CAUSAL LINKAGES AMONG SHANGHAI, SHENZHEN, AND HONG KONG STOCK MARKETS

Transcription

1 International Journal of Theoretical and Applied Finance Vol. 7, No. 2 (2004) c World Scientific Publishing Company CAUSAL LINKAGES AMONG SHANGHAI, SHENZHEN, AND HONG KONG STOCK MARKETS HONGQUAN ZHU, ZUDI LU and SHOUYANG WANG Institute of Systems Science, Academy of Mathematics and System Sciences, Chinese Academy of Sciences, Beijing , China zhu hq@yahoo.com.cn zdlu@mail.iss.ac.cn swang@mail.iss.ac.cn ABDOL S. SOOFI Department of Economics and Office of Institutional Research University of Wisconsin, Platteville, WI 53818, USA soofi@uwplatt.edu. Received 9 May 2003 Accepted 14 October 2003 In this paper, we test for causal relationship between China s stock markets by using returns and a measure of volatility for the Shanghai Composite index, the Shenzhen Composite Subindex, and the Hong Kong Hang Seng Index. We also show that the stock index series are nonstationary and that cointegrating vectors and error correction models do not exist for the series. Based on these tests, for the return series, we conclude that Shenzhen Granger caused Shanghai before For the volatility data, we find that there exists a positive feedback relationship between Shanghai and Shenzhen stock markets, and that Hong Kong volatility Granger causes Shanghai volatility, but not vice versa. Keywords: Financial integration; volatility; Granger causality; Shanghai stock market; Shenzhen stock market. JEL classification code: G10; G15; G20 1. Introduction The rapid cross-country transmission of financial crisis is the subject of intensive research in recent years (e.g. Sola et al. [21]; Brooks and Henry [1]; Van Rijckeghem and Weder [26]). The interest in the subject is due to the massive welfare losses and economic dislocations that are caused by successive devaluations of currencies, collapse of the stock markets, and the ensuing economic depression in countries that experience financial crisis. Given the volatility of international financial markets, the accession of China to the World Trade Organization (WTO), opening up of China s financial industry to international competitions, and the on-going capital account liberalization efforts in that country give understanding the causal relationships 135

2 136 H. Zhu et al. and transmission mechanism, if any, among three Chinese equity markets (Hong Kong, Shanghai, and Shenzhen) strong policy impetus. Understanding of the transmission mechanism among the stock markets in the Mainland and Hong Kong is interesting both for academic researchers and for investment professionals. The choice of Hong Kong for inclusion in this study is due to the following considerations. Although Hong Kong SAR is politically unified with the P.R. of China, its economy is structurally different from the economy of the Mainland. In addition to having a very open economy, Hong Kong has had strong economic linkages with Mainland China. Apparently, the linkages have been reinforced since China began the economic reforms(chan et al. [2]). Moreover, as a major financial center, Hong Kong has been an important source of funds for the Mainland China s state-owned enterprises. Furthermore, the trading of the B-shares in Shenzhen stock exchange is settled in HK dollars, and Hong Kong is the closest fully developed market economy to the Mainland. The main purpose of this paper is to test for the presence of causal linkages between the stock markets in Shanghai (SHSM), Shenzhen (SZSM), and Hong Kong (HKSM). We will examine the relationships both from returns and volatility perspectives. The paper is organized as follows. Section 2 provides some background on the institutional nature of the stock markets in China. Section 3 gives a brief literature review of the empirical causality tests among the Asian stock markets and discusses the methods of testing for causality in the framework of cointegrating series and vector error correction models. In Sec. 4, we discuss the data. Section 5 introduces the theory of Granger causality. In Sec. 6, we present the empirical results, and finally Sec. 7 concludes the study. 2. Institutional Arrangement of the Stock Markets in China The SHSM and SZSM, though founded only ten years ago, have experienced rapid growth and have played very important roles in the economy of Mainland China. With the impressive economic growth of China over the last two decades and her entry into the WTO, it is expected that these two markets would develop faster and potentially could become two highly prominent stock markets in Asia and the Pacific Rim. By the end of 2000, the combined market capitalization of these two markets had become equal to that of HKSM totaling RMB 4,809 billion a (about USD 580 billion). The market capitalization in these two markets in 2001 was RMB 4,352 billion (about USD 527 billion), about 2400 times larger than what it was in The number of listed companies in both SHSM and SZSM increased from 14 to 1160, a RMB is the abbreviation for Renminbi, the Chinese currency. The exchange rate of RMB to US dollar was about 8.25 in early 2004.

