Working Paper INVESTMENT DYNAMICS OF GREATER CHINA SECURITIZED REAL ESTATE MARKETS AND INTERNATIONAL LINKAGES Kim Hiang LIOW* Department of Real Estate National University of Singapore 4 Architecture Drive Singapore 117566 Tel: (65)65161152 Fax: (65)67748684 Email : rstlkh@nus.edu.sg and Graeme NEWELL School of Economics and Finance University of West Sydney Email: G.newell@uws.edu.au January 29, 2010 * Corresponding author Abstract This paper focuses on evidence on securitized real estate markets as it investigates the volatility spillovers and correlation dynamics among the three real estate securities markets: China, Hong Kong and Taiwan in Greater China (GC) as well as their international linkages with the US securitized real estate markets over 1995-2009. Overall, results indicate the conditional volatility linkages and correlations among the GC securitized real estate markets have outweighed those relationships with the US, implying closer real estate market integration among the GC markets. Diversification potential across the markets is still good due to lower cross-market volatility interactions and correlation. Finally, there is also evidence supporting a higher level of conditional correlation and volatility spillover index in the post Asian aw well as the global financial crisis periods. INVESTMENT DYNAMICS OF GREATER CHINA SECURITIZED REAL ESTATE MARKETS AND INTERNATIONAL LINKAGES Abstract This paper focuses on evidence on securitized real estate markets as it investigates the volatility spillovers and correlation dynamics among the three real estate securities markets: China, Hong Kong and Taiwan in Greater China (GC) as well as their international linkages with the US securitized real estate markets over 1995-2009. Overall, results indicate the conditional volatility linkages and correlations among the GC securitized real estate markets have outweighed those relationships with the US, implying closer real estate market integration among the GC markets. Diversification potential across the markets is still good due to lower cross-market volatility interactions and correlation. Finally, there is also evidence supporting a higher level of conditional correlation and volatility spillover index in the post Asian aw well as the global financial crisis periods. 1. Introduction This paper analyzes the investment dynamics of Greater China (GC) securitized real estate markets (Mainland China, Hong Kong and Taiwan) and examines their volatility linkages and timevarying correlations with those of the United States (US), the world‘s largest and most transparent securitized real estate market, over the last fifteen years (1995-2009). As economic growth continues in the Mainland China, it is expected that the GC region is growing into an important player in the global financial markets (Johansson and Ljungwall, 2009); with their real estate markets attracting the interest of domestic and international investors. The longest study period adopted in this study is particularly meaningful as it covers the post Asian financial crisis (AFC) and post global financial crisis (GFC) periods and thus permit a comparative assessment of the market dynamics during the two different crisis periods. We first study the time-series properties of real estate securities returns and time-varying volatility using a univariate Threshold Autoregressive GARCH (TGARCH) specification. The short-term linkage is then considered in the context of conditional return and volatility spillovers across the four real estate markets and with the US as well as their time-varying co-movement via a BEKK time-varying covariance model of Engle and Kroner (1995). Following literature, we interpret more volatility spillovers and increased time-varying correlations to be indicative of closer relationship among the four securitized real estate markets. Examining cross-market dynamics could be important from a portfolio diversification perspective; for if the GC securitized real estate markets were linked together, the gains from diversification across these markets would be reduced. Similarly, 1 the US investors will be more interested in the region if there is little or insignificant cross-volatility spillovers or correlations between the US and the real estate markets within the GC region. At the broader level, our research contributes to the ongoing scholarly work in globalization and financial market integration (Bardhan et al. 2008). Our specific contributions are in enhancing international investors with additional knowledge regarding the volatility spillovers and time-varying correlations of the three GC securitized real estate markets and their cross-market relationships with those of the US. Whilst Groenewold et al. (2004) and Cheng and Glascock (2005) assessed the dynamic interrelationships between the GC stock markets, fuller studies involving the dynamics of the GC real estate securities markets and their international linkages have not been carried out. As such, the dynamics of the GC real estate securities markets in an international context is the focus of this paper. The main novelty in the current paper is the application of advanced time series technology, i.e.: multivariate volatility and correlation analysis, volatility spillovers index and correlation time trend analysis to the four securitized real estate markets. We also analyze the effects of the GFC and AFC on our results regarding their impact on the cross-market relationships. Specifically, we evaluate and compare the nature and magnitude of the volatility and correlation linkages as well as of the dynamic responses due to the two crisis events. We find the conditional volatility linkages and correlations among the GC securitized real estate markets have outweighed those relationships with the US, implying closer real estate market integration among the GC markets. Diversification potential across the markets is still good due to lower cross-market volatility interactions and co-movements. Finally, there is also evidence supporting a higher level of conditional correlation and volatility spillover index in the post Asian and global financial crisis periods. The structure of the paper is as follows. Section two highlights the significance of the securitized real estate markets within the GC region. Section 3 provides a concise review of the relevant stock and real estate literature. Sections 4 and 5 describe the data and econometric methodologies used in the present study. Section 6 presents the empirical results with the final section concludes the study. 2 2. Significance of Real Estate Securities in Greater China The significance of real estate in GC has taken on increased importance in recent years, given its strong economic growth and improved real estate market maturity and transparency in recent years (Chin et al 2006; JLL, 2008). This sees 2010 GDP forecasts for China (10.1%), Hong Kong (4.0%) and Taiwan (4.3%) ahead of the global growth forecast (2.6%) (JLL, 2009). These factors have seen GC take on more significance amongst global real estate investors during the GFC, accounting for 13.0% of global commercial real estate transactions in 2008 compared to only 7.6% in 2007 (Newell and Razali, 2009).This sees the major cities in GC of Hong Kong, Shanghai, Beijing and Taipei as becoming major international cities, with significant regional and international economic functions. In particular, the listed real estate securities markets in GC account for 29.0% of global real estate securities at September 2009, comprising Hong Kong (20.3%), China (7.8%) and Taiwan (0.9%) (Macquarie Securities,2009). This sees Hong Kong as the largest real estate securities market globally, China as the 3rd largest and Taiwan as the 16th largest. This high level of securitized real estate in GC is further reflected in the GC stock markets accounting for 11% of global stock markets (WFE, 2009). REIT markets in GC are still in the early stage, with REITs in Hong and Taiwan only established in 2005 and only accounting for 1.9% of REITs globally (Macquarie Securities, 2009) and REITS are yet to be established in China. The stature of the GC real estate securities sector is further reinforced by its strong performance relative to the other major real estate securities markets and global stocks at September 2009 (see Exhibit 1). In particular, a strong recovery from the global financial crisis is clearly evident; significantly ahead of the mature US, UK and global markets. (Exhibit 1 here) 3. Selective Literature Review The extent of price and volatility spillovers and correlation between international stock markets has received much attention in the literature (e.g.: Hamos et al. 1990; Panayiotis et al. 1997). 