IRES2012-022
IRES Working Paper Series
Dynamics of Economic Globalization and Capital Market Integration: Evidence from Greater China and International Linkages
Kim Hiang LIOW Qing YE
August 2012
Dynamics of Economic Globalization and Capital Market Integration: Evidence from Greater China and International Linkages By: Kim Hiang LIOW (
[email protected])1 and Qing YE (
[email protected])2 Department of Real Estate, National University of Singapore
1
Professor of real estate, Department of Real Estate, National University of Singapore, email:
[email protected] 2
PhD student, Department of Real Estate, National University of Singapore, email:
[email protected]
Dynamics of Economic Globalization and Capital Market Integration: Evidence from Greater China and International Linkages Abstract In this paper, the dynamics and current status of economic globalization and capital market correlation, as well as their interdependence, among the three Greater China (GC) economies (mainland China, Hong Kong and Taiwan) and across the GC areas with their regional and international partners (Japan, Singapore, the US and the UK) is assessed, using monthly data on onemonth interbank rates, exchanges rates, inflation and realized correlation. Despite the limited extent of economic globalization and inconclusive time-trend evidence, the unit root and mean reversion results imply that real interest rates, uncovered interest rates and relative purchasing power in the GC context tend to converge and therefore the three parity conditions are likely to hold as a long-run equilibrium condition, respectively. The increased realized cross-real estate securities market correlations and realized cross-stock market correlations imply that international capital markets have become increasingly integrated over the study period. Finally, the integration spillover index and plot investigations have detected some evidence of interdependence between economic globalization and capital market integration. Our results should be regarded as preliminary, but indicative. They suggest that economic globalization could be one of the key driving forces of capital market integration. Further studies are definitely required to order to confirm whether the GC results reported in this paper could be generalizable to e developed and emerging economies. Keywords: economic globalization, capital market integration, realized correlation, international parity, integration spillover index
1.
Introduction With the trend of economic globalization since 1980’s, China has become one of the fastest
growing economies in the world, an informal economic region that embraces Mainland China, Hong Kong and Taiwan is rapidly emerging as a new epicenter for industrial, commerce and Finance. The accession of China to the WTO in November 2001 further marks a distinctive break in China’s economic relation with the rest of the world. The economic globalization between the Mainland China and Hong Kong has been enhanced since Hong Kong returned to China in 1997. In addition, China is the largest recipient of Taiwan’s overseas investment and Taiwan is China’s third largest source of direct investment. The mainland China, Hong Kong and Taiwan, which are often identified as the Greater China (GC) region, has emerged as one of the most dynamic economic regions in the world, and contributes 23.70% of world GDP in 20103. The export in the GC region accounts for 13.9% of world’s total export while the import value
3
Data source: World Bank and National Statistics (Taiwan).
accounts for 12.08% in 20094. The foreign exchange reserve in these economies is also among the largest in the world. Increasing globalization of financial and economic activities is expected to impact real estate market which is a significant asset component in the three GC economies. Economic literature has pointed out that in addition to positively impact trade activities and investment flows, global integration is capable of influencing real and financial markets through the pricing of and investment in local real estate markets as well as in international capital markets (Bardhan et al. 2008). Specifically, with continuing strong economic growth, massive urbanization and the growth of private real estate ownership in China, the scope for Hong Kong REITs to provide more pure-play property investment opportunities in China, as well as Taiwan’s growing economic ties with China, the GC region is expected to grow into an important player in the global capital markets, with their direct real estate and securitized real estate markets attracting the interest of domestic and international investors; although Fung et al (2006) have noted that the three GC real estate securities markets are quite different in terms of their macroeconomic conditions, degree of market openness and transparency, legal system, size and maturity level, as well as levels of government intervention on the real estate market. As real estate companies invest in real properties and are themselves real estate vehicles for other investors in international capital markets, it is timely to investigate the nature and extent of interdependence between economic globalization and capital market integration as literature has suggested international capital markets are closely correlated in today’s global economy. Following literature, increased capital market correlations imply, respectively, higher across-stock and higher acrossreal estate stock integration. The core objective of this paper is to assess the real interest parity (RIP), uncovered interest parity (UIP) and deviations from relative purchasing power parity (RPPP) hypotheses within the GC economies, as well as across the GC areas with their selected regional and world partners (the US, the UK, Japan and Singapore); to assess whether the corresponding capital markets (i.e. stock and real estate securities markets) have become increasingly correlated; and to examine whether there is significant interdependence between economic globalization and capital market integration, as well their dynamic changes over time.
