Vous êtes sur la page 1sur 18

Journal of International Money and Finance 25 (2006) 894e911

www.elsevier.com/locate/jimf

A currency union for Hong Kong and Mainland China?


Xinpeng Xu*
School of Accounting and Finance, The Hong Kong Polytechnic University, Kowloon, Hong Kong,
Peoples Republic of China

Abstract

This paper investigates the degree and the evolution of business cycle synchronization between Hong
Kong SAR and Mainland China, as well as Hong Kong SAR and the United States with a two-step pro-
cedure. A structural VAR model is first set up to capture the dynamics of shocks to different regions. Then
a state-space model is used to gauge the degree of symmetry of regional shocks, and to examine whether
business cycles have been more synchronized over time. Our results cast doubt on the feasibility of a cur-
rency union between Hong Kong and Mainland China in the foreseeable future.
2006 Elsevier Ltd. All rights reserved.

JEL classification: E32; F15; O18; O53

Keywords: Hong Kong; China; Business cycle synchronization; Structural VAR; State-space model

1. Introduction

The currency board system in Hong Kong SAR,1 which was put in place in October 1983,
has recently been the focus of intense debate.2 While the linked exchange rate system has con-
tributed significantly to the development of Hong Kong as an international financial and trade
centre, by instilling stability in the foreign exchange market and ensuring credibility in

* Tel.: 852 2766 7139; fax: 852 2774 9364.


E-mail address: afxxu@inet.polyu.edu.hk
1
On July 1, 1997, Hong Kong became a special administrative region (SAR) of China after more than a hundred years
of British colonial rule. We use region instead of country to refer to the three economies under study, Hong Kong
SAR, Mainland China and the United States.
2
See, for example, Devereux (2002) and Kueh and Ng (2002).

0261-5606/$ - see front matter 2006 Elsevier Ltd. All rights reserved.
doi:10.1016/j.jimonfin.2006.07.001
X. Xu / Journal of International Money and Finance 25 (2006) 894e911 895

government policy, its inflexibility in both exchange rate and interest rate policies in facilitating
external and internal adjustment has been called into question, especially in times of economic
difficulty such as the Asian financial crisis and the deflation in 2000 through to 2004.3 By virtue
of the US dollar peg exchange rate system, Hong Kong has to forfeit its own monetary policy
and follow that of the United States. If Hong Kongs business cycle shocks are synchronized
with those of the United States, by following the monetary policy of the United States,
Hong Kong would be able to minimize the destructive effect of business cycle fluctuation.
However, as suggested by various observers,4 the China Factor has been playing an increas-
ingly important role in Hong Kongs economy, and Hong Kongs business cycle synchroniza-
tion may have been deviating, if at all, from that of the United States towards that of Mainland
China. The key issue is, therefore, the extent of business cycle synchronization between Hong
Kong and the United States, as well as Hong Kong and Mainland China, over time.
Although the issue of business cycle synchronization between Hong Kong and other econ-
omies has been the centre of debate recently, there is a paucity of research to empirically quan-
tify the extent of business cycle synchronization between Hong Kong and the United States, as
well as Hong Kong and Mainland China. For example, Kueh and Ng (2002) in their detailed
accounts of the interplay of the China Factor and the US dollar peg in the Hong Kong econ-
omy, provide only descriptive statistics about GDP growth and inflation between Hong Kong
and the US. Ma and Tsang (2002) constitute the first attempt in examining whether the two
economies have progressed to a stage where monetary unification is justified. The principal
component approach allows them to quantify the degree of business synchronization between
Hong Kong and China. However, the use of annual data that spans from Chinas central plan-
ning period in the 1950s to the 1990s, without taking into account structural breaks, may intro-
duce bias in the estimation, as China experiences significant structural changes since the reform
in 1978. As admitted by the authors, the GDP data after 1978 are too short to apply the prin-
cipal component analysis for China and Hong Kong as a whole (p. 11). Nevertheless, their
results point to a sceptical answer even for the post-1997 era. Liang (1999), on the other
hand, looks at the behaviour of the long-run real exchange rate (RER) of Hong Kong and China
by testing the generalized-purchasing power parity (G-PPP) hypothesis. The rationale is that if
the fundamental variables determining RERs are sufficiently integrated, as in a currency area,
the RERs should share common trends. The limitation of this approach, however, is that the
nominal exchange rate in China and, to a lesser extent, market prices were under control by
the government before the mid 1990s. With the period of research running from January
1979 to June 1998, movements of RER in Mainland China may not reflect underlying market
forces in early years, and therefore trends in RER between Hong Kong and Mainland China
may not reveal meaningful information about the integration between them.
In contrast with the previous studies, this paper attempts to examine quantitatively both the
degree and the evolution of business cycle synchronization between Hong Kong and the United
States, as well as Hong Kong and Mainland China. In making use of the recent advance in time

3
Various suggestions have been made, ranging from totally de-linking the peg exchange rate system into a flexible
exchange rate system, to more modest reform by devaluation but still maintaining the currency board system. Possible
future peg with Chinas Renminbi (RMB) when the RMB becomes fully convertible has also been put forward. Of
course, the official stance has always been firm in maintaining the current system, which is essential for a currency
board to function effectively.
4
See Kueh and Ng (2002).
896 X. Xu / Journal of International Money and Finance 25 (2006) 894e911

