CAUSALITY IN MEAN AND VARIANCE: EVIDENCE FROM REAL ESTATE MARKETS
Kim Hiang LIOW and Haihong Zhu, Department of Real Estate, National University of Singapore
Corresponding Author
Dr. Kim Hiang LIOW
Associate Professor
Department of Real Estate
National University of Singapore
4 Architecture Drive
Singapore 117566
Tel: (65)68743420
Fax: (65)67748684
Email: rstlkh@nus.edu.sg
24 May 2005
1
CAUSALITY IN MEAN AND VARIANCE: EVIDENCE FROM REAL ESTATE MARKETS
Abstract
While much work has been undertaken on causality-in-mean returns causality-in-variance has not been
received any attention in real estate literature. In this paper, we examine the presence or absence of causality-in-mean
and causality-in-variance across seven Asian and the US and UK public real estate markets following the procedures
developed by Cheung and Ng (1996). Our main results are that international real estate markets are generally
correlated in returns and volatilities contemporaneously and with lags. Causality in both mean and volatility are
observed in some markets, with dynamic adjustments take place on average between one and five weeks. The US and
UK markets also significantly affect some Asian markets such as Singapore, Hong Kong, Japan and Malaysia in either
the mean or return volatility at different lags. However at least half of the Asian real estate markets are still largely
segmented implying that institutional investors would likely to benefit from diversifying public real estate portfolios
internationally across Asia and the US/UK in the short- and /or medium term. This knowledge would further help fund
managers in managing their exposure in Asian real estate markets and constructing better asset allocation models.
1.
INTRODUCTION
Understanding of the transmission mechanism and causal linkages among real estate markets is of great
interest to academic researchers, investment professionals and regulators given the increasing trend in integration and
globalization of capital and real estate markets. Furthermore, with increasing re-emergence of real estate as an asset
class in the Asian economies after the major financial and economic slowdowns in the late 90’s, it is important for
institutional investors and regulators to revisit and obtain fresh insights into the dynamic linkages and
interdependencies across the various Asian and the developed public real estate markets of the US and UK. In this
study, we test for the presence or absence of causal linkages in mean and variance across the real estate markets of
the US, UK and seven Asian economies; Singapore, Hong Kong, Malaysia, Indonesia, Thailand, Philippines and
Japan. Whilst the examination of causation in mean using Granger causality test has commonly appeared in studies
relating to financial market movements; the issue of variance causality has received less attention in international
finance. Causality tests and results provide investors with additional insights into how and when information is
impacted on different real estate markets and design more objective pricing models with the appropriate lag structure.
Additionally, regulators may use the causality results to identify the foreign market(s) that drive(s) movements in the
domestic real estate market and take appropriate corrective measures.
Following the procedures developed by Cheung and Ng (1996), this study uses their methodology to detect
the causation patterns in mean and volatility and compare differences across the nine major national real estate
markets. Traditional Granger causality focuses on the mean changes, while the causality-in-variance examines the
2
conditional volatility dependence between two markets. The knowledge of how volatility between two markets are
related is important in that the second-moment or variance is directly linked to information flow. Specifically, causalityin-variance tests are able to detect the direction of causality as well as the number of leads/lags involved.
Methodologically, the main advantage of the test procedures is the flexible specifications of the innovation process and
the non-dependence on return normality (Chung and Ng, 1996). Consequently, the causality-in-variance tests are of
also of special interest in instances of highly volatile and non-normally distributed returns which are typically of Asian
economies and real estate markets. Essentially, the analysis involves a two-step procedure which requires estimating
as a first step a time-varying conditional mean and variance models, and computing and testing as a second step the
cross-correlation functions (CCF) of the standardized and squared standardized residuals.
Our main results are that international real estate markets are generally correlated in returns and volatilities
contemporaneously and with lags. Causality in both mean and volatility are observed in some markets, with dynamic
adjustments take place on average between one and five weeks. The US and UK markets also significantly affect
some Asian markets such as Singapore, Hong Kong, Japan and Malaysia in either the mean or return volatility at
different lags. This knowledge would further help fund managers in managing their exposure in Asian real estate
markets and constructing better asset allocation models.
Our study is organized as follows. Section 2 contains a selected literature review. This is followed by an
explanation of the research data and methodology in Sections 3 and 4 respectively. Section 5 discusses the empirical
results and implications. The last section concludes the study with the summary of main results.
2.
RELATED LITERATURE
Of late, there has been increasing interest in the causation in conditional variance across stock market price
movements. The study of causation pattern in variance provides useful insights into the relationship between
information flow and volatility and the characteristics and dynamics of financial asset prices (Cheung and Ng, 1996).
Together with evidence regarding the causality in mean, the overall results will yield a more complete picture regarding
the dynamics of interactions among the financial markets involved. The policy implication regarding causality would
clearly be important, as the evidence would suggest that financial market policies of one country should not be
3
implemented without taking into account the impacts on other markets (say, in Asian or European regional real estate
markets), and vice versa. From the institutional investors’ perspective, there would be little diversification benefits in the
short-term if two markets are causally linked in both returns and volatilities.
Cheung and Ng (1996) develop a test for causality-in-variance (CIV) that allow researchers to assess how a
market evaluates and assimilates new information, and examine the temporal dynamics of return volatilities across
national stock markets. The CCF testing procedure does not require modeling of the dynamics of the interaction of the
series involved. Instead it is based on the residual cross-correlation function (CCF) and is robust to distributional
assumptions. The null hypothesis of no causality in variance is tested through asymptotic normal t statistic.
Mathematically, given two stationary time series X t and Yt , the first task is to impose a conditional mean and
variance specification for the time series. Then the squared standardized residuals ( ε t and ξ t ) are estimated for
2
2
series X t and Yt , namely:
U t = {( X t − μ x ,i ) 2 / hx ,t } = ε t
Vt = {(Yt − μ y ,t ) 2 / h y ,t ) = ξ t
Where
μ and
2
2
h are the conditional mean and variance of the time series respectively. Assume further
rUV (k ) and rεξ (k ) are the sample cross-correlations of the squared standardized residuals and standardized
residuals at lag k respectively.
The CCF testing procedure is employed to test the hypothesis of no causality in variance against the
alternative hypothesis of causality at lag k. Given that T is the number of time series observations, the CCF-statistic is
give by:
CCF − statistic = T * rUV (k )
Similarly, to test the null hypothesis of no causality in mean against the alternative hypothesis of causality at
lag k, the CCF statistic is given by:
4
CCF − statistic = T * rεξ (k )
To-date, the CCF testing methodology has been applied in several studies on stock markets, foreign
currency markets and interest rate markets. Hu et al (1997) examine the CIV and volatility spillover effects among two
developed markets (US, Japan) and four emerging markets in the South China Growth Triangular (Hong Kong,
Taiwan, Shanghai and Shenzhen). Employing the CCF methodology, they find significant relationships among the
variances of the markets and that there is a feedback relationship between the Hong Kong and the US stock markets.
In a study that covers five Pacific-Rim stock markets and the US stock market returns with the CCF testing procedure
plus a TARCH model, a MGARCH specification and VAR analyzes, Tay and Zhu (2000) find that in most instances
there is bi-directional causality (both) in mean and variance among the markets. Furthermore, the dynamic adjustment
of the market return volatilities can take as much longer time than expected. In a foreign exchange study, Kanas and
Kouretas (2002) examine the issue of mean and variance causality across four Latin American markets over 1976 1993. Using an EGARCH-M model and the CIV methodology, they find substantial evidence of causality in both mean
and variance with the causality in mean largely driven by the causality in variance. Alaganar and Bhar (2003) employ
the CCF testing procedure to examine the linkages between the financial sector returns and interest rates of the G7
countries. They demonstrate that the CIV procedure is able to model the direction and lags in information flow between
two time series. They are able to find feedback effects both at the mean and the volatility level. Finally, Caporale et al
(2002) provide some empirical evidence on the casual relationship between stock prices and exchange rates volatility
in four East Asian countries using daily data from January 1987 to March 2000 by appealing to the MGARCH
methodology with BEKK representation. They also analyze the effects of the 1997 East Asia crisis by splitting the
sample in two, before and after the onset of the crisis.