4
Data source: International Trade Statistics 2009 of WTO.
The US, UK and Japan have well-developed financial markets and open capital accounts. In particular, the US and Japan are the main investors and trading partners with the GC economies. Singapore, due to its geographical proximity and cultural similarity, has had close economic ties with the GC economy. We consider RIP, UIP and RPPP as the basis of economic globalization because these three parity conditions are popularly used to evaluate the degree of economic globalization in international finance (Cheung et al. 2003). We use realized correlation measure as an alternative to the dynamic conditional correlation (DCC) of Engle (2002), to estimate the time-varying cross-stock market correlation and cross-real estate securities market correlation. Finally, we hypothesize that higher capital market integration is positively linked to the level of economic globalization (as measured by the three parity conditions) for the markets under examination. Finding a positive relationship between (say), RIP and capital market integration within the GC region implies that capital mobility can be enhanced through facilitating and financial and commodity arbitrage that will in turn eliminate the differentials between these economies’ rate of return on investment among the three GC economies. Although there is a wealth of literature on economic globalization and capital market integration5 and that our work is similar in spirit to that of Bracker et al (1999) that links some macroeconomic factors to stock market co-movements, to our knowledge this present work is probably one of very few to explore the joint dynamics of economic globalization and capital market integration within and across the GC context, using advanced time series and econometric methodologies. Specifically, our study is able to contribute significantly in at least three different aspects. First, previous economic globalization studies mainly focused on countries in Europe and other developed countries. Our study using a different dataset is thus an addition to the already large body of literature on economic globalization on an extended period (including era of global financial crisis). Results from this study are expected to enrich the economic globalization literature on RIP, UIP and RPPP within and across the GC areas. Second, instead of relying on parametric procedures (e.g. GARCH) to estimate variance and correlation, we appeal to the realized correlation methodology to estimates ex-post realized correlations across the sample stock and securitized real estate markets. In consistent to Anderson et al. (2003), correlations so constructed are model free and
5
See Section 2 for a brief literature review.
contain little measurement error as the sampling frequency of the returns approaches infinity. These econometric merits motivate our use of this approach which has received less formal attention in the real estate literature. Finally, our panel regression and integration spillover index methodologies link economic globalization (as measured by the three international parity differentials) to capital market integration (as measured by the cross-real estate realized correlations and cross-stock realized correlations) and is consistent with the economic convergence and market integration literature. Our contribution is based on the belief that the more the economies of two countries are globalized, the more interdependent their real estate securities markets and stock markets will be. In addition to the conventional panel investigation, we motivate the gerneralized volatility spillover index methodology of Diebold and Yilmaz (2012) and develop an “integration spillover index” to investigate whether there is increasing connection (from the spillover perspective) between economic globalization and capital market integration in the GC context. Our study can thus be regarded as a good supplement to, as well as provides a new perspective on the economic globalization and market integration literature within and across the GC regions. The plan of the paper is as follows. The next section introduces briefly the international parity relations framework and related literature, as well as explains briefly how the RIP, UIP and RPPP can be applied to assess economic globalization. Section 3 explains the data sample and econometric procedures of the research. This is followed by Section 4 which provides report and discussion on evidence of economic globalization and stock/real estate securities market correlations, as well as the interdependence between economic globalization and capital market integration from panel regression and integration spillover analysis. Section 5 concludes the paper.
2.
