Findings show that the fluctuations in HNX-Index have no influence on VN-Index, yet, the fluctuations in VN-Index impact on the HNX-Index. This is to say, such the causality is unidirectional, from VN-Index to HNX-Index.
RESEARCHES & DISCUSSIONS By using Granger Causality Test, this study is to investigate whether there is the causality between VN-Index and HNX-Index The weekly time series data set of VN-Index and HNX-Index quoted from July 20, 2005 to Dec 30, 2009 shall be employed in this study Findings show that the fluctuations in HNX-Index have no influence on VN-Index; yet, the fluctuations in VN-Index impact on the HNX-Index This is to say, such the causality is unidirectional, from VN-Index to HNX-Index Keywords: VN-Index, HNX-Index, causality Introduction After 10 years, Vietnamese stock market has accomplished considerable achievements, greatly contributing to the national economic growth It has provided a channel of capital mobilization and rotation, meeting the need for capital of the whole economy Besides, it has also seen strong fluctuations in the two main indexes (i.e VN-Index and HNX-Index) In such a situation, the ever-most concerned question raised by investors is whether there is a causality between the VN-Index and HNX-Index In recent years, the causality in terms of price fluctuation among stock markets has attracted special considerations of many researchers Most results point out that the fluctuation in stock price of a market may affect others and vice versa (Huang et al., 2000; Zhu et al 2004; Floros, 2005; Fan et al., 2009) Such the research on this topic, although carried out in many markets in the world, has not ever been published for the case of Vietnam Therefore, the aim of this research is to define the causality in terms of price fluctuation between stocks listed in HOSE and those in Hanoi Stock Exchange (HNX) Literature review So far, many a study on price linkages among different stock markets has been published Due to incapably mentioning a full list of all studies on * Cần Thơ University ** BIDV Haäu Giang this topic, in this item, we just sest forth here several outstanding studies which lay a foundation for our research Huang et al (2000) have used the cointegration test to investigate the causality and cointegration between stock markets of the US, Japan and the Hong Kong-Shanghai-Taiwan Growth Triangle (Southern China Growth Triangle) The numerical data used for the research is daily indexes of Down Jones, Nikkei 225, Hang– Seng, Shanghai Composite, Shenzhen and Taiwan Weighted Volume quoted from Oct 2, 1992 to June 20, 1997 The test results show that there is no cointegration among stock markets of the US, Japan and the Southern China Growth Triangle However, the Granger causality test shows the price fluctuations of the US market have effect on Hong Kong’s and Taiwan’s but not on Shanghai’s Whereas, price changes in Japan market completely have no effect on that of the remainder Besides, the evidences of cointegration between Shanghai’s and Shenzhen’s have also been found In Latin America, Tabak et al (2002) tested the price linkage between stock markets of the US and that of Latin America Numerical data used in the research are a series of daily closing price of stocks, viz MERVAl (Argentina), IBOVESPA (Brazil), IBB (Colombia), IGPA (Chile), IPC (Mexico), IBC (Venezuela), IGBVL (Peru), and Down Jones (USA), quoted from January 1996 to April 2001 The results show that the causality between Economic Development Review - April 2011 47 RESEARCHES & DISCUSSIONS the US stock market and those in Latin America except for Mexico is unilateral The Mexico market maintains a bilateral causality with the US market In another research, Zhu et al (2004) also used both Granger causality test and cointegration test to explore the price linkages among stock markets of Shanghai, Shenzhen, and Hong Kong The research results show that there is a causality between Shanghai’s stock market and Shenzhen’s Besides, this research also indicates that the price changes in Hong Kong stock market will influence that in Shanghai’s Nonetheless, there is not any price linkage between stock markets of Hong Kong and Shenzhen Moreover, the cointegration test also shows that there exists no cointegration equation between these stock markets Floros (2005) uses the Granger causality test and cointegration test to study the price linkages among stocks markets of the US, Japan and UK Numerical data used for this research are daily indices of the US’s S&P 500, Japan’s NIKKEI 250 and UK’s FTSE 100 listed from September 1988 to September 2003 This research shows that there is a bidirectional causality between NIKKEI 225 and FTSE 100 and a unidirectional causality between S&P 500 and