3 Similarly, much work has been conducted on the stock market integration in the Asian region (e.g.: Chan, Gup and Pan, 1992; Chowdhury, 1994; Daratt and Zhong, 2002; Johnson and Soenen, 2002). In particular, several studies have focused on the GC stock markets. Groenewold, Tang and Wu (2004) find a strong contemporaneous relationship between the Shanghai and Shenzhen stock markets; but the two mainland markets are relatively isolated from the Hong Kong and Taiwan stock markets. Cheng and Glascock (2005) detect weak nonlinear relationships between the three stock markets; however, these markets are not co-integrated with either US or Japan. Their results from the innovation accounting analysis reveal that Hong Kong is the most influential among the three GC markets. After the Asian financial crisis, both Hong Kong and Taiwan have strong contemporaneous relationships with each other. Qiao, Chiang and Wong (2008) analyze the long-run equilibrium, short-term adjustment and spillover effects across Chinese segmented stock markets (A-share and B-share) and Hong Kong stock markets using a bivariate FIVECM-BEKK GARCH model. According to the authors, the FIVECM (fractionally integrated vector error correction model) has two main appealing features. First it helps investors observe long-run equilibrium relationships among co-integrated variables as well as shortterm adjustment. Second, it accounts for the possible long memory in the co-integration residual series. Moreover, combining the FIVECM with a bivariate BEKK-GARCH formulation allows the analysis of the first and second moment spillover effects across the stock markets and the time-varying correlations between the markets simultaneously. Two main findings from their analyses are: (a) the A-share markets are most influential in the spillover effects, and (b) the effects of the Asian crisis on the dynamic return correlations vary across the stock markets. Although Johansson and Ljungwall (2009) find no long-run relationship among the three GC stock markets, there are short-run spillover effects in both returns and volatility in the region from their VAR-MVGARCH model. These results lead them to conclude that there are significant interdependencies among the three markets. 4 In a different context, Fan, Lu and Wong (2009) investigate the dynamic linkages between the China and the US, the UK, Japan and Hong Kong using a Markov-Switching Vector Error Correction Model (MS-VECM) which considers the three regimes of depression, boom and speculation in the market. In the short term, they find that the China stock market has been influenced by the four stock international stock markets and the impacts vary under different regimes. Finally, Tan and Zhang (2009) investigate the spillovers from the US to mainland China and Hong Kong stock markets during the subprime crisis. With the univariate and multivariate GARCH models, they find that China stock market has experienced price and volatility spillovers from the US. Moreover, the price and volatility spillovers from the US are more significant to the HK’s equity returns than China’s returns. The conditional correlation between China and HK is stronger than their conditional correlations with the US, indicating increasing financial integration between China and Hong Kong. In the real estate literature, adequate attention has been given regarding the diversification benefits from the US securitized real estate markets (e.g. Ling and Naranjo, 1999; Worzala and Sirmans, 2003; Cotter and Stevenson, 2006; Michayluk et al. 2006). In contrast, empirical research regarding the dynamics of the China commercial real estate markets is still limited. Newell et al (2005) and Newell et al (2009) demonstrate the risk-adjusted performance, added value and portfolio diversification benefits of China commercial real estate in a pan-Asia portfolio for both listed and direct real estate, while Ke (2008) assesses the performance of state-owned listed real estate companies in China. Most other studies concerning China have not been China-specific, but are largely concerning China real estate securities in a pan-Asia real estate securities portfolio (e.g.: Liow and Adair,2009; Liow and Sim,2006).Similarly, studies regarding listed real estate securities in Taiwan have focused on their role in a pan-Asia portfolio (e.g. :Liow and Adair,2009; Liow and Sim,2006). Most of the previous property research regarding Hong Kong commercial property has focused on Hong Kong property companies, including linkages between the Hong Kong listed and direct property markets (e.g.: Chau et al 2001; He and Webb, 2000; Newell and Chau, 1996; Newell et 5 al 2004; Schwann and Chau, 2003; Tse and Webb, 2000), Hong Kong property company performance (e.g.: Chau et al 2003; Newell et al 2007) and the role of Hong Kong property companies in a panAsia or global property company portfolio (e.g.: Liow, 2007, 2008; Liow and Alastair, 2009; Liow and Sim, 2006; Wilson et al 2006); as well as the dynamics of the Hong Kong direct property market (e.g.: Brown and Chau, 1997; Chau, 1997; Ganesan and Chiang, 1998; Newell et al 1996). 4. Data Data used in this paper are weekly real total return indexes for China, Taiwan and Hong Kong and the US securitized real estate markets over the period January 6, 1995 to December 31, 2009, making a total of 783 observations, the longest data period for which all four real estate indices are available in an international study. Weekly data are preferred because they provide more accurate estimates of the mean and volatility spillovers (Panayiotis et al, 1997) and avoid the problems caused by non-synchronous trading associated with the use of daily data. The data are from S&P/Citygroup Global Property database.1 Weekly real estate index returns are calculated as natural logarithm of the total return index relative, measured in terms of the local dollars as investors in the four markets are assumed to have different hedging abilities. Over this full period, two important events (AFC and GFC) took place respectively in July 1997 and June 2007; this allows us to capture and assess possible changes in the cross-market linkages in the respective post crisis periods. The AFC has reduced real estate returns and increased real estate volatility and correlation with other asset classes (Kallberg et al, 2002).The GFC began in the US with the bursting of sub-prime mortgage market. This rapidly propagated across different asset classes and financial markets. By end October 2008, stock prices in China and Hong Kong had experienced the largest decline of 71 percent and 56 percent, respectively, since their peak in October 2007 (compared with about 30 percent for the major economies) (Tao and Zhang,2009). Consequently, the full study period (783 weeks) is divided into four shorter sample periods and we focus on the AFC (Sub-period 2) and GFC (Sub-period 4). (a) Sub-period 1 (pre AFC): January 1995-June 1997 (130 weeks), 6 (b) Sub-period 2 (post AFC): July 1997-December 1999 (131 weeks) (c) Sub-period 3 (pre GFC): January 2000 to June 2007 (391 weeks) (d) Sub-period 4 (post GFC): July 2007 to December 2009 (131 weeks) Exhibit 2 plots the four securitized real estate indices over the full study period. In addition, Exhibit 3 presents the summary statistics for the weekly returns of the