An Empirical Study on Market Segmentation and Information Diffusion in Chinese Stock Markets

(1)

An Empirical Study on

Market Segmentation and Information Diffusion in Chinese Stock Markets

Author: Chen Cao Supervisor: Johan Lyhagen

Master thesis in statistics Department of Statistics Uppsala University, Sweden

May, 2010

(2)

Abstract

The efficacy and accuracy of information is very important for making decision in stock markets. In this paper, we study on the effect of information diffusion in Chinese stock market before and after the owership release in February 19, 2001, by testing the stationary of A share premium and cointegration between A and B share prices. The panel unit root tests we propose on A share premium are Augmented Dickey-Fullar (ADF) tests for individual firm and Fisher tests for the panel, based on combining p- values from each individual cross-section. The panel cointegration tests on A and B shares we use is Johansen’s likelihood ratio tests for individual firm and likelihood- based panel cointegraion tests for panel, based on combining the test statistics. The results show that before the opening of B share markets to domestic investors, A share premiums have a unit root and there is no cointegration relationship between A and B share markets. On the contrary, after ownership release, A share premium is stationary and there is cointegration relationship between A and B share markets.

Key words: Information diffusion; Premium; A and B shares; Panel data

(3)

1. Introduction

The modern Chinese stock market was opened in the early 1990s. Chinese stock market has many unique characteristics. For instance, different shareholders such as enterprises, states and individual shareholders, trade shares with different purchasing costs and circulation regulation. Another major feature is that the market is segmented for domestic and foreign investors by the stock-ownership restrictions. From 1992, domestic investors can only trade A shares, while foreign investors can only trade B shares, so the shares of a firm will be traded at the same time, at the same stock exchange, but at different prices for different groups of investors. On February 19, 2001, B share markets are opened to both domestic and foreign investors, that is, domestic investors can trade both A and B shares, while foreign investors are still restricted to trade B shares only. Particular to Chinese stock market, A share prices always have a premium over B share prices. Since A share prices and B share prices of a firm are of the same time and stock exchange, there exists interdependence between these two prices.

The effect of information flow is quite important for two investor groups making decision. The existence of information flow between domestic and foreign investors is understandable due to the following factors. First, domestic investors have information advantage. They can be easier and faster to get information from local sources, like local news and firm press releases. Second, the foreign investors in Chinese stock market are always large institutional investors. Compared with domestic investors, they have more experience since they not only participate in Chinese stock market, but also other stock markets. Their reactions on the information are faster and more accurate.

They also have more advanced analysis technology.

Several early studies have worked on the market segmentation and information diffusion in Chinese stock market. Chakravarty, Sarkar and Wu (1998) develop a model to explain the pricing of A shares and B shares, by incorporating both asymmetric information and market segmentation. They argue that the reason for A share premiums is that domestic investors have more information then foreign investors. By a cross- sectional regression, they also show that B share discounts are significantly related to a model-based proxy for information asymmetry.

Sjöö and Zhang (2000), study on the information diffusion between Chinese A and B shares. They used Johansen’s likelihood cointegration tests and found that information flows from foreign investors to domestic investors in the Shanghai stock exchange while in the opposite way in less liquid and smaller Shenzhen stock exchange. Another

(5)

conclusion is the most important factor that determines the direction of information flows is the choice of stock exchange.

Ahlgren, Sjöö and Zhang (2003) proposed panel data tests when testing unit root and cointegration on A and B share prices. Their panel data tests are based on Fisher (1932) tests introduced by Maddala and Wu (1999). Their empirical result is that A share premium is stationary and there exists cointegration relationship between A and B shares for most firms. They also do a probit analysis of the individual firms and show that cointegration exists with higher probabilities for firms in the service and manufacturing sectors and listing B shares in more recent years.

In this study, based on the idea of Ahlgren, Sjöö and Zhang (2003), we are going to check the effect of information flow before and after the ownership release in 2001, by testing the stationarity of A share premiums and cointegration between A and B share prices. When testing the stationarity of A share premiums, we use panel data unit root tests. And we are going to test the cointegration by panel data cointegration tests. If A share premiums are stationary and A and B share prices are cointegrated, we can make the conclusion that information flows effectively between the two investment groups.

The data we use include 40 firms listing on both A and B stock markets, 20 from SHSE and the other 20 from SZSE. The sample period is from January 1995 to December 2005. The sample period includes five years before and after the ownership release. The panel data tests we use allow us to pool the information from each individual firm. The panel data unit root tests we use are standard Augmented Dickey-Fuller (ADF) tests and Fisher (1932) tests introduced by Maddala and Wu (1999). The Fisher tests are based on combining the p-values from individual cross-section. The panel data cointegration tests we use are Johansen (1995)’s likelihood ration tests and likelihood-based cointegration tests suggested by Larsson, Lyhagen & Löthgren (2001). The likelihood-based cointegration tests are based on pooling the statistics from individual cross-sectional unit. The first advantage of these tests for panel data is that they allow for unbalanced panel data we use. The second advantage is that they solve the problem with heteroscedasticity in the stock data. But the tests require independence between firms.

The paper is organized as follows. Section 2 introduces the general background of Chinese stock market and the structure of A and B share markets. Section 3 gives a description on the data. Section 4 discusses the several reasons for share prices differences. Section 5 discusses academic basis and Section 6 shows the empirical study on the Chinese stock markets using the methods we present in Section 5. Finally, Section 7 gives the conclusion.

