International Stock Return Linkages : Evidence from Latin American Markets

Transcription

1 Inernaional Sock Reurn Linkages : Evidence from Lain American Markes Mohamed El Hedi AROURI a a LEO, Universié d Orléans mohamed.arouri@univ-orleans.fr Fredj JAWADI b b ESC Amiens School of Managemen and EconomiX fredj.jawadi@supco-amiens.fr Duc Khuong NGUYEN c c ISC Paris dnguyen@groupeisc.com. Absrac This aricle focuses on he exen of sock reurn linkages beween he main Lain American markes and he world marke. We firs esimae he condiional correlaions beween he Lain American markes and he world marke using he DCC-GARCH model of Engle (00). Then, we apply he mulivariae coinegraion es of Johansen (1988) o he esimaed condiional correlaions o check wheher hey move ogeher. Finally, we esimae a VECM and invesigae shor and long-run marke comovemens. Our resuls show ha here is an increase in shor and long erm linkages beween he Lain American markes and he world marke. JEL Classificaion: F37; G15 Keywords: marke comovemens, emerging markes, mulivariae GARCH, coinegraion and VECM. 1

2 1 Inroducion Inernaional diversificaion as a sock marke invesmen sraegy o improve porfolio performance has been exensively discussed since he early papers of Grubel (1968) and Solnik (1974). In fac, correlaions beween asse reurns from differen naional markes are lower han correlaions wihin he same marke [Solnik (1974)]. However, recenly financial markes have become more inegraed in response o he reducion of many resricions and barriers [Bekaer (1995), Gerard e al. (003) and Carrieri e al. (006)]. Many sudies have invesigaed he implicaions of his increasing marke inegraion for gains from inernaional porfolio diversificaion. The earlies papers have concenraed on he idenificaion of shor-run benefis of inernaional porfolio diversificaion. For example, Longin and Solnik (1995) have shown ha correlaions beween major markes have increased in recen years, which may have decreased he benefis of inernaional diversificaion. Bekaer and Harvey (1997) have argued ha correlaions in emerging markes have slighly raised bu remained relaively low, so inernaional invesors seel have opporuniies for porfolio diversificaion by invesing in hese markes. However, i has been also documened ha he comovemens of naional sock markes are sronger during he crisis periods [Forbes and Rigobon (00) and Bekaer e al. (005)]. Oher works have focused on examining long-erm marke linkages by applying he coinegraion mehodology o naional sock marke indices. For example, Kasa (199) and Darra and Zhong (005) have poined ou ha conemporaneous correlaions may no reflec genuine informaion on markes comovemens and could be misleading if invesors have long holding periods. In paricular, he auhors have esablished ha mos developed sock marke indices share a single common sochasic rend, which implies limied gains from long horizon inernaional diversificaion. However, if many sudies have shown ha correlaions beween inernaional sock reurns have increased in recen periods and ha naional marke indices share a common rend, here have been no work on he join behavior of he naional correlaions wih he world marke hemselves. This is he focus of he presen sudy. Precisely, we examine shor and long erm behavior of correlaions beween six major Lain American markes and he world marke. We firs esimae he condiional correlaions beween he Lain American markes and he world marke using he dynamic condiional correlaion GARCH (DCC-GARCH) model. Then, we apply he coinegraion mehodology o he esimaed condiional correlaions o check wheher hey move ogeher. Finally, we esimae a vecor error correcion model (VECM) for heses correlaion coefficiens and invesigae shor and long-run marke linkages. We choose o invesigae he comovemen issue in he conex of Lain American emerging markes since hey rank among he mos maure markes wihin he universe of emerging counries and hey acually arac a paricular aenion from global invesors hanks o heir grea marke openness. The comovemens beween hese markes have been sudied by several recen papers and significan linkages have been found [Chen e al. (00), Johnson and Soenen (003) and Fujii (005)]. As we have menioned before, here have been no work on he common behavior of he naional marke correlaions wih he world marke hemselves. Our analysis of sock marke linkages in Lain American emerging markes has indicaed ha he naional correlaions wih he world marke are relaively low. Thus, inernaional invesors seel have shor-erm opporuniies for porfolio diversificaion

