Applied Economics

ISSN: 0003-6846 (Print) 1466-4283 (Online) Journal homepage: https://www.tandfonline.com/loi/raec20

Capital account liberalization and dynamic price

discovery: evidence from Chinese cross-listed
stocks

Marc K. Chan & Simon S. Kwok

To cite this article: Marc K. Chan & Simon S. Kwok (2016) Capital account liberalization and
dynamic price discovery: evidence from Chinese cross-listed stocks, Applied Economics, 48:6,
517-535, DOI: 10.1080/00036846.2015.1083087

To link to this article: https://doi.org/10.1080/00036846.2015.1083087

View supplementary material

Published online: 15 Sep 2015.

Submit your article to this journal

Article views: 370

View Crossmark data

Citing articles: 9 View citing articles

Full Terms & Conditions of access and use can be found at

https://www.tandfonline.com/action/journalInformation?journalCode=raec20
APPLIED ECONOMICS, 2016
VOL. 48, NO. 6, 517–535
http://dx.doi.org/10.1080/00036846.2015.1083087

Capital account liberalization and dynamic price discovery: evidence from

Chinese cross-listed stocks
Marc K. Chana and Simon S. Kwokb
a
University of Technology Sydney, UTS Business School, Sydney, Ultimo NSW 2007, Australia; bSchool of Economics, The University of
Sydney, Sydney, Australia

ABSTRACT KEYWORDS
We analyse the effects of a recent financial reform (Shanghai-Hong Kong Stock Connect) that enables Capital account
cross-market investment between Hong Kong and Shanghai stock exchanges. Using a VECM, we find liberalization; cointegration;
that the reform announcement considerably narrows the equilibrium level of price disparity and VECM; cross-listing; Chinese
strengthens the price comovement of shares that are cross-listed in both markets. The estimated A-H shares
equilibrium relationship is in support of the relative law of one price. We find that both markets adjust
JEL CLASSIFICATION
in response to a disequilibrium in price disparity, leading to a sizeable error correction activity. The F36; G18; C32
Shanghai market contributes to approximately two-thirds of the price discovery process. Competition
and informativeness of trading affect the relative role of price discovery in each market. Finally, the
reform implementation reinforces the long-run cointegration relationship and strengthens the short-
run price comovements of cross-listed stocks despite the widening price disparity during the period.

I. Introduction Since these shares have the same dividend and voting
rights, the presence of price disparity, which can be
Capital account liberalization has played an integral
substantial, has been one of the most interesting puz-
role in China’s reform agenda in recent years.
zles in the Chinese financial market. By exploiting the
Although China’s capital account remains relatively
uniqueness of the reform and the prevalence of cross-
closed – it accounts for less than 3% of global
listed firms between Shanghai and Hong Kong, we
holdings of cross-border assets and liabilities –
formally assess the implications of capital account lib-
ambitious attempts have recently been launched to
eralization for the dynamics of price discovery and
liberalize its financial and foreign exchange systems.
financial market integration in China.
However, it remains a highly controversial issue as
At a general level, our study integrates two impor-
to how financial markets will be affected by such
tant strands of literatures. The first literature is
liberalizations.
related to the effects of capital account liberalization
This study presents new evidence on the topic by
in emerging markets. In recent years, the policy-
examining a unique natural experiment from a recent
experiment approach has been advocated as a clean
financial reform in China. The financial reform,
source of identification of policy effects (Henry
known as Shanghai-Hong Kong Stock Connect, enables
2007).1 While few studies have pursued this
cross-market investment between two of the largest 10
approach, an exception is Chari and Henry (2004),
stock exchanges in the world – Shanghai and Hong
who disentangle the effects on stock prices using
Kong. Hong Kong investors are allowed to invest in
firms that are eligible and ineligible for purchase by
selected stocks in the Shanghai market, and vice versa,
foreigners in liberalizations. However, their model
subject to a daily quota. An interesting feature of the
does not incorporate pricing dynamics. The second
reform is that while it restricts cross-market invest-
literature is related to the mechanism of price dis-
ment to designated stocks it includes shares of all
covery among cross-listed firms in developed
firms that are concurrently listed in both markets.

CONTACT Marc K. Chan marc.chan@uts.edu.au

1
A prevailing empirical approach is to analyse cross-country time-series data using various measures of liberalization as the principal source of policy
variation (e.g. Henry 2000; Harrison, Love, and McMillan 2004; Galindo, Schiantarelli, and Weiss 2007). The findings are often sensitive to country coverage,
sample periods and indicators of liberalization (Eichengreen 2001).
Supplemental data for this article can be accessed here.
© 2015 Taylor & Francis
518 M. K. CHAN AND S. S. KWOK
markets. In particular, our model is similar to Eun
and Sabherwal (2003), who analyse Canadian stocks
that are listed on both the Toronto and US stock
exchanges.2 They find that the home market’s share
of price discovery is around 60%. However, as in
most studies in the literature, they focus on pre-
existing patterns of pricing dynamics. There remains
a large knowledge gap between pricing dynamics
and important policy issues such as financial market
globalization.
Our study also extends the literature on cross-
listings in Hong Kong (H-shares) and Mainland
(A-shares) markets. In 2014, there were more than
80 cross-listings, around three-quarters of which
were firms that were concurrently listed in the
Hong Kong and Shanghai stock exchanges.3 These
firms constitute a sizeable proportion of the total
market capitalization in both markets. A large body
of the literature, which is predominantly based on
static models, has focused on the level of price dis-
parity between H-shares and A-shares.4 By contrast,
there are fewer studies that analyse the price comove-
ment between H-shares and A-shares. Su, Chong,
and Yan (2007) find that there were more firms
with cointegrated H-share and A-share prices in
2004 when earlier episodes of liberalization were
launched (Qualified Foreign Institutional Investor
(QFII) and CEPA). Choi et al. (2013) document a
stronger cointegration between H-share and A-share
prices in the post-global financial crisis period from
2009 to 2011. Both papers do not formally model the
dynamics of price disparity, however. An exception
is Cai, McGuinness, and Zhang (2011), who focus on
daily closing prices between January 1999 and
March 2009. Using an ECM for H-shares, they find
that the relationship between A-shares and H-shares
strengthened during the 2000s both in terms of price
disparity and pricing dynamics. In particular, they
reveal a marked uptrend in cointegration effects over
the period, while controlling for key sub-period
events. This finding is notable in light of strong
barriers to arbitrage during the 1999–2009 period
due to market segmentation. In this respect, the
institutional background is consistent with their
2
For an earlier study that apply a similar model to one firm (IBM), see Harris et al. (1995).
3
The remaining firms were concurrently listed in the Hong Kong and Shenzhen stock exchanges. While these firms are not included in our baseline sample,
we will conduct a similar empirical analysis as a natural control group. More details can be found in the section ‘Shenzhen market as a control group and
recent developments’.
4
Examples of recent studies include Arquette, Brown, and Burdekin (2008), Chang, Luo, and Ren (2013) and Chung, Hui, and Li (2013).
finding that the error correction activity was gener-
ally very low during the period.
Our study deals with a very different market set-
ting from Cai, McGuinness, and Zhang (2011), Su,
Chong, and Yan (2007) and Choi et al. (2013) in
terms of the overall market and regulatory environ-
ment, as well as available stock pairings. In particu-
lar, significant developments have taken place since
the time frame considered in Cai, McGuinness, and
Zhang (2011), reflecting a deeper integration
between Hong Kong and Mainland markets. These
include not only specific financial market policies
such as Shanghai-Hong Kong Stock Connect but
also broader developments such as Renminbi
Qualified Foreign Institutional Investor (RQFII) in
2011 and the development of the ‘CNH’ or offshore
RMB market. Market de-segmentation can poten-
tially unveil a richer set of pricing dynamics but at
the same time also presents a challenge to the under-
lying modelling approach.
To analyse the dynamic effects of the financial
reform, we adopt a more general modelling
approach by estimating a VECM (Engle and
Granger (1987)) for firms that are concurrently
listed in the Hong Kong and Shanghai markets.
For each firm, we first use cointegration tests to
formally test for the existence of equilibrium rela-
tionship between the prices of H-shares and
A-shares. Since the VECM captures the joint price
movements of the firm’s H-shares and A-shares, we
are able to derive estimates of the equilibrium price
disparity, as well as the magnitude and share of the
disequilibrium price adjustment process in each
market. In this respect, our approach can be consid-
ered as an extension to Cai, McGuinness, and Zhang
(2011). They estimate a univariate model in which
the return of H-shares is affected by the return of
A-shares and the level of price disparity, while the
return of A-shares is assumed to be exogenous as a
martingale difference sequence. The presence of
strong arbitrage barriers is central to their model
design. Given the recent policy developments, how-
ever, we are confronting a new market environment
that encourages two-way trading activities and
 APPLIED ECONOMICS 519