FTSE 100 and even NIKKEI 225 In addition, the cointegration test shows a cointegration between these markets as proven by the existence of a cointegration equation with the significance at 5% By means of Granger causality test, Ozdemir et al (2007) has found evidences of the bidirectional causality in terms of price fluctuations between the US stock market and UK one Moreover, the research results also figure out that price changes of US stock market influencing those of France’s and Japan’s is unilateral In this research, besides the Granger causality test, the cointegration test developed by Johansen is also employed to investigate the cointegration among these markets However, the test results point out a fact that there is no cointegration between the US stock market and those of UK, Japan and France Recently, Fan et al (2009) have tested the price linkage between China’s stock market and those of the US, UK, Japan and Hong Kong as from Jan 1, 1999 to Dec 26, 2008 This research provides with evidences of the cointegration be- 48 Economic Development Review - April 2011 tween the Shanghai stock market of China and those of the US, UK, Japan and Hong Kong Data and research methods a Numerical data: The numerical data used in this research are the weekly time series data of VN-Index and HNX-Index (formerly HASTC-Index) quoted from July 20, 2005 to Dec 30, 2009 on each Wednesday’s closing If there is not any transaction on a certain Wednesday of a certain week, the index of Thursday’s closing (or that of Tuesday’s closing in case of no transaction in Thursday) will be chosen for a replacement If there is even no transaction on Tuesday or Thursday, stocks indexes of that week will be ignored and treated as lack of information The fact that the indexes on each Wednesday’s closing are chosen to study is to avoid effects of transactions conducted on weekends (Huber, 1997) Based on closing indexes, changes in the market yield (i) of the two consecutive weeks will be measured as follows: Ri,t = log(pi,t) - log(pi,t-1) = log(pi,t / pi,t-1) Where, Ri,t: Changes of market yield i within two consecutive weeks pi,t: The market yield i on Wednesday’s closing of week t pi,t-1: The market yield i on Wednesday’s closing of week t-1 b Research methods: In order to study the linkage between changes in VN-Index and HNX-Index, Granger causality test will be run in this research The Granger test requires a stationary time series Therefore, before conducting Granger test, a unit root test must be conducted to test whether a time series is stationary or not Unit root test The unit root test can be conducted by using the Dickey–Fuller test (DF), or Augmented Dickey-Fuller test (ADF), or Phillips-Person test (PP) Yet, in this research, ADF alone will be employed The ADF test model can be written as follows: + et (1) Dyt = a0 + byt-1+ Dyt = a0 + dt + byt-1+ + et (2) Model (2) is different from (1) in that it is included with a trend variable dt Signs in (1) and RESEARCHES & DISCUSSIONS (2) are explained as follow: D = yt - yt-1 yt: time series data considered k: lag time et: white noise Due to the fact that the results of ADF test are very sensitive to choice of the lag time (k), Akaike Information Criterion (AIC) will be employed to select the optimal k for ADF model (i.e k will be chosen so that AIC is minimum) The hypothesis H0 in ADF test is that there exists a root unit (b=0) which will be rejected if its critical value is smaller than ADF test value Yet, the ADF test value follows the non-normal distribution; therefore the critical value is based on the available values made by Mackinnon (1991) Comparing ADF test value with critical values of Mackinnon, it will be an easy task to conclude whether the time series is stationary or not Granger causality test The Granger test used in this research is to ponder whether changes in X influence changes in Y and vice versa The regression equation in Granger test can be written as follows: Z k k ]]Yt = a0 + / bl Yt - l + / dl Xt - l + ft l=1 l=1 [ k k ]] Xt = a1 + / zl Xt - l + / tl Yt - l + ot l=1 l=1 \ In this research, X and Y represent VN-Index and HNX-Index respectively In Granger test, the lag time (k) is also chosen according to AIC - If dl is not equal to zero and has the statistical significance and dl without, then it can be concluded that X is the reason of fluctuations in Y (unidirectional causality) - If dl has no statistical significance and but rl is statistically significant and not equal to zero, then it can be concluded that X is influenced by changes in Y (unidirectional causality) - If both dl and rl are statistically significant and not equal to zero, then it can be concluded that there is a mutual influence between X and Y (bidirectional