Interrelationships among regional stock indices

Review of Financial Economics 11 (2002) 91 – 108

Interrelationships among regional stock indices$ Orawan Ratanapakorna, Subhash C. Sharmab,* a

Faculty of Economics, Bangkok 10200, Thailand Department of Economics, Southern Illinois University, Carbondale, IL 62901-4515, USA

b

Received 10 April 2001; received in revised form 13 October 2001; accepted 22 January 2002

Abstract This study investigates the short-run and long-run relationships among stock indices of the US, Europe, Asia, Latin America, and Eastern Europe –Middle East for the pre-Asian crisis and for the crisis period. The findings from these two periods are compared and contrasted. No long-run relationship is observed among these indices during the pre-Asian crisis period. However, during the crisis period, one significant cointegrating vector is observed and more short-run (i.e., causal) relations are observed in this period as compared to the pre-crisis period. Based on the analysis, we infer that during the Asian crisis period, the globalization increased and only the European markets directly effected the US market, while the other regional markets indirectly influenced the US market via the European market. As regards the effect of shocks, we observe that during the pre-Asian crisis period, the response of all regional markets to shocks in other markets is transitory, whereas during the crisis period, the response of the US stock market is transitory but that of EU market is permanent to all other markets. D 2002 Published by Elsevier Science Inc. JEL classification: G15 Keywords: Cointegration; Vector error correction model; Granger causality; US stock markets; European stock markets; Asian stock markets; Latin American stock markets; Eastern Europe – Middle East stock markets

$ We would like to thank the editors for their patience. We would also like to thank the referee for helpful comments and suggestions which has greatly improved this study. * Corresponding author. Tel.: +1-618-453-5070; fax: +1-618-453-2717. E-mail address: [email protected] (S.C. Sharma).

1059-0560/02/$ – see front matter D 2002 Published by Elsevier Science Inc. PII: S 1 0 5 9 - 0 5 6 0 ( 0 2 ) 0 0 1 0 3 - X

92

O. Ratanapakorn, S.C. Sharma / Review of Financial Economics 11 (2002) 91–108

1. Introduction Increased economic integration among economies of the world has brought increased attention of investors and academic scholars to the issue of interrelationships among these markets around the world. While integration in banking and financial markets provides some advantages in terms of gains in market efficiency and portfolio diversification, it also offers potential pitfalls. The October 1987 crash of US financial markets led to the doom and gloom in the financial markets around the world. Moreover, the dramatic decline in the Japanese stock market at the beginning of 1990s, as well as the recent recession in mid-1998, and the recent financial crisis in the emerging markets point to an important pitfall for financial institutions investing globally. Peek and Rosengren (1997) argue that the Japanese stock market decline resulted from a decrease in lending by Japanese banks and was transmitted internationally to the US. The economic downturn in Japan originated during the same quarter that the Thai baht was coming under increased pressure in mid-1997. The financial crisis in Thailand rapidly spread to Indonesia, Malaysia, the Philippines, and Korea. In October 1997, the crisis began to affect other economies when speculative pressures intensified against the Hong Kong dollar, the Korean won, and the Taiwanese dollar resulting in drops in their stock markets. A severe drop in Hong Kong’s financial market dramatically pushed the global equity prices (affecting Eastern Europe, Latin America, Japan, Europe, and the United States) into a downturn. Consequently, in mid-1998, the East Asian crisis became a worldwide financial and economic crisis hitting developing economies in Latin America, Middle East, Eastern Europe, and North Africa. There are several factors that explain the repercussions from financial crises. First, common shocks such as a steep rise in world interest rates, a sharp decline in world aggregate demand, a slowdown in commodity prices, or large changes in exchange rates between major currencies can induce pressure on currencies of several countries simultaneously. Second, a significant currency depreciation in one country experiencing a financial crisis may affect other countries through trade spillovers due to the improved price competitiveness of the crisis country. Third, the occurrence of a crisis in one or more countries may induce investors to rebalance their portfolios for risk management or other reasons. Fourth, a crisis in one country may wake up other financial markets to reassess their countries’ circumstances. Thus, many researchers have investigated the short-term and long-term interrelationships among worldwide financial markets. The primary focus of the empirical research has been the G-7 and other industrialized countries. Lai, Lai, and Fang (1993) observe both the short-run and the long-run feedback relationships between the New York and Japanese stock markets. Kasa (1992) finds a single common stochastic trend among G-7 countries. Among the major European markets, Choudhry (1996), Koutmos (1996), and Serletis and King (1997) all discover long-run relationships. Booth, Martikainen, and Tse (1997) notice a weak relationship among the Scandinavian markets. There are some studies available for the Asian and Pacific markets. Engle and Susmel (1993) test for common volatility in 18 nations’ stock markets and report two closed-boundary groups

