Physica A 387 (2008) 25472560 www.elsevier.

com/locate/physa

Parametric and nonparametric Granger causality testing: Linkages between international stock markets
Jan G. De Gooijer a, , Selliah Sivarajasingham b
a Department of Quantitative Economics and Tinbergen Institute, University of Amsterdam Roetersstraat 11,

1018 WB Amsterdam, The Netherlands

b Department of Economics, University of Peradeniya, Peradeniya, Sri Lanka

Received 12 July 2007; received in revised form 2 November 2007 Available online 11 January 2008

Abstract This study investigates long-term linear and nonlinear causal linkages among eleven stock markets, six industrialized markets and ve emerging markets of South-East Asia. We cover the period 19872006, taking into account the on-set of the Asian nancial crisis of 1997. We rst apply a test for the presence of general nonlinearity in vector time series. Substantial differences exist between the pre- and post-crisis period in terms of the total number of signicant nonlinear relationships. We then examine both periods, using a new nonparametric test for Granger noncausality and the conventional parametric Granger noncausality test. One major nding is that the Asian stock markets have become more internationally integrated after the Asian nancial crisis. An exception is the Sri Lankan market with almost no signicant long-term linear and nonlinear causal linkages with other markets. To ensure that any causality is strictly nonlinear in nature, we also examine the nonlinear causal relationships of VAR ltered residuals and VAR ltered squared residuals for the post-crisis sample. We nd quite a few remaining signicant bi- and unidirectional causal nonlinear relationships in these series. Finally, after ltering the VAR-residuals with GARCH-BEKK models, we show that the nonparametric test statistics are substantially smaller in both magnitude and statistical signicance than those before ltering. This indicates that nonlinear causality can, to a large extent, be explained by simple volatility effects. c 2008 Elsevier B.V. All rights reserved.
Keywords: GARCH-BEKK; Granger causality; Hypothesis testing; Nonparametric; Stock market linkages

1. Introduction Since the late 1980s many national stock exchange markets in industrial countries have become aware of the increased competitiveness among these markets. This, in conjunction with a less restrictive climate toward capital movements has brought about the view among economists that the major nancial markets of the world are systematically interrelated. This interrelationship may indicate a growing similarity in reactions toward external developments in macroeconomic policies and in the world nancial environment. In addition, it may also reect a temporary, or perhaps more lasting, causal relationship between various individual stock exchanges.
Corresponding author.

E-mail addresses: j.g.degooijer@uva.nl (J.G. De Gooijer), ssivaraj@pdn.ac.lk (S. Sivarajasingham). 0378-4371/$ - see front matter c 2008 Elsevier B.V. All rights reserved. doi:10.1016/j.physa.2008.01.033

2548

J.G. De Gooijer, S. Sivarajasingham / Physica A 387 (2008) 25472560

Causal linkages among stock markets have important implications for security pricing, hedging and trading strategies, and nancial market regulations. Also the presence of long-term linear and nonlinear relationships may be used to achieve nancial gains from international portfolio diversication and to reduce systematic local risks. Consequently, there exists a large body of literature examining the presence of causal linkages between developed (less risky) markets. They typically nd that the US market leads other developed markets (e.g. Ref. [31]). However, there is substantially less literature on stock market linkages between developed markets and emerging markets; see Section 2 for a selective overview. Moreover, quite a few studies relied on the restrictive assumption of a causal linear relationship between stock markets through the use of Grangers [21], parametric, causality test. But, as noted by Hsieh [26,27] and many others, nancial time series exhibit signicant nonlinear features. Indeed, Hiemstra and Jones [25] argue that a nonlinear and nonparametric Granger causality test (hereafter HJ test), based on the work of Baek and Brock [4], is more effective in uncovering certain nonlinear causal relationships in daily stock prices. The HJ causality test seems to be the one most used in economics and nance. Examples include stockpricevolume relationships [25,43], futures and cash markets [15], stock price dividend relationships [29], fundamentals and exchange rates [35], equity volatility returns [7]. However, Diks and Panchenko [12,13] demonstrate that the HJ test can severely over-reject if the null hypothesis of noncausality is not true. In addition, with instantaneous dependence, the HJ test has serious size distortion problems. As an alternative the authors of Ref. [13] (hereafter DP) develop a new test statistic which does not suffer from these limitations. Their empirical results suggest that some of the rejections of the Granger noncausality hypothesis, using the HJ causality test, may be spurious. The objective of the current paper is two-fold. The rst one is to explore the existence of linear and nonlinear causal relationships among eleven stock markets. Six of these (Germany, Hong Kong, Japan, Singapore, UK, and US) belong to the group of worlds major stock markets, while ve markets (India, Malaysia, South Korea, Sri Lanka, and Taiwan) are emerging stock markets in South-East Asia. Clearly South-East Asia as a region has undergone rapid market liberalization in the past decade, resulting in increased investment ows. A possible consequence of this nancial openness is an increase in the causal linkages between these emerging markets and the worlds major nancial markets. In particular, the time period after the 1997 Asian nancial crisis may have changed the direction and strength of the causal relationships among the markets under study. A second objective is to explore the ability of the DP test to detect nonlinear causal relationships. The paper has ve remaining sections. Section 2 presents a brief overview of the relevant literature. Also we point out some limitations of the reviewed studies. In Section 3 we present some selected stock market indicators jointly with a discussion of the eleven stock market indices. Section 4, entitled Testing methodology, introduces (i) a multivariate test of nonlinearity; and (ii) the nonparametric DP causality test. The empirical ndings are reported in Section 5. The nal section closes the paper by discussing some of the main implications of the results and providing directions for future research. 2. Literature review There is a wealth of literature on stock market interdependence and integration. However, depending on the data, methodology, and theoretical models used there is no clear resolution of the issue yet. Some previous work has have found that international stock markets are integrated; see, e.g., Refs. [2,23]. Others have found that stock markets are not interlinked; see, e.g., Refs. [41,44]. Most of the studies on stock market interdependence in emerging markets have been done on geographical groups of markets, such as markets in Central and Eastern Europe [20,22], Latin America [10,11,9], and in Asian countries. Since stock markets in South-East Asia form a substantial part of the set of markets considered here, we summarize some of the most recent ndings. Masih and Masih [33,34] found cointegration in the pre-nancial crisis period of October 1987 among the stock markets of Thailand, Malaysia, the US, the UK, Japan, Hong Kong and Singapore. But there were no long-run relationships between these markets for the period after the global stock market crash of 1987. By contrast, Phylaktis and Ravazzolo [39] found no linkages and dynamic interactions amongst a group of Pacic-Basin stock markets (Hong Kong, South Korea, Malaysia, Singapore, Taiwan and Thailand) and the industrialized countries of Japan and US for the period 19801998. Further, Arshanapalli et al. [3] noted an increase in stock market interdependence after the 1987 crisis for the emerging markets of Malaysia, the Philippines, Thailand, and the developed markets of Hong Kong, Singapore, the US and Japan for the period 19861992. Likewise, when testing for causality-in-variance,