causality) - If both dl and rl are not statistically significant, then it can be concluded that X and Y are independent from each other a Unit root test results: As mentioned above, before carrying out Granger test, it is oblidged to conduct the unit root test with a view to examining whether the observed time series is stationary In case such the time series are non-stationary, the unit root test will be in turn carried out with the first difference of the time series (rt) The results of ADF unit root test with and without time trend variable are shown in Table Table 1: The results of ADF unit root test Data series Without time trend With time trend VN-Index and HNX-Index VN-Index (k=1) -0,27 -1,55 HNX-Index (k=1) -0,65 -2,06 -11,69a -11,74a -6,26a -6,26a The first difference of VNIndex and HNX-Index Changes in VN-Index (k=0) Changes in HNX-Index (k=0) has statistical significance at 1% The result of ADF test shows that the hypothesis H0 about the non-stationariness of VN-Index and HNX-Index series cannot be rejected because its value is smaller than Mackinnon’s critical value However, when the first difference of time series (changes registered in two consecutive weeks) is tested, H0 is rejected with the statistical significance at 1% This allows us to draw a conclusion that the first difference of time series for VN-Index and HNX-Index is stationary This can be meant these two series meet the requirements of Granger test In order to run Granger test, besides testing the stationariness of the series, it is a must to determine whether the lag time (k) is suitable for variables in the model In this research, the most appropriate value of k for Granger model is b Granger test results: Based on the results of ADF unit root test and AIC, Granger test is conducted to determine the causality between VN-Index and HNX-Index The result of Granger test is presented in Table a Findings Economic Development Review - April 2011 49 RESEARCHES & DISCUSSIONS Table 2: Granger test result Hypotheses (H0) F-test Changes in VN-Index have no influence on HNX-Index 2,72b Changes in HNXIndex have no influence on VN-Index 1,69 b Lag time (k) Conclusion H0 rejected H0 accepted retaining the statistical significance at 5% Table shows that “changes in HNX-Index have no influence on VN-Index” cannot be rejected By accepting H0, it is meant that fluctuations in HNX-Index not influence VN-Index Yet, the hypothesis “changes in VN-Index have no influence on HNX-Index” is rejected with the statistical significance at 5%; or in other words, its null hypothesis is accepted In short, the causality in terms of price fluctuations between VN-Index and HNX-Index is unidirectional from VN-Index to HNX-Index In other words, changes of VNIndex are the reason for changes in HNX-Index The reason is that HOSE is the largest trading floor in Vietnam, consequently, and influential in trends of the Vietnam’s stock market Thus, changes of VN-Index have certain effects on HNXIndex (with same direction) Vice versa, because of the small scale of the Hanoi Stock Exchange, changes in HNX-Index are not strong enough to influence VN-Index Besides, according to many stock experts, the net buying and selling from foreign investors have strong effect on fluctuations in VN-Index due to the fact that they often attend much to blue chips in HOSE Additionally, being exacerbated by the herd behavior, such the fluctuations also cast impacts on HNX-Index with a certain lag time Via what we have analysed so far, it is possible to sum up that present fluctuations in VN-Index is a salient indicator for the future value of HNXIndex The findings of this research have a significant meaning to investors, especially individual investors, because this is the foundation for them to make a wiser decision prior to investing in stocks listed in HNX c Effects of changes of VN-Index on HNXIndex: regression analysis the results: Based on the above Granger test results the regression analysis shall be run to identify impacts 50 Economic Development Review - April 2011 of VN-Index on HNX-Index with different lag times (k=4) A regression equation can be written as follows: Where, Yt: Changes in HNX-Index at the time t Xt: Changes in VN-Index at the time t k: lag time The regression analysis results presented in Table shows that changes in HNX-Index at the time t are directly proportional to changes in HNX-Index three weeks later This means if the current HNX-Index is increasing, it will continuously increase three weeks later and vice versa More specifically, if current HNX-Index is increasing by 1%, it will continue to increase by 0.23% in the third week to come Such the relation retains a statistical significance at 5% Moreover, Table shows that changes in HNXIndex