information flow between the two markets. This
motivates a more general model set-up that would
allow us to extract the richer mutual dynamics
between the H- and A-share prices.5
Our empirical analysis is performed using data
before the policy was announced, after the policy
was announced and after the policy was implemen-
ted. We first look at announcement effects by focus-
ing on data prior to the implementation of the
reform, which was originally scheduled to launch 6
months after the announcement. The announcement
effect is important – if the market is efficient, it
should already reflect the information related to the
reform starting from the date of policy announce-
ment instead of implementation. We also look at
implementation effects by using data after the
reform officially started. There are two features of
the data that make our empirical analysis particu-
larly interesting. First, the firms in the data have
widely different pre-existing levels and signs of
price disparity. In particular, just before the policy
was announced, one-third of the firms had a price
premium and two-thirds had a discount in the
H-share market. This allows us to assess intuitively
(through the lens of a natural experiment) how the
reform affected the price convergence process.
Second, the overall price disparity between the
A-share and H-share market changed dramatically
during the sample period. This presents both an
opportunity and challenge to the estimation and
interpretation of our model.
Our results indicate that the reform strengthens
the degree of financial integration between the Hong
Kong and Shanghai markets but in intricate ways.
The major findings are listed as follows. First, the
cointegration tests show that substantially more
firms have cointegrated A-share and H-share prices
during the post-announcement period between
April and July 2014. Among these firms, which
constitute 40% of the baseline sample, the estimated
cointegrating coefficients are close to unity.
Therefore, the equilibrium relationship formed
after the policy announcement is consistent with
the relative law of one price.6 Second, the estimated
post-announcement equilibrium level of price dis-
parity is considerably narrower than the price dis-
parity just before the announcement. In particular,
among firms whose H-shares were traded at a pre-
mium to A-shares, our model predicts that the price
disparity will narrow by an average of 40% in equili-
brium. Third, both markets adjust in response to a
disequilibrium in price disparity, leading to a size-
able error correction process. On average, a 1%
deviation from equilibrium will generate a next-day
price response of 0.27% in H-shares and 0.14% in
A-shares. Therefore, the Shanghai market contri-
butes to approximately two-thirds of the price dis-
covery process. Regression analysis indicates that
competition and informativeness of trading affect
the relative role of price discovery in both markets,
a finding that is consistent with Eun and Sabherwal
(2003). Fourth, despite a widening overall level of
price disparity during the post-implementation per-
iod, we find that the implementation of the reform
strengthens comovements between the A-share and
H-share prices of cross-listed firms during this per-
iod. The intricate relationship between the two mar-
kets can only be fully revealed using firm-level data
and a rich dynamic model.
The article proceeds as follows. Section II pro-
vides a policy background for the empirical analysis.
Section III describes the VECM and its relationship
with the literature. Section IV discusses the data.
Section V reports the estimation results of the
VECM, regression analysis and sensitivity analysis.
Section VI concludes.
II. Policy background
The Hong Kong and Shanghai markets are ranked
among the largest 10 stock exchanges in the world.
However, there is a strong asymmetry in institu-
tional characteristics and the degree of openness
between both markets. An interesting feature of
both markets is the prevalence of firms that are
concurrently listed – in 2014, 68 firms were concur-
rently listed in both markets, and their shares con-
stitute approximately 50% and 18% of the total

5
Otherwise, this may lead to a biased estimate of the error correction process if both H-share and A-share prices adjust in response to a disequilibrium in
price disparity. Another advantage of the VECM is that it can perform hypothesis tests on the law of one price. The univariate model restricts the
cointegrating coefficient to unity, which assumes the validity of the relative law of one price. Supplemental data on the relationship between our model
and Cai, McGuinness, and Zhang (2011) can be assessed in the section ‘Supplemental data’.
6
As an illustration, the cointegrating relationship can be written as logðPtH Þ þ β logðPtA Þ þ c ¼ 0, where PtH is the H-share price, PtA is the A-share price, β is
the cointegrating coefficient and c is a constant. The relative law of one price implies that β ¼ 1.
520 M. K. CHAN AND S. S. KWOK

market capitalization of the Shanghai and Hong
Kong markets, respectively.7
For each cross-listed firm, its A-shares (in
Shanghai market) and H-shares (in Hong Kong
market) are nonfungible and can be traded in their
respective stock exchanges only. Although A-shares
and H-shares have the same dividend and voting
rights, their price disparity is highly heterogeneous
across firms – for instance, in March 2014, one-third
of the firms had A-shares that were at least 50%
more expensive than H-shares. The price disparity
between A-shares and H-shares is one of the most
interesting puzzles in the study of Chinese financial
markets.
The recent announcement of a pilot programme,
called Shanghai-Hong Kong Stock Connect, funda-
mentally changed the prospect of market segmenta-
tion between Shanghai and Hong Kong markets. The
policy announcement, which was made on 10 April
2014, described in detail how the pilot programme
will enable cross-market investment between Hong
Kong and Shanghai’s stock markets. In particular,
Hong Kong investors can invest in the Shanghai
market, and vice versa, subject to an overall quota
of 250 billion yuan (40 billion USD) and 300 billion
yuan (48 billion USD), respectively.8 The quotas
constitute 2% and 1.6% of the market capitalization
in the Shanghai and Hong Kong stock exchanges,
respectively. The programme also restricts cross-
market investment to designated stocks in both mar-
kets. These include constituents of major indices and
shares of firms that are concurrently listed in the
Shanghai and Hong Kong stock exchanges. The pilot
programme was scheduled to launch 6 months after
the policy announcement. It represents a significant
step towards China’s capital control liberalization.
Prior to the pilot programme, cross-border invest-
ments in stock markets between Mainland China
and the rest of the world were dominated by the
QFII and Qualified Domestic Institutional Investor
(QDII) programmes. The QFII programme started
in 2002 and allows foreign institutional investors to
invest in Mainland China’s securities markets. The
QDII programme started in 2006 and allows institu-
tional investors in Mainland China to invest in
financial markets abroad. Quotas were allocated to
both programmes for institutional investors only,
and the quotas were gradually increased over the
past decade. As of 2014, the total quotas for QFII
and QDII are 53.5 and 86.5 billion US dollars,
respectively. Therefore, the combined sizes of QFII
and QDII programmes are similar to the pilot
programme.
The QFII programme was further extended in late
2011. Formally known as the RQFII programme, it
allows foreign investors to invest in Mainland
China’s securities via offshore Renminbi accounts.
The programme made its debut in Hong Kong in
December 2011, with an initial quota ceiling of 20
billion yuan (3.1 billion USD) allocated to a number
of companies, all of which were subsidiaries of main-
land asset managers. The quota ceiling was increased
to 270 billion yuan (43.2 billion USD) in November
2012.9 By the end of February 2014, the outstanding
quotas issued to RQFIIs in Hong Kong amount to
180.4 billion yuan (29.4 billion USD), compared to
the corresponding quotas of 52.3 billion USD to all
QFIIs. The RQFIIs are mainly composed of Hong
Kong subsidiaries of Mainland fund management
companies and securities companies.10
The implementation of the aforementioned capi-
tal liberalization policies and the launch of Shanghai-
Hong Kong Stock Connect culminated a whole new
policy environment relative to the past. It is expected
that the expanding cohort of RQFIIs and the grow-
ing offshore Renminbi market in Hong Kong would
greatly facilitate investments from Hong Kong into
China. Indeed, the development of the RQFII pro-
gramme saw the creation of several A-share
exchange traded funds and bond funds in Hong

7
The Mainland market has two stock exchanges (Shanghai and Shenzhen). Firms can only list in either the Shanghai or the Shenzhen market. In 2014, 16
firms were concurrently listed in Hong Kong and Shenzhen markets. Therefore, a total of 68 + 16 = 84 firms were cross-listed between Hong Kong and
Mainland China. Firms that are concurrently listed in the Shenzhen and Hong Kong stock exchanges are unaffected by the programme and can form a
natural control group for analysis. We will discuss this point in further detail in the section ‘Shenzhen market as a control group and recent developments’.
8
The daily quotas in the Shanghai and Hong Kong markets are 10.5 and 13 billion yuan, respectively. Investors in Mainland China who are allowed to invest
in the Hong Kong market must have at least 500 000 yuan (80 645 USD) in securities or cash.
9
Since then, the RQFII programme has been extended to financial institutions in London, Singapore and Paris, to name a few.
10
In addition, the offshore Renminbi market in Hong Kong expanded further as the daily yuan conversion cap (20 000 yuan) faced by Hong Kong residents
was technically removed by the Hong Kong Monetary Authority at the launch of the Shanghai-Hong Kong Stock Connect on 17 November 2014. However,
the lifting of the cap is restricted to the offshore RMB market. Transferring of funds to and from mainland China is still subject to the same restriction as
before.