3 Causal Linkages Among Shanghai, Shenzhen, and Hong Kong Stock Markets 137 and the number of individual investors exceeded 66 million by the end of The reader is also referred to Cheng (2000) for the background of the two markets. b The Stock Exchange of Hong Kong Ltd. was founded on April 2, The main index used in HKSM is Hang Seng Index, composed of 33 major stocks. By the end of 2001, there were 756 companies listed on the Main Board with a total market capitalization of HKD 3,885 billion, and the number of market participants (firms that pay transaction levy and trading fees to the exchange) was 507. The Growth Enterprize Market (GEM) commenced operations on 15 November 1999 to provide capital formation facilities for growth companies that are not listed on the Main Board. Up to the end of 2001, 111 companies were listed on the GEM with a total market value of HKD 61.0 billion. Hong Kong, with a total market capitalization of HKD 3,946 billion (Main Board and GEM) at the end of 2001, was the 10th largest stock exchange in the world and the second in Asia. c 3. Empirical Linear Causality Tests of Stock Markets We use Granger causality test to determine the causal relationship among the China s stock markets. The empirical works on testing for causality among stock markets include Masih and Masih [16] that examined the dynamic linkages among stock prices of six major stock markets before and after globalization of financial markets and found evidence of causal relationships among the markets both before and after globalization. In an earlier study Masih and Masih [15] also find significant interdependencies between the stock markets of OECD and the emerging Asian stock markets. Tay and Zhu [24] examined the correlations in returns and volatilities in the Pacific- Rim stock markets. They noticed that the information resulting in volatility in one market seems to be transmitted more rapidly to the other markets located closer geographically and to the markets that are efficiently organized and open. Liu et al. [14] provided evidence of a bi-directional causality between Shanghai and Shenzhen stock markets. Chowdhury [5] studied the interdependencies of the Asian Newly Industrialized Economies stock markets, and found that the Hong Kong and Singapore markets are significantly linked with the Japanese and the U.S. markets, while the Korea and Taiwan markets are less responsive to outside innovations. The development of a linear causality test by Granger [7] was popularized by Sims [20], and was further developed by Granger [8]. Soon it was realized that for systems containing variables with stochastic trends, the traditional inference procedures in causality tests that often involve Wald criteria become eminently more complicated. The use of Wald criteria creates complications because in the presence of variables with stochastic trends the limit theory involves nuisance parameters and b The data are from China Securities Regulatory Commission ( c The data are from the Stock Exchange of Hong Kong Ltd. (