series modelling techniques, together with the recently available high frequency data for China,
we first derive structural supply and demand shocks from a structural vector auto-regression
model to capture the dynamics of shocks to different economies,5 that is, Hong Kong, Mainland
China and the United States. We then investigate the degree of synchronization of both supply
and demand shocks between these economies over time. The setup of a state-space model in the
second step allows us to decompose the shocks, both supply and demand, into common and
region-specific shocks, to gauge the degree of symmetry of regional shocks, and to examine
whether regional shocks have been more synchronized over time.
Our investigation is related to two strands of large and growing literature: one is optimal
currency areas (OCA),6 pioneered by Mundell (1961) and McKinnon (1963) and, the other
is intra-national business cycle (see, e.g., Hess and van Wincoop, 2000). The theory of OCA
asks what the appropriate domain of a currency area is. If two regions experience the same
shocks, they will presumably favour the same policy responses and thus they are good candi-
dates for a currency union. Specifically, business cycle synchronization makes a currency union
an optimal monetary arrangement. Recent empirical research also finds that the correlations of
business cycle shocks are endogenous, in that stronger trade ties may lead to stronger correla-
tions of business cycle disturbances (Frankel and Rose, 1998),7 which has significant implica-
tions for Hong Kong as the China Factor is becoming increasingly important.
Our study of the degree of asymmetry of regional business cycle disturbances between Hong
Kong and Mainland China is also related to the literature on intra-national business cycles with
which most of the research to date focuses e on cycles within and/or among the developed
countries (see, e.g., Hess and van Wincoop, 2000; Xu, 2002). The recent advance of integration
in Europe and elsewhere has blurred national economic boundaries. International business cy-
cles are likely to be increasingly more like intra-national business cycles. Indeed, an under-
standing of the empirics of intra-national business cycles may prove useful for predicting the
features of international business cycles in the future (Hess and Shin, 1997).
Our paper is structured as follows. Section 2 describes in detail the time series techniques we
use in our empirical estimation. This is followed by a discussion of data. In Section 4, the
econometric results are reported and interpretations are provided. Section 5 provides an alter-
native setup to check the sensitivity of our empirical findings. Section 6 concludes the paper.

2. Methodology

To examine, firstly, the degree of asymmetry of business cycle disturbances between Hong
Kong and Mainland China, as well as Hong Kong and the United States, and secondly, the dy-
namics of the asymmetry over time, we employ a two-step procedure in our estimation. A struc-
tural vector auto-regressive model (VAR) is first set up to capture the dynamics of shocks to
different regions. We then investigate the correlations of both real and nominal shocks between
Hong Kong and Mainland China, as well as Hong Kong and the United States, over time. The
setup of a state-space model in the second step allows us to decompose the shocks into common
and region-specific shocks, to gauge the degree of symmetry of regional shocks, and to examine

5
We use country, region and economy interchangeably in this study.
6
For a discussion on optimal currency areas, see Bayoumi and Eichengreen (1994).
7
A recent paper by Corsetti and Pesenti (2002) offers a theoretical model on endogenous optimal currency areas.
X. Xu / Journal of International Money and Finance 25 (2006) 894e911 897

whether regional shocks have been converging over time. This section describes the procedure
in detail.

2.1. Structural VAR to extract business cycle disturbances

We first investigate the sources and symmetries of disturbances to output and inflation in the
regional dimension. Following Blanchard and Quah (1989) and Bayoumi and Eichengreen
(1994), consider a bivariate system of output and inflation that has a structural vector moving
average (VMA) representation of two orthogonal shocks, supply and demand shocks. Using the
lag operator L, the model in matrix form8 can be written as:

Xt m qL3t ; 1

where the matrix polynomials q(L) denote the impulse response functions of the structural
shocks 3t 3Rt ; 3Nt 0 to the elements of X Dy; Dp0 , and m is a vector of constant.
In the bivariate system we study that yt and pt denote log of output and inflation, respec-
tively, Dyt and Dpt are therefore their rates of change at specified frequency. The structural
shocks 3t 3Rt ; 3Nt 0 are assumed to be mean-zero serially uncorrelated with diagonal covari-
ance matrix D E3t 30t .
To estimate the model defined by Eq. (1), we invert Eq. (1) into a structural vector auto-
regression representation:

BXt G0 G1 Xt1 / Gp Xtp 3t ; 2

where B is a 2  2 matrix that describes the contemporaneous relationship between the two en-
dogenous variables.
Direct estimation of the SVAR model is not possible due to the endogeneity of variables in
Xt, which is determined by the vector B. Pre-multiplication of Eq. (2) by B1 allows us to obtain
the reduced form vector auto-regression (VAR):

Xt A0 A1 Xt1 / Ap Xtp et ; 3

where A0 B1 G0 , A1 B1 G1 , Ap B1 Gp , and the reduced form errors et B1 3t are linear
combinations of the structural errors 3t with mean-zero and covariance matrix U B1 DB10 .
Although the reduced form VAR model of Eq. (3) can now be estimated, it is still not pos-
sible to uniquely derive the structural parameters B, G0, G1, . , Gp and D, given estimates of A0,
A1, . , Ap, and U. There are a total of 4p 8 structural parameters with 4p in the vector of
G1, . , Gp and 8 in B, G0 and D. However, estimation of the reduced form VAR model of

8
Specifically, the model can be written as

X
N X
N
Dyt qi11 3Rti qi12 3Nti ; f1
i0 i0

X
N X
N
Dpt qi11 3Rti qi12 3Nti ; f2
i0 i0

where yt and pt denote log of output and inflation, Dyt and Dpt are their rates of change at specified frequency, and 3Rt
and 3Nt denote structural orthogonal real and nominal shocks at time t, respectively.
898 X. Xu / Journal of International Money and Finance 25 (2006) 894e911

Eq. (3) yields only 4p 5 parameters. Thus, three restrictions are required for identification of
the structural parameters. Normalization of the elements of the vector B yields one on the di-
agonal and provides another two restrictions. Clearly, at least one restriction on the parameters
of the SVAR is required for the identification of all of the structural parameters.
There are broadly two approaches in the literature in imposing an additional restriction in
order to recover the structural parameters from estimates of reduced form VAR. The first is
to impose ad hoc restrictions such as recursive ordering on the parameters of SVAR. This
amounts to imposing short-run restriction on the SVAR model in that one of the endogenous
variables has no short-run impact on the other. The resulting estimates of the SVAR model can
be sensitive to the choice of the ordering of variables (Enders, 1995).
The second approach is to impose restrictions based on standard economic theory, such as
Blanchard and Quah (1989). We follow the second approach in assuming that aggregate de-
mand shocks have no long-run effects on real output, which implies that the coefficient matrix
of q(1) in Eq. (1) is lower triangular.
It can be shown that q(1) can be recovered from the Choleski factorization of the long-run
variance matrix X given that the matrix q(1) is lower triangular. Once the matrix q(1) is ob-
tained, B can be computed from the equation q1 I  A11 B1 . We can, therefore, re-
cover the vector of structural residuals by et B1 3t.