J.G. De Gooijer, S. Sivarajasingham / Physica A 387 (2008) 25472560

2549

Caporale et al. [8] found some empirical evidence on the bi-directional causal relationships between stock prices and exchange rate volatility in the case of Indonesia and Thailand for the post-crisis period 19872000. Also, Najand [36], using linear state space models, detected stronger interactions among the stock markets of Japan, Hong Kong, and Singapore after the 1987 stock market crash. Linkages among national stock markets before and during the period of the Asian nancial crisis in 1997/98 were explored by Sheng and Tu [42]. In particular, adopting multivariate cointegration and error-correction tests, these authors focused on 11 major stock markets in the Asian-Pacic region (Australia, China, Hong Kong, Indonesia, Japan, Malaysia, Philippines, Singapore, South Korea, Taiwan and Thailand) and the US. Using daily closing prices, they found empirical evidence that cointegration relationships among the national stock indices has increased during, but not before, the period of the nancial crises. More recently, Weber [46] revealed various causality-in-variance effects between the volatilities in the national nancial markets in the Asian-Pacic region (Australia, Hong Kong, Indonesia, India, Japan, South Korea, New Zealand, Philippines, Singapore, Taiwan and Thailand) for the post-crises period 19992006. Also, allowing for structural breaks such as the Asian nancial crisis, Narayan et al. [37] found that stock prices in Bangladesh, India and Sri Lanka Granger-cause stock prices in Pakistan for the period 19952001. All the above studies rely on the restrictive assumption of linearity either through the use of linear causality tests or via linear time series methodology. Moreover, some of these studies fail to notice that parametric linear Granger causality tests have low power against nonlinear alternatives; see Ref. [4]. Recognition of the nonlinear property of stock prices, and subsequently exploring for possible long-run nonlinear relations among national stock markets, came after publication of the study by Hiemstra and Jones [25]. For instance, using the HJ causality test, Hunter [28] focuses on the emerging markets of Argentina, Chile, and Mexico. Similarly, Ozdemir and Cakan [38] examine the dynamic relationships between the stock market indices of the US, Japan, France, and the UK. This latter study reports that there is a strong bi-directional nonlinear causal relationship between the US and the other countries, which has also been documented in the literature using linear causality tests. But, as explained in Section 1, the HJ test may lead to false inference. Clearly the causality and nonlinearity tests to be introduced in Section 4 provide a useful way to extend and update much of the above empirical knowledge on causal (non)linear relationships. 3. Data Data consists of eleven time series of daily closing (5 days) stock market price indices, measured in domestic currencies. The data covers two periods: P1, from November 2, 1987 June 30, 1997, denoting the pre-Asian nancial crisis period (2521 observations), and P2, the post-Asian nancial crisis period, with data from June 1, 1998 December 1, 2006 (2220 observations). Recall that the on-set of the Asian nancial crisis started with a 15%20% devaluation of Thailands Baht which took place on July 2, 1997. This was subsequently followed by devaluations of the Philippine Peso, the Malaysian Ringgit, the Indonesian Rupiah, and the Singaporean Dollar. In addition, the currencies of South Korea and Taiwan suffered. Further in October, 1997 the Hong Kong stock market collapsed with a 40% loss. In January 1998, the currencies of most South-East Asian countries regained parts of the earlier losses. The data are taken from DataStream. Table 1 presents some basic information about the eleven stock markets. Among the ve emerging markets, the Taiwan (TAW) stock market is the largest in terms of market capitalization, followed by Malaysia (MAL) and India (IND). By contrast Sri Lanka (SL) is a relatively small market. Also, in terms of listed companies, the Sri Lankan market is the smallest. As can be seen from the listed trading values, the Asian nancial crisis is clearly visible with a drop in the 1998 gures. The six developed stock markets have been deregulated and liberalized for quite a very signicant period of time. For the emerging markets, Bekaert et al. [6] dates Indias integration into the world equity market as 1992. Malaysia, Taiwan and South Korea, however, are still deregulating and liberalizing their markets, a process which began in the late 1980s. Sri Lankas stock market (Colombo Stock Exchange (CSE)), on the other hand, underwent a rapid increase in foreign investment following liberalization in 1989. According to Ariff and Khalid [1] and Elyasiani et al. [16] the CSE was one of the best performing markets in the 19891994 period, with a 15-fold increase in annual turnover and an eight-fold increase in market capitalization. The trading hours of the eleven stock exchanges are not perfectly synchronized, though there are several overlapping hours in each trading day for the developed markets. But, within the group of emerging markets, the trading activity is to a large extent concurrent. Nevertheless, the differences in closing times could cause sequential

2550 Table 1 Some selected stock market indicators Country GER HNG JAPa SNG UK US IND SL MAL SK TAW GER HNG JAPa SNG UK US IND SL MAL SK TAW GER HNG JAPb SNG UK US IND SL MAL SK TAW 1997 1998

J.G. De Gooijer, S. Sivarajasingham / Physica A 387 (2008) 25472560

1999 933 717 2470 355 1945 7651 5863 239 757 1178 462

2000 1022 779 2561 418 1904 7524 5937 239 795 1308 531 68.1 383.4 66.3 164.8 180.3 154.0 32.4 6.6 130.4 37.5 79.8 1,069,120 377,866 2,693,856 91,485 1,835,278 31,862,485 509,812 144 58,500 1,067,999 983,491

2001 749 657 2471c 386 1923 6355 5795 238 809 1409 584 57.8 310.8 54.1 138.3 151.3 137.9 23.1 8.5 136.4 45.7 104.1 1,419,579 196,361 1,826,230 63,385 1,861,131 29,040,739 249,298 153 20,772 703,960 544,808

2002 715 968 3058 434 2405 5685 5650 238 865 1518 638 34.8 289.5 53.5 115.4 119.2 106.4 25.7 10.2 130.0 45.6 90.4 1,233,056 210,662 1,573,279 56,129 1,909,716 25,371,270 197,118 318 27,623 792,165 631,931

2003 684 1027 3116 456 2311 5295 5644 244 897 1563 689 44.2 460.7 70.9 158.0 136.8 130.2 46.5 14.9 162.0 54.2 132.4 1,147,209 331,615 2,272,989 87,864 2,211,533 15,547,431 284,802 769 50,135 682,706 592,012

2004 660 1086 3220 489 2486 5231 4730 245 962 1573 697 43.6 528.5 79.6 160.6 132.6 139.4 56.1 18.2 160.6 63.1 144.5 1,406,055 438,966 3,430,420 81,314 3,707,191 19,354,899 39,085 582 59,878 638,891 718,619