4 138 H. Zhu et al. nonstandard distributions. Accordingly, these earlier works were further developed in the contexts of the time series econometrics of co-integrating systems, the general vector autoregressive (VAR) models, and error correction models (see Toda and Phillips [25], Engle and Granger [6]) Causality, co-integration, and vector error correction Since Granger causality test in the framework of a VAR model captures only the short-run temporal causality among the series, the presence of cointegrating relationship among the variables of interest, one could use a vector error correction(vec) model for test of short run and long run causal relationship among the variables (see Toda and Phillips [25]). A time series Y = {Y 1,..., Y n }, is said to be integrated of order d, signified as I(d), if it has a stationary, invertible autoregressive moving average ARIM A(p, d, q) representation after applying differencing operator (1 L) d. Here p is the order of the autoregression, d is the differencing parameter, and q is the order of the moving average. The series is fractionally integrated when d is not an integer. Define two I(d) series y 1t and y 2t as Y t = (y 1t, y 2t ). Generally, a linear combination of the series Z t = αy t (1) is also an I(d). However, if a vector α exists such that Z t is I(d b) with b > 0, then series y 1t and y 2t are co integrated of order (d, b) Johansen s method of testing for co-integration In this study we use Johansen s full-information maximum likelihood estimation method (Johansen [12, 13]) in determining whether any cointegrating vector exists for the stock markets. Let Y t denote an (n x 1) vector. We maintain the hypothesis that vector Y t follows a V AR(p) in levels, and write it Y t = ξ 1 Y (t 1) + ξ 2 Y (t 2) + + ξ p 1 Y (t p+1) + β + ξ 0 Y (t 1) + ζ t (2) where β is a contant and ζ t are the error terms. Given a sample of T + p observations and Gaussian disturbances ζ t, we select the parameters (Ω, ξ 1, ξ 2,..., ξ p 1, β, ξ 0 ) such that the log likelihood function of Y t = (Y 1, Y 2,..., Y T ), conditional on (Y p+1, Y p+2,..., Y 0 ), that is, L(Ω, ξ 1, ξ 2,..., ξ p 1, β, ξ 0 ) is maximized subject to the constraint that under the hypothesis of h cointegrating relations, only h separate linear combinations of the levels of Y (t 1) appear in (2). The null hypothesis H o is that there are exactly h cointegrating relations among the elements of Y t subject to the restriction stated in the previous paragraph. The alternative hypothesis H A is that there are n cointegrating relationships, where n

5 Causal Linkages Among Shanghai, Shenzhen, and Hong Kong Stock Markets 139 is the number of elements of Y t. One method of testing the hypothesis employs a likelihood ratio test of H o against H A which is L A L o = (T/2) n i=h+1 log(1 ˆλ i ) (3) where ˆλ i is the ith largest eigenvalue of the sample variance-covariance matrices of the ordinary least square residuals Error correction model and estimation of cointegration vector If it is established that the variables of Y t are cointegrated, then to show the dynamics of adjustment of the series one must estimate the restricted vector autoregressive model, or the VEC model. The VEC model as the error correcting model restricts the long-run behavior of the endogenous variables to converge to their cointegrating relationship but simultaneously allows presence of short-run dynamics (see Engle and Granger [6]). The fundamental idea behind the error correction model is that a portion of the disequilibrium from one period is corrected in the next period. In a two variables model, this translates into the change in one variable to past equilibrium errors and to the past changes in both variables. Formalizing these ideas we write the VEC model for the co-integrating vector in the present context as follows. Consider a VAR model in levels Y t = µ + Π 1 Y t Π k Y t k + η t, (4) where Y t and η t are (p x 1) matrices, η t N(0, σ 2 ), Π i s are (p x p) matrices, and µ is (p x 1) regression coefficients. If there exists at least one cointergating vector for the variables of Y t, the error-correction model for (4) can be written as Y t = µ + Γ 1 Y t Γ k 1 Y t k+1 + αβ Y t k + η t, (5) where α and β are (p x 1) matrices. The matrix β contains the h cointegration vectors and β Y t k represents h error correction terms measuring the deviations from the h long run equilibrium relationships. Note 1: The matrices Γ 1 through Γ k 1 capture the short-run dynamics of the model. Note 2: Each row of matrix α represents the speed of adjustment of an individual series from the h cointegrating relationships. If it is known that the series are I(1) with no cointegrating vectors, then causality tests can be performed using differences VAR s. In these tests the usual chi-square critical values are employed.