markets. Over the full study period, the average weekly returns for China, Taiwan and Hong Kong and the US are, respectively, 0.192%, 0.154%, -0.069% and 0.176%; and the standard deviations are about 5.764%(China), 4.678%(Hong Kong), 5.007% (Taiwan) and 3.32% (US) respectively. Except for the US real estate securities returns which over-performed the US stock market, all other average returns and risk performances were inferior compared with the respective local stock markets Skewness and excess kurtosis indicate that positive shocks are more common for China and negative shocks are more common for Hong Kong, Taiwan and the USA. In general, the distribution properties of all four real estate return series appear to be non-normal. There is sufficient evidence of linear and non-linear dependence as revealed by the Portmanteau tests for serial correlation for the return and squared return (proxy for volatility) series. Hence there is presence of conditional heteroskedasticity in the returns and volatilities of all four real estate indices in the sample. These descriptive statistics are thus in favor of a model that incorporates GARCH features. The numbers in Exhibit 4 indicate while all four real estate markets derive a positive average return each for the 1st sub-period (0.679%-China; 0.546%-HK; 0.410%-TW and 0.365%-US), the returns were negative for them in the 2nd sub-period, due to the adverse effect of the AFC. This period was also associated with higher investment risk for all the markets. Finally, as expected, the 4th sub-period is characterized by lower returns and much higher standard deviation in all four markets, reflecting the serious effect on the global financial market consequent to the sub-prime turmoil that began in the summer of 2007 in the USA. (Exhibits 2, 3 and 4 here) Exhibit 5 contains the results of the unconditional correlations between the returns and squared returns (a common proxy for risk) of the four securitized real estate markets. Over the full study period, the return correlations among the GC markets (between 0.2938 and 0.3777) are higher 7 than those between the GC markets and the USA (between 0.1152 and 0.2584) indicating closer relationships within the GC region. Nevertheless, these low correlation figures illustrate that opportunities for diversification in the GC securitized real estate markets are many for both domestic and international investors, at least before July 2007. In accordance with the literature, the 2nd subperiod (AFC period) and the 4th sub-period (GFC period) experienced significant increases in many correlation series, with average correlation reach as high as 0.6526 for the CH-HK pair in the post AFC period. This indicates that the crisis might have exerted a statistically significant and positive impact on the unconditional correlation with a large structural break in cross-market relationship. We will explore this issue again with the conditional correlation series generated from the BEKK model. Finally, results for the squared return correlation are similar and weaker, with the strongest correlation happened for the six market pairs in the post GFC period. (Exhibit 5 here) 5. Methodology This study uses four different methodologies. First, a univariate TGARCH-in mean model is investigated for the time-series behavior of real estate securities return and volatility. Second, we model simultaneously the volatility spillovers and time-varying correlations among the four real estate markets using a asymmetric VAR-BEKK-MGARCH-in mean model. Lastly, we measure volatility spillovers and time trend in correlations using, respectively, a volatility spillover index and two time trend tests. The empirical methodologies are briefly explained below. (a) Univariate AR (1)-TGARCH (1, 1)-in mean model Following Chiang and Doong (2001), we estimate an AR (1)-TGARCH (1,1)-in mean model to investigate the relationship between real estate securities returns and conditional volatility as well as whether there is any asymmetric effect on each real estate market’s conditional variance. In the conditional mean equation presented below, the real estate securities market return, R jt , is linearly 8 1/ 2 related to its first order autoregressive components and its conditional standard deviation h jt and the innovation jt is conditional on the information set t 1 with zero mean and variance h jt . Any significant relationship between volatilities and real estate returns is captured by the estimated coefficient , with a significant and positive value implies that investors are compensated by higher returns for bearing a higher level of risk. In the conditional variance equation, the conditional variance is a linear function of four components: the previous variance ( ht 1 ), the square of the lagged shock term ( t 1 ), the asymmetric indicator variable I t 1 that takes a value of 1 when the previous shock is 2 negative and zero otherwise, and the GFC dummy ( (D ) that takes a value of 1 for subperiod 4 and zero otherwise. Note that the asymmetric effect is captured by the hypothesis 0 ; a positive implies that a negative innovation increases conditional volatility. AR(1) TGARCH (1, 1) -in mean model R jt 0 1 R j ,t 1 h1j / 2 jt jt t 1 ~ N (0, h jt ) h jt 0 1 h j ,t 1 ( 2 I t 1 ) 2j ,t 1 (b) An asymmetric VAR (1)-BEKK-MGARCH (1, 1) – in mean model Following stock market literature, we consider a multivariate asymmetric framework and use the BEKK parameterization (Engle and Kroner, 1995) to model the conditional variance-covariance matrix. The main advantage is that within this framework, it allows for a dynamic structure of conditional correlation and empirical lead-lag relationship in the mean and volatility transmission in a cross-market setting to be simultaneously estimated and captures potential asymmetries in the volatility spillovers mechanism. In addition, the BEKK imposes restrictions on across and within equations as well as guarantees a positive variance-covariance matrix. Hence it is preferred to alternative multivariate models such as the VECH specification.2 9 A BEKK model is specified by the following equations: Ri ,t 0 1 Ri ,t 1 2 R j ,t 1 i ( H i ,t ) i ,t (1a) R j ,t 0 1 R j ,t 1 2 Ri ,t 1 2 ( H j ,t ) j ,t (1b) t t 1 N (0, H t ) (2) H t C ' C A' t 1 t 1 A B ' H t 1 B G ' t 1 t 1G (3) Equations (1a) and (1b) show the first order moments expressed according to a bivariate VAR-in mean model, where t ( i ,t , j ,t ) is assumed to follow a normal bivariate distribution with mean 0 and variance H t and t 1 is the information set in t-1. Return linkages between a pair of markets are ascertained by the VAR parameters. The conditional matrix of variance-covariance is given by H t , and C, A, B, G are N x N parameters with C upper triangular (Equation 3). Volatility spillovers effects are examined from the GARCH estimates ( a ij , bij ).The asymmetrical response of the volatility to news of different signs is given by Min(0, t ) (Kroner and Ng, 1998). Hence this specification allows us study volatility transmissions between the two markets as well as the signs of the shocks, when the innovations are either positive or negative. Further examination using the Akaike and Bayesian information criteria suggests a BEKK-GARCH (1, 1) specification is sufficient for the conditional variance. The parameters of the bivariate-BEKK system are estimated by maximizing the conditional log-likelihood function of: ' TN 1 T L( ) ln(2 ) (ln H t ( ) H t1 ( ) t ) 2 2 t 1 t 10 Where T is the number of observations, N is the number of variables in the system and is the vector of all the parameters to be estimated. The estimation is carried out using the quasi maximumlikelihood estimation with the optimization algorithm of BFGS. (c) Volatility spillover index We focus on variance decomposition which permits the aggregation of volatility spillovers effect across the four securitized real estate markets and distills the information into a single volatility spillover measure. Under the variance decomposition methodology, the