Kong. Furthermore, the Shanghai-Hong Kong Stock
Connect enables regular northbound investment
activities, thus realizing the large demand for main-
land securities by Hong Kong-based investors. This
new policy environment forms the backdrop of our
two-way dynamical analyses of cross-listed shares in
this article.
The increase in northbound investments occurred
around the time when A-shares were traded at a
growing premium relative to H-shares. This can be
seen from the trajectory of the Hang Seng China AH
premium Index. Launched on 9 July 2009, the Index
keeps track of the value-weighted average price dif-
ference of A-shares over H-shares for the largest and
most liquid companies cross-listed in the Shanghai
and Hong Kong stock exchanges.11 Figure 1 plots
the time series of the Index from January 2013 to
early April 2015. The four shaded regions denote the
sub-periods used in our analysis (discussed in
Section IV). As shown by Fig. 1, cross-listed H-
and A-shares were trading approximately at par in
2013 (after adjusted for the exchange rate) while the
Index was lingering around the 100 level. The Index
slides gradually below 100 from February 2014 to
July 2014, but then it reverts to an upward trend
from the end of July 2014. The Index continues to go
up for the rest of 2014. A sharp increase in the Index
160
140
120
100
80
Jan 2013
Figure 1. Hang Seng China A/H premium index (grey areas denote key sample sub-periods).
11
As of 2015, there are 58 pairs that form the constituents of the Index. The index contains 15 financial stocks, which constitute over half of the weight of the
index.
APPLIED ECONOMICS 521
occurs shortly after the implementation of the
reform in November 2014, and the Index surges
beyond 120 in early 2015. While the increasingly
favourable market sentiment in mainland China
could be part of the reason, it is believed that the
launch of the Shanghai-Hong Kong Stock Connect
could act as the catalyst to the observed change. In
fact, as our empirical analysis suggests, the pricing
dynamics of cross-listed firms after the programme
implementation turn out to be richer than what was
depicted in Fig. 1.
III. VECM
We consider the joint pricing dynamics of H-shares and
A-shares for a firm that is concurrently listed on both
the Hong Kong and Shanghai stock exchanges. Let the
price of the firm’s H-shares and A-shares at time t be PtH
and PtA , respectively. Both prices are exchange-rate-
adjusted to Hong Kong dollars. The log-prices are
denoted by ht ¼ logðPtH Þ and at ¼ logðPtA Þ, respec-
tively, so that Δht ¼ ht  ht1 and Δat ¼ at  at1
are the relative returns. We are interested in modelling
the evolution of the bivariate log-price process over the
period t ¼ 0; :::; T. The VECM of order p is given as
follows (Engle and Granger 1987):
Oct 2013 Jul 2014 Apr 2015
Date
522 M. K. CHAN AND S. S. KWOK

0
X
p
innovation vector ut ¼ ½uht uat  is independent and
Δ ht ¼ c þ α wt1 þ
h h
πhh
j Δ htj identically distributed with mean 0 and has an
j¼1
unrestricted covariance matrix u . Therefore, the
X
p
þ πha
j Δ atj þ ut
h
(1) short-run dynamics allow for both contemporaneous
j¼1 change and two-way temporal feedback between
H-share and A-share prices. The former is achieved
X
p
Δ at ¼ ca þ αa wt1 þ πah by the potential correlation in innovations uht and uat ,
j Δ htj
j¼1 while the latter is captured by the lagged returns (up
X
p to lag p) of both types of shares.
þ πaa
j Δ atj þ ut
a
(2)
j¼1

where the cointegrating equation, which represents IV. Data

the equilibrium relationship between the log-prices
ht and at , is given by
wt1 ¼ ht1 þ βat1 þ μ þ γt
If wt ¼ 0, the pair of log-prices is in equilibrium.
Otherwise, a deviation from equilibrium occurs,
which will lead to a subsequent adjustment of both
log-prices according to Equations 1 and 2. In parti-
cular, the rates of price adjustments from dis-
equilibrium will depend on the error correction
adjustment coefficients αh and αa . We expect that
αh < αa or, more specifically, αh < 0 and αa > 0, so
that prices will converge to the equilibrium relation-
ship in the long run should a deviation from equili-
brium occurs.
The cointegrating equation allows for an unrest-
ricted value of the cointegrating coefficient β. The
equation also contains a constant μ, as well as a
linear time trend γt (Johansen 1991, 1995). A special
case of interest is β ¼ 1, which implies the relative
law of one price. The component μ þ γt can then be
interpreted as the equilibrium level of discount of
H-shares relative to A-shares, a point which we will
discuss in detail in the section ‘Convergence of price
disparity in equilibrium’.
When the log-prices are cointegrated, fwt gTt¼0 will
be a trend stationary process. The cointegration
tests, as well as unit root tests for ht and at , will be
discussed in the section ‘Cointegration tests’. In
addition to the adjustments due to departure from
the long-run equilibrium, the VECM representation
in Equations 1 and 2 captures a rich array of short-
run price dynamics in terms of a VAR process. The
12
One stock suspended trading for 1 week, while the others suspended trading for at least 3 months in 2014. Except for the stock that suspended trading for
1 week (utilities), the other stocks were ‘small-cap’ stocks.
13
An observation is included if the stock is traded in at least one of the markets. In days where the stock is traded in one market only (e.g. public holiday in
one market), the missing price in the other market is filled in by interpolation.
Our data sample consists of daily close prices of
stocks that are cross-listed on both Hong Kong and
Shanghai stock exchanges. In 2014, 68 firms were
(3)
concurrently listed in Hong Kong and Shanghai
markets. We exclude seven stocks from the raw
sample based on the following criteria: (1) two
stocks of firms that are primarily involved in broker-
age activities, and (2) five stocks with an extended
nontrading period (1 week or more) in at least one
of the markets during the post-announcement per-
iod in 2014.12 In the end, there are 61 firms in the
analysis sample, which contains 122 stock price
series.
The cointegration tests are carried out separately
using data from the pre- and post-announcement
periods. The sample is divided into four roughly
equal sub-periods.13 The pre-announcement period
lies between 2 January 2014 and 9 April 2014. The
first post-announcement period, which covers more
than 3 months after policy announcement, lies
between 10 April 2014 and 29 July 2014. As dis-
cussed in the previous section, this sub-period ends
with a reversal of trend in the increase of A-share
discount. The second post-announcement period
begins on 30 July 2014 and ends on 16 November
2014. The final sub-period, i.e. the post-implementa-
tion period, begins on the reform implementation
date of 17 November 2014 and ends on 27 February
2015.
Estimation of the VECM is carried out using the
first post-announcement period, which spans
77 days of trading, and the post-implementation
period, which spans 75 days of trading. In the
 APPLIED ECONOMICS 523

baseline model, the maximum lag order in Equations
1 and 2 is chosen to be p ¼ 2, and sensitivity analysis
on lag length will be discussed in the section ‘Further
empirical specifications’. The model is estimated by
price discovery process. The Hong Kong market
takes up an average of 38.9% of the combined trad-
ing volumes in both markets ðHvolShare ¼
HVolþAVolÞ. While the mean percentage bid–ask
HVol

the method of maximum likelihood with Gaussian spread of H-shares (HBas) is slightly higher than
innovations (e.g. Johansen 1988, 1991). that of A-shares (ABas), the average spread
Table 1 reports summary statistics of the market  
ratio is 1.54 SpreadRatio ¼ HBas
ABas . Following Cai,
variables related to the H- and A-shares of the 61 McGuinness, and Zhang (2011), we construct mea-
firms during the first post-announcement period. sures of differential market sentiment and informa-
After adjusting for the exchange rate, A-shares tion asymmetry, respectively:
have a slightly higher average price than H-shares
and they have a similar average dollar amount of jHVol  AVolj
trading volume. We follow Eun and Sabherwal VolRelative ¼
HVol þ AVol
(2003) and construct measures of volume shares
and bid–ask spread as potential determinants of the jHBas  ABasj
BasRelative ¼
HBas þ ABas

Table 1. Summary statistics of market variables. If there is any difference in the trading volume or
Percentiles bid–ask spread between both markets in either direc-
Variables Mean SD 25th Median 75th
tion, the measures above will be strictly positive. The
Closing price (in HKD)
H-share 10.08 12.17 3.18 4.94 12.92 sample averages of VolRelative and BasRelative are
A-share 10.43 10.71 4.05 5.75 12.80 0.40 and 0.21, respectively. Table 1 also reports the
Volume (in million
HKD) firms’ market capitalization on April 9. The average
H-share 180.52 283.88 23.36 52.87 187.48 market capitalization is rather sizeable at 65.79 billion
A-share 170.50 210.51 55.44 126.32 185.48
HKD, with an SD of 184.69 billion HKD.
Volume (in million
shares) A key summary measure of interest, called the HA
H-share (HVol) 27.31 55.03 3.65 10.32 19.52
A-share (AVol) 20.73 18.51 6.95 13.25 29.99
premium, is defined as follows:
Per cent bid–ask
spread PtH
H-share (HBas) 0.26% 0.15% 0.16% 0.21% 0.32% HApremiumt ¼ 1
A-share (ABas) 0.23% 0.14% 0.10% 0.22% 0.32%
PtA
HVolSharea 38.90% 20.49% 24.46% 36.30% 51.84%
SpreadRatio b
1.54 1.07 0.87 1.12 1.75 The measure compares the exchange-rate-
VolRelativec 0.40 0.23 0.24 0.36 0.54 adjusted prices of H-shares (PtH ) and A-shares
BasRelatived 0.21 0.18 0.06 0.17 0.27
MktCap on 9 65.79 184.69 5.36 11.97 42.52 (PtA ) of the firm. If the firm’s H-shares are traded
e
April 2014 at a relative discount to A-shares, the HA premium
HApremium on selected datesf
2 January 2014 –0.13 0.28 –0.36 –0.14 0.11 is negative. If the H-shares are more expensive than
9 April 2014 –0.13 0.29 –0.33 –0.13 0.10 A-shares, the HA premium is positive. If both are
10 April 2014 –0.12 0.24 –0.29 –0.11 0.08
(announcement) traded at price parity, the HA premium is zero.
29 July 2014 –0.08 0.22 –0.25 –0.10 0.13 The bottom panel of Table 1 reports summary
14 November 2014 –0.17 0.19 –0.35 –0.19 0.03
17 November 2014 –0.19 0.21 –0.37 –0.21 0.02 statistics of the HA premium on selected key dates
(implementation) during the sample period. The average (unweighted)
27 February 2015 –0.35 0.21 –0.50 –0.41 –0.15
Notes: Unless otherwise stated, the summary statistics are computed from
HA premium on 2 January 2014 is –0.13, which
a
the post-announcement period between 10 April 2014 and 29 July 2014. implies that H-share prices were on average 13%
HVolShare is defined as HVol/(HVol + AVol), where HVol and AVol are the
traded volumes (both in million HKD) of H-shares and A-shares, cheaper than A-share prices. The SD is 0.28, which
b
respectively. indicates a relatively wide cross-sectional variation of
SpreadRatio is defined as HBas/ABas, where HBas and ABas are the
percentage bid–ask spreads of H-shares and A-shares, respectively. price disparity (either in terms of premium or
c
VolRelative is defined as |HVol – AVol|/(HVol + AVol). discount). The summary statistics on the HA pre-
d
BasRelative is defined as |HBas – ABas|/(HBas + ABas).
e
MktCap is the market capitalization (in billion HKD) as of 9 April 2014. mium are similar on 9 April 2014. By contrast, on 10
f
HApremium is defined as PH/PA –1, where both the price of H-shares
(PH) and A-shares (PA) are expressed in Hong Kong dollars.
April 2014, the average HA premium changed
524 M. K. CHAN AND S. S. KWOK

