regions over ime; an increasing influence of he Japanese marke a he expense of he US marke was also deeced. Anoher branch of research concenraes on he ransmission of inernaional equiy movemens by sudying he spillover of reurn and volailiy across markes. For example, Hamao e al. (1990), who sudied hree major sock markes (London, New York and Tokyo,) using univariae GARCH-in-mean models, found volailiy spillovers (i) from New York o Tokyo and London and (ii) from London o Tokyo. Theodossiou and Lee (1993) used mulivariae GARCH-in-mean models o analyse he markes in Canada, Germany, Japan, he UK and he US; hey discovered ha he US marke was he major exporer of volailiy. More recenly, Scheicher (2001) analysed hree Easern European markes (Czech Republic, Hungary and Poland) and repored ha alhough he equiy reurns were affeced by boh regional and global facors, he volailiies were impaced by only regional influences. Frazscher (2002) and Baele (2002) arrived a a similar conclusion; hey documened evidence ha 3

4 shock ransmissions from he aggregae European Union marke o domesic European equiies had become more pronounced in recen years 1. A number of researchers have addressed he quesion of wheher he quaniy of news (i.e. he size of an innovaion) and he qualiy of he informaion (i.e. he sign of an innovaion) are imporan deerminans of he degree of volailiy spillover across markes. This quesion has been moivaed by findings of an asymmeric or leverage effec associaed wih equiy reurns; bad news has a differen degree of predicabiliy on fuure volailiy compared o is good news counerpar 2. This asymmeric effec has been examined in sudies of volailiy spillovers across markes. For example, Bae and Karolyi (1994), who examined he join dynamics of overnigh and dayime reurn volailiy for he New York and Tokyo sock markes over he period , noed ha he magniude and persisence of shocks originaing in New York or Tokyo ha ransmied o he oher marke were significanly undersaed if his asymmeric effec was ignored; bad news from domesic and foreign markes appeard o have a much larger impac on subsequen reurn volailiy han good news. Koumos and Booh (1995) invesigaed he asymmeric impac of marke advances and marke declines (i.e. good and bad news respecively) on volailiy ransmission across he New York, Tokyo and London sock markes. Using daily open-o-close reurns, hey found unidirecional price spillovers (i) from New York o Tokyo, (ii) from New York o London, and (iii) from Tokyo o London. They also uncovered bidirecional volailiy spillovers among he hree markes. In all 1 Some researchers have exended his invesigaion o foreign exchange and spo and fuure markes and uncovered evidence for he exisence of spillovers among major currency markes (Baillie and Bollerslev, 1990; Engle e al., 1991; Chin e al., 1991; Cheung and Fung, 1997). 2 This phenomenon was originally moivaed by he work of Black (1976), Chrisie (1982), French e al. (1987) and Nelson (1991) and is significance was evaluaed by Pagan and Schwer (1990), Braun e al. (1992), Glosen e al. (1993) and Engle and Ng (1993) by employing differen variaions of volailiy models. Nelson (1991), Cheung and Ng (1992), Koumos (1992) and Poon and Taylor (1992), among ohers, provide empirical evidence for he exisence of a leverage effec. 4

5 insances, he volailiy ransmission mechanism was asymmeric - i.e. negaive innovaions in one marke increased volailiy in he oher marke considerably more han heir posiive counerpars. Booh e al. (1997) looked a he four Scandinavian markes and found significan and asymmeric volailiy spillovers among Swedish, Danish, Norwegian, and Finnish securiies. Similar evidence has been repored for oher European markes - London, Paris, and Frankfur by Kanas (1998). Only a minoriy of sudies have focussed on he reurn and volailiy spillovers from developed o emerging capial markes 3. In paricular, he evidence on marke ineracions and informaion ransmissions in Souh Asian capial markes is hard o find. The Capial markes in Souh Asia have generaed a considerable ineres among local and foreign invesors as a resul of he increased economic aciviy in hese counries arising from economic reforms and he liberalisaion of capial markes. In his research exercise, we invesigae how informaion is ransmied from developed capial markes o hree recenly libaralised Souh Asian capial markes he Bombay Sock Exchange (BSE) of India, he Karachi Sock Exchange (KSE) of Pakisan and he Colombo Sock Exchange (CSE) of Sri Lanka; reurn and volailiy spillover models are esed on marke index daa. Our sudy differs from he previous research on his opic in hree respecs. Firs, unlike many exising sudies which focus on how a single inernaional marke (ofen he US or a world marke) influences oher sock markes 4, we consider he innovaions from boh he US and Japanese markes in an aemp o analyse he impacs of boh regional and world shocks on Souh Asian equiies. Second, we recognise ha volailiy ransmission may be asymmeric in 3 See, Ng (2000), Chan-Lau and Ivaschenko (2002) and Worhingon and Higgs (2004) for some evidence on his opic. 4 Many early sudies failed o disinguish beween world and regional facors as hey were predominanly occupied wih esing he influence of he world marke (ofen US) on oher markes. For example, see Hamao e al. (1990), Campbell and Hamao (1992), Bekaer and Hodrick (1992), Bekaer and Harvey (1995), Harvey (1995), Karolyi (1995) and Karolyi and Sulz (1996). 5

