Abstract
This article uses linear and nonlinear Granger causality tests to study Granger causal relations among the stock markets of Greater China. In sharp contrast to the results disclosed by its linear counterpart, a nonlinear causality test provides evidence of isolated bi-directional causal relations between two Chinese stock markets and between the HKSE and the TWSE in the earlier period of 1992 to 1999. The nonlinear causality test further shows that, in the later period of 2000 to 2008, Granger causal relations have been strengthened and that Chinese stock markets are well connected with their neighbour markets, the HKSE and the TWSE, and they are playing a most influential role among the stock markets of Greater China.
Acknowledgements
The authors are grateful to an anonymous referee, Professor Tho Nguyen and Bartosz Gebka for their comments and help. The first author gratefully acknowledges financial support from University of Macau (research grant number RG003/08-09S/QZ/FBA).
Notes
1 In 2007, among all foreign countries and regions, Hong Kong and Taiwan were ranked as the third and fifth largest trading partners of China, respectively. China is the first ranked trading partner of both Hong Kong and Taiwan. The total export and import merchandise traded in the Greater China area ranked third in the world, while those of China, Hong Kong and Taiwan ranked third, twelfth, and seventeenth, respectively, in 2007.
2 In 2008 alone, there were 49 Chinese firms newly listed on the Hong Kong stock market, raising US$8.5 billion for China.
3 Examples of firms with ties to Taiwan listed on the Hong Kong stock exchange are Uni-President China, Tingyi and Want Want China. The main businesses of these three Hong Kong-listed companies operate in China. In addition, a few Hong Kong-listed companies have also invested in firms listed in China. For instance, the Hong Kong-listed Hang Seng Bank Ltd. and HSBC group acquired 15.98% and 19.9% shares of China Industrial Bank and Ping An Insurance in 2004 and 2005, respectively.
4 At the end of 2008, the stock markets of China, Hong Kong and Taiwan ranked sixth, seventh, and nineteenth in the world by market capitalization.
5A separate line of research has been advanced by examining the linkages among different shares (A and B shares) in Chinese stock markets. See, for example, Laurence et al. (Citation1997), Kim and Shin (Citation2000), Sjoo and Zhang (Citation2000), Tian and Wan (Citation2004), Brooks and Ragunathan (Citation2003), Chelley-Steeley and Qian (Citation2005), Shen et al. (Citation2007), Qiao et al. (Citation2008a) and Gebka (Citation2009).
6 Weekday effects are documented in the Greater China area. For example, McGuinness (Citation1997) finds a Friday effect in Hong Kong. Yakob et al. (Citation2005) find a Friday effect in Taiwan. Cai et al. (2006) find a Monday effect and a Tuesday effect in Chinese stock markets. Ogunc et al. (Citation2009) find weekend effects in the Shanghai A share market.
7 The Granger causality test examines the null hypothesis that or
for all i (i = 1, 2, … , m). If, for example, the null hypothesis that
for all i (i = 1, 2, … , m) is rejected, it implies that there is a Granger causality relationship running from market 2 to market 1.
8 The BDS test was suggested by Brock et al. (Citation1987) and revised by Brock et al. (Citation1996).
9 Diks and Panchenko (Citation2006) report that the HJ test may have an over-rejection bias when the sample size is very large. Their simulation results show that, under certain assumptions, when sample size is very big (e.g. 100 000), the rejection probability may tend to be one. However, the relatively much smaller sample size of this paper justifies our analysis.
10 The ADF test does not allow a structural break and the PP test allows only an exogenous structural break. Based on the referee's suggestion, we have also adopted the Zivot and Andrews (Citation1992) unit root test, which allows the endogenous structural break. We would like to thank the referee for pointing out the limitations of the ADF and PP unit root tests here.
11 To save space, the complete estimation results for the VAR and ECM–VAR are not reported here but are available upon request.
12 Based on the referee's suggestion, we have also adopted the Gregory and Hansen (Citation1996) test for cointegration with structural breaks. The cointegration test results are qualitatively the same for all pairs except for HKSE and TWSE in the second sub-period. However, both linear and nonlinear Granger causality test results for this pair are qualitatively unchanged. To save space, these results are not reported here but are available upon request.
13 Similar to the analyses in the last section, the Ljung–Box -test and the BDS test are performed on the associated residual series before the adoption of the nonlinear HJ test. The corresponding results reported in Panel B of and justify the appropriateness of adopting the nonlinear HJ test.