(6)

2. Background and Structure of A and B share markets

There are two main stock exchanges in China: Shanghai Stock Exchange (SHSE) and Shenzhen Stock Exchange (SZSE). Shanghai Stock Exchange has a long history and was first opened in the late 1860s. After shutting down several times for different reasons including the war, SHSE was ultimately reopened in 1990. One year later, China opened the second stock exchange in Shenzhen for the increasing in technology and government securities. In order to attract international investors to participate in Chinese stock market, B share markets were opened in both SHSE and SZSE in 1992.

The amount of outstanding B shares of a firm is restricted to be small, so foreign investors can only hold a small part of shares. As a result, China can absorb international capitalization, and at the same time, assure the ownership under Chinese control. Both two exchanges are open for trading from Monday to Friday, with daily price change limits of 10%. The number of listed companies rocketed from 10 in 1990, to over 1800 by April of 2010. Simultaneously, the market capitalization had a one- hundred-fold increase from less than 10 billion to more than 3210 billion RMB. These numbers show that Chinese stock market has become one of the largest stock markets in Asia.

In Chinese stock market, a listed company can issue two classes of shares: A shares and B shares. A shares are valued by Yuan Renminbi (RMB) in both stock exchanges, however, B shares are valued by US dollars in SHSE whereas by Hong Kong dollars (HKD) in SZSE. In the total 1821 listed companies, 1714 companies issues only A shares, 24 companies issues only B shares, 83 companies issued both A shares and B shares by April of 2010. From 1992 to 2001, A shares were only available to domestic investors, while B shares could be only traded by foreign investors, which made the two share markets totally segmented. And the two investor groups can only exchange their information by the observed prices, but not by arbitrage. Now, the structure of Chinese equity markets has been changed since the ownership restriction has been released. On February 19, 2001, B share markets are opened to domestic investors, but foreign investors can still only trade B shares. This policy makes the A and B share markets partially merge. However, since domestic currency, Renminbi, is not freely convertible, arbitrage is still restricted. At the same time, domestic investors cannot freely exchange Renminbi for foreign money. Moreover, a domestic investor has to open an account only for trading B shares. And the foreign exchanges in that account have to be

(7)

transferred from a foreign bank. For these reasons mentioned above, we can still regard A and B share markets as segmented markets.

3. Data description

In this study, we are doing the research on the information diffusion between A and B share markets, so we only focus on the firms that are listing both A and B shares in SHSE and SZSE. So we choose a sample of 40 daily listed firms, 20 of which are from SHSE and the remaining 20 are from SZSE. The data are monthly closing prices for this section and weekly closing prices for the empirical study in Section 6. Monthly closing prices are the prices on the last trading day of every month. Similarly, weekly closing prices are the prices on the last trading day of every week. The sample period is from January 1995 to December 2005. All B shares prices are converted to Chinese RMB.

More information about the 40 firms included in this study is shown in Appendix Table A1. The reason why we use monthly and weekly closing prices rather than the monthly and weekly average prices in this study is that they are easily collected and can still directly present the trend of the prices in a long period.

Figure 1 and Figure 2 present the trend of A and B shares prices in Shanghai and Shenzhen stock markets. Unweighted indices PSH_At, PSH_Bt, PSZ_At and PSZ_Bt are respectively computed by the average of the monthly closing prices of 20 firms in Shanghai and Shenzhen stock markets in month t.

From Figure 1 and 2 we can directly find that there exist A share premiums over B share in both Shanghai and Shenzhen stock markets. As mentioned in Chakravarty, Sarkar and Wu (1998), in other stock markets, which are also segmented by domestic and foreign investors, foreign shares always trade with premiums; while in Chinese stock market quite differently, there exist obvious A share premiums. They also argue that the lack information in foreign investors causes the premiums in A shares.

We also denote indices PSH_ABt and PSZ_ABtrespectively as the percentage A share premiums over B shares prices of Shanghai and Shenzhen markets, which are computed by

At Bt

ABt

Bt

PSH PSH

= − (1)

(8)

Figure 1. Average of 20 firms’ monthly closing prices in Shanghai stock market, respectively in A and B share markets. The sample period is from January, 1995 to December,2005.

Figure 2. Average of 20 firms’ monthly closing prices in Shenzhen stock market, respectively in A and B share markets. The sample period is from January, 1995 to December,2005.

0 2 4 6 8 10 12 14 16 18 20

1995/01/27 1995/07/31 1996/01/31 1996/07/31 1997/05/30 1997/11/28 1998/05/29 1998/11/30 1999/05/31 1999/11/30 2000/05/31 2000/11/30 2001/05/31 2001/11/30 2002/05/31 2002/11/29 2003/05/30 2003/11/28 2004/05/31 2004/11/30 2005/05/31 2005/11/30

SHSE A share prices SHSE B share prices

0 2 4 6 8 10 12 14 16 18

1995/01/27 1995/07/31 1996/01/31 1996/07/31 1997/05/30 1997/11/28 1998/05/29 1998/11/30 1999/05/28 1999/11/30 2000/05/31 2000/11/30 2001/05/31 2001/11/30 2002/05/30 2002/11/29 2003/05/30 2003/11/27 2004/05/31 2004/11/30 2005/05/31 2005/11/30

SZSE A share prices SZSE B share prices

(9)

At Bt ABt

Bt

PSZ PSZ

= − (2)

Figure 3 displays the PSH_ABtand PSZ_ABt. From Figure 3, we can directly find that after the ownership release in February 2001, the indices seem to become stationary and in the latter section, we will check it by panel unit root tests. Sjöö and Zhang (2003) indicated unweighted indices cause problems with outliers and heteroscedasticity problems. So we choose panel data to solve this problem.