On the contrary, research studies on real estate market volatility and its transmission across international
real estate markets are relatively limited and none of them has examined the issue of causality-in-variance. The lack of
research in this area is a peculiar oversight, given the increase in indirect global property investment over the last
decade. Garvey et al (1991) examine the linkages between securitized real estate markets of Australia, Japan,
Singapore and Hong Kong on both short and long-term basis. The short term analysis covers linkages in both mean
and variance of the return series using Granger causality and GARCH models that incorporate volatility spillovers.
5
They find little evidence of co-movements or influences between the markets on a bivariate basis. Stevenson (2002)
investigates the influences on volatility in the US REITs. He finds that the volatility in equity REITs has significant
influence on other sub-sectors of the REIT market. More recently, Liow et al (2005) examine the long-term and shortrun relationships in mean and variance spillover effects across four Asian property stock markets (Japan, Hong Kong,
Singapore and Malaysia) and four European real estate stock markets (the UK, France, Germany and Italy). Their
evidence of minimal cointegration, weak mean transmission and little cross-volatility spillovers among the Asian and
European real estate stock markets imply that diversification opportunities exists among the regional public real estate
markets. However, they do not examine the issue of causality in mean and variance across the real estate markets.
Additionally, Liow (2005) focuses on the dynamics of volatility persistence and systematic risk in Asian-Pacific real
estate markets. Again, the issues of interdependence in term of the causal links among the returns and variances
among the real estate markets were not investigated. Finally, this research represents an attempt to transfer a novel
technique developed by Cheung and Ng (1996) from other investment markets to the international real estate markets.
3.
RESEARCH DATA
As in many previous academic real estate studies, we use returns on real estate stocks to proxy for real
estate performance. This choice is mainly justified by the availability of longer time series data and higher frequency
data (such as monthly and weekly) for real estate stocks. Whilst the adequacy of this proxy has been extensively
debated amongst real estate practitioners and researchers, it remains the only substantive “real estate” series
appropriate for any rigorous statistical analysis.
We include nine public real estate markets for a US-based investor. The focus on these markets is also of
significant interest to the world investors and policy makers. Apart from the popularly known US and UK real estate
markets that have different institutional and market structures from the developing economies, the remaining seven are
Asian markets of Japan, Hong Kong, Singapore, Malaysia, Indonesia, Thailand and the Philippines.1 Our sampling thus
cover almost all the Asian public real estate markets and two other key developed economies that meet the
requirement of the study to provide a comprehensive study of national real estate markets. The Asian markets are
however in different stages of development as revealed by the key macroeconomic and stock indicators reported in
Table 1. Japan is a significant world economy and has a long history of listed real estate. Similarly, Hong Kong and
6
Singapore have track records of listed real estate companies that play a relatively important role in general stock
market indexes. These markets plus the US and UK markets have the world’s most significant listed real estate
markets in the respective regions (UBS Bank, 2003). The remaining four markets (Malaysia, Thailand, Indonesia and
the Philippines) are classified as emerging markets and hence needs longer time to develop. Finally, as many of the
Asian economies have shown good signs of recovery, it is expected that increasing foreign investments in these
markets will enhance the role that these markets play in the world economy and becomes more important in due
course.
(Table 1 here)
All data are Dow Jones real estate stock indexes 2 extracted from the Datastream International and the
sample period is from January 1992 till December 2004, the longest period for which all data indexes are available.
Furthermore any potential bias due to the specific time period should be minimal as the study period covers the boom
and bust phases of the most recent real estate and economic cycles in the Asia-Pacific region. Weekly real estate
stock returns (R) are obtained by taking the logarithmic difference of the stock index (P) times 100. That is, R t = 100 *
(log P t – log P t – 1). All data are expressed in US dollars. Figure 1 displays the index movement over the study period.
(Figure 1 here)
To provide a general understanding of the nature of each real estate market return, Table 2 contains sample
characteristics of the weekly real estate stock returns. These include the mean, standard deviation, coefficient of
variation, maximum and minimum of returns and the measures of skewness, Kurtosis and normality. As the figures
indicate, the UK real estate market has the highest average weekly return (0.160%) and is followed by the US
(0.129%); while Malaysia has the lowest mean (-0.099%). Judging from the sample standard deviations, in general,
developing Asian real estate markets are characterized by a higher unconditional volatility, compared to the developed
markets of the UK and the USA. In particular, the Thailand real estate market appears to be the most volatile (standard
deviation = 17.12%), followed by Indonesia (standard deviation = 15.26%), while the US real estate market is the least
volatile (standard deviation = 1.95%). Consequently, the US and UK markets also have the two lowest coefficient of
variation (i.e. risk per unit of return). Of the five markets (Japan, Indonesia, Malaysia, Thailand and the Philippines)
whose returns are positively skewed, Thailand has the highest skewness (20.39) and is followed by Indonesia (15.88).
7
On the other hand, all the five negative skewness measures are small (range between -0.664 for Hong Kong and 0.142 for the USA). The kurtosis measure is more than three in all the return series (between 4.16 for the UK and
487.79 for Thailand). The distribution of returns has thus fat tails compared with the normal distribution. In addition, the
index patterns of the Asian developing real estate markets have a larger value of excess kurtosis and conform less to
normality assumption than the developed markets of Australia, Japan and the UK. Finally, all the series’ JB values are
highly significant confirming that the distributions of return are not normal.
(Table 2 here)
4.
RESEARCH METHODOLOGY
The principal task in this research are to model the temporal dynamics of real estate market returns (1st
moment) and return volatilities (2nd moment) and investigate their causality relationships across the real estate
markets. In addition to the usual correlation analysis, we employ the causality-in-variance test developed by Cheung
and Ng (1996) to detect causal relations and identify patterns of causation in the first and second moments
respectively. The test involves a two-step procedure. The first stage requires the estimation of univariate time-series
models that permits time variations in both conditional means and conditional variances. In the second stage the
resulting series of residuals and squared residuals standardized by conditional variances are, respectively,
constructed. The cross-correlation function (CCF) of these standardized residuals and squared-standardized residuals
are used to test the null hypothesis of no causality in mean and variance respectively. In the present study, the
statistical procedures are outlined below.
(a)
Preliminary examination of the correlation in raw returns and volatilities (proxied by mean squared deviation)
of returns. Pearson Correlation analysis has been frequently used in the initial literature on international stock market
linkages.
(b)
To determine an appropriate univariate model to describe the real estate series that allow for time-varying
conditional means and variances. In addition, to capture the leverage effect found in many developing real estate stock
indices and to avoid imposing non-negativity on the values of the GARCH parameters to be estimated, we employ the
8
EGARCH representation developed by Nelson (1991). Specifically, the widely used EGARCH (1, 1) process is
employed to model the real estate stock returns, R t. and is given by:
Conditional mean equation
Rt = u t + φD97 + ε t or Rt = ARMA( p, q) + φD97 + ε t
Conditional variance equation
log
Where
εt
σ
2
t
= c + β log
is the unexpected return and
σ
2
t − 1
σ 2 t is
+ α
ε
σ
t − 1
t − 1
+ γ
ε
σ
t − 1
t − 1
the conditional variance. The conditional mean equation
includes a time dummy D 97 that takes a value of one for the period July 1997 – August 1998 and zero otherwise. The
intent is to control for regime swifts in many Asian real estate markets following the eruption of Asian financial crisis in
July 1997. This dummy is however not included in the US and UK mean equations. The main feature of the conditional
variance equation is that the leverage effect is exponential rather than quadratic. Hence forecasts of the conditional
variance are guaranteed to be non-negative. The presence of leverage effect can be tested by the hypothesis that
γ<
0. The impact is asymmetric if γ ≠ 0 .
In addition to EGARCH (1, 1), we also estimate another three models, namely EGARCH (1, 1)-M, EGARCH
(1, 1) - MA (1) and a variant of EGARCH (1, 1)-ARMA (p, q). For each market, the log-likelihood test is conducted to
select the best model to characterize the return series and estimate standardized residuals squared-standardized
residual series
(c)
Once appropriate time-series models are identified, sample cross-correlations (CCF) of the resulting
standardized residuals and squared standardized residuals at k lags are determined. Tests of causality in mean
(through standardized residuals) and causality-in variance (through squared standardized residuals) can be carried out
by a t-test for a particular lag, i.e.