3 by invesing in hese markes. However, wo oher imporan resuls have been esablished. On he one hand, he naional correlaions have significanly increased in recen years, indicaing higher degrees of financial inegraion. Especially, higher comovemens in imes of crisis have been found, suggesing conagion effecs. On he oher hand, he resuls from coinegraion analysis have shown ha many correlaions share a common rend, limiing room for long-run gains from diversificaion in hese markes. In fac, naional correlaions wih he world marke move ogeher and anyone naional correlaion will be represenaive of he behavior of ha group of markes correlaions wih he world marke. From his i follows ha long horizon inernaional invesors in Lain American markes will benefi he leas from inernaional diversificaion when hey will need i he mos. The res of he aricle is organized as follows. Secion describe he daa. Secion 3 presen he mehodology and discuss he empirical resuls. Concluding remarks are summarized in secion 4. - Daa We use he S&P s IFCG oal reurn indices for he six major Lain American emerging markes (Argenina, Brazil, Chile, Colombia, Mexico and Venezuela) plus he MSCI World sock marke index. Monhly reurns are compued over he period 1985:01 o 005:08. All he indices are obained from DaaSream Inernaional and expressed in American dollar. Descripive saisics and sochasic properies of monhly reurns are presened in Table I. Panel A reveals a number of ineresing facs. Compared o he world marke, Lain American emerging markes have higher reurns and risk. The es for condiional heeroscedasiciy rejecs he null hypohesis of no ARCH effec for Argenina, Brazil, Colombia and Mexico. Skewness is mosly negaive and kurosis is above hree. The Jarque-Bera es saisic (JB) srongly rejecs he hypohesis of normally disribued reurns. These facs suppor our decision o use he quasi-maximum likelihood (QML) approach of Bollerslev and Wooldridge (199) while esimaing he model. The null hypohesis of no auocorrelaion of order 1 is rejeced for Chile, Colombia and Mexico. The firs auocorrelaion (repored in Panel B) is highly significan for hese counries, which suggess ha we need o include an auo-regressive correcion in he mean equaions. Panel C repors he uncondiional correlaions among markes. As expeced, here is a posiive correlaion beween Lain American sock markes. The highes uncondiional correlaion is beween Chile and Mexico (43.34%) and he lowes one is beween Argenina and Colombia (7.00%). The values of he uncondiional correlaions are clearly low. This suggess ha here are sill benefis from diversificaion across Lain American emerging markes. 3- Mehodology and Resuls Firs, we will esimae he condiional correlaions beween Lain American markes and he world marke using he DCC-GARCH model and discuss he obained resuls. Then, we will es for coinegraion beween hese correlaions and esimae a VECM. 3

4 3.1- Modelling dynamic condiional correlaions We use he ime-varying correlaions obained from he esimaion of a mulivariae DCC-GARCH model developed by Engle (00) o measure he comovemens beween he sudied markes and he world marke. Assuming ha sock reurns from he k series are mulivariae normally disribued wih zero mean and condiional variancecovariance marix H, our mulivariae DCC-GARCH model can be wrien as follows: r = µ + ε, ε I 1 N(0, H ) H D R D where r is he (k 1) vecor of he reurns; ε is a (k 1) vecor of zero mean reurn innovaions condiional on he informaion available a ime -1; µ δ + δ r δ r for marke i and µ w, = δwo + δw1r w, 1 for he world marke wih i, = i0 i1 i, 1 + i w, 1 r w denoing he world reurn; diagonal being he condiional sandard deviaions of he reurns and condiional correlaion marix and. The marices (1) D is a (k k) diagonal marix wih elemens on is main D and R is he (k k) R are compued as follows: 1/ 1/ D = diag( h... h ) () 11 kk where h ii is chosen o be a univariae GARCH(1,1) process; 1/ 1/ R = ( diagq ) Q ( diagq ) (3) ' where Q = ( 1 α β) Q + α u 1 u 1 + β Q 1 refers o a (k k) symmeric posiive definie marix wih u i = εi / hii, Q is he (k k) uncondiional variance marix of u, and α and β are non-negaive scalar parameers saisfying α + β p 1. The condiional correlaion coefficien compued as follows : (1 α β ) q ρ ij beween wo markes i and j is hen ij i, 1 j, 1 ij, 1 ρ ij = (4) ((1 α β ) q + αu + βq ) 1/ ((1 α β ) q + αu + βq ) 1/ ii i, 1 ii, 1 + αu u + βq where q ij refers o he elemen locaed in he i h row and j h column of he symmeric posiive definie marix Q. jj j, 1 jj, 1 The DCC-GARCH model is esimaed using a wo-sage procedure. Firsly, a univariae GARCH(1,1) model is esimaed for each marke. Secondly, he sandardized residuals from he firs sage are used o infer he condiional correlaion esimaors. The log-likelihood funcion can be rewrien as follows: 1 L = T ' 1 ( nlog(π ) + log D + log R + u R u ) = 1 4

