diffusion might exist only during the ESM period and disappear, if or when the institutions behind the markets become more credible. We find it unlikely that the
political risk premium is an integrated stochastic process. It is more likely that the premium shifts at discrete intervals, if it shifts at all
4
.
4. Data and empirical results
4
.
1
. Data Chinese enterprises began to raise capital by issuing bonds and stocks
in the 1980s. Since then, China’s financial markets have evolved quickly. There are two stock exchanges in China, the Shanghai stock exchange and the Shenzhen
stock exchange. Both were inaugurated in the early 1990s. The Shenzhen exchange is a relatively smaller and less liquid market. The market for B shares
opened in 1992, which was more than 1 year after the A shares were first listed in the Shanghai stock exchange. Table 1 presents basic statistics of the two
exchanges
5
. The sample in this study includes weekly time series of 41 firms issuing
both A and B shares from July 1993 to June 1997 in either the Shanghai stock exchange or the Shenzhen stock exchange. Among them, 22 are from
the Shanghai stock exchange and 19 from the Shenzhen stock exchange. Using firm-specific data, we construct average price series in natural log form for
different exchanges, by evenly weighting the share prices of the firms in each exchange.
Table 1 Descriptive statistics
a
Shenzhen stock Shanghai stock
exchange exchange
Number of A shares listed 300
328 44
45 Number of B shares listed
A- and B-share market capitalization billion RMB 101.2
76.3 4.26
Average daily trading volume of B to A shares 2.99
a
This table contains basic statistics of China’s stock markets. The sample period is July 1993–June 1997.
4
A possible consequence of changes in the political risk premium is segmented trends in the price series. This could bias our tests towards finding I1 processes see Perron, 1989. Without any detailed
information about when these possible shifts might occur, we have not investigated this idea further.
5
A detailed institutional background of the Chinese stock markets is given in Zhang 1999.
4
.
2
. Results for the Shanghai and Shenzhen stock markets Empirical tests of cointegration are sensitive to the properties of the estimated
model, and tabulated critical values are valid only for normally distributed white noise residuals. To construct a VAR representation of our samples, such that the
assumption of normally and independently distributed white noise residuals cannot be rejected, takes around 10 – 20 dummy variables. A closer investigation of our
models reveals that cointegration depends critically on two observations, 51 and 52
6
. Figs. 1 and 2 reveal a downward trend in the prices of B shares during the first
year of the sample, up to observation 52. Imposing a dummy for what is the end of a downward trend, after which the prices quickly adjust back again, seems ad hoc.
Removing these critical outliers will make A and B shares look more alike than they really are
7
. The approach here is to avoid a huge number of dummy variables. We focus on
a sufficient number of lags to ensure that the null of no autocorrelation in the models cannot be rejected. This is achieved by using two and three lags in the
models. For the aggregated Shanghai and Shenzhen price series, Johansen’s trace test and
the max test statistics reject cointegration at the 5 risk level, in Table 2, based on Johansen 1995 and Hendry and Doornik 1996. The result changes if we include
some dummy variables, but we cannot reject the hypothesis of no cointegration with a margin. Thus, we find the cointegration test inconclusive about whether the series
are cointegrating or not.
The fact that we cannot easily establish cointegration is an important result because it tells us that there are substantial differences between A and B share prices.
A long-run stationary relation between A and B share prices cannot be taken for granted. As discussed above, we are skeptical to the no-cointegration hypothesis. In
this situation, with just two variables, an alternative test is to impose a reduced rank of the P matrix in Eq. 1, and test the significance of the adjustment parameters
8
. After imposing one cointegrated vector in the system, we find one significant error
correction mechanism. The a
1
parameter is significant for the Shanghai market, showing that foreign B-share prices drive the domestic A-share prices. In the
Shenzhen market, the relationship is reversed. In this market, the A-share prices drive the B-share prices.
Tables 3 and 4 show the estimated parameters of the VECMs. In both markets, B shares affect A shares, but there are important differences between the short and
the long run. In the short run, we find an uni-directional link from historical returns
6
These observations are c 51, July 22, 1994 and c 52, July 27, 1994.