O. Ratanapakorn, S.C. Sharma / Review of Financial Economics 11 (2002) 91–108

93

that share similar volatility characteristics. Cheung (1995) observes a long-run relationship among five emerging stock markets: Hong Kong, Korea, Malaysia, Singapore, and Thailand. Furthermore, Corhay, Rad, and Urbain (1995) find that in the long run, there exists a geographical separation between the Asia and the Pacific Markets. Recently, Sharma and Wongbangpo (2002) investigate the long-term trends and cycles of stock markets in Indonesia, Malaysia, Singapore, and Thailand. They observed that the stock markets of Indonesia and Thailand are cycle dominated, and those of Malaysia and Singapore are trend dominated. However, a few studies have investigated the relationships among markets across regions. Masih and Masih (1997) conclude that the markets of Japan, US, UK, and Germany drive the fluctuations in the markets of Taiwan, South Korea, Singapore, and Hong Kong. Kwan, Sim, and Cotsomitis (1995) note that the markets of Hong Kong, Singapore, Korea, and Taiwan are not cointegrated among themselves but they are cointegrated with G-7 countries. Furthermore, Cheung, He, and Ng (1997) investigate the existence of common movement and interaction among the Pacific Rim return, European return, and the North American return from 1970 to 1991. They observe that the North American markets have the most predictive power for the European and Pacific Rim’s stock-market movements. The European and Pacific Rim markets have weak ability to influence other regional stock market movements. Specifically, the Pacific Rim markets can explain only about 3% of the North American return variability. Moreover, the European markets significantly affected Pacific Rim stock market movements in the 1980s. The only study that has investigated the effect of local and global events on the international stock markets is that of Aggarwal, Inclan, and Leal (1999). For the period 1985–1995, Aggarwal et al. (1999) observe that global events such as the October 1987 crash and the Gulf War had an impact on many countries. In addition, they also note that local events (the Mexican peso crisis, periods of hyperinflation in Latin America, the stock market in India, and the Marcos–Aquino conflict in the Philippines) in a specific country, which affect the volatility of that country’s index, are also reflected in the high volatility of its regional index. For instance, the Mexican peso crisis caused large shifts in the volatility of the Mexican stock market and, hence, the Latin American index. Fama (1981) noted that stock returns are positively related to real variables such as rate of return on capital and output and many studies (e.g., Dhakal, Kandil, & Sharma, 1993; Gallinger, 1994; Mahdavi & Sohrabian, 1991; Wongbangpo & Sharma, 2002, among others) have supported empirically that stock prices lead economic activities. We believe that the stock index of a country is an indicator of the economic activities of the country and the regional stock index is an indicator of the economic activities of the region. Thus, the objectives of this study are twofold. Since all the studies cited above have investigated the interrelationships among financial markets of different countries in one region, or, the markets across regions, the first objective is to investigate the short-run and long-run relationships among the regional stock market indices, that is, between the indices of US, Europe, Asia, Latin America, and Eastern Europe–Middle East. This is done for the preAsian crisis period and during the Asian crisis. Constrained by the availability of the data, for the pre-Asian crisis period, we have considered the time period from Jan. 1, 1990 to

94

O. Ratanapakorn, S.C. Sharma / Review of Financial Economics 11 (2002) 91–108

Dec. 31, 1996. The Asian crisis period is considered from July 2, 1997 to March 10, 2000. It would be of interest to compare and contrast the findings of these two time periods to learn if any meaningful changes occurred during the crisis period.1 Towards this goal, cointegration analysis, Granger causality tests, and innovation accounting analysis are performed. Cointegration and causal analyses will shed light on the long-term and shortterm relationships and impulse response analysis will reveal the speed of transmission of shocks. Since we do not have any theory to specify the structural econometric model, in such situations, it is more appropriate to use time series techniques to investigate the longterm and short-term relations among regional stock indices. Note that in time series analysis, no a priori model is specified, and the inference is made based on analysis of the series. Cointegration, causal, and innovation analyses are the appropriate time series techniques to investigate the long-run and short-term relationships. The second objective of this study is to examine the impact of regional and global crises on the US economy. The analyses of stock prices grouped into regional indices will shed some light on the impact of regional and global crises on the US economy, since the regional stock index is an indicator of the economic activities of the region. A significant long-run relationship among different regional stock indices could be explained due to one or more of the following reasons. Strong economic ties, policy coordination, and trade among the relevant regions may indirectly link their stock indices (Cheung & Lai, 1999; Choudhry, 1996; Ripley, 1973). Economic conditions resulting in lower interest rates and decreased inflation or reflecting the world’s general financial condition may, also, lead to co-movement of different stock indices. The real interest rate linkage among regions can facilitate long-run relationships between different stock indices due to its important role in stock markets and due to international capital flows. The growing importance of international investors, the substantial improvements in communication technology, the innovations in financial products and services, and deregulation and liberalization, as well as, the increase in the activities of multinational corporations can further induce long-run relationships among regional stock indices. It is argued that comovement of regional stock indices may increase after some turmoil in the markets owing to (local and global) contagion effects (Chan, Gup, & Pan, 1997). As mentioned above, there could be many reasons to induce interrelationships among regional stock indices; it would be of interest to investors, policy makers, and academicians to know whether the data reveal any such long-run and short-run relationships. The rest of the study is organized as follows. The model and data are given in Section 2. Long-run relationships and short-run relationships are discussed in Sections 3 and 4, respectively. The findings of innovation analyses are presented in Section 5, and Section 6 concludes this study.

1

In an earlier version, only the Asian crisis period was considered. We are thankful to the referee for suggesting to include the pre-Asian crisis period and compare and contrast the results of two periods. This has greatly improved this study.

O. Ratanapakorn, S.C. Sharma / Review of Financial Economics 11 (2002) 91–108

95

2. Model and data To investigate relationships among stock indices across geographical regions, the following model is analyzed, that is, W ¼ ðUS, EU, AP, LA, EMÞ0