5 The esimaion resuls indicae ha he DCC-GARCH specificaion we use is flexible enough o capure he dynamics of he condiional covariance marix. 1 The Dynamic condiional correlaions beween he Lain markes and he world marke are ploed in Figure 1. The average of correlaions beween he sudied Lain American markes and he world marke is 5.30%. However, hese condiional correlaions vary considerably over ime. Especially, i is observed ha he evoluion of hese correlaions winesses some periods on negaive values. This should mean high diversificaion gains from invesing in Lain American markes. The mos ineresing paerns found is ha here is a clear upward rend in correlaion from 1994 and onwards as a resul of marke liberalizaion and increased globalizaion. The lowes average correlaion beween for sudied markes can be found beween 1985 and Ineresingly, i can be seen ha here are sudden increases in condiional correlaion following he Asian and Brazilian financial crises in and, o less exen, he sock marke crash in 1987 and he Lain American markes crises in 1994 and 001. The increased correlaion in period of crises may be a sympom of inernaional conagion [Bekaer e al. (005)]. To sum up, he esimaed condiional correlaions have increased in recen years, implying higher linkages beween he Lain American markes and he world marke. Furhermore, i seems ha hese correlaions have a common posiive rend. In oher words, he linkages beween he Lain American markes and he world marke end o move ogeher. We es his purpose in he nex secion. 3.- Are he inernaional reurn condiional correlaions coinegraed? The inspecion of figure 1 sugges wo remarks. On he one hand, i seems ha all correlaion coefficiens are characerized by a posiive linear rend, which should indicae ha hese ses are no saionary. On he oher hand, hese correlaions are evolving ogeher noably a he end of he period and hey seem o have he same adjusmen dynamic a long-run. In order o check he coinegraion hypohesis beween heses series, we es, firsly, he inegraion order of hese ses. According o he DF ess, he six series are I(1), which means ha he firs differencing is required o achieve saionariy. Secondly, we proceed o es for coinegraion for he six naional condiional correlaions wih he world marke using Johansen (1988) race saisics. The essence of a coinegraing relaionship is ha he variables in he sysem share a common uni roo process. This mehodology is paricularly suiable in our conex because i provides a flexible funcional form for modeling condiional naional correlaions behavior under a long-erm equilibrium condiion. The resuls of his es are presened in Table II. According o hese resuls, he null hypohesis of absence of coinegraion is rejeced, indicaing ha he series are coinegraed. Besides, we accep he hypohesis for which here is a mos one coinegraion relaionships, showing ha he sudied series evolve ogeher a long-erm around an equilibrium ha is defined by a long-run relaionships. Furhermore, we have esimaed he coinegraion relaionships beween hese series while defining he Argeninean correlaion wih he world marke (ARG) as he endogenous variable. The resuls are given in Table III. These resuls show ha all 1 The resuls of he DCC-GARCH model esimaion are available upon reques from he auhors. 5

6 esimaors excep ha of Chile are saisically significan indicaing he presence of a meaningful economic relaionships a long-erm beween he condiional correlaions of Lain American markes wih he world marke. Finally, we have esimaed a VECM wih six equaions by he maximum likelihood. The resuls are presened in Table IV. The analysis of hese resuls shows ha he error correcion erm is negaive for all markes bu saisically significan a 5% only for Argenina, Brazil and Colombia. This implies he presence of an acive mean reversion a long run in reason of he presence of a significan adjusmen srengh. Bu, his convergence owards he equilibrium is sronger and he resuling mean reversion srengh is more imporan for Colombia. Besides, hese resuls also show ha a shor erm, here is a significan dependence beween he sudied correlaions, for example, beween Argenina and Colombia. This indicaes ha he adjusmen dynamic of he condiional correlaion of one counry wih he world marke is depending no only on is previous dynamic bu also i is funcion of hose of he oher neighbors counries. Overall, hese resuls sugges a significan relaionships beween hese counries, bu his inerdependency is higher for Argenina, Brazil and Colombia. 4- Conclusion This aricle focused on he exen of inernaional sock reurn linkages beween he main Lain American markes and he world marke. We firs esimaed he condiional correlaions beween Lain American markes and world marke using he DCC- GARCH model of Engle (00). Then, we applied he mulivariae coinegraion es of Johansen o he esimaed correlaion coefficiens in order o check wheher hey move ogeher. Finally, we esimaed a VECM and examined shor and long-run marke comovemens. Our resuls showed ha he naional correlaions wih world marke are relaively low, so here is sill furher shor-run room for global invesors o ge in Lain American markes. However, wo oher imporan resuls have been found. On he one hand, he naional correlaions have significanly increased in recen years, indicaing a higher sock marke inegraion. Furhermore, hese correlaions are significanly higher in periods of crisis, suggesing conagion effecs. On he oher hand, our coinegraion ess have esablished ha many correlaions share a common rend, limiing room for longhorizon gains from diversificaion in hese markes. The correlaions of Lain American markes wih he world marke move ogeher and anyone naional correlaion will be represenaive of he behavior of he oher correlaions. Taken ogeher, our findings sugges ha he long-horizon inernaional invesors in Lain American markes will benefi he leas from inernaional diversificaion when hey will need i he mos. Finally, i would be imporan o exend he presen analysis o find ou he inerrelaionship among oher markes and among oher economic variables such as ineres raes, exchange raes and inflaion in order o beer undersand he greaer marke linkages. 6

