Transmission of Stock Returns and Volatility: the Case of Korea *

Transcription

1 THE JOURNAL OF THE KOREAN ECONOMY, Vol. 5, No. 1 (Spring 4), Transmission of Sock Reurns and Volailiy: he Case of Korea * Sang-Moon Hahm ** The exen of inernaional financial inegraion among he developed economies has been well documened in he lieraure. This paper examines wheher here are lagged spillovers in reurn and volailiy beween he U.S. and Korea, an emerging economy, for a sample period including he financial crisis of Using open-oclose KOSPI and S&P 5 reurns, his paper finds saisically significan lagged volailiy spillovers from Korea o he U.S. bu no from he U.S. o Korea. This paper also finds ha saisically significan lagged reurn spillovers do no exis in neiher he Korean nor he U.S. sock markes. Thus, ha domesic marke efficienly adjuss o foreign informaion holds even for an emerging marke. Finally, his paper finds ha when KOSPI reurns measured in U.S. dollars are used, saisically significan lagged reurn spillovers exis from he U.S. o Korea bu no from Korea o he U.S. This paper concludes ha he lagged reurn spillovers wih reurns measured in U.S. dollars may resul from he way he Korean governmen has inervened in he KRW/USD foreign exchange marke. JEL Classificaion: G15 Keywords: spillover, sock reurns, volailiy, GARCH * I am graeful for he helpful commens and suggesions from he paricipans of he Ausralasian Meeing of he Economeric Sociey, Sydney, July 3, and he KAEA Meeing, San Diego, January 4, and from wo anonymous referees. ** Professor, KDI School of Public Policy and Managemen and Senior Fellow, Korea Developmen Insiue, P.O. Box 113, Chong-Nyang, Seoul 13-1, Korea. Tel.: , Fax.: , smhahm@kdischool.ac.kr.

2 18 Sang-Moon Hahm 1. INTRODUCTION Wih increased globalizaion of financial markes, invesors in a given marke incorporae ino heir financial decisions no jus heir own domesic informaion bu he informaion of foreign financial markes. The exen of inernaional financial inegraion has received much aenion in recen years. For example, dividing daily close-o-close reurns ino dayime and overnigh reurns, Hamao, Masulis, and Ng (199) sudy he shor run inerdependence of reurns and reurn volailiies across New York, London, and Tokyo sock markes. They find pre-crash evidence of lagged volailiy spillovers from New York and London ino Tokyo, bu no spillover effec is found from Tokyo ino London and New York. Wih he inclusion of he Ocober 1987 crash period, hese spillovers become significan in almos all direcions and moreover, hey find significan lagged reurn spillovers using open-o-close sock reurns. Engle, Io, and Lin (199) find ha news which is revealed when one foreign exchange marke is open conribues o he reurn volailiy of he nex marke o open rading. They do no find any evidence ha news in one marke could predic he mean reurn in subsequen markes. Lin, Engle and Io (1994) find ha excep for lagged spillovers from New York o Tokyo for he period afer he 1987 crash, here are no significan lagged spillovers in reurn or volailiy. These sudies examine inerdependancies in daily sock reurns focusing on he financial markes in developed counies such as he U.S., he U.K., and Japan. Recenly, a number of sudies examine he sock marke behavior of some emerging economies. For example, Harvey (1995) finds ha emerging marke reurns are more likely o be influenced by local informaion han hose in developed counries. Wang, Rui, and Firh () sudy he reurn and volailiy behavior of socks of 15 Hong Kong companies lised boh on he Hong Kong and London sock markes and find ha he saellie marke (London) reacs o informaion from he dominan marke (Hong Kong) wih a delay. Lee, Rui, and Wang () find srong spillovers from he lagged NASDAQ reurns and volailiies o Asian second board marke reurns and

3 Transmission of Sock Reurns and Volailiy: he Case of Korea 19 volailiies for a period ha includes he Asian financial crisis of Ji, Cho, and Yang (1) find srong reurn spillovers from he lagged S&P 5 reurns o he Korean sock reurns. Nam and Yuhn (1) find volailiy spillovers from he U.S. sock marke o he Korean sock marke. This paper sudies he exen of financial marke inegraion beween he U.S. and Korea using daily and inraday sock reurns over he 5-monh period, saring November 1, 1997 hrough December 31, 1. The sample period includes he financial crisis in lae During he crisis period, he Korean securiies marke underwen an overhaul. The Korean financial reform was focused on srenghening marke infrasrucure, acceleraing deregulaion and enhancing invesor proecion. For our purpose, he mos relevan changes were he deregulaion of he Korean governmen s resricions on foreign porfolio invesmen. In November 1997, he aggregae foreign porfolio invesmen ceiling was raised from 3 percen o 6 percen of he ousanding shares of a company. In December 1997, i was raised o 5 percen on he 1 h and raised again o 55 percen on he 3 h. Finally in May 1998, he Korean governmen compleely eliminaed he aggregae foreign porfolio invesmen ceiling. Wih he eliminaion of he resricion on aggregae foreign porfolio invesmen, foreign invesors now have much less difficuly in adjusing heir emerging marke porfolios. To separae he effecs of deregulaion and/or he financial crisis, his paper esimaes he models over a full sample period, a period saring from November 1, 1997 o December 31, 1 and a period saring from May 1, 1998 o December 31, 1 (he pos crisis period). 1) Following Hamao e al. (199), Engle e al. (199), and Lin e al. (1994), his paper uses daily open and close price daa o sudy wheher here are lagged spillovers in reurn and volailiy across counries using lagged reurns and esimaed squared residuals from he previously open foreign sock 1) The Korean governmen requesed o IMF a bailou package on Nov 1, Sandard & Poors changed is sovereign credi raing on he Governmen of Korea from AA- o A on Oc 4, 1997, A- on Nov 5, BBB- on Dec 11, B on Dec 3, 1997; BB on Feb 17, 1998; BBB- on Jan 5, BBB on Nov 11, 1999; BBB on Nov 13, 1; and A- on Jul 4,. Noe ha BBB- or beer is considered invesmen grade raing.