slightly to –0.12, but the SD decreased substantially on April 9: (1) lower than –10%; (2) between –
by 5 percentage points to 0.24. This suggests that the 10% and +10%; and (3) larger than +10%. There
cross-sectional variation in price disparity reduced are 32, 14 and 15 firms in the above groups,
upon the policy announcement. At the end of the respectively. The daily stock returns in both mar-
first post-announcement period on 29 July 2014, the kets did not exhibit any anomaly on April 8 and 9.
average HA premium was –0.08 with an SD of 0.22. However, on April 10, the Hong Kong market
Beyond that point, A-shares tended to become more experienced a 9.46% price increase among firms
expensive – the average HA premium was –0.17, whose H-shares were at a relative discount, and
–0.19 and –0.35 on November 14, 17 November the Shanghai market experienced a 4.87% price
2014, 2014, and 27 February 2015, respectively. increase among firms whose A-shares were at a
However, despite the widening of the average level relative discount. On April 11, the price disparity
of price disparity, the cross-sectional variation in narrowed further, as the Hong Kong market
price disparity (as measured by SD) remained con- experienced a 3.24% price reduction among firms
siderably lower than the pre-announcement period whose H-shares were at a relative premium, and
(between 0.19 and 0.22). the Shanghai market experienced a mild 1.13%
Table 2 reports summary statistics related to the price reduction among firms whose A-shares were
daily returns of A-shares and H-shares during the at a relative premium.
week surrounding the policy announcement on
April 10. The results are reported by three separate
subgroups as defined by the firm’s HA premium V. Empirical results
Cointegration tests
Table 2. Summary statistics of daily returns of H-shares and
A-shares. Table 3 summarizes the results of cointegration tests
By HA premium on 9 April 2014 a that are conducted on all 61 firms in the analysis
Lower than Between Higher than sample. For each firm, the test is run separately
−10% −10% and +10% +10% All
using data from (1) the pre-announcement period
Number of firms 32 14 15 61
Daily returns of H-shares from 2 January 2014 to 9 April 2014, (2) the first
On 8 April 1.11% 2.51% 1.03% 1.37% post-announcement period from 10 April 2014 to 29
On 9 April 0.14% 0.79% 0.46% 0.35%
On 10 April 9.46% 1.16% –0.49% 5.22% July 2014, (3) the second post-announcement period
On 11 April –0.67% –1.87% –3.24% –1.58% from 30 July 2014 to 16 November 2014 and (4) the
On 12 April –0.94% –0.20% 0.72% –0.36%
Daily returns of A-shares post-implementation period from 17 November
On 8 April 1.53% 2.03% 1.90% 1.73% 2014 to 27 February 2015. We report results from
On 9 April 0.22% 0.47% 0.12% 0.24%
On 10 April 1.13% 1.06% 4.87% 2.10% two test methods: (1) Schwarz Bayesian information
On 11 April –1.13% –0.46% –0.49% –0.83% criterion (SBIC) (Gonzalo and Pitarakis (1998),
On 12 April –0.05% 0.07% –1.12% –0.31%
Note: aHA premium is defined as PH/PA – 1, where both the price of
Aznar and Salvador (2002)) and (2) trace statistic
H-shares (PH) and A-shares (PA) are expressed in Hong Kong dollars. method (Johansen 1991, 1995) at the 1% significance

Table 3. Results of cointegration tests.

Number of firms with cointegrated share prices
Pre-announcement Post-announcement Post-announcement Post-implementation
(2 January 2014 to (10 April 2014 to (30 July 2014 to (17 November 2014 to
9 April 2014) 29 July 2014) 16 November 2014) 27 February 2015)
HA premium on the Schwarz Trace Schwarz Trace Schwarz Trace Schwarz
day prior to policy Bayesian statistic Bayesian statistic Bayesian statistic Bayesian Trace statistic
announcement Number information method (1% information method (1% information method (1% information method (1%
(9 April 2014): of firms criterion sig. level)a criterion sig. level)a criterion sig. level)a criterion sig. level)a
Lower than −10% 32 4 0 9 6 1 0 15 11
−10% to +10% 14 2 0 5 3 2 0 3 0
Larger than +10% 15 2 1 12 7 0 0 3 2
All 61 8 1 26 16 3 0 21 13
Notes: The HA premium is PH/PA – 1, where both the price of H-shares (PH) and A-shares (PA) are expressed in Hong Kong dollars. A positive HA premium
implies that H-shares are more expensive. A negative premium implies that A-shares are more expensive.
a
See Johansen (1995).

level.14 The results are classified by three firm sub-
groups as defined by the firm’s HA premium on the
day prior to policy announcement.
The test results show that there are substantially
more firms with cointegrated share prices during
the first post-announcement period. Prior to policy
announcement, the SBIC and trace statistic meth-
ods detect cointegration between the prices of
H-shares and A-shares in eight firms and one
firm, respectively. By contrast, during the first
post-announcement period, the number of firms
with cointegrated share prices increases to 26 and
16, respectively. Therefore, as a whole, there is
stronger evidence for an equilibrium relationship
between the prices of H-shares and A-shares after
the policy announcement. In addition, all three
firm subgroups experience an increase in the num-
ber of firms with cointegrated share prices. The
effect is particularly notable among firms with a
large positive HA premium just prior to the policy
announcement.
The cointegration effects were quite different in
subsequent sub-periods. Interestingly, there is
almost no cointegration detected during the
second post-announcement period prior to reform
implementation. This appears to be in line with a
major confounding factor that was in play during
that period – the civil disobedience campaign
‘Occupy Central’ (or ‘Umbrella Revolution’), which
culminated in September and October, affected the
Hong Kong market exclusively during the period.
Beginning in September 2014, student strikes
developed into waves of demonstrations which led
to severe disruption of economic activity.15 By con-
trast, there is reasonably strong evidence for cointe-
gration during the post-implementation period. The
number of firms with cointegrated share prices
becomes 21 and 13 according to the SBIC and
trace statistic methods, respectively. This result is
somewhat surprising given the strong shift in market
sentiment towards the Shanghai market, and the
sharp widening of the average level of price disparity
between the Hong Kong and Shanghai markets
during the period. Nevertheless, the evidence for
14
We also test for cointegration using the trace statistic method at the 5% significance level. The results are similar and are available upon request.
15
The protests were a result of dissatisfaction with the decision by the National People’s Congress regarding proposed reforms to the Hong Kong electoral
system. The total number of protesters routinely exceeded 100 000, and key areas in central business districts were occupied by protesters and remained
closed to traffic for over 70 days. During the month prior to the official implementation of the campaign, the Hang Seng Index dropped by 10%, while the
Shanghai Stock Exchange Composite Index (SSE) increased by 6%.
APPLIED ECONOMICS 525
post-implementation cointegration is not as strong
as that in the first post-announcement period.
Pricing dynamics during the first post-
announcement period
Key VECM estimates
We will first focus on pricing dynamics during the
post-announcement period from 10 April 2014 to 29
July 2014, which contains the strongest evidence for
cointegration. Table 4 summarizes the estimation
results of 26 VECMs that use data from firms with
cointegrated share prices as determined by SBIC.
The firms are classified into three subgroups accord-
ing to their level of HA premium on the day prior to
policy announcement: (1) lower than –10%; (2)
between –10% and +10%; and (3) larger than
+10%. For each of the above subgroup and as a
whole, the table reports the mean of parameter esti-
mates, the SD (in parentheses) and the number of
firms with a parameter estimate that is different
from zero at the 5% significance level (in squared
brackets).
The top panel of the table reports summary sta-
tistics of parameter estimates from the cointegrating
equation, which represents the long-run equilibrium
relationship between the price of H-shares and
A-shares. The average cointegrating coefficient (β)
for all 26 firms is –1.237, and the SD is 0.611.
Therefore, as a whole, the result is consistent with
the theoretical prediction that the cointegrating coef-
ficient is –1 or, in other words, the relative law of
one price. While firms with a negative initial HA
premium (subgroup 1) tend to have a more negative
cointegrating coefficient, the other subgroups are
strongly consistent with this theoretical prediction.
In the cointegrating equation, the intercept coeffi-
cient tends to be smaller among firms with a higher
initial HA premium. This is broadly consistent with
theoretical predictions on the level of equilibrium, a
point which we will return to when the restricted
model is discussed. The average time trend coeffi-
cient is negative at –0.0012, which implies that
526 M. K. CHAN AND S. S. KWOK