decomposition of forecast error variance determines explicitly how much of the conditional volatility movement in one market can be explained by other markets in term of the percentage of the forecast variance of that market. Following the procedures developed by Diebold and Yilmaz (2007), a volatility spillover index each is constructed for the full period as well as for the AFC and GFC periods, thereby allowing the measurement and comparison of the magnitude of volatility interactions among the four securitized real estate markets (d) Time trend in time-varying correlation From the BEKK estimation, we assess whether correlations display trending behavior by first studying their autocorrelation structure and ADF unit root results. We then use Vogelsang’s (1998) simple linear time trend test with the basic model specified as: Correlationt * TREND i We use the “t-PS” test in Vogelsang (1998) to test if =0. Additionally, we also conduct the “t-dan” test developed by Bunzel and Vogelsang (2005) to check the consistency of the time-trend results.3 6. Results 11 First, the results of estimating the TAR-GARCH (1, 1)–in mean model of the four securitized real estate markets are reported in Exhibit 6. We first consider the mean equation. No evidence of positive serial correlation in the four weekly return series is detected. In addition, we are unable to find any statistical significance by including a conditional standard deviation term in the weekly series. With respect to the estimates of the variance equation, all the GARCH parameters are statistically significant. Moreover, the estimated results show that the sum of estimated equations in the variance equation is all more than 0.9; and is close to unity (0.9983) for China, implying the volatility shocks may persist over a longer period of time. The hypothesis of no asymmetric effect is strongly rejected for the US and Hong Kong weekly series at least at the five percent level of significance. The GFC dummy estimates for all four markets are positive, and the estimated coefficient is statistically significant for the US and Hong Kong markets, indicating that the AFC has created a significant regime shift in volatilities. Finally, we examine the Ljung-Box (LB) statistics for diagnostic checking. Results indicate the four estimated TGARCH models are generally well-specified with the removal of serial dependence in the return and squared return series. (Exhibit 6 here) Second, Exhibit 7 reports the results of the multivariate return and volatility relationships among the four securitized real estate markets using a asymmetric VAR-BEKK-MGARCH- M model.4 The impact of a market’s own effects is represented by 11, 22, 33 and 44 for the market US, China, Hong Kong and Taiwan. Cross-market effects are given by 21 and 12, 34 and 43…..etc. In the conditional variance equation, the A coefficients represent ARCH effects, while the B coefficients are GARCH effects, and the G coefficients indicate asymmetric effects. The VAR examines the direction and magnitude of the return linkages. There are few instances of significant cross-return transmission among the four markets; although the results are in general less satisfactorily. Focusing on the volatility transmission, there is some strong evidence of volatility spillovers between all markets considered. With respect to volatility innovations, the “own “spillovers coefficients are significant in three cases. In addition, there are five significant crossmarket spillovers; between China and Taiwan (coefficient A24), Hong Kong and China (coefficient 12 A32), Taiwan and US (coefficient A41), Taiwan and China (coefficient A42), Taiwan and Hong Kong (coefficient A43), Hence we are able to verify that a bidirectional cross-market spillover transmission only exists between the China and Taiwan markets. With regard to the effects of variance (GARCH), all four “own” spillover coefficients are significant and of higher magnitude, results which are reasonably expected. The schema of volatility transmission when the lagged variances are analyzed is bidirectional between China and Taiwan (B24 and B42) and between Hong Kong and Taiwan (B34 and B43), with unidirectional from the US to China (B12), US to Hong Kong (B13) and from China to Hong Kong (B23). The results of the G asymmetries matrix indicate that the negative innovations of the markets increase significantly the volatility in the same market in 3 cases (coefficients G11, G22 and G33). In addition, the volatility transmission mechanism is asymmetric. Negative innovations in the first market increase volatility in the second market considerably more than positive innovations do. Nevertheless, this transmission is only unidirectional, from the US to Hong Kong (G13), US to Taiwan (G14), China and Taiwan (G24) and Hong Kong and Taiwan (G34). Based on this statistical evidence, significant and reciprocal lead/lag linkages emerge among the three GC securitized real estate markets. There is also some weaker unidirectional volatility interactions exist between the US and GC markets. The conditional volatility relationships among the GC securitized real estate markets have outweighed their volatility linkages with the US, implying closer real estate market integration among the GC markets. (Exhibit 7 here) To compare the impact of the AFC and GFC on the volatility spillover dynamics, we estimate a volatility spillover index to measure the magnitude of volatility spillovers across the four securitized real estate markets. The conditional volatility spillover tables for the full study period, the post AFC and the post GFC periods are reported in Exhibit 8. The volatility spillover index is shown in the lower right corner of each table. The spillover table provides an “input-output” decomposition of the spillover index forecasted 12 weeks ahead. Over, the spillover indices are 0.467 (full period), 0.490 (post AFC period) and 0.635 (post GFC period). Therefore, the volatility spillover index registers its highest in the post GFC period, a result which is consistent with expectation. Focusing on the post 13 GFC period (Panel C), we find the US market is the strongest in term of volatility transmission to other markets (173.55%). This is not surprising as, during this period, the subprime turmoil that began in the summer of 2007 subsequently boiled over and set off dramatic changes not only in the US financial markets, but also to the global markets. Consequently, none of the GC securitized real estate markets is immune to the financial crisis, as evidenced by the volatility spillovers from the USA. Finally, Hong Kong ranks 2nd second in transmitting volatility to other markets (40.6%) due to its developed real estate market status. (Exhibit 8 here) The third section examines the implied time-varying correlations derived from the multivariate asymmetric BEKK model. Descriptive statistics (mean, standard deviation, maximum and minimum) for the weekly conditional correlations and volatilities series are provided in Exhibit 9. Over the full study period, the conditional correlation among the three GC markets ranges between 0.1878 (CH/TW) and 0.3519 (CH/HK); while their conditional correlations with the US markets are lower, between 0.0498 (US/TW) and 0.2079 (US/HK). The finding that the conditional correlations between the GC markets have outweighed their conditional correlations with the Us market present little surprise and support closer integration between the GC markets. Among the three GC markets, US has the highest correlations with HK (0.2079) than with China (0.1238) and Taiwan (0.0498), reflecting HK’s role as an international financial centre. Overall, the low correlation coefficients of the GC markets with the US provide some diversification motivation to attract the US investors to invest in the GC securitized real estate markets. Comparing with the pre-AFC period, the post AFC experiences an increase (and in some cases, a significant increase) in correlation in all markets, with the highest correlation