9
t k = T ruv (k )
where rUV is the CCF of the squares of standardized residuals U t and V t at lag k . This statistic is used to
evaluate the causal relationship at a specific lag k comparing with the standard normal distribution.3
5.
RESULTS
5.1
Correlations in Returns and Volatilities
The sample correlations of the real estate stock returns and return volatilities are first estimated. The low
correlation between returns of international real estate markets has been regarded as an incentive to hold
internationally diversified real estate portfolios. In Table 3, the correlation matrices are calculated for the return (Panel
A) and mean squared deviation of the returns (Panel B) (used to represent return volatilities). This is because
international real estate markets might be related through their returns, or their return volatilities or both.
(Table 3 here)
As Panel A of Table 3 shows, correlations between the returns of Asian real estate markets are between low
and moderate ranges. Here we see that only the highest correlation in returns is 0.671 (between Singapore and Hong
Kong) and only another two pairs’ correlation coefficients are between 0.40 and 0.50. Additionally, none of the Asian
developing real estate markets shows a correlation higher than 0.25 with Japan. Finally, correlations between the
returns of all individual Asian markets and the returns of the USA and UK are considerably low, with only 3 out of 16
correlation coefficients are higher than 0.20. The highest return correlation in only 0.230 (between HK and the UK) and
the lowest correlation coefficient is a negative of 0.010 (between Indonesia and the UK)
In Panel B of Table 3, it is observed that the correlation coefficients for the mean squared deviations (return
volatilities) are lower. Here we see that only 3 pairs of the 21 Asian markets’ pair-wise correlation coefficients are
higher in return volatilities than in returns (i.e. Singapore and HK, Singapore and Philippines, Japan and Philippines).
Again, Singapore and Hong Kong report the highest correlation coefficient of 0.727. Thailand has a negatively weak
correlation coefficient with all Asian markets except Singapore and the Philippines. Similar results hold for both the
10
USA and UK, where all the correlations of return volatilities are lower or almost uncorrelated with the Asian markets.
These findings are generally in agreement with what appears in the literature. Similar to stock markets, International
real estate markets show small to moderate correlation in returns, but some markets are not independent because they
are related through their second moments.
5.2
Univariate Specification
For each market, Table 4 reports the log-likelihood ratios for an EGARCH (1, 1), EGARCH (1, 1) – M, MA (1)
- EGARCH (1, 1) and an ARMA (p, q) - EGARCH (1, 1) model. The model with the largest increase in log-likelihood
value over the EGARCH (1, 1) model is selected to calculate standardized residuals.
(Table 4 here)
The maximum likelihood estimates of the appropriate time-series models for each market are reported in
Table 5. They are EGARCH (1, 1) (HK, Japan, Indonesia, UK and US); MA (1) - EGARCH (1, 1) (Thailand); ARMA (1,
1) - EGARCH (1, 1) (Malaysia) and ARMA (2, 2) - EGARCH (1, 1) (Singapore and Philippines). The log-likelihood
values (between -1334.32 and -2822.25) are large enough to suggest the respective volatility models are able to
capture the temporal dependence of volatility reasonably well. Specifically, all ten index returns display a significant
GARCH (1, 1) effect (between 0.840 for the US and 0.984 for Indonesia). With minor exceptions, all coefficients of the
conditional variance equation are significantly greater than zero. Consistent with expectation, we find a significant and
negative Asian financial crisis coefficient ( φ ) each for Singapore, Hong Kong, Indonesia, Malaysia, Thailand and the
Philippines real estate markets. Finally, all leverage coefficients are negative and are, with the exception for Japan,
statistically significant at least at the 5% level.
(Table 5 here)
The results of Table 5 are generally consistent with those of other stock market empirical work on timevarying volatility. However, the similarities between developing and developed real estate markets documented in
Table 6 hide some interesting differences. Specifically, the estimated conditional volatilities for all developing Asian real
estate markets are considerably larger than those of the two developed real estate markets of the US and UK. As the
figures in Table 6 indicate, the average values of the conditional standard deviation confirm the fact that developing
11
Asian-Pacific real estate markets of Thailand (mean = 15.62%), Indonesia (mean = 9.27%), Philippines (mean =
5.58%), Malaysia (mean = 5.16%), Hong Kong (mean = 4.61%) and Singapore (mean = 4.42%) are significantly higher
than those of the UK (mean = 2.55%) and US (mean = 1.81%).4 Moreover, the estimated conditional volatility series for
the developing Asian real estate markets show a higher degree of dispersion and suggests that large changes in
volatility are more frequent than in developed real estate markets. Finally, both the maximum and minimum values of
the conditional volatility are considerably larger in the developing Asian Pacific real estate markets. Overall, this
conditional evidence is supportive of the unconditional statistics reported in Table 2.
(Table 6 here)
Finally, Table 7 contains the results of the diagnostic tests of the various univariate country models. A LjungBox test with 36 lags (i.e. LB (36)) is conducted on standardized residuals from the respective models to detect serial
correlation. In addition, LB2 (36) is used to determine whether there is any serial correlation in the squared
standardized residual series. The results reveal that with the exception of LB (36) for Indonesia, the remaining eight LB
(36) and LB2 (36) statistics for all 9 markets support the chosen EGARCH model specification at a 5 percent level of
significance.
(Table 7 here)
Since the choice of lag length is likely to affect the causality results, following Cheung and Ng (1996) we use
up to five leads and five lags (i.e. lags = ± 5 ) to carry out the CCF tests. Tables 8 and 9 report the t-statistics for the
causality-in-mean test and causality-in-variance test respectively.
(Tables 8 and 9 here)
5.3
Causality-in-mean
Out of 36 pairs of the conditional return series, 16 pairs are contemporaneously correlated. For example,
Hong Kong is significantly correlated with all other markets except with Thailand. Similarly, Singapore real estate return
is significantly correlated with all other real estate return series except with Indonesia and Thailand. Whilst the US and
UK returns are contemporaneously correlated with Singapore and Hong Kong, the remaining Asian markets are not
significantly correlated with the UK and USA markets. Finally, the ranges of significant correlations are only between
8.4% (Singapore and the US) and 33.4% (Singapore and Hong Kong).
12
All significant causality in returns between some markets takes places between one to five lags.5 Among the
Asian markets, there are statistically significant causal relations in return from the real estate market of Singapore and
Hong Kong, respectively, to Japan, Indonesia, Malaysia and Philippines at different lags. Similarly, the null hypothesis
of no significant causal relation from Japan to Malaysia and Indonesia to Malaysia is also rejected. Other significant
one-way causal relations in return are from the UK to Singapore at lag 1 and from the US to UK at lag 4.
A feedback relationship at the mean level implies that there are simultaneous adjustments in the returns of
the markets. It exists between the real estate markets of (i) Singapore and Hong Kong; and (ii) Indonesia and
Philippines. For example, the Singapore real estate index affects Hang Seng real estate index (Hong Kong) at lag 2,
and the Hang Seng real estate index affects Singapore real estate index at lag 5. Additionally, the impact from
Singapore to HK is stronger than that from HK to Singapore. Likewise, Indonesia’s real estate returns seem to cause
the Philippines returns at lag 2 and vice versa.