Table 4. Summary statistics of key VECM estimates: post- equations, which represent short-run dynamics.
announcement period (10 April 2014–29 July 2014). The key results are the adjustment coefficients in
By HA premium on 9 April
2014 the H-share and A-share equations, which have an
Mean of parameter Lower Between Higher average value of –0.273 (αh ) and 0.134 (αa ), respec-
estimates than –10% and than tively. Therefore, as a whole, the price of H-shares
(SD in parentheses) All –10% +10% +10%
Cointegrating equation tends to adjust twice as fast as the price of A-shares
Log-price of A-shares (β) –1.237 –1.632 –1.158 –0.973 towards the equilibrium, and the overall magnitude
(0.611) (0.665) (0.597) (0.439)
Intercept (μ) 0.490 1.588 0.302 –0.255 of the error correction process is quite sizeable (i.e.
(1.434) (1.482) (0.929) (1.074) 0.273 + 0.134 = 0.407). Moreover, there is stronger
Linear time trend (γ) –0.0012 –0.0014 –0.0017 –0.0009
(0.0013) (0.0016) (0.0013) (0.0009) evidence that the H-share market is subject to such
[17] [5] [4] [8] an adjustment – 21 firms have a statistically signifi-
Daily return of H-shares
equation cant estimate in the H-share equation, but there are
Deviation from –0.273 –0.355 –0.265 –0.215
equilibrium (αh)
only 14 such firms in the A-share equation. Turning
(0.143) (0.101) (0.155) (0.145) to the subgroups, firms with a negative initial HA
[21] [9] [3] [9]
Lagged daily return of 0.065 0.026 0.201 0.036 premium tend to have a strong price adjustment in
H-shares (πhh1) H-shares but minimal price adjustments in A-shares;
(0.176) (0.165) (0.142) (0.180)
[6] [2] [2] [2] by contrast, firms with a positive initial HA pre-
Lagged daily return of –0.154 –0.037 –0.324 –0.170 mium tend to have symmetric price adjustments in
A-shares (πha1)
(0.181) (0.185) (0.122) (0.138) both markets.
[4] [0] [1] [3] The remaining coefficients in the VAR equations
Intercept (ch) 0.0008 0.0002 0.0012 0.0010
(0.0012) (0.0003) (0.0019) (0.0011) suggest that the relationship of prices in both mar-
[0] [0] [0] [0]
Daily return of A-shares
kets is largely determined by the cointegrating equa-
equation tion rather than short-run dynamics. For most firms,
Deviation from 0.140 0.075 0.101 0.204
equilibrium (αa) the stock return in one market has an insignificant
(0.134) (0.069) (0.136) (0.148) effect on the stock return in the other market the
[14] [4] [2] [8]
following day (πha 1 and π1 ). The equations also
ah
Lagged daily return of –0.009 0.029 –0.058 –0.017
A-shares (πaa1) suggest that the current stock return generally has
(0.178) (0.166) (0.305) (0.126)
[2] [1] [1] [0] an insignificant effect on the stock return in the
–0.035
Lagged daily return of
H-shares (πah1)
0.018 0.144 0.007
same market the following day (πhh 1 and πaa 1 ).
(0.128) (0.093) (0.165) (0.107) Similarly, the second-order lagged terms (πj for
[3] [1] [2] [0]
Intercept (ca) 0.0007 0.0007 0.0006 0.0007 j ¼ 2) are statistically insignificant in almost all the
(0.0008) (0.0009) (0.0011) (0.0007) cases (results not shown). Both markets do not have
[0] [0] [0] [0]
Number of firms with 26 9 5 12 a significant trend in price levels according to the
cointegrated share prices
short-run equations (ch and ca ).
Notes: The HA premium is PH/PA – 1, where both the price of H-shares (PH)
and A-shares (PA) are expressed in Hong Kong dollars. Numbers in Figures 2 to 4 focus on the 26 firms with coin-
parentheses are the SDs of parameter estimates among all firms with tegrated share prices, and plot each firm’s key model
cointegrated shares. Numbers in squared brackets are the number of
firms with a parameter estimate that is different from zero at the 5% estimates against its level of HA premium on the day
significance level. Coefficient estimates on second-order lagged daily prior to policy announcement. Figure 2 reports the
returns are not reported
point estimates and 95% confidence intervals of the
cointegrating coefficient β. Almost all the firms have
H-shares tend to become relatively more expensive a point estimate between 0 and –2; in addition, the
than A-shares over time. While 17 firms have a majority of the firms have a cointegrating coefficient
statistically significant estimate, the estimates are that is not different from –1 at the 5% significance
quite heterogeneous as indicated by a relatively level. Interestingly, firms with a highly negative
large SD. We will revisit the time trend coefficients initial HA premium tend to have a more negative
when the restricted model is discussed. cointegrating coefficient.
The next two panels of the table report summary Figures 3 and 4 report the point estimates and 95%
statistics of parameter estimates from the VAR confidence intervals of the price adjustment coefficient
 APPLIED ECONOMICS 527

1
cointegrating coefficient (β)
0
–1
–2
–3

–1 –.5 0 .5
HA premium on April 9, 2014

Figure 2. Estimate of cointegrating coefficient β (error bars denote 95% confidence interval).
0 .2
coefficient for H-shares (αH)
error-correction

–.6 –.4 –.2 –.8

–1 –.5 0 .5
HA premium on April 9, 2014

Figure 3. Estimate of the error correction coefficient in the H-share market αh (error bars denote 95% confidence interval).

in the H-share market (αh ) and the A-share market
(αa ), respectively. The point estimates for αh typically
range from –0.2 and –0.5, and the point estimates for
αa are typically smaller in magnitude (between +0.1
and +0.4). While the adjustment coefficients are rela-
tively homogeneous across firms, there is a slight
tendency for firms with a highly negative initial HA
premium to experience a more sensitive price adjust-
ment in the H-share market than the A-share market.
Determinants of error correction dynamics
The results above indicate that both the Hong Kong
and Shanghai markets contribute to the price dis-
covery process during the first post-announcement
period. In this section, we present further evidence
of the mechanism of price adjustment dynamics.
Figure 5 plots the price adjustment coefficients in
the A-share market (αa ) versus the negative value of
the price adjustment coefficients in the H-share
528 M. K. CHAN AND S. S. KWOK

.6 .8
coefficient for A-shares (αA)
error-correction

0 .2 .4 –.2

–1 –.5 0 .5
HA premium on April 9, 2014

Figure 4. Estimate of the error correction coefficient in the A-share market αa (error bars denote 95% confidence interval).
.6
error-correction coefficient for A-shares (αA)
.5
.4
.3
.2
.1
0
–.1

–.1 0 .1 .2 .3 .4 .5 .6
negative value of error-correction coefficient for H-shares (–αH)

45 degree line

Figure 5. Plot of error correction coefficients in both markets (αa versus –αh).

market (–αh ) for 26 firms with cointegrated share We further analyse how various factors contribute
prices. In 21 firms, the price of H-shares adjusts by a to the role of price discovery in both markets.
larger degree than the price of A-shares whenever Following Eun and Sabherwal (2003), we construct
the price levels are in disequilibrium. Thus, for the a measure for the Shanghai share of total adjustment
majority of the firms, the Hong Kong market plays a in prices:
smaller role in price discovery than the Shanghai
αa
market. This is consistent with the fact that the ShAdj ¼ (5)
Mainland market is often considered as the ‘home αa  αh
market’ of these firms where substantial information A special case is no price adjustment in the
is produced. Shanghai market (i.e. αa = 0). Then, ShAdj will be