6 140 H. Zhu et al Method of Granger causality test To investigate further, we consider Granger causality analysis between the three markets. According to Granger [7], X causes Y if the past values of X can be used to predict Y more accurately than simply using the past values of Y only. In other words, if past values of X improve the prediction of Y, then we say that X Granger causes Y. Let {r (1) t }, {r (2) t }, {r (3) }, be the returns or volatility series of SHSM, SZSM, and HKSM respectively. To examine whether {r (j) t we define the following equation: r (i) t = a 0 + p l=1 t a (i) il r(i) t l + q l=1 } Granger causes {r (i) t }, b (j) il r (j) t l, i j, (6) where p, q are the lag length for series {r (i) t } and {r (j) t }, respectively. For the equation above, the null hypothesis of Granger causality is stated such that market j does not Granger cause market i, that is, all the coefficients b (j) il, l = 1, 2,..., q in Eq. (6) are zero. (Detailed discussions on the test can be found in Granger [7] and Patterson [17]). In general, one important practical problem in implementing Ganger causality tests is lag length selection. A suitable choice of the lag length is often important. In this paper we use Schwarz information criterion (SIC) [19] to choose an appropriate lag length. 4. The Data We will investigate the relationship among China s stock markets using the Shanghai Composite Index (SCI), the Shenzhen Sub-Component Index (SSI, this index has been recently renamed as Shenzhen Composite Subindex which is now used by Datastream and Shenzhen Stock Exchange), and the Hong Kong Hang Seng Index (HSI). We use the Shenzhen Composite Subindex instead of the the Shenzhen Composite Index because of the popularity of the former in China. As an example of popularity of this index, note that all financial news broadcasts in China announce the daily changes in the Shenzhen Composite Subindex. It should be pointed out that the B shares listed on SHSM and SZSM contain a very small portion of the trading stocks listed in both markets. We select these indices for the following reasons. First, the selected indexes are major indicators of the respective markets. Second, SCI, SSI and HSI are the most commonly used indices of SHSM, SZSM and HKSM, respectively. Third, these data sets are the most complete for these stock markets in comparison with other indices. SCI is a weighted average of the stock prices listed in the SHSM with the value as the weights. Dec. 19th, 1990 is the base for the index. SSI is a weighted stock price average of the selected companies listed in SZSM using the value as the weights and taking the index of April 4th, 1991 as the base.

7 Causal Linkages Among Shanghai, Shenzhen, and Hong Kong Stock Markets 141 In this study, we use daily returns, r t, defined as the difference of the natural logarithm of the daily closing price, that is, r t = ln P t ln P t 1, where P t and P t 1 are the closing prices of an index on days t and t 1, respectively. We choose the daily closing prices of the indexes from the beginning of 1993, to the end of 2001 as the whole sample. This choice follows from two considerations. First, before October of 1992 when China Securities Regulatory Commission was founded, there was no government commission to supervise SHSM and SZSM consistently, and SHSM and SZSM were operated separately. Second, the choice of the sample period is based on the high volatility of the markets before Before October, 1992, the two markets fluctuated heavily. The maximum change in the index in a single day was more than 0.7 and took place on May 21, 1992 in SHSM when the price constraints were removed. Some stock prices went up more than 500 percent in that trading day, and SCI jumped from 617 to But in the later five months SCI fell to 386 points. To gain more insights into the microstructure and the relationship of the three markets, we divide our chosen sample period into two sub-periods: from the first trading day of 1993 to the last trading day of 1994, and from the start of 1995 to the end of Such a division is due to the relative size of the two markets undergoing a major change. The SZSM, though larger and more active than the SHSM, was outsized by the SHSM by the end of 1994 owing to Chinese government s policy normalized price index HK SH SZ time Fig. 1. The normalized price indexes of the three stock markets.