Finally, there are three instances where the contemporaneous correlations are statistically insignificant, the
correlations at other lags are statistically significant. One example is the contemporaneous correlation between
Singapore and Indonesia is merely 0.031, but the cross-correlation at lag 2 is 0.086 and is statistically significant at the
5% level. The implication is that this lag real estate return of Singapore causes the Indonesia real estate returns
Nevertheless, all significant cross-correlations are much weaker than their contemporaneous correlation coefficients.
The largest cross-correlation coefficient is only 8.6% in this instance.
5.4
Causality-in-variance
First, 13 pairs of the conditional variance series are contemporaneously correlated. Again, Singapore and
Hong Kong have the highest number of cases of significant correlation coefficient with other markets (five each). Whilst
the UK real estate market is significantly correlated (in return variance) with Singapore and Japan, the US market is
significantly correlated with Hong Kong only. Again, the highest volatility correlation coefficient is only 36.4% (between
Singapore and Hong Kong). Also, as in the case of return, the contemporaneous correlation coefficients between the
respective pairs of Asian-Pacific conditional volatility are generally higher than those with the UK and US.
13
As in the cases of mean causality, causality in variance between some markets also takes places between
one to five weeks lags. Among the Asian markets, variance causality exists from the real estate market of Singapore to
those of Hong Kong (at lag 2), Japan (at lags 2 and 3), Malaysia (at lag 1) and Philippines (at lag 3); from the real
estate market of Hong Kong to that of Singapore (at lag 5); from the Japanese real estate market to that of Philippines
(at lag 2); from the real estate market of Malaysia to that of Japan (at lag 1) and from the real estate market of
Philippines to that of Indonesia (at lag 2). Variance causality also exists from the US real estate market to those of
Singapore (at lags 2 and 5), Hong Kong (at lag 1) and Malaysia (at lag 1); and from the UK real estate market to those
of Singapore (at lag 5), Hong Kong (at lag 2) and Japan (at lag 2); implying that the lagged variances of the two highly
developed public real estate markets’ returns (i.e. USA and UK) cause the variances of developed Asian markets in
Japan, Hong Kong, Singapore and Malaysia.
A feedback relationship exists between six pairs of markets. This implies that there are simultaneous
adjustments in the return volatilities of these real estate markets. They are respectively, Singapore and Hong Kong,
Singapore and the UK, Hong Kong and the UK, Japan and the UK, and Indonesia and Philippines. Of these, there are
two instances (HK ↔ UK and Indonesia ↔ Philippines) where the contemporaneous correlations are statistically
insignificant.
Finally, similar to stock markets, International real estate markets also display bilateral causal
relationships in variances among some countries.
5.5
Composite picture
Table 10 presents the composite picture regarding the significance of both causality-in-mean and causality-
in-variance across the nine real estate markets. Out of the 36 market pairs, 16 pairs (44.4%) have no causal
relationship in both mean and variance. In addition, the majority of these 16 pairs have insignificant correlation
coefficients in contemporaneous return and variance. It is further observed that many of these pairs involve markets of
Indonesia, Thailand, Philippines and (to a lesser degree) Malaysia whose public real estate markets are still largely
segmented with the developed real estate markets. Causal relations in both mean and volatility are observed in 9 pairs
(25%). This is in contrast with Tay and Zhu (2000)’s evidence that both bi-directional causality in mean and variance
are observed in most cases in Asian Pacific-Rim stock markets. Our results indicates that six pairs of the significant
causal relations are observed among the Asian markets of Singapore, Hong Kong, Japan and Malaysia suggesting
14
geographical proximity and economic linkage do have a possible impact on creating causal relationships in returns and
volatility. The evidence also seem to suggest that information giving rise to returns and volatilities in the public real
estate markets are transmitted more rapidly among markets that are more efficiently organized and more open.
Additionally, causality in both mean and variance are also observed between Singapore and Philippines and between
Indonesia and Philippines. On the contrary, there is only one significant pair of both mean and volatility causality
observed between the developed markets and Asian markets (i.e. between the UK and Singapore). Our real estate
market evidence is thus different from some international stock market studies (such as Eun and Shum, 1989) that
have documented a stronger degree of interdependence in returns and volatilities between the stock markets of Japan,
Singapore, Hong Kong, the UK and the USA.
Finally, there are three other important observations and implications derived from the results. First, the US
real estate market exerts significant influence on Singapore, Hong Kong, Malaysia and the UK but not vice versa. This
evidence is consistent with the understanding that the US is the world’s most influential real estate market. Second it
appears that Singapore and Hong Kong, backed by their strong economic performance and leading financial centre
status in Asia, contribute significantly to the gradual integration of world real estate markets Not only that both real
estate markets are significantly correlated in their contemporaneous and lead-lag means and volatilities, the means
and variances of these two markets at different lags also have significantly impact on the current return and volatility of
some other markets. Third, at least half of the sample real estate markets (in particular, developing Asian real estate
markets) are still largely segmented implying that institutional investors would likely to benefit from diversifying public
real estate portfolios internationally across Asia and the US /UK markets in the short- and /or medium term. Our results
complement those of Liow et al. (2005) that there is lack of integration in the international public real estate markets.
In sum, the above causality results indicate that real estate markets in developing countries in Asian
(Thailand, Indonesia, Philippines and to a lesser degree, Malaysia) are perhaps caused more by country-specific
macroeconomic and market-specific factors rather than movements in the real estate markets of other economies . On
the other hand, causality effects exist in some Asian developed real estate markets. Notably, Singapore and Hong
Kong “lead’ other Asian real estate markets (including Japan) in most instances. Although this finding is a departure
from the usual understanding that Japanese market is the most influential Asian real estate market, the cross-market
15
differences and learning would definitely help fund managers in managing their exposure in Asian public real estate
markets.
(Table 10 here)
6.
CONCLUSION
The causal linkages among international real estate markets are of great interest from both academia and
investors. In this study, we focus on the causality in return and volatility among two highly developed real estate
markets (US and UK) and seven Asian developed and developing real estate markets (Singapore, HK, Japan,
Malaysia, Thailand, Indonesia and Philippines). We document the general direction and the lead / lag structure of
causality in the mean and the variance for the public real estate markets in the nine economies.
We successfully transfer a novel technique of Cheung and Ng (1996) from other investment markets to the
public real estate markets. In Particular, the causality- in-variance test, based on the residual cross-correlation function,
is robust to distributional assumptions. We find that Asian real estate markets show small to moderate correlation in
returns, but some markets are not independent because they are related through their second moments. There is
some evidence of causality-in-mean and / or causality-in variance among some Asian real estate markets. Based on
the causality results, the seven Asian real estate markets can be broadly classified into two groups. The first group is
composed of Singapore, Hong Kong, Japan and to a lesser extent, Malaysia. The second group is composed of
Thailand, Indonesia and Philippines. Specifically, the majority of the significant causal relations are observed among
the markets in the first group suggesting geographical proximity, economic linkage and market efficiency and openness
do have a possible impact on creating causality relationships in returns and volatilities. On the contrary, there is little
interaction between the real estate markets of the first and second groups. This classification seems to be consistent
with the stage of economic development and degree of stock market efficiency in the respective Asian economies.
Results of this study also support the notion that there is lack of integration in the international real estate markets as at
least half of the real estate markets are not causally linked. Consequently, a “pure” Asian real estate portfolio and a
“mixed” Asian real estate portfolios (i.e. with US and /or UK included) can be appropriately constructed using the
knowledge of presence or absence of causality (as well as cointegration) between the markets.
16
To conclude, our study is very significant as to-date no other research has considered comprehensively the
linkages across international real estate markets from both the causality-in-mean and causality-in-variance
perspectives. The study provides adequate support for fund managers to include volatility as a factor in the real estate
asset pricing models. Moreover, the regulators may use the causality results to identify the foreign market(s) that
drive(s) movements in the domestic real estate market and take appropriate corrective measures. Finally, the causality
results obtained in this study give academic researchers useful insights to design more appropriate bivariate models
(compared with univariate) with the appropriate lag structure. As more and more Asian economies are interested in
developing REIT type securitized real estate products, our study provide investors and regulators with additional
insights into the diversification potential of investing in major real estate markets. It also reinforces the increased
potential importance of Asian securitized real estate in an investment portfolio for both local and international investors.