has a directly proportional relation to VNIndex with k=2 This means when VN-Index increases or decreases at the week t, then HNXIndex will increase or decrease accordingly two weeks later In details, when VN-Index increases by 1% at the week t, HNX-Index will increase by 0.43% at the week t+2 This relation is statistically significant at 1% However, changes in VNIndex have no effect on HNX-Index with k=1, k=3 and k=4 (the correlation coefficient has no statistical significance) In other words, changes in VNIndex at week t will just influence HNX-Index in the week t+1 but not in other weeks Y(-1), Y(-2), Y(-3), Y(-4): Changes of HNX-Index with k = 1, 2, 3, respectively X(-1), X(-2), X(-3), X(-4): Changes of VN-Index with k = 1, 2, 3, respectively t-test value is presented in parentheses Conclusion This research investigates the price linkage of stocks listed in HOSE and HNX from July 20, 2005 to Dec 30, 2009 The Granger test results show that changes in VN-Index influence HNXIndex However, changes in HNX-Index have no influence on VN-Index Accordingly, it is possible to conclude that the causality between VN-Index and HNX-Index is unidirectional from VN-Index to HNX-Index In details, regression analysis re- RESEARCHES & DISCUSSIONS Table 3: Effects of changes in VN-Index on HNX-Index Variables Constant (a) Y(-1) Y(-2) Y(-3) Y(-4) X(-1) X(-2) X(-3) X(-4) Number of observed times F-value Estimate coefficients 0.0002 -0.11 0.0093 -0.08 -0.3035 (-0.26) 0.2329 (2.00)b 0.0568 -0.48 0.1574 -1.08 0.429 (2.92)a -0.0949 (-0.63) 0.0501 -0.34 225 3,30a : Having statistical significance at 1% and 5% respectively a,b sults indicate that if VN-Index increases by 1% at the week t, HNX-Index will increase by 0.43% in the week t+2 with a statistic significance at 1% Yet, this research still has its limitations Firstly, this research has not excluded the disparity in price margin and liquidity of two stock exchanges, which can affect the relation of those two indexes Secondly, in order to leave out influences of transactions conducted on the first days and weekends of a week, weekly time series data are employed in this research However, by observing fluctuations in VN-Index and HNX-Index in recent time, it is supposed that the causality between these two indexes can occur with a lag time of several transactions Therefore, further studies may investigate this relation with daily time series data Thirdly, during the development of VN- Index and HNX-Index, the causality of these two indexes may vary; yet, this point has not been counted in this research Therefore, testing the causality between VN-Index and HNX-Index for each period of their development will be an interesting topic for further studies in the futuren References Fan, Kui; Z Lu & S Wang (2009), “Dynamic Linkages between the China and International Stock Markets”, Asia-Pacific Financial Markets 16, p.211-230 Floros, Christos (2005), “Price Linkages between the US, Japan and UK Stock Markets”, Financial Markets and Portfolio Management, 19 (2), p.169-178 Huang Bwo-Nung, Yang Chin-Wei & John WeiShan Hu, (2000), “Causality and Cointegration of Stock Markets among the United States, Japan, and the South China Growth Triangle”, International Review of Financial Analysis, 9(3), p.281-297 Huber, Peter (1997), “Stock Market Returns in Thin markets: Evidence from the Vienna Stock Exchange”, Applied Financial Economics 7, p.493-498 MacKinnon, J.G (1991), “Critical Value for Cointegration Tests” in Long- Run Economic Relationships: Readings in Cointegration, ed R.F Engle and C.W.J Granger, Oxford: Oxford University Press, p.267-276 Ozdemir, Z A & E Cakan (2007), “Non-Linear Dynamic Linkages in the International Stock Markets”, Physical A: Statistical Mechanics and its Applications, 377(1), p.173-180 Tabak, B M & E J A Lima (2002), “Causality and Cointegration in Stock Markets: the Case of Latin America”, Working Paper Series 56, p.1-28 Zhu, Hongquan, Z Lu & S Wang (2004), “Causal Linkages among Shanghai, Shenzhen, and Hong Kong Stock Markets”, International Journal of Theoretical and Applied Finance, (2), p.135-149 Economic Development Review - April 2011 51 ... bidirectional causality between NIKKEI 225 and FTSE 100 and a unidirectional causality between S&P 500 and FTSE 100 and even NIKKEI 225 In addition, the cointegration test shows a cointegration between. .. causality between VN-Index and HNX-Index is unidirectional from VN-Index to HNX-Index In details, regression analysis re- RESEARCHES & DISCUSSIONS Table 3: Effects of changes in VN-Index on HNX-Index. .. of VN- Index and HNX-Index, the causality of these two indexes may vary; yet, this point has not been counted in this research Therefore, testing the causality between VN-Index and HNX-Index for