6 characer i.e. he negaive innovaions in one marke may produce higher volailiy spillovers in anoher marke, han he posiive innovaions of equal magniude. Finally, we address he possible effec of he Asian financial crisis 5 on he ransmission mechanism by disaggregaing he daa ino hree sample periods: (i) precrisis, (ii) in-crisis and (iii) pos-crisis. The reminder of his paper is organized as follows. Secion II provides a brief overview abou he Souh Asian sock markes. Secion III describes he spilover models used o analyse he daa in his sudy. Secion IV oulines he daa employed in he sudy and presens he empirical resuls. The final secion offers some conclusions. II. An Overview of Souh Asian Capial Markes The Souh Asian region is noable for is large populaion (more han one-fifh of he world oal s inhibians) which coninues o grow rapidly. India is by far he larges Souh Asian counry, in erms of populaion, GDP, and land area. Sri Lanka has he mos open economy. Indian, Pakisan, and Sri Lankan sock exchanges are also he hree bigges markes in his region in erms of marke capializaion. Souh Asia has experienced fas economic growh in recen years because of he economic reforms implemened by hese counries governmens; i was he fases growing region of he world in The emerging capial markes in his region have generaed considerable ineres among regional as well as global invesors because of he rapid growh of hese counries economies and he concessions provided o foreign invesors hrough radical liberalizaion processes. The Bombay Sock Exchange is he oldes sock marke in Asia - even older han he Tokyo Sock Exchange - and was esablished in 1875 as a volunary non- 5 Bollerslev e al. (1992) sugges ha he asymmeric response of volailiy o innovaions may be he 6

7 profi making associaion. I is one of 25 sock markes hroughou India. Wih over 20 million shareholders, India has he hird larges invesor base in he world afer he US and Japan. India's marke capializaion was he 6 h highes among he emerging markes. Share rading on Colombo Sock Exchange daes back o he 19 h cenury; in 1896 Colombo Brokers Associaion commenced he rading of shares in limied liabiliy companies. By conras, he sock marke in Pakisan is relaively new. The Karachi Sock Exchange only came ino exisence in These capial markes exhibi a number of common feaures; hey did no play a prominen role in he economic developmen of heir counries unil he respecive governmens sared a programme of deregulaion and economic liberalizaion. For example, India iniiaed financial reforms in conjuncion wih economic deregulaion and permied foreign companies o own a majoriy sake in quoed Indian firms from many differen indusries. The liberalizaion policies of he Pakisan governmen have led o rapid deregulaion of he economy and he removal of impedimens o privae invesmen. The secondary sock marke in Pakisan is now open o foreign invesors; nonnaionals are reaed equally wih local paricipans when rading shares. The Sri Lankan governmen ook a number of seps including he opening of he banking secor o foreign owners, repealing he business acquisiion ac and privaizing governmen-owned business underakings in an aemp o creae a well-funcioning capial marke in he counry. TABLE 1 ABOUT HERE resul of a few exreme observaions such as hose associaed wih he Ocober 1987 crash. 7

8 As a resul of hese changes, share markes in hese hree Souh Asian counries recorded a remarkable rae of growh in heir rading aciviies. Table 1 repors some marke saisics for he sock exchanges in hese counries. According o his able, during he 10-year period ending 2003, hese markes have repored a phenomenal growh in marke capializaion: BSE per cen, KSE 1, per cen and CSE per cen. Similar growh paerns can be observed wih respec o marke capializaion as a percenage of GDP, annual urnover, he number of lised companies, and he marke price index. The number of companies lised on he BSE a he end of December 1998 was 5,860. This was more han he aggregae oal of companies lised in 9 emerging markes (Malaysia, Souh Africa, Mexico, Taiwan, Korea, Philippines, Thailand, Brazil and Chile) a he same dae. The number of lised companies was also larger han ha in several developed markes: Japan, UK, Germany, France, Ausralia, Swizerland, Canada and Hong Kong. The KSE has also grown quickly, especially in recen years. I was declared he bes performing sock marke of he World for he year 2002 by Business Week. The findings of his sudy will be ineresing as lile evidence appears in he finance lieraure on Souh Asian capial markes. III. Mehodology EGARCH Model Esimaion GARCH (generalized auoregressive condiional heeroskedasiciy) models are generally used o explore he sochasic behavior of several financial ime series and, in paricular, o explain he behavior of volailiy over ime. However, such models do no work wih negaive daa. The exponenial GARCH (EGARCH) model 8

9 developed by Nelson (1991) 6 overcomes his limiaion and allows researchers o capure he leverage effec or asymmeric impac of shocks on volailiies. I herefore, avoids he imposiion of non-negaiviy resricions on he values of he GARCH parameers o be esimaed. Specifically, ime series of share reurns are modeled in EGARCH (p, q) as follows: r 0 + α i i + ε i= 1 R = α R [1] 2 ε Ω 1 ~ N(0, σ ) [2] q p 2 ( ) = a0 + ai f ( z i ) + bi log( log σ σ ) [3] i= 1 i= 1 f ( ) = θ z z Ε( z ) z i i [ i i 2 i + ] [4] ( ) Ε = z i 2 π 0.5 [5] where R is he reurn series in ime (i.e. coninuously compounded reurns generaed aking he naural logarihm of he raio of curren price o he lagged price), ε is he 2 sochasic error, Ω is he informaion se a ime -1, is he condiional (ime 1 σ varying) variance, and z is he sandardized residual which is derived from ε / σ condiional on Ω 1. The ermε is assumed o be normally disribued wih a zero mean and variance ( σ 2 ). The erm 2 π 0.5 is a consan employed o make sure ha he inegral under he curve of he normal disribuion of he residual from negaive o posiive infiniy is equal o one. 6 A compeing model which also capures he asymmeric leverage effec is he Quadraic GARCH model proposed by Engle (1990). However, Engle and Ng (1993) find ha he EGARCH performs beer. Moreover, a significan body of previous evidence, summarized by Hamilon (1994, P. 672), suppors he use of he EGARCH model. On he basis of his evidence, he EGARCH model is employed in his sudy. 9