Number of listed companies 700 741 671 693 2387 2416 303 321 2157 2087 8851 8450 5843 239 708 1135 404 5860 233 736 1079 437

Market capitalization as a % of GDP 39.1 51.0 68.1 241.7 211.2 385.1 51.4 63.3 101.2 112.5 114.9 240.1 91.4 166.7 201.2 137.0 154.3 180.7 30.5 13.9 93.4 9.7 101.6 24.5 10.9 136.0 37.8 99.8 41.5 10.1 184.0 97.4 130.6 814,740 247,428 1,849,228 97,985 1,377,859 18,574,100 278,828 209 48,512 825,828 910,016

Trading value ($USm) 535,745 761,888 489,365 205,918 1,251,750 948,522 63,954 50,735 829,131 1,167,382 10,216,074 13,148,480 158,301 309 153,292 172,019 1,291,524 148,239 281 29,889 145,572 890,491

Source: Standard & Poors [45]. a From 2002 gures include data from both the Tokyo Stock Exchange and JASDAQ. b Figures include data from the Tokyo Stock Exchange and Osaka Stock Exchange. From 2002, JASDAQ gures are included. c Indicates a break in the series.

price responses to common information that could be mistaken for causal linkages. Intra-day market data may be used to distangle these sequential responses from causal transmissions within a particular day. Regrettably these data are not available for the emerging markets under study. In the rst part of the study, we analyze daily returns Rt = ln Pt ln Pt1 , where Pt is the closing price of an index on day t. In the second part, we focus on the squared and unsquared residuals from linear VAR models tted to {Rt }. The squared residuals may be considered as a useful proxy of volatility. Application of three augmented DickeyFuller tests (no trend and no intercept, intercept, and trend plus intercept) indicated that the series {Rt } are integrated of order zero, i.e. stationary, with p-values less than 0.01. The appropriate lag lengths for the tests were selected by minimizing AIC. The JarqueBera test for normality indicated that the returns are not normally distributed. Initial exploratory analysis of the sample cross-correlation matrix at lag 0 (contemporaneous correlation) indicated that almost all series are positively correlated, and signicantly different from zero at the 1% level, for the series {Rt }, and for both time periods. Very low (insignicant) cross-correlations at lag 0 were obtained for the pairs SLJAP,

J.G. De Gooijer, S. Sivarajasingham / Physica A 387 (2008) 25472560

2551

SLTAW, SLUK, and SLUS. Signicant sample cross-correlations at lag 1 were noted for the UK and US stock markets having uni-directional links with almost all other stock markets. However, it is well-known that correlations cannot fully capture the long-term dynamic linkages between the markets in a reliable way. Hence, these results should be interpreted with caution. Consequently, they are not included in the paper. Indeed, what is needed is a long-term causality analysis between the markets. Empirical experience indicates that it is rather hard to absorb simultaneously large tables with numbers. To overcome this difculty, and to save space, we use the following simplifying notation: means that the corresponding p-value of a particular test (causality and nonlinearity) is smaller than 1%; means that the corresponding p-value of a test is in the range 1%5%; and denotes that the corresponding p-value of a test is larger than 5%.1 Bi- and uni-directional causalities will be denoted by the functional representations and , respectively. 4. Testing methodology 4.1. A multivariate test of nonlinearity Let Yt = (Y1,t , . . . , Yk,t ) (t = 1, . . . , n) denote a stationary k-variate time series of length n. A multivariate nonlinear model can be expressed as Yi,t = i + f i (1,t1 , 2,t1 , . . .) + i,t , where {i,t } are serially uncorrelated, but may be cross-correlated at lag zero, and identically distributed random variables, and f i () are measurable real-valued functions. In general, f i () (i = 1, . . . , k) can be represented by a discrete-time Volterra series of the form
k k

f i (1,t1 , 2,t1 , . . .) =
s=1 l=1 k

bi,sl s,tl +
s,u=1 l,m=1

bi,sulm s,tl u,tm

+
s,u,v=1 l,m,r =1

bi,suvlmr s,tl u,tm v,tr + .

(1)

This is a generalization of the Volterra representation of a nonlinear stationary univariate time series. A k-variate linear process results if all the coefcients of the second and higher-order terms in (1) equal zero. In practice the upper and lower limits in the summations are replaced by nite numbers. For convenience, assume that each component of Yt has mean zero. Then the idea for the test is that if a vector time series process is nonlinear, this structure will be reected in the residuals of a tted linear pth-order VAR model. The test procedure consists of the following steps. 1. Fit a VAR( p) model to Yt by regressing Yt on the pk 1 vector Zt = (Y1,t1 , . . . , Yk,t1 , . . . , Y1,t p , . . . , Yk,t p ). Compute the k 1 vector of tted values Yt , (t = p + 1, . . . , n), the k 1 vector of residuals et = Yt Yt , and the corresponding k k matrix SS R1 of sum of squared and cross-product terms for the regression. 2. Compute a k 1 vector of squares of tted values, say Xt , from the k-variate AR( p) regressions in step (1). Remove the linear dependence of Xt on Zt by a second k-variate AR( p) regression of Xt on Zt . Obtain the k 1 vector of tted values Xt , and the k 1 vector of residuals ut = Xt Xt . 3. Regress the vector of residuals et from step (1) on the vector of residuals ut from step (2). 4. Compute the corresponding k k sum of squared regressions matrix, SS R2 , and sum of squared errors matrix, SS E 2 . Let SS R2|1 = SS R2 SS R1 , i.e. SS R2|1 is the extra sum of squares due to the addition of the second-order term to the model. 5. Compute the F-statistic F: F= n p pk k k 1 1/2 1/2 , (2)

1 The actual p-values of all tests used in the paper are available upon request.

2552

J.G. De Gooijer, S. Sivarajasingham / Physica A 387 (2008) 25472560

where = |SS E 2 |/|SS R2|1 + SS E 2 | is Wilks lambda statistic. Under the assumption that Yt follows a zero-mean Gaussian VAR( p) process, and if the sample size n is large, F follows approximately an F1 ,2 distribution with degrees of freedom 1 = k and 2 = (n p) pk k.2 Simulation results of Harvill and Ray [24] indicate that, in general, the multivariate version of Keenans [30] test is more powerful than the univariate tests for at least one of the component series, in particular when the nonlinearity in one series of the vector process is due solely to terms from the other series. 4.2. The nonparametric DP causality test The general setting for a causality test is as follows. Assume {X t , Yt ; t 1} are two scalar-valued strictly stationary time series. Then {X t } is a strictly Granger-cause of {Yt } if past and current values of X t contain additional information on future values of Yt that is not contained in the past and current Yt -values alone. More formally, let F X,t and FY,t denote the information sets consisting of past observations of X t and Yt up to and including time t, and let denote equivalence in distribution. Then {X t } is a Granger-cause of {Yt } if, for some k 1, (Yt+1 , . . . , Yt+k )|(F X,t , FY,t ) (Yt+1 , . . . , Yt+k )|FY,t . (3)