4 Sang-Moon Hahm marke. Some previous sudies sugges ha changes in exchange raes may affec he behavior of sock index reurns. See, e.g., Roll (199). Reurns in erms of U.S. dollars may be relevan o inernaional invesors. Korean invesors may also wan o have informaion on he U.S. sock marke ne of any movemens in he foreign exchange marke. This consideraion migh be paricularly relevan for Korea during he financial crisis. During November and December of 1997, Korean won depreciaed dramaically from 961 won per U.S. dollar o 196 won per U.S. dollar (USD). Such a change in he value of dollar wih respec o Korean won (KRW) migh have quie an impac on he behavior of Korean sock marke paricipans. In order o sudy he impacs of changes in foreign exchange raes on cross marke informaion ransmission, his paper also examines open-o-close KOSPI reurns measured in U.S. dollars. In each marke, his paper uses he mos comprehensive and diversified sock index. For he Korea Sock Exchange, he Korea Composie Sock Price Index (KOSPI) is used. I is a marke value weighed index and i accouns for 1 percen of he equiy capializaion of he Korea Sock Exchange (KSE). There were 689 firms lised on he Korea Sock Exchange a he end of 1. Unil December 6, 1998, KSE opened is rading a 9:3 A.M. and a 9: A.M. hereafer, ) and closes a 3: P.M., Korean Sandard Time. 3) For he New York sock marke, he Sandard & Poor s 5 Composie Index (S&P 5) is used. I is also a marke value weighed index. As of 1, S&P 5 accouns for 85 percen of he equiy capializaion of he New York Sock Exchange (NYSE) and some NASDAQ and AMEX socks are also represened. NYSE opens a 9:3 A.M. and closes a 4: P.M. EST. Noe ha Korea is ahead of New York ) Noe ha invesors can place orders one hour before KSE officially opens. Orders placed from 8: A.M. hrough 9: A.M. and 1 minues before he marke closing (:5 P.M. 3: P.M.) are aggregaed and mached a a single price ha minimizes he imbalance beween buy and sell. From 9: A.M. hrough :5 P.M., orders are mached by coninuous aucions. 3) Unil Dec 6, 1998, KSE closed for lunch breaks beween 11:3 A.M. and 1: P.M., and unil May 1,, beween noon and 1: P.M. Lunch breaks were abolished on May,.

5 Transmission of Sock Reurns and Volailiy: he Case of Korea 1 Figure 1 Exchange Trading Hours (Korean Sandard Time, Summer) :3 : 5: 9: 15: NYSE opens NYSE closes KSE opens KSE closes by eiher 13 hours (in he summer) or 14 hours (in he winer). Thus, he rading aciviy on he KSE and he NYSE are no concurren. Holidays across he U.S. and Korean sock markes are no synchronous and Saurday rading was allowed on he KSE unil December 6, For such cases, he model is esimaed wihou domesic reurns for any days where he foreign marke is closed. As is well known, he use of index prices near he open of rading may cause some difficulies. When individual socks of he index have no ye opened rading, he previous day s closing price quoes are subsiued ino he index. This subsiuion procedure may arificially induce serial correlaion in reurn daa (see, e.g., Cohen e al., 198; Lo and MacKinley, 1988). To avoid such nonsynchronous rading problem or sale quoe problem, he opening quoe is chosen as a price index quoed 3 minues afer NYSE or KSE officially opens (see Lin e al., 1994). 4) The KRW/U.S. dollar exchange raes used in his paper are he ones deermined in he inerbank marke of he Korean foreign exchange marke. The Korean foreign exchange marke is an over-he-couner (OTC) marke, which opens a 9:3 A.M. and closes a 4:3 P.M. Alhough hese open and close foreign exchange daa are no perfecly synchronized wih he corresponding open and close KOSPI daa, his paper uses hem due o daa availabiliy o conver KOSPI inraday reurns in Korean won ino hose in 4) For open prices, 1: A.M. quoe is used for NYSE. For KSE, 1: A.M. quoe is used hrough December 6, 1998 and 9:3 A.M. quoe is used hereafer.

6 Sang-Moon Hahm U.S. dollars. 5) This paper is organized as follows. Secion discusses he saisical characerisics of KOSPI and S&P 5 reurns and heir implicaions on empirical mehodologies. Secion 3 discusses preliminary OLS esimaion of he co-movemens beween he U.S. and Korean sock reurns. Secion 4 repors he GARCH esimaion resuls for KOSPI and S&P 5 reurns. Lagged volailiy and reurn spillovers are examined in Secion 5 and Secion 6, respecively. Secion 7 summarizes he main findings.. STATISTICAL CHARACTERISTICS OF KOSPI AND S&P 5 RETURNS AND THEIR IMPLICATIONS This paper begins wih an examinaion of he serial correlaion of he close-o-close and open-o-close reurns on he KSE and he NYSE for he full sample period and he pos crisis subperiod. 6) See Table 1. For he full sample period saring November 1, 1977 hrough December 31, 1, he close-o-close KOSPI reurns exhibi saisically significan serial correlaion a every lag, some posiive and some negaive. The open-o-close KOSPI reurns also exhibi serial correlaion a every lag, mosly negaive. The close-o-close KOSPI reurns measured in U.S. dollars again exhibi serial correlaion a every lag. 7) The open-o-close reurns exhibi serial correlaion a every lag excep lag 1. The close-o-close S&P 5 reurns do no exhibi any significan serial correlaion. The open-o-close S&P 5 reurns do no exhibi serial correlaion excep a he lag 1. The close-o-close Korean won/u.s. dollar foreign exchange reurns exhibi serial correlaion a every lag. The open-o-close foreign exchange reurns also exhibis serial 5) Prior o Dec 16, 1997, he Korean foreign exchange rae sysem was called he Marke Average Rae Sysem, which consised of he previously deermined KRW/USD base rae and he daily foreign exchange flucuaion band. Beween Dec 1, 1995 and Nov 19, 1997, he daily allowable foreign exchange band widh was.5% (up or down from he base rae); and beween Nov and Dec 15, 1997, he band widh became 1%. Since Dec 16, 1997, he foreign exchange flucuaion band was abolished and he exchange rae sysem became free floaing. 6) S&P 5 and KOSPI daa were direcly obained from S&P and KSE, respecively. 7) KRW/USD exchange rae daa were obained from he Bank of Korea.