Figure 3. Average of percentage of A share premiums over B share prices in Shanghai (SHPA, 20 firms) and Shenzhen (SZPA, 20 firms) stock markets.

4. Share price differences and segmented investment groups

In this section, we discuss the causations of the observed A share premium. First we introduce two well known securities pricing models: capital asset pricing model and discounted dividend model. Then we explain the causations of the A share premium based on these models.

Capital asset pricing model

0 1 2 3 4 5 6 7 8

1995/01/27 1995/06/30 1995/11/30 1996/04/30 1996/09/27 1997/06/27 1997/11/28 1998/04/30 1998/09/30 1999/02/09 1999/07/30 1999/12/30 2000/05/31 2000/10/31 2001/03/30 2001/08/31 2002/01/31 2002/06/28 2002/11/29 2003/04/30 2003/09/30 2004/02/27 2004/07/30 2004/12/31 2005/05/31 2005/10/31

SHPA SZPA

(10)

The capital asset pricing model (CAPM) describes the relationship between risk and expected rate of return. It is used in pricing the risky securities and was independently introduced by Treynor (1961, 1962), Sharpe (1964), Lintner (1965) and Mossin (1966).

We consider the extended CAPM modeled as ( _i) _f _i( ( _m) _f)

E R =R +β E R −R + (3) ω where:

( )E R is the expected rate of return _i

R is the risk-free rate of return f

βiis Beta of security¹ ( _m)

E R is the expected rate of return ( _m) _f

E R −R is the risk premium

ωis the excess return caused by liquidity cost.

Discounted dividend model [14][15]

The discounted dividend model (DDM) is a financial tool of evaluating the stock prices based on the net present value of the future dividends. The model is

, 0, ,

, 1

( )

(1 ) (1 )

N

t i i N i

o i t N

t i i

E D P

P = r r

= Γ +

+ +

∑

(4)

where:

( _{t i},)

E D is the expected dividend of firm i

,

P is current stock price of firm o i i

r is the required rate of return, also ( )i E R in the CAMP model _i

Γ is the current information of firm 0,i i.

1 β_iis the sensitivity of the expected excess asset returns to the expected excess market returns, or also

( , ) ( )

i m

i

m

Cov R R Var R

β = , see from http://en.wikipedia.org/wiki/Capital_asset_pricing_model.

(11)

The existence of A share premiums in Chinese stock market is first caused by the different liquidity in A and B share markets. Generally, the lower liquidity causes higher liquidity cost, thereby, investors need higher expected rate of return, which bring on the demand for lower share prices.² The monthly average trading volumns of B share market are obviously lower than those of A share market. Consquently, there exist the discounted B share prices compared with A share prices.

Second, the difference in price elasticity of demand between domestic and foreign investors brings on the A share premium. In Chinese stock market, for a domestic investor, due to lack of types of investment tools, A shares play a chief rool in the investment portfolio, so the price elasticity of demand for A shares is low for domestic investors. On the contrary, Chinese B shares are merely a small part in the total portfolio of foreign investors due to a variety of investment tools. The price elasticity of demand for B shares for foreign investors is high. As a consquence, according to price discrimination, customers with lower elasticity of demand have to pay more than those with higher elasticity of demand do. This also brings on A share premiums in Chinese stock market.

Third, foreign investors face to much more other risk than domestic investors, such as inflation, exchange risk and political risk. As a consquence, they have higher discounted rate r in (4). _i

Another important factor is the asymmetric information. The accounting and disclosure standards of Chinese firms are of lower quality than those in developed markets (Sze,1993, and Sjöö and Zhang, 2003). The information prompt and validity of Chinese firms also need to be improve. By reason of low quality informationm, the share prices need to be discounted. Compared with domestic investors, foreign investors have more information search cost. So B share prices have higher discounted rate r in (4). _i Chakravatr, Sarkar and Wu (1998), set up a theoretical model to prove that the asymmetric information is the major reason for A share premium in Chinese stock market. They find that B share prices discount was significantly negatively related to English media converge variable.

These factors explain the reason for the observed premium in A shares. Overall, foreign investors have lack of information problem and more risks. One interesting idea is

2 Higher liquidity cost causes higher ωin (3), so investors need higher expected rate of return. Higher expected rate of return r_iresults in lower share prices.

(12)

whether these factors cause A and B share prices moving independently before and after the ownership release in 2001. Sjöö and Zhang (2003) argue that if there exists a stochastic trend in A share premium, at least one investment group use wrong or inefficient information. If information flows efficiently between two investment groups, A and B share prices will be cointegrated. Sjöö and Zhang (2003) use Fisher tests in both panel unit root tests and cointegration tests and got the conclusion that information diffusion after 2001 becoming more effectively than before. They find after Chinese government lifting restrictions, A share premiums became stationary. They cannot do panel cointegration tests particular to sample period after 2001, due to lack of data. In our study, we will work on this problem on a longer sample period enough for panel cointegration tests.