10 Equaion [1] (he condiional mean equaion) is modeled as an auoregressive process of order r [AR (r)], following Theodossiou and Lee (1993), and Karolyi (1995). To specify he lag lengh r for each reurn series, he auocorrelaion funcion (ACF) and parial auocorrelaion funcion (PACF) of each series are considered, and residuals from he mean equaions are hen esed for whieness using he Ljung-Box saisic. For he enire period (01/01/ /12/2003), we use 1 lag for he US, Japanese and Indian reurn series and 2 lags for Pakisan and Sri Lankan series o yield uncorrelaed residuals. For he sub-periods, whieness in he residuals for each series is achieved using 1 lag excep for he Pakisan series in he pre-crisis period and Indian series in he pos-crisis period; in each of hese wo excepions, 2 lags are needed. Equaion [3] (he condiional variance equaion) reflecs he EGARCH (p,q) represenaion of he variance of ε. According o his EGARCH represenaion, he variance is condiional on is own pas values as well as on pas values of a funcion of z, or he sandardized residuals ( ε / σ ). The persisence of volailiy implied by p b i i= 1 p b i i= 1 Equaion [3] is measured by. The uncondiional variance is finie if <1 in absolue erms (see Nelson, 1991). If p b i i= 1 =1, hen he uncondiional variance does no exis and he condiional variance follows an inegraed process of order one. As noed by Hsieh (1989), he exponenial specificaion is less likely o produce p b i i= 1 inegraed variances. The smaller he, he less persisen he volailiy is afer a shock. 10

11 In equaion [4], asymmery is presen if θ is negaive and saisically significan. Asymmery in volailiy ransmission can be convenienly examined using is parial derivaives: The erm z Ε( ) f ( z ) 1 + θ, for z = z 1+ θ, for z z > 0 [6] < 0 [ measures he size effec of an innovaion whereas θ z measures ] he corresponding sign effec. If θ is negaive, a negaive z ends o reinforce he size effec, whereas a posiive z ends o parially offse i. If θ = 0, a posiive shock has he same effec as a negaive shock of he same magniude. If -1<θ <0, a negaive shock increases volailiy more han a posiive shock and, hus, θ measures he asymmeric effec of shocks on volailiy. If θ < -1, a negaive (posiive) shock increases (reduces) volailiy. The relaive imporance of he asymmery or he leverage effec can be measured by he raio 1+θ. Lag runcaion lenghs, p and q, 1+ θ are deermined using likelihood raio (LR) ess of alernaive specificaions 7. Based on hese ess, EGARCH (1, 1) models were deermined o be opimal. The Univariae EGARCH Models of Price and Volailiy Spillovers Esimaion In his sudy, he univariae EGARCH model is used o es for reurn and volailiy spillovers from he wo developed sock markes of he US and Japan o a hird small sock marke (India, Pakisan, and Sri Lanka respecively). We assume unidirecional reurn and volailiy spillovers o be relevan because hese small sock 7 Likelyhood raios are calculaed as follows: LR = 2* [ ln( log likelihood of unresriced model) ln( log likelihood of resriced model) ] The unresriced model refers o eiher he EGARCH(1,2) or EGARCH(2,1) model and he resriced model refers o he EGARCH(1,1) model. Since we have a very small LR saisic for all markes, he 11

12 markes are no hough o have a subsanial impac on he wo developed markes considered. To es for spillovers from a foreign marke o he domesic marke, he approach adoped by Hamao e al. (1990) and Theodossiou and Lee (1993) is followed. According o his approach, he mos recen squared residuals from he condiional mean condiional variance formulaion of he foreign marke are inroduced as an exogenous variable in he condiional variance equaion of he domesic marke. The univariae EGARCH (1, 1) models of reurn and volailiy spillovers for marke j are specified as follows: The condiional mean equaion for India, Pakisan and Sri Lanka becomes; R IND, = α IND, 0 + α + β IND,1 IND,2 R R IND, 1 JAP, 1 + β + ε IND,1 IND, R US, 1 [7a] R PAK, = α PAK, 0 + α + β PAK,1 PAK,2 R R PAK, 1 JAP, 1 + α + ε PAK,2 PAK, R PAK, 2 + β PAK,1 R US, 1 [7b] R SRI, = α SRI, 0 + α + β SRI,1 R SRI,2 SRI, 1 R JAP, 1 + α + ε SRI,2 R SRI, SRI, 2 + β SRI,1 R US, 1 [7c] The condiional variance equaion for hese hree markes is: where, log log log 2 ( σ IND, ) 2 ( σ PAK, ) 2 ( σ SRI, ) 2 = a + a f ( z ) + b log( σ IND, 1) IND,0 + c IND,1 PAk,1 IND,1 log( U IND, 1 US, US, ) + c IND,2 PAK,2 IND,1 log( U JAP, JAP, ) [8a] 2 = apak + a,1 (, 1),1 log( PAK, 1),0 PAK f zpak + bpak σ [8b] + c log( U ) + c log( U ) 2 = a + a f ( z ) + b log( σ SRI, ) SRI,0 + c SRI,1 SRI,1 log( U SRI, 1 US, ) + c SRI,1 SRI,2 log( U JAP, ) 1 [8c] lag runcaion lenghs p = 2 or q = 2 are no saisically significan. Based on hese ess, we fi EGARCH(1,1) models for all markes. 12