ð1Þ

where US is the S&P 500 index, EU is the European index, AP is the Asian–Pacific index, LA is the Latin American index, and EM is the Eastern European–Middle East index. The European index includes 14 markets including Austria, Belgium, Denmark, Finland, France, Germany, Ireland, Italy, Netherlands, Norway, Portugal, Spain, Sweden, and the UK. The Asian–Pacific index includes the following 12 countries: Australia, China, Hong Kong, Indonesia, Japan, Korea, Malaysia, New Zealand, Philippines, Singapore, Taiwan, and Thailand which were affected by the recent Asian crisis. The Latin American index includes seven countries: Argentina, Brazil, Chile, Colombia, Mexico, Peru, and Venezuela. The Eastern European–Middle East index consists of eight markets: Czech Republic, Greece, Hungary, Israel, Jordan, Poland, Russia, and Turkey. The latter grouping is used to explore the Russian crisis. Based on availability, daily data from January 1, 1990 to March 30, 2000 are obtained for all regional indices. The S&P 500 is obtained from Standard and Poor’s, a division of McGraw-Hill, while data for other indices are obtained from Morgan Stanley Capital International. All data are in US dollars and market capitalization weighted. The base year for all indices is March 31, 1995 = 100. To compare and contrast the effects of the Asian crisis and pre-Asian crisis periods, the data set is divided into two time periods. Time Period I: from January 1, 1990 to December 31, 1996, that is, the pre-Asian crisis period; and Time Period II: from July 2, 1997 to March 10, 2000, that is, the Asian crisis period. Time period I covers certain important events, such as the Gulf War, the Japanese stock market decline, and the Mexican peso crisis. The starting date of time period II is taken to coincide with the announcement of the Thai government’s intentions to let the baht float which led to financial crises in emerging markets. The second period also covers the downturn in the Japanese economy in the mid-1997, the Brazilian crises, and the repercussions from the Asian crisis on other regions.2 All data are transformed to natural logarithms prior to analysis.

3. Long-run relationships First, the order of integration in each series is tested, using Dickey–Fuller, Augmented Dickey–Fuller (ADF) (Said & Dickey, 1984) and Phillips and Perron (PP) (Perron, 1988; Phillips & Perron, 1988) unit root tests. To save space, only selected test statistics for time periods I and II are reported in Table 1. The unit root test statistics reveal that each series is 2

We are aware of the fact that for cointegration analysis, one should have at least 30 years of data. One of the prerequisites for Granger causality is to test for cointegration among variables. In time period II, even though we have enough number of effective observations, the cointegration results should be viewed with caution.

96

O. Ratanapakorn, S.C. Sharma / Review of Financial Economics 11 (2002) 91–108

Table 1 Unit root test statistics Phillips – Perron Variable

ADF(tt)

Z(a ˜)

Z(ta˜ )

Time period I: pre-Asian crisis period (1/1/1990 – 12/31/1996) Log levels: US  2.54 2.34  2.26 EU  2.74 2.59  2.50 AP  3.29 3.27  3.17 LA  1.33 1.39  1.01 EM  1.46 1.40  1.13 Log first differences: DUS  40.24 * 2.22  41.49 * DEU  39.55 * 1.34  41.12 * DAP  38.91 * 0.38  40.01 * DLA  35.08 * 1.68  35.12 * DEM  35.77 *  0.08  36.14 * Time period II: Asian crisis period and thereafter (7/2/1997 – 3/10/2000) Log levels: US  3.63 *  24.53  3.50 EU  2.29  11.03  2.31 AP  2.15  4.66  2.15 LA  1.08  2.98  0.98 EM  0.74  2.22  0.76 Log first differences: DUS  27.34 *  679.63 *  27.46 * DEU  19.72 *  549.42 *  23.48 * DAP  24.07 *  559.57 *  23.97 * DLA  23.44 *  599.72 *  23.37 * DEM  23.51 *  639.57 *  23.60 *

Z(F3)

Z(F2)

3.58 4.03 5.96 1.42 1.11

4.30 3.38 4.03 1.95 0.74

805.75 * 775.64 * 751.61 * 614.84 * 636.05 *

435.17 * 517.10 * 501.07 * 409.89 * 424.03 *

6.15 2.68 5.38 1.88 1.44

4.66 2.66 3.58 1.25 1.28

376.94 * 275.67 * 287.25 * 272.99 * 278.55 *

251.29 * 183.78 * 191.50 * 181.99 * 185.70 *

For the augmented Dickey – Fuller and Phillips – Perron tests, a model with a nonzero mean and a linear trend is used. For the details of test Z(a ˜ ), Z(ta˜ ), Z(F3), and Z(F2), see Perron (1988, pp. 302 – 3 and 308 – 9).The optimal lag length for the ADF regression is selected based on the Bayesian information model selection criterion. The truncation lag parameter p in the PP tests is selected using a Newey – West method. US is S&P 500 index, EU is European index, AP is Asian – Pacific index, LA is Latin American index, and EM is Eastern Europe – Middle East index. * Indicates rejection of the corresponding null hypothesis at the 5% level of significance.

nonstationary in log levels but stationary in log first differences. Thus, we note that all regional index series are integrated of order one, I(1), in both time periods. Since the series are integrated of order one, the number of significant cointegrating vectors is tested by using the maximum likelihood-based l-max and l-trace statistics introduced by Johansen (1988, 1991) and Johansen and Juselius (1990). In a set of m series, if there are r cointegrating vectors, then there are (m  r) common stochastic trends. Prior to testing for the number of significant vectors, the likelihood ratio (LR) tests are performed to determine the lag length of the vector autoregressive system. The LR tests yield lag lengths of 12 and 8 for

O. Ratanapakorn, S.C. Sharma / Review of Financial Economics 11 (2002) 91–108

97

Table 2 Tests for the number of cointegrating vectors Hypotheses

Hypotheses

Critical values (95%)

l-Max

H0

H1

l-Trace

l-Max

l-Trace

Time period I: 1/1/1990 – 12/31/1996 .0191 r=0 r=1 .0076 r=1 r=2 .0047 r=2 r=3 .0024 r=3 r=4 .0002 r=4 r=5