reported for the CH/HK pair. During this period, the correlations are also less stable reflecting turbulent investment climates. In accordance with the finance literature, this increase in the mean and standard deviation of the market correlation indicates a “contagion” effect due to the crisis. Focusing on the GFC, there appears to be a significant shift in the conditional correlation, as a higher level of correlation for several market-pairs is observed in the post GFC period compared with the earlier sample period. Comparing with the pre-GFC period, the correlations among the three GC 14 markets report an increase of about 49% (CH/HK), 88% (CH./HK) and 165%(HK/TW) respectively; the increase in the correlation between the US and GC markets is 3% (US/HK), 100%(US/CH) and 131% (US/TW) respectively, highlighting that the securitized real estate markets are vulnerable to external shocks with a large and positive shift in correlation. Together with significant increases in the conditional volatilities (between 41% and 167%) of the four markets, any possible diversification gains will quickly disappear in this volatile market environment. (Exhibit 9 here) To formally test the impact of the AFC and GFC events on dynamic correlations, we regress the time-varying correlation (from the BEKK estimation) with its one-period lag and to crisis dummy variables, dummy one for Jul97to Dec 99 (131 weeks) for the AFC and for Jul07 to Dec09 (131 weeks) for the GFC, and zero otherwise. Due to the collinearity problem, the two dummies are included one at a time, and the results are reported in Exhibit 10. It is clear with the exception for the US/HK series, the GFC has exerted a statistically highly significant and positive impact on other five correlation series implying there are significant structural break happened in the dynamic correlation. Thus the five correlation series are on average significantly higher than the normal level. Results for the AFC are weaker with only three cases of significant regime shift in correlation detected within the GC markets. These results are in general consistent with the understanding that the AFC event was mainly regional and did not impact significantly on the dynamic correlations between the US and GC securitized real estate markets. Overall, these dummy regression results are in agreement with the average correlation results reported in Exhibit 9. (Exhibit 10 here) Exhibit 11 plots six graphs regarding the evolution of the conditional correlations among the three GC markets and with the US markets. Three observable trends are noted. First, the conditional correlations deviate substantially from the point estimates of the unconditional correlations and in most cases the unconditional correlations are different from the conditional correlations as estimated by the time-varying covariance BEKK model. Second, there is generally a similar pattern of correlation observed around 1997-98 with the occurrence of the AFC, where the correlations tend to 15 be significantly higher for China, Hong Kong and Taiwan; however, the AFC has exerted much lower impact on the correlations between the US and GC markets. A similar pattern of correlation is observed around 2007-2008 with the occurrence of the GFC and is accompanied by significantly higher level of correlations, and in some cases, the correlations tends to peak. This is consistent with the finance literature that documents international correlation increases when global factors dominate domestic ones and affect all asset markets. Finally, other than periods of AFC and GFC, the pattern of correlation seems to diverge among the economies due to some market-specific movements. (Exhibit 11 here) Finally, results for detecting a deterministic linear time trend for all six correlation series appears in Exhibit 12. 5 Based on the t-dan tests, we detect one case of a significant time trend coefficient (0.00014, significant at the 5% level) over the full study period. Specifically, the US/China correlation, which is stationary, has increased significantly by a total of about 10.96 percent over the full study period, implying increasing integration between the two securitized real estate markets. We do not detect either significant upward or downward time trends for other correlation series. These findings imply the process of globalization over the past 15 years has not resulted in any significant increase in the co-movements over time among the GC real estate markets and with the US market. Accordingly, these real estate markets have not become more integrated among themselves, implying diversification potential across the markets is still good. Of course, it is possible that significant deterministic trends are not detected because we have disregarded a structural break in the correlations reported in Exhibit 10. Additional estimations indicate while there is no significant time trend coefficients detected in the post AFC period, there are two cases of a significant coefficient estimates for the CH/TW and HK/TW correlation series in the post GFC period, translating to an increase of 18.08 percent (CH/TW) and 30.52 percent (HK/TW) in dynamic correlation respectively. (Exhibit 12 here) 7. Conclusion 16 Although the dynamics of volatility spillovers and time-varying correlation in international securitized real estate markets have received greater attention in the recent years, fuller studies involving the dynamics of the GC securitized real estate markets and their international linkages have not been carried out. These three markets have drawn great attention from investors all over the world and it is likely they will form an even a stronger merged market in the future. In the full sample period from beginning July 1995 and covers a 15-year history of GC securitized real estate markets up to end December 2009, the main objective of this study was to provide a comprehensive study of their investment dynamics and how the covariance and correlation structure have developed among them and with the USA in the context of globalization and two major crisis events: AFC and GFC. We estimate an asymmetric time-varying covariance multivariate BEKK model that takes into account the asymmetric volatility phenomenon to simultaneously capture the lead-lag volatility spillovers and time-varying correlation among the four markets. Using a volatility spillover index, we measure and compare the magnitude of volatility spillovers across the four securitized real estate markets during the post-AFC and post GFC periods. We also investigate whether the real estate securities markets have become more closely linked by assessing whether the respective correlation series display trending behavior. This in-depth analysis and understanding regarding the nature and degree of volatility transmission and correlation relationship as well as their potential changes triggered by two major crises is important for international investors in their portfolio management and risk diversification effort that include real estate investing in the Greater China region. Similarly, policy makers may find the time-varying market interdependence results useful in their crisis management using macroeconomic tools. Subject to the usual empirical limitations, this study produces several valuable and significant findings, notably: (a) significant and reciprocal lead/lag linkages emerge among the three GC securitized real estate markets exist, together with some weaker unidirectional volatility interactions exist between the US and GC markets; (b) During the GFC period, the US market is the strongest in term of volatility transmission to other markets and is followed by Hong Kong. None of the GC securitized real estate markets is immune to the financial crisis, (c) A higher level of correlation is 17 observed in the post-GFC and post AFC periods, with a large regime shift in some correlation pairs; (d) The conditional volatility interactions and dynamic correlations among the GC securitized real estate markets have