5. Academic basis

Consider a panel data set contains a sample of N cross-sections. Each cross-sectioniare observed through a period T . So this panel dataset is unbalanced. In general, panel data _i unit root tests and cointegration tests are based on three procedures: pooling the data, pooling the statistic and pooling the p-values. In this paper, we are going to applying the last two procedures: pooling the test statistic and pooling the p-values. We do ADF unit root tests for individual cross-section and propose combining p-value tests for the panel data. Pooling the p-values is first introduced by Fisher (1932), and applied in panel data unit root test by Maddala and Wu (1999). We choose likelihood based cointegration tests, introduced by Larsson, Lyhagen & Löthgren (2001), for testing cointegration relationships in panel data. The tests are based on pooling the test statistic, by computing the averages of individual trace statistic from Johansen’s LR cointegration tests. The panel data unit root tests and cointegration tests we choose, are both suitable for our unbalanced panel data. In Section 5.1, we introduce panel data unit root tests while in Section 5.2, we introduce panle cointegration tests.

5.1 Panel data unit root testsCombining pvalue tests（Fisher tests）

There are three commonly used unit root tests for panel data: Levin & Lin (1992) tests, Im, Pesaran & Shin(1997) tests, and Combining p-value tests (Maddala and Wu, 1999).

(See Baltagi, B.H. and Kao,C. 2000)

(13)

The Levin & Lin tests are based on pooling the data, and require a homogeneous coefficient for y_it₋₁ in the model³. On the contrary, Im, Pesaran & Shin tests (IPS) and combining p-value tests relax the above restriction and allow for different coefficients ofy_it₋₁. The main difference between these two tests is: IPS are based on pooling the unit root test statistics, while combining p-value tests, introduced by Maddala and Wu (1999), suggest combining the p-values from each individual test, and then proposing a Fisher type test. Both the IPS and Fisher tests pool the information from individual unit root test, but Fisher tests are more suitable for the unbalanced dataset I use.

The individual unit root test for each cross-section I use is Augmented Dickey-Fuller (ADF) unit root test, which is based on the model

, 1 ,

1 ki

it i i t ij i t j i it

j

y ρ y ₋ γ y ₋ μ ε

=

Δ = +

∑

Δ + + ^，^ε^it ^~^NID^(0,^σⁱ²⁾， (5) 1,...,

i= N，t=1,...,T_i,

with the null hypothesisH_0,_i: ρ_i = against the alternative hypothesis 0 H_1,_i: ρ_i < for 0

each individual cross-sectional unit. Letp be the p-value of the unit root test for cross-_i sectional unit i. Maddala and Wu suggest a Fisher type test with the test statistic

1

2 ln

N i i

P p

=

= −

∑

, (6)

which is combining the p-values from each cross sectional ADF test.

With the assumptions that each cross-sectional unit is independent and the p-value of a test statistic follows a uniform distributionU(0,1), Phas a Chi-square distribution with 2N degreed of freedom. Since 2 ln− p_ihas a Chi-square distribution with 2 degree of freedom (Bickel and Doksum, 2001, problem 4.1.5, P.270).

When N is large, Choi (1999a) suggests a Z-test, with the test statistic

3 The LL tests are based on the model y_it =ρ_iy_it₋₁+z_it^′γ_i+u_it. And the tests assume ρ_i =ρ for all i, that is, require a homogeneous coefficient for y_it₋₁.

(14)

1

1 ( 2 ln 2)

2

N

i i

p

Z N ⁼

− −

=

∑

, (7)

which converges in distribution to a standard normal distributionN(0,1)through the central limit theorem, when p s are iid. Ahlgren, Sjöö and Zhang (2003) suggest _i constructing an asymptotic test using the above Z statistic when the independence assumption is violated. Maddala and Wu (1999) suggest using the bootstrap when dealing with the independence between cross-sectional units.

5.2 Panel data cointegration tests—Likelihood based cointegration tests

Larsson, Lyhagen & Löthgren (2001) propose likelihood-based cointegration tests in heterogeneous panels. These cointegration tests are based on the averages of trace statistic from Johansen’s likelihood ratio tests on each cross-sectional unit. This new proposed LR-bar statistic they suggested is quite similar to the test statistic in IPS since they are both based on the idea of pooling the test statistic. But the number of observations in time series can vary between the cross-sectional units, that is, the likelihood-based cointegration tests allow for unbalance panels.

Johansen’s likelihood ratio methods start from a vector autoregressive (VAR) model:

, 1 ki

it ij i t j it

j

Y Y ₋ ε

=

∑

Π + , (8)

with the error termε_it s are independent identically distributed ε_it ~NID(0,Ω ,_i) 1,...,

i= N,t=1,...,T_i.

The VAR model can be rewritten as an error correction representation. The heterogeneous error correction model is

1

, 1 ,

1 ki

it i i t ij i t j it

j

Y Y ₋ ⁻ Y ₋ ε

=

Δ = Π +

∑

Γ Δ + , (9) where Π is a _i p×pmatrix with rank r_i≤ . If p Π has a reduced rank _i r , _i Π can be _i

written as Π =_i α β_i _i^′, where α_i is known as the matrix of adjustment parameters and

βi′ is the cointegration vector. Note that T should be large enough to make sure the _i

(15)

upper VECM can be estimated separately for each cross sectional unit (see Larsson, Lyhagen & Löthgren, 2001).

The cointegration rank null hypothesis is:

0: ( _i) _i

H rank Π = ≤ for r r i=1,...,N.

The alternative hypothesis is:

1: ( _i)

H rank Π = for p i=1,...,N.