8 142 H. Zhu et al. shift. Beginning with a similar size at the end of 1992, the market capitalization of SHSM increased to 4.5 times as large as that of SZSM at the end of d The normalized HSI, SCI and SSI indices from 1991 to 2001 are shown in Fig. 1. e 5. Empirical Results 5.1. Unit root and cointegration tests To determine whether the indexes are cointegrated, we first test for stationarity of the series. The results of the unit root test using Phillips and Perron [18] method including intercept, intercept and trend, or no intercept or trend for the whole considered period are shown in Table 1. We used Newey-West automatic truncation lag of 7 in these analyses. Accordingly, the series at level are nonstationary. The results for the two subperiods and are similar, which are not reported here to save space. Table 1. The unit root Phillips and Perron test statistic of the three stock indexes, using Phillips and Perron [18] method with Newey-West automatic truncation lag, 7. A. Price level: Series Shanghai Shenzhen Hong kong 5% critical value Cases Intercept Intercept and trend None B. Log Price level: Series Shanghai Shenzhen Hong kong 5% critical value Cases Intercept Intercept and trend None C. First difference Price : Series Shanghai Shenzhen Hong kong 5% critical value Cases Intercept Intercept and trend None D. First difference Log Price: Series Shanghai Shenzhen Hong kong 5% critical value Cases Intercept Intercept and trend None d Cheng [4] gives a detailed description of the microstructure of the Chinese stock markets. e We normalize all series by dividing all elements of each time series by the first observation of that series. This has the effect of making the first observation of each series equal to unity.

9 Causal Linkages Among Shanghai, Shenzhen, and Hong Kong Stock Markets 143 Table 2. Results of the co-integration testing of the three stock indices using Johansen [13] method. A. Index series: SHANGHAI, SHENZHEN, HONGKONG Lag p = 2 in (2) is chosen by Schwarz Criteria. Data Trend: None None Linear Linear Quadratic Cointegrating No Intercept Intercept Intercept Intercept Intercept Model: No Trend No Trend No Trend Trend Trend Result of L.R. Test: 5% critical value in parenthesis Rank = (24.31) (34.91) (29.68) (42.44) (34.55) Rank = (12.53) (19.96) (15.41) (25.32) (18.17) Rank = (3.84) (9.24) (3.67) (12.25) (3.74) Conclusion: Rank = 0 Rank = 0 Rank = 0 Rank = 0 Rank = 0 Schwarz Criteria by Model and Rank Rank = Rank = Rank = Rank = Conclusion: Rank = 0 Rank = 0 Rank = 0 Rank = 0 Rank = 0 B. Logarithmic index series: LOG(SHANGHAI), LOG(SHENZHEN), LOG(HONGKONG) Lag p = 2 in (2) is chosen by Schwarz Criteria. Data Trend: None None Linear Linear Quadratic Cointegrating No Intercept Intercept Intercept Intercept Intercept Model: No Trend No Trend No Trend Trend Trend Result of L.R. Test: 5% critical value in parenthesis Rank = (24.31) (34.91) (29.68) (42.44) (34.55) Rank = (12.53) (19.96) (15.41) (25.32) (18.17) Rank = (3.84) (9.24) (3.67) (12.25) (3.74) Conclusion: Rank = 0 Rank = 0 Rank = 0 Rank = 0 Rank = 1 Schwarz Criteria by Model and Rank Rank = Rank = Rank = Rank = Conclusion: Rank = 0 Rank = 0 Rank = 0 Rank = 0 Rank = 0

10 144 H. Zhu et al. Moreover, we used Johansen cointegration method testing for cointegration of three indices. The results of the LR test statistic, 2 (L A L 0 ) defined in (3), together with Schwarz criterion by model and rank are reported in Table 2, where by Schwarz criterion, the lag p = 2 in (2) is chosen for both panels A and B. Clearly, the models without trend are preferred according to the Schwarz criterion, for which both the LR statistic (at 5% level) and the Schwarz criterion indicate that there are no cointegrating vectors. The traditional cointegration tests are based on the notion that the degree of intergation of a series is an integer. However, two series could be cointegrated even if the differencing parameter of the residuals of the regression of the series is a fraction in the interval 0.5 < d < 0.5 [Granger and Joyeux [9], Soofi [22]]. To test for fractional cointegration, we used Hurvich and Deo [11] plug-in method to estimate the differencing parameters ds that are based on the optimal periodogram ordinate [Soofi and Payesteh [23]]. Based on this test and the results that are presented in Table 3, we cannot detect any evidence that the series are fractionally cointegrated either. All estimated d ± SE > Two-way Granger causality test of the returns Based on these findings, we proceed to use the traditional Granger causality method to test for two-way causal linkages among the stock markets. Though there are no significant weekend-day effects in China stock markets as that exists in U.S. and other stock markets, although Cheng et al. [3] found... negative returns (on Tuesday) after January 1, So we use calendar-adjusted returns in Granger causality test. The method used to remove calendar effects is the same as the one used in Hiemstra [10]. Table 3. The estimated differencing parameter, d, using Hurvich and Deo [11] method, for testing fractional co-integration of the series pair, based on the residuals of LS regression. In this and the following tables, SH, SZ and HK are the abbreviation of Shanghai, Shenzhen and Hong Kong stock markets respectively. A. Price level: Series pair (SH, SZ) (SZ, SH) (SZ, HK) (HK, SZ) (HK, SH) (SH, HK) Optimal m Estimated d Standard dev B. Log price level: Series pair (SH, SZ) (SZ, SH) (SZ, HK) (HK, SZ) (HK, SH) (SH, HK) Optimal m Estimated d Standard dev