34.98 * 13.85 8.60 4.35 0.44

r=0 r1 r2 r3 r4

r>0 r>1 r>2 r>3 r=4

62.22 27.25 13.40 4.79 0.44

34.40 28.14 22.00 15.67 9.24

76.07 53.12 34.91 19.96 9.24

Time period II: 7/2/1997 – 3/10/2000 .0544 r=0 r=1 .0351 r=1 r=2 .0270 r=2 r=3 .0074 r=3 r=4 .0024 r=4 r=5

38.79 * 24.81 18.97 5.16 1.69

r=0 r1 r2 r3 r4

r>0 r>1 r>2 r>3 r=4

89.41 * 50.63 25.81 6.84 1.69

Eigenvalues

H0

H1

The critical values for l-max and l-trace statistics are from Osterwald-Lenum (1992). Same critical values are used in both periods. * Indicates rejection of the null hypothesis at the 5% level of significance.

time periods I and II, respectively. The l-max and l-trace statistics are reported in Table 2. Since the l-trace statistic takes into account all (m  r) of the smallest eigenvalues, it tends to have more power than the l-max statistics (Kasa, 1992; Serletis & King, 1997). Moreover, Johansen and Juselius (1990) also emphasize the use of the l-trace statistics in cases where a conflict between these two test statistics occurs. We note that for time period I, at the 5% level of significance, no significant cointegrating vector is indicated by l-trace statistics while l-max statistics indicate only one cointegrating vector. Based on l-trace statistics, we conclude that there is no significant cointegrating vector in time period I. In other words, during the pre-Asian crisis period considered here, the five regional stock indices do not have any longrun relationships, that is, they move far apart in the long run. For the time period II, based on both the l-trace and l-max statistics, there is one significant cointegrating vector at the 5% level of significance. The normalized cointegrating vector and the residual analysis statistics for time period II are reported in Table 3. Table 3 indicates that the US stock prices during the Asian crisis are positively related to European, Asian, and Eastern Europe–Middle East indices but negative long-run relationships are found between US and Latin American indices. Next, we test the significance of each variable in the cointegrating relation by using the LR test statistics given by Johansen (1991), which is asymptotically chi-square with one degree of freedom. Based on this statistic, all variables are statistically significant at the 1% level of significance and thus contribute to the long-run relationship. Furthermore, the LM(1) and LM(4) statistics in Table 3 indicate that there is no autocorrelation in the residuals at the 5% level of significance, and thus the adequacy of the model during time period II is confirmed. The evidence of cointegration has several implications. First, the market efficiency hypothesis is violated in a multivariate context since the variables share common stochastic trend(s), that is, variables do

98

O. Ratanapakorn, S.C. Sharma / Review of Financial Economics 11 (2002) 91–108

Table 3 Normalized cointegrating vector, test for exclusion, and residual analysis for time period II The first normalized eigenvector US = 1.21 + 0.69EU + 0.23AP  0.19LA + 0.12EM Test of exclusion of each variable Variable

US

EU

AP

LA

EM

2 c(1)

20.24 *

20.77 *

24.27 *

7.49 *

8.27 *

Residual analysis: autocorrelation Test-statistics 2 c(25)

LM(1) 18.19 (.83)

LM(4) 23.03 (.58)

The P values corresponding to LM(1) and LM(4) statistics are reported in parentheses. * Indicates significance at the 5% level of significance.

not drift too far apart from each other since the Asian crisis. Second, it guarantees some significant Granger causality in the system (Granger, 1988). Third, there is a common force such as arbitrage activity that brings the regional stock markets together in the long run. Fourth, international portfolio diversification is less effective across regions because the investment risk cannot be reduced. Thus, we observe that during the pre-Asian crisis period, the five regional composite stock indices had no long-run relationship, that is, did not move together in the long run. However, during the crisis period, a long-run relationship is observed among these indices, that is, they shared common stochastic trends and thus the regional market efficiency hypothesis is also violated during this period.

4. Short-run relationships Granger (1969) introduced a testable definition of causality in terms of predictability in a set of non-cointegrated variables. This definition of causality was extended to a set of cointegrated variables by Granger (1988). Granger (1988) suggested that in a set of cointegrated variables, the short-term causal relations among these variables should be examined within the framework of the error correction model (VECM). Let Wt = (USt EUt APt LAt EMt)0 denote a five-component vector. A five variable VECM (with a deterministic term) can be written as DWt ¼ F0 þ FðLÞDWt1 þ dECt1 þ et

ð2Þ

where F0=(f10 f20 f30 f40 f50)0 is a constant term and F(L)=(fij(L)) is a 5  5 polynomial p P fij,‘ L‘ , where p is the degree of the matrix of coefficients to be estimated, that is, fij ðLÞ ¼ ‘¼1 polynomial, ECt  1 is the vector of error correction term which represents the deviations from

O. Ratanapakorn, S.C. Sharma / Review of Financial Economics 11 (2002) 91–108

99

long-run relationships and et is a vector of error term with E(et) = 0 and E(etes0) = V, for t = s and zero otherwise. For the US stock index, Eq. (2) can be written as DUSt ¼ f10 þ