2.2. A state-space model to gauge the importance of common shocks

The second step in our empirical investigation involves an examination of the degree of asym-
metry of the set of structural disturbances across different economies. The degree of asymmetry
of structural shocks is often measured by a simple contemporaneous correlation coefficient be-
tween economies in most previous studies (e.g., Bayoumi and Eichengreen, 1994). This ap-
proach, however, has several limitations. First, a lot of factors that affect the co-movement of
disturbances between two economies are not and cannot be controlled by the simple correlation
coefficient. For example, shocks that are common to all economies, i.e., world business cycle dis-
turbances, or shocks that are common to all economies in a region, may lead to co-movement
of structural disturbances in the two economies under study. In such a case, a simple correlation
coefficient may turn out to be high but it would be difficult to treat the co-movement as having
a high degree of two-economies-specific common shocks, which is the basis for evaluating the
plausibility of forming an optimal currency area.
Second, the calculation of a simple correlation coefficient requires an anchor region (re-
gions), which makes it difficult to assess the degree of symmetry for a country, or a region
that consists of several economies. Third, simple correlation coefficients are sensitive to the pe-
riods chosen. Finally, studies along the lines of simple correlation do not provide dynamics of
the convergence (divergence) process, which is very important as one region might have little in
common in terms of structural shocks with the other in the first few years under study but may
have been converging in the last few years, or vice versa.
To measure the degree of asymmetry and to characterize the evolution of it over time, we set
up a state-space model that decomposes the structural shocks (observable variables) in each
country (region) into two unobservable stochastic components: one common to all countries
(regions) and the other country (region)-specific. We then calculate the share of common
shocks in total shocks in a specific period to gauge the degree of symmetry of shocks in that
period. Finally, we compare the changing share of common shocks to provide a dynamic pic-
ture of convergence of different shocks over time.
X. Xu / Journal of International Money and Finance 25 (2006) 894e911 899

Formally, for the purpose of the decomposition into common and specific shocks, the mea-
surement equation of the state-space model is set up as follows9:
2 3
 1   Z0t
st a 1 0 4 5
11 Z1t ; 4
s2t a21 0 1
21 23
Z2t
31

where the observable variables sit are supply or demand shocks obtained from the structural
VAR decomposition in the first step for country (region) i at time t. Z0t is the unobservable com-
mon component while Zit are the unobservable country (region)-specific components for region
1 or 2 at time t.
The unobservable components to be recovered follow a transition equation as:
2 3 2 32 3
Z0t 1 0 0 60t
4 Z1t 5 4 0 1 0 5 4 61t 5 ; 5
Z2t 0 0 1 62t
31 33 31

where the 60t and 6it are assumed to have no serial correlation with varianceecovariances ma-
trix specified as follows:
2 3
1 0 0
E660 4 0 P11 0 5; 6
0 0 P22

where for purpose of identification the common and specific shocks are assumed to be orthog-
onal, the variance of common shocks is set to unity and the variances of province-specific
shocks are Pii.
The state-space model outlined above allows us to decompose structural shocks (supply or
demand) facing province i into a component that is common to all provinces (Z0t) and one that
is province-specific (Zit), with coefficients ait measuring how strongly structural shocks facing
province i are related to the common component. The state-space model is estimated using the
Kalman filter.10
With the model coefficients having been estimated, it is straightforward to assess the relative
importance of common shocks with respect to total shocks for a particular province in a partic-
ular period. As the structural shocks from VAR estimation are the results from common and
province-specific shocks, the variances of the structural shocks are the sum of the variances
of common shocks a211 and that of the province-specific shocks (Pii). The proportion of var-
iance of the shocks that is therefore explained by the common component is expressed as
a211 =a211 Pii . Finally, we compare the changing share of common shocks to provide a dy-
namic picture of convergence of different shocks over time. This completes the description
of the methodology.

9
Although more complex dynamics can be set up in a state-space model, the current setup serves our purpose in de-
composing structural shocks into common and specific components. Here we follow the approach by Chamie et al.
(1994), which uses a similar setup for a three variable VAR model in the study of shock asymmetry in Europe and
the United States.
10
We refer readers to Harvey (1989) for an introduction to the estimation of the Kalman filter.
900 X. Xu / Journal of International Money and Finance 25 (2006) 894e911