7 Transmission of Sock Reurns and Volailiy: he Case of Korea 3 correlaion a every lag. For he pos crisis subperiod saring May 1, 1998 hrough December 31, 1, he close-o-close KOSPI reurns exhibi posiive serial correlaion a lag 1 and negaive serial correlaion a lags and 5. The open-o-close KOSPI reurns exhibi serial correlaion a every lag, some posiive and some negaive. The reurns exhibi large negaive correlaion a lag 1 and diminished serial correlaion a lag and higher. The close-o-close KOSPI reurns measured in U.S. dollars exhibi serial correlaion a lags 1 hrough 7, and he magniude of he serial correlaion a lag 1 is he larges. For he open-o-close KOSPI reurns measured in U.S. dollars, he serial correlaion a lag 1 is large and negaive and higher order lagged auocorrelaion are less imporan. The close-o-close S&P 5 reurns do no exhibi any serial correlaion. The open-o-close S&P 5 reurns exhibi negaive serial correlaion a lag 1. The close-o-close Korean won/u.s. dollar foreign exchange reurns exhibi serial correlaion a every lag. The open-o-close foreign exchange reurns also exhibi serial correlaion a every lag. For he open-o-close S&P 5 reurns, esimaing a GARCH(1,1)-M model wih higher order MA processes specified produced no evidence supporing he significance of moving-average parameers of a higher order han an MA(1). I has been also considered wheher o include a dummy variable for he rading day following a weekend or holiday o capure poenial day of he week effecs. Ye, inclusion of he Monday dummy does no significanly help he model in explaining he behavior of he openo-close S&P 5 reurns. For he open-o-close KOSPI reurns, a Figure(1, 1)-M model wih ARMA(1, 1) is posied o accoun for he significan serial correlaion in he open-o-close KOSPI reurns. Inclusion of he Monday dummy does no significanly help he model for he enire sample period or for he pos crisis subperiod. For he open-o-close KOSPI reurns measured in U.S. dollars, a GARCH-M model wih ARMA(1, 1) is also used o accoun for he significan serial correlaion and auoregressive condiional heeroskedasiciy in residuals. In his case, inclusion of he Monday dummy in he condiional

8 4 Sang-Moon Hahm variance equaions helps he model in explaining he KOSPI reurns for he full sample period bu no for he pos crisis subperiod. The severe serial correlaion in sock reurns does no seem specific o he Korean case. Raher, i seems common for he emerging economies. See, e.g., Harvey (1995) and Lee e al. (). Some may consider i as evidence of marke inefficiency regarding informaion ransmission. Ye, i may also resul from he liquidiy consrains facing individual raders which seem much severe in he emerging economies, he governmen inervenions in he sock markes, or he explici or implici governmen and/or he exchange imposed resricions on rading aciviies. Figure The Behavior of KOSPI, S&P 5, KRW/USD, and Foreign Reserves Korea Composie Sock Price Index (KOSPI) S&P Exchange Rae (Korean won/u.s dollar) 1 Official Foreign Reserves (Bank of Korea) (billions of U.Sdollars) Figure coninued

9 Transmission of Sock Reurns and Volailiy: he Case of Korea 5 15 KOSPI close-o-close daily reurns 3 KOSPI close-o-close reurns in U.S dollars KOSPI open-o-close reurns KOSPI open-o-close Reurns in U.S. dollars S&P 5 close-o-close reurns 8 S&P 5 open-o-close reurns

10 6 Sang-Moon Hahm Table 1 Daa Summary

11 Transmission of Sock Reurns and Volailiy: he Case of Korea 7 3. OLS ESTIMATION: PRELIMINARY To accoun for he co-movemens beween he U.S. and he Korean sock reurns, he following OLS model is esimaed using close-o-close domesic sock reurns. RKc RUSc = α β 1 RKc β RUSc ε = α β1 RUSc 1 β RKc η, (1) where RKc = close-o-close KOSPI daily reurn and RUSc = close-oclose S&P 5 daily reurn. Table presens he OLS esimaion resuls of Table OLS Esimaion Using Close-o-Close Domesic Sock Reurns RKc = α β1 RKc 1 β RUSc ε RUSc = α β1 RUSc 1 β RKc η RKc close-o-close KOSPI daily reurn and = where = RUSc close-o-close S&P 5 daily reurn. KOSPI S&P 5 Panel A: Sample period: Nov 1, Dec 31, 1 Number of obs Log-likelihood Coeff. -sa coeff -sa α β β F-sa (Prob=.) 1.83 (Prob=.) R-squared Ljung-Box(1) for residuals 9.7 (Prob=.) (Prob=.1) Ljung-Box(1) for residuals squared (Prob=.) 5.41 (Prob=.) Panel B: Sample period: May 1, 1998 Dec 31, 1 Number of obs Log-likelihood Coeff. -sa coeff -sa α β β F-sa 4.1 (Prob=.) 9.8 (Prob=.) R-squared Ljung-Box(1) for residuals 4.44 (Prob=.) (Prob=.7) Ljung-Box(1) for residuals squared 18. (Prob=.1) (Prob=.)

12 8 Sang-Moon Hahm close-o-close sock reurns in erms of own lagged reurns and he mos recen foreign close-o-close reurns. According o Table, he mos recen close-o-close U.S. sock reurns help predic he curren Korean reurns and he mos recen close-o-close Korean sock reurns help predic he curren U.S. sock reurns. These resuls hold for he full sample period and also for he pos crisis period. They resul from he fac ha by using close-o-close reurns, heir rading periods overlap in ime. These resuls are conrary o he popular belief ha he U.S. sock reurns influence Korea bu no vice versa. Noe ha hey seem consisen wih he findings of Lin e al. (1994) ha cross marke inerdependence in reurns is bi-direcional for he wo developed markes, he U.S. and Japan. Now, o examine wheher news from a foreign marke has lasing effecs, he following OLS model is esimaed using open-o-close reurns. RK RUS = α β1 RK β RUS ε = α β1 RUS β RK η, () where RK = open-o-close KOSPI reurn, and RUS = open-o-close S&P 5 reurn. Table 3 presens he OLS esimaion resuls using open-o-close sock reurns. In his case, he mos recen foreign sock reurns do no significanly help predic he curren domesic sock reurns for neiher he U.S. nor Korea. This is consisen wih our previous discussion ha wih open-o-close sock reurns, heir rading periods do no overlap in ime. Noe ha for he open-o-close KOSPI reurns, he Ljung-Box saisic for firs 1 normalized residuals indicae significan serial correlaion and ha for he residuals squared indicae significan auoregressive condiional heeroskedasiciy (ARCH) a convenional levels for he full sample period. For he open-o-close S&P 5 reurns, he Ljung-Box saisics for he residuals squared also indicae significan condiional heeroskedasiciy in he residuals for he full sample period (see Panel A, Table 3). Such resuls end o also hold for he pos crisis period (see Panel B, Table 3). These resuls indicae ha a GARCH model is an appropriae specificaion.