7 References Bekaer G. (1995), Marke Inegraion and Invesmen Barriers in Emerging Equiy Markes, The World Economic Review, vol.9, n 1, Bekaer G. and Harvey C. (1997), Emerging Equiy Marke Volailiy, Journal of Financial Economics 43, Bekaer G., Harvey C. and Ng A. (005), Marke Inegraion and Conagion, Journal of Business, vol. 78, Bollerslev T. and Wooldrige J.M. (199), Quasi-maximum Likelihood Esimaion and Inference in Dynamic Models wih Time-Varying Covariances, Economeric Review, n 11, Carrieri F, Errunza V. and Hogan K. (006), Characerizing World Marke Inegraion Through Time, forhcoming, Journal of Financial and Quaniaive Analysis. Chen G-M., Firh M., and Rui O.M. (00), Sock Marke Linkages: Evidence from Lain America, Journal of Banking & Finance, 6, DARRAT A. and ZHONG M. (005), Equiy Marke Linkage and Mulinaional Trade Accords: The Case of NAFTA, Journal of Inernaional Finance and Money, vol.4,5, Engle R.F. (00), Dynamic Condiional Correlaion: a New Simple Class of Mulivariae GARCH Models, Journal of Business and Economic Saisics, 0, Forbes K. and Rigobon R. (00), No Conagion, Only Inerdependence: Measuring Sock Markes Comovemens, The Journal of Finance, Fujii E. (005), Inra and Iner-regional Causal Linkages of Emerging Sock Markes: Evidence from Asia and Lain America In and Ou of Crises, Inernaional Financial Markes, Insiuions and Money, 15, Gerard B., K. Thanyalakpark and J. Baen (003), Are he Eas Asian Markes Inegraed? Evidence from he ICAPM, Journal of Economics and Business, 55, Grubel H. (1968), Inernaionally diversified porfolios: welfare gains and capial flows, American Economic Review, 58, Johansen S. (1988), Saisical analysis of coinegraion vecors, Journal of Economic Dynamics and Conrol, 1, Johnson R. and Soenen L. (003), Economic Inegraion and Sock Marke Comovemen in he Americas, Journal of Mulinaional Financial Managemen, 13, Kasa K. (199), Common sochasic rends in inernaional sock markes, Journal of Moneary Economics, 9, Longin F. and Solnik B. (1995), Is he Correlaion in Inernaional Equiy Reurns Consan: ?, Journal of Inernaional Money & Finance, 14, 3-6. Solnik B. (1974), The Inernaional Pricing of Risk: an Empirical Invesigaion of he World Capial Marke Srucure, Journal of Finance, 9,