The trace statistic from Johansen’s likelihood ratio test under the null hypothesis ( _i)

rank Π ≤ for cross-section r iis denoted as

1

2 ln(1 ˆ )

p

ir i im

m r

LR T λ

= +

= −

∑

− , (10) The asymptotic distribution of the trace statistic is

LR→Zk, where

( )

¹

1 1 1

0( ) 0 0 ( ) '

Zk tr dW W WW W dW

⎧ ′ ′ − ⎫

≡ ⎨ ⎬

⎩

∫ ∫ ∫

⎭^,

and W is a k= −p r dimensional Brownian motion (see Larsson, Lyhagen &

Löthgren,2001).

Let (E Z_k)be the mean and Var Z( _k)be the variance of the above asymptotic distribution.

The simulated mean and variance of Z for_k k=1,..., 6 are shown in Table 1.

Denote the LR-bar statistic as the average of trace statistic under the null hypothesis ( _i)

rank Π ≤ for individual cross-sectional unit as r

1

1 ^N

r i

i

LR LR

N ₌

=

∑

, (11) The standardized LR-bar statistic is defined as

(16)

( ( )) ( )

k LRr

k

N LR E Z Var Z

γ = ⁻ , (12)

which follows a standard normal distribution N(0,1). The test is one-side, so the criterion of a level α is z₁₋_α(see Larsson, Lyhagen & Löthgren,2001). The test will begin with the null hypothesis r= , and proceed sequentially as suggested by 0 Johansen (1998), that is, the test will stop when the null is not rejected or the hypothesis

1

r= −p is rejected.

Table 1. Simulated mean and variance of Z . The mean and variance are obtained k

from Johansen (1995, Chapter 15).

k = −p r E Z( _k) Var Z( _k)

1 1.137 2.212

2 6.086 10.535

3 14.955 24.733

4 27.729 45.264

5 44.392 71.284

6 6.960 103.452

6. Empirical study on Chinese stock markets

As mentioned in above section, there exists a premium of A share prices over B share prices in Chinese stock markets. In Section 6.1, we are going to propose a panel data unit root test to check the stationary of the premium. We are hoping for the results that the premiums are stationary then get the idea of information from domestic and foreign investors is transferred virtually between two groups. In Section 6.2, we are going to do a panel data cointegration test to find if there exists cointegration relationship between A share prices and B share prices. We expect that there is one cointegration relationship between A and B share prices, which also shows there exists interdependenc between A and B share prices and the information flows between two investment groups effectively.

(17)

6.1 Panel data unit root tests for A share premium

First we denote the A share premium as

ln ln

it Ait Bit

d = P − P , i=1,...,N, t=1,...,T_i

where P and _Ait P respectively denote the A share prices and B share prices of firm _Bit ion time t . First we apply unit root tests on individual firm. We do the ADF tests on the all 40 firms in 2 different sample periods: from January 1995 to January 2001 and from March 2001 to December 2005. Since from Feb.19, 2001, B share stock markets are open to domestic investors. We choose a intercept in the test equation and specify the lag length k= , since 1 k= is sufficient for our tests. We denote the ADF statistic for 1 firm iasADF , and the corresponding p-values as_i p . The results for the ADF tests on _i

each firm can be found in Appendix Table A2. Then we use (6) to combine the p s _i together to compute the Fisher type Pstatistic for both markets, and Shanghai and Shenzhen separately, and we use (7) to obtain the asymptotic Zstatistic as well. The FisherPstatistic, the asymptotic Z statistic and the relative p-values are reported in Table 2.

Table 2. Panel data unit root tests on the A share premium. Results from Fisher tests and asymptotic Z-tests for both markets and Shanghai and Shenzhen separately. N is the number of firms included in the sample.

Period Market N P p-value Z p-value

1995(1)- 2001(1)

Shanghai&Shenzhen 38 22.17925 1 -4.4789 0.999

Shanghai 18 8.43181 0.999 -3.4247 0.999

Shenzhen 20 13.74744 0.999 -2.93513 0.998 2001(3)-

2005(12)

Shanghai&Shenzhen 40 146.9155 0.000 5.290135 0.000 Shanghai 20 83.33947 0.000 4.845501 0.000

Shenzhen 20 63.57603 0.010 2.63588 0.004

From Table 2, we can find that for the period 1995 to 2001, all the FisherPstatistic, the asymptotic Z statistic show that we should accept the null hypothesis of a unit root on the A share premium, that is, A share premium is not stationary, but has a stochastic trend. The results imply at least one investment group use wrong or ineffective information.

(18)

But there is a complete change after February 2001, when domestic investors were allowed to buy B shares. The results from the Fisher tests and asymptotic Z tests for sample period 2001 to 2005 tell that we can reject the null hypothesis that there exists a unit root on the A share premium, namely, A share premium is stationary for both market, Shanghai and Shenzhen Market separately. Also A share premium decreases after this reform. Consequently, the opening of B stock markets to domestic investors makes the information flows between two groups of investors effectively.

We also draw the histograms of p-values from ADF unit root tests for both markets, Shanghai market and Shenzhen market separately, displayed in Figure 4. These histograms directly show the test results above. The histograms on the left side of Figure 4 are of sample period January 1995 to January 2001. From these three histograms, we find that there is large proportion of large p-values. The three histograms on the right side, on the other hand, from the sample period March 2001 to December 2005, have lots of smaller p-values, which make the rejection on hypothesis of a unit root.