13 f f f, IND [ z IND, Ε( z IND, )] ( z IND 1 ) = θ IND ( z PAK 1 ) = θ PAK z [9a] +, 1 1 1, PAK [ z PAK, Ε( z PAK, )] ( z SRI 1 ) = θ SRI z [9b] +, 1 1 1, SRI [ z SRI, Ε( zsri, )] z [9c] +, U, and are he conemporaneous squared residuals (from he AR(1) US U JAP, EGARCH(1,1) models) for he US and Japanese reurns respecively, and lagged sandardized residuals. z 1 is he Reurn spillovers occur when pas informaion abou he US and Japanese markes have persisen effecs on small marke reurns, and volailiy spillovers are relaed o he presen informaion flows from he foreign markes. The univariae EGARCH model permis us o differeniae beween he relaive influence of he US and Japan on he hree small markes. Exisence of reurn spillovers is indicaed by he saisical significance of β 1 (reurn spillovers from he US) and β 2 (reurn spillovers from Japan). Exisence of volailiy spillovers is indicaed by he saisical significance of c (volailiy spillovers from he US) and 1 2 c (volailiy spillovers from Japan). Saisical inference regarding and c is based on robus sandard errors c1 2 derived by Bollerslev and Wooldridge (1992) 8. A significan c (or c ) coupled wih a 1 2 significanly negaive θ implies ha negaive innovaions in he US marke (or Japanese marke) have a higher impac on he volailiy of marke j han posiive innovaions, i.e. he volailiy spillover mechanism is asymmeric. Given a sample of T observaions and condiional normaliy for he sock reurns in each marke, he log likelihood funcion for he univariae EGARCH can be wrien as: 8 Convenional sandard errors end o underesimae he rue sandard errors, especially for he parameers in he condiional variance equaion (Susmel and Engle, 1994). 13

14 T L( Θ) = ( T / 2) log(2π ) 0.5 log( σ 2 ) [10] where Θ is he parameer vecor ( α 0 α 1 α 2 a0 a1 b1 c1 c2 = 1 θ ) o be esimaed. IV. Empirical Findings Daa and Preliminary Saisics The daa used in he sudy consis of daily sock indices for five counries - he USA, Japan, India, Pakisan, and Sri Lanka for he period 1 January 1993 o 31 December 2003; a oal of 2869 observaions are employed for each marke. The sample period is divided ino hree sub-periods pre-crisis (01/01/ /06/1997), in-crisis (01/07/ /12/1999), and pos-crisis (01/01/ /12/2003). The index daa are obained from Daasream. The sock marke indices used in his sudy are he S&P 500 (he US), he Nikkei 500 (Japan), he BSE Naional Price Index (India), he Karachi 100 Price Index (Pakisan), and he Colombo All Share Price Index (Sri Lanka). In each marke, we choose he mos comprehensive and diversified sock index. The S&P 500 index consiss of he 500 larges, publicly-held companies represening approximaely 76 percen of oal marke capializaion. The Nikkei 500 index incorporaes 500 Japanese companies lised in he Firs Secion of he Tokyo Sock Exchange. The BSE Naional Index comprises 100 socks lised a five major Indian sock exchanges (Mumbai, Calcua, Delhi, Ahmedabad and Madras). The Karachi 100 includes he larges 100 companies in he exchange (27 companies represening 27 secors and 73 companies represening he enire marke) covering abou 83 percen of marke capializaion of he exchange. The Colombo All Share 14

15 Price Index consiss of all he shares raded on he sock exchange 9.Wih he excepion of he Nikkei 500, all indices are calculaed on a value-weighed basis. The Japanese index is a share price-weighed index which does no ake dividend reinvesmen ino accoun. However, cash dividends paid on mos Japanese socks are relaively small, so his dividend omission is of lile consequence. 10 The variable analysed in he sudy is he daily reurn which is calculaed by aking he naural logarihm of he raio of curren price o he lagged price 11. TABLE 2 ABOUT HERE Table 2 repors summary saisics for he daily reurns of he five naional sock markes. The mean reurns are posiive for four markes wih he excepion of Japan. The Pakisan marke earned he highes mean reurn bu wih he larges risk as measured by he sandard deviaion. However, he sample means for all five markes are no saisically differen from zero. The measures for skewness show ha wih he excepion of he disribuion of reurns for he Sri Lankan marke, he reurn series are negaively skewed. The excess Kurosis measures indicae ha he disribuions of all he reurn series are highly lepokuric. Likewise, he Jarque-Bera saisics rejec normaliy for each of he reurn series a he 1 percen level of significance. The Ljung-Box q-saisics - LB(k) and LB 2 (k) - for lag lenghs of 6 and 12 days are used o es for serial correlaion in he reurn and squared reurn series. The 9 Even hough he CSE had a blue chips index represening he op companies in he marke, is composiion changed in Therefore, i was decided o use he All Share Price Index. 10 See Campbell and Hamao (1989) for evidence on he dividend-price raio for he Tokyo marke. 11 Since Easern rading ime leads Wesern rading ime by one day, we consider US reurns wih a one day lag in order o overcome problems associaed wih non-synchroneous rading across five markes analysed. All hree Souh Asian markes have overlapping rading hours wih he Japanese marke bu no wih he US marke. Recen spillover invesigaions deal wih his problem using open-o-close reurns (Hamao e al. 1990; Bae and karolyi, 1994; and Koumos and Booh, 1995). However, his 15

16 null hypohesis of uncorrelaed reurns is rejeced a he 1 percen level of significance for he markes of Japan, India, Pakisan, and Sri Lanka a boh lag lenghs used. The null hypohesis of homoskedasic reurns (uncorrelaed squared reurns) is also rejeced a he 1 percen level for all markes a boh lag levels. Linear dependencies may be due eiher o non-synchronous rading of he socks ha make up each index 12 or o some form of marke inefficiency. Non-linear dependencies may be due o auoregressive condiional heeroskedasiciy, as documened by several recen sudies for boh he US and foreign sock markes 13. The ARCH Lagrange Muliplier (LM) ess (Engle, 1982) indicae ha each marke s reurns srongly depend on heir pas values and exhibi srong ARCH effecs, implying ha he ARCH model is appropriae for daa analysis in his sudy 14. The ARCH effecs may explain (a leas parially) he observed hicker han normal disribuional ails. Since he Jarque-Bera normaliy ess show ha all he reurn series are no normally disribued, we examine he relaionship among reurns using nonparameric correlaions. All reurn series are posiively correlaed, bu he cross-correlaions among reurns are relaively low. Univariae EGARCH Model Esimaion We firs esimae a univariae EGARCH (1, 1) model for each of he five indices by resricing all cross-marke coefficiens measuring reurn and volailiy spillovers o be zero. An EGARCH (1, 1) model was deermined o offer he bes fi for he daa series. The resuling coefficiens from hese models are presened in Table opion was no available o us due o he difficuly of obaining opening and closing prices for he Souh-Asian capial markes. 12 See Scholes and Williams (1977) and Lo and MacKinley (1988). 13 See, for example, Nelson (1991), Akgiray (1989) and Booh e al. (1992). 14 The LM es approach requires he esimaion of he auxiliary regression model of e 2 p = cons an + δ e i= i i error, where e s are he OLS residuals, i =1,2,..p;and = p +1, p +2,, 16