8 Table I : Descripive Saisics Panel A: Summary Saisics Argenina Brazil Chile Colombia Mexico Venezuela World Mean (% per monh) Sd-Deviaion (% per monh) Skewness ** * ** * -.45 * * * Kurosis * * * * * * * JB Q(1) ARCH(1) Panel B: Auocorrelaions Lag Argenina Brazil Chile Colombia Mexico Venezuela World * * 0.84 * *** *** * *** Panel C: Uncondiional correlaions of r i (in percenage) Argenina Brazil Chile Colombia Mexico Venezuela World Argenina Brazil Chile Colombia Mexico Venezuela MSCI World 1.00 Noes: The es for Kurosis coefficien has been normalized o zero. JB is he Jarque-Bera es for normaliy based on excess skewness and Kurosis. Q(1) is he Ljung-Box es for auocorrelaion of order 1. ARCH is he Engle (198) s es for condiional heeroscedasiciy. *, ** and *** indicae he significance of coefficiens a he 1%, 5% and 10% respecively. +, ++ and +++ indicae he rejecion of he null hypoheses of no auocorrelaion, normaliy and homocedasiciy a he 1%, 5% and 1% levels of significance respecively for saisical ess. Table II : Coinegraion Trace Tes Hypohesized Trace 5 Percen 1 Percen No. of CE(s) Eigenvalue Saisic Criical Value Criical Value None ** A mos A mos A mos A mos A mos Noes: *(**) denoes rejecion of he hypohesis a he 5%(1%) level Table III : Esimae of he coinegraion relaionships 1 Coinegraing Equaion(s): Log likelihood Normalized coinegraing coefficiens (sd.err. in parenheses) ARG BRA CHI COL MEX VEN (0.6111) (0.5378) ( ) ( ) (0.816) Noes : ARG, BRA, CHI, COL, MEX and VEN are respecively he condiional correlaion of Argenina, Brazil, Chile, Colombia, Mexico and Venezuela wih he world marke. 8

9 Table IV : VECM Esimaion Error Correcion: D(ARG) D(BRA) D(CHI) D(COL) D(MEX) D(VEN) CoinEq ( ) (0.0189) (0.0149) (0.0157) (0.013) ( ) D(ARG(-1)) (0.0934) ( ) ( ) ( ) ( ) ( ) D(ARG(-)) ( ) (0.0954) ( ) ( ) ( ) ( ) D(ARG(-3)) ( ) ( ) ( ) (0.0968) ( ) ( ) D(ARG(-4)) ( ) ( ) ( ) (0.0915) ( ) ( ) D(BRA(-1)) ( ) ( ) ( ) (0.0846) ( ) ( ) D(BRA(-)) ( ) ( ) ( ) ( ) ( ) ( ) D(BRA(-3)) ( ) ( ) ( ) ( ) ( ) ( ) D(BRA(-4)) (0.0865) ( ) ( ) (0.0888) (0.0871) ( ) D(CHI(-1)) ( ) ( ) ( ) ( ) ( ) (0.1051) D(CHI(-)) ( ) ( ) ( ) ( ) ( ) (0.1095) D(CHI(-3)) ( ) ( ) ( ) ( ) ( ) ( ) D(CHI(-4)) ( ) ( ) (0.114) ( ) ( ) ( ) D(COL(-1)) ( ) (0.0879) (0.0958) ( ) ( ) ( ) D(COL(-)) ( ) (0.0851) ( ) ( ) ( ) ( ) D(COL(-3)) ( ) (0.083) ( ) ( ) (0.0847) (0.0986) D(COL(-4)) ( ) ( ) (0.0994) ( ) (0.0830) ( ) D(MEX(-1)) (0.0885) (0.0874) ( ) ( ) ( ) (0.0985) D(MEX(-)) ( ) ( ) ( ) ( ) ( ) ( ) D(MEX(-3)) ( ) ( ) (0.0993) ( ) ( ) ( ) D(MEX(-4)) ( ) ( ) (0.0986) (0.0878) ( ) ( ) D(VEN(-1)) ( ) (0.0661) ( ) ( ) ( ) ( ) D(VEN(-)) ( ) ( ) ( ) ( ) ( ) ( ) D(VEN(-3)) ( ) (0.0671) ( ) ( ) ( ) ( ) D(VEN(-4)) ( ) ( ) (0.0784) ( ) (0.0698) ( ) C (0.0004) (0.0001) (0.0033) ( ) (0.0006) (0.007) R-squared Adj. R-squared F-saisic Log Likelihood (d.f. adjused) Akaike Informaion Crieria Schwarz Crieria Noes : D noes he firs difference, ARG, BRA, CHI, COL, MEX and VEN are respecively he condiional correlaion of Argenina, Brazil, Chile, Colombia, Mexico and Venezuela wih he world marke. Sandard errors are in parenheses. As daa are monhly, we defined he delay for he esimaed VECM as p equal o 4. 9

10 Figure 1 : Condiional correlaions beween Lain markes and World marke A R G B R A C H I C O L M E X V E N 10