11 Causal Linkages Among Shanghai, Shenzhen, and Hong Kong Stock Markets 145 Table 4 presents these results. Based on the P-values, we conclude that at a 5 percent or lower significance levels, there is no causal relationship among these three stock market returns with the exception that Shenzhen Granger causes (GC) Shanghai before Though the p-value of Shenzhen GC Shanghai in the entire period is less than 0.05, we believe this is mainly caused by the first sub-period s effect. One possible explanation for this observation is the relative activities in the markets. Before 1994, Shenzhen market was more active than Shanghai market, especially in 1991 and 1992, while after 1994, Shenzhen market s activities droped below that of Shanghai market owing to Chinese government s policy shift. See Sec. 4 in the above and Cheng [4] (the first two paragraphs of part two, and Table 1) who provides some information on these two market microstructures. Table 4. Linear Granger Causality of Stock Returns. This table reports the results of the linear Granger Causality test of stock returns. In this and the following tables, SH, SZ and HK are the abbreviation of Shanghai, Shenzhen and Hong Kong stock markets respectively. SIC is the Schwarz information criterion. For each panel we test whether the stock market return listed in the row Granger Causes the stock market return in the column. For example, in the first panel, we test whether Shanghai stock market return Granger Causes the returns of Shenzhen and Hong Kong stock markets. The data (1, 1) is the lag lengths set with SIC. The lag length of Shanghai is 1 and Shenzhen is 1 respectively. The F-statistics is 3.462, and the p-value is We remove the day-of-the-week effects of daily returns independently in the three sample periods. Panel A. Non-calendar-adjusted return: SH SZ HK SH SZ HK SH SZ HK SIC (1, 1) (1, 1) (1, 1) (1, 1) (1, 1) (1, 1) SH F-Stat Prob SIC (1, 13) (4, 9) (3, 1) (1, 1) (1, 1) (1, 1) SZ F-stat Prob SIC (1, 13) (1, 1) (1, 1) (1, 1) (1, 1) (1, 1) HK F-stat Prob Panel B. Calendar-adjusted return: SH SZ HK SH SZ HK SH SZ HK SIC (1, 1) (1, 1) (1, 1) (1, 1) (1, 1) (1, 1) SH F-stat Prob SIC (1, 13) (4, 9) (3, 1) (1, 1) (1, 1) (1, 1) SZ F-stat Prob SIC (1, 13) (1, 1) (1, 1) (1, 1) (1, 1) (1, 1) HK F-stat Prob