6.2 Panel data cointegration tests between A share prices and B share prices

In this part, we test the cointegration relationship between A share prices and B share prices in Chinese stock markets. The tests we use is the likelihood based cointegration tests (Larsson, Lyhagen & Löthgren, 2001). We first do Johansen’s likelihood ratio test on each firm, and then average the trace statistic from the tests.

Define Yit =

(

^lnPAit^{, ln}PBit

)

′as a bivariate time series, whereP and_Ait P are respectively _Bit A share price and B share price of firm iin time t . We use the error correction model (8) with a unrestricted intercept term for each individual firm. For an error correction model like (9), the error term should be white noise for effect Johansen’s LR tests, so we choose different optimal lag length k for each firm, using the VAR model (8). The _i

optimal lag length k for each firm can be found in Appendix Table A3. We are going to _i test the cointegration between A share prices and B share prices, so p=2. Due to the sequential test procedure of Johansen’s LR test, in our tests, we begin with the test under the null hypothesis r= , if the null is not rejected, the test stops; if not, we do 0 the test again under the null hypothesisH₀:r≤ and the test stops no matter whether the 1 null is rejected or not. So we denote the individual trace statistic for firm ias LR , with

(19)

Figure 4. Histograms of p-values from ADF unit root tests for A share premium.

0 2 4 6 8

0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1.0

Frequency

Both markets, 1995(1)-2001(1)

0 5 10 15 20 25

.0 .1 .2 .3 .4 .5 .6 .7 .8

Frequency

Both markets, 2001(3)-2005(12)

0 2 4 6 8 10

0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1.0

Frequency

Shanghai market, 1995(1)-2001(1)

0 4 8 12 16

.00 .05 .10 .15 .20 .25 .30 .35 .40

Frequency

Shanghai market, 2001(3)-2005(12)

0 1 2 3 4 5 6

.1 .2 .3 .4 .5 .6 .7 .8

Frequency

Shenzhen market, 1995(1)-2001(1)

0 2 4 6 8 10 12

.0 .1 .2 .3 .4 .5 .6 .7 .8

Frequency

Shenzhen market, 2001(3)-2005(12)

(20)

the null hypothesis H₀:r= . Similarly, we denote the individual trace statistic for firm 0 ias LR , with the null hypothesis _i₁ H₀:r≤ . The critical values for the significant level 1 0.05 in trace statistic with null hypothesis H₀:r= and 0 H₀:r≤ are respectively 15.49 1

and 3.84. To investigate the effect of the opening of B share markets to domestic investors in February 2001, I am going to do the cointegration tests in two separate sample periods: from January 1995 to January 2001 and from March 2001 to December 2005.

Table 3 reports the LR-bar statisticsLR0andLR1calculated by using the trace statistics from individual Johansen’s LR test and (11), respectively under the null r= and 0 r≤1. The individual trace statistics from Johansen’s LR tests can be found in Appendix Table A3. The standard LR bar statistics γ_LRare computed by LR-bar statistics and (12), and

simulated (E Z_k)and Var Z( _k)used in (12) can be found in Table 1. For instance, under the nullr = , 0 E Z( ₂)=6.086and Var Z( ₂) 10.535= . The asymptotic criterion of level 0.05 for γ_LR is 1.645. The corresponding p-values for γ_LR are also shown in Table 3.

Table 3. Panel data cointegration tests for A share and B share prices. N is the number of firms included in the sample.

Sample period

Stock market N _LR₀ _LR₁

LR0

γ ^p-

value ^LR¹

γ ^p-

value 1995-

2001

Shanghai market 18 12.437 8

1.9131 8.3026 0.000 0

2.2138 0.013 4 Shenzhen

market

20 12.450 8

3.1284 8.7697 0.000 0

5.9880 0.000 0 Both markets 38 12.444

6

2.5527 12.076 4

0.000 0

5.8678 0.000 0 2001-

2005

Shanghai market 20 19.000 0

1.0076 17.793 4

0.000 0

-0.3891 0.651 4 Shenzhen

market

20 13.615 4

0.9504 10.374 3

0.000 0

-0.5610 0.712 6 Both markets 40 16.307

7

0.9790 19.917 6

0.000 0

-0.6718 0.749 1

(21)

From the results in Table 3, we know that during the sample period January 1995 to January 2001, the null r= and 0 r ≤1are all rejected for both markets, Shanghai and Shenzhen market separately. This means there is no cointegration between the A and B share prices, and the time series are no longer suitable for an error correction model. On the other hand, after the opening of B share markets, the null r= are rejected but 0 r≤1 are accepted for all markets, that is, there exist cointegraion between A and B share prices in these markets. This result means after the markets were partially merged, the information flows between domestic and foreign investors more effectively.

7. Conclusion

In this paper, we have studied on the effect of information diffusion before and after the ownership release in February 19, 2001, by testing the stationary of A share premium and cointegration between A and B share prices. The panel unit root tests we propose on A share premium are Augmented Dickey-Fullar (ADF) tests for individual firm and Fisher tests for the panel, based on combining p-values from each individual cross- section. The panel cointegration tests on A and B shares we use is Johansen’s likelihood ratio tests for individual firm and likelihood-based panel cointegraion tests for panel, based on combining the test statistics.

Our empirical result is that before the opening of B share markets to domestic investors, A share premium has a unit root and there is no cointegration relationship between A and B share markets. However, after ownership release, A share premium is stationary and there is cointegration relationship between A and B share markets. This conclusion shows that the policy release on B share markets have good effect on the information flows between A and B shares.