3. Data

To estimate the above structural vector auto-regression model we need data on output and
prices for Hong Kong SAR, Mainland China and the United States. The frequency of our
data is quarterly. Data for gross domestic product (GDP) and GDP deflator for Hong Kong
are drawn from Hong Kongs Census and Statistics Departments publication Gross Domestic
Product, various issues. US data on GDP and GDP deflator are extracted through the Bureau of
Economic Analysis website. The selection of output and price indicators for China requires
some discussion, however, given that the reliability of national GDP data for China has been
challenged recently (Rawski, 2001).
We choose gross output value of industry (GOVI) as an indicator for output while retail price
index is an indicator of price for China. GOVI, instead of national GDP, is chosen for several
reasons. First, data for the industry sector is the most reliable among all the sectors. It is col-
lected routinely and directly from the primary source of the National Bureau of Statistics
(NBS) long-established statistical reporting system (from directly reporting enterprises for
State Owned Enterprises (SOEs) and other enterprises satisfied with certain criteria, see below).
It accounts for more than 66% of Chinas output and can therefore proxy the national output. In
contrast, output data for the tertiary sector are more questionable as the statistical reporting sys-
tem is highly underdeveloped in collecting data on tertiary sector economic activities. During
the central planning period, most activities in the service sector were not regarded as contrib-
uting to output and the statistical reporting system only collected data on material output. Sec-
ond, this indicator has the advantage over other indicators such as industrial output value
added (i.e., GDP for secondary industry), which may be theoretically more appealing than
GOVI but practically not so as in the context of the Chinese statistical system since it is a sec-
ondary source of data, derived from GOVI, by assuming arbitrarily some value added to output
ratio, especially when the data are reported at monthly and quarterly frequency.
Third, it has the longest data available with high frequency compared with other output indi-
cators such as industrial output value added, for which the quarterly data are only available since
the mid 1990s. Although the National Bureau of Statistics (NBS) stopped publishing monthly
data on GOVI in 1995 (replaced by industrial output value added), it is still a routine for the sta-
tistical bureau to collect data on GOVI as a primary source of data, among other indicators.
Quarterly data on GOVI at constant prices for the period 1989:Ie2001:IV are collected from
China Monthly Statistics (CMS). China Monthly Statistics continued to publish data on GOVI
until the beginning of 2000. As data on GOVI continues to be collected by NBS, we are able to
assemble the data for the period 2000:Ie2001:IV from the China Economic and Statistics Ex-
press (Zhong Guo Jing Ji Tong Ji Kuai Bao).
A reclassification occurred in 1999 for the indicator of GOVI. Data for GOVI before 1999
were collected directly, through the long-established enterprises reporting system, for all SOEs
and those non-SOEs at the level of township and above, while after 1999, it includes all SOEs
and those non-SOEs that have a sales value above 5 million Yuan.11 Although the new

11
According to National Bureau of Statistics (1999), these firms are selected at the beginning of each year and will not
be changed throughout the year unless one of the following occurs. (1) New firms that fall into the above category will
be included; (2) changes in ownership structure, for example, from SOEs to non-SOEs, will be adjusted; (3) new firms
that emerge from mergers of existing firms will be adjusted; and (4) bankruptcy and acquisition by other firms will be
dropped.
X. Xu / Journal of International Money and Finance 25 (2006) 894e911 901

classification of reporting enterprises leaves out a large number of small and medium enter-
prises in the non-SOE sector, output and other data for non-SOEs that have a sales value below
5 million Yuan are still routinely collected, by survey rather than direct reporting, through the
Urban and Rural Social-Economic Survey Division of the NBS.
Although we have argued that GOVI may be an appropriate output indicator for China, we
have performed robustness checking of our results using alternative output indicators such as
GDP, which is available from 1993 onwards.12
Quarterly data on retail price index (RPI) are derived as the average of monthly RPI, which
cover the same period as GOVI, from China Monthly Statistics, with unpublished data for De-
cember 1997 obtained from the publisher directly. Data for retail prices are collected routinely,
through survey, by the Urban and Rural Social-Economic Survey Division of the NBS. We
choose retail price index as an indicator for inflation for the purpose of our study as it covers
goods that are tradable, as compared with consumer price index (CPI) that includes both trad-
ables and non-tradables, and that only covers a smaller basket of commodities in the measure-
ment. However, we do use CPI data as an alternative to check the robustness of our results.
Our dataset thus assembled consists of output and price indicators for Hong Kong SAR,
Mainland China and the US for the period between 1988:I and 2001:IV. For the purpose of
comparison, we report simple correlations of output growth between Hong Kong, Mainland
China and the US in Table 1. During the period between 1988:I and 2001:IV, the correlation
of output growth between Hong Kong and the US is low, with the simple correlation coefficient
at 0.03, while that with China is relatively high (with the coefficient at 0.41). If the whole pe-
riod is separated into two sub-periods, we see that the correlation of output growth in Hong
Kong with both China and the US has been declining.
Simple correlations of output deflator between Hong Kong, Mainland China and the US re-
veal a somewhat different pattern (Table 2). Although Hong Kongs correlation in output de-
flator with that of the US is declining, and with that of China increasing over the two sub-
periods, the correlation coefficients are higher for the whole period than the simple average
of the two sub-periods, suggesting that the use of simple correlation coefficients may be highly
sensitive to the periods chosen and therefore a more robust technique is required. In Sections 4
and 5, a structural VAR model is used to separate the business cycle disturbances into supply
and demand side shocks, and a new time series technique is used to capture the dynamic of
business cycle correlations.

4. Estimation results

4.1. Univariate properties of the variables

Our dataset consists of quarterly data for output value and price index ranging from 1988:II
to 2001:IV for each country (region). Proper estimation of the VAR model requires that the data
generating process is stationary. Given the low power of the various unit root tests, we imple-
ment a number of unit root tests (Table 3). First, the augmented DickeyeFuller tests, both in
demeaned and detrended forms, are conducted for all three regions, with Bayesian Information
Criteria (BIC) being used to determine the choice of appropriate lag lengths in the regression.

12
Data for quarterly GDP are collected from Peoples Bank of China Statistics Quarterly, various issues. I thank one
referee for pointing out this to me.
902 X. Xu / Journal of International Money and Finance 25 (2006) 894e911

Table 1
Simple correlations of output growth between Hong Kong, China and the United States
HK GDP growth US GDP growth
1988:Ie1995:I
US GDP growth 0.23
China GDP growth 0.69 0.32
1995:IIe2001:IV
US GDP growth 0.01
China GDP growth 0.49 0.37
1988:Ie2001:IV
US GDP growth 0.03
China GDP growth 0.41 0.12
Note: Chinas GOVI refers to gross output value of industry.
Sources: Authors calculations.

Alternative criteria, such as a step-wise regression technique, in which, starting with some max-
imum number of lagged differences, insignificant lags are omitted until the last included lag is
significant, are also supplemented to check the robustness of the choice of lag length.
As a check of the robustness of our unit root tests using conventional augmented Dickeye
Fuller tests, we implement a powerful univariate test based on generalized least square (GLS),
proposed by Elliot et al. (1996). The results from augmented DickeyeFuller tests and Elliot
et al.s (1996) GLS tests indicate that growth of gross industrial output value is stationary while
changes in retail price are integrated of order one but a first difference is enough to make it
stationary.