12 146 H. Zhu et al. The results that other stock market returns do not add significantly to the predictive power of a model to forecast the returns of a stock market index are consistent with common observation that stock market returns are difficult to predict Two-way causality test of volatility We perform causality tests for the volatility of the markets also. We define the volatility of an index as the absolute value of the first-difference of the index. The results are reported in Table 5. From Table 5 we conclude that there exists a significant positive feedback relationship between Shanghai and Shenzhen stock market volatility. Moreover, Hong Kong volatility Granger causes Shanghai volatility in the period before 1994 and also in the whole time period, but not vice versa. These are quite different from stock market returns and signify that the volatility shock in SHSM has a positive effect on volatility in SZSM, and vice versa. HKSM volatility does not GC SZSM volatility is not in line with the result of Tay and Zhu [24] that the information resulting in volatility in one market seems to be transmitted more rapidly to other markets located closer geographically, though SZSM is much closer geographically to HKSM than SHSM. Why does the volatility of the Hong Kong market leads the volatility of the Shanghai market but not the volatility of the Shenzhen market? We attribute this differential performance of the Hong Kong market to the listing of some Chinese companies in the Shanghai, Shenzhen, and Hong Kong markets. A total of 23 Chinese enterprises that issue both A and H shares before Of these enterprises, 17 are listed both at Shanghai and Hong Kong Stock Exchanges, and the rest six are listed both at Shenzhen and Hong Kong. Of the 17 enterprises, 10 were listed before All the six enterprises listed at Shenzhen and Hong Kong issued H shares after Table 5. Linear Granger Causality of Stock Volatility. This table reports the results of the linear Granger Causality test of stock volatility. The stock volatility is defined as the absolute value of the returns of the indexes. The symbols used in this table are the same as Table SH SZ HK SH SZ HK SH SZ HK SIC (2, 8) (1, 14) (2, 4) (1, 5) (3, 9) (1, 7) SH F-stat Prob SIC (2, 15) (1, 14) (2, 4) (1, 5) (2, 7) (2, 7) SZ F-stat Prob SIC (1, 20) (1, 8) (1, 4) (1,4) (2, 3) (1, 9) HK F-stat Prob

13 Causal Linkages Among Shanghai, Shenzhen, and Hong Kong Stock Markets 147 Before 1994, there are ten Mainland enterprises that issued both A and H stocks in Shanghai and Hong Kong stock markets, respectively, while none of enterprises issued both A and H stocks in Shenzhen and Hong Kong at that time. The activity of those H stocks in Hong Kong would influence the decision-making of the investors, and hence the activity, of the corresponding A stocks listed in Shanghai. This may be the reason why, in term of volatility, the Hong Kong market leads the Shanghai market but not the Shenzhen market before Summary and Conclusion Financial liberalization in Mainland China and recurring financial crises that have been experienced in all sundry countries in recent years, particularly in Asia during , give important policy impetus for an understanding whether there are powerful linkages between three stock markets in China. This study uses Granger causality test to determine whether there are any causal relationships, in the Granger sense, between the three markets. However, given that in the presence of stochastic trends in the series, inference about Granger causality becomes problematic, we test for the presence of cointegrating vectors for the stock market time series. The empirical results show that (1) all three stock market series are intergated processes, (2) the series are not cointegrated, and (3) therefore, the error correction model does not exist for the series. Accordingly, we proceed with the traditional approach for testing for the presence of Granger causal linkages among the markets, using two-way test of the returns and a volatility measure of the series. We perform causality tests for the returns and volatility of the markets also. Based on these tests, we conclude that at a 5 percent or lower significance level, there is not a causal relationship among these three stock market returns except Shenzhen Granger causes Shanghai before For the volatility data, we find that there exists a significant positive feedback relationship between Shanghai and Shenzhen stock market volatility, and Hong Kong s volatility Granger causes Shanghai s volatility before 1994 and in the whole sample, but not vice versa. The results based on this preliminary study have a pronounced policy implication. The volatility of the equity markets in Mainland China are linked to the global economy through the Hong Kong exchange. Empirical works by Brooks and Henry [1] indicate that financial crisis maybe nonlinearly transformed across economies. Even though, GC does not necessarily imply cause and effect relationship between the markets, nevertheless, the results point to strong connections between the Hong Kong market and the markets in the Mainland. Given such a linkage, the possibility of transmission of cross-country financial crisis and contagion to the Mainland is present. Further tests and analyzes, particularly with respect to whether the stock market series are nonlinear and for presence of nonlinear causal relationship among the series are required.