zero, and the Hong Kong market is a ‘pure satellite’
of the Shanghai market (e.g. Garbade and Silber
1982). In our subsample of 26 firms, the mean of
ShAdj is 31.2% (the median is 32.2%), and the SD is
36.1%. Therefore, on average, the Shanghai market
contributes to approximately two-thirds of the price
discovery process. The estimate is strikingly similar
to Eun and Sabherwal (2003), who find that the
Canadian (home) market contributes to 61.9% of
price discovery among Canadian stocks that are
cross-listed in the US market. The similarity of the
results is particularly surprising given the vast insti-
tutional differences – the US and Canada markets
are open to each other, while Hong Kong and
Shanghai markets have been segmented until very
recently. In addition, the sample in Eun and
Sabherwal (2003) contains 62 cross-listed stocks. In
their analysis, market segmentation is not an issue so
all stocks can be used to compute the price adjust-
ment parameters. By contrast, although we have 61
cross-listed stocks in the sample, only 26 exhibit
cointegration. Only among these stocks can the
price adjustment parameters be meaningfully
calculated.
We focus on the following measures as potential
determinants of the price adjustment mechanism. In
a similar spirit to Eun and Sabherwal (2003), we
construct variables on the Hong Kong share of trad-
ing volume (HVolshare) and the bid–ask spread ratio
between both markets (Hong Kong divided by
Shanghai, denoted by SpreadRatio). A larger Hong
Kong share of trading volume implies more intense
competition from Hong Kong as well as greater
informativeness of the Hong Kong trading, which
should drive ShAdj higher. By contrast, a higher
spread ratio suggests that the Hong Kong stock
exchange poses a smaller competitive threat and
carries less informative trading, which should drive
ShAdj lower. As in Cai, McGuinness, and Zhang
(2011), we also include two variables that capture
differential market sentiment effects (VolRelative)
and information asymmetries (BASRelative) between
both markets, respectively. A larger value of either
variable may reduce the magnitude of the price
adjustment process. We include the log of market
16
Sample size restrictions preclude us from further classifying the stocks into multiple industries.
17
The coefficient on BASRelative is positive and statistically significant. Since the dependent variable is the share of price adjustment, the expected sign of the
coefficient is ambiguous in this regression.
APPLIED ECONOMICS 529
capitalization (LMktCap) to control for firm size. In
addition, we control for the level of HA premium
just prior to policy announcement, which may
reflect other factors that are not explicitly controlled
for in the regression analysis. Finally, for each set of
regression, we present estimates with and without
controlling for industry effects. In particular,
stocks from the banking and insurance sector
(‘financials’) tend to have a larger market capitaliza-
tion and have generally traded at lower relative
valuations in Shanghai in recent times. Stocks
in this sector are captured by a dummy vari-
able (BankingandInsurance).16
Table 5 reports estimation results from four sets
of regression analysis. The even columns control for
industry while the odd columns do not. In columns
1 and 2, the dependent variable is the Shanghai share
of total price adjustment (ShAdj). The coefficient
estimates (in column 2) on the Hong Kong share
of trading volume and the bid–ask spread ratio are
0.826 and –0.332, respectively, which have expected
signs and are statistically significant at the 10% and
5% levels. Both results confirm the key findings in
Eun and Sabherwal (2003) – increased competition
and greater informativeness from the Hong Kong
market will increase Shanghai’s price adjustment
share and therefore reduce its relative role of price
discovery.17 All else being equal, stocks in the bank-
ing and insurance sector tend to have a lower
Shanghai share of total price adjustment. Failure to
control for the sector has a minimal effect on the key
coefficients of the regression; however, the coeffi-
cient on market capitalization becomes statistically
significant.
The next four columns (3–6) report estimation
results related to the size of price adjustment
coefficients in the Shanghai and Hong Kong markets
(αa and  αh ), respectively. The results allow us to
evaluate how the above factors are related to price
adjustment in each individual market. The size of
the price adjustment coefficient in the Shanghai
market tends to be more related to the explanatory
variables, as more coefficients are statistically signif-
icant or at the borderline of being significant at the
10% level. For instance, the Hong Kong share of
530 M. K. CHAN AND S. S. KWOK

Table 5. Determinants of error correction dynamics.

ShAdj αa/(αa-αh) ShAdj αa/(αa-αh) αa αa –αh –αh αa–αh αa–αh
Dependent variable (1) (2) (3) (4) (5) (6) (7) (8)
Intercept 0.235 0.156 0.082 0.053 0.234** 0.226* 0.316*** 0.279**
(0.231) (0.223) (0.087) (0.085) (0.110) (0.115) (0.117) (0.115)
HA premium on April 9 0.300 0.277 0.174 0.166 –0.072 –0.075 0.102 0.092
(0.291) (0.276) (0.110) (0.105) (0.139) (0.142) (0.148) (0.142)
LMktCap –0.186** –0.113 –0.064** –0.037 0.006 0.014 –0.058 –0.024
(0.067) (0.076) (0.025) (0.029) (0.032) (0.039) (0.034) (0.039)
HVolShare 1.043** 0.826* 0.343** 0.263 –0.046 –0.067 0.297 0.195
(0.413) (0.410) (0.156) (0.156) (0.197) (0.211) (0.209) (0.210)
SpreadRatio –0.361** –0.332** –0.102* –0.092 0.153* 0.156* 0.051 0.064
(0.155) (0.147) (0.058) (0.056) (0.074) (0.076) (0.079) (0.076)
VolRelative 0.448 0.421 0.236 0.227 –0.006 –0.008 0.231 0.218
(0.371) (0.352) (0.140) (0.133) (0.177) (0.181) (0.188) (0.181)
BasRelative 2.726** 2.610** 0.853** 0.809** –0.841 –0.853 0.012 –0.044
(1.028) (0.977) (0.388) (0.370) (0.491) (0.504) (0.521) (0.501)
Banking and Insurance –0.303* –0.112 –0.030 –0.142
(0.171) (0.065) (0.088) (0.088)
R2 0.536 0.605 0.519 0.587 0.328 0.333 0.198 0.300
Notes: Number of observations = 26. The dependent variables are based on coefficients from the VEC model using data from the first post-announcement
period. LMktCap is the logarithm of market capitalization (MktCap). HVolShare is defined as HVol/(HVol + AVol), where HVol and AVol are the trading
volumes of H- and A-shares, respectively. SpreadRatio is defined as HBas/ABas, where HBas and ABas are the percentage bid–ask spreads of H- and
A-shares, respectively. VolRelative is defined as |HVol – AVol|/(HVol + AVol). BasRelative is defined as |HBas – ABas|/(HBas + ABas). Banking and Insurance is a
dummy variable that is equal to 1 if the stock belongs to the sector, 0 otherwise. SEs are given in parentheses.
*Significant at the 10% level; **significant at the 5% level; ***significant at the 1% level.

trading volume is positively related to the size of price
adjustment in the Shanghai market, but it has a very
weak relationship with the size of price adjustment in
the Hong Kong market. The coefficient on the bid–ask
spread ratio has expected sign and is statistically sig-
nificant in both sets of regressions – a higher relative
bid–ask spread in the Hong Kong market will reduce
the size of price adjustment in the Shanghai market
and increase the size of price adjustment in the Hong
Kong market. By contrast, the coefficients on differen-
tial market sentiment (VolRelative) and information
asymmetries (BASRelative) deliver mixed evidence, as
both have positive signs in the Shanghai market, but
have negative signs in the Hong Kong market. The last
column of the table uses the total magnitude of price
adjustment (i.e. αa  αh ) as the dependent variable.
Not surprisingly, since the coefficients in the previous
two regressions have opposite signs, the coefficients in
this regression are generally close to zero and statisti-
cally insignificant.
The initial HA premium is not statistically sig-
nificant at the 10% level in all regressions. As a
robustness check, we also run the above regres-
sions using the initial HA premium and log of
market capitalization as the only explanatory vari-
ables (results not shown). The signs of the coeffi-
cients are the same as the original set of
regressions, and the magnitudes remain similar.
Interestingly, the initial HA premium becomes
more statistically significant in the first three
regressions. Therefore, the level of initial HA pre-
mium may merely reflect information about the
stock’s trading characteristics and does not appear
to play a critical role in directly determining the
price adjustment process when other factors are
taken into account.
Convergence of price disparity in equilibrium
To further investigate the implications of the VECM,
we estimate a restricted version of the model in
which the cointegrating coefficient β is set a priori
to –1. Under this restriction, the cointegrating equa-
tion can be written as
   
log PtH  log PtA þ μ þ γt ¼ 0
which implies that the equilibrium level of HA pre-
mium is a simple exponential function of the time
trend γt:
PtH
 1 ¼ eμγt  1
PtA
Both the intercept and the trend coefficients have
a straightforward interpretation. The equilibrium
HA premium on the first day of policy announce-
ment (April 10, or t ¼ 0) is equal to a constant
denoted by eμ  1. The trend coefficient γ can be
interpreted as the average daily rate of decrease of
the equilibrium HA premium level.
The restricted model is estimated for each of the
above 26 firms with cointegrated share prices.

.5
on April 10, 2014 ( exp(–μ)–1)
Equilibrium HA premium
0
–.5
–1
–1.0
Figure 6. Convergence of price disparity in equilibrium.
Figure 6 plots the restricted model’s equilibrium HA
premium on the first day of policy announcement
(eμ  1) against the actual HA premium one day
prior to announcement. The points form a positive
relationship that is flatter than the 45° line, which
implies that the equilibrium HA premia of these
firms are closer to zero than the actual HA premia
prior to policy announcement. Therefore, the price
disparity between H-shares and A-shares has
converged not only in the observed data but also in
terms of the level of the equilibrium implied by the
model. The results also reveal some interesting asym-
metries regarding the degree of convergence. For
instance, among firms with an initial HA premium
of larger than +10% on April 9, the equilibrium is on
average 40% lower than the initial premium; among
firms with an initial HA premium of smaller than
−10% on April 9, the equilibrium is on average 12%
closer to zero than the initial premium.
The trend coefficients (γ) from the restricted model
are very similar to the estimates from the unrestricted
model. Among firms with an initial HA premium of
larger than +10% on April 9, the average trend coeffi-
cient is −0.0009; among firms with an initial HA pre-
mium of lower than –10% on April 9, the average trend
coefficient is −0.0015. While both numbers imply that
the equilibrium HA premium becomes larger over
18
We also attempt to classify the stocks according to their level of HA premium just prior to reform implementation. However, comparison with Table 4 then
becomes difficult, as only one firm belongs to subgroup 3 under this classification.
APPLIED ECONOMICS 531
–0.5 0.0 0.5
HA premium on April 9, 2014
45 degree line
time, the change occurs 1.5 times as fast among firms
with a negative initial HA premium. Interestingly,
among firms whose H-shares were traded at a deep
discount to A-shares, the trend coefficients are sub-
stantially more negative (and also highly statistically
significant) – they range from –0.003 to –0.005 among
the three firms with the most negative HA premium.
The above results indicate that net of the overall mar-
ket effect, the equilibrium price discount of H-shares
diminishes over time, a finding that is broadly consis-
tent with Cai, McGuinness, and Zhang (2011).
Pricing dynamics during the post-implementation
period
We now focus on pricing dynamics during the post-
implementation period from 17 November 2014 to
27 February 2015. As indicated in Table 3, 21 firms
exhibit cointegrated share prices during this period
according to the SBIC method. Table 6 summarizes
the estimation results of 21 VECMs that use data
from these firms during the post-implementation
period. To facilitate comparison, these firms are
classified into the same three subgroups as in
Table 4 (i.e. according to their level of HA premium
on 9 April 2014): (1) lower than −10%; (2) between
−10% and +10%; and (3) larger than +10%.18 For
532 M. K. CHAN AND S. S. KWOK