Acknowledgement
The original version of this paper entitled “An international study of causality in mean and variance: evidence from real
estate markets” was presented at the at the 21st ARES Annual Meeting, 13-16 April 2005, Santa Fe, USA. We are
grateful to Professor Alastar Adair and conference participants for their helpful comments in improving the paper.
Thank you.
17
REFERENCES
Alaganar, V.T. and R. Bhar (2003), An international study of causality in variance: interest rate and financial sector
return, Journal of Economics and Finance 27: 39-55.
Cheung, Y. W. and L. K. Ng (1996), A causality-in-variance test and its application to financial market prices, Journal of
Econometrics 72: 33-48
Caporale, G.M., N. Pittis and N. Spagnolo (2002), Testing for causality-in-variance: an application to the East-Asian
Markets, International Journal of Finance and Economics 7: 235-245
Eun, G.S. and S. Shum (1989), International transmission of stock market movements, Journal of Financial and
Quantitative Analysis 24(2): 241-56
Garvey, R., G. Santry and S. Stevenson (2001), The linkages between real estate securities in the Asia-Pacific, Pacific
Rim Property Research Journal 7(4): 240-258
Hu, J.W.S., M.Y. Chen, R.C.W. Fok and B. N. Huang (1997), Causality in volatility and volatility spillover effects
between US, Japan and four equity markets in the South China Growth Triangular, Journal of International Financial
Markets, Institutions and Money 7: 351-367
Kanas, A. and G.P Kouretas (2002), Mean and variance causality between official and parallel currency markets:
Evidence from four Latin American countries, Financial Review 37: 137-164
Liow, K.H., O.T.L., Ooi and Y. Gong (2005), Cross-market dynamics in property stock markets: some international
evidence, Journal of Property Investment and Finance 23(1): 55-75
Liow, K.H. (2005), Real estate return volatility and systematic risk: evidence from international markets, working paper,
Department of Real Estate, National University of Singapore
Nelson, D. (1991), Conditional Heteroscedasticity in Asset Returns: A New Approach, Econometrica 59: 323-70
Tay, N.S.R and Z. Zhu (2000), Correlations in Returns and Volatilities in Pacific-Rim Stock Markets, Open Economic
Review 11: 27-47
UBS Investment Bank (2003), Global Real Estate Market Conditions, Research report
18
Table 1 Economic & stock market statistics (2003)
Hong
Kong
GDP*
Exchange rate*
US $ Billion
Local per
USD
Lending rate*
%
Consumer Price*
Unemployment rate*
%
Stock Market Captilization**
US $ Million
Value Traded **
US $ Million
Value Traded (/market cap)**
156.67
Singapore
93.56
Japan
4648.19
Malaysia
103.16
Indonesia
211.07
Thailand
149.79
Philippines
USA
UK
78.144
11004.1
1797.809742
55.569
NA
0.6118
7.787
1.7008
107.1
3.8
8465
39.591
5
5.31
1.82
6.3
16.94
5.94
9.472
4.12
3.69
92.9
101.1
98.1
104.3
130.1
102.3
112.541
106.8
106.5
11.4
6
3.1
7.9
5.4
5.3
3.5
NA
2.4
463,108
101,900
2,126,075
123,872
29,991
46,084
39,021
14,266,266
2,412,434
15,547,431
2,150,753
210,622
56,129
1,573,279
27,623
13,042
47,612
3,103
0.45
1
1
0
0
1
0.08
1.09
0.89
5,295
2,311
968
434
3,058
865
331
466
235
Average firm Size**
US $ Million
478.4
235
695
143
91
99
166.05
2694.29
1043.89
Real Estate stock % of stock market***
%
11.44
8.49
1.27
2.68
0.20
3.72
16
1.27
1.75
22.60
21.60
35.50
14.20
NA
37.80
16.5
25.9
22.7
0.49
0.99
5.54
2.88
No. of companies**
P/E ratio of real estate stock***
Dividend yield of real estate stock***
%
2.47
2.58
0.94
2.96
NA
Source: * data from IMF country database, the other data are extracted from Stock Market Factbook
** Source: Standard & Poor's Emerging Stock Markets Factbook 2003 and IMF
*** Source: Datastream International
19
Figure 1
Real Estate Market Index Movement: 1992-2004
400
350
300
250
200
150
100
50
Singapore
HK
Indonesia
Maylaysia
20
04
20
03
20
02
20
01
20
00
19
99
19
98
19
97
19
96
19
95
19
94
19
93
19
92
0
Philippine
350
300
250
200
150
100
50
Thailand
UK
U.S.
20
04
20
03
20
02
20
01
20
00
19
99
19
98
19
97
19
96
19
95
19
94
19
93
19
92
0
Japan
Notes: the market indexes are presented by two groups: emerging markets: Singapore, Hong Kong, Indonesia,
Malaysia, Philippine, Thailand and developed markets: UK, U.S. and Japan.
20
Table 2
Summary statistics of weekly real estate stock market returns: Jan 92-Dec 04
Mean(%)
Max(%)
Min(%)
Std(%)
Skewness
Kurtosis
Coefficient
of Variation
Jarque-Bera
(JB)
Sing
0.066
27.083
-29.026
5.004
-0.182
8.593
75.818
HK
0.117
22.331
-34.891
4.942
-0.664
8.992
42.239
JP
-0.026
23.230
-15.070
4.618
0.471
4.926
-177.615
Indo
0.036
338.378
-54.854
15.256
15.878
359.726
423.777
May
-0.099
42.226
-28.716
5.811
0.639
10.894
-58.697
Phil
-0.091
27.672
-22.447
5.930
0.384
5.699
-65.164
Thai
-0.026
409.978
-27.608
17.119
20.389
487.786
-658.423
UK
0.160
9.669
-12.162
2.604
-0.145
4.155
16.275
US
0.129
12.263
-10.024
1.950
-0.142
9.108
15.116
186.26**
106.24**
129.68**
361.82**
180.31**
222.19**
367.61**
40.02**
105.45**
Notes: ** Indicates two-tailed significance at 1% level.
Table 3
Correlation Results: Jan 92-Dec 04
Panel A: sample correlation of returns
Sing
HK
JP
Indo
May
Phil
Thai
UK
US
Sing
1.000
HK
0.671
1.000
JP
0.288
0.276
1.000
Indo
0.232
0.179
0.028
1.000
May
0.467
0.433
0.227
0.184
1.000
Phil
0.488
0.420
0.165
0.177
0.322
1.000
Thai
0.156
0.139
0.036
0.091
0.094
0.121
1.000
UK
0.175
0.230
0.221
-0.010
0.127
0.163
0.027
US
0.174
0.191
0.115
0.050
0.144
0.121
0.025
0.220
1.000
Panel B: Sample correlation in the variances (measured by mean squared deviation) of the returns
Sing
HK
JP
Indo
May
Phil
Thai
UK
US
Sing
1.000
HK
0.727
1.000
JP
0.261
0.238
1.000
Indo
0.016
0.004
0.003
1.000
May
0.322
0.389
0.179
0.004
1.000
Phil
0.532
0.373
0.250
0.005
0.226
1.000
Thai
0.003
-0.004
-0.012
-0.002
-0.005
0.001
1.000
UK
0.052
0.060
0.075
0.019
-0.006
0.022
-0.022
1.000
US
0.110
0.106
0.092
-0.007
-0.013
0.049
-0.013
0.021
1.000
21
Table 4
Likelihood Ratio Tests and Model Selection
EGARCH(1,1)
(A)
Singapore
Hong Kong
Japan
Indonesia
Philippine
Malaysia
Thailand
UK
US
-1914.82
-1967.55
-1966.15
-2538.35
-2109.57
-2021.79
-2832.49
-1585.08
-1334.32
Log Likelihood
EGARCH(1,1)MA(1)M
EGARCH(1.1)
(B)
(C)
-1914.59
-1915.39
-1969.38
-1967.04
-1965.73
-1966.33
-2583.44
-2799.61
-2108.61
-2109.02
-2024.02
-2016.66
-2846.43
-2822.25
-1586.33
-1584.86
-1332.92
-1338.12
ARMA (p, q)EGARCH(1,1)
(D)
-1903.67
N.A.