7
Some empirical studies deal with this problem by deleting observations according to some statistical properties, for example, Xu and Wang 1997.
8
With only one I1 or I0 variable, the t-statistics of the adjustment vector will have an asymptotic N0, 1 distribution under the null, see Banerjee et al. 1993.
B .
Sjo ¨o
¨ ,
J .
Zhang J
. of
Multi .
Fin .
Manag .
10 2000
421 –
438
429 Table 2
Cointegration test statistics of the Shanghai and Shenzhen average price series
a
Adjustment parameters aˆ for Eigenvalues mˆ
Cointegration test statistics Normalized Eigenvectors
b .
r = 1 H
m ˆ maximum
m maximum
Trace Trace 95
B shares SH –A
A shares leading SH –B
B shares a
2
95 leading A
shares a
1
Panel A
:
Shanghai stock exchange 14.10
14.85 15.40
0.066 0.120
0.011 −
0.010 1.000
− 0.348
r = 0 12.80
3.80 2.05
3.80 3.27
0.46 2.05
1.000 −
0.073 r51
Panel B
:
Shenzhen stock exchange r = 0
11.16 14.10
13.31 15.40
− 0.032
− 0.071
1.000 0.059
− 1.186
0.012 r51
2.15 3.80
2.15 3.80
1.29 3.27
1.017 1.000
a
This table reports the Johansen cointegration test statistics. The variables are SH –A, Shanghai A-share average price; SH–B, Shanghai B-share average price; SZ –A, Shenzhen A-share average price; SZ–B, Shenzhen B-share average price. The t-values are in parentheses.
Significant at the 0.01 level or better.
Table 3 Vector error correction model results for the Shanghai stock exchange
a
Dependent variable D
SH –B ii D
SH –A i −
0.002 D
SH –A
t−1
0.015 −0.031
0.339 0.014
− 0.038
D SH –A
t−2
−0.900 0.191
0.082 0.470
D SH –B
t−1
3.507 1.059
− 0.405
D SH –B
t−2
− 0.064
0.810 −2.942
0.216 −
0.016 Constant
2.955 −0.389
− 0.110
0.008 b
x
t−1
−2.971 0.389
Vector residual tests Vector AR 1-2 F8, 344 = 1.756 [0.085]
Vector normality x
i 2
4 = 91.72 [0.000] Vector X
i 2
F42, 475 = 2.384 [0.000] Vector X
i
× X
j
F105, 417 = 2.505 [0.000]
a
This table reports the VECM results for Shanghai. The variables are, DSH –A, Shanghai A-share average price, first difference; DSH –B, Shanghai B-share average price, first difference. The t-values are
in parentheses. The P-values are in brackets. Significant at the 0.01 level or better.
on B shares to A shares. In the long run, B-share prices drive the A-share prices in Shanghai. In Shenzhen, the long-run effect is just the opposite; here A-share prices
drive the B-share prices. These results support the assumption that foreign investors in the Shanghai stock
exchange have better information, and that domestic investors adjust towards the prices of B shares. However, in the smaller and less liquid Shenzhen market, the
domestic investors have better information about the future long-run prospects of the firms. This could be a type of neglected firm effect, if foreign institutional investors
do not find it worthwhile or too costly to examine the firms listed in this exchange.
4
.
3
. Sensiti6ity tests In the following, we test the sensitivity of these results with respect to various
assumptions regarding the information process. First, the sample is split into different regimes
9
. Second, the series from Shanghai and Shenzhen are pooled into one model. Pooling the data will permit us to ask more detailed questions about the information
flow.
9
When viewing the results based on different sub-periods, it is important to remember that the tests of cointegration are based on the asymptotic properties of assumed infinite processes. Therefore, when
we split the sample into sub-periods, we are not formally testing for structural breaks; we are only demonstrating the consequences, assuming that there are different regimes in the sample period.
Figs. 1 and 2 reveal that the prices of A shares fall during the first 52 weeks. It could be that foreign investors play a larger role in the early stages of an ESM.