14 148 H. Zhu et al. Acknowledgments This work was supported by the National Natural Science Foundation of China ( , ) and Chinese Academy of Sciences ( ). We are grateful for helpful comments by an anonymous referee. References [1] C. Brooks and T. Henry, Linear and non-linear transmission of equity return volatility: Evidence from the US, Japan, and Australia, Economic Modeling 17 (2000) [2] W. S. Chan, W. C. Lo and S. H. Cheung, Return transmission among stock markets of greater China, Mathematics and Computers in Simulation 48 (1999) [3] G. M. Cheng, C. Y. Kwok and O. Rui, The day-of-the-week regularity in the stock markets of China, Journal of Multinational Financial Management 11 (2001) [4] K. X. Cheng, The microstructure of the Chinese stock market, China Economic Review 11 (2000) [5] A. R. Chowdhury, Stock market interdependencies: Evidence from the Asian NIEs, Journal of Macroeconomics 16 (1994) [6] R. Engle and C. W. J. Granger, Co-integration and error correction: Representation, estimation and testing, Econometrica 35 (1987) [7] C. W. J. Granger, Investigating causal relationships by econometrics models and cross-spectral methods, Econometrica 37 (1969) [8] C. W. J. Granger, Some recent developments in a concept of causality, Journal of Econometrics 39 (1988) [9] C. W. J. Granger and R. Joyeux, An introduction to long memory time series models and fractional differencing, Journal of Time Series Analysis 1 (1980) [10] C. Hiemstra and J. Jones, Testing for linear and nonlinear Granger causality in the stock price-volume relation, Journal of Finance 49 (1994) [11] C. Hurvich and R. Deo, Plug-in selection of the number of frequencies in regression estimates of the memory parameter of a long-memory time series, Journal of Time Series Analysis 20 (1999) [12] S. Johansen, Statistical analysis of co-integration vectors, Journal of Economic Dynamics and Control 12 (1988) [13] S. Johansen, Estimation and hypothesis testing of co-integration vectors in Gaussian vector autoregressive models. Econometrica 59 (1991) [14] X. M. Liu, H. Y. Song and P. Romilly, Are Chinese stock markets efficient? A cointegration and causality analysis, Applied Economics Letters 4, (1997) [15] A. M. Masih and R. Masih, Long and short term dynamic causal transmission among international stock markets, Journal of International Money and Finance 20 (2001) [16] A. M. Masih and R. Masih, Propagative causal price transmission among international stock markets: Evidence from the pre- and post globalization period, Global Finance Journal 13 (2002) [17] K. Patterson, An Introduction to Applied Econometrics: A Time Series Approach (Macmillan Press Ltd., London, 2000) p [18] P. C. B. Phillips and P. Perron, Testing for a unit root in time series regression, Biometrika 75 (1988)

15 Causal Linkages Among Shanghai, Shenzhen, and Hong Kong Stock Markets 149 [19] G. Schwarz, Estimating the dimension of a model, The Annals of Statistics 5 (1978) [20] C. Sims, Money, income and causality, American Economic Review 62 (1972) [21] M. Sola, F. Spagnolo and N. Spagnolo, A test for volatility spillovers, Economics Letters 76 (2002) [22] A. Soofi, A fractional co-integration test for purchasing power parity: the case of selected members of OPEC, Applied Financial Economics 8 (1998) [23] A. Soofi and S. Payesteh, ARFIMA modeling and persistence of shocks to the exchange rates: Does the optimal periodogram ordinate matter? Advanced Modeling and Optimization: An Electronic International Journal 4 (2002) [24] S. P. Tay and Z. Zhu, Correlations in returns and volatilities in Pacific-Rim stock markets, Open Economies Review 11 (2000) [25] H. Y. Toda and P. C. B. Phillips, Vector autoregressions and causality, Econometrica 61 (1991) [26] C. Van Rijckeghem and B. Weder, Sources of contagion: Finance or trade? IMF Working Paper (WP/99/146), (1999).

16