4.2. Estimation of a VAR

We estimated Eq. (3) for each of the three regions. The likelihood ratio tests indicated that
a lag length of 4 quarters for the VAR model was most appropriate (reducing the length from 8
to 4 quarters yielded a Chi square value that was insignificant at the conventional level). Other
tests based on information criteria such as AIC and BIC do not present a uniform pattern. The
VAR lags were therefore set to 4 quarters in all cases in order to preserve symmetry of

Table 2
Simple correlations of output deflators between Hong Kong, China and the United States
HK GDP deflator US GDP deflator
1988:Ie1995:I
US GDP deflator 0.36
China retail price index 0.01 0.22
1995:IIe2001:IV
US GDP deflator 0.13
China retail price index 0.53 0.41
1988:Ie2001:IV
US GDP deflator 0.62
China retail price index 0.56 0.37
Sources: Authors calculations.
X. Xu / Journal of International Money and Finance 25 (2006) 894e911 903

Table 3
Unit root test statistics: 1988:IIe2001:IV
Augmented DickeyeFuller tests DFeGLS tests
Constant Lag Trend Lag Constant Lag Trend Lag
Level
Log US GDP 0.239 0 1.251 0 3.393 0 1.076 0
Log HK GDP 1.402 5 1.775 5 0.461 5 1.244 5
Log China GOVI 0.442 4 3.167 5 0.229 4 2.756 4
First difference
Log US GDP 5.028** 0 5.015** 0 4.548** 0 4.865** 0
Log HK GDP 3.812* 4 3.896* 4 2.247* 4 3.638** 4
Log China GOVI 1.984 3 7.284** 1 4.582** 1 6.33** 1
Level
US GDP deflator 0.94 0 2.333 1 0.792 0 2.191 1
Hong Kong GDP deflator 0.993 0 1.798 0 0.675 0 1.721 0
China retail price index 1.098 0 3.887* 2 1.012 0 4.119** 2
First difference
US GDP deflator 4.372** 0 4.671** 0 3.889* 0 4.139** 0
Hong Kong GDP deflator 5.852** 0 5.796** 0 5.832** 0 5.89** 0
China retail price index 4.584** 0 4.531** 0 3.595** 0 4.223** 0
Notes: DFeGLS test is a DickeyeFuller test based on GLS-detrended series proposed by Elliot et al. (1996). The test
statistics reported under the heading of constant include a constant term while those reported under the heading of
trend involve a constant term and a linear trend term. The lag lengths are chosen based on a procedure that adds
lags until a Lagrange Multiplier test fails to reject no serial correlation at 5% significant level. ** denotes that the
test is significant at 1% level while * represents significance at 5% level.
Source: Authors calculations.

specification across regions. Our estimation of the structural VAR model (3) with long-run re-
strictions imposed generates quarterly structural supply and demand disturbances for each of
the regions, which are then subject to further investigation on the degree of synchronization
in the next step.
As a benchmark for comparison with more sophisticated time series techniques in measuring
the correlations in Section 4.3, we present in Table 4 simple correlation coefficients of struc-
tural supply and demand shocks that are derived from the first stage VAR analysis, among
Hong Kong, Mainland China and the US. Interestingly, correlation of supply shocks between
Hong Kong and the US has been declining but remains still high relative to that with Mainland

Table 4
Simple correlations of structural shocks: supply and demand shocks
Supply shock Demand shock
US China US China
Hong Kong 1989:IIe1995:I 0.18 0.07 0.07 0.49
1995:IIe2001:IV 0.07 0.01 0.10 0.19
1989:IIe2001:IV 0.12 0.02 0.06 0.16
United States 1989:IIe1995:I 0.00 0.20
1995:IIe2001:IV 0.39 0.08
1989:IIe2001:IV 0.17 0.10
Source: Authors calculations.
904 X. Xu / Journal of International Money and Finance 25 (2006) 894e911

China, while correlation with Mainland China has been increasing but remains weak. Correla-
tion of demand side disturbance of Hong Kong with Mainland China is higher than that with the
US but has also been declining. Furthermore, business cycle correlation between Mainland
China and the US has been drifting away, for both supply and demand side shocks.

4.3. Degree of asymmetry from the estimation of a state-space model

Table 5A reports the results of the state-space model estimation for the share of common
shocks in total supply and demand shocks. To gauge the degree of asymmetry of shocks and
how it evolves over time, we decompose the entire dataset into two sub-periods,13 with the first
quarter of 1995 being the cut-off date for the period between the second quarter of 1989 and the
fourth quarter of 2001 when the data are available.
Several features emerge from Table 5A. First, from the last column of Table 5A, we can see
that, for the period between 1989:II and 2001:IV, supply disturbances to Hong Kong, on aver-
age, share only 3% in common with those to Mainland China, while 15% are in common with
those to the US. However, when it comes to the demand side, the pattern of common distur-
bances is reversed. Hong Kong shares almost 15% in common with those shocks to Mainland
China while only 5% with those shocks to the US. In all cases, Hong Kong shares little in com-
mon with shocks to both Mainland China and the US. We have performed robustness checking
on our results using alternative indicators. With Chinas GOVI and retail price index replaced
by quarterly GDP and consumer price index, respectively, supply disturbances to Hong Kong,
on average, share about 35% in common with those to Mainland China, while 8% are in com-
mon for the demand disturbances, for the period between 1995:II and 2001:IV (Table 5B).
Turning to the time series pattern, Hong Kong shares increasingly less in common with both
Mainland China and the US, in both demand and supply shocks. Hong Kong has more, albeit
declining, in common in supply shocks with the US relative to Mainland China, while more in
common in demand shocks with Mainland China as compared with the US.
The low degree of synchronization between Hong Kong SAR and Mainland China, as well
as Hong Kong SAR and the US is in sharp contrast to the studies on the US regions. For ex-
ample, a study by Chamie et al. (1994) shows that the variance of supply shocks facing US re-
gions (excluding the New England region) explained by the common component varies from 59
to 89% for the period between the first quarter of 1970 and the final quarter of 1991 while de-
mand shocks exhibit even greater symmetry than supply shocks for the US regions, ranging
from 71 to 99%.
Although a thorough study of the determinants of the degree of business cycle synchroniza-
tion is beyond the scope of the current paper and would be an interesting topic for future re-
search, it may be useful here to provide some discussions on the underlying structural forces
that have shaped Hong Kongs economic performance and that have contributed to the observed
patterns of business synchronization.
From entrepot to manufacturing. Throughout much of its history, Hong Kong had been func-
tioning mainly as Chinas entrepot. That function was, however, interrupted between 1949 and
1979, during which Hong Kong was able to develop, based on its comparative advantage,