Suppose that this first part of the sample represents a different regime, with different behavior of domestic investors. To analyze this possible regime change,
the models are re-estimated with the first 60 observations truncated from the sample. The results are reported in Table 5. The cointegration test statistics are now
significant for both exchanges. More interesting, the significance of the a parame- ters is changing. As for the whole sample, foreign investors determine domestic
A-share prices in Shanghai, and domestic investors determine long-run B-share prices in Shenzhen. The change is that foreign investors also drive A shares in
Shenzhen. If we assume a regime shift, the role of foreign investors seem more important over time, in the sense that their influence spreads to the Shenzhen stock
exchange as the markets develop
10
.
Table 4 Vector error correction model results for the Shenzhen stock exchange
a
Dependent variable D
SZ –B ii D
SZ –A i D
SZ –A
t−1
0.038 −
0.055 −0.680
0.536 −
0.029 0.057
D SZ –A
t−2
−0.366 0.823
D SZ –A
t−3
− 0.071
− 0.025
−0.365 −0.921
0.178 D
SZ –B
t−1
− 0.141
−1.876 2.073
D SZ –B
t−2
− 0.131
− 0.153
−1.750 −1.705
− 0.145
− 0.102
D SZ –B
t−3
−1.672 −1.346
− 0.025
− 0.010
Constant 1.213
−2.721 0.072
0.030 b
x
t−1
3.320 −0.923
Vector residual tests Vector AR 1-2 F8, 344 = 0.734 [0.662]
Vector normality x
i 2
4 = 90.63 [0.000] Vector X
i 2
F48, 470 = 1.463 [0.027] Vector X
i
× X
j
F132, 390 = 2.185 [0.000]
a
This table reports the VECM results for Shenzhen. The variables are, DSZ –A, Shenzhen A-share average price, first difference; DSZ –B, Shenzhen B-share average price, first difference. The t-values are
in parentheses. The P-values are in brackets. Significant at the 0.01 level or better.
10
Another observation from Figs. 1 and 2 is that the markets can be characterized as bear markets until the end of 1995, and bull markets thereafter. We also estimated these periods separately, but the
results did not lead us to change our conclusions from above.
Table 5 Vector error correction model results for both Shenzhen and Shenzhen stock exchanges — sub-
sample
a
Shanghai stock exchange Shenzhen stock exchange
3 3
Lag Yes
Yes Cointegration
ECMs −
0.067 2.28 0.181 3.01
B shares leading A shares a
1
A shares leading B shares a
2
− 0.011 0.31
− 0.113 3.67
AR 1-2 = 0.902 [0.516] AR 1-2 = 0.839 [0.570]
Vector residual tests Normality = 100.7 [0.000]
Normality = 62.64 [0.000] Vector X
i 2
= 1.827 [0.036]
Vector X
i 2
= 0.779 [0.816]
Vector X
i
× X
j
= 2.112 [0.000]
Vector X
i
× X
j
= 0.805 [0.876]
a
This table summarizes the cointegration tests and the VECM results for both Shanghai and Shenzhen in the period of October 1994–June 1997, with 130 observations. The t-values are in
parentheses. The P-values are in brackets. Significant at the 0.01 level or better.
Our second sensitivity test is to pool the A and B shares of the two exchanges in one model. The cointegration test statistics from this ‘pooled’ model suggest two, or
possible one cointegrating vector, depending on the choice of risk level, see Table 6. In the following, we explore the different hypotheses that follow by assuming one
or two cointegrating vectors.
Suppose there is only one cointegrating vector in the system. The system would consist of three common stochastic trends and one stationary relation. The latter
could be a common risk premium for A shares over B shares in both exchanges. Alternatively, there is a stationary risk premium between the Shanghai and the
Shenzhen markets.