Table 6. Summary statistics of key VECM estimates: post-imple- equation, which represents the long-run equilibrium
mentation period (17 November 2014–27 February 2015). relationship between the prices of H-shares and
By HA premium on 9 April
2014 A-shares. The average cointegrating coefficient (β)
Mean of parameter Lower Between Higher for all 21 firms is –0.302, and the SD is 0.525.
estimates than –10% and than Therefore, as a whole, the stocks exhibit cointegra-
(SD in parentheses) All –10% +10% +10%
Cointegrating equation tion during this period but there is mixed evidence
Log-price of A-shares (β) –0.302 –0.337 –0.173 –0.259 for the relative law of one price. The average time
(0.525) (0.616) (0.241) (0.076)
Intercept (μ) –0.858 –0.614 –1.600 –1.334 trend coefficient is positive at 0.0007. Both results
(1.326) (1.473) (0.847) (0.250) are consistent with sharp widening of the overall
Linear time trend (γ) 0.0007 0.0009 0.0007 –0.0007
(0.0017) (0.0016) (0.0028) (0.0004) level of HA price disparity during the period.
[13] [9] [2] [2] The next two panels of the table report summary
Daily return of H-shares
equation statistics of parameter estimates from the VAR equa-
Deviation from –0.376 –0.318 –0.462 –0.583
equilibrium (αh)
tions, which represent short-run dynamics. The key
(0.210) (0.151) (0.121) (0.408) results are the adjustment coefficients in the H-share
[19] [14] [3] [2]
Lagged daily return of 0.143 0.128 0.161 0.200 and A-share equations, which have an average value of
H-shares (πhh1) –0.376 (αh ) and –0.085 (αa ), respectively. These results
(0.126) (0.135) (0.136) (0.080)
[5] [4] [1] [0] are driven in part by the more extreme values in sub-
Lagged daily return of –0.039 –0.032 –0.017 –0.094 group 3; the average adjustment coefficients in sub-
A-shares (πha1)
(0.092) (0.090) (0.128) (0.079) group 1 are –0.318 and –0.011, respectively.
[1] [1] [0] [0] Nevertheless, the H-share market exhibits much stron-
Intercept (ch) 0.0001 0.0001 –0.0009 0.0013
(0.0023) (0.0020) (0.0015) (0.0040) ger price adjustments from disequilibrium – as a
[1] [1] [0] [0]
Daily return of A-shares
whole, 19 firms have a statistically significant adjust-
equation ment coefficient in the H-share equation, but there are
Deviation from –0.085 –0.011 –0.172 –0.367
equilibrium (αa) only 4 such firms in the A-share equation. The remain-
(0.279) (0.164) (0.237) (0.604) ing coefficients in the VAR equations suggest that the
[4] [2] [1] [1]
Lagged daily return of –0.027 –0.034 0.021 –0.039 relationship of prices in both markets is largely deter-
A-shares (πaa1) mined by the cointegrating equation rather than short-
(0.148) (0.167) (0.118) (0.091)
[1] [1] [0] [0] run dynamics. However, there is some evidence that
Lagged daily return of 0.088 0.083 0.087 0.114 the stock return in the H-share market has a positive
H-shares (πah1)
(0.259) (0.286) (0.105) (0.297) effect on the stock return in A-share the following day
[5] [4] [0] [1]
Intercept (ca) 0.0029 0.0033 0.0028 0.0015 (πah
1 ). This is consistent with the growing northbound
(0.0024) (0.0027) (0.0014) (0.0006) investment activities during the period.
[0] [0] [0] [0]
Number of firms with 21 15 3 3
cointegrated share prices
Notes: The HA premium is PH/PA – 1, where both the price of H-shares (PH)
and A-shares (PA) are expressed in Hong Kong dollars. Numbers in Sensitivity analysis
parentheses are the SDs of parameter estimates among all firms with
cointegrated shares. Numbers in squared brackets are the number of
firms with a parameter estimate that is different from zero at the 5%
Further empirical specifications
significance level. Coefficient estimates on second-order lagged daily A number of sensitivity analyses are conducted on
returns are not reported.
the baseline results. For instance, we estimate a
VECM with more lags in the VAR component,
each of the above subgroup and as a whole, the table with little change in the estimation results. While
reports the mean of parameter estimates, the SD (in our cointegration test results (e.g. trace statistic
parentheses), and the number of firms with a para- method in Table 3) already rule out stationarity of
meter estimate that is different from zero at the 5% both series, we conduct separate unit root tests on
significance level (in squared brackets). We will price data for robustness check and confirm that
focus on discussing the results as a whole and in almost all the data series have a unit root. We also
subgroup 1, which contains 15 stocks. conduct a two-step cointegration test (Engle and
The top panel of Table 6 reports summary statis- Granger (1987)), which reveals a similar pattern to
tics of parameter estimates from the cointegrating Table 3. Interestingly, the two-step method tends to

identify more firms with cointegrated share prices
than the trace statistic method at the same signifi-
cance level.
We also perform cointegration tests based on an
earlier pre-announcement period from 2 January
2013 to 31 December 2013. During the period, there
are three and two firms with significant cointegrated
share prices under the SBIC and trace statistic meth-
ods, respectively. Therefore, these results are consistent
with the negative result during the pre-announcement
period in Table 3. Another interesting result is based
on estimation using a shorter post-announcement per-
iod. For instance, using 9 June 2014 as the last observa-
tion for the first post-announcement period, there are
34 and 19 firms with significant cointegrated share
prices under the SBIC and trace statistic methods,
respectively. While the tests detect more firms with
cointegrated share prices, estimation results from the
VECM suggest that the cointegrating coefficient β
tends to have larger SEs and is generally farther away
from the hypothesized value of −1.
Cointegration analysis in Table 3 shows that
among the 61 firms in the analysis sample, 35
firms do not have cointegrated share prices during
the first post-announcement period. We extend the
empirical analysis to these 35 remaining firms.19
Since these firms do not pass the cointegration
tests, the results should be treated with caution
because the VECM may be misspecified.
Nevertheless, since a number of firms are on the
borderline of passing the tests, the model estimates
could be informative. Upon estimation, we find that
there are seven firms with rather extreme values of
the cointegrating coefficient β (larger than 5 or
smaller than –5) and large SEs. To minimize the
impact of outliers, these firms are removed from
subsequent analysis. Of the remaining subsample,
the average cointegrating coefficient is –1.26 with
an SD of 1.40. Therefore, the results confirm the
validity of the relative law of one price, even though
the distribution of the estimates tends to be wider.20
We also investigate the price adjustment coeffi-
cients (αa and αh ) of these firms. Interestingly, the
price adjustment coefficients tend to be smaller in
19
As discussed earlier, the raw sample contains 68 firms concurrently listed in Hong Kong and Shanghai markets, 7 of which are dropped from the raw
sample. We also attempt to extend the empirical analysis to these firms (when trading is not suspended). There is little difference in the result across the
sub-periods, as few of these stocks exhibit cointegration.
20
Seven firms have a cointegrating coefficient of lower than –2, 18 firms have a value between 0 and –2, and three firms have a small positive value. The
average SE is 0.50, which is considerably larger than firms with cointegrated share prices (0.19).
APPLIED ECONOMICS 533
magnitude. The average size of the price adjustment
coefficients in the Hong Kong and Shanghai markets
are −0.14 and 0.01, respectively. Therefore, the aver-
age total magnitude of price adjustment (αa  αh ) is
only 0.13, which is one-third as large as in firms with
cointegrated share prices (0.407). The slow error
correction dynamics is also fully consistent with the
weaker equilibrium relationship as indicated by the
cointegration tests.
Shenzhen market as a control group and recent
developments
In 2014, there were 16 firms concurrently listed in
the Hong Kong and Shenzhen markets. Since the
financial reform does not allow for cross-market
investment between Shenzhen and Hong Kong mar-
kets, these firms are unaffected and can therefore
form a natural control group for analysis. We per-
form an empirical analysis on these firms as well.
However, the results should be treated with caution
and should be interpreted in the context of recent
policy developments. Unlike Shanghai-Hong Kong
Stock Connect, which has a clean policy announce-
ment date, investors may anticipate that the liberal-
ization will be extended to the Shenzhen market
eventually, even before the policy is officially
announced.
The cointegration tests confirm our baseline
results. During the pre-announcement period, four
firms are found to exhibit cointegration under the
SBIC method. The number of firms with cointe-
grated share prices are three and zero during the
first and second post-announcement periods, respec-
tively. Therefore, firms concurrently listed in the
Shenzhen and Hong Kong markets do not exhibit
stronger price comovements after the policy
announcement and before reform implementation.
Interestingly, during the post-implementation
period, eight firms are found to exhibit cointegrated
share prices. Although the implementation date of
Shenzhen-Hong Kong Stock Connect has never
been announced, the stronger price comovement is
consistent with key policy developments that
534 M. K. CHAN AND S. S. KWOK
emerged in late 2014 and early 2015. On 4 December
2014, the Authority of Qianhai, which oversees the
development of the prospective Qianhai Shenzhen-
Hong Kong Modern Service Industry Cooperation
Zone, announced a key strategic work document
regarding the deepening of integration between
Hong Kong and Shenzhen markets. Shenzhen-
Hong Kong Stock Connect was first described offi-
cially in this document. The Qianhai New District,
scheduled to complete by 2020 in Qianhai,
Shenzhen, is a pilot zone for cooperation between
mainland China and Hong Kong and innovation in
the service industry, particularly financial policies. It
is also one of the few Free Trade Zones in China. On
5 January 2015, Chinese Prime Minister Li Keqiang
visited Shenzhen and emphasized that Shenzhen-
Hong Kong Stock Connect should come after
Shanghai-Hong Kong Stock Connect. Routine meet-
ings were subsequently held between Shenzhen and
Hong Kong municipalities regarding the implemen-
tation and logistics of the prospective policy.
VI. Conclusion
In this article, we studied the impact of Shanghai-
Hong Kong Stock Connect on the dynamic relation-
ship between H- and A-share prices of firms that
were concurrently listed on both stock exchanges.
Through a VECM and a microscopic analysis on 61
pairs of price series, we obtained the following find-
ings. First, more cross-listed firms exhibited a coin-
tegrating relationship between their H- and A-share
prices after the policy announcement, particularly
between the announcement date of 10 April 2014
and 29 July 2014. During this period, the cointegrat-
ing coefficients were close to unity, indicating that
the relative law of one price held in the post-
announcement equilibrium. Second, by using the
policy as a natural experiment and leveraging on
the highly heterogeneous price disparity before the
announcement, our analysis revealed that the price
disparity between H- and A-shares tightened under
equilibrium right after the policy announcement,
which affected subsequent pricing dynamics. Third,
we compared the roles played by Hong Kong and
Shanghai markets in price discovery. The analysis
showed that, on average, a greater portion of dise-
quilibrium adjustment occurred on H-share prices,
suggesting that the Shanghai market acted as the
‘home market’. We also found that the price discov-
ery role of the Shanghai market was positively
related to its share of trading volume and negatively
related to its percentage bid–ask spread compared to
the Hong Kong market. This corroborated with ear-
lier studies on the price discovery of cross-listed
stocks, which found that the home market played a
larger role in information dissemination (e.g. Eun
and Sabherwal 2003). Finally, we found substantial
price comovements (especially in the form of a coin-
tegrating relationship) between A-shares and
H-shares of cross-listed firms after the reform
started on 17 November 2014. This highlighted the
importance of modelling pricing dynamics –
although the level of price disparity between
A-shares and H-shares generally widened during
this period, it does not necessarily imply that both
markets have become more segmented. Indeed, our
model revealed rich equilibrium relationships and
short-run dynamics between both markets, which
would have been overlooked in a static analytical
framework.
Our findings shed new light on the effects of
capital liberalization policy on financial markets.
The announcement of Shanghai-Hong Kong Stock
Connect set the perfect stage for analysing the
implications of capital control for pricing disparity
and dynamics. Although the policy did not launch
until 7 months after the announcement, investors’
expectations seemed to have a quick impact on the
prices of cross-listed stocks in both markets. At
least in the short run, the prices of H- and
A-shares reacted to disequilibrium shocks asym-
metrically and adjusted to an equilibrium relation-
ship defined by a cointegrating equation. This was
in line with the original aims of the reform, which
were to enhance financial integration and lower
the barrier of capital flow between both financial
markets. While the reform imposed upper limits
on the daily and overall amounts of capital flow
between both markets, our study will have impor-
tant implications for the price dynamics of cross-
listed stocks when there are further unanticipated
attempts to lower the capital flow barrier in the
future.
As a final note, our study only focused on the
policy’s impact on two markets. It would be inter-
esting to extend the analysis to a multimarket level
and investigate the policy implication for closely