This denition is general and does not involve model assumptions. In practice one often assumes k = 1, i.e. testing for Granger noncausality comes down to comparing the one-step-ahead conditional distribution of {Yt } with and without past and current observed values of {X t }, which will also be the case considered here. Note the testing framework introduced above concerns conditional distributions given an innite number of past observations. In practice, however, tests are usually conned to nite orders in {X t } and {Yt }. To this end, dene the delay vectors Xt X = (X t X +1 , . . . , X t ), and Yt Y = (Yt Y +1 , . . . , Yt ), ( X , Y 1). If past observations of Xt X contain no information about future values, it follows from (3) that the null hypothesis of interest is given by H0 : Yt+1 |(Xt X ; Yt Y ) Yt+1 |Yt Y . (4)

For a strictly stationary bivariate time series, (4) comes down to a statement about the invariant distribution of the X + Y + 1-dimensional vector (Xt X , Yt Y , Z t ) where Z t = Yt+1 . To simplify notation we drop the time index t. Further, it is assumed that X = Y = 1. Hence, under the null, the conditional distribution of Z given (X, Y ) = (x, y) is the same as that of Z given Y = y. Then (4) can be restated in terms of ratios of joint distributions. Specically, the joint probability density function f X,Y,Z (x, y, z) and its marginals must satisfy the relationship f X,Y,Z (x, y, z)/ f Y (y) = ( f X,Y (x, y)/ f Y (y))( f Y,Z (y, z)/ f Y (y)). Thus X and Z are independent conditionally on Y = y, for each xed value of y. DP [13] show that this reformulated H0 implies q E[ f X,Y,Z (X, Y, Z ) f Y (Y ) f X,Y (X, Y ) f Y,Z (Y, Z )] = 0. Let fW (Wi ) denote a local density estimator of a dW -variate random vector W at Wi dened by fW (Wi ) = (2n )dW (n 1)1 j, j=i IiW , where IiW = I ( Wi W j < n ) with I () the indicator function and n the j j bandwidth, depending on the sample size n. Given this estimator, the test statistic of interest is given by Tn (n ) = n1 n(n 2)
2 fY (Yi )( fX,Z |Y (X i , Z i |Yi ) fX |Y (X i |Yi ) fZ |Y (Z i |Yi )). i 1 4 1 < < 3 ). Then DP [13] prove that (5) satises

(5)

Suppose that n = Cn (C > 0,

(Tn (n ) q) D n N (0, 1), Sn

D

(6)

where denotes convergence in distribution, and Sn is an estimator of the asymptotic variance of Tn () as discussed in detail by DP [13, Appendix A].3
2 FORTRAN-code for computing F can be obtained from the rst author. 3 C-code for computing the T () test statistic can be downloaded from: http://www1.fee.uva.nl/cendef/upload /15/GCTtest.zip.
n

J.G. De Gooijer, S. Sivarajasingham / Physica A 387 (2008) 25472560

2553

Table 2 Multivariate nonlinearity test results (above diagonal) and orders of tted VAR( p) models (below diagonal): Periods P1 (pre-Asian nancial crisis) and P2 (post-Asian nancial crisis) GER P1 GER HNG JAP SNG UK US IND MAL SK SL TAW 10 2 1 1 7 1 7 1 9 1 P2 3 1 4 10 3 10 5 3 2 4 HNG P1 10 10 10 7 10 5 5 10 5 P2 1 8 6 5 1 6 5 2 5 JAP P1 2 2 6 10 3 3 9 10 P2 1 5 5 6 7 1 2 4 SNG P1 1 8 1 5 1 6 3 P2 8 4 1 10 1 2 10 UK P1 8 1 1 1 8 1 P2 8 5 6 5 3 6 US P1 1 7 1 7 2 P2 1 7 3 2 4 IND P1 1 1 10 1 P2 7 1 2 4 MAL P1 1 6 3 P2 5 7 5 SK P1 6 2 P2 2 5 SL P1 6 P2 4 TAW P1 P2

5. Empirical results 5.1. Multivariate nonlinearity test Before investigating specic linear and nonlinear causal relationships between pairs of bivariate time series, it is desirable to test for the general vector nonlinear structure. Using linear VAR( p) models having p = 0, 1, . . . , 10, the AIC model selection criterion was used to select the best model tted to pairs of time series. The optimal orders p are listed in Table 2 below the main diagonal. Applying the multivariate Keenan test of Section 4, a summary of information about the results of the nonlinearity test statistic is presented above the main diagonal. Some observations are in order. The Asian nancial crisis has an obvious impact on the orders of the selected VAR( p) models. The total number of pairs of series having the same VAR orders in both periods is 8 (14.5%) while in 47 cases (85.5%) there has been a change between the orders p selected for period P1 and period P2. Within the set of developed (emerging) markets these percentages are 6% (10%) and 94% (90%), respectively. Nonlinearity, at the 1% level, is detected for 35 (64%) pairs of indices in period P1, and for 23 (42%) pairs in period P2. No evidence of general nonlinear structure, at the 1% level, in the data is indicated for 12 (22%) pairs in period P1, and for 29 (53%) pairs in period P2. Clearly, this is an overall view of the data. But these percentages suggest that there is weak evidence that the number of paired nonlinearities have changed between both periods. At a more detailed level, however, there appear to be more differences between both time periods. In particular, we see that for 17 pairs (31%) the simplifying notation for the signicance of the multivariate nonlinearity test changed from to , and for only 3 (5%) pairs it changed from to . Among the whole set of nonlinearity test results, those presented for Malaysia are interestingly different. Indeed, there is evidence that this market has signicant () nonlinear relationships with almost all other markets. Moreover, these relationships remain fairly persistent across both time periods. Broadly, the signicant nonlinearity test results reported in Table 2 motivate the search for causal and persistent nonlinear linkages between pairs of stock indices by a nonparametric test. Nevertheless these results should be interpreted with caution since high kurtosis in the returns may affect the null distribution of the test. 5.2. Parametric and nonparametric causality tests: No pre-ltering For each of the eleven series {Rt } pairwise causality testing was carried out using the Wald variant of Grangers test. Simulation results provided by Geweke et al. [19] show that this type of Granger causality test has a number of advantages over eight alternative tests of causality. In each case, the optimal lag orders reported in Table 2 are used for the unrestricted VAR model. It is well-known that the results of the Granger causality test are sensitive to the choice of the lag length, even when a sophisticated search routine for the lag length has been implemented in the process of nding the best VAR specication. Therefore it is safe to restrict the discussion below to causality results obtained at