17 3. Panel A provides esimaes for equaion [1]. The firs order auoregressive coefficien, α 1, is saisically significan for he Japanese, Indian, Pakisan, and Sri Lankan markes, indicaing ha eiher non-synchronous rading or marke inefficiency induces auocorrelaion in he reurn series. The second order auoregressive coefficien, α 2, is also saisically significan for he Pakisan and Sri Lankan markes. However, for he US marke, α 1, is insignifican; his finding is consisen wih previous sudies such as Theodossiou and Lee (1993) and Koumos and Booh (1995), indicaing ha he US marke is more efficien han oher markes. Condiional hereoskedasiciy is perhaps he single mos imporan propery describing he shorerm dynamics of all markes. TABLE 3 ABOUT HERE The condiional variance is a funcion of pas innovaions and pas condiional variances. Panel B provides esimaes for equaion [3]. The relevan coefficiens, a 1 (measuring he ARCH effec) and b 1 (measuring he degree of volailiy persisence) are all saisically significan for all he markes. Furhermore, he values of b 1 coefficiens are all close o one indicaing a high degree of persisence in volailiy. This volailiy persisence is highes for he US, followed by he Japanese, Indian, Pakisan, and Sri Lankan markes. The leverage effec, as measured by θ, or he asymmeric impac of pas innovaions on curren volailiy, is negaive and saisically significan for he US, Japanese, and Indian markes indicaing ha he volailiy spillovers may also be asymmeric. The relaive imporance of he 2 m. From he resuls of his auxiliary regression, a es saisic is calculaed as ( N p) * R which is expeced o be disribued as Chi-squared (p) under he null hypohesis of no ARCH effecs. 17

18 1+ θ asymmery, or leverage effec, can be measured by he raio. Thus, he degree 1+ θ of asymmery, on he basis of he esimaed θ coefficiens, equals for he US marke, for Japanese marke, and for Indian marke. These raios indicae ha he degree of asymmery is highes for he US marke (negaive innovaions increase volailiy 1.23 imes more han posiive innovaions), followed by he Japanese marke (1.13 imes) and he Indian marke (1.06 imes). The hypohesis ha he reurn series are homoskedasic (i.e. a 1 = b 1 = θ) is rejeced a any significance level on he basis of he Wald es 15. Panel C repors he diagnosics on sandardized and squared sandardized residuals. The esimaed Ljung-Box saisics show ha he EGARCH model fully capures all linear and non-linear dependencies presen in he US and Japanese reurn series, bu only successfully accouns for he non-linear dependencies of he Indian, Pakisan and Sri Lankan reurn series. Our auoregressive formulaions of he condiional mean and condiional variance equaions appear o absorb all he nonlinear serial correlaions presen in he original reurn series 16. On he basis of Jarque- Bera saisics, he hypohesis of univariae normaliy is rejeced for he all markes. Price and Volailiy Spillovers We nex esimae he univariae EGARCH (1, 1) model given by equaions [7], [8], and [9] for each marke o es for reurn and volailiy spillovers. The resuls are shown in Table 4. Panels A, B and C repor he reurn spillover coefficiens, volailiy spillover coefficiens, and he diagnosics on sandardized and squared 15 A non-linear Wald es is used o es for he join significance of he EGARCH model, as sandard - saisics do no work since we have a non-linear ML. The very large Wald saisic indicaes he presence of an EGARCH volailiy model. 18

19 sandardized residuals respecively. The full model considers boh reurn and volailiy spillovers from he world source of shocks (he US) and he regional source of shocks (Japan) o he hree small emerging markes. In erms of firs momen inerdependencies (reurn spillovers), here are posiive significan reurn spillovers from he US o India, Pakisan, and Sri Lanka respecively; all hree US reurn spillover coefficiens (0.0989, , and ) are saisically significan a convenional levels. There is a posiive significan reurn spillover from Japan o Pakisan, bu here are negaive significan reurn spillovers from Japan o India and Sri Lanka. Again, all hree Japanese reurn spillover coefficiens ( , , and ) are saisically significan. Moreover, he magniude of he spillover coefficiens varies from a low of from he US o Sri Lanka o a high of , from he US o India. TABLE 4 ABOUT HERE Turning o second momen inerdependencies (volailiy spillovers), a saisically significan spillover effec exiss from US o India a he 10 per cen level of significance, from US o Sri Lanka a he 1 per cen level, and from Japan o Pakisan a he 5 per cen level. The magniude of he volailiy spillover coefficiens also varies. Specifically, he coefficien from he US o Sri Lanka (0.0209) is greaer han is counerpars from Japan o Pakisan (0.0097), and from he US o India (0.0056); hese findings indicae ha he US, proxying for he world facor as a source of shocks, has more impac on he Asian small markes. In addiion, he coefficien measuring asymmery, θ, is significan for he Indian and Sir Lankan markes, which 16 Higher-order lags could no eliminae he linear serial correlaion presen in he Indian, Pakisan and 19