knitted trading platforms such as Shenzhen and
other international markets, on which stocks are
cross-listed. In fact, our preliminary analysis on the
Shenzhen market revealed interesting patterns that
were in line with recent developments of the pro-
spective Shenzhen-Hong Kong Stock Connect. This is
left for future research.
Acknowledgement
We would like to thank an anonymous referee for providing
comments that substantially improved the article.
Disclosure statement
No potential conflict of interest was reported by the authors.
References
Arquette, G., W. O. Brown Jr, and R. Burdekin. 2008.
“US ADR and Hong Kong H-share Discounts of
Shanghai-Listed Firms.” Journal of Banking & Finance
32: 1916–1927. doi:10.1016/j.jbankfin.2007.12.019.
Aznar, A. and M. Salvador. 2002. “Selecting the Rank of the
Cointegration Space and the Form of the Intercept
using an Information Criterion.” Econometric Theory 18:
926–947. doi:10.1017/S0266466602184064.
Cai, C. X., P. B. McGuinness, and Q. Zhang. 2011. “The
Pricing Dynamics of Cross-Listed Securities: The Case of
Chinese A- and H-shares.” Journal of Banking & Finance
35: 2123–2136. doi:10.1016/j.jbankfin.2011.01.010.
Chang, E. C., Y. Luo, and J. Ren. 2013. “Cross-Listing and
Pricing Efficiency: The Informational and Anchoring
Role Played by the Reference Price.” Journal of
Banking & Finance 37: 4449–4464. doi:10.1016/j.
jbankfin.2012.12.018.
Chari, A. and P. Henry. 2004. “Risk-Sharing and Asset Prices:
Evidence from a Natural Experiment.” The Journal of
Finance 59: 1295–1324. doi:10.1111/jofi.2004.59.issue-3.
Choi, O. T., H. Wong, C. K. F. Yiu, and M. Yu. 2013. “In Depth
Analysis of the Dually Listed Companies in Hong Kong and
China Stock Markets Prior and Posterior to the Global
Financial Turmoil.” International Journal of Economics and
Finance 5: 100–110. doi:10.5539/ijef.v5n10p100.
Chung, T.-K., C.-H. Hui, and K.-F. Li. 2013. “Explaining
Share Price Disparity with Parameter Uncertainty:
Evidence from Chinese A- and H-Shares.” Journal of
Banking & Finance 37: 1073–1083. doi:10.1016/j.
jbankfin.2012.11.004.
APPLIED ECONOMICS 535
Eichengreen, B. 2001. “Capital Account Liberalization: What
Do Cross-Country Studies Tell Us?” The World Bank
Economic Review 15: 341–365. doi:10.1093/wber/15.3.341.
Engle, R. F. and C. W. J. Granger. 1987. “Co-integration and
Error Correction: Representation, Estimation, and
Testing.” Econometrica 55: 251–276. doi:10.2307/1913236.
Eun, C. S. and S. Sabherwal. 2003. “Cross-Border Listings
and Price Discovery: Evidence from U.S.-Listed Canadian
Stocks.” The Journal of Finance 58: 549–576. doi:10.1111/
jofi.2003.58.issue-2.
Galindo, A., F. Schiantarelli, and A. Weiss. 2007. “Does
Financial Liberalization Improve the Allocation of
Investment? Micro Evidence from Developing
Countries.” Journal of Development Economics 83:
562–587. doi:10.1016/j.jdeveco.2005.09.008.
Garbade, K. and W. Silber. 1982. “Price Movements and
Price Discovery in Futures and Cash Markets.” Review of
Economics and Statistics 64: 289–297.
Gonzalo, J. and J.-Y. Pitarakis. 1998. “Specification via
Model Selection in Vector Error Correction Models.”
Economics Letters 60: 321–328. doi:10.1016/S0165-1765
(98)00129-3.
Harris, F. H., T. H. McInish, G. L. Shoesmith, and R. A.
Wood. 1995. “Cointegration, Error Correction and Price
Discovery on Informationally Linked Security Markets.”
The Journal of Financial and Quantitative Analysis 30:
563–579. doi:10.2307/2331277.
Harrison, A., I. Love, and M. McMillan. 2004. “Global
Capital Flows and Financing Constraints.” Journal of
Development Economics 75: 269–301. doi:10.1016/j.
jdeveco.2003.10.002.
Henry, P. 2000. “Stock Market Liberalization, Economic
Reform, and Emerging Market Equity Prices.” The Journal
of Finance 55: 529–564. doi:10.1111/jofi.2000.55.issue-2.
Henry, P. 2007. “Capital Account Liberalization: Theory,
Evidence, and Speculation.” Journal of Economic
Literature 45: 887–935. doi:10.1257/jel.45.4.887.
Johansen, S. 1988. “Statistical Analysis of Cointegration
Vectors.” Journal of Economic Dynamics and Control 12:
231–254. doi:10.1016/0165-1889(88)90041-3.
Johansen, S. 1991. “Estimation and Hypothesis Testing of
Cointegration Vectors in Gaussian Vector Autoregressive
Models.” Econometrica 59: 1551–1580. doi:10.2307/
2938278.
Johansen, S. 1995. Likelihood-Based Inference in Cointegrated
Vector Autoregressive Models. Oxford: Oxford University
Press.
Su, Q., T. T.-L. Chong, and I. K.-M. Yan. 2007. “On the
Convergence of the Chinese and Hong Kong Stock
Markets: A Cointegration Analysis of the A and
H Shares.” Applied Financial Economics 17: 1349–1357.
doi:10.1080/09603100600993760.