outweighed their equivalent relationships with the US, implying closer real estate market integration among China, Hong Kong and Taiwan; and (e) Globalization and financial integration over the past 15 years has not resulted in any significant increase in the co-movements over time among the GC securitized real estate markets and between them and the US markets, implying diversification potential across the markets is still good. Some of the above findings motivate further research. For instance, it would be useful to modify the BEKK specification (although a difficult task) to focus on the fundamental determinants of volatility spillovers and time-varying covariance in the context of macro-economy. Little or no direct evidence is available on the cross-market volatility and correlation dynamics between securitized real estate markets and business cycles. Expanding this body of evidence in the GC securitized real estate markets is a fruitful area for future research. References Antonios, A. and Pescetto, G.M. (2007) Market-wide and sectoral integration – evidence from the UK, USA and Europe Managerial Finance 33(3): 173-193 Arago-Manzana, V. and Fernandez-Izquierdo, M.A. (2007) Influence of structural changes in transmission of information between stock markets: A European empirical study Journal of Multinational Financial Management 17: 112-124 Bardhan , A., Edelstein, R. and Tsang D. (2008), Global Financial Integration and real estate security returns, Real Estate Economics 36(2), 285-311 Bhar, R. and Nikolova, B. (2009) Return, volatility spillovers and dynamic correlation in the BRIC equity markets: an analysis using a bivariate EGARCH framework Global Finance Journal 19: 203218 Brown, G. and K.W. Chau (1997) Excess returns in the Hong Kong commercial real estate market, Journal of Real Estate Research, 14, 91-105. Bunzel, H. and Vogelsang, T.J. (2005), Powerful trend function tests that are robust to strong serial correlation, with an application to the Prebisch-Singer hypothesis, Journal of Business and Economic Statistics 23, 381-394 Caporale, G.M., Pettis, N. and Spagando, N. (2006) Volatility transmission and financial crisis Journal of Economics and Finance 30(3): 376-390 18 Chan, K.C., Gup, B. and Pan, M. (1992) An empirical analysis of stock prices in major Asian markets and the United States, Financial Review, 27, 289-307 Chau, K.W. (1997) Political uncertainty and the real estate risk premiums in Hong Kong, Journal of Real Estate Research, 13, 297-316. Chau, K.W., B. MacGregor and G. Schwann (2001) Price discovery in the Hong Kong real estate market, Journal of Property Research, 18, 187-216. Chau, K.W., S.K. Wong and G. Newell (2003) Performance of property companies in Hong Kong: a style analysis approach, Journal of Real Estate Portfolio Management, 9, 29-44. Cheng, H. and Glascock, J. (2005), Dynamic linkages between the Greater China Economic Area stock markets – Mainland China, Hong Kong and Taiwan, Review of Quantitative Finance and Accounting 24: 343-357 Chin, W., P. Dent and C. Roberts (2006) An exploratory analysis of barriers to investment and market maturity in Southeast Asian cities, Journal of Real Estate Portfolio Management, 12, 49-58. Chou, Ray Y., Lin, Jin-Lung and Wu, Chung-Shu (1999), Modeling the Taiwan stock market and international linkages, Pacific Economic Review 4(3): 305-319 Chowdhury, A.R. (1994) Stock market interdependencies: evidence from the Asian NIEs, Journal of Maxroeconomics 16, 629-651 Cotter, J. and Stevenson, S. (2006) Multivariate modelling of daily REIT volatility, Journal of Real Estate Finance and Economics 32: 305-325 Darrat, A. and Zhong M. (2002) Permanent and transitory driving forces in the Asia-Pacific stock markets, Financial Review 37, 35-52 Engle, R.F. and Kroner, K.F. (1995), Multivariate simultaneous generalized ARCH, Economic Theory 11, 122-150 Fan, K. Lu, Z. and Wang S. (2009), Dynamic linkages between the China and International stock markets, Asia-Pacific Financial Markets DOI 10.1007/s10690-009-9093-5 Ganesan, S. and Chiang, Y. H. (1998) The inflation-hedging characteristics of real and financial assets in Hong Kong, Journal of Real Estate Portfolio Management, 4, 55-67. Groenewold, Nicolaas, Tang, Sam H.K. and Wu, Yanru (2004), The dynamic interrelationships between the greater China share markets, China Economic Review 15: 45-62 Hamos, Y., Masulis, R.W. and Ng, V. (1990) Correlation in price changes across international stock markets, Review of Financial Studies, 281-307 He, L.T. and J. Webb (2000) Causality in real estate markets: the case of Hong Kong, Journal of Real Estate Portfolio Management, 6, 259-271. Johansson, Anders C. and Lijunwall, Christer (2009), Spillover effects among the Greater China stock markets, World Development 37(4): 839-851 19 Johnson, R. and Soenen, L. (2002) Asian economic integration and stock market co-movement, Journal of Financial Research 25, 141-157 Jones Lang LaSalle (2008) Real Estate Transparency Index, JLL. Jones Lang LaSalle (2009) Asia Pacific Property Digest Q3:2009, JLL. Kallburg, J.G., Liu, C.H. and Pasquariello, P. (2002), Regime shifts in Asian equity and real estate markets, Real Estate Economics 30,2, 263-291 Ke, Q. (2008) Are state-owned companies underperforming? A case study of Chinese listed property companies, Journal of Real Estate Literature, 16,2, 183-200 Kroner, K.F. and Ng, V.K. (1988), Modeling asymmetric co-movements of asset returns, Review of Financial Studies 11, 817-844 Ling, D.C. and Naranjo, A. (1999), “The integration of commercial real estate markets and stock markets”, Real Estate Economics 27(3): 483-515 Liow, K.H. (2007) The dynamics of return volatility and systematic risk in international real estate security markets, Journal of Property Research, 24, 1-29. Liow, K.H. (2008) Financial crisis and Asian real estate securities market interdependence: some additional evidence, Journal of Property Research, 25, 127-155. Liow, K.H. and A. Adair (2009) Do Asian real estate companies add value to investment portfolios, Journal of Property Investment and Finance, 27, 42-64. Liow, K.H. and M.C. Sim (2006) The risk and return profile of Asian real estate stocks, Pacific Rim Property Research Journal, 12, 283-310. Macquarie Securities (2009) Global Property Securities Analytics: June 2009, Macquarie Securities Michayluk, D., Wilson, P. and Zurbruegg, R. (2006), “Asymmetric volatility, correlation and return dynamics between the US and UK securitized real estate markets”, Real Estate Economics 34(1): 109131 Newell, G. and K.W. Chau (1996) Linkages between direct and indirect property performance in Hong Kong, Journal of Property Finance, 7, 9-29 Newell, G. and M.N. Razali (2009) The impact of the global financial crisis on commercial property investments in Asia, Pacific Rim Property Research Journal, 15, 430-452 Newell, G., Chau, K.W. and F. Pretorius (1996) Adjusting the volatility of the Hong Kong property market, Journal of Real Estate and Construction, 6, 1-16. Newell, G., K.W. Chau and S.K. Wong (2004) The level of direct property in Hong Kong property company performance, Journal of Property Investment and Finance, 22, 512-532. Newell, G., K.W. Chau, S.K. Wong and K. McKinnell (2007) Factors influencing the performance of Hong Kong real estate companies, Journal of Real Estate Portfolio Management, 13, 75-86. 