13 Transmission of Sock Reurns and Volailiy: he Case of Korea 9 Table 3 OLS Esimaion Using Open-o-Close Sock Reurns RK = α β1 RK β RUS ε RUS = α β1 RUS β RK η, RK open-o-close KOSPI reurn and = where = RUS open-o-close S&P 5 reurn. KOSPI S&P 5 Panel A: Sample period: Nov 1, Dec 31, 1 Number of obs Log-likelihood Coeff. -sa coeff -sa α β β F-sa 8.69 (Prob=.).7 (Prob=.6) R-squared Ljung-Box(1) for residuals 1.44 (Prob=.4) 7.54 (Prob=.8) Ljung-Box(1) for residuals squared (Prob=.) 9.63 (Prob=.) Panel B: Sample period: May 1, 1998 Dec 31, 1 Number of obs Log-likelihood Coeff. -sa coeff -sa α β β F-sa 6.1 (Prob=.).79 (Prob=.6) R-squared Ljung-Box(1) for residuals (Prob=.7) 7.3 (Prob=.86) Ljung-Box(1) for residuals squared (Prob=.) 1.91 (Prob=.4) 4. GARCH MODELS FOR OPEN-TO-CLOSE RETURNS To examine he behavior of sock reurns furher, he following R = α βh ρr γε ε, ARMA(1,1)-GARCH(1,1)-M model is esimaed for KOSPI open-o-close reurns.

14 3 Sang-Moon Hahm h ε, (3) = a bh 1 c where R = open-o-close sock index reurn measured in domesic currency x 1, and h = he condiional variance of he open-o-close sock index reurn. An ARMA(1, 1) srucure is posied o accoun for severe serial correlaion in he Korean open-o-close sock reurns. For he open-o-close U.S. sock reurns, he above GARCH model wih ρ = is esimaed as he pas sock reurns do no help he model in describing he curren open-oclose U.S. sock reurn behavior. The resuls of his esimaion for he full sample period are shown in Table 4-a, Panel A. In order for a GARCH model o be sable, he sum of b and c mus be less han one. For a model describing he Korean sock reurns, he sum is.9888 and for S&P 5 reurns,.9746, and hey are boh significanly differen from zero and less han one. 8 For KOSPI reurns, an increase in volailiy ends o lower openo-close reurns, while for S&P 5 reurns, an increase in volailiy ends o raise is inraday reurns. The likelihood raio saisics, LR(4) for U.S. and LR(5) for Korea, are boh significan a he one percen level. None of he Ljung-Box saisics for he firs 1 residuals or residuals squared are significan a convenional levels. The coefficien of kurosis is 3.61 for he KOSPI reurns, bu i is 6.8 for he S&P 5 reurns, much greaer han he prediced value of 3. for normaliy. For he open-o-close KOSPI reurns measured in USD, he following model is esimaed Ra α βh ρra γε ε = = a bh cε dd, (3 ) h 8) Unlike Hamao e al. (199), none of he coefficiens in he condiional variance equaions in his paper violae he nonnegaiviy assumpion.

15 Transmission of Sock Reurns and Volailiy: he Case of Korea 31 where Ra = open-o-close KOSPI reurn measured in USD 1, h = condiional variance of Ra, and D = a dummy variable ha akes a value of one on days following weekends and holidays and is zero oherwise. The GARCH model (3 ) is esimaed for he full sample period, a period saring November 1, 1997 hrough December 31, 1. The resuls of he esimaion are shown in able 4-b. The likelihood raio saisic LR(6) is significan a he one percen level, indicaing ha he GARCH model is well specified. The sum of b and c is.9497, significanly differen from zero and less han one. As before, he coefficien of he condiional variance in he mean equaion is negaive and significanly differen from zero. Thus, a rise in volailiy ends o reduce inraday KOSPI reurns in U.S. dollars. The coefficien for he Monday dummy is significanly differen from zero and is negaive. So unlike Fama (1965) for he U.S. case, open-o-close KOSPI reurn variances in U.S. dollars end o be lower on Mondays. A decrease in volailiy ends o raise he open-o-close reurns on average, which implies posiive mean reurns for Korean socks measured in U.S. dollars on Mondays, which is conrary o French (198) and Gibbons and Hess (1981) for he U.S. case. Ljung-Box saisics for he firs 1 normalized residuals or residuals squared are no significan a he convenional level. The skewness for he normalized residuals is.87 and he Kurosis is Thus, here is no indicaion for serious model misspecificaion. For he pos crisis subperiod saring May 1, 1998, hrough December 31, 1, he Monday dummy does no help he GARCH-M model in explaining he behavior of he open-o-close KOSPI sock reurns measured in U.S. dollars. Thus, for his subperiod he model (3 ) is esimaed wih d =. The resul of he esimaion is repored in Table 4-b, panel B. The likelihood raio saisic LR(5) is significan a he one percen level, indicaing ha he GARCH-M model is well specified. Unlike he full sample case, he condiional variance does no have a significan effec on he condiional mean in he pos crisis period. Again, he sum of b and c

16 3 Sang-Moon Hahm Table 4-a GARCH Esimaion Using Open-o-Close Reurns where = R = α βh ρr γε ε h = α bh 1 cε, R open-o-close domesic reurn x1 and = h condiional variance of R. KOSPI S&P 5 Panel A: Sample period: Nov 1, Dec 31, 1 Number of obs Log-likelihood Coeff. -sa Coeff -sa α β γ ρ a b c LR (4) for H : β = γ = b = c = LR (5) for H : β = γ = ρ = b = c = 79.6 Ljung-Box(1)for residuals 11.68(Prob=.39) 5.91 (Prob=.88) Ljung-Box(1)for residuals squared 5.87 (Prob=.88) 5.6 (Prob=.9) Skewness Kurosis Panel B: Sample period: May 1, 1998 Dec 31, 1 Number of obs Log-likelihood Coeff. -sa Coeff -sa α β γ ρ a b c LR (4) for H : β = γ = b = c = 55.4 LR (5) for H : β = γ = ρ = b = c = Ljung-Box(1) for residuals 9.45 (Prob=.58) 5.1 (Prob=.93) Ljung-Box(1) for residuals squared 7.79 (Prob=.73) 4.61 (Prob=.95) Skewness Kurosis Table 4-b GARCH Esimaion Using Open-o-Close KOSPI Reurns

17 Transmission of Sock Reurns and Volailiy: he Case of Korea 33 in U.S. Dollars Ra = α βh ρra γε ε h = a bh cε dd, Ra open-o-close KOSPI reurn in U.S. dollars x 1 and h = Ra. where = of KOSPI in U.S. dollars condiional variance Panel A: Sample period: Nov 1, Dec 31, 1 Number of obs 987 Log-likelihood Coeff. -sa α β γ ρ a b c d LR (1) for H : d = 7.9 LR (6) for H : β = γ = ρ = b = c = d = 3.58 Ljung-Box (1) for residual (Prob=.6) Ljung-Box (1) for residuals squared 16.8 (Porb=.14) Skewness -.87 Kurosis Panel B: Sample period: May 1, 1998 Dec 31, 1 Number of obs 87 Log-likelihood Coeff. -sa α β γ ρ a b c LR (5) for H : β = γ = b = c = Ljung-Box(1) for residuals 1.96 ( Porb=.3) Ljung-Box(1) for residuals squared 1.9 (Prob=.5) Skewness -.79 Kurosis Noes: χ (1) criical values:.71 (1%), 3.84 (5%), 6.64 (1%). χ (5) criical values: 9.4 (1%), 11.7 (5%), 15.8 (1%). χ (6) criical values: 1.64 (1%), 1.59 (5%), (1%). is.9653, significanly differen from zero and less han one. Ljung-Box saisics for he normalized residuals or residuals squared are no significan

18 34 Sang-Moon Hahm a he convenional level. The skewness for he residuals is -.79 and he Kurosis is There is no much indicaion of serious model misspecificaion. 5. LAGGED VOLATILITY SPILLOVERS This secion examines wheher here are lagged volailiy spillovers from he previously open foreign sock marke ino he domesic sock marke. Following Engle e al. (199), Hamao e al. (199), and Lin e al. (1994), le us define x as he mos recen squared residual from model (3), using open-oclose reurns of he previously open foreign marke. Wih he inclusion of x in he condiional variance equaion, i becomes: R α βh ρr γε ε = h a bh 1 cε fx, (4) = where x is he mos recen volailiy surprise observed in he foreign marke. Noe ha for he U.S. sock reurns, he above GARCH model wih ρ = is esimaed. The resuls of esimaion using open-o-close reurns measured in domesic moneary unis are shown in Table 5-a. The parameer esimaes repored in Table 5-a are no significanly differen from hose of he model (3) repored in Table 4-a. For he full sample period, a lagged volailiy surprise from he U.S. sock marke o he Korean sock marke is no saisically significan. However, a lagged volailiy surprise from he Korean sock marke o he U.S. sock marke is saisically significan. More specifically, he parameer esimae of he Korean volailiy surprise is negaive and significan a he one percen level. Noe ha he -saisics and likelihood raio saisics can be regarded as a causaliy es. The saisical significance implies ha he Korean open-o-close reurns provide addiional informaion in predicing he U.S. open-o-close reurns. The coefficien of he condiional variance in he mean equaion for he U.S. sock reurns is posiive hough i is significan only a he en percen level is p-value is.61. A volailiy

19 Transmission of Sock Reurns and Volailiy: he Case of Korea 35 surprise from he Korean sock marke ends o lower he U.S. open-o-close sock reurns for he full sample period. 9) For he pos crisis subperiod, he effec of a foreign volailiy surprise is no saisically significan for boh sock exchanges. 1) The remainder of his secion examines wheher significan volailiy spillovers sill exis in he case where all he open-o-close reurns are measured in U.S. dollars. Le xa be he mos recen squared residual from he model (3 ), using open-o-close reurns measures in U.S. dollars of he previously open foreign marke. Wih he inclusion of xa in he variance equaion, i becomes: Ra α βh ρra γε ε = h a bh 1 cε dd fxa, (4 ) = where xa = he mos recen foreign volailiy surprise derived from a model using open-o-close reurn in U.S. dollars. The above model is used o esimae he effec of a volailiy spillover from U.S. o Korea for he full sample period. For he pos crisis subperiod, he above model wih d = is esimaed. For he U.S. sock reurns, he above GARCH model wih ρ = d = is esimaed. The resuls of esimaion using reurns measured in U.S. dollars are shown in Table 5-b. The same qualiaive resuls are found. For he full sample period, saisically significan volailiy spillovers are observed from Korea o he U.S., bu no from he U.S. o Korea. For he 9) Since he kurosis of he normalized reurns is large paricularly for he S&P 5 reurns, his paper has re-esimaed -values for he U.S. case using he Bollerslev and Wooldrige (199) robus sandard errors and covariance procedure, and go he same qualiaive resuls. Noe ha his procedure does no change he parameer esimaes. For he full sample period, he -saisic of he coefficien f is.551 and is p-value is.39; he -saisic of he coefficien β is.1545 and is p-value is.31. 1) As indicaed, Nam and Yuhn (1) find volailiy spillovers from he U.S. sock marke o he Korean sock marke for he period saring 1999 hrough, and Lee, Rui, and Wang () also find volailiy spillovers from NASDAQ o Asian second board markes. Their findings probably resul from heir used of daily sock reurn daa and no properly adjusing for he non-synchronous or sale quoe problem.

20 36 Sang-Moon Hahm Table 5-a GARCH Esimaion of Lagged Volailiy Spillovers Using Open-o-Close Reurns R = α βh ρr γε ε h = α bh cε fx, R open-o-close reurn x 1 and = x mos recen squared residual derived from where = a GARCH model applied o he open-o-close reurn of he previously open foreign marke. Panel A: Sample period: Nov 1, Dec 31, 1 From U.S. o Korea From Korea o U.S. Number of obs Log-likelihood α Coeff. -sa Coeff -sa β γ ρ a b c d LR (1) for H : d =. 5. LR (5) for H : β = γ = b = c = f = 9.14 LR (6) for H : β = γ = ρ = b = c = f = 79.6 Ljung-Box (1) for residuals 11.59(Prob=.4) 6.6(Prob=.86) Ljung-Box (1) for residuals squared 5.8(Prob=.89) 6.1(Prob=.87) Skewness Kurosis Panel B: Sample period: May 1, 1998 Dec 31, 1 From U.S.o Korea From Korea o U.S. Number of obs Log-likelihood α Coeff. -sa Coeff -sa β γ ρ a b c f LR (1) for H : d = LR (5) for H : β = γ = b = c = f = LR (6) for H : β = γ = ρ = b = c = f = Ljung-Box (1) for residuals 9.11(Prob=.61) 5.5(Prob=.9) Ljung-Box (1) for residuals squared 7.69(Prob=.74) 5.1(Prob=.93) Skewness Kurosis Noes: χ (1) criical values:.71 (1%), 3.84 (5%), 6.64 (1%). χ (5) criical values: 9.4 (1%), 11.7 (5%), 15.8 (1%). χ (6) criical values: 1.64 (1%), 1.59 (5%), (1%). Table 5-b GARCH Esimaion of Volailiy Spillovers Using Open-o

21 Transmission of Sock Reurns and Volailiy: he Case of Korea 37 Close Turns Measured in U.S. Dollars Ra = α βh ρra γε ε h = a bh cε dd fxa, Ra open-o-close reurn in U.S. dollars x 1 and xa = where = mos recen squared residual derived from a GARCH model applied o he open-o-close reurn of he previously open foreign marke, measured in U.S. dollars Panel A: Sample period: Nov 1, Dec 31, 1 From U.S. o Korea From Korea o U.S. Number of obs Log-likelihood α Coeff. -sa Coeff -sa β γ ρ a b c d f LR (1) for H : d = LR (5) for H : β = γ = b = c = f = LR (7) for H : β = γ = ρ = b = c = f = 4.68 Ljung-Box(1)for residuals 13.81(Prob=.4) 5.76(Prob=.89) Ljung-Box(1)for residuals squared 15.14(Prob=.18) 5.31(Prob=.9) Skewness Kurosis Panel B: Sample period: May 1, 1998 Dec 31, 1 From U.S.o Korea From Korea o U.S. Number of obs Log-likelihood α Coeff. -sa Coeff. -sa β γ ρ a b c f LR (1) for H : d =.. LR (5) for H : β = γ = b = c = f = 55.4 LR (6) for H : β = γ = ρ = b = c = f = 55.4 Ljung-Box (1) for residuals 1.57(Prob=.3) 5.4(Prob=.93) Ljung-Box (1) for residuals squared 1.(Prob=.53) 4.6(Prob=.95) Skewness Kurosis Noes: χ (1) criical values:.71 (1%), 3.84 (5%), 6.64 (1%). χ (5) criical values: 9.4 (1%), 11.7 (5%), 15.8 (1%). χ (6) criical values: 1.64 (1%), 1.59 (5%), (1%). pos crisis subperiod, saisically significan volailiy spillovers are no

22 38 Sang-Moon Hahm observed. This secion has found ha wheher reurns are measured in domesic currency unis or in U.S. dollars, for he full sample period saisically significan lagged volailiy spillovers are observed from Korea o he U.S., bu no from he U.S. o Korea. 11) Such volailiy spillovers from an emerging economy o he Unied Saes is surprising given he relaive size of he KSE: he marke value of NYSE is sixy imes greaer han ha of he KSE. 1) However, since no such volailiy spillovers are observed for he pos crisis subperiod, he findings in his secion imply ha he lagged volailiy spillovers from Korea o he U.S. are concenraed during he financial crisis period. Furhermore, i may be he resul of he financial inegraion across he sock markes in Eas Asia. The rading aciviies of he Tokyo Sock Exchange (TSE), he Hong Kong Exchange (HKEx), he Singapore Exchange (SGX), and he Korea Sock Exchange are mosly concurren. Thus, any informaion ha may cause he volailiy spillovers from any of he Eas Asian sock marke may be refleced in he Korean sock reurns. Tha is, he volailiy spillovers from Korea o he U.S. migh be regarded as he volailiy spillovers from Eas Asia o he U.S. 6. LAGGED RETURN SPILLOVERS This secion examines wheher here are lagged spillovers on he condiional mean reurn, using open-o-close reurns. Following Hamao e al. (199) and Lin e al. (1994), he GARCH model (4) is modified o include he open-o-close reurn of he mos recen foreign marke, y in he condiional mean equaion. For he open-o-close reurns measured in domesic currency unis, i becomes 11) Wang e al. () have indicaed lagged volailiy spillovers from Hong Kong o London from inspecing sock reurns of 15 Hong Kong firms lised boh on he Hong Kong and London sock exchanges. Their resuls are no derived from examining he reurns of major marke indices such as FTSE 1 and Hang Seng Index. Thus, i is no clear wheher heir resuls are specific o he fac ha heir sample firms are all headquarered in Hong Kong. 1) The marke value of KSE lised securiies amouned o 195 billion USD and ha of NYSE lised securiies amouned o 11,714 billion USD a he end of 1.

23 Transmission of Sock Reurns and Volailiy: he Case of Korea 39 R α βh ρr φy γε ε = h a bh 1 cε fx, (5) = where y = open-o-close reurn of he previously open foreign marke. In acual esimaion, several variaions of he model described above are used. To check wheher here are reurn spillovers from he U.S. in he Korean open-o-close daa, he above model wih f = is esimaed since he volailiy surprise from he U.S. is no saisically significan regardless of he sample period, according o Table 5-a. For he U.S. open-o-close reurns, ρ = is posied since he own pas reurns do no help explain he curren reurns, as indicaed before. For he full sample period he model (5) is esimaed wih he Korean volailiy surprise x, and for he pos crisis subperiod he model wihou x is examined since he Korean volailiy surprise is no saisically significan, according o Table 5-a. The resuls of he esimaion are shown in Table 6-a. In comparison wih he esimaes repored in Table 5-a, he corresponding parameer esimaes do no differ significanly. According o he resuls in Table 6-a, saisically significan lagged reurn spillovers do no exis in neiher he Korean sock marke nor he U.S. sock marke regardless of he sample period, using open-o-close reurns measured in domesic currency. 13) Noe ha Hamao e al. (199) find significan reurn spillovers from he U.S. o Japan. In conras o heir findings, afer adjusing for nonsynchronous rading a open, Lin e al. (1994) find lile evidence agains hypohesis ha domesic marke Table 6-a GARCH Esimaion of Reurn Spillovers Using Open-o- Close Reurns Measured in Domesic Currency 13) For he reasons indicaed in foonoe 9, his paper has re-esimaed -values for he U.S. case using he Bollerslev - Wooldrige procedure, and go he same qualiaive resuls. For he full sample period, he -saisic of he coefficien φ is 1.9 and is p-value is.76; he -saisic of he coefficien f is.354 and is p-value is.4.

24 4 Sang-Moon Hahm R = α βh ρr φy γε ε 1 1 h = a bh cε fx, R open-o-close reurn x 1 and = where = foreign sock marke x 1. y open-o-close reurn of he previously open Panel A: Sample period: Nov 1, Dec 31, 1 From U.S. o Korea From Korea o U.S. Number of obs Log-likelihood α Coeff. -sa Coeff -sa β γ ρ φ a b c f LR (1) for H : φ = LR (5) for H : β = γ = ρ = φ = b = c = 8.4 LR (6) for H : β = γ = φ = b = c = f = 91.3 Ljung-Box (1) for residuals 11.84(Prob=.37) 6.4(Prob=.84) Ljung-Box (1) for residuals squared 6.15(Prob=.86) (Prob=.89) Skewness Kurosis Panel B: Sample period: May 1, 1998 Dec 31, 1 From U.S.o Korea From Korea o U.S. Number of obs Log-likelihood α Coeff. -sa Coeff -sa β γ ρ φ a b c LR (1) for H : φ = LR (5) for H : β = γ = φ = b = c = LR (6) for H : β = γ = ρ = φ = b = c = 6.64 Ljung-Box (1) for residuals 9.9(Prob=.54) 5.1(Prob=.9) Ljung-Box (1) for residuals squared 7.53(Prob=.76) 5.1(Prob=.96) Skewness Kurosis Noes: χ (1) criical values:.71 (1%), 3.84 (5%), 6.63 (1%). χ (5) criical values: 9.4 (1%), 11.7 (5%), 15.8 (1%). χ (6) criical values: 1.64 (1%), 1.59 (5%), (1%). Table 6-b GARCH Esimaion of Reurn Spillovers Using Open-o- Close Reurns Measured in USD

25 Transmission of Sock Reurns and Volailiy: he Case of Korea 41 Ra = α βh ρra φya γε ε h = a bh cε dd fxa, Ra open-o-close reurn in USD x1 and = where = previously open foreign marke in USD x1. ya open-o-close reurn of he Panel A: Sample period: Nov 1, Dec 31, 1 From U.S. o Korea From Korea o U.S. Number of obs Log-likelihood α Coeff. -sa Coeff -sa β γ ρ a b c d f LR (1) for H : φ = LR (5) for H : β = γ = φ = b = c = 9.48 LR (7) for H : β = γ = ρ = φ = b = c = d = 3.1 Ljung-Box(1)for residuals 15.49(Prob=.16) 5.99(Prob=.9) Ljung-Box(1)for residuals squared 13.86(Prob=.4) 5.5(Prob=.96) Skewness Kurosis Panel B: Sample period: May 1, 1998 Dec 31, 1 From U.S.o Korea From Korea o U.S. Number of obs Log-likelihood α Coeff. -sa Coeff -sa β γ ρ a b c f LR (1) for H : φ = LR (5) for H : β = γ = φ = b = c = LR (6) for H : β = γ = ρ = φ = b = c = 31.6 Ljung-Box (1) for residuals 14.73(Prob=.) 5.3(Prob=.9) Ljung-Box (1) for residuals squared 9.3(Prob=.6) 4.7(Prob=.96) Skewness Kurosis Noes: χ (1) criical values:.71 (1%), 3.84 (5%), 6.63 (1%). χ (5) criical values: 9.4 (1%), 11.7 (5%), 15.8 (1%). χ (6) criical values: 1.64 (1%), 1.59 (5%), (1%). χ (7) criical values: 1. (1%), 14.7 (5%), (1%).

26 4 Sang-Moon Hahm efficienly adjus o foreign informaion for developed markes, such as he U.S. and Japan. This paper indicaes ha heir resuls also hold for an emerging marke. 14) Finally, his paper examines wheher saisically significan reurn spillovers exis when all he open-o-close reurns are measured in U.S. dollars. Le ya be he open-o-close reurn of he mos recen foreign marke, measured in U.S. dollars. For he open-o-close reurns measured in U.S. dollars, i becomes Ra α βh ρra φya γε ε = h a bh cε dd fxa (5 ) = To examine he reurn spillovers from he U.S. ino Korea, he model (5 ) wih f = is esimaed since he volailiy surprise from he U.S., when KOSPI reurns are measured in USD, is no saisically significan according o Table 5-b. To examine such effecs from Korea ino he U.S., his paper esimaes model (5 ) wih ρ = d = for he full sample period and wih ρ = d = f = for he pos crisis subperiod, incorporaing he resuls in Table 5-b. The resuls of he esimaion for full sample are repored in Table 6-b. In boh of he markes, he parameer esimaes do no change significanly from hose obained in Table 5-b. Unlike he previous case where reurns are measured in domesic currency unis, saisically significan reurn spillovers are observed in he Korean sock marke. More specifically, reurn spillovers from he U.S. ino Korea are significan a he 5 percen level. Ye, hose from Korea ino he U.S. are no. The exisence of reurn spillovers from he U.S. o Korea, when open-oclose KOSPI reurns are measured in USD, does no seem consisen wih he 14) Unlike his paper, Ji, Cho, and Yang (1) find srong reurn spillovers from he lagged U.S. sock reurns o he Korean sock reurns and conclude ha he Korean sock reurns respond inefficienly o he U.S. reurn changes. Lee, Rui, and Wang () also find spillovers from he lagged NASDAQ reurns o Asian second board marke reurns. Judging from Lin e al. (1994) s findings, Ji e al. s and Lee e al. s findings probably resul from heir using sock reurn daa wihou adjusing for he non-synchronous or sale quoe problem.

27 Transmission of Sock Reurns and Volailiy: he Case of Korea 43 predicions of inernaional asse pricing model. One migh consider i as evidence of inefficien use of informaion by he invesors paricipaing in he Korean sock marke. However, he exisence of reurn spillovers wih sock reurns measured in USD mus be explained in erms of he behavior of he KRW/USD exchange rae since saisically significan reurn spillovers do no exis wih reurns measured in domesic currency. One of he major paricipan in he KRW/USD exchange marke is he Korean governmen. The official foreign reserve holdings of he Korean governmen (or he Bank of Korea) amouned o $4.4 billion in November 1997, and $.4 billion in December Ever since he governmen s official foreign reserve holdings have increased seadily -- $31.6 billion in 1998, $. billion in 1999, $. billion in, and $6.6 billion in 1, so ha is foreign reserve holdings amouned o $1.8 billion a he end of 1 (see Figure ) The Korean governmen raised is official foreign reserve holdings mosly o lower he counry defaul risk in he firs wo or hree years afer he financial crisis and hen o relieve he appreciaion pressure on he Korean won. The governmen s acion mus have remendous impacs on he KRW/USD foreign exchange rae since he major porion of he Korean governmen s foreign reserve holdings is in USD. Thus, he exisence of lagged reurn spillovers wih reurns measured in U.S. dollars may reflec he way he Korean governmen has accumulaed is U.S. dollar based asses. 7. CONCLUSION The exen of inernaional financial inegraion among he developed economies has been well documened in he lieraure. This paper has examined wheher here are lagged spillovers in reurn and volailiy beween he U.S. and Korea, an emerging economy, for a sample period including he financial crisis of Using open-o-close KOSPI and S&P 5 reurns, his paper has found saisically significan lagged volailiy spillovers from Korea o he U.S., bu no from he U.S. o Korea. Such spillovers are no observed for he pos crisis subperiod. These findings imply ha volailiy

28 44 Sang-Moon Hahm spillovers are concenraed during he crisis period. Furhermore, since rading aciviies in he Eas Asian sock markes are mosly concurren, any informaion ha may cause he spillover from any of he Asian sock markes may be refleced in he Korean sock reurns. This paper has also found ha saisically significan lagged reurn spillovers do no exis in neiher he Korean sock marke nor he U.S. sock markes. This is consisen o he finding of Lin e al. (1994) ha domesic marke efficienly adjus o foreign informaion for he U.S. and Japan. This paper indicaes ha heir resuls also hold for an emerging economy. Finally, his paper has found ha saisically significan lagged reurn spillovers exis from he U.S. o Korea when reurns measured in USD are used. Given ha he amoun of he Korean governmen s official foreign reserve holdings increased by 4 percen beween December 1997 and December 1, he lagged reurn spillovers wih reurns measured in U.S. dollars may resul from he way he Korean governmen has inervened in he KRW/USD foreign exchange marke. REFERENCES Bollerslev, T., and J. Wooldridge, Quasi-Maximum Likelihood Esimaion and Inference in Dynamic Models wih Time Varying Covariance, Economeric Reviews, 11, 199, pp Cohen, K., G. Hawawini, S. Maler, R. Schwarz, and D. Whicomb, Implicaions of Microsrucure Theory for Empirical Research on Sock Price Behavior, Journal of Finance, 35, 198, pp Engle, E. F., T. Io, and W. Lin, Meeor Showers or Hea Waves? Heeroskedasic Inra-daily Volailiy in he Foreign Exchange Marke, Economerica, 58, 199, pp Fama, E., The Behavior of Sock Marke Prices, Journal of Business, 38, 1965, pp French, K., Sock Reurns and The Weekend Effec, Journal of Financial

29 Transmission of Sock Reurns and Volailiy: he Case of Korea 45 Economics, 8, 198, pp Gibbons, M. R. and P. Hess, Day of The Week Effecs and Asse Reurns, Journal of Business, 54, 1981, pp Hamao. Y., R. W. Masulis, and V. Ng, Correlaions in Price Changes and Volailiy across Inernaional Sock Markes, Review of Financial Sudies, 3, 199, pp Harvey, C., Predicing Risk and Reurns in Emerging Markes, Review of Financial sudies, 8, 1995, pp Ji, C., D. Cho, and C. Yang, Effecs of The U.S. Sock Price Change on The Korean Sock Price, Journal of he Korean Securiies Associaion, 8, 1995, pp. 1-. Lee, B., O. Rui, and S. Wang, Inernaional Transmission beween NASDAQ and Asian Second Board Markes, unpublished manuscrip,. Lin, W. R. Engle, and T. Io, Do Bulls and Bears Move across Borders? Inernaional Transmission of Sock Reurns and Volailiy, Review of Financial Sudies, 7, 1994, pp Lo, A. W., and A. C. MacKinlay, An Economeric Analysis of Nonsynchronous Trading, Journal of Economerics, 5, 199, pp Nam, J. and K. Yuhn, Volailiy Spillover Effecs from The U.S. Sock Marke o The Korean Sock Marke, Kukje Kyungje Yongu, 7, 1, pp (in Korean). Roll, R., Indusrial Srucure and The Comparaive Behavior of Inernaional Sock Marke Indices, Journal of Finance, XLVII, 199, pp Wang, S., O. Rui, and M. Firh, Reurn and Volailiy Behavior of Duallylised Socks: The Case of Hong Kong, Journal of Inernaional Money and Finance, 1,, pp