20 means ha any negaive news (innovaions) from he US marke increase volailiy more han posiive news of similar size from he same marke. Thus, boh he Indian and Sri Lankan markes presen evidence consisen wih an asymmeric response of volailiy o innovaions from he US marke. Numerically, bad news from he US marke for Indian and Sri Lankan markes have 1.06 imes, and 1.07 imes he impac of good news as indicaed by he relaive asymmery raio. The spillovers are symmeric for he Pakisan marke since he coefficien measuring asymmery is insignifican. Comparing he coefficiens from he univariae EGARCH model (resriced model) wih hose of he spillover model (unresriced model) (i.e. Tables 3 and 4, respecively), we can see ha boh ses of resuls are consisen. The coefficiens α 1, α 2 (for he one-lag and wo-lag condiional means) and b 1 (for he one-lag condiional variances) all are highly significan; b 1 is close o uniy as well. These findings clearly indicae ha boh he reurns and volailiy of all hree small markes respond o heir own pas informaion. Thus, curren informaion for a marke remains imporan for all fuure forecass of he condiional mean and condiional variance of ha marke. Condiional volailiies of he reurns in he Pakisan and Sri Lankan markes respond symmerically o heir own pas innovaions; he θ coefficiens repored in Table 3 for hese wo markes are insignifican. Also, evidence of asymmeric volailiy ransmission from eiher of he developed markes o he Pakisan marke is no presen; he θ coefficien repored in Table 4 for his marke is insignifican. However, afer aking ino accoun volailiy spillover, he Sri Lankan marke becomes sensiive o news originaing from he US marke more srongly when he news is bad han when he news is good. The Indian marke also responds Sri Lankan reurn series. 20

21 asymmerically o is own pas innovaions and also o world shocks; boh he θ coefficiens repored in Tables 3 and 4 are negaive and significan. We use he likelihood raio (LR) saisic o es he hypohesis ha reurn and volailiy spillovers from he wo developed markes o hree small markes are joinly zero (i.e. he univariae EGARCH model versus he spillover model). The null hypohesis canno be rejeced a any significance level, implying he imporance of reurn and volailiy spillovers. The Ljung-Box saisics for he sandardized and squared sandardized residuals repored in his unresriced model indicae he presence of limied spillover effecs as he values repored in he able are very close o hose calculaed for he resriced model. The Jarque-Bera normaliy es saisics indicae ha sandardized residuals for all hree indices exhibi srong deviaions from normaliy. In shor, he exisence of firs and second momen inerdependencies poins o he presence of a global markeplace; however, he degree of inerdependencies is limied. Subperiod Price and Volailiy Spillovers The Asian financial crisis sared in mid-1997 and lased unil he end of The mos direcly affeced naions were from Souheas Asia, namely Malaysia, Thailand, he Philippines and Indonesia. However, oher counries soon became affeced. Due o financial conagion, markes fell across he globe and he implicaions of he Asian financial urmoil became far-reaching. For example, in he US he Dow Jones Indusrial Average fell by 554 poins on Ocober 27, The crisis badly affeced Japan which was he bigges rading parner of he main Asian counries affeced and he main supplier of foreign capial o Asian markes. TABLE 5 ABOUT HERE 21

22 The resuls for he unresriced model (i.e. univariae EGARCH(1,1) wih spillover effec) for he hree sub-periods are repored in Table 5. Coefficien a 1 (measuring ARCH effec) and b 1 (measuring volailiy persisence) are significan for almos all markes in he hree periods. The α 1 coefficien (measuring he reurn persisence) is significan on average, excep for India during in-crisis period and for Pakisan during in- and pos-crisis periods. The findings are consisen wih he resuls repored in Table 4 for he enire period; ha is, for hese small emerging markes, pas informaion can be used o forecas boh sock marke reurns and variance. Finally, he Ljung-Box saisics for he sandardized and squared sandardized residuals indicae ha he univariae EGARCH model wih spillover effecs are correcly specified. For he pre-crisis period, here is evidence of reurn spillovers from Japan o all hree small markes. There is also evidence of volailiy spillovers from Japan o he Pakisan and Sri Lankan markes. However, hese spillovers are symmeric since he θ coefficiens (measuring he asymmery) for boh markes are insignifican. For he in-crisis period, he Indian marke shows evidence of reurn spillovers from boh he US and Japanese markes; he Pakisan marke also shows signs of reurn spillovers from he Japanese marke. There is no evidence of reurn spillovers for Sri Lankan markes and also no evidence of volailiy spillovers for any marke. However, he θ coefficien is significan for he Indian and Pakisan markes, implying boh markes respond asymmerically o heir own pas innovaions. For he pos-crisis period, here is evidence of boh reurn spillovers and volailiy spillovers from he US marke o he Indian, Pakisan, and Sri Lankan markes. In addiion, here is some evidence of reurn spillovers from he Japanese o he Pakisan marke and volailiy 22

23 spillovers from he Japanese o he Indian marke. However, he volailiy spillovers are only asymmeric in he Indian marke as he coefficien θ is only significan for India. Thus, he Indian marke appears o respond asymmerically o is own pas innovaions and o innovaions from he wo developed markes as well. A comparison of he resuls from he hree sub-periods reveals ha during he crisis he small markes are comparaively isolaed. In more recen years, however, hese markes have grown more inerdependen in he sense ha informaion affecing asse prices has become more global in naure. We also find ha during he pre-crisis period, hese small markes are more responsive o price changes in he Japanese marke which suggess ha a regional facor dominaes he source of spillovers. However, during he pos-crisis period, he small markes have become more sensiive o news originaing in he US marke which indicaes ha he world facor is he source of spillovers. Even hough we find significan volailiy spillovers in hese markes, he volailiy ransmission is no all asymmeric in he sense he bad news (a marke decline) in one marke has a greaer impac on he volailiy of he nex marke o rade. Discussion Since governmens have implemened financial liberalisaion policies, he capial markes in Souh Asian counries have become more dependen upon news from heir developed marke counerpars which are ofen he sources of capial ouflows. This fac is confirmed by he findings of significan reurn spillovers from boh he world s larges (he US) and he region s larges (he Japanese) sock markes o all hree Souh Asian sock markes. 23