20 Newell, G., K.W. Chau and S.K. Wong (2009) The significance of Chinese commercial property in an Asian property portfolio, Journal of Property Investment and Finance, 27,2, 102-119 Panayiotis, T., Kahya, E., Koutmos, G. and Christofi, A. (1997) Volatility reversion and correlation structure of returns in major international stock markets, The Financial Review 32(2), 205-224 Qiao, Z., Chiang, Thomas C. and Wong W. (2008) Long-run equilibrium, short-term adjustment, and spillover effects across Chinese segmented stock markets and the Hong Kong stock market, Int. Fin. Markets., Inst. & Money 18: 425-437 Schwann, G. and K.W. Chau (2003) News effects and structural shifts in price discovery in Hong Kong, Journal of Real Estate Finance and Economics, 27, 257-271 Serrano, C. and Hoesli, M. (2009), Global securitized real estate benchmarks and performance, Journal of Real Estate Portfolio Management 15(1): 1-19 Skintzi, V.D. and Refenes, A.N. (2006), Volatility spillovers and dynamic correlation in European bond markets, Int. Fin. Markets, Inst. and Money 16:23-40 Standards & Poor’s (2009) Global Property and REIT Report: 3rd Quarter 2009, S&P. Tao, S. and Zhang, X. (2009), Spillovers of the US subprime financial turmoil to Mainland China and Hong Kong SAR: evidence from the stock markets, IMF working paper: monetary and capital market, Chinese Academy of Social Sciences Tse, R. and J. Webb (2000) Public versus private real estate in Hong Kong, Journal of Real Estate Portfolio Management, 6, 53-60. Vogelsang, T.J. (1998), Trend function hypothesis testing in the presence of serial correlation, Econometrica 66, 123-148 World Federation of Exchanges (2009) Focus: July 2009, WFE. Worzala, E. and Sirmans, C.F. (2003), “Investing in international real estate stocks: a review of the literature”, Urban Studies 40(5/6): 1115-1149 Zhu, H., Lu, Z. and Wang, S. (2004), Causal linkages among Shanghai, Shenzhen, and Hong Kong stock markets, International Journal if Theoretical and Applied Finance 7(2): 135-149 21 Exhibit 1 Performance of Greater China real estate securities markets: Sept 2009 Average annual returns 3Y 5Y 1Y China 10Y 101.1% 20.7% 35.0% 25.0% Hong Kong 41.5% 9.8% 13.6% 9.9% Taiwan 76.2% 9.0% 11.3% 1.1% US -28.3% -13.1% 1.0% 9.6% UK -35.8% -28.0% -7.0% 3.1% Global -4.8% -8.8% 4.8% 9.4% Global stocks 1.5% -2.4% 5.6% 3.3% Source: S&P (2009) Exhibit 2 Real estate securities total return index movement 800 700 600 500 400 300 200 100 0 95 96 97 98 99 00 01 China Taiwan 02 03 04 05 06 07 08 09 Hong Kong US Source: S& P Property 22 Exhibit 3 Sample statistics of weekly securitized real estate returns: Jan95-Dec09 Statistic Mean (%) Std deviation (%) Maximum (%) Minimum (%) CV (sd/mean) Skewness Ex Kurtosis Jarque-Bara Arch(10) LB(10) LB(50) LB2(10) LB2(50) China 0.192 5.764 28.775 -24.731 30.021 0.0369 2.213*** 159.97*** 5.477*** 18.367** 57.669 65.029*** 207.720*** Hong Kong 0.154 4.678 19.855 -27.107 30.377 -0.3536*** 3.8828*** 508.19*** 4.567*** 10.083 59.911 59.061*** 241.881*** Taiwan -0.069 5.007 17.101 -24.206 -72.565 -0.1865** 2.1593*** 156.66*** 4.737*** 20.729** 45.936 64.874*** 112.509*** US 0.176 3.325 21.527 -21.056 18.892 -0.4234*** 11.970*** 4698.1*** 53.992*** 17.705* 123.005*** 957.545*** 1837.05*** Notes: The respective stock markets returns are: China (0.206%), Hong Kong (0.167%), Taiwan (0.027%), US (0.154%); the standard deviations of the stock market returns are: China (4.699%), Hong Kong (3.544%), Taiwan (3.688%) and US (2.605%). ***, **, * - indicates significant at the 1%, 5% and 10% levels. Exhibit 4 Statistic Risk and return of weekly securitized real estate returns in different periods China Hong Kong Taiwan Panel A: Pre-Asian financial crisis (Jan95-June97: 130 weekly returns) Average return (%) 0.679 0.546 0.41 Std deviation (%) 5.246 3.281 3.782 Panel B: Post Asian financial crisis (July 97 - Dec 99: 131 weekly returns) Average return (%) -1.071 -0.216 -1.139 Std deviation (%) 6.597 6.528 5.014 Panel C: Pre Global financial crisis (Jan00-Jun07: 391 weekly returns) Average return (%) 0.586 0.155 0.078 Std deviation (%) 4.848 3.665 5.009 Panel A: Post global financial crisis (July 07 - Dec 09: 131 weekly returns) Average return (%) -0.200 0.137 0.087 Std deviation (%) 7.486 6.146 5.893 US 0.365 1.026 -0.086 2.018 0.352 2.069 -0.275 6.939 23 Exhibit 5 Unconditional correlations in returns and volatilities (squared returns) among the securitized real estate markets Full period Jan95-Dec09 783 weeks CH and HK CH and TW HK and TW US and CH US and HK US and TW 0.3777*** 0.2145*** 0.2938*** 0.1687*** 0.2584*** 0.1152*** CH and HK CH and TW HK and TW US and CH US and HK US and TW 0.3087*** 0.1131** 0.1128** 0.0604* 0.1866** 0.0651* Subperiod 1 Subperiod 2 Jan95-Jun97 Jul97-Dec99 130 weeks 131 weeks Panel A: return 0.1462* 0.3599*** 0.0414 0.2429*** 0.0083 0.2804*** 0.0128 0.0591 0.3163*** 0.1585*** 0.0976 -0.0151 Panel B: squared returns 0.0199 0.3380*** -0.0789 0.0222 0.0088 -0.2001 0.0863 0.0849 0.2629*** 0.1017 0.0166 0.0194 Subperiod 3 Jan00-Jun07 391 weeks Subperiod 4 Jul07-Dec09 131 weeks 0.2358*** 0.1066** 0.2097*** 0.1690*** 0.2291*** 0.0478 0.6526*** 0.4292*** 0.5520*** 0.2445*** 0.3667*** 0.2310*** 0.0784 0.0344 0.1260** 0.0311 0.1178** -0.0378 0.5535*** 0.3520*** 0.3729*** -0.0105 0.2976*** 0.0817 Notes: China (CH), Hong Kong (HK), Taiwan (TW) and the US. ***, **, * - indicates statistical significance at the 1%, 5% and 10% level Exhibit 6 Estimates of univariate AR (1)-TARCH (1,1)-M model of weekly securitized real estate returns with the impact of Global financial crisis (GFC): January 1995 – December 2009 Coefficient China Constant AR(1) In-mean 0.0027 0.0179 -0.4363 Constant ARCH(1) GARCH(1) Leverage effect Dummy (GFC) Log-likehood 0.000033 0.0926*** 0.9057*** -0.0035 0.000034 1207.41 Hong Kong Taiwan Conditional mean euqation 0.0038* -0.0019 -0.0023 0.0318 -0.7788 -0.0738 Condtional variance equation 0.000023* 0.00025** 0.0159 0.2043*** 0.9318*** 0.7361*** 0.0694** -0.0682 0.00011** 0.00027 1385.65 1300.34 US 0.0028 0.0307 -0.1631 0.000008*** 0.0398** 0.9086*** 0.0667*** 0.000076** 1887.37 Notes: (1) The Ljung Box (LB) statistics are computed for returns and squared returns to conduct diagnostic checking. Results indicate that the four univariate T-GARCH-M models are generally well specified with the removal of serial dependence in the square and square returns; (2)another four TGARCH-M models are estimated with Dummy (AFC-Asian financial crisis). Except for the HK market, the other three markets’ AFC dummy estimates are statistically insignificant;(3) ***, **, * - indicate statistical significance at the 1%, 5% and 10% level 24 Exhibit 7 Variable Results of the Asymmetric VAR-BEKK-MGARCH-M model: January 1995 – December 2009 (full period) US (i =1) China (i=2) HK (i=3) Taiwan (i=4) i ,1 0.0111 Panel A: Conditional Mean 0.1183** 0.0844* 0.0722 i,2 -0.0125 -0.0057 -0.0462* -0.0563** i ,3 -0.0046 0.0194 0.0014 0.0794** i,4 0.0233* -0.0178 0.0249 0.0210 i -1.8587** -4.3411** -1.6996* -3.7781 Ai ,1 0.2905*** Panel B: Conditional Variance -0.0065 -0.0218 0.0383** Ai , 2 0.0190 -0.0546 0.2579*** -0.1079*** Ai ,3 -0.0262 -0.0015 0.1616*** 0.0650** Ai , 4 -0.0023 -0.1437*** 0.0404 0.1802*** Bi ,1 0.8950 *** 0.0146 -0.0187 0.0256 Bi , 2 0.1717** -0.3531*** -0.0303 1.0943*** Bi ,3 -0.0667*** 0.2076*** 0.6853*** 0.2923*** Bi , 4 0.0450 -0.7418*** -0.0801*** 0.9506*** Gi ,1 0.4780*** -0.0252 -0.0932 -0.0192 Gi , 2 -0.0889 0.2092*** 0.0353 0.0573 Gi , 3 0.0063*** 0.1059 0.2874* -0.0068 Gi , 4 0.1307*** 0.0916* 0.1404* -0.0185 Log likelihood Skewness Excess Kurtosis ARCH (10) LB(10) LB(50) LB2(10) LB2(50) 5818.12 Panel C: Some Residual diagnostic -0.047 0.264*** -0.0381 0.995*** 1.096*** 1.807*** 0.803 1.632 0.426 1.931 13.751 9.850 45.43 41.826 55.200 8.166 16.393 4.313 45.60 101.982*** 60.334 0.073 1.996*** 1.759 11.206 38.375 11.028 71.218*** Note: The impact of a market’s own effects is represented by 11, 22, 33 and 44 for the market US, China, Hong Kong and Taiwan respectively. Cross-market effects are given by 21 and 12, 34 and 43…..etc. In the conditional variance equation, the A coefficients represent ARCH effects, while the B coefficients are GARCH effects, and the G coefficients indicate asymmetric effects The respective coefficients are estimated with heteroskedasticity/misspecification adjusted standard errors using WINRATS 7.0, with convergence of the Asymmetric VAR-BEKK-MGARCH-M model reached at 235 iterations. The Ljung-Box level (LB), statistics are for 10 and 50 lags for the residuals and the squared residuals. ***, **, * - indicate significance at the 1%, 5% and 10% level 25 Exhibit 8 TO TO TO Volatility spillovers index Panel A: Full study period (Jan95 - Dec09) FROM 0 