To test for these hypotheses, we start by testing for exclusion of exchanges or types of shares from the vector. All these hypotheses are rejected; means all four
variables are needed to form the stationary relation. Next, we test if the premium of A shares over B shares in Shanghai together with the premium in Shenzhen form
a stationary relation. This joint hypothesis, the vector is made up of two premia, is rejected by the data. The x
2
2 statistics is 10.702, with probability value of 0.0047
11
. The alternative is that the two non-stationary A-share series cointegrate with the
two non-stationary B-share series. This hypothesis assumes a joint risk premium of A shares over B shares in the two markets. To test this hypothesis, we impose the
restrictions of a ratio of A-share series and a ratio of B-share series on the
11
The exclusion tests are not presented here since they are all insignificant. In the tests, x = [SH – A, SH –B, SZ–A, SZ–B], the b vector is restricted as [b
1
= − b
2
] and [b
3
= − b
4
= 1]. The test [b
1
= − b
2
] is not rejected with probability of 0.1928, [b
3
= − b
4
] is rejected with probability of 0.0011.
B .
Sjo ¨o
¨ ,
J .
Zhang J
. of
Multi .
Fin .
Manag .
10 2000
421 –
438
433
Table 6 Cointegration test statistics of the Shanghai and Shenzhen pooled price series
a
Cointegration test statistics Restricted cointegration vectors
Adjustment parameters aˆ for r = 2 H
m ˆ maximum
m maximum
SH –
B A Trace
Trace 95 SH
– A B
SH –
A SH
– B
SZ –
A B SZ
– A
SZ –
B A SZ
– B
shares leading 95
shares leading shares leading
shares leading B shares
A shares A shares
B shares 27.10
54.04 0.000
47.20 a
11
, −0.146 −
1.235 1.000
a
21
, 0.037 0.000
r50 a
31
, −0.061 26.33
a
41
, −0.154 1.01
2.93 1.09
2.38 21.10
29.70 29.70
a
12
, 0.108 −
1.446 a
22
, −0.019 1.000
a
32
, 0.047 0.000
0.000 a
42
, 0.178 r51
15.61 1.35
3.76 0.60
1.20 14.10
14.09 15.40
r52 8.83
r53 5.26
3.80 5.26
3.80 Vector residual tests
Vector Vector
Vector AR 1-2 Vector
normality x
i 2
X
i 2
= 1.756
F32, 602 X
i
× X
j
= 2.00
[0.000] [0.000]
= 1.001
8 = 127.8 [0.000]
[0.468]
a
This table reports the Johansen cointegration test statistics. The variables are: SH –
A, Shanghai A-share average price; SH –
B, Shanghai B-share average price; SZ –
A, Shenzhen A-share average price; SZ
– B, Shenzhen B-share average price. The optimal lag length in this model is 3. The t-values are in parentheses. The P-values are in brackets.
Significant at the 0.05 level. Significant at the 0.01 level or better.
cointegrating vector. The data does not reject the hypothesis that these two ratios cointegrate, x
2
2 = 1.570, with probability value of 0.4566. Our conclusion from these tests is that one cointegrating vector is not sufficient to correctly describe the
system. There is a stationary risk premium between A and B shares, but this premium is not necessarily of the same magnitude in both markets.
Assume that there are two cointegrating vectors, one vector represents the stationary premium in the Shanghai stock exchange and the other premium in the
Shenzhen stock exchange
12
. The next question is how the two markets interact with each other in the long run.
The non-significant adjustment parameters of the pooled system suggest that there are two long-run exogenous prices in the system, the B-share prices in
Shanghai and the A-share prices in Shenzhen. The significant parameters confirm the findings above that the foreign investors drive the Shanghai market, and that
the domestic investors are more important in Shenzhen. The new result from the pooled system is that the price information in the Shanghai exchange spills over to
the B share market in Shenzhen, as suggested by the significant a
41
-parameter. Since the first vector occurs in two exchanges, but for different types of stocks, the
foreign investors seem to use the same information to price B shares in Shenzhen as domestic investors used in Shanghai.
4
.
4
. The flow of information between the markets Why should the prices of B shares lead the prices of A shares in Shanghai? Chui
and Kwok 1998 suggest that foreign investors are better informed and receive news faster than domestic investors because of the information barriers in China.
An additional factor is that B-share investors are mostly big financial institutions, while domestic A-share investors are relatively smaller. Thus, the returns of the
institutional favored shares could lead those of institutional unfavored shares, as suggested by Badrinath et al. 1995.