24 The reurn and volailiy spillovers observed from he US marke o he Indian marke are hardly surprising as he US is India s bigges foreign rade parner as well as is larges cumulaive invesor - boh in Foreign Direc Invesmen (FDI) and Foreign Porfolio Invesmen (FPI). According o he Inernaional Financial Saisics Yearbook, for example, he FDI inflows from he US consiued abou 16 percen of he oal acual inflow ino he economy in Ou of he 538 Foreign Insiuional Invesors (FIIs) regisered wih he BSE, 220 were from he US. An invesmen of nearly USD7 billion ou of a oal of USD13 billion by FIIs in he Indian capial markes was from he US. This accouns for abou 47 percen of he ne invesmens made by he FIIs since However, FPI inflows are very volaile. For example, in 1998, FDI inflows from he US were negaive. As Granger e al. (1999) highlighed, foreign invesmens o emerging markes are exremely volaile and depend on changing economic condiions. Since independence, Pakisan has had o depend on foreign assisance in is developmen effors. Japan is is larges donor and he bigges invesor. According o he Inernaional Financial Saisics Yearbook, he share of financial flows from Japan o Pakisan amouned o 91.9 percen, 39 percen, and 59 percen of oal donaions in 1998, 1999, and 2000 respecively. The oal cumulaive amoun of ne disbursemen from Japan's Official Developmen Assisance (ODA) o Pakisan reached USD4 billion hrough As a resul, i is no oo surprising o find ha he volailiy of he Japanese capial marke influences he volailiy of Pakisan equiy values. Due o he small size of Sri Lankan economy, expor-oriened indusries are exremely imporan. Sri Lanka and he US enjoy cordial rade relaions. Since he proporion of expors o he US as a percenage of oal expors has reached an average of 40 percen during , according o Inernaional Trade Saisics 24

25 Yearbook, we would herefore expec he volailiy of he US economy o be ransmied o he Sri Lankan marke. I is ineresing o see ha he Souh Asian sock markes do no show any volailiy spillovers from he US and/or Japan during he in-crisis period. The Souh Asian counries ha were examined in his sudy have been relaively insulaed from he 1997 financial crisis. One reason migh be ha he financial secors of hese counies migh no have been libaralised o he exen ha is eviden in Eas Asian counries. Also, hese counries, and in paricular heir companies, are less exposed o foreign deb. V. Conclusion This sudy invesigaes he magniude and changing naure of he reurn and volailiy spillovers from he US and Japan o he hree small Souh Asian sock markes: namely India, Pakisan and Sri Lanka. We use a univariae Exponenial Generalized Auoregressive Condiionally Heeroskedasic (EGARCH) spillover model o accoun for asymmeries in he volailiy ransmission mechanism, i.e. he possibiliy ha bad news in a given marke has a greaer impac on he volailiy of he reurns of an oher marke han good news. We also aemp o disinguish world forces (he US) from regional facors (Japan). The ess cover he period 01/01/ /12/2003. To examine wheher or no here are srucural shifs in he inernaional marke dynamics, he ess are also conduced for hree sub-periods pre-crisis, incrisis, and pos-crisis. A number of findings emerge from he analysis. Firs, for he enire period analysed, boh world and regional facors are imporan in explaining reurns and volailiy in he hree Souh Asian counries examined, alhough he world marke 25

26 influence ends o be greaer. We find evidence of significan reurn spillovers from he US and Japan o all hree small markes. We also documen evidence of volailiy spillovers from he US o Sri Lanka and from Japan o Pakisan a he 5 per cen significance level and from he US o India a he 10 per cen significance level. Second, he volailiy ransmission mechanism is asymmeric bu only from he US sock marke, i.e. negaive innovaions in US equiy prices increase volailiy in he Indian and Sri Lankan sock markes considerably more han posiive innovaions. Third, no volailiy spillovers exised during he period of Asian crisis. More spillovers, and spillovers of greaer inensiy, were uncovered during he pos-crisis period. In mos cases, spillovers during he pos-crisis period were no asymmeric. Finally, he relaive imporance of he world and regional marke facors was influenced by he Souheas Asian financial crisis. The sub-period analysis revealed ha before he crisis, regional facors were more imporan han heir world facor counerpars; in oher words, he Japanese sock marke was he source of price and volailiy spillover for he Souh Asia region. However, afer he crisis, he world facors dominae he regional facors; ha is, he US sock marke has had a larger impac on small Souh Asian sock markes. 26

27 References Akgiray, V., Condiional Heeroskedasiciy in Time Series of Sock Reurns: Evidence and Forecass. Journal of Business 62, Bae, K.H., Karolyi, G.A., Good News, Bad News and Inernaional Spillovers of Sock Reurn Volailiy Beween Japan and he U.S. Pacific-Basin Finance Journal 2, Baele, L., Volailiy Spillover Effecs in European Equiy Markes: Evidence from a Regime Swiching Model. Working Paper, Ghen Universiy. Baillie, R.T., Bollerslev, T., Inra-Day and Iner-Marke Volailiy in Foreign Exchange Raes. Review of Economics Sudies 58, Bekaer, G., Harvey, C.R., Time-Varying World Marke Inegraion. Journal of Finance 50, Bekaer, G., Hodrick, R.J., Characerizing Predicable Componens in Excess Reruns on Equiy and Foreign Exchange Markes. Journal of Finance 47, Becker, K.G., Finnery, J.E., Gupa, M., The Inernaional Relaion beween he U.S. and Japanese Sock Markes. Journal of Finance 45, Black, F., Sudies of Sock Marke Volailiy Changes. Proceedings of he American Economerics 31, Bollerslev, T., Chou, R., Kroner, K., ARCH Modeling in Finance: A Review of he Theory and Empirical Evidence. Journal of Economerics 52, Bollerslev, T., Wooldridge, J., Qusai-maximum Likelihood Esimaion and Inference in Dynamic Models wih Time-Varying Covariances. Economeric Reviews 11,