International Parity Relations Framework and relevant literature The relevant theoretical framework underpinning the concept of economic globalization in this
study is the international parity relations that provide an intuitive framework explaining the relationship among exchange rates, inflation and interest rates. Following literature, the RIP, UIP and RPPP conditions are adopted to assess the parity relations within and across the GC areas. According to Cheung et al. (2003), the RIP condition depends on whether capital flows equalize real interest rates across economies, UIP involves financial arbitrage between money and foreign exchange markets and RPPP entails arbitrage
in goods and services. Mathematically, the RIP condition is derived by requiring that UIP, RPPP and the ex-ante Fisher equation in both the domestic and foreign currency to hold. In this way, the RIP condition encompasses elements of both real and financial integration. This is further explained below: (a)
Given
and
are the expected j period RI, nominal interest rate and inflation rate in the
first economy respectively and “*” indicates the second economy, then: The real interest parity (RIP), which is the ex-ante RIP differential between two economies, is:
(1)
(b)
Annualized expected depreciation
is defined as:
(2)
Where
is the expected nominal foreign exchange rate in logarithm form between two
countries at time
(c)
Assume further
while
and
is the nominal foreign exchange rate in logarithm at time
are respectively the price in logarithm expected at time t+1 and the
price at t. Annualized expected inflation rate at time t is given by:
(3)
Finally, the term in the first square bracket on the right side of equation (1) is the expected UIP differential while the term in the second square bracket is the expected RPPP differential. Since we do not have observations on market expectations, we will employ an operational version based on ex post differentials6. Hence equation 4 is the ex post international parity relations that implies: RIP = UIP + RPPP:
6
Under the rational expectation hypothesis, the ex post realizations should be unbiased predictors of the ex ante counterparts.
(4)
Empirically, this international parity relations framework and the three separate components (i.e. RIP, UIP and RPPP) have attracted a great deal of attention and has been explored extensively in the international finance literature using recent advances in the field of time series econometrics. Empirical studies include Goodwin and Grennes (1994); Chinn and Frankel (1995); Moosa and Bhatti (1996); Wu and Chen (1998); Fountas and Wu (1999); Holmes and Maghrebi (2004); Cheung, Chinn and Fujii (2003, 2005); Holmes and Wang (2008); Serletis and Gogas (2010) and Cuestas and Harrison (2010). Similarly, with increasing globalization of the world financial markets, there are considerable studies investigating the correlation structure across GC and international stock markets (e.g. Cheng and Glascock, 2005; Caporale et al. 2006; Qiao et al, 2008 and Johansson and Ljungwall, 2009), as well as among international securitized real estate markets (e.g. Liow et al, 2009; Liow, 2012; Liow and Newell, 2012, forthcoming). In these studies, increased time-varying correlations imply higher stock/ real estate stock market integration. Specifically, Liow and Newell (2012, forthcoming) investigate simultaneously the effects of volatility spillover and time-varying conditional correlation on the cross-market relationships between the three GC securitized real estate markets, as well as their international links with the US securitized real estate markets. They find that the conditional correlations between the GC securitized real estate markets have outweighed their conditional correlations with the US market, supporting closer integration between the GC markets due to geographical proximity and closer economic links.
3.
Data and Methodology The research agenda is captured in the form of three research questions. Our first research question
asks to what extent the RIP, UIP and RPP parity conditions will hold in the long run. We use monthly observations on one-month interbank interest rates, exchange rates and consumer price indices on seven economies, namely, China (CN), Hong Kong (HK), Taiwan (TW), Japan (JP), Singapore (SG), the United Kingdom (UK) and the United States (US). The interbank interest rate is regarded as the most flexibly determined interest rate available. The sample period is from February 1996 to June 2011, the longest time series data that is available for each country since China’s one-month interbank rate was only available
only from February 19967. For each of the 15 pairs8, the ex post RI differential ( differential (
) and ex post RPP differential {
), ex post UI ) are constructed. An