(22)

Reference

[1] Ahlgren, N. Sjöö, B. and Zhang, J. (2003), ‘ Market Segmentation and Information Diffusion in China’s Stock Markets: Panel Data Unit Root and Cointegration Tests on A and B Share Prices’,

[2] Baltagi, B.H. and Kao, C. (2000), ’Nonstationar Panels, cointegraion in Panels and Dynamic Panel: A Survey’, Advances in Econometrics, 15, 7-52.

[3] Chakravarty, S. Sarkar, A.and Wu, L. (1998), ’Information Asymmetry, Market Segmentation and the Pricing of Cross-Listed Shares: Theory and Evidence from Chinese A and B shares’, Working Paper, Federal Reserve Bank of New York.

[4] Choi, I. (1999a), ’Unit Root Tests for Panel Data’, Working paper, Department of Economics, Kookmin University, Korea.

[5] Fisher, R. A. (1932), statistical Methods for Research Workers (Edinburgh: Oliver and Boyd).

[6] Im, K.S., Pesaran, M.H, and Shin, Y. (1997), ’Testing for Unit Roots in Heterogenous Panels’, Discussion paper, Department of Applied Economics, University of Cambridge.

[7] Johansen, S. (1996), Likelihood-Based Inference in Cointegrated Vector Autoregressive Models (Oxford: Oxford Universiy Press).

[8] Larsson, R. Lyhagen, J. and Löthgren, M. (2001), ‘Likelihood-Based cointegration Tests in Heterogeneous Panels’, Econometrics journal, 4, 109-142.

[9] Levin, A. and Lin, C. (1992), ‘Unit Root Tests in Panel Data: Asymptotic and Finite Sample Properties’, Discussion paper 92-93, Department of Economics, University of California at San Diego.

[10] Maddala, G. S. and Wu, S. (1999), ‘A Comparative Study of Unit Root Tests with Panel Data and a New Simple Test’, Oxford Bulletin of Economics and Statistics, Special issue, 631-652.

[11] Sharpe, William F. (1964), Capital asset prices: A theory of market equilibrium under conditions of risk, Journal of Finance, 19 (3), 425-442

[12] Sjöö, B. and Zhang, J. (2000), ‘Market Segmentation and Information Diffusion in China’s Stock Markets’, Journal of Multinational Financial Management, 10, 421-438.

[13] Shen, C.H., Chen, C.F. and Chen, L.H. (2007), ‘An empirical study of the asymmetric cointegration relationships among the Chinese stock markets’, Applied Economics, 2007, 39, 1433-1445.

[14] http://en.wikipedia.org/wiki/Dividend_Discount_Model#cite_note-0

[15] http://mylifedump.com/2008/01/04/evaluating-stocks-using-dividend-discount- model/

(23)

Appendix

Table A1. Detailed information of the firms in the sample.

Firm

No. Name A Share

Code

B Share Code SHANGHAI STOCK EXCHANGE

1 SVA ELECTRON 600602 900901

2 DAZHONG TRSP. 600611 900903

3 CHINA FIRST PENCIL 600612 900905

4 SHAI.RUB.BELT 600614 900907

5 SHAI.TYRE & RUBBER 600623 900909

6 SHAI.HAIXIN GP. 600851 900917

7 SHAI.DIESEL ENGR. 600841 900920

8 HERO GP.CO. 600844 900921

9 SHAI.FRIENDSHIP 600827 900923

10 SHAI.SHANGLING 600835 900925

11 SHAI.BAOSIGHT

SOFTWARE 600845 900926

12 SHAI.LUJIAZUI DEV. 600663 900932

13 HUAXIN CEMENT CO. 600801 900933

14 SHAI.NEW ASIA 600754 900934

15 INMONG.EERDUOSI

CASHMERE 600295 900936

16 HEILONGJIANG ELEC.PWR 600726 900937

17 EASTERN COMMS.CO. 600776 900941

18 HUANGSHAN TOURISM

DEV. 600054 900942

19 SHAI.KAIKAI INDL. 600272 900943

20 HAINAN AIRLINES 600221 900945

SHENZHEN STOCK EXCHANGE

(24)

21 SHN.LIONDA HDG. 000012 200012

22 KONKA FP.CO. 000016 200016

23 SHN.CHIN.BICYCLES 000017 200017

24 SHN.VCT.ONWARD 000018 200018

25 SHN.SHENBAO INDL. 000019 200019

26 SHNHUAFA ELTN. 000020 200020

27 SHN.CHIWAN WHARF HDG. 000022 200022

28 CHINA MER.SHEKOU 000024 200024

29 SHN.TELLUS HDG. 000025 200025

30 SHENZHEN FIYTA HDG. 000026 200026 31 SHEN.ACCORD PHARM.CO. 000028 200028 32 SHN.SEZ REAL ESTATE 000029 200029

33 CHONGQING CHANGAN 000030 200030

34 SHN.NANSHAN PWR. 000037 200037

35 CHINA INTL.MARINE 000039 200039

36 SHN.TEXTILE HLDG. 000045 200045

37 CHINA FANGDA GP.CO. 000055 200055

38 SHN.INTL.ENTERPRISE 000056 200056

39 SHENZHEN SEG CO. 000058 200058

40 SHIJIAZHUANG BAOSHI 000413 200413

Table A2. The results from ADF unit root tests for individual firm on A share premiums d . _it

Firm No.