Hong Kong China 69.93 6.73 22.4 56.97 39.3 2.12 58.22 16.82 23.67 16.29 3 0.4 131.48 26.55 24.92 201.41 65.85 48.59 Taiwan 0.94 1.62 1.3 80.31 3.86 84.17 Contribution from others 30.07 60.71 76.34 19.69 186.81 spillovers index =0.4670 US Hong Kong China Taiwan Contribution to others Contribution including own Panel B: Post Asian Financial crisis period (Jul97 -Dec99) FROM US Hong Kong China 69.41 12.76 14.71 58.58 39.7 1.07 61.84 23.67 13.06 10.1 2.99 4.97 130.52 39.42 20.75 199.93 79.12 33.81 Taiwan 3.11 0.65 1.42 81.94 5.18 87.12 Contribution from others 30.58 60.3 86.93 18.06 195.87 spillovers index = 0.4897 US Hong Kong China Taiwan Contribution to others Contribution including own Panel C: Post Global financial crisis period (Jul07-Dec 09) FROM US Hong Kong China 72.49 5.94 13.23 63.63 30.73 0.9 53.1 16.11 21.52 56.82 18.55 3.32 173.55 40.6 17.45 246.04 71.33 38.97 Taiwan 8.32 4.74 9.26 21.31 22.32 43.63 Contribution from others 27.49 69.27 78.47 78.69 253.92 spillovers index = 0.6348 US Hong Kong China Taiwan Contribution to others Contribution including own Notes: The ijth entry in the table is the estimated contribution to the forecast error variance of market i, coming from innovations from country j. Therefore the “contribution to others” (off-diagonal column sums) or “contribution from others” (off-diagonal row sums), when total across markets, give the numerator of the spillover index. The denominator of the spillover index is given by the column sums or row sums (including diagonals) and totaled across markets. 26 Exhibit 9 Summary statistics of weekly conditional correlation and volatility estimates CH and HK Mean Std deviation Maximum Minimum CH and TW Mean Std deviation Maximum Minimum HK and TW Mean Std deviation Maximum Minimum US and CH Mean Std deviation Maximum Minimum US and HK Mean Std deviation Maximum Minimum US and TW Mean Std deviation Maximum Minimum China Hong Kong Taiwan US Mean Std deviation Mean Std deviation Mean Std deviation Mean Std deviation Full period Subperiod1 Subperiod 2 Subperiod3 Subperiod 4 Jan95-Dec09 Jan95-Jun97 Jul97-Dec99 Jan00-Jun07 Jul07-Dec09 783 returns 130 returns 131 returns 391 returns 131 returns Panel A: Weekly conditional correlations 0.3519 0.2690 0.4162 0.3177 0.4720 0.1394 0.1067 0.1432 0.1141 0.1300 0.7362 0.4785 0.7362 0.6236 0.7322 -0.0917 -0.0849 0.0190 -0.0917 0.0805 0.1878 0.1186 0.2484 0.1555 0.2924 0.1431 0.1161 0.1634 0.1162 0.1411 0.6582 0.4248 0.5845 0.5855 0.6582 -0.2028 -0.2028 -0.1840 -0.1996 -0.1452 0.2124 0.0307 0.3451 0.1587 0.4201 0.2034 0.1577 0.1918 0.1430 0.1513 0.6923 0.3832 0.6923 0.5689 0.6900 -0.3702 -0.3702 -0.1443 -0.2694 0.0511 0.1238 0.0753 0.1268 0.1080 0.2161 0.1509 0.1383 0.1377 0.1225 0.2060 0.6528 0.3366 0.4796 0.5082 0.6528 -0.3115 -0.2820 -0.2424 -0.3115 -0.3000 0.2079 0.2153 0.2087 0.2041 0.2109 0.1648 0.1683 0.1678 0.1358 0.2272 0.6863 0.6495 0.6234 0.5848 0.6863 -0.2157 -0.1758 -0.1508 -0.1507 -0.2157 0.0498 0.0137 0.0716 0.0402 0.0928 0.1488 0.1453 0.1548 0.1248 0.1942 0.6024 0.3363 0.6024 0.3540 0.5192 -0.3967 -0.3706 -0.2180 -0.3865 -0.3967 Panel B: Weekly conditional volatilities 0.0033 0.0025 0.0045 0.0029 0.0041 0.0014 0.0005 0.0013 0.0011 0.0015 0.0023 0.0010 0.0041 0.0015 0.0040 0.0020 0.0004 0.0027 0.0008 0.0022 0.0028 0.0019 0.0037 0.0022 0.0033 0.0012 0.0003 0.0012 0.0009 0.0012 0.0011 0.0002 0.0005 0.0005 0.0045 0.0025 0.0002 0.0005 0.0005 0.0050 Source: Derived from the Asymmetric VAR-BEKK-MGARCH-M model 27 Exhibit 10 Impact of AFC and GFC on the conditional correlations We examine the impact of the Asian financial crisis (AFC) and Global financial crisis (GFC) on the dynamic correlation by regressing the time-varying correlation (from the BEKK estimation) with its oneperiod lag and dummy variables, dummy one for Jul97to Dec 99 (131 weeks) for the AFC and for Jul07 to Dec09 (131 weeks) for the GFC, and zero otherwise. Due to the collinearity problem, the two dummies are included one at a time, and the results are reported in this Exhibit Correlation China-Hong Kong AFC (July1997-Dec1999) 0.0588** (0.0188) (0.0208) China-Taiwan 0.0744*** 0.1383*** (0.0251) (0.0263) Hong Kong-Taiwan 0.0431*** 0.0817*** (0.0138) (0.0157) 0.0029 0.0961*** US-China GFC (July 2007-Dec2009) 0.1219*** (0.0163) (0.0271) US-Hong Kong 0.00036 -0.0021 (0.0081) (0.0097) US=Taiwan 0.0182 0.0374** (0.0194) (0.0169) 28 Exhibit 11 Conditional correlations and volatilities graphs Panel A: Among the Greater China securitized real estate markets Correlation: Hong Kong and China .8 .4 .0 -.4 1996 1998 2000 2002 2004 2006 2008 Correlation: China and Taiwan .8 .4 .0 -.4 1996 1998 2000 2002 2004 2006 2008 Correlation: Hong Kong and Taiwan .8 .4 .0 -.4 1996 1998 2000 2002 2004 2006 2008 29 Panel B: Between the US and Greater China securitized real estate markets Correlation: US and China .8 .4 .0 -.4 1996 1998 2000 2002 2004 2006 2008 Correlation: US and Hong Kong .8 .4 .0 -.4 1996 1998 2000 2002 2004 2006 2008 Correlation: US and Taiwan .8 .4 .0 -.4 -.8 1996 1998 2000 2002 2004 2006 2008 30 Panel C: Conditional volatility China .010 .005 .000 95 96 97 98 99 00 01 02 03 04 05 06 07 08 09 04 05 06 07 08 09 03 04 05 06 07 08 09 03 04 05 06 07 08 09 Hong Kong .02 .01 .00 95 96 97 98 99 00 01 02 03 Taiwan .010 .005 .000 95 96 97 98 99 00 01 02 US .03 .02 .01 .00 95 96 97 98 99 00 01 02 Source: Derived from the Asymmetric- VAR (1, 1)-Mean-BEKK-MGARCH (1,1) model 31 Exhibit 12 Deterministic time trend in correlation, allowing for serial correlation This table reports the estimated deterministic linear trends in the time series of correlations: ( Correlationt * TREND i ) among the securitized real estate markets of China (CH), Hong Kong (HK), Taiwan (TW) and the US for the full sample period, post Asian financial crisis (subperiod 2) and post global financial crisis (sub-period 4). For each correlation series, we report the t-ps test from Vogelsang (1998) and the t-dan test from Brunzel and Vogelsang (2005). The 5% critical value (twosided) for t-tps is 2.152, and for t-dan, is 2.052. % increase/decrease in the time trend is estimated where either the t-ps or t-dan coefficient is significant (% increase/decrease = coefficient x no of weeks over the period). * - denotes significance at the 5% level Correlation CH - HK CH - TW HK - TW US - CH US - HK US - TW CH - HK CH - TW HK - TW US - CH US - HK US - TW CH - HK CH - TW HK - TW US - CH US - HK US - TW t-dan test coefficient t-stat Full period (Jan 95 - Dec 09) 0.00002 0.128 0.00012 1.202 0.00000 0.028 0.00010 1.309 0.00005 0.097 0.00023 0.427 0.00009 1.284 0.00014* 2.332 -0.00002 -0.241 -0.00004 -0.415 0.00003 0.561 0.00004 0.711 Subperiod 2 (Jul97-Dec99) - Post Asian financial crisis -0.00147 -1.032 -0.00157 -2.22800 -0.00153 -1.125 -0.00161 -1.76900 -0.00244 -0.349 -0.00274 -0.58200 -0.00098 -2.105 -0.00083 -1.06500 -0.00190 -0.215 -0.00170 -0.29700 -0.00139 -0.899 -0.00151 -1.43000 Subperiod 4 (Jul07-Dec09) - Post global financial crisis 0.00167 1.241 0.00116 1.838 0.00187 1.598 0.00138* 2.762 0.00287 1.762 0.00233* 2.446 0.00107 0.329 0.00008 0.048 -0.00051 -0.008 -0.00157 -0.006 0.00003 0.006 -0.00072 -0.437 coefficient t-ps test t-stat % increase/decrease na na na 10.96 na na na na na na na na na 18.08 30.52 na na na \ 32 Endnotes 1 This S&P/Citygroup global property database, the latest international public real estate database in the market, is designed to reflect components of the broad universe of investable international real estate stocks reflecting their risk and return characteristics. In total, the database has indices (both capitalization weighted and float adjusted) comprised of over 500 companies from more than 35 developed and emerging markets with a minimum market value of $100 million (Serrano and Hoesli, 2009). 2 Stock market studies uding the BEKK methodology include Arago-Manzana and Fernandez-Izquierdo (2007) (asymmetric bivariate BEKK), Caporale, Pitttis and Spagnolo (2006) (bivariate BEKK), Bhar and Nikolova (2009) (bivariate BEKK), Skintzi and Refenes (2006) (bivariate BEKK), Qiao, Chiