Stock market integration and financial crises: the case of Asia JIAN YANG1*, JAMES W. KOLARI2, and INSIK MIN3 1

Department of Accounting, Finance & Information Systems, Prairie View A&M University, Texas 77446, USA

2

Department of Finance, Texas A&M University, College Station, Texas 77843, USA

3

Department of Economics, Texas A&M University, College Station, Texas 77843, USA

May 2002

ABSTRACT This study examines long-run relationships and short-run dynamic causal linkages among the U.S., Japanese, and ten Asian emerging stock markets, with the particular attention to the 1997-1998 Asian financial crisis. Extending related empirical studies, comparative analyses of pre-crisis, crisis, and post-crisis periods are conducted to comprehensively evaluate how stock market integration is affected by financial crises. In general, the results for the case of Asia show that both long-run cointegration relationships and shortrun causal linkages among these markets were strengthened during the crisis and that these markets have generally been more integrated after the crisis than before the crisis. Detailed country-by-country analyses are provided, which yield a variety of new results concerning the roles of individual countries in international stock market integration. An important implication of our findings is that the degree of integration among countries tends to change over time, especially around periods marked by financial crises.

JEL Classification: G15, C32 * Corresponding author. Email: [email protected]

Applied Financial Economics (forthcoming)

Stock market integration and financial crises: the case of Asia I. INTRODUCTION A number of empirical studies have examined long-run relationships and short-run dynamic causal linkages among the emerging Asian financial markets and major developed markets.1 Cointegration studies that investigated long-run relationships focus on the extent to which nascent stock markets in Asian countries are internationally integrated and, in turn, have important implications to diversification potential in Asian stock markets (Chan, Gup, and Pan, 1992; Hung and Cheung, 1995; DeFusco, Geppert, and Tsetsekos, 1996; Masih and Masih, 2001). Studies that have estimated short-run dynamic causal linkages seek to better understand the propagation mechanism driving stock market fluctuations in different countries, especially with respect to market crashes (Masih and Masih, 1997a,1999; Ghosh, Saidi, and Johnson, 1999; Sheng and Tu, 2000). These and other related studies (Chung and Liu, 1994; Arshanapalli, Doukas, and Lang, 1995; Cheung, 1997; Janakiramanan and Lamba, 1998; Dekker, Sen, and Young, 2001) employ vector autoregression (VAR) techniques, including cointegration, Granger causality, impulse response analysis, and forecast error variance decomposition. In general, the empirical evidence presented in these studies is mixed with respect to both long-run relationships and short-run dynamic causal linkages. This study re-examines Asian stock market integration around the time of the 1997-1998 Asian financial crisis. We contribute to extant literature in several ways. First, we comparatively document the impact of the Asian financial crisis on those countries’ stock market integration for three different stages of the crisis – namely, pre-

1

crisis, crisis, and post-crisis periods. To our knowledge, no previous work has addressed the issue of how the crisis altered market integration among Asian countries over time. Previous studies of market integration invariably assume that markets are either perfectly integrated, perfectly segmented, or partially integrated but the extent of integration is constant over time (Bekaert and Harvey, 1995). By focusing on how the Asian financial crisis affected market integration over time, we seek to determine if this common assumption holds. Second, we employ the recently developed vector autoregression (VAR) technique of generalized impulse response analysis (Koop, Pesaran, and Potter, 1996; Pesaran and Shin, 1998) to estimate short-run dynamic causal linkages across stock markets. This empirical approach is motivated by the existence of strong contemporaneous correlations among Asian stock market innovations as documented in the literature (Cheung, 1997; Janakiramanan and Lamba, 1998). In such instances it is well known that traditional orthogonalized impulse response analysis (or forecast error variance decomposition) based on the widely-used Choleski factorization of VAR innovations is sensitive to the ordering of the variables. By contrast, generalized impulse response analysis is invariant to the ordering of the variables. Third and last, while previous studies conduct analyses of major Asian markets, we provide more comprehensive information by including the developed markets of the U.S. and Japan as well as ten Asian emerging stock markets. Several small emerging stock markets included in this study, such as India, Indonesia, Pakistan, and the Philippines, have received scant attention in prior research, with the exception of Ghosh, Saidi, and Johnson (1999). In this regard, we model twelve markets in one system rather than using

2

several smaller subsystems, which is important in making accurate inferences concerning cointegration and causality (Hassapis, Pittis, and Prodromidis, 1998). In general, the results for the case of Asia show that both long-run cointegration relationships and short-run causal linkages among these markets strengthened during the financial crisis and that these markets have generally been more integrated after the crisis than before the crisis. More specifically, our results reveal that, while the U.S. exerted substantial influence on most Asian stock markets in all three sample periods, Japan had little or no impact on most markets in the region except during the crisis period. Further results show that, during the non-crisis periods, Japan, Taiwan, and the Philippines were fairly isolated markets, but Thailand and Indonesia were fairly interactive rather than fairly isolated as reported previously in the literature. Korea, India and Pakistan were fairly endogenous, which has not yet been reported in the literature. Singapore instead of Hong Kong appeared to be a regional leader. An important implication of our findings is that, consistent with Bekaert and Harvey, the degree of integration among countries tends to change over time, especially around periods marked by financial crises. The next section briefly reviews the literature. Section 3 describes the empirical framework. Section 4 presents the empirical results. The last section gives the summary and conclusions. II. LITERATURE REVIEW As mentioned above, the empirical findings of previous studies on the integration of Asian stock markets are mixed. Chan, Gup, and Pan (1992) and DeFusco, Geppert, and Tseksekos (1996) reported no cointegration between U.S. and many Asian emerging stock markets (i.e., Hong Kong, Korea, Singapore, Taiwan, Malaysia, Thailand, and the

3

Philippines) in the 1980s and early 1990s. By contrast, Arshanapalli, Doukas, and Land (1995) and Masih and Masih (1997a, 1999, 2001) reported only one cointegrating vector among several major Asian emerging markets (i.e., Hong Kong, Korea, Singapore, and Taiwan) and major developed markets. Chung and Liu (1994) documented two cointegrating vectors between the U.S. and these larger Asian-pacific stock markets. Ghosh, Saidi, and Johnson (1999) documented pairwise cointegration between U.S. (and Japan) and some Asia-pacific stock markets during the 1997-1998 international financial crisis. Sheng and Tu (2000) reported no cointegration in the year before the Asian financial crisis but one cointegrating vector during the crisis between the U.S. and many Asian stock markets. Most of the aforementioned studies are conducted in local currency terms. Masih and Masih (1999, 2001) did utilize U.S. dollars as the base currency, and only Hung and Cheung (1995) employ data collected in terms of both local currency and U.S. dollars. Using U.S. dollar denominated stock prices, the authors found that five major Asiapacific stock markets (i.e., Hong Kong, Korea, Malaysia, Singapore, and Taiwan) were cointegrated after but not before the 1987 stock crash. Interestingly, they found no cointegration throughout the whole sample period using stock prices in local currency terms. While currency numeraire is one explanation for differences in conclusions reached by different studies on the question of Asian stock market integration, other reasons could be frequency of data, model specification, and sample time period. Frequency of data (i.e., daily, weekly, and monthly) likely has only limited effects on the cointegration analysis, as Hakkio and Rush (1991) have shown that, given a fixed sample

4

period, frequency of data did not affect cointegration results. With regard to model specification, many previous studies typically used several smaller subsystems (e.g., bivariate systems) to model relationships among a large number of markets. This practice may ignore potential indirect channels of stock market linkage and generate spurious patterns of stock market linkages (Janakiramanan and Lamba, 1998). Recently, Hassapis, Pittis, and Prodromidis (1998) have demonstrated that cointegration and causality inferences are strongly affected by the omission of an important causing variable in the system. Finally, different sample time periods could account for the different findings. Asian stock markets may (or may not) have been more integrated with each other and with the world for various reasons, including extensive stock market liberalization, increased economic integration within the region and with the world, technological advances in communication, and stock market crashes. Stock market crashes such as the one in 1987 have been widely argued to strengthen major international as well as Asian stock market linkages. This study contributes to the literature by comprehensively examining Asian stock market integration using a twelvecountry VAR system with different currency numeraires and different sample periods surrounding the Asian financial crisis. III. EMPIRICAL FRAMEWORK We model long-run and short-run relationships between emerging Asian and the U.S. and Japanese stock markets using a cointegrated vector autoregression (VAR) framework. Let Xt denote a vector which includes m nonstationary variables (m = 12 countries). If k price series in Xt are cointegrated, they can be expressed using a reduced form error correction model (ECM) with a lag of k-1 (i.e., a lag of k for a levels VAR):

5

∆X t = Γ 1 ∆X t −1 + ... + Γ k −1 ∆X t −k +1 + ΠX t −1 + µ + ε t

(1)

where Π = αβ ' , the matrix α contains short-run adjustment parameters to the long-run relationships reflected in the matrix β, and the rank of Π determines the r number of cointegrating vectors. To comprehensively investigate the impact of the Asian financial crisis on longrun relationships among sample stock markets, we compare numbers of cointegrating vectors among these markets in three periods: pre-crisis (January 2, 1995 – December 31, 1996), crisis (July 1, 1997 – June 30, 1998), and post-crisis (July 1, 1998 – May 15, 2001). As suggested by Kamin (1999, p. 506) and Corsetti, Pesenti, and Roubini (1999, pp. 370-371), the first year after the crisis was marked by stabilization of exchange rates but continued instability in the macroeconomy as a whole. As such, it can be considered as a second phase in the financial crisis. For this reason we further divide the post-crisis period into two sub-periods: July 1, 1998 – June 30, 1999 (transition period) and July 1, 1999 – May 15, 2001 (post-crisis period). Before testing whether the price series are cointegrated, one should establish that each univariate series is nonstationary, or I(1). The existence of a unit root in Asian stock market prices is well established in the literature. In particular, Masih and Masih (1999, 2001) have conducted extensive tests which verify the existence of a unit root for all Asian stock market index prices. Following Yang and Leatham (2001) and Bessler and Yang (2003), we apply a relatively new procedure for testing stationarity of data series (see also Hansen and Juselius (1995)). Unlike commonly-used augmented Dickey-Fuller (ADF) and Philips-Perron (PP) tests, in which the null hypothesis is nonstationarity, the null under this new procedure is stationarity. Specifically, in the presence of one

6

cointegrating vector, Johansen (1991) likelihood ratio tests are applied to test whether any eleven (out of twelve) price series do not enter the cointegrating vector. This procedure is equivalent to testing the possibility that one cointegrating vector might arise simply because the remaining variable is itself stationary. Similarly, in the presence of two cointegrating vectors, likelihood ratio tests are applied to test whether any ten price series are excluded from both cointegrating vectors. No cointegration among twelve variables implies that all variables are nonstationary. The likelihood ratio test results (available upon request from the authors) show that stationarity of each price series can be rejected at either 1 percent or 5 percent levels for all subperiods and for stock prices in both local currency and U.S. dollar terms. Thus, it can be concluded that each price series is I(1). A priori, we expect that Asian stock markets are most cointegrated during the crisis period, which would be captured by the highest number of cointegrating vectors. As mentioned earlier, the number of cointegrating vectors is determined by the rank of Π = αβ ' . In this respect trace test statistics of Johansen (1991) can be calculated to test

the null hypothesis that there are at most r cointegrating vectors. Additional testing of the cointegration space spanned by β can produce further information on long-run market linkages. We are particularly interested in how many markets are actually excluded in all of the identified long-run relationships (if any). This hypothesis can be tested by examining whether β ij = 0 for all i (i = 1, ..., r) cointegration vectors for the jth (j = 1, ..., 12) market. Short-run stock market linkages are reflected in the parameter matrixes α and Γi . While the α parameter matrix defines the short-run adjustments to long-run relationships,

7

the parameter matrixes ( Γ1 ,…, Γp −1 ) define the short-run adjustments to changes in the process. However, it is well recognized that, like standard VAR, the individual coefficients of the ECM can be difficult to interpret (Lutkepohl and Reimers, 1992). In view of this drawback, innovation accounting is the most appropriate method of summarizing the short-run dynamic structure of market linkages. It is common practice in VAR analyses to rely on a Choleski factorization to orthogonalize VAR innovations so that they are uncorrelated contemporaneously. Unfortunately, innovation accounting results based on the Choleski factorization are sensitive to the ordering of variables when the residual covariance matrix is nondiagonal. Here we apply generalized impulse response analysis as developed in Koop, Pesaran and Lee (1996) and Pesaran and Shin (1998), which is invariant to the ordering of the variables in the VAR model.2 Under the generalized VAR approach, ∆X t is rewritten as the infinite moving average version of equation (1), or ∞ ∆X = ∑ C ε , t = 1,2, …, T . t i t −i i=0

(2)

The (scaled) generalized impulse response function which measures the effect on ∆X t + n of the shock to the jth equation in equation (1) can be specified as follows: ψ ( n ) = σ − 1 / 2 C Σe j , j jj n

where

n = 0,1,2,…,

σ jj is jj th element in the variance-covariance matrix Σ , and e j is m x 1 vector

with unity as its jth row and zeros elsewhere.

8

(3)

IV. EMPIRICAL RESULTS Data Our data consist of daily stock index closing prices of two developed markets and ten Asian emerging markets, i.e., Hong Kong Hang Seng (HK), India BSE National(ID), Indonesia Jakarta SE composite (IN), Japan Nikkei 225 Stock Average (JP), Korea SE Composite (KR), Malaysia Kuala Lumpur Composite (ML), Pakistan Karachi SE 100 (PK), Philippines SE Composite (PH), Singapore Strait Times (SG), Thailand Bangkok S.E.T. (TH), Taiwan SE weighted (TW), and U.S. S&P 500 composite (US). The sample period is January 2, 1995, to May 15, 2001, which includes 1,662 daily observations for each series. All stock indices are expressed in both local currency and U.S. dollar terms. The source of data was Datastream. Cointegration analysis results The choice of optimal lags for the VAR system is selected based on the Akaike information criterion (AIC). Two lags are chosen for all periods in this study. Diagnostic statistics reveal that the residuals are generally well behaved and in particular free from autocorrelation problems. Trace test results in terms of local currencies and U.S. dollars are reported in Tables 1 and 2, respectively. As shown in Table 1, using local currencies and excluding a linear trend term, trace test results indicate no cointegrating vector exists in the pre-crisis and transition periods but two cointegrating vectors exist in both the crisis and post-crisis periods. With a linear trend term, similar results are found across periods. Our finding in the pre-crisis period is consistent with Chan, Gup, and Pan (1992), DeFusco, Geppert, and Tsetsekos (1996), and Sheng and Tu (2000); however, unlike Sheng and Tu (2000), we found two

9

rather than one cointegrating vector during the crisis. These results suggest that the longrun integration was intensified in the crisis and post-crisis periods relative to the precrisis period. By implication, we infer that the Asian financial crisis altered the degree of market integration in the region over time. Our results also differ from those of Masih and Masih (1997b), who concluded that the 1987 stock crash did not affect the number of common stochastic trends in major stock markets. Also, it is interesting to note that longrun integration was weaker during the transitional period. We interpret this finding as indirect evidence that linkages among real economic variables may partly account for stock market linkage. Table 2 shows that the results are unchanged using U.S. dollars. The only difference is that one rather than zero cointegrating vector is found in the pre-crisis period, which is consistent with Masih and Masih (1999, 2001). Thus, allowance for the exchange rate adjustment can affect the number of cointegrating vectors, in line with studies by Hung and Cheung (1995) and Bessler and Yang (2003). Table 3 gives the results to whether certain stock market index prices are excluded from the long run relationships identified by the above trace tests. Based on local currencies, only the Hong Kong stock market is excluded from the long run relationships during the crisis period, whereas the Philippines and Taiwan stock markets are ruled out from the long-run relationships in the post-crisis period. The results using U.S. dollars as numeraire reveal that three markets are excluded from the long-run relationships during the crisis and post-crisis periods, four markets are excluded during the pre-crisis period, and eight markets are excluded in the transition period. The latter

10

result confirms our earlier finding that long-run linkage was weakest among the Asian stock markets in the transition period. Generalized impulse response analysis results Due to the weak stock market linkages found in the transition period, we estimated ECMs only for the pre-crisis, crisis, and post-crisis periods. Generalized impulse response functions estimated from the ECMs provide insight into how innovations in a particular market in the system affected other markets through dynamic interactions among markets. Table 4 summarizes these results. To conserve the space, only the results for day 20 based on the local currency terms are reported here due to little difference of dollar-denominated results (results on days 0 to 20 are available from the authors upon request). Consistent with Bessler and Yang (2003), we find that exchange rate adjustments can affect long-run cointegration relationships but do not significantly affect the short-run dynamic causal linkage pattern of stock markets. Following Brocato (1994) and Bessler and Yang (2003), our analyses employ a forecast horizon of 20 days ahead3. Similar to Dekker, Sen, and Young (2001), significant responses are defined as those that exceed 0.20 unit standard deviations on day 20. Following Kamin (1999), we organize Asian markets into three groups: serious victim markets (Indonesia, Korea, Malaysia, Philippines, and Thailand), less serious victim markets (Hong Kong, Singapore, and Taiwan), and non-victim markets (the U.S., Japan, India, and Pakistan). Previous work by Janakiramanan and Lamba (1998) found that the Indonesian market was a relatively isolated market. However, Table 4 shows that quite a few markets, including Hong Kong, India, Malaysia, Philippines, Pakistan, and even the U.S., responded significantly to shocks from Indonesia in the pre-crisis period. In the post-

11

crisis period Hong Kong, Japan, Korea, Malaysia, Philippines, Pakistan, Singapore, and Thailand showed significant responses to shocks from the Indonesian market. Interestingly, different from what is observed in the case of other victim markets, the Indonesian market did not increase its influence on other markets during the period of the crisis. This evidence agrees with the fact that the Indonesian market did not initiate the crisis. From Table 4 it can also be seen that the Indonesian market responded to shocks from other markets, including the U.S. and Japan. Turning to Korea, only Malaysia, Philippines, and Singapore responded to shocks from the Korean market in the pre-crisis period, and only Pakistan, Thailand, and Taiwan was responsive in the post-crisis period. Further evidence that the Korean market held a passive role in Asian markets is reflected in the fact that only Thailand and Taiwan reacted to news from the Korean market during the crisis period. This pattern is an exception relative to other victim Asian markets. It is also shown in Table 4 that the Korean market responded strongly to shocks from other markets, particularly after the crisis. Overall, the Korean market appears to be a fairly endogenous market, which has not been documented in the previous literature. With respect to Malaysia, several markets, such as India, Indonesia, Philippines, Pakistan, and Thailand, had substantial responses to shocks from Malaysia in the precrisis period. While Singapore showed little response to shocks from Malaysia in the pre- and post-crisis periods, Malaysia responded to shocks from Singapore in these periods. We infer that Singapore is more dominant from an informational perspective than Malaysia. This finding is consistent with Janakiramanan and Lamba (1998) but contrary to Dekker, Sen, and Young (2001). The influence of the Malaysian market

12

decreased in the post-crisis period, affecting only the Indian and Indonesian markets, perhaps due to its weakened economy after the crisis. India, Indonesia, Pakistan and Thailand were responsive to shocks from the Philippines market before the crisis and, with the exception of Pakistan, remained responsive after the crisis. It is also obvious that the Philippines market had little response to shocks from other markets. Overall, like Dekker, Sen, and Young (2001), we find that the Philippines market was fairly isolated in non-crisis periods. Also, all other Asian emerging markets under study (but not the U.S. and Japan) responded to shocks from the Philippines market. Such an increased influence on other markets during the crisis period is similar to what most other victim markets experienced. Another market identified in previous studies (Ghosh, Saidi, and Johnson, 1999; Dekker, Sen, and Young, 2001) as relatively isolated is Thailand. However, we find that: Pakistan, India, Indonesia, Korea, Philippines, and Taiwan, significantly responded to shocks from Thailand before the crisis; India, Indonesia, Korea, Philippines, and Singapore responded to shocks from Thailand after the crisis; and all markets, except Taiwan and the U.S., sharply responded to the shocks from Thailand during the crisis. These findings imply that Thailand played an important role in Asian markets, especially during the crisis. The Hong Kong market is believed to be the most interactive and influential market in the Asian region (Masih and Masih, 1999; Dekker, Sen, and Young, 2001). Our results reveal that only Pakistan and Taiwan responded to shocks from Hong Kong in the pre-crisis period. However, the Hong Kong market does become a more influential market after the crisis, with India, Indonesia, Korea, Pakistan, and Singapore all

13

exhibiting strong reactions to shocks from this market. As expected, compared to the non-crisis periods, more Asian markets are significantly responsive to innovations in the Hong Kong market during the crisis (i.e., Indonesia, Japan, Malaysia, Philippines, Pakistan, Singapore, Thailand, and Taiwan, all of which are victim markets except Japan and Pakistan). Table 4 shows that the Singapore market is an influential market in the Asian region. India, Indonesia, Malaysia, Philippines, Pakistan, and Thailand had strong reactions to shocks from the Singapore market before the crisis, while Hong Kong, India, Indonesia, Korea, Malaysia, and Thailand responded to innovations from this market after the crisis. Like most other victim markets, the Singapore market became more influential market during the crisis, with strong reactions from Hong Kong, India, Japan, Malaysia, Philippines, Pakistan, Thailand and Taiwan. Thus, we can infer that the most integrated markets during the crisis are victim markets. Similar to the case of Korea, few markets, with the exception of Thailand, responded significantly to a shock from Taiwan before the crisis. Such a pattern is also observed after the crisis, with the only exception of Korea responding to the shock from Taiwan. This suggests that Taiwan had little influence on other markets in the non-crisis periods. Consistent with Ghosh, Saidi, and Johnson (1999) and Dekker, Sen, and Young (2001), we find that Taiwan normally is an isolated market. However, it is found that most markets, except non-victim countries such as India, Japan and the U.S., significantly responded to shocks from Taiwan during the crisis period. Examining the responses of the victim markets to shocks from non-victim markets provides further evidence on how the crisis affected linkages among the Asian

14

markets. Table 4 reports the size of responses to a U.S. shock for each of the other countries. In general, all Asian markets significantly responded to U.S. shocks throughout all the three periods. Responses to U.S. shocks among most Asian markets was intensified during the crisis; however, some countries such as Japan, India, and Malaysia experienced no substantial change in response to U.S. shocks across the three different sample periods. Conversely, the U.S. market normally did not respond to Asian markets in the three sample periods, which is consistent with previous literature (e.g., Masih and Masih, 1999; Sheng and Tu, 2000). The role of Japan as the leader in the Asian region has been a contentious issue. While Ghosh, Saidi, and Johnson (1999) as well as Maish and Masih (2001) argue that Japan is a market leader, our results suggest that it did not play a pivotal role in the noncrisis periods. Only India and Pakistan exhibited an appreciable response to the shocks from Japan before the crisis. Also, the Japanese market did not become much more influential after the crisis, as only India, Indonesia, and Korea responded to shocks from this market. Thus, consistent with Masih and Masih (1997a,1999), Japan did not have much influence on other Asian markets in non-crisis periods. Also, Japan showed little or no response to shocks from other markets in such periods. We infer that Japan is a relatively isolated market under normal market conditions, as found in earlier studies by Dekker, Sen, and Young (2001) as well as Bessler and Yang (2003). Once again, during the financial crisis, more Asian markets significantly responded to innovations in the Japanese market, including Hong Kong, Indonesia, Korea, Malaysia, Philippines, Singapore, and Thailand, all of which are victim markets.

15

Both India and Pakistan are small Asian emerging markets and not identified as victim markets. Only two markets (Hong Kong and Taiwan) before the crisis and another two markets (Korea and Taiwan) after the crisis responded to shocks from India. Thus, it is obvious that India has little influence on other Asian markets under normal market conditions. By contrast, during the crisis, eight victim markets (Hong Kong, Indonesia, Korea, Malaysia, Philippines, Pakistan, Singapore, and Thailand) exhibited strong reactions to innovations from India. Lastly, Table 4 reports the Asian markets’ responses to shocks from the Pakistan market. No other markets showed significant responses to shocks from this market in the non-crisis periods, with the exception of Korea after the crisis. However, during the non-crisis periods, Pakistan, as well as India, responded substantially to shocks from quite a few other markets, including the US, Singapore, Hong Kong, Thailand, Indonesia, and Philippine. In this sense, India and Pakistan are not isolated from other markets; on the contrary, they are quite endogenous as their stock movements are to a large extent driven by other market movements during the non-crisis periods while the reverse may not hold. During the crisis, however, four victim markets, including Hong Kong, Indonesia, Singapore, and Thailand, significantly responded to innovations from the Pakistan market. Overall, the crisis caused the victim markets to be more responsive to external shocks from non-victim markets. Thus, an important new finding is that the crisis not only led markets affected by the crisis to be more integrated with each other but also caused them to be more responsive to the outside world.

16

V. SUMMARY AND CONCLUSIONS This study examines the long-run relationship and short-run dynamics among the U.S., Japanese, and ten Asian stock markets, with the particular attention to the 19971998 Asian financial crisis. Extending related empirical studies, comparative analyses of pre-crisis, crisis, and post-crisis periods are conducted to comprehensively evaluate how stock market integration is affected by financial crises. An error correction model (ECM) is employed to estimate long-run relationships between markets, and generalized impulse response functions are utilized to provide insights into short-run causal dynamic linkages among Asian and developed stock markets. In general, the empirical results reveal that long-run cointegration relationships among these markets were strengthened during the crisis and that these markets have been more integrated after the crisis than before the crisis. Our results for the U.S. and Japanese stock markets’ impact on emerging Asian markets agree with the previous studies (e.g., Masih and Masih, 1999; Bessler and Yang, 2003) on the roles of these two markets in the international stock markets. The U.S. substantially influenced the Asian markets in all three sample periods but was almost unaffected by the Asian markets. Conversely, Japan has little or no influence on the Asian markets except during the financial crisis. Further empirical evidence indicates that Japan, Taiwan and Philippines are fairly isolated markets, which is consistent with Dekker, Sen, and Young (2001), among others. Unlike prior studies (e.g., Janakiramanan and Lamba, 1998;Ghosh, Saidi, and Johnson, 1999; Dekker, Sen, and Young, 2001), we find that Indonesia and Thailand are integrated with several other Asian markets during non-crisis periods, rather than being isolated markets. Korea, India and Pakistan appear to be fairly endogenous

17

markets, which has not been documented in the previous literature. Also, Hong Kong is not as interactive in non-crisis periods as reported in prior studies and, therefore, is not as influential as previously believed (e.g., Masih and Maish, 1999; Dekker, Sen, and Young, 2001). In this regard, Singapore appears to be a market leader in the Asian region. An important implication of our findings is that the degree of integration among countries tends to change over time, especially around periods marked by financial crises. As Bekaert and Harvey (1995) have noted, previous research assumes that stock markets are either perfectly integrated, perfectly segmented, or partially integrated but the extent of integration is constant over time. Based on evidence gathered from regime-switching models, they showed that this assumption does not hold. Extending their proposition to the case of the Asian financial crises, we also find that Asian stock market integration can be time variant.

18

REFERENCES Arshanapalli, B., Doukas, J. and Lang, L. (1995) Pre and post-October 1987 stock market linkages between U.S. and Asian markets, Pacific-Basin Finance Journal, 3, 57-73. Bekaert, G. and Harvey, C. R. (1995) Time-varying world market integration, Journal of Finance, 50, 403-444. Bessler, D. A. and Yang, J. (2003) The structure of interdependence in international stock markets, Journal of International Money and Finance, 22, forthcoming. Brocato, J. (1994) Evidence on adjustments in major national stock market linkages over the 1980s, Journal of Business Finance & Accounting, 21, 643-667. Chan, K.C., Gup, B.E. and Pan, M. (1992) An empirical analysis of stock prices in major Asian markets and the United States, Financial Review, 27, 289-307. Cheung, D. (1997) Pacific Rim stock market integration under different federal funds rate regimes, Journal of Business Finance and Accounting, 24, 1343-1351. Chung, P. and Liu, D. (1994) Common stochastic trends in Pacific Rim stock markets, Quarterly Review of Economics and Finance, 34, 241-259. Corsetti, G., Pesenti, P. and Roubini, N. (1999) What caused the Asian currency and financial crisis? Japan and the World Economy, 11, 305-373. Dekker, A., Sen, K. and Young, M. (2001) Equity market in the Asia Pacific region: A comparison of the orthogonalized and generalized VAR approaches, Global Finance Journal, 12, 1-33. DeFusco, R.A., Geppert, J. M. and Tsetsekos, G. P. (1996) Long-run diversification potential in emerging stock markets, Financial Review, 31, 343-363.

19

Ghosh, A., Saidi, R. and Johnson, K. (1999) Who moves the Asia-Pacific stock marketsUS or Japan? Empirical evidence based on the theory of cointegration, Financial Review, 34, 159-170. Hansen, H. and Juselius, K., (1995) CATS in RATS: Cointegration Analysis of Time Series, Evanston, Illinois: ESTIMA. Hakkio, C.S. and Rush, M. (1991) Cointegration: how short is the long-run? Journal of International Money and Finance,10, 571-581. Hassapis, C., Pittis, N. and Prodromidis, K. (1999) Unit roots and Granger causality in the EMS interest rates: the German Dominance Hypothesis revisited, Journal of International Money and Finance, 18, 47-73. Hung, B. and Cheung, Y. (1995) Interdependence of Asian emerging equity markets. Journal of Business Finance and Accounting, 22, 281-288. Janakiramanan, S. and Lamba, A.S. (1998) An empirical examination of linkages between Pacific-Basin stock markets, Journal of International Financial Markets, Institutions and Money, 8, 155-173. Johansen, S. (1991) Estimation and Hypothesis Testing of Cointegration Vectors in Gaussian Vector Autoregressive Models, Econometrica, 59, 1551-1580. Kamin, S.B. (1999) The current international financial crisis: how much is new? Journal of International Money and Finance, 18, 501-514. Koop, G., Pesaran, M.H. and Potter, S.M. (1996) Impulse response analysis in nonlinear multivariate models, Journal of Econometrics, 74, 119-147. Lutkepohl, H. and Reimers, H. (1992) Impulse response analysis of cointegrated systems. Journal of Economic Dynamics and Control, 16, 53-78.

20

Masih, A. M. M. and Masih, R. (1997a) A comparative analysis of the propagation of stock market fluctuations in alternative models of dynamic causal linkages, Applied Financial Economics, 7, 59-74. Masih, A. M. M. and Masih, R. (1997b) Dynamic linkages and the propagation mechanism driving major international stock markets, Quarterly Review of Economics and Finance, 37, 859-885. Masih, A. M. M. and Masih, R. (1999) Are Asian stock market fluctuations due mainly to intra-regional contagion effects? Evidence based on Asian emerging stock markets, Pacific-Basin Finance Journal, 7, 251-282. Masih, R. and Masih, A. M. M. (2001) Long and short term dynamic causal transmission amongst international stock markets, Journal of International Money and Finance, 20, 563-587. Ng, A. (2000) Volatility spillover effects from Japan and the US to the Pacific-Basin, Journal of International Money and Finance, 19, 207-233. Pesaran, M.H. and Shin, Y. (1998) Generalized impulse response analysis in linear multivariate models, Economics Letters, 58, 17-29. Phillips, P. (1998) Impulse response and forecast error variance asymptotics in nonstationary VARs, Journal of Econometrics, 83, 21-56. Sheng, H. and Tu, A. (2000) A study of cointegration and variance decomposition among national equity indices before and during the period of the Asian financial crisis, Journal of Multinational Financial Management, 10, 345-365.

21

Yang, J. and Leatham, D. J. (2001) Currency convertibility and linkage between Chinese official and swap market exchange rates, Contemporary Economic Policy, 19, 347-359.

22

Table 1. Trace tests for cointegration in local currency terms Panel A: Without linear trend Number of cointegrating vectors

r=0

r =1

r=2 r =3 r=4 r =5 r=6 r=7 r =8 r =9

r = 10 r = 11

Pre-crisis period

Crisis Period

Transition Period

Post-crisis period

Critical value (5 percent)

328.12 251.69 196.39 148.92 117.04 93.59 71.11 50.82 34.28 23.09 12.70 6.07

380.04 292.52 220.43 174.62 137.73 104.10 75.70 54.06 33.59 20.81 10.94 3.85

334.69 263.73 211.20 166.42 129.36 95.92 70.04 48.48 32.04 18.60 8.86 2.69

368.47 295.08 226.94 172.22 134.86 105.95 78.58 51.87 31.96 20.06 10.54 5.16

338.09 289.70 244.56 203.34 165.73 132.00 101.83 75.73 53.42 34.79 19.99 9.13

Pre-crisis period

Crisis Period

Transition Period

Post-crisis period

Critical value (5 percent)

305.42 231.72 178.38 131.89 103.17 79.74 58.48 41.54 28.24 17.07 6.86 0.77

363.32 277.21 205.68 160.07 125.36 91.86 64.60 44.13 25.43 13.76 4.51 0.24

326.16 255.36 205.71 160.93 123.99 90.71 64.85 43.35 28.01 14.94 5.44 0.28

361.13 288.02 219.87 165.78 130.11 101.22 73.86 47.17 29.43 18.94 9.42 4.18

323.93 276.36 232.60 192.30 155.74 123.03 93.91 68.86 47.20 29.37 15.34 3.84

Panel B: With linear trend Number of cointegrating vectors

r=0

r =1

r=2 r =3 r=4 r =5 r=6 r=7 r =8 r =9

r = 10 r = 11

23

Table 2. Trace tests for cointegration in U.S. dollar terms Panel A: Without linear trend Number of cointegrating vectors

r=0 r =1

r=2 r =3 r=4 r =5 r=6 r=7 r =8 r =9

r = 10 r = 11

Pre-crisis period 350.36 266.72 212.22 160.43 124.22 95.51 70.57 51.96 35.15 24.69 15.07 7.04

Crisis Period

Transition Period

369.12 295.74 233.02 182.69 140.29 105.97 79.55 57.51 38.36 23.38 11.97 4.25

339.87 269.07 215.01 169.18 129.29 92.09 64.65 46.24 31.70 17.65 8.23 2.72

Post-crisis period 387.76 292.39 234.54 178.64 133.16 101.24 72.69 49.40 29.90 19.50 11.29 4.45

Critical value (5 percent) 338.09 289.70 244.56 203.34 165.73 132.00 101.83 75.73 53.42 34.79 19.99 9.13

Panel B: With linear trend Number of cointegrating vectors

r=0 r =1

r=2 r =3 r=4 r =5 r=6 r=7 r =8 r =9

r = 10 r = 11

Pre-crisis period 325.51 242.60 190.31 144.52 110.73 82.99 61.71 43.99 28.53 18.42 9.43 1.88

Crisis Period

Transition Period

351.88 278.94 216.35 166.65 125.51 91.44 66.14 44.20 27.67 15.18 4.97 0.35

331.45 261.64 208.90 163.09 123.55 86.37 58.97 42.10 27.66 15.14 5.72 0.32

24

Post-crisis period 380.29 284.92 227.32 171.91 127.54 96.43 67.88 45.68 26.60 16.26 8.48 3.16

Critical value (5 percent) 323.93 276.36 232.60 192.30 155.74 123.03 93.91 68.86 47.20 29.37 15.34 3.84

Table 3. Market exclusion tests Likelihood ratio test statistic (p-value) Excluded market

Local currency

U.S. dollars

Period 2

Period 3b

Period 1

Period 2

Period 3a

Period 3b

HK

4.70 (0.094)

16.90 (0.000)*

4.55 (0.032)*

6.71 (0.034)*

0.67 (0.411)

33.51 (0.000)*

ID

16.63 (0.000)*

9.14 (0.001)*

11.80 (0.000)*

2.01 (0.364)

0.57 (0.446)

22.31 (0.000)*

IN

11.27 (0.003)*

9.59 (0.008)*

17.21 (0.000)*

11.43 (0.003)*

3.17 (0.074)

13.16 (0.001)*

JP

12.83 (0.001)*

15.63 (0.000)*

6.77 (0.009)*

7.62 (0.022)*

5.28 (0.021)*

38.48 (0.000)*

KR

29.12 (0.000)*

13.41 (0.001)*

29.50 (0.000)*

2.76 (0.250)

0.22 (0.635)

19.68 (0.000)*

ML

16.62 (0.000)*

10.28 (0.005)*

2.08 (0.148)

13.58 (0.001)*

0.39 (0.528)

14.29 (0.000)*

PH

19.04 (0.000)*

3.05 (0.217)

29.38 (0.000)*

1.07 (0.584)

11.43 (0.000)*

14.02 (0.000)*

PK

15.51 (0.000)*

15.29 (0.000)*

3.64 (0.056)

6.07 (0.047)*

3.23 (0.072)

8.05 (0.017)*

SG

11.03 (0.004)*

12.90 (0.001)*

25.35 (0.000)*

6.01 (0.049)*

1.18 (0.275)

0.37 (0.830)

TH

37.99 (0.000)*

8.03 (0.01)*

21.69 (0.000)*

16.23 (0.000)*

0.87 (0.350)

2.58 (0.274)

TW

11.29 (0.003)*

4.33 (0.114)

0.05 (0.815)

14.15 (0.000)*

5.19 (0.023)*

1.35 (0.507)

US

17.90 (0.000)*

11.30 (0.003)*

0.14 (0.701)

15.17 (0.000)*

12.76 (0.000)*

16.49 (0.000)*

Note: For Period 1 and Period 3a in US dollars terms, the test is performed given one cointegrating vector. For other periods the test is performed given two cointegrating vectors. An asterisk denotes the rejection of the null hypothesis at the 5 percent significance level.

25

Table 4. Impulse responses by country to one standard deviation innovations in each stock market (day 20) Subperiod Indonesia (IN) (a) (b) (c)

Responses by country HK

ID

IN

JP

KR

ML

PH

PK

SG

TH

TW

US

0.27 0.47

0.29 0.20

0.63 0.75

0.14 0.18

0.11 -0.06

0.30 0.68

0.26 0.23

0.50 0.40

0.19 0.46

-0.01 0.32

0.07 0.19

0.28 0.03

0.31

0.18

0.98

0.27

0.48

0.27

0.24

0.20

0.45

0.33

0.15

0.07

0.18 0.08 0.08

-0.14 0.10 0.14

0.19 0.08 0.11

0.48 0.89 0.75

-0.20 0.10 -0.05

-0.23 0.01 0.06

0.05 0.19 -0.27

-0.21 0.07 -0.01

0.06 0.54 0.20

-0.18 0.20 0.24

0.07 0.06 -0.02

0.41 0.04 0.26

0.49 0.48 0.29

0.05 0.07 0.01

0.10 0.05 -0.06

0.42 0.98 1.02

0.49 0.37 0.07

0.77 0.14 0.04

0.10 0.43 -0.11

0.29 0.52 0.10

0.18 0.23 0.10

0.00 0.03 -0.03

0.26 0.30 0.22

0.30 -0.25 0.25

0.08 0.15 0.01

-0.05 -0.21 0.19

0.16 0.24 -0.01

0.57 0.67 1.04

0.29 0.64 -0.01

0.00 0.50 0.00

0.21 0.33 0.23

-0.18 0.32 0.03

0.12 0.04 0.03

0.35 0.21 0.43

0.24 0.31 0.27

-0.03 0.22 0.15

0.30 0.71 0.60

0.08 0.97 0.07

0.24 0.49 0.33

0.56 0.50 0.18

0.11 0.59 0.31

0.59 1.04 0.94

-0.25 0.18 -0.02

-0.08 0.10 0.11

0.09 0.18 0.78

0.15 0.22 0.40

-0.19 0.20 0.18

-0.00 0.02 0.74

0.07 0.52 -0.09

0.02 0.32 0.16

0.41 0.40 0.25

0.03 0.57 0.26

-0.07 0.26 0.16

-0.25 0.25 0.01

0.07 0.02 0.05

0.11 0.85

0.56 0.28

0.61 0.13

0.05 0.21

-0.01 -0.01

0.48 0.80

0.56 0.43

0.70 0.70

0.48 0.97

0.66 0.53

0.06 0.36

-0.01 0.05

0.21

0.66

0.70

0.14

0.46

-0.20

0.06

-0.18

0.67

0.31

0.09

-0.05

0.04 0.38 0.02

0.08 -0.09 0.20

-0.04 0.41 0.00

-0.14 0.16 0.08

-0.08 0.73 0.29

0.06 0.61 0.09

0.04 0.43 0.02

0.04 0.30 0.11

-0.04 0.42 0.02

-0.23 0.55 0.00

0.52 1.00 0.99

-0.04 -0.02 0.02

0.73 1.02 0.84

0.36 0.34 0.36

0.72 0.40 0.26

0.58 0.58 0.56

0.00 0.53 1.05

0.44 0.39 0.40

0.82 0.60 0.45

0.15 0.47 0.14

0.29 0.71 0.62

0.32 1.19 0.54

-0.06 0.63 0.52

0.70 0.89 0.99

0.01 0.38 -0.04

0.15 -0.18 0.66

0.21 0.90 0.26

0.62 0.88 0.87

0.06 0.42 0.51

-0.01 0.40 -0.04

0.14 0.50 -0.01

0.25 -0.17 0.04

0.14 0.41 0.03

0.08 0.67 0.01

0.12 0.15 0.16

0.04 0.00 0.08

Korea (KR) -0.08 (a) 0.08 (b) 0.18 (c) Malaysia (ML) 0.01 (a) 0.27 (b) -0.07 (c) Philippines (PH) 0.11 (a) 0.46 (b) 0.04 (c) Thailand (TH) -0.07 (a) 0.61 (b) 0.17 (c) Hong Kong (HK) 0.18 (a) 0.89 (b) 0.71 (c) Singapore (SG) (a) (b) (c) Taiwan (TW) (a) (b) (c)

United States (US) (a) (b) (c) Japan (JP) (a) (b) (c)

26

Table 4, continued India (ID) (a) (b) (c)

-0.22 0.77 0.19

0.59 1.06 1.06

-0.18 0.72 0.00

-0.07 0.18 0.16

0.11 0.44 0.33

0.03 0.88 0.11

0.11 0.60 0.12

0.02 0.28 0.02

-0.12 0.67 0.14

0.04 0.59 0.10

0.36 0.05 0.27

-0.12 0.12 -0.02

-0.14 0.22 0.12

0.06 0.06 -0.17

-0.10 0.23 0.08

-0.05 0.01 -0.02

-0.07 -0.05 -0.26

-0.13 0.18 0.13

-0.09 0.14 0.02

0.44 1.05 0.77

-0.06 0.22 -0.12

-0.05 0.37 0.09

-0.15 0.15 0.12

-0.09 -0.02 -0.03

Pakistan (PK) (a) (b) (c)

Note: The three sub-periods are denoted as follows: (a) pre-crisis period, (b) crisis period, and (c) postcrisis period. Countries are abbreviated as follows: HK (Hong Kong), ID (India), IN (Indonesia), JP (Japan), KR (Korea), ML (Malaysia), PH (Philippines), PK (Pakistan), SG (Singapore), TH (Thailand), TW (Taiwan), and US (United States).

27

FOOTNOTES 1

See Bessler and Yang (2003) for a summary of such analysis conducted in major

international stock markets. Also, some researchers focus on stock return volatility spillover in Asian stock markets, which is not pursued in this study. See Ng (2000) for an excellent example in this line of research. 2

In a comparative study of the traditional orthogonalized and generalized VAR analysis,

Dekker, Sen, and Young found that the generalized approach provided more reliable results than the traditional orthogonalized approach based on the Choleski factorization. Also, the generalized impulse response analysis instead of the generalized forecast error variance decomposition is used in this study for an important reason. As pointed out by Masih and Masih (1999, p.269), the generalized forecast error variance decomposition can not be strictly used to isolate responses of a particular market, assuming that all other shocks are not present or not running in conjunction with the particular shock in question. In other words, when applying and interpreting generalized forecast error variance decomposition, one should not attribute the shock to the sole variable in the system, as would be appropriate in traditional forecast error variance decomposition. 3

Different from Dekker, Sen, and Young (2001) but similar to Masih and Masih (2001),

this study allows for long-run relationships (if any) when conducting generalized impulse response analysis. In such a case, the impact of a shock on other markets may not be transitory and die away within a few days. Instead, as demonstrated in Masih and Masih (2001), the impact of a shock is most likely to be lasting for a long period after it returns to its long-run level. This study is able to explore the possible lasting effect of a shock by employing a longer horizon of day 20 ahead. Also, Phillips (1998) has recently proved

28

that the imposition of cointegration constraints in the nonstationary VAR analysis is crucial in yielding consistent results on impulse response analysis and forecast error variance decompositions.

29