p X

f11,‘ DUSt‘ þ

‘¼1

þ

p X ‘¼1

f14,‘ DLAt‘ þ

p X ‘¼1

p X

f12,‘ DEUt‘ þ

p X

f13,‘ DAPt‘

‘¼1

f15,‘ DEMt‘ þ dECt1 þ et

ð3Þ

‘¼1

where ECt ¼ USt  1:21  0:69EUt  0:23APt þ 0:19LAt  0:12EMt

ð4Þ

Note that ECt  1 is the vector of error correction terms which represents the deviation from the long-run relationships at time t and d represents the response of the dependent variable to departures from equilibrium. Granger (1988) points out that in a VECM, Eq. (3), there are two channels of causality: one through the lagged values of DEU, DAP, DLA, and DEM and the other through the error correction term, ECt  1. For example, European index, EU, Granger causes US stock index if either f12,‘’s are jointly significant (i.e., H0: f12,1 = f12,2 = . . .f12,p = 0 is rejected) or the error correction term, d, is significant. The joint significance of the lags of each variable is tested by the F-statistics3 and the coefficient of the lagged error correction term is tested by the t-statistics. Similarly, AP causes US stock index if f12,‘’s are jointly significant or if d is significant. In period II, where five stock indices are cointegrated, Eq. (3) is used to detect Granger causality. Moreover, the VECM can indicate econometric exogeneity of the variables if both the F-statistics and t-statistics are insignificant. Since there is only one significant cointegrating vector, the error correction term is obtained by using this vector (Eq. (4)). However, in time period I, Granger causality is detected by using Eq. (3) without the ECt  1 term, since the stock indices are not cointegrated. Thus, in a non-cointegrated system, there is only one channel of causality, that is, through the lagged values of EU, AP, LA, and EM. The causality test statistics, that is, F- and t-statistics for both time periods, are reported in Table 4, and the causal relations are summarized qualitatively in Table 5. From Table 5, we note that during the pre-Asian crisis, none of the regional indices caused the US and LA indices. This implies that local events such as the Gulf War, the Japanese stock market decline, and the Mexican peso crisis did not affect the US stock market during the pre-Asian crisis. On the other hand, the US index caused all other indices except the Latin American index. However, the US index caused AP and EM indices also indirectly via EU. Furthermore, the EU index caused the AP and EM indices, and the LA index also caused 3

Note that whether we estimate one equation in model (3) by ordinary least square (OLS) method, or as a system of all five equations together by seemingly unrelated procedure, the estimates are equivalent. Since in a seemingly unrelated regression (SUR) model, SUR estimates are equivalent to OLS estimates if all the variables on the right hand side are identical.

100

O. Ratanapakorn, S.C. Sharma / Review of Financial Economics 11 (2002) 91–108

Table 4 Granger causality tests F-Statistics Dep Var

DUS

Period I: 1/1/1990 – 12/31/1996: DUS – DEU 11.88 ** DAP 7.60 ** DLA 1.01 DEM 6.33 **

DEU VAR model 0.58 – 2.43 ** 1.14 3.63 **

Period II: 7/2/1997 – 3/10/2000: VECM model DUS – 2.52 * DEU 1.45 – DAP 1.58 4.14 ** DLA 1.13 0.73 DEM 0.37 2.18 *

DAP 1.16 1.01 – 1.46 1.25

0.32 4.24 ** – 2.65 ** 2.93 **

DLA 0.90 0.89 1.46 – 1.99 *

1.02 9.03 ** 5.31 ** – 13.42 **

DEM

t-Stat ECTt  1

1.53 1.24 1.45 1.37 –

– – – – –

0.28 3.30 ** 1.64 0.41 –

 0.49 3.86 ** 5.13 ** 2.68 ** 1.29

* Indicates significance at the 5% level of significance.

** Indicates significance at the 1% level of significance.

the EM index. It is worth noting that the AP and EM indices did not cause any of the regional indices. During time period II, only the European index caused the US index. In fact, feedback is observed between the European and the US index. One possible explanation is that the US stock market reacted to the launch of the Euro dollar on January 1, 1999. In the long run, the Euro currency promises gains in economic efficiency but there is no guarantee that the conversion will go smoothly. The other explanation could be that during the Asian crisis, the Asian/Pacific markets affected the European markets and caused US investors to panic for fear that the Asian financial crisis would spread to the US. All other regional indices, that is, AP, LA, and EM, did not cause the US index. However, the US index caused the AP Table 5 Causal relationships Direct causality

Indirect causality

Period I: pre-Asian crisis, that is, 1/1/1990 – 12/31/1996 US ) EU, US ) AP, US ) EM EU ) AP, EU ) EM, LA ) EM

US ! AP via EU US ! EM via EU

Period II: crisis period and thereafter, that is, 7/2/1997 – 3/10/2000 US () EU, US ) AP, US ) LA EU () AP, EU () LA, EU () EM AP () LA, AP () EM, LA () EM

US ! EM via EU, AP, and LA AP ! US via EM, LA, and AP LA ! US via EM, AP, and EU EM ! US via LA, AP, and EU

The causality is based on a VAR model with 12 lags for time period I and the VECM with 8 lags for time period II. X ) Y indicates X causes Y. X () Y indicates bidirectional causalities, that is, a feedback relationship and X ! Y indicates indirect causality.

O. Ratanapakorn, S.C. Sharma / Review of Financial Economics 11 (2002) 91–108

101

and LA indices but not the EM index. The direct causal relation from the US market to the markets in Europe, Asia, and Latin America can be explained due to close trade relations among these regions. The European index, EU (besides the US index) also had feedback relations with all other regional indices. Furthermore, the Asian index had feedback relations with the LA and EM indices and the LA index had feedback relation with the EM index. Basically, during the Asian crisis, all regional stock indices affected the stock markets in Europe, Asia, Latin America, and Eastern Europe–Middle East. The feedback relationships between the European index and all other indices can be explained by the close trade relations of European nations with countries in other regions. This feedback relationship may also be explained by the branching out of the European companies in other regions of the world. A significant portion of the income of these companies may be dependent on economic outcomes in countries in other regions. The Asian crisis, Brazilian crisis, and Russian crisis, therefore, individually transmitted their repercussions to each region and Europe. The indirect causal relationships in Table 5 indicate that the regional stock markets in Asia, Latin America, and Eastern Europe–Middle East indirectly influenced the US stock market by transmitting the information through the European market during the Asian crisis. In summary, the effects of the Asian crisis spread to either Latin America or Eastern Europe–Middle East and was then transmitted to Europe and finally to the US. Similar causal directions for Russian and Brazilian crises could be found from Table 5. Thus, our analyses indicate stronger interactions between regional markets during the Asian crisis than during the pre-Asian crisis. Since the US index has a feedback relationship only with the EU, whereas the EU has feedback relationships with AP, LA, and EM, this also somewhat supports the view that the European economies are more globalized than the US economy.