13
The choice of a cut-off date may seem somewhat arbitrary but it is chosen in such a way to balance the estimate
between the two sub-periods. We use an alternative specification, i.e., time varying coefficient model in Section 5 to
check the robustness of our results.
X. Xu / Journal of International Money and Finance 25 (2006) 894e911 905

Table 5A
Variance decomposition of structural shocks: share of common shock (%)
1989:IIe1995:I 1995:IIe2001:IV 1989:IIe2001:IV
Supply shock
Mainland China and Hong Kong SAR
Mainland China 5.13*** 1.14*** 2.26***
Hong Kong SAR 8.90*** 1.98*** 3.08***
USA and Hong Kong SAR
USA 10.44*** 7.13*** 7.76***
Hong Kong SAR 29.67*** 7.67*** 15.29***

Demand shock
Mainland China and Hong Kong SAR
Mainland China 58.17*** 5.50*** 10.19***
Hong Kong SAR 40.52 32.44*** 14.93***
USA and Hong Kong SAR
USA 5.16*** 6.98*** 5.25***
Hong Kong SAR 9.77*** 2.73*** 5.11***
Note: *** denotes shocks that are statistically significant at 1% level in relation to the common component.
Source: Authors calculations.

a competitive export-oriented labour-intensive sector, thanks to its lassiez faire policy and a fa-
vourable regional and international environment. As a result, Hong Kongs GDP growth sus-
tained an average of 8.9% for two decades since the 1960s, which is now part of the
successful story of the so called East Asian miracle. Rising labour and land cost in the late
1970s, resulting from two decades of rapid economic growth, were the fundamental forces
that started to push Hong Kongs labour-intensive firms out of the territory.

Table 5B
Variance decomposition of structural shocks: alternative estimates of share of common shock (%)
1995:IIe2001:IV
Supply shock
Mainland China and Hong Kong SAR
Mainland China 30.5***
Hong Kong SAR 35.0***
USA and Hong Kong SAR
USA 7.13***
Hong Kong SAR 7.67***

Demand shock
Mainland China and Hong Kong SAR
Mainland China 0.06***
Hong Kong SAR 8.09***
USA and Hong Kong SAR
USA 6.98***
Hong Kong SAR 2.73***
Note: *** denotes shocks that are statistically significant at 1% level in relation to the common component. Shocks for
China are derived using alternative data (quarterly GDP and consumer price index).
Source: Authors calculations.
906 X. Xu / Journal of International Money and Finance 25 (2006) 894e911

From manufacturing to services. The opening up of China to foreign trade and foreign direct
investment since 1979 has had a profound impact on the process of Hong Kongs economys
structural transformation. It provides golden opportunities for Hong Kongs labour-intensive
firms to relocate to the other side of the border, which has an abundant supply of labour and
land. More importantly, the lack of a well-functioning producer services sector, such as export-
ing and importing, transportation, telecommunication and financial sectors, in Mainland China,
facilitated a rapid expansion of the services sector in Hong Kong, to serve not just those Hong
Kong firms in Mainland China, but also Chinese firms now interacting with overseas markets.
Between 1986 and 1995, despite the massive relocation of manufacturing to China, Hong
Kongs GDP grew at an average annual rate of 6.6%, while the unemployment rate remains
at an average of 1.9%, a remarkably low rate given the dramatic structural change in the
economy.
The relocation of labour-intensive firms to China and the need for a well-functioning pro-
ducer services sector to serve the growing manufacturing activities in the Mainland helped
to transform Hong Kong from a manufacturing-oriented to a services-oriented economy. In
the mean time, Hong Kongs traditional role as an entrepot was resurrected. Hong Kongs
re-exports as a percentage of its total exports has risen, from 65% in 1990 to 90% in 2001.
In 1980, the manufacturing sector accounts for 20.7% of GDP while the services sector took
up 69.1%. In 2001, the share of manufacturing fell to 5.2% while the services sector increased
to 86.5%. By the mid 1990s virtually all Hong Kong manufacturers had migrated to China
(Kueh and Ng, 2002). Employment in the manufacturing sector declined from a high of
45.9% in 1980 to just about 10% in 2000.
The similarity in production structure has been identified as a major determinant of co-
movement of output in so far as industry-specific disturbances are the dominant factors in driv-
ing the business cycles (Imbs, 1999). Hong Kong and Mainland China are at different stages of
economic development and therefore, reveal distinct production pattern, with Hong Kong being
a service dominated economy and China a world manufacturing center. Even within the
manufacturing sector, most of the labour-intensive manufacturing activities previously located
in Hong Kong have moved across the border to Mainland China, leaving only those human-
capital intensive or capital-intensive activities in Hong Kong. This distinct production pattern
may have contributed to the low degree of business cycle synchronization between Hong
Kong SAR and Mainland China.
Although several empirical studies have identified that more trade is associated with higher
output synchronization (Frankel and Rose, 1998), theoretically, the trade-output co-movement
link may go either way. If more trade leads to regional specialization based on comparative ad-
vantage, industry-specific shocks may lead to less synchronization in output co-movement. It is
well known that trade between Hong Kong and Mainland China comprises mainly re-exports,
which makes the trade-output co-movement nexus less powerful. This may contribute to some
extent to the low degree of output synchronization between Hong Kong and China.
The on-going integration process of Hong Kong, initially with its close neighbour Pearl
River Delta, then with the rest of Mainland China, also brought on tremendous challenges to
Hong Kongs economy, through another fundamental force, i.e., factor price equalization.
Hong Kongs property market has been declining for several years since 1998 while that of
Mainland China has been increasing. Although there is still a border between Hong Kong
and Mainland China, the border is disappearing fast. It has been estimated that there is now
about 10% of Hong Kongs labour force working on Mainland China. More and more Hong
Kong residents are now living in Shenzhen and its neighbouring regions. The Closer Economic
X. Xu / Journal of International Money and Finance 25 (2006) 894e911 907