assessment of the mean, standard deviation and range of the differential series will allow us make useful preliminary inferences regarding their respective parity characteristics. We then evaluate the parity conditions via two perspectives. First, we test for the presence of zero mean reversion characteristics in these three differential series. Second, we examine whether the deviations from the three parity conditions are shrinking over time or not. For the mean reversion property, we employ a modified Dickey Fuller test known as the ADF-GLS test (Elliot, Rothenburg and Stock, 1986), a panelbased unit-root test provided by Im, Pesaran and Shih (IPS, 1995) 9 and the variance ratio test. The common argument underlying all the tests is that if the deviations from ex post parity are transitory and stationary, then even though the condition does not hold in the short run, deviations from parity are stationary. In contrast, if the deviations from parity are not stationary, there is permanent disequilibrium resulting from shocks and consequently there is no guarantee to restore the parity condition in the long run. For example, if the non-stationary null is rejected for the RI differential series, this implies that there is a tendency for the said series to be mean reverting which is consistent with the implication of RIP. Finally, the variance ratio test has been regarded as more powerful to detect the presence of the stationary component in any random series. For the purpose of this test, the three differential series are decomposed into permanent and transitory components to examine their stationary properties at different frequencies. For the time-trend test, we use a cross-dispersion approach to assess if differential convergence (i.e. decreasing trend) or divergence (i.e. increasing trend) exists over the full period for the RI, UI and RPP
7
A unified national interbank market was only established in January 1996; prior to that the interbank market in the mainland China was substantially controlled (Cheung et al, 2005). 8
This is restricted to three within GC pairs (CN/HK, CN/TW and HK/TW), four CN and international pairs (i.e. CN/US, CN/UK, CN/JP and CN/SG), four HK and international pairs, as well as four TW and international pairs. 9
Panel unit root tests have been widely applied in the empirical literature, especially in the RPPP literature. For example, the Im, Pesaran and Shih (IPS) W-statistic (with trend) tests the null hypothesis of joint non-stationary on a pooled cross-sectional dataset and can provide “dramatic improvement in statistical power” (Im, Pesaran and Shih, 1995)
series. The cross-dispersion is the standard deviation of the various differential series relative to three group average series {overall, within GC, across GC (GC/international)}. The Hodrick-Prescott technique then follows to estimate the long-term trend component of the differential series. For each series, the final level (as at 2011M06) is compared with its initial level (as at 1996M02) and its average time trend over the study period is estimated. For a particular series, a statistically negative time trend imply that the differentials are narrowing (converging) during the sample period and is an indication of increasing economic globalization. Our second research question asks to what extent the real estate securities markets of the three GC economies are correlated, as well between each of the three GC economies and their regional/international partners. We also examine the correlations among the corresponding stock markets. Our data are the daily common stock and real estate stock returns of indices for the seven economies derived from the Broad Market Index (for common stocks) and Global Property (for real estate stocks) sections of the Standard and Poor (S&P) database. Daily stock returns are computed as the natural logarithmic of the total return indices relative, I, in successive days, over February 1996 to June 2011. Next, to obtain a consistent estimate of correlation at monthly frequency, we appeal to the concept of realized moments to compute a monthly estimate of the cross-market correlation10. The construction process is briefly explained here. Define the daily
stock
return
at
country
i
as
Ri ,t ,d ln( I i ,t ,d I i ,t ,d 1 ) x100 and
country
j
as
R j ,t ,d ln( I j ,t ,d I j ,t ,d 1 ) where I are the daily stock (real estate stock) price indices, the realized variances is given by:
Dt
Dt
d 1
d 1
t2,i ( Ri ,t ,d ) 2 and t2, j ( R j ,t ,d ) 2
, where d = (d= 1,….D), D
t
is the
total business day in the month t. Then, the realized covariance between cross-country stock (real estate stock) returns is measured as ij ,t
Dt
( Ri ,t ,d xR j ,t ,d ) . Finally, the realized correlation ij,t measure d 1
between the cross-country returns is obtained as
10
ij ,t
ij ,t i2,t x 2j ,t
. The correlation series are then
This is an alternative approach to the use of parametric models such as the GARCH or the multivariate GARCH models. See Andersen et al. (2003), Kim et al. (2006) and Cappiello et al. (2006).