2554

J.G. De Gooijer, S. Sivarajasingham / Physica A 387 (2008) 25472560

Table 3 Parametric Granger causality testing for periods P1 (pre-Asian nancial crisis) and P2 (post-Asian nancial crisis) using unltered returns {Rt } GER P1 GER HNG JAP SNG UK US IND MAL SK SL TAW P2 HNG P1 P2 JAP P1 P2 SNG P1 P2 UK P1 P2 US P1 P2 IND P1 P2 MAL P1 P2 SK P1 P2 SL P1 P2 TAW P1 P2

Signicant ( or ) entries indicate that stock market X (top row) has a causal linear relationship with stock market Y (left column), i.e. X Y .

the 1% level, denoted by , rather than the 5% level (). Table 3 reports the summarized results. They permit the following observations. Not surprisingly, the six leading markets Germany, Hong Kong, Japan, Singapore, UK, and US have a strong degree of causal linear relationship which also affects four of the ve emerging markets. A clear exception is Sri Lanka. Interestingly none of Sri Lankas ten major trading partners have a signicant causal linear relationship with the CSE. This applies to both time periods. In fact, the Sri Lankan market behaves completely through its own internal dynamics. This is in accordance with the study by Elyasiani et al. [16] who considered the period 19891994. This lack of integration allows CSE to provide additional prot opportunities and diversication benets to global portfolios. In period P1 six signicant () bi-directional causal linear relationships exist, i.e. GER MAL, JAP UK, JAP SK, US GER, US SNG, and MAL SNG. In period P2 there are ve signicant () bi-directional linear causalities: UK GER, US TAW, US UK, SNG HNG, and SNG TAW. Thus there is no interdependence among indices of the emerging markets. Interestingly, in period P2, strong evidence of bi-directional linear relationship can be seen between the stock indices of Taiwan and those of the US and Singapore. The total number of signicant uni-directional causal linear relationships is 34 (31%) in period P1, and 43 (39%) in period P2. Not surprisingly, the US stock market is dominating all other markets. Further, there is no strong evidence of the inuence of Japan on any other market in period P2, apart from Taiwan. This parallels early ndings by Wu and Su [47], using a linear Granger causality test. In what follows we discuss results for the Tn () test for lags X = Y = 1. Only in this case the asymptotic result (6) has been proved. To implement the DP test, the constant C for the bandwidth n was set at 7.5. With the theoretical optimal rate = 2/7 given by DP [13], this implies a bandwidth value of approximately 0.75, for both n = 2520 (P1) and n = 2219 (P2). Selected bandwidth values smaller (larger) than 0.75 resulted, in general, in larger (smaller) p-values. Summary results of the DP nonparametric causality test are reported in Table 4. These results permit the following observations. In period P1 signicant () bi-directional causal nonlinear relationships exist between the following 10 (18%) stock markets: GER MAL, US GER, MAL SNG, UK GER, UK MAL, UK SNG, SNG HNG, MAL HNG, MAL SL, and IND JAP. In period P2, however, the number of bi-directional nonlinear relationships has considerably increased to 41 (75%). This strengthening of linkages exists across almost all countries, except for Sri Lanka. This suggest that the Asian nancial crisis had a profound effect on the relationships among the stock markets. Note the striking difference between the results of Taiwan in the pre- and post-Asian nancial crisis periods. 5.3. Nonparametric causality testing: VAR-ltering The results in Tables 2 and 3 suggest that there are signicant and persistent linear and nonlinear causal linkages between the stock markets. However, the results obtained for the DP test may be blurred by the presence of the combined effect. In other words, even though we found nonlinear causality, the DP test should be reapplied to ltered

J.G. De Gooijer, S. Sivarajasingham / Physica A 387 (2008) 25472560 Table 4 Nonparametric causality testing for periods P1 (pre-Asian nancial crisis) and P2 (post-Asian nancial crisis) using unltered returns {Rt } GER P1 GER HNG JAP SNG UK US IND MAL SK SL TAW P2 HNG P1 P2 JAP P1 P2 SNG P1 P2 UK P1 P2 US P1 P2 IND P1 P2 MAL P1 P2 SK P1 P2 SL P1 P2

2555

TAW P1 P2

Signicant ( or ) entries indicate that stock market X (top row) has a causal nonlinear relationship with stock market Y (left column), i.e. X Y . Table 5 Nonparametric causality testing for period P2 (post-Asian nancial crisis): Residuals { t } and squared residuals { t } are obtained after ltering 2 with linear bivariate VAR models GER t GER HNG JAP SNG UK US IND MAL SK SL TAW t 2 HNG t t 2 JAP t t 2 SNG t t 2 UK t t 2 US t t 2 IND t t 2 MAL t t 2 SK t t 2 SL t t 2 TAW t t 2

Signicant ( or ) entries indicate that stock market X (top row) has a causal nonlinear relationship with stock market Y (left column), i.e. X Y .

VAR-residuals to ensure that any causality found is strictly nonlinear in nature. Table 5 presents the DP nonparametric causality test for period P2, using residuals { t } and squared residuals { t2 } obtained from the linear VAR models reported in Table 2. Comparing the summary results for { t } in Table 5 with the results reported in Table 4 for the returns {Rt } in period P2 it is interesting to see that both tables show identical pairwise signicant () causal nonlinear relationships, except for the following cases: The causal nonlinear relationships from JAP UK, JAP SK, UK IND, SNG IND, SL TAW have changed from in Table 4 to in Table 5. Given the results presented in Table 2, it suggests the absence of nonlinear causality for the rst three pairs above, while the last two pairs have no signicant causal linear or nonlinear relationships. This is also the case for the signicance of the relationship HNG US which changed from in Table 4 to in Table 5. On the other hand, the signicance of the causal nonlinear relationship JAP IND changed from in Table 4 to in Table 5. Similarly, no signicant () causal nonlinear relationship could be detected from MAL JAP in Table 4. But the result in Table 5 clearly indicates: MAL JAP. These reverse changes in relationships are hard to interpret. In particular, since according to Table 2 no signicant causal linear relationship could be detected between these two specic pairs (JAPIND and MALJAP) at the 1% level. When we exclusively focus on the results in Table 5, the following additional observations emerge.