Partnership Agreement (CEPA), which came into effect in January 2004, further, integrates
Hong Kong with the Mainland economy. The process of factor price equalization between
the two economies within one country entity may have led to a divergence of business cycles
between Hong Kong and the Mainland.14
Having said that, one has to bear in mind that although with different production patterns,
Hong Kongs specialization in the services sector, in particular, producer related services
such as insurance, importeexport trade, financing, and business services, which have long
been serving the growing manufacturing activities in Mainland China since its opening up,
has meant that its business cycle may have been increasingly more synchronized with that
of Mainland China.
With these distinctive forces at work, the extent of business cycle synchronization between
Hong Kong and Mainland China is therefore an empirical question. Our empirical results sug-
gest that different production patterns may be a more powerful factor than the integration fac-
tor, at least in the period under study.

5. Estimating time varying parameters: an alternative specification

A number of existing studies have emphasized that the degree of business cycle synchroni-
zation is an on-going process. For example, in their empirical study, Frankel and Rose (1998)
conclude that more trade leads to closer output co-movements across trading parties. Canova
and Dellas (1993) also find that higher trade interdependence increases the extent of output
co-movement. It is therefore tempting to estimate time varying parameters to capture the on-
going dynamic process of business cycle synchronization among different countries (regions).
Although Section 4 has provided estimates of common shocks for two separate periods, the dy-
namic process from year to year is yet to be uncovered. This dynamic process may be of great
interest, in particular, for the recent period when Hong Kong has returned to the rule of China
and there are increasing ties in both trade and investment between the two sides.
In this section, we present a standard state-space model to estimate the time varying param-
eters, following Harvey (1989), Haldane and Hall (1991), and Boone (1997). Let

Yt d0 Zt yt ; 7

be the measurement equation, where Yt is a measured variable, Zt is the state vector of unob-
served variables, d is a vector of parameters and yt wNID0; 1. The state equation follows a ran-
dom walk process:

Zt Zt1 ut ; 8

where ut wNID0; Qt .
Given the structural supply and demand shocks that we have already derived from the struc-
tural VAR model in Section 4 for Hong Kong, China and the US, we will be able to estimate
a time varying relationship of the shocks between Hong Kong and Mainland China, as well as
Hong Kong and the US. We first specify d as a vector of known variables, in this case the struc-
tural shock variable for Mainland China and the US, and Zt as a vector of time varying coef-
ficients (with constant term expected to be zero since all structural shocks from VAR have

14
I wish to thank one referee for pointing out this to me.
908 X. Xu / Journal of International Money and Finance 25 (2006) 894e911

mean-zero). Yt is a series of structural shocks to Hong Kong, also derived from VAR estimates.
Qt is specified as a diagonal matrix, the elements of which are to be estimated using maximum
likelihood by the Kalman filtering technique. Of particular interest is the time varying coeffi-
cient Zt, whereby parameters converging to one over time may imply Hong Kong increasingly
shares business cycle synchronization with Mainland China (or the US), and vice versa.
Time varying coefficients from Eqs. (7) and (8) are first estimated for structural supply
shocks between Hong Kong, China and the US, with results shown in Fig. 1. Interestingly,
Hong Kongs business cycles are idiosyncratic to those of the US from 1989:II to 2001:IV.
As shared by our common belief, the synchronization of Hong Kong business cycles with those
of Mainland China has been increasing over the period. However, in contrast with the common
notion that Hong Kong may have integrated closely with Mainland China, our time varying co-
efficient suggests that the on-going process of business cycle synchronization between Hong
Kong and Mainland China is still at a very low level, and becoming rather volatile after the
Asian financial crisis. The results are not sensitive to the choice of output and price indicators
for China.
The low degree of business cycle synchronization between Hong Kong and Mainland China
is also largely confirmed by the demand side shocks, as shown in Fig. 2. A sharp break took
place in early 1996 in the association of demand side disturbances between Hong Kong and

0.5 HK with China

0.3

0.1

-0.1

-0.3

-0.5

-0.7
89:II 90:II 91:II 92:II 93:II 94:II 95:II 96:II 97:II 98:II 99:II 00:II 01:II

0.5
HK with USA
0.3

0.1

-0.1

-0.3

-0.5

-0.7
89:II 90:II 91:II 92:II 93:II 94:II 95:II 96:II 97:II 98:II 99:II 00:II 01:II

Fig. 1. Convergence of Hong Kongs structural supply shocks: China vs. the United States. Sources: Authors
calculations.
X. Xu / Journal of International Money and Finance 25 (2006) 894e911 909

1.5

0.5

-0.5
HK vs China
-1

-1.5

-2

-2.5
89:II 90:III 91:IV 93:I 94:II 95:III 96:IV 98:I 99:II 00:III 01:IV

1.5
HK vs USA
1

0.5

-0.5
HK vs China
-1

-1.5

-2

-2.5
89:II 90:III 91:IV 93:I 94:II 95:III 96:IV 98:I 99:II 00:III 01:IV

Fig. 2. Convergence of Hong Kongs structural demand shocks: China vs. the United States. Sources: Authors
calculations.