N.A.
N.A.
-2095.41
-2012.92
N.A.
N,A.
N.A.
A
versus
B
0.46
-3.66
0.84
-90.18
1.92
-4.46
-27.88
-2.5
2.8
LR Test
A versus
C
A versus
D
-1.14
1.02
-0.36
-522.52
1.10
10.26**
20.48**
0.44
-7.6
22.3**
N.A.
N.A.
N.A.
28.32**
17.74**
N.A.
N,A.
N.A.
Notes: The LR statistic is computed as: LR = −2( Lu − Lr ) . H 0 is EGARCH (1, 1) model. Lr & Lu are log
likelihood of the unrestricted and restricted models. The LR statistic has an asymptotic χ distribution with degrees of
freedom equal to the number of restrictions (the number of added variables). Finally, the appropriate specifications
are: Singapore ARMA (2,2)-EGARCH(1,1);Hong Kong and Japan EGARCH(1,1), Indonesia EGARCH(1,1), Philippine
ARMA (2,2)-EGARCH(1,1),Malaysia ARMA (1,1)-EGARCH(1,1); Thailand MA (1) - EGARCH(1,1);; UK and U.S.
EGARCH(1,1)
2
Table 6 Summary Statistics for Conditional Standard Deviation
Index
Singapore
Hong Kong
Japan
Indonesia
Malaysia
Philippines
Thailand
UK
USA
Mean
4.4175
4.6138
4.5339
9.2655
5.1594
5.5762
15.6163
2.5509
1.8100
Std Dev Maximum Minimum
2.0353
13.0118
2.0413
1.4787
14.2084
2.3475
1.0060
7.2677
2.6036
4.1729
17.1958
1.6743
2.3927
17.0351
2.4104
1.5432
10.8635
3.0378
2.4407
17.7781
0.0051
0.4660
5.2707
1.8352
0.6031
5.5546
1.0670
22
Table 5
Estimates of EGARCH models
Mean equation: rt = u t + φD97 + ε t or rt = ARMA( p, q ) + φD97 + ε t
Variance equation: log
Singapore
EGARCH(1,1)-ARMA(2,2)
Hong Kong
EGARCH(1,1)
Japan
EGARCH(1,1)
Indonesia
EGARCH(1,1)
Philippine
EGARCH(1,1)-ARMA(2,2)
Malaysia
EGARCH(1,1)-ARMA(1,1)
Thai
EGARCH(1,1)-MA(1)
UK
EGARCH(1,1)
US
EGARCH(1,1)
σ
2
t
= c + β log
t −1
+ γ
t −1
ε
σ
t −1
t −1
α
γ
LogL
N.A.
N.A.
N.A.
N.A.
-0.086**
(0.03)
0.017
(0.04)
0.003
(0.03)
0.105**
(0.001)
-0.012
(0.027)
-0.096**
(0.035)
0.056**
(0.007)
0.031
(0.033)
-0.068
(0.046)
0.969**
(0.008)
0.935**
(0.01)
0.969**
(0.08)
0.984**
(0.004)
0.982**
(0.10)
0.973**
(0.08)
0.882**
(0.09)
0.948**
(0.019)
0.840**
(0.029)
0.227**
(0.04)
0.226**
(0.03)
0.112
(0.03)
0.036**
(0.01)
0.099**
(0.028)
0.245**
(0.039)
0.065*
(0.03)
0.079*
(0.036)
0.322**
(0.053)
-0.065**
(0.015)
-0.093**
(0.02)
-0.091
(0.02)
-0.054**
(0.01)
-0.078**
(0.017)
-0.046**
(0.019)
-0.067**
(0.012)
-0.053**
(0.021)
-0.163**
(0.033)
-1903.68
0.473**
(0.18)
-0.25**
(0.08)
0.37*
(0.18)
N.A.
-0.117*
(0.05)
-0.579*
(0.28)
-0.024
(0.36)
-0.225*
(0.10)
-0.377**
(0.11)
-0.356*
(0.145)
0.443**
(0.08)
N.A.
N.A.
-0.175**
(0.03)
N.A.
0.277**
(0.04)
N.A.
N.A.
N.A.
N.A.
N.A.
-0.360*
(0.18)
0.369
(0.27)
N.A.
0.160*
(0.08)
0.129*
(0.06)
ε
σ
φ
ut
N.A.
+ α
β
MA
N.A.
2
t −1
c
AR
0.445*
(0.21)
-0.012
(0.27)
-0.167*
(0.09)
N.A.
σ
-1967.55
-1966.15
-2538.35
-2095.41
-2012.92
-2822.25
-1585,08
-1334.32
Note: **,* indicate two-tailed significance level at the 1% and 5% respectively.
23
Table 7
Univariate model diagnostics
Real estate market
Singapore
Hong Kong
Japan
Indonesia
Philippines
Malaysia
Thailand
UK
USA
Model
ARMA(2,2)-EGARCH(1,1)
EGARCH(1,1)
EGARCH(1,1)
EGARCH(1,1)
ARMA(2,2)-EGARCH(1,1)
ARMA(1,1)-EGARCH(1,1)
MA(1)-EGARCH(1,1)
EGARCH(1,1)
EGARCH(1,1)
LB (36)
27.02
47.19
26.10
124.76*
33.53
25.99
0.613
28.18
33.98
LB2 (36)
21.95
34.18
25.34
0.112
30.89
23.48
0.061
40.61
41.40
Notes
LB (36) is a Ljung Box statistic at lag 36 to detect serial correlation in the standardized residual series. LB2(36) is a
Ljung Box statistic using squared standardized residual series to detect remaining heteroskedasticity
* Indicates two-tailed significance at the 5% level
24
Table 8
t-statistics for the causality-in-mean test
Real estate
market
Sing->HK
HK->Sing
Sing->JP
JP->Sing
Sing->indo
Indo->Sing
Sing->May
May->Sing
Sing->Phil
Phil->Sing
Sing->Thai
Thai->Sing
Sing->UK
UK->Sing
Sing->US
US->Sing
HK->JP
Japan->HK
HK->Indo
Indo->HK
HK->May
May->HK
HK->Phil
Phil->HK
HK->Thai
Thai->HK
HK->UK
UK->HK
HK->US
US->HK
JP->Indo
Indo->JP
JP->May
May->JP
JP->Phil
Phil->JP
JP->Thai
Thai->JP
JP->UK
UK->JP
JP->US
US->JP
Indo->May
May->Indo
Indo->Phil
Phil->Indo
Indo->Thai
Thai->Indo
Lag length
0
1
2
3
4
5
0.334**
.