2556

J.G. De Gooijer, S. Sivarajasingham / Physica A 387 (2008) 25472560

The total number of signicant () bi-directional causal nonlinear relationships is 33 (60%) for { t }, and 17 (31%) for { t2 }. Only 16 pairs have signicant common causal nonlinear relationships for both squared and unsquared residuals. Again, there is almost no signicant interdependence between the Sri Lankan market and the other markets. Only, in terms of squared residuals, signicant () causal nonlinear relationships can be seen from MAL SL and from SK SL. The nature and source(s) of the detected nonlinearities may be different from that of the linear Granger causality test. For instance, exchange rate volatility might induce nonlinear causality in the emerging markets. Caporale et al. [8] provide empirical evidence for this proposition using daily exchange rates (in local currencies) and stock price indices for Indonesia, Japan, South Korea, and Thailand. Also, persistent nonlinear causality can be induced by relatively high transaction costs. Research by Domowitz et al. [14] shows that trading costs in the emerging markets are signicantly higher than those in more developed markets, with South Korea being one of the most expensive markets. This result still holds after correcting for factors affecting costs such as market capitalization and volatility. Consequently, investors in these markets pool their information until it is protable to trade. Needless to say, there may be other sources of the observed nonlinearities. 5.4. Nonparametric causality testing: GARCH-BEKK ltered VAR-residuals Given the causality results for { t2 } in Table 5, it is interesting to reinvestigate the hypothesis of nonlinear noncausality after controlling for conditional heteroskedasticity in the data. Many vector linear and vector nonlinear (asymmetric) time series models can be used for this purpose. As an illustration, we restrict our attention here to the bivariate GARCH-BEKK [17] processes of order ( p, q), in which (t ) = (1,t , 2,t ) follows the equations
p q

t = Ht t , with parameter matrices C= c11 0 c12 c22 ,

1/2

Ht = C C +
i=1

Ai ti ti Ai +
j=1

B j Ht j B j

(7)

Ai =

a11 a21

(i) (i)

a12 a22

(i)

(i)

Bj =

b11

( j) ( j)

b12

( j) ( j)

b21

b22

and {t } is a sequence of i.i.d. random variables with mean zero and 2 2 covariance matrix I. Note, Ht is the conditional covariance matrix of {t }, i.e. t |Ft1 (0, Ht ) with Ft1 the information set at time t 1. Through the matrices Ai and B j , (7) allows for own-market- and cross-market interactions. The residuals are obtained by the whitening matrix transformation H1/2 t . Weak restrictions on Ai and B j guarantee that Ht is always positive denite; see, e.g., Ref. [5] for a full account of the properties of multivariate GARCH models.4 Using the results in Table 5, we initially selected 32 pairs of stock market indices showing signicant (i.e. (, ), (, ), or (, )) causal nonlinear relationships in the squared and unsquared residuals. As a rst step, we tted GARCH-BEKK( p, q) models to ( 1,t , 2,t ) with values of the order ( p, q) in the range [1, 2]. At this point it is worth mentioning that tted GARCH-BEKK models are only useful if the process under study is covariance stationary. This p q property can be veried by computing the eigenvalues of the expression i=1 (Ai Ai ) + j=1 (B j B j ) with the Kronecker product. If all the eigenvalues are less than one in modulus, the covariance-stationarity condition is satised. Only for eight pairs of residuals the maximum eigenvalue was less than one. Table 6 reports the best (in terms of minimal AIC values, and eigenvalues less than one) tted models. Asymptotic standard errors are in parentheses. No efforts were made to improve the tted models by setting insignicant parameters equal to 0. Table 6 also shows results of the nonparametric DP causality test statistic applied to the VAR-residuals before and after GARCH-BEKK-ltering.5
4 Parameters of (7) will be estimated by the routine mvBEKK.est (default settings) contained in the R-package mgarchBEKK. The package can be downloaded from: http://www.vsthost.com. 5 As far as we know it is unclear how the asymptotic distribution of the nonparametric DP causality test statistic is affected by the use of estimated model (e.g. VAR, GARCH-BEKK, etc.) residuals.

J.G. De Gooijer, S. Sivarajasingham / Physica A 387 (2008) 25472560

2557

Table 6 GARCH-BEKK estimation results applied to the VAR-residuals, and nonparametric causality test results applied to the VAR-residuals before and after GARCH-BEKK-ltering: Period P2 (post-Asian nancial crisis) GERHNG C Estimation results 0.0004 0.0065
(0.0005) (0.0006)

HNGUK 0.0119
(0.0003)

UKUS 0.0041
(0.0002)

MALUK 0.0011
(0.0004)

0.0049
(0.0005)

0.0002
(0.0002)

0.0000
(0.0004)

0 A1 0.3449
(0.0334)

0.0018
(0.0013)

0 0.2029
(0.0453)

0.0038
(0.0002)

0 0.3149
(0.0478)

0.0001
(0.0004)

0 0.3619
(0.0263)

0.0029
(0.0003)

0.4627
(0.0397)

0.2616
(0.0369)

0.2862
(0.0298)

0.0009
(0.0292)

0.1099
(0.0266)

0.1312
(0.0394)

0.3952
(0.0615)

0.4671
(0.0218)

0.0138
(0.0675)

0.0990
(0.1266)

0.0280
(0.0269)

0.2615
(0.0573)

A2

0.1065
(0.0509)

0.1174
(0.0449)

0.2374
(0.0764)

0.0143
(0.0363)

0.1716
(0.0449)

0.5469
(0.0272)

0.0458
(0.0122)

0.4631
(0.0318)

B1

0.8724
(0.0108)

0.0806
(0.0118)

0.2551
(0.0704)

0.0141
(0.0444)

0.1022
(0.0497)

0.2812
(0.0694)

0.6486
(0.2142)

0.0054
(0.0155)

0.1429
(0.0232)

0.7544
(0.0184)

0.0777
(0.0199)

0.4687
(0.0375)

0.2486
(0.0701)

0.6863
(0.0469)

0.1128
(0.0475)

0.6546
(0.2005)

B2

0.1789
(0.0362)

0.3386
(0.0309)

0.2593
(0.1530)

0.0579
(0.0986)

0.6173
(0.1955)

0.0182
(0.0171)

0.0457
(0.0523)

0.5065
(0.0530)

0.4927
(0.0915)

0.0264
(0.1221)

0.0653
(0.0299)

0.4794
(0.2388)

Before After

Nonparametric causality test results: Before and after GARCH-BEKK-ltering GER HNG HNG GER HNG UK UK HNG UK US 3.603 () 2.804 () 4.257 () 6.456 () 4.504 () 0.764 () 0.177 () 1.674 () 3.166 () 1.303 () INDTAW SNGUS 0.0031
(0.0003)

US UK 5.276 () 1.914 ()

MAL UK 4.652 () 0.614 () SKSL

UK MAL 4.702 () 1.955 ()

INDUK 0.0018
(0.0007)

Estimation results 0.0064 0.0115

(0.0015) (0.0006)

0.0053
(0.0004)

0.0009
(0.0004)

0.0192
(0.0004)

0.0018
(0.0025)

0 A1 0.5058
(0.0341)

0.0009
(0.0020)

0 0.3433
(0.0277)

0.0039
(0.0004)