1995.1-2001.1 2001.3-2005.12

1 -1.3962 0.585 -2.1196 0.2372

2 -1.6757 0.443 -4.1626 0.0009

3 -1.2482 0.6547 -2.5944 0.0408

4 -1.203 0.6746 -3.1062 0.0274

ADFi pⁱ p_i

ADFi

(25)

5 -1.514 0.5259 -1.8579 0.3519 6 -1.9859 0.2931 -2.0163 0.2798

7 -1.3825 0.5917 -2.0585 0.2619

8 -1.4784 0.5439 -4.6285 0.0002

9 -1.3034 0.6294 -2.8311 0.0554

10 -1.179 0.685 -3.136406 0.0253

11 -1.3055 0.6284 -2.09133 0.2485

12 -1.619 0.472 -3.587314 0.0067 13 -1.2445 0.6563 -5.083744 2.28E-05 14 -1.2929 0.6342 -3.881819 0.0025

15 NA NA -2.425246 0.1359

16 0.468 0.9854 -3.47645 0.0094

17 -1.1423 0.6999 -3.31303 0.0154

18 1.8132 0.374 -3.826693 0.0031

19 NA NA -3.205117 0.0209

20 -0.9226 0.7803 -6.678492 1.03E-08 21 -1.263185 0.6479 -1.000076 0.7536 22 -1.956198 0.3064 -2.996353 0.0367 23 -1.432995 0.5667 -4.083317 0.0012 24 -1.458561 0.5538 -1.979978 0.2956

25 -1.722969 0.419 -2.757915 0.066

26 -1.416245 0.5749 -3.920908 0.0022 27 -1.374349 0.5956 -1.84741 0.357 28 -0.99356 0.7568 -2.416256 0.1383 29 -1.777328 0.3917 -1.875789 0.3434 30 -2.001659 0.2862 -2.516974 0.1127 31 -1.281458 0.6394 -3.969554 0.0019 32 -1.619546 0.4717 -2.246474 0.1906 33 -1.488716 0.5386 -5.911161 4.89E-07 34 -2.058467 0.2619 -1.595263 0.4834 35 -1.210265 0.6714 -1.811833 0.3742 36 -1.320923 0.621 -2.457888 0.1273 37 -2.115528 0.2387 -2.854978 0.0523 38 -2.450844 0.1286 -3.73754 0.0041 39 -1.608792 0.4773 -4.366325 0.0004 40 -1.193428 0.6786 -3.276461 0.0171

Table A3. The results from Johansen’s cointegration tests for individual firm between A and B share prices.

Firm No.

1995.1-2001.1 2001.3-2005.12 lag

length

lag length

1 10.62019 0.828976 2 12.06994 0.296216 1

LR0 LR1 LR0 LR1

(26)

2 14.55045 3.737186 1 19.23631 0.511441 1

3 13.92311 2.929079 3 27.33262 0.643103 1

4 4.113293 0.399072 2 14.02562 0.628437 11

5 5.94716 1.6517 2 17.08012 1.440036 2

6 15.21796 1.415883 1 5.033702 0.07468 1

7 10.95621 1.609352 1 9.916442 1.447441 2

8 12.94917 1.951747 2 26.05079 2.665417 1

9 10.10745 2.190847 2 13.88126 1.300369 4

10 10.62697 1.602627 2 10.63605 0.110091 1

11 4.884348 0.003284 2 15.57876 1.259054 1

12 13.22942 1.711934 3 17.05783 0.225258 1

13 8.381965 1.219809 3 30.79134 1.585682 1

14 11.40805 2.347557 1 20.55029 2.279343 1

15 NA NA NA 10.70599 1.868903 1

16 8.010224 3.154147 1 21.72704 0.137574 1

17 24.88061 1.380722 1 15.46864 0.323599 1

18 13.81636 2.873749 1 30.90255 2.692723 3

19 NA NA NA 15.84718 0.215174 3

20 30.25692 3.427472 12 46.10802 0.447202 1

21 19.08066 5.130762 2 8.625526 0.725472 1

22 16.34758 5.127597 1 12.60713 1.41636 1

23 12.08676 1.637612 2 34.5047 0.699981 2

24 15.93672 5.666242 3 12.17827 0.07832 1

25 18.18003 3.706278 3 12.1552 0.632186 1

26 8.184996 2.741735 2 17.0923 0.134768 1

27 17.25761 2.345693 2 7.557158 2.609884 1

28 8.982619 1.838018 2 10.60153 3.960469 1

29 15.98205 4.52337 1 6.876411 1.40764 3

30 12.43979 3.467493 1 9.214788 0.004604 1

31 6.937999 0.737894 2 19.52573 0.967543 1

32 10.03713 2.75962 4 9.822523 1.106776 2

33 6.344111 1.225281 2 25.09945 0.896212 2

34 14.4002 3.085636 6 6.876806 1.195248 1

35 7.778402 2.148216 1 5.860015 1.151189 4

36 11.00529 3.905926 2 9.405983 0.208666 2

37 9.327981 3.912029 1 18.67106 0.416878 1

38 12.66656 1.50805 8 8.335838 1.260472 7

39 12.46413 3.396327 8 21.13573 0.135443 1

40 13.57542 3.704436 1 16.1618 0.000569 1

An Empirical Study on Market Segmentation and Information Diffusion in Chinese Stock Markets