0.141**
0.021
-0.047
0.046
-0.041
0.025
0.047
0.093*
-0.011
0.004
-0.023
-0.015
-0.012
0.061
0.130**
0.039
0.061
0.118**
0.014
0.058
0.033
0.144**
-0.006
0.027
0.059
0.001
-0.037
0.061
0.050
-0.031
0.020
-0.058
0.006
0.098**
0.066
0.020
0.004
-0.018
-0.003
0.008
-0.050
-0.012
-0.030
0.081*
-0.028
0.043
0.016
-0.017
-0.018
-0.095**
-0.057
0.075*
-0.048
0.086*
0.037
0.023
-0.031
0.026
0.048
-0.022
-0.010
0.009
0.030
0.003
0.074
0.066
-0.045
0.071*
0.063
0.045
-0.021
0.031
0.065
-0.027
0.034
0.009
0.042
-0.002
0.018
-0.066
0.021
-0.070
0.041
0.001
0.031
-0.023
-0.019
-0.011
0.081
0.011
0.060
0.069
0.025
0.122**
0.154**
-0.018
-0.018
-0.004
-0.063
0.049
-0.012
-0.007
-0.005
0.013
0.048
0.091*
0.001
0.011
0.005
-0.043
-0.021
0.035
-0.026
0.008
-0.014
0.013
0.017
0.047
-0.002
0.084*
0.002
-0.024
-0.019
-0.043
-0.018
-0.005
0.000
0.006
-0.007
-0.003
-0.014
-0.070
-0.004
-0.039
0.024
0.024
0.009
0.017
0.020
0.024
0.001
0.042
0.046
-0.017
-0.018
-0.054
-0.083
-0.017
-0.010
0.097
-0.008
0.011
-0.009
0.026
0.008
-0.025
-0.043
-0.001
0.021
-0.036
-0.012
-0.020
0.006
0.075*
0.053
0.046
0.009
0.037
0.055
0.011
0.019
-0.001
0.043
-0.002
0.024
-0.010
-0.017
0.015
0.011
-0.051
0.010
-0.044
0.015
0.045
-0.041
-0.005
-0.024
0.039
-0.005
0.037
0.044
-0.018
-0.018
0.016
-0.073*
0.086
-0.030
0.005
0.046
0.016
-0.047
-0.001
0.039
-0.044
-0.038
0.018
0.039
0.002
0.080
0.097**
0.002
0.064
0.027
0.022
-0.005
0.043
0.047
0.003
-0.012
0.018
0.020
0.018
0.045
-0.016
-0.031
-0.005
0.030
0.064
0.057
0.032
-0.042
0.063
0.042
0.057
0.050
-0.020
0.067
0.003
0.014
-0.018
-0.018
0.031
.
0.190**
.
0.234**
.
-0.008
.
0.111**
.
0.084**
.
0.161**
.
0.113**
.
0.242**
.
0.254**
.
0.004
.
0.085*
.
0.101**
.
-0.023
0.127**
.
0.097**
0.026
.
0.071
.
0.072
.
0.065
0.129**
-0.017
25
Indo->UK
UK->Indo
Indo->US
US->Indo
May->Phil
Phil->May
May->Thai
Thai->May
May->UK
UK->May
May->US
US->May
Phil->Thai
Thai->Phil
Phil->UK
UK->Phil
Phil->US
US->Phil
Thai->UK
UK->Thai
Thai->US
US->Thai
UK->US
US->UK
0.024
-0.023
0.149**
-0.035
.
0.064
-0.022
-0.013
0.026
.
0.000
.
-0.012
.
-0.039
.
0.069
.
-0.007
0.037
-0.060
0.067
-0.018
0.022
-0.009
0.008
0.008
0.019
0.042
-0.015
-0.012
-0.043
0.015
0.018
-0.043
0.052
-0.007
-0.036
-0.014
-0.038
-0.002
0.036
0.069
-0.025
-0.014
-0.001
0.035
0.017
-0.035
0.058
-0.036
0.020
0.002
0.090
-0.033
0.041
-0.003
-0.006
-0.020
-0.028
-0.042
-0.013
0.050
-0.015
0.010
0.063
-0.039
-0.036
-0.020
-0.036
0.065
-0.020
-0.016
0.047
-0.013
0.034
0.012
-0.048
-0.001
-0.037
0.021
0.054
-0.025
0.058
0.004
-0.029
0.058
-0.018
-0.017
0.052
-0.043
-0.043
-0.061
0.045
-0.023
0.040
-0.007
0.061
0.081
0.004
-0.002
-0.005
0.009
-0.013
0.015
-0.002
-0.027
-0.018
0.053
0.028
-0.011
-0.010
-0.033
0.074*
0.034
0.003
-0.054
-0.042
0.034
-0.012
-0.008
-0.031
0.013
0.002
0.053
-0.036
0.031
-0.024
0.081
-0.044
-0.061
0.067
0.002
-0.032
-0.006
-0.011
0.020
0.035
Notes: Lag 0 is the contemporaneous correlation; ** and * indicate significance at the 1% and 5% levels. Test statistics
is defined as t k =
^
T γ uv (k ) , where k=0, 1, 2, 3, 4, 5. γ uv = C uv (k )(C uu (0)C vv (0)) −1 / 2
26
Table 9
t-Statistics for the causality-in-variance test
Lag length
Real estate
market
Sing->HK
HK->Sing
Sing->JP
JP->Sing
Sing->indo
Indo->Sing
Sing->May
May->Sing
Sing->Phil
Phil->Sing
Sing->Thai
Thai->Sing
Sing->UK
UK->Sing
Sing->US
US->Sing
HK->JP
Japan->HK
HK->Indo
Indo->HK
HK->May
May->HK
HK->Phil
Phil->HK
HK->Thai
Thai->HK
HK->UK
UK->HK
HK->US
US->HK
JP->Indo
Indo->JP
JP->May
May->JP
JP->Phil
Phil->JP
JP->Thai
Thai->JP
JP->UK
UK->JP
JP->US
US->JP
Indo->May
May->Indo
Indo->Phil
Phil->Indo
Indo->Thai
0
1
2
3
4
5
0.364
0.020
-0.023
0.063
0.004
-0.012
0.036
0.077*
0.001
-0.024
0.008
-0.016
-0.015
0.123**
0.066
0.029
0.051
0.146**
-0.003
-0.002
0.001
0.101**
-0.003
0.005
0.039
-0.009
-0.019
0.066
0.024
-0.015
0.082*
-0.025
-0.008
0.075*
0.103**
-0.002
0.010
-0.018
-0.012
0.000
-0.049
-0.004
-0.021
0.026
-0.020
-0.009
-0.015
-0.003
0.083*
-0.042
0.076*
-0.010
0.021
0.016
-0.015
-0.030
0.026
0.031
-0.018
-0.014
0.008
0.024
-0.021
0.075*
0.045
-0.035
0.005
-0.011
0.006
-0.020
0.006
0.051
-0.018
0.014
0.024
0.084*
0.019
0.030
-0.021
0.010
-0.031
0.042
-0.004
0.010
-0.019
-0.019
-0.018
0.076*
-0.022
0.049
0.013
-0.013
0.114**
0.194**
-0.003
-0.012
-0.056
0.103**
-0.021
-0.014
-0.015
0.001
0.018
0.113**
0.007
-0.002
-0.006
-0.029
-0.032
0.019
-0.022
-0.006
-0.022
-0.014
-0.016
0.019
0.009
0.097
-0.010
-0.018
-0.017
-0.032
-0.020
-0.013
0.000
-0.001
-0.016
-0.020
-0.016
-0.057
0.021
-0.022
0.008
0.091*
0.011
0.018
-0.008
-0.002
-0.013
0.013
-0.010
-0.003
-0.034
-0.042
0.008
0.002
0.061
-0.019
0.022
-0.011
0.066
-0.009
-0.018
-0.021
-0.004
0.042
-0.028
-0.004
-0.032
0.091
0.010
-0.008
0.048
0.000
0.009
0.038
-0.004
0.002
0.042
0.017
-0.010
0.009
-0.018
-0.011
0.039
-0.013
-0.056
0.004
-0.023
0.000
0.018
-0.034
-0.044
0.005
-0.010
-0.017
0.002
-0.009
-0.003
0.000
0.082*
0.047
-0.039
-0.016
0.021
-0.008
-0.007
-0.003
0.033
-0.021
-0.021
0.028
0.083*
-0.014
0.098*
0.057
0.019
-0.006
-0.013
0.002
0.044
0.005
-0.002
-0.007
-0.014
0.083*
-0.004
0.017
0.048
-0.015
-0.016
0.004
0.033
0.081*
0.043
0.013
-0.023
0.073
0.003
0.100
0.108
-0.017
-0.003
-0.011
-0.017
-0.003
0.143**
-0.012
0.354**
.
0.208**
-0.014
0.130**
.
0.067
.