28 Booh, G.G., Haem, J., Viranen, I., Yli-Olli, P., Sochasic Modeling of Securiy Reurns: Evidence from he Helsinki Sock Exchange. European Journal of Operaional Research 56, Booh, G.G., Marikainen, T., Tse, Y., Price and Volailiy Spillovers in Scandinavian Sock Markes. Journal of Banking and Finance 21, Braun, P., Nelson, D., Sunier, A., Good News, Bad News, Volailiy and Beas. Working paper, Universiy of Chicago. Campbell, J.Y., Hamao, Y., Predicable Sock Reurns in he Unied Saes and Japan: A Sudy of Long-erm Capial Marke Inegraion. Working Paper, Naional Bureau of Economic Research. Campbell, J.Y., Hamao, Y Predicable Sock Reurns in he Unied Saes and Japan: A Sudy of Long-erm Capial Marke Inegraion. Journal of Finance 47, Chan-Lau, J.A., Ivaschenko, I., Asian Flu or Wall Sree Virus? Price and Volailiy Spillovers of he Tech and Non-Tech Secors in he Unied Saes and Asia. IMF Working Paper. Chueng, Y.W., Ng, L.K., Sock Price Dynamics and Firm Size: An Empirical Invesigaion. Journal of Finance 47, Cheung, Y.W., Fung, H.G., Informaion Flows beween Eurodollar Spo and Fuures Markes. Mulinaional Finance Journal 1, Chin, K., Chan, K.C., Karolyi, A., Inra day Volailiy in he Sock Index and Sock Index Fuures Markes. Review of Financial Sudies 4, Chrisie, A.A., The Sochasic Behavior of Common Sock Variances: Value, Leverages and Ineres Rae Effecs. Journal of Financial Economics 10,

29 Engle, R.F., Auoregressive Condiional Heeroscedasiciy wih Esimaes of he Variance of Unied Kingdom Inflaion. Economerica 50, Engle, R.F., Discussion: Sock Volailiy and he Crash of 87. Review of Financial Sudies 3, Engle, R.F., Io, T., Lin, W., Meeor Shower or Hea Waves? Heeroscedasic Inra-Daily Volailiy in he Foreign Exchange Marke. Economerica 58, Engle, R.F., Ng, V., Measuring and Tesing he Impac of News on Volailiy. Journal of Finance 48, Eun, C., Shim, S., Inernaional Transmission of Sock Marke Movemens. Journal of Financial and Quaniaive Analysis 24, Frazscher, M., Financial Marke Inegraion in Europe: on he Effecs of EMU on Sock Markes. Inernaional Journal of Finance and Economics 7, French, K.R., Schwer, G.W., Sambaugh, R.F., Expeced Sock Reurns and Volailiy. Journal of Financial Economics 19, Glosen, L., Jagannahan, R., Runkle, D., Seasonal Paerns in he Volailiy of Sock Index Excess Reurns. Journal of Finance 48, Granger, C.W. J., Huang, B.N., Yang, C.W., A Bivariae Causaliy Beween Sock Prices and Exchange Raes: Evidence from Recen Asian Flu. Universiy of California a San Diego (UCSD) Economics Discussion Paper. Hamao, Y., Masulis, R. Ng, V., Correlaions in Price Changes and Volailiy Across Inernaional Sock Markes. Review of Financial Sudies 3, Hamilon, J.D., Time Series Analysis. Princeon Universiy Press, Princeon, New Jersey. Harvey, C.R., Predicable Risk and Reurns in Emerging Markes. Review of Financial Sudies. 8,

30 Hilliard, J., The Relaionship beween Equiy Indices on World Exchanges. Journal of Finance 34, Hsieh, D., Modeling Heeroskedasiciy in Daily Foreign Exchange Raes. Journal of Business and Economics Saisics 7, Joen, B.N., von Fursenberg, G.M., Growing Inernaional Co-movemen in Sock Price Indexes. Quarerly Review of Economics and Business 30, Kanas, A., Volailiy Spillovers Across Equiy Markes: European Evidence. Applied Financial Economics 8, Karolyi, G.A., A Mulivariae GARCH model of Inernaional Transmissions of Sock Reurns and Volailiy: he Case of he Unied Saes and Canada. Journal of Business and Economics Saisics 13, Karolyi, G.A., Sulz, R.M., Why do Markes Move Togeher? An Invesigaion of US-Japan Sock Reurn Comovemens Using ADRs. Journal of Finance 51, Koch, P.D., Koch, T.W., Evaluaion in Dynamic Linkages across Daily Naional Sock Indexes. Journal of Inernaional Money and Finance 10, Koumos, G., Asymmeric Volailiy and Risk Reurn Tradeoff in Foreign Sock Markes. Journal of Mulinaional Financial Managemen 2, Koumos, G., Booh, G.G., Asymmeric Volailiy Transmission in Inernaional Sock Markes. Journal of Inernaional Money and Finance 14, Lo, A.W., MacKinley, A.C., Sock Marke Prices do no Follow Random Walks. Review of Financial Sudies 1, Ng, A., Volailiy Spillover Effecs from Japan and he US o he Pacific-Basin. Journal of Inernaional Money and Finance 19,