0 0.4509
(0.0246)

0.0022
(0.0005)

0 0.2876
(0.0481)

0.0031
(0.0083)

0.0744
(0.0363)

0.1616
(0.0401)

0.0195
(0.0190)

0.0117
(0.0153)

0.0881
(0.0543)

0.2478
(0.0523)

0.3674
(0.0241)

0.0877
(0.0432)

0.0400
(0.0380)

0.4127
(0.0204)

0.0512
(0.0585)

0.7379
(0.0651)

B1

0.1470
(0.0722)

0.6005
(0.0082)

0.7909
(0.0279)

0.1981
(0.0566)

0.8368
(0.0169)

0.0057
(0.0125)

0.1407
(0.1614)

0.4271
(0.1263)

0.7191
(0.0623)

0.0993
(0.0779)

0.1394
(0.0202)

0.0344
(0.0036)

0.0151
(0.0182)

0.8923
(0.0128)

0.1007
(0.1284)

0.2127
(0.1155)

B2

0.1600
(0.1175)

0.6131
(0.0630)

0.0634
(0.0534)

0.5790
(0.0273)

Before After

Nonparametric causality test results: Before and after GARCH-BEKK-ltering IND TAW TAW IND SNG US US SNG IND UK 2.559 () 3.218 () 3.733 () 3.527 () 3.840 () 2.168 () 2.391 () 0.191 () 2.139 () 2.154 ()

UK IND 1.570 () 0.426 ()

SK SL 1.772 () 1.058 ()

SL SK 1.126 () 0.858 ()

Note that the GARCH-BEKK-ltering yields substantially smaller values of the nonparametric causality test statistic. In all cases the statistical signicance, as denoted by the short-hand notations , , and , is much weaker after ltering than before. These differences in statistical signicance indicate that the nonlinear causality is largely due to simple volatility effects. The GARCH-BEKK estimates are for the most part signicantly different from zero. Signicant bi-directional cross-market dependence can be noted, through checking the signicance of the off-diagonal

2558

J.G. De Gooijer, S. Sivarajasingham / Physica A 387 (2008) 25472560

elements in the matrices Ai and B j (i, j = 1, 2), in the case of GERHNG, HNGUK, UKUS, INDTAW, and SNGUS. 6. Summary and concluding remarks Several interesting conclusions with respect to the internationalization of the stock exchanges have already emerged from this study. In particular, it was shown that almost all stock markets considered here have become more internationally integrated after the Asian nancial crisis. But no signicant long-term causal linkages were found between Sri Lanka and any of the other countries during both the pre- and post-Asian nancial crisis period. Although this nding is in line with Elyasiani et al. [16] it is hard to explain this phenomenon.6 It may be due to the relatively small size of the CSE. Testing for linear and nonlinear causality in the stock-pricevolume relation for Sri Lanka may provide some information on this matter. In addition, in their 1998 paper Elyasiani et al. noted that the Sri Lankan market suffers from a lack of market liquidity and from high concentration in blue chip stocks. As a consequence, local traders hold on to their stocks, especially blue chips, for longer time periods. Whether this phenomenon is still the case is worth investigating. However, given the fact that the Sri Lankan market is found to be relatively isolated from other markets in this study, it provides prot opportunities and diversication benets to global investors. The leading role of the US market in the world stock market is clearly visible throughout all causality tests and in both time periods. This is consistent with earlier ndings by Eun and Shim [18]. In the post-Asian period the US market was inuenced by Germany, Japan, Singapore, the UK, India, Malaysia, South Korea, and Taiwan; see Tables 4 and 5. This conrms the increased nancial links between the US and the world economy. On the other hand, the Japanese market has relatively little inuence on the stock price movements in other markets once linear effects have been removed through VAR-ltering; see Table 5. This nding gives further support to an earlier study by Malliaris and Urrutia [32] who concluded that the Japanese market plays a passive role in transmitting information to other stock markets. Another nding is that no major differences exist between the persistence and strength of the bi-directional causal linkages among the industrialized markets and the emerging markets in the post-Asian nancial crisis period. These results, apart from offering a much better understanding of the dynamic linear and nonlinear relationships underlying the movements of emerging Asian stock markets than that has been noted until now, have many implications for market efciency. For instance, they may be useful in future research to quantify the process of nancial integration of emerging markets. Also, the presence of causal nonlinear relationships may inuence the greater predictability of these markets. This may, for instance, be investigated through a trading strategy using multivariate (non)parametric methodology. Another topic for future research concerns the source(s) or cause(s) of the nonlinear causal linkages. Previously we conjectured that volatility effects might induce nonlinear causality. The tted GARCH-BEKK models provide some guidance on this matter, but only in some special cases. Alternative specic parametrized structural models may be employed. Related to this, is a need for determining the economic factors driving the interdependence of emerging stock markets. Pretorius [40] studies this topic empirically by using cross-section and linear time series regression models. But more research is needed. Finally, the empirical nding of nonlinear causal relationships between emerging stock markets may be analyzed in terms of their implications for the efciency of these markets. The fact that there are long-term links between various markets may imply that excess (risk- and transactions cost-adjusted) returns exist. Given such inefciency, authorities in emerging markets may then reconsider their policies regarding, for instance, brokerage fees, stock exchange charges, and other equity trading costs. Acknowledgments We acknowledge comments from a referee which were very useful in improving the presentation of this paper. Selliah Sivarajasinghams research was supported in part by a grant from the National Center for Advanced Studies in Humanities and Social Sciences (NCAS), Sri Lanka, and the Department of Quantitative Economics, University of Amsterdam. SS would like to thank the members of the Department for their hospitality during his stay.
6 There was no civil war in Sri Lanka during the period 20022006.