0.152**
0.000
.
0.290**
.
0.223**
-0.009
.
0.053
.
0.120**
.
0.000
.
0.094**
.
0.088**
.
0.010
.
0.079*
.
0.063
.
-0.002
.
0.005
.
-0.003
27
Thai->Indo
.
-0.003
-0.003
-0.003
-0.003
-0.003
Indo->UK
-0.013
-0.023
0.059
-0.028
-0.027
0.025
UK->Indo
.
0.047
-0.023
-0.028
-0.027
-0.003
Indo->US
-0.018
-0.023
0.002
-0.020
-0.023
-0.021
US->Indo
.
0.087
-0.004
-0.018
0.040
-0.020
May->Phil
0.093**
-0.031
0.022
0.037
-0.040
0.052
Phil->May
.
0.020
-0.009
-0.005
0.008
-0.021
May->Thai
-0.018
-0.011
-0.018
-0.014
-0.011
-0.012
Thai->May
.
-0.004
0.036
0.023
0.038
-0.017
May->UK
0.071
0.018
-0.009
-0.019
0.060
0.002
UK->May
.
0.011
0.000
-0.001
-0.022
0.000
May->US
-0.014
-0.013
0.010
-0.044
0.041
-0.005
US->May
.
0.076*
-0.002
-0.012
-0.020
0.060
Phil->Thai
-0.016
-0.015
-0.021
-0.010
-0.005
0.013
Thai->Phil
.
-0.022
0.024
-0.021
-0.015
-0.019
Phil->UK
-0.012
-0.014
-0.003
-0.009
0.015
0.062
UK->Phil
.
0.019
-0.010
0.048
-0.021
-0.056
Phil->US
0.016
-0.026
-0.027
-0.039
-0.002
-0.062
US->Phil
.
0.043
-0.038
0.038
-0.022
0.048
Thai->UK
-0.016
-0.014
-0.025
-0.007
0.042
-0.010
UK->Thai
.
-0.024
-0.017
-0.023
0.013
-0.024
Thai->US
-0.019
-0.014
0.031
0.038
-0.014
-0.012
US->Thai
.
-0.019
-0.015
-0.016
-0.013
-0.013
UK->US
0.050
0.029
-0.004
-0.008
-0.045
0.061
US->UK
.
0.007
0.037
0.030
-0.062
0.051
Notes: Lag 0 is the contemporaneous correlation; ** and * indicate significance at the 1% and 5% level. Test statistics
is defined as t k =
^
T γ uv (k ) , where k=0,1, 2, 3, 4, 5. γ uv = C uv (k )(C uu (0)C vv (0)) −1 / 2
28
Table 10
Summary of causality results
Panel A
A
B
Causality-in-mean
A→B
B→A
correlation
(Y/N)
(Y/N)
(Y/N)
Y
Y(-2)
Y(-5)
Y
Y(-2)
N
N
Y(-2)
N
Y
Y(-1)
N
Y
Y(-3)
N
N
N
N
Y
N
Y(-1)
Y
N
N
Y
Y(-1,-5)
N
Y
Y(-2,-4)
N
Y
Y(-1)
N
Y
Y(-3)
N
N
N
N
Y
N
N
Y
N
N
N
N
N
Y
Y(-1)
N
Y
N
N
N
N
N
N
N
N
N
N
N
N
Y(-1)
N
Y
Y(-2)
Y(-2)
N
N
N
N
N
N
N
N
N
Y
N
N
N
N
N
N
N
N
N
N
N
N
N
N
N
N
N
N
N
N
N
N
N
N
N
N
N
N
Y(-4)
16
10*
2*
1Significant
SIN
HK
SIN
JP
SIN
INDO
SIN
MAL
SIN
PHI
SIN
THAI
SIN
UK
SIN
US
HK
JP
HK
INDO
HK
MAL
HK
PHI
HK
THAI
HK
UK
HK
US
JP
INDO
JP
MAL
JP
PHI
JP
THAI
JP
UK
JP
US
INDO
MAL
INDO
PHI
INDO
THAI
INDO
UK
INDO
US
MAL
PHI
MAL
THAI
MAL
UK
MAL
US
PHI
THAI
PHI
UK
PHI
US
THAI
UK
THAI
US
UK
US
Overall (No of
YES)
A↔B
(Y/N)
Y
N
N
N
N
N
N
N
N
N
N
N
N
N
N
N
N
N
N
N
N
N
Y
N
N
N
N
N
N
N
N
N
N
N
N
N
2
Causality-in-variance
A→B B→A
correlation
(Y/N)
(Y/N)
(Y/N)
Y
Y(-2)
Y(-5)
Y
Y(-2,-3)
N
N
N
N
Y
Y(-1)
N
Y
Y(-3)
N
N
N
N
Y
Y(-1)
Y(-5)
N
N
Y(-2,-5)
Y
Y(-1)
N
N
N
N
Y
Y(-1)
N
Y
N
N
N
N
N
N
Y(-5)
Y(-2)
Y
N
Y(-1)
N
N
N
Y
Y(-1)
Y(-1)
Y
Y(-5)
N
N
N
N
Y
Y(-3)
Y(-2)
N
N
N
N
N
N
N
Y(-2)
Y(-2)
N
N
N
N
N
N
N
N
N
Y
N
N
N
N
N
N
N
N
N
N
Y(-1)
N
N
N
N
N
N
N
N
N
N
N
N
N
N
N
N
N
N
13
6*
3*
1Significant
A↔B
(Y/N)
Y
N
N
N
N
N
Y
N
N
N
N
N
N
Y
N
N
Y
N
N
Y
N
N
Y
N
N
N
N
N
N
N
N
N
N
N
N
N
6
Notes: This summary is constructed based on the results reported in Exhibits 9 and 10; 1significant correlation (Y/N)
indicates whether significant contemporaneous correlation (in mean / volatility) between two markets exists; Y (-k)
indicates the presence of correlation at lag k. For example, Y (-2) (SIN → HK) indicates that the return (/ volatility) of
the Singapore real estate market at lag 2 has an impact on the current return (/ volatility) of the Hong Kong market. *
exclude the feedback cases.
29
Panel B Further analysis
(a)
Total number of market-pairs
:
36 (100%)
(b)
Number of pairs that display both causality-in-mean and causality-in-variance :
9 (25.0%)
(c)
Number of pairs that display causality-in-mean only
:
5 (13.9%)
(d)
Number of pairs that display causality-in-variance only
:
6 (16.7%)
(e)
Number of pairs that have NO causality- in-mean and causality-in-variance
:
16 (44.4%)
Notes
1 Two other Asian economies, Taiwan and South Korea are not included in this study as no equivalent real estate stock
index series are publicly available.
The Dow Jones Global Indexes (DJGI) provides comprehensive world indexes to help international investors in
portfolio management and benchmarking. DJGI calculates indexes on 80% of the investable market capitalization in 34
countries including both developed and developing markets. In addition, the DJGI family includes indexes for each of
the 10 economic sectors, 18 market sectors, 51 industry groups and 89 subgroups defined by the Dow Jones Global
Classification Standard. Real estate index (code: 8730) is one of the industry sector indexes and comprises two subsector indexes: (a) code 8733: real estate holding and development; and (b) code 8737: real estate investment trusts
(http://djindexes.com). Datastream only has the aggregate real estate sector price indexes but does not have the index
data for the two real estate sub-sectors.
2
Using Monte Carlo simulation, Cheung and Ng (1996) have shown that “….this test has the ability to identify causality
and reveal useful information on the causality patterns…..” Furthermore, it is robust to non-symmetric and leptokurtic
errors and asymptotically robust to distributional assumptions.
3
The only developed Asian-Pacific real estate markets is Japan whose average conditional standard deviation (mean
= 4.53%) is comparable to that of Hong Kong and Singapore.
4
5 In their study of six Pacific-Rim and US stock markets, Tay and Zhu (2000) report that statistically significant causal
effects can arise from observations at lags as long as 11 weeks
30