examined for their descriptive statistics, mean reversion property and trending behavior over the full sample period. Our main objective is to confirm that the sample stock markets and real estate securities markets) have become increasing correlated among themselves over the sample period, as causal observation has suggested. Finally, our third research question asks whether there is a significant positive link between capital market integration (represented by cross-real estate and cross-stock realized correlations, respectively) and economic globalization (as defined by RIP, UIP and RPPP). Given that economic globalization has greatly enhanced the significance and performance of the global real estate securities investment and management over the last two decades (Bardhan et al, 2008), we are motivated to examine scientifically, whether economic globalization and capital market integration (in particular, real estate securities market coerrelation) are positively linked within and across the GC areas, as well as understand the evolution of their interdependence over time. In addition to the usual panel regression approach, we appeal to the return spillover index methodology proposed by Diebold and Yilmaz (2009, 2012), which is based on decomposition of return forecast error variances obtained from a vector auto-regression (VAR)11. In our context, we model the three economic globalization differential series (RI, UI and RPP) and two market integration series (cross-stock and cross-real estate realized correlation: CORR
S
and CORR
RE)
as a five-
factor VAR. Then we conduct a variance decomposition analysis to 36-month long rolling windows of the five variables. For each factor i we add the shares of its return forecast error variance due to shocks come from four other factors j. Then we sum across all i = 1, 2, 3, 4, 5 to obtain the spillover index. In other words, the “integration spillover index”, as we wish to label, is the sum of all non- diagonal in the return forecast error variance matrix. Our integration spillover analysis covers two aspects; (1) an aggregate integration spillover index which measures what proportion of the return forecast error variance comes
11
Diebold and Yilmaz (2012) introduced a generalized VAR methodology and the concept of directional spillovers in volatility transmission research. Their approaches represent significant improvement over the traditional Cholesky-factor identification of VAR. According to Diebold and Yilmaz (2012), the Cholsesky factorization method is able to achieve orthogonality; but the variance decompositions depend on the ordering of the variables. Instead, the generalized VAR framework of Pesaran and Sim (1998) produces variance decompositions which are invariant to the ordering by allowing correlated shocks and using the historically observed distribution of the errors to account for the shocks.
from spillovers; and (2) integration spillover plots which are constructed from the rolling-samples of the spillover indices to assess the extent and nature of the integration spillover variations over time.
4.
Results
4.1
Evidence of economic globalization The three differential series (RI, UI and RPP) for the 15 economy pairs are first adjusted for the
effect of Asian financial crisis (AFC) and Global financial crisis (GFC) and are expressed in annualized percentages. They are graphed in Figures 1-3. In addition, Table 1 reports their mean, standard deviation and range (maximum-minimum) over the full period. Some observations are documented. First, within the GC areas, the summary statistics indicate of the nine average differential series, only three series (two RI and one UI) are statistically significant at least at the 5% level. However, an evaluation of the band of the differential series reveals very large range, spanning 33 (CN/HK, UI) or more percentages points. For the GC/international group, the average differentials are significantly different from zero in 7 (RI), 3(UI) and 2 (RPP) of the 12 comparisons. The significant RI differential averages imply the existence of persistent opportunities for arbitrage activities and may thus provide evidence against financial and real integration between some GC and non-GC economies. (Figures 1-3 and Table 1 here) The ADF-GLS unit root test results (Table 2, Panel A) support stationary in every RI, UI and RPP case, although the degree of support for stationary differentials in two cases of RI (HK/UK and HK/international) and one case of UI (China/US) is not so strong as that offered to other pairs in that the test indicates stationary only at the 10% level of significance. Similarly, the IPS unit root test results (Table 2, Panel B) indicate the null hypothesis that the three differential series are joint non-stationary can be rejected consistently in all groups (GC, China/international, HK/international and TW/international). Additional variance ratio test results (Table 3) indicate at up to 16th lags, the estimated values of the variance ratio are smaller than 1 at the 5% significance level in the majority of the cases, indicating quite strong mean reversion in the data under examination.
(Tables 2 and 3 here) Finally, Figure 5 (Panel A and Panel B) graphs the Hodrick-Prescott filtered cross-dispersion, classified by two broad groups (GC, GC/international). In addition, Table 5 estimates the average annual trend over the full sample period. The results are mixed. They indicate that, on average, the magnitude of the deviation from the RIP condition is declining at between 1.43% and 2.96% per annum. In contrast, the magnitude of the deviation from RPPP is increasing at between 1.31% and 3.29% per annum. (Figure 5 and Table 5 here) Based on the results reported, we are inclined to conclude that that the three parity conditions hold in the long run within and across the GC areas. The unit root and variance ratio null are rejected for all the series/groups and, thus, the deviations from the three parity conditions are stationary. Despite the limited extent of economic globalization and inconclusive time-trend results, the mean reversion results indicate that the RI, UI and RPP in the GC context tend to converge and therefore the three parity conditions tend to hold in the long run. Thus, there is some evidence of economic globalization, in particular, real and financial integration within the three GC economies and across the GC areas regionally and globally. However, there is very little evidence of short-run equilibrium detected in the three differential series which are characterized by the existence of profitable arbitrage opportunities.