If information barriers are crucial, domestic investors have a problem in obtain- ing information, mainly because of the low creditability of domestic media. The
cost of obtaining information about the stock market in general and the prospects for individual firms is high for domestic investors. Therefore, a cost-effective way of
getting information is to observe the price movements of the foreign B shares. Then, the question is why A-share prices follow B-share prices in Shanghai, but not
in Shenzhen. The answer could be that the Shenzhen exchange is relatively smaller in terms of total market capitalization and number of listed firms, or because the
Shenzhen stock exchange is dominated by small firms.
In Table 1, we see that the total market capitalization of the Shanghai stock exchange is 101.2 billion RMB, and that of the Shenzhen stock exchange is 76.3
billion. By June 1997, in Shanghai, the number of A-share listing firms is 328, while
12
This hypothesis is not rejected by the data. The test statistic is x
2
2 = 0.4735, with probability value of 0.7892. Here, the two cointegrating vectors are restricted as [b
11
= 1, b
13
= b
14
= 0] and [b
23
= 1,
b
21
= b
22
= 0].
Table 7 Summary of the cointegration test results of individual firms
Stock exchange Cointegrated
Not cointegrated N = 22
Shanghai stock exchange 45.5
55.5 Shenzhen stock exchange
N = 19 73.7
26.3 N = 41
Total 58.5
41.5
in Shenzhen, this number is 300. If we calculate the ratio of the average daily trading volume of B shares to A shares in 1997, we find that this ratio is 4.26 for
Shanghai, and 2.99 for Shenzhen
13
. The Shanghai market is bigger and the B shares are much more liquid than those in Shenzhen. The result that foreign
investors are leading domestic investors in Shanghai could be in line with various small firms and liquidity effects found in other markets. The next section analyses
the firm size effect in detail.
4
.
5
. The firm size effect The lead-lag effect and the information hypotheses suggest that firm size could be
an important factor for foreign institutional investors. Therefore, we test if the prices of B shares have a tendency to lead those of A shares for firms with larger
market capitalization. Table 7 summarizes the cointegration test results, which show that more than half of the A and B shares are cointegrated. The share of firms
with cointegration among the assets is 58.5.
Table 8 Summary of the vector error correction model results — classified by firm size and stock exchange
a
a
2
Significant: A a
1
and a
2
a
1
and a
2
a
1
Significant: B shares leading B
significant insignificant
shares leading A shares
shares Panel A
:
firm size 62.5
– 12.5
Large 25.0
36.0 36.0
Medium 20.0
22.0 Small
87.5 12.5
– –
Panel B
:
stock exchange 72.7
9.1 4.5
13.6 Shanghai stock
exchange 52.7
26.3 10.5
Shenzhen stock 10.5
exchange
a
This table summarizes the estimated results from individual firms classified by firm size and stock exchange, respectively. Large firms are in the top 20; small firms are in the bottom 20; and in the
middle 60, they are the medium ones. Significant stands for significant at the 0.05 level or better.
13
Source, Shanghai and Shenzhen Stock Market Data, 1997, respectively.
Table 8 summarizes the VECM results, classified by firm size and exchange, respectively
14
. Firm size is measured by adding the market capitalization of A and B shares all in local currency, RMB at the end of June 1997. The sample is then
split into three groups, big firms are the top 20; small firms are the bottom 20; medium firms 60. Panel A of Table 8 reveals that B shares lead A shares for big
firms 62.5 as we expected. However, for small firms, B shares lead A shares as well 87.5, which is inconsistent with our expectation. We check firm size in
different exchanges in our sample and find that most of the firms in both the top and the bottom 20 are from the Shanghai stock exchange. We proceed to test if
the choice of exchange determines the investment decisions of the foreign investors. Panel B of Table 8 shows that in the Shanghai stock exchange most of the B shares
lead A shares 72.7, while in the Shenzhen stock exchange, on the contrary, most of the firms’ A shares lead B shares 52.7. The results demonstrate that it is the
Shanghai stock exchange that determines that B shares are leading A shares. In Shanghai, a larger number of A shares is driven by B shares compared with the
Shenzhen stock exchange. Thus, the choice of stock exchange is the most important factor behind the conclusion of B shares driving A shares.
5. Summary and conclusions