Mainland China, as well as Hong Kong and the US. Before 1996, demand side disturbances
between Hong Kong and the US are strong. As Hong Kong was experiencing high inflation
with strong GDP growth before 1996, the US economy was also growing robustly at the
same time (Kueh and Ng, 2002). By virtue of the US dollar peg, Hong Kongs demand side
disturbances went hand-in-hand with those of the US, with persistent inflationary pressure fu-
elled further by the loose monetary policy in the US. This process underwent a complete over-
haul in 1996, with the degree of association coming down to a level on par with China,
suggesting more powerful factors might have come into play. Again, from the demand shocks
perspective, the process of Hong Kongs business cycle synchronization with China has been
increasing, but is still at a very low level.
To summarize, our results from a time varying coefficient model point to an on-going pro-
cess between Hong Kong and China, with increasingly more synchronized business cycle dis-
turbances, from both supply and demand sides, albeit with the fact that this synchronization
process is still at a very low level. Furthermore, Hong Kong experiences increasingly more id-
iosyncratic supply side disturbances with the US throughout the period under study, and a sim-
ilar picture for the demand side after 1996.
910 X. Xu / Journal of International Money and Finance 25 (2006) 894e911

6. Conclusion

Making use of the recent advance in time series econometrics modelling techniques, we ex-
amined, in this paper, the degree of business cycle synchronization among Hong Kong SAR,
Mainland China and the United States, as well as the dynamics of the on-going process of syn-
chronization over time with a two-step procedure: a VAR model to derive structural business
shocks and a state-space model to identify the degree of synchronization over time.
Several important findings emerge from our analysis. First, contrary to the common notion
that Hong Kong SAR is increasingly integrated with the Mainland China, for the period be-
tween 1989 and 2001, Hong Kong, on average, shares little in common in both supply and de-
mand shock with China, as well as with the US. The reasons behind this may be the distinct
production pattern and trade structure between them. Second, the time series pattern reveals
that Hong Kong shares increasingly less in common with both Mainland China and the US,
in both demand and supply shocks. The low degree of business cycle synchronization is robust
to various model specifications. Our results cast doubt on the feasibility of a currency union
arrangement between Hong Kong and Mainland China in a foreseeable future, even when
the RMB becomes fully convertible.
We have only documented in this paper the stylized facts for the business cycle synchro-
nization between Hong Kong and Mainland China. Future research would identify the under-
lying causes of this synchronization process, in particular, the role of production structure
similarity, trade intensity, and mobility of factors, goods and services across the border. Re-
search along this line would be able to contribute to our understanding of business cycles in
Hong Kong, which carries important implications for the future of Hong Kongs exchange
rate regime.

Acknowledgements

The author wishes to thank James Lothian (the editor) and an anonymous referee for con-
structive comments on earlier versions of this paper. All remaining errors are mine. Financial
support from the Hong Kong Polytechnic University (Grant Account # G-U077) is gratefully
acknowledged.

References

Bayoumi, T., Eichengreen, B., 1994. One Money or Many? Analyzing the Prospects for Monetary Unification in Var-
ious Parts of the World. In: Princeton Studies in International Finance No. 76.
Blanchard, O.J., Quah, D., 1989. The dynamic effects of aggregate demand and supply disturbances. The American
Economic Review 79, 655e673.
Boone, L., 1997. Symmetry and asymmetry of supply and demand shocks in the European Union: a dynamic analysis.
CEPII Working Paper.
Canova, F., Dellas, H., 1993. Trade interdependence and the international business cycle. Journal of International Eco-
nomics 34, 23e47.
Chamie, N., DeSerres, A., Lalonde, R., 1994. Optimal currency areas and shock asymmetry: a comparison of Europe
and the United States. Bank of Canada Working Paper 94/1.
Corsetti, G., Pesenti, P., 2002. Self-validating optimal currency areas. NBER Working Paper No. 8783.
Devereux, M., 2002. A tale of two currencies: the Asian crisis and the exchange rate regimes of Hong Kong and Sin-
gapore. Review of International Economics 11, 38e54.
X. Xu / Journal of International Money and Finance 25 (2006) 894e911 911

Elliot, G., Rothenberg, T.J., Stock, J.H., 1996. Efficient tests for an autoregressive time series with a unit root. Econ-
ometrica 64, 813e836.
Enders, W., 1995. Applied Econometric Time Series. John Wiley & Sons, Inc.
Frankel, J.A., Rose, A.K., 1998. The endogeneity of the optimum currency area criteria. Economic Journal 108, 1009e
1025.
Haldane, A.G., Hall, S.G., 1991. Sterlings relationship with the Dollar and the Deutschemark: 1976e89. Economic
Journal 101 (406), 436e443.
Harvey, A., 1989. Forecasting, Structural Time Series Models and the Kalman Filter. Cambridge University Press, Cam-
bridge, New York.
Hess, G., van Wincoop, E., 2000. Intranational Macroeconomics. Cambridge University Press, Cambridge, UK.
Hess, G., Shin, K., 1997. International and intranational business cycles. Oxford Review of Economic Policy 13 (3),
93e109.
Imbs, J., 1999. Co-fluctuations. CEPR Discussion Paper No. 2267.
Kueh, Y.Y., Ng, R., 2002. The interplay of the China Factor and US dollar peg in the Hong Kong economy. China
Quarterly 170, 387e412.
Liang, H., 1999. Do Hong Kong SAR and China constitute an optimal currency area? An empirical test of the gener-
alized purchasing power parity hypothesis. IMF Working Paper No. 99/79.
Ma, Y., Tsang, S., 2002. Do China and Hong Kong constitute an optimal currency area? Lingnan University Working
Paper No. 130, Hong Kong.
McKinnon, R.I., 1963. Optimal currency areas. The American Economic Review 53, 717e725.
Mundell, R.A., 1961. A theory of optimal currency areas. The American Economic Review 51, 657e665.
National Bureau of Statistics (NBS), 1999. Xin Bin Gong Ye Tong Ji Gong Zuo Zi Nan. (Working Guide for Industrial
Statistics, a New Version). Industry and Transportation Department, NBS. China Statistical Publishing House,
Beijing.
Rawski, T., 2001. What is happening to Chinas GDP statistics? China Economic Review 12, 347e354.
Xu, X., 2002. Have the Chinese provinces become integrated under reform? China Economic Review 13 (2/3), 116e
133.