J.G. De Gooijer, S. Sivarajasingham / Physica A 387 (2008) 25472560

2559

References
[1] M. Ariff, A. Khalid, Liberalization, Growth and the Asian Financial Crisis, Edward Elgar, Cheltenham, 2000. [2] B. Arshanapalli, J. Doukas, International stock market linkages: Evidence from the pre- and post October 1987 period, Journal of Banking and Finance 17 (1993) 193208. [3] B. Arshanapalli, J. Doukas, L.H.P. Lang, Pre- and post-October 1987 stock market linkages between US and Asian makets, Pacic-Basin Finance Journal 3 (1995) 5773. [4] E. Baek, W. Brock, A general test for non-linear Granger causality: Bivariate model, Working paper, Iowa State University and University of Wisconsin, Madison, WI, 1992. [5] L. Bauwens, S. Laurent, J.V. Rombouts, Multivariate GARCH models: A survey, Journal of Applied Econometrics 21 (2006) 79109. [6] G. Bekaert, C.R. Harvey, R. Lumsdaine, Dating the integration of world equity markets, Working paper, Stanford University, Stanford, CA, 2001. [7] C. Brooks, T. Henry, Linear and non-linear transmission of equity return volatility: Evidence from the US, Japan, and Australia, Economic Modeling 17 (2000) 497513. [8] G.M. Caporale, N. Pittis, N. Spagnolo, Testing for causality-in-variance: An application to the East Asian markets, International Journal of Finance and Economics 7 (2002) 235245. [9] G. Chen, M. Firth, O.M. Rui, Stock market linkages: Evidence from Latin America, Journal of Banking and Finance 26 (2002) 11131141. [10] T. Choudhry, Stochastic trends in stock prices: Evidence from Latin American markets, Journal of Macroeconomics 19 (1997) 285304. [11] A. Christo, A. Pericli, Correlation in price changes and volatility of major Latin American stock markets, Journal of Multinational Financial Management 9 (1999) 7993. [12] C. Diks, V. Panchenko, A note on the HiemstraJones test for Granger non-causality, Studies in Nonlinear Dynamics and Econometrics 9 (2005) art. 4. [13] C. Diks, V. Panchenko, A new statistic and practical guidelines for nonparametric Granger causality testing, Journal of Economic Dynamics & Control 30 (2006) 16471669. [14] I. Domowitz, J. Glen, A. Madhavan, Liquidity, volatility, and equity trading costs across countries and over time, Working paper 322, William Davidson Institute and International Finance, 4, 22155, 2001. [15] G.P. Dwyer, P. Locke, W. Yu, Index arbitrage and nonlinear dynamics between the S&P 500 futures and cash, Review of Financial Studies 9 (1996) 301332. [16] E. Elyasiani, P. Perera, T.N. Puri, Interdependence and dynamic linkages between stock markets of Sri Lanka and its trading partners, Journal of Multinational Financial Management 8 (1998) 89101. [17] R.F. Engle, F.K. Kroner, Multivariate simultaneous generalized ARCH, Econometric Theory 11 (1995) 122150. [18] C. Eun, S. Shim, International transmission of stock market movements, Journal of Financial and Quantitative Analysis 24 (1989) 241256. [19] J. Geweke, R. Meese, W. Dent, Comparing alternative tests of causality in temporal systems, Journal of Econometrics 21 (1983) 161194. [20] C.G. Gilmore, G.M. McManus, International portfolio diversication: US and Central European equity markets, Emerging Markets Review 3 (2002) 6983. [21] C.W.J. Granger, Investigating causal relations by econometric models and cross-spectral methods, Econometrica 37 (1969) 424438. [22] L. G nd z, A. Hatemi-J, Stock price and volume relation in emerging markets, Emerging Markets Finance and Trade 41 (2005) 2944. u u [23] Y. Hamao, R. Masulis, V. Ng, Correlation in price changes and volatility across international stock markets, Review of Financial Studies 3 (1990) 281307. [24] J.L. Harvill, B.K. Ray, A note on tests for nonlinearity in a vector time series, Biometrika 86 (1999) 728734. [25] C. Hiemstra, J.D. Jones, Testing for linear and nonlinear Granger causality in the stock price-volume relation, Journal of Finance 49 (1994) 16391664. [26] D.A. Hsieh, Testing of non-linear dependence in daily foreign exchange rates, Journal of Business 62 (1989) 339368. [27] D.A. Hsieh, Chaos and non-linear dynamics; application to nancial markets, Journal of Finance 5 (1991) 18391877. [28] D.M. Hunter, Linear and nonlinear dynamic linkages between emerging market ADRs and their underlying stocks, 2003. Downloadable as: SSRN-ID586542-code97395.pdf. [29] A. Kanas, Nonlinearity in the stock pricedividend relation, Journal of International Money and Finance 24 (2005) 583606. [30] D.M. Keenan, A Tukey nonadditivity-type test for time series nonlinearity, Biometrika 72 (1985) 3944. [31] M. King, S. Wadhwani, Transmission of volatility between stock markets, The Review of Financial Studies 3 (1990) 533. [32] A.G. Malliaris, J.L. Urrutia, The international crash of October 1987: Causality tests, Journal of Financial and Quantitative Analysis 27 (1992) 353364. [33] A.M. Masih, R. Masih, Dynamic linkages and the propagation mechanism driving major international stock markets: An analysis of the preand post-crash eras, Quarterly Review of Economic Finance 37 (1997) 859885. [34] A.M. Masih, R. Masih, Are Asian stock market uctuations due mainly to intra-regional contagion effect? Evidence based on Asian emerging stock markets, PacicBasin Finance Journal 7 (1999) 251282. [35] Y. Ma, A. Kanas, Testing for nonlinear relationship among fundamentals and exchange rates in the ERM, Journal of Financial Economics 19 (2000) 135152. [36] M. Najand, A causality test of the October crash of 1987: Evidence from Asian stock markets, Journal of Business Finance & Accounting 23 (1996) 439448. [37] P. Narayan, R. Smyth, M. Nandha, Interdependence and dynamic linkages between the emerging stock markets of South Asia, Accounting and Finance 44 (2004) 419439. [38] Z.A. Ozdemir, E. Cakan, Non-linear dynamic linkages in the international stock markets, Physica A 377 (2007) 173180.

2560

J.G. De Gooijer, S. Sivarajasingham / Physica A 387 (2008) 25472560

[39] K. Phylaktis, F. Ravazzolo, Stock market linkages in emerging markets: Implications for international portfolio diversication, 2003. Downloadable from: http://www.cass.city.ac.uk/emg/workingpapers/stock market linkages.pdf. [40] E. Pretorius, Economic determinants of emerging stock market interdependence, Emerging Markets Review 3 (2002) 84105. [41] E. Roca, Short-term and long-term price linkages between the equity markets of Australia and its major trading partners, Applied Financial Economics 9 (1999) 501511. [42] H.C. Sheng, A.H. Tu, A study of cointegration and variance decomposition among national equity indices before and after the period of the Asian nancial crisis, Journal of Multinational Financial Management 10 (2000) 345365. [43] P. Silvapulle, J.-S. Choi, Testing for linear and nonlinear Granger causality in the stock price-volume relation: Korean evidence, The Quarterly Review of Economics and Finance 39 (1999) 5976. [44] R. Smyth, M. Nandha, Bivariate causality between exchange rates and stock prices in South Asia, Applied Economics Letters 10 (2003) 699704. [45] Standard and Poors, Global Stock Markets Factbook, New York, 2005. [46] E. Weber, Volatility and causality in Asia Pacic nancial markets, SFB 649 Discussion paper 2007004, 2007. Downloadable from: http://sfb649.wiwi.hu-berlin.de. [47] C. Wu, Y. Su, Dynamic relations among international stock markets, International Review of Economics and Finance 7 (1998) 6384.