4.2
Capital market integration Table 5 provides the cross-market realized correlation estimates of real estate securities market
and stock markets for the 15-economy pairs. Figure 5 graphs the average time series variations over the study period for the two groups (GC, GC/international). As the numbers in Panel A indicates, average cross-real estate market securities correlations range between 0.028 (Taiwan/UK) and 0.361 (Hong Kong/Singapore). Only one pair (HK/Singapore) is above 0.3. The average correlation among the three GC economies is 0.152, which is about, respectively, 43% and 103% higher than the averages between China/international (0.106) and between Taiwan/international (0.075); yet about 36% lower than the average Hong Kong/international correlations (0.207). The corresponding cross-stock realized correlations are consistently higher than their real estate securities counterpart (between 0.135-TW/UK and 0.469-
CN/HK). Except for the TW/UK real estate securities market correlation, all other correlation estimates are statistically significant at the 1% level. Our results are thus in agreement with those reported by Liow et al. (2009) that indicates (significantly) lower cross-real estate securities market correlation (and hence implies lower market integration) than the corresponding cross-stock market correlation in international developed countries. The unit root test (Panels A and B) and variance ratio test (Panel C) results consistently reject the null of non-stationary in every case, and offer strong evidence in support of stationarity in capital market correlation whereby cross-real estate securities correlations and cross-stock market correlations do not persistently diverge from one another in the long run. Finally, the Hodrick-Prescott filtered cross-market correlation dispersion (Figure 6 and Panel D, Table 5) indicates that the sample real estate securities markets have become increasingly correlated, at the rate of between 3.85% and 4.50% per year; the magnitude of increase in stock market correlations within and across the GC areas; however; is on the lower end, ranging between 0.58% and 0.67% per annum. (Table 5, Figure 5 and Figure 6 here)
4.3
Link between economic globalization and capital market integration We formally investigate whether there is a significant link between economic globalization and
capital market integration for our dataset. Based on the usual perception, we expect a negative relationship between the two sets of indicators because higher correlations should be associated with declining RI/UI/RPP differentials (which implies economic globalization). This outcome will in turn imply that economic globalization and capital market integration are positively linked, and is consistent with practical observation. We adopt two formal empirical procedures. First, based on Model (A) and Model (B), Table 6 presents six panel regression results (full sample, GC, GC/international, CN/International, HK/International and TW/international) over the full study period. This means that in each panel, the influence of each explanatory variable is constrained to be identical across all constituent equations while the intercepts are allowed to vary across all equations. The two models are: Model (A):
CORR RE- t = f (RID t /UID t /RPPD t)
Model (B):
CORR RE- t = f (RID t /UID t /RPPD t , CORR ST-t )
In the above, Model (A) tests the relationship between real estate securities market realized correlation (CORR RE- t) and three economic globalization parity differentials (i.e. RID, UID and RPPD). To avoid the influence of multicollinearity only one of the three parity factors is included in the regression at a time. Of the 18 regressions, the results reported in Table 6 confirm the expected significant negative relation between CORR RE- t and three cases of RID, four cases of UID and four cases of RPPD. Moreover, since international correlation of real estate securities and the broader stock market are synchronized with each other (Liow et al, 2009), the inclusion of the stock correlation factor in Model B will serve to control for any stock market effect on the real estate securities markets’ correlations, so that any residual relationship between the detected real estate securities market correlation and economic globalization can be reasonably attributed to the real estate securities markets per se. As the numbers indicate, when the effect of the underlying stock market correlation factor is controlled, there are only four regressions (i.e. two RID, one UID and one RPPD) that display a significantly negative relationship each between real estate securities market correlation and economic globalization The stock market correlation factor is always significantly positively linked to real estate securities market correlation factor (p