5. Innovation accounting analysis For the innovation accounting analysis, a VAR model is used in period I and VECM is used in period II where a significant cointegrating vector is used in the analysis. Orthogonalization is achieved by using Choleski decomposition where the order of variables corresponds to the order in Eq. (1): US index, European index, Asian index, Latin American index, and Eastern European–Middle East index. 5.1. Variance decomposition The causal findings in Table 5 give only qualitative relations. However, the decomposition of variance gives a quantitative measure to these causal relations indicating how much the movement in one market can be explained by other markets in terms of the percentage of the forecast error variance of that market. Table 6 summarizes the variance decomposition findings of 1-day, 6-day, 12-day, and 24-day ahead forecasts in each regional market for both time periods. Our discussion focuses on the 24-day ahead forecast results. During the preAsian crisis period, the results show that the US and LA stock markets are exogenous with almost 97% and 88%, respectively, of its own variance explained by its own shock even after

102

O. Ratanapakorn, S.C. Sharma / Review of Financial Economics 11 (2002) 91–108

Table 6 Forecast of error variance of regional stock markets Market explained

Percentage of forecast error variance by innovations in: DEU

DAP

DLA

DEM

0.00 0.26 0.31 0.65 91.82 83.54 82.38 82.12 13.18 13.13 13.10 13.08 1.21 1.42 1.68 1.88 9.77 10.95 11.12 11.13

0.00 0.38 0.65 0.85 0.00 0.22 0.64 0.72 84.97 78.36 76.97 76.69 0.00 0.50 0.86 0.88 0.67 0.99 1.07 1.15

0.00 0.19 0.41 0.57 0.00 0.30 0.54 0.63 0.09 0.28 1.04 1.22 90.90 88.86 88.04 87.70 0.01 0.17 0.41 0.46

0.00 0.21 0.63 0.98 0.00 0.32 0.86 0.92 0.00 0.38 0.91 0.94 0.00 0.45 0.72 0.75 89.26 85.98 84.80 84.47

Period II: crisis period and thereafter (7/2/1997 – 3/10/2000) DUS 1 100.00 0.00 6 96.77 2.33 12 92.37 6.49 24 92.53 5.70 DEU 1 0.43 99.57 6 1.89 91.55 12 1.48 89.92 24 5.55 83.80 DAP 1 0.05 7.48 6 0.82 22.88 12 0.72 16.25 24 7.49 9.63 DLA 1 0.72 15.43 6 1.57 14.14 12 1.05 11.27 24 3.06 6.97 DEM 1 0.30 25.17 6 2.71 37.09 12 2.44 36.81 24 4.52 31.72

0.00 0.04 0.35 0.47 0.00 2.38 2.81 3.64 92.47 73.17 74.99 73.51 0.35 0.32 0.79 0.90 2.78 0.67 0.45 0.41

0.00 0.68 0.52 0.48 0.00 3.78 2.71 2.20 0.00 2.66 1.88 1.10 83.50 83.85 86.81 89.03 1.27 15.82 21.41 26.04

0.00 0.18 0.28 0.82 0.00 0.41 3.07 4.82 0.00 0.47 6.16 8.27 0.00 0.11 0.08 0.05 70.47 43.71 38.89 37.31

Days

DUS

Period I: pre-Asian crisis (1/1/1990 – 12/31/1996) DUS 1 100.00 6 98.96 12 98.00 24 96.94 DEU 1 8.18 6 15.62 12 15.56 24 15.60 DAP 1 1.76 6 7.85 12 7.98 24 8.07 DLA 1 7.89 6 8.76 12 8.71 24 8.79 DEM 1 0.29 6 1.92 12 2.60 24 2.80

O. Ratanapakorn, S.C. Sharma / Review of Financial Economics 11 (2002) 91–108

103

24 days. These findings are consistent with the causality results. This implies that the Gulf War, the Japanese stock market decline, and the Mexican peso crisis did not affect the US and the Latin American markets during the pre-Asian crisis. The variance decomposition for period I also supports the other causal findings of this period. For example, the US stock market tends to have the most consistent effect on the European market because 15.60% of its own variance is explained by movements in the US stock prices, and the European market can explain more than 11% of the Asian–Pacific and Eastern Europe–Middle East indices. For period II, the results indicate that the US stock market is more self-dependent (or exogenous) with almost 93% of its own variance explained by its own shocks even after 24 days. The US stock market is the most influential in the world indicating that no regional market can explain more than 6% of the US error variance. It also implies that the Asian, Russian, and Brazilian crises did not have an impact on the US economy since each of these regional indices explains less than 1% of the US error variance. However, there is evidence that the European markets caused the US market. Almost 6.5% of the variance in the US market is explained by European markets. In contrast, shocks in the US stock market are ineffectively transmitted to the other regional markets. They can explain from 3.06% for the Latin American markets to 7.49% for the Asian markets. Note that innovations in European markets explain the most forecast variance in other markets. The forecast error variance explained by the European index in other markets ranges from 5.70% for the US to 31.72% for the Eastern Europe–Middle East markets. This implies that a European shock has the strongest impact on each regional economy, particularly on the Eastern Europe–Middle East markets. However, the US stock market tends to have the most consistent effect on the European markets because 5.55% of its own variance is explained by movements in US stock prices. The movements in Asian markets do not tend to have any significant effect on any other regional markets. Since the Asian crisis, Eastern Europe–Middle East is the most interactive market as it accounts for a relative large part of the movements in foreign markets. This implies that Eastern Europe–Middle East markets are obviously weak if shocks from other regions occur because they are more dependent on other regions. Note that only European and Latin American innovations are strong. Interestingly, movements in foreign markets explain 11% of the forecast error variances for the Latin American markets. This regional economy is more dependent on its own shocks. 5.2. Impulse response function Next, to shed some light on the duration of the effect of shock in one regional stock index to other regional indices, the impulse response functions (IRF) are analyzed. The IRF are obtained from the VAR model for time period I and from VECM model for time period II. If Notes to Table 6: The VDC is sensitive to the ordering of the variables. Here, the ordering of the variables is based on the correlation between the stock prices and regional stock markets, that is, US, EU, LA, AP, and EM, respectively, for the time period I; and US, EU, AP, LA, and EM, respectively, for the time period II.

104

O. Ratanapakorn, S.C. Sharma / Review of Financial Economics 11 (2002) 91–108

the effect of a shock in one of the variables does not die out in the long run (even if no further shocks occur), that is, it shifts the system to a new equilibrium, it is called the permanent effect. On the other hand, if the system returns to its previous equilibrium value after some time, it is called the transitory effect. For time period I, the impulse response function of each market is reported in Fig. 1. This IRF is due to one standard deviation shock in each of the five regional markets for a horizon of 36 days. The ordering of the variables is the same as in the VDC. In general, the US and the European markets do not respond to a one standard deviation shock in the European, Eastern Europe–Middle East, Latin American, and Asian–Pacific markets. This implies that the US and the European markets are more stable than the others. On the other hand, the Latin American and the Asian–Pacific markets quickly react to a one standard deviation innovation not only in the US market but also in European markets. One possible explanation is that the

Fig. 1. Impulse response functions for time period I: from VAR model.

O. Ratanapakorn, S.C. Sharma / Review of Financial Economics 11 (2002) 91–108

105

Latin American and Asian–Pacific countries have closed trade relationships with the US and Europe during the pre-crisis period. The Eastern Europe–Middle East stock market suddenly responds to the European market as a result of a closed trade relationship between the two. Importantly, the effect of shocks in other markets on all regional markets is transitory, that is, an individual stock market moves to its original equilibrium within 10–15 days after being affected by the shocks in the other markets. For time period II, Fig. 2 represents the impulse response of each stock market to one standard deviation shock in each of the five regional markets for a horizon of 100 days. The ordering of the variables is the same as in the VDC. In general, the response of the US market to a shock in the European and Eastern Europe–Middle East markets is rapid, but not so rapid for the Asian and Latin American markets. Note that the response of the US market is transitory to all regional markets. A shock in Asian–Pacific, Eastern Europe–

Fig. 2. Impulse response functions for time period II: from VECM model.

106

O. Ratanapakorn, S.C. Sharma / Review of Financial Economics 11 (2002) 91–108

Middle East, Latin American, and European markets moves the US market away from its original equilibrium for a short period of time. On the other hand, a shock in European market is the most effective because it takes the longest time span to move back to its original equilibrium. The shocks in US, AP, LA, and EM markets affect the EU market permanently as they shift the EU market to a new equilibrium. However, the short-term effect of shocks in US, AP, LA, and EM markets varies (in terms of number of days) on the EU market. The AP market also responds to shocks in other markets and in the long run it moves to a new equilibrium. The LA market does not rapidly react to innovations in other markets. However, it is permanently influenced by markets in the US and EU and to some extent by the AP market. The EM market permanently responds to all other markets. Moreover, the EM market quickly reacts to all other markets from Day 1. Thus, the EM market is not stable to external shocks. It is interesting to note that the AP, LA, and EM markets move in the opposite direction to a one standard deviation shock in the US or EU market.

6. Concluding remarks In this study, short-term and long-term relationships are investigated in five regional stock indices for the pre-Asian crisis (January 1, 1990 to December 31, 1996) and Asian crisis (July 2, 1997 to March 10, 2000) periods. No long-term relationship is observed before the preAsian crisis. However, one significant cointegrating vector is observed during the crisis period and each market contributed significantly (in a statistical sense) to the long-run relationship. It can be interpreted that the degree of globalization increased during and after the crisis period. The short-term relations are analyzed through Granger causality and innovation analyses. Note that the local events such as the Gulf War, the Japanese stock market decline, and the Mexican peso crisis did not effect the US market in a causal sense in the pre-Asian crisis period. During the crisis, only the European market Granger causes the US market, while the other regional markets indirectly cause the US market via the European market. The direction of causality for the whole system implies that the Asian crisis spread to either Latin American or Eastern Europe-Middle East markets, then to Europe, and finally to the US market. It can be inferred that the outlook for the US economy can be hinted by examining European markets. The innovation analysis supports the causality findings. For the variance decomposition analysis, it can be concluded that local events and financial crises in recent years do not have enough power to influence the US stock market due to the robust US economy. According to the impulse response functions, local events such as the Gulf War, the Japanese stock market decline, and the Mexican peso crisis do not have powerful impacts on other regional markets, that is, the effects will die out within 10 days. The Asian crisis temporarily influenced the US stock market with a time span of 6 days. The repercussions of the Asian crisis to markets in EU, LA, and EM, however, are long lasting. This implies that the US stock market is the most influential among regional markets.

O. Ratanapakorn, S.C. Sharma / Review of Financial Economics 11 (2002) 91–108

107

References Aggarwal, R., Inclan, C., & Leal, R. (1999). Volatility in emerging stock markets. Journal of Financial and Quantitative Analysis, 34 (1), 33 – 55. Booth, G. G., Martikainen, T., & Tse, Y. (1997). Price and volatility spillovers in Scandinavian stock markets. Journal of Banking and Finance, 21, 811 – 823. Chan, K. C., Gup, B. E., & Pan, M.-S. (1997). International stock market efficiency and integration: a study of eighteen nations. Journal of Business and Accounting, 24 (6), 803 – 813. Cheung, A. C. S. (1995). The UK stock market and economic factors: a new approach. Journal of Business Finance and Accounting, 22 (1), 129 – 142. Cheung, Y.-W., He, J., & Ng, L. K. (1997). Common predictable components in regional stock markets. Journal of Business and Economic Statistics, 15 (1), 35 – 42. Cheung, Y.-W., & Lai, K. S. (1999). Macroeconomic determinants of long-term stock market comovements among major EMS countries. Applied Financial Economics, 9, 73 – 85. Choudhry, T. (1996). Interdependence of stock markets: evidence from Europe during the 1920s and the 1930s. Applied Financial Economics, 6, 243 – 249. Corhay, A., Rad, A. T., & Urbain, J.-P. (1995). Long run behavior of Pacific-Basin stock prices. Applied Financial Economics, 5, 11 – 18. Dhakal, D., Kandil, M., & Sharma, S. C. (1993). Causality between the money supply and share prices: a VAR investigation. Quarterly Journal of Business and Economics, 32 (3), 52 – 74. Engle, R. F., & Susmel, R. (1993). Common volatility in international equity markets. Journal of Business and Economics Statistics, 11, 167 – 176. Fama, E. F. (1981). Stock returns, real activity, inflation, and money. The American Economic Review, 71 (4), 545 – 565. Gallinger, G. W. (1994). Casuality tests of the real stock return — real activity hypothesis. The Journal of Financial Research, 17 (2), 271 – 289. Granger, C. W. J. (1969). Investigating causal relations by econometric models and cross-spectral methods. Econometrica, 37, 424 – 438. Granger, C. W. J. (1988). Some recent development in a concept of causality. Journal of Econometrics, 39, 199 – 211. Johansen, S. (1988). Statistical analysis if cointegration vector. Journal of Economic Dynamics and Control, 12 (2/3), 231 – 254. Johansen, S. (1991). Estimation and hypothesis testing of cointegration vectors in Gaussian vector autoregressive models. Econometrica, 59 (6), 1551 – 1580. Johansen, S., & Juselius, K. (1990). The full information maximum likelihood procedure for inference on cointegration with applications to the demand for money. Oxford Bulletin of Economics and Statistics, 52 (2), 169 – 210. Kasa, K. (1992). Common stochastic trends in international stock markets. Journal of Monetary Economics, 29, 95 – 124. Koutmos, G. (1996). Modeling the dynamic interdependence of major European stock markets. Journal of Business Finance and Accounting, 23 (7), 975 – 988. Kwan, A. C. C., Sim, A.-B., & Cotsomitis, J. A. (1995). The causal relationships between equity indices on world exchanges. Applied Economics, 28 (1), 33 – 37. Lai, M., Lai, K. S., & Fang, H. (1993). Dynamic linkages between the New York and Tokyo stock markets: a vector error correction analysis. Journal of International Financial Markets, Institutions and Money, 3 (2), 73 – 96. Mahdavi, S., & Sohrabian, A. (1991). The link between the rate of growth of stock prices and the rate of growth of GNP in the United States: a Granger causality test. The American Economist, 35 (2), 41 – 48. Masih, A. M. M., & Masih, R. (1997). Dynamic linkages and the propagation mechanic driving major international markets: an analysis of the pre- and post-crash eras. The Quarterly Review of Economics and Finance, 37 (4), 859 – 885.

108

O. Ratanapakorn, S.C. Sharma / Review of Financial Economics 11 (2002) 91–108

Osterwald-Lenum, M. (1992). A note with quartiles as the asymptotic distribution of the maximum likelihood cointegration rank test statistics. Oxford Bulletin of Economics and Statistics, 54, 461 – 472. Peek, J., & Rosengren, E. S. (1997). The international transmission of financial shocks: the case of Japan. The American Economic Review, 87 (4), 495 – 506. Perron, P. (1988). Trend and random walks in macroeconomic time series: further evidence from a new approach. Journal of Economic Dynamics and Control, 12, 297 – 332. Phillips, P. C. B., & Perron, P. (1988). Testing for a unit root in time series regression. Biometrika, 75, 335 – 346. Ripley, D. M. (1973). Systematic elements in the linkage of national stock market indices. Review of Economics and Statistics, 55, 356 – 361. Said, S., & Dickey, D. (1984). Testing for unit roots in autoregressive-moving average models with unknown order. Biometrika, 71, 599 – 607. Serletis, A., & King, M. (1997). Common stochastic trends and convergence of European union stock markets. The Manchester School, 65 (1), 44 – 57. Sharma, S. C., & Wongbangpo, P. (2002). Long term trends and cycles in ASEAN stock markets. Forthcoming, Review of Financial Economics. Wongbangpo, P., & Sharma, S. C. (2002). Stock market and macroeconomic fundamental dynamic interactions: ASEAN-5 countries. Journal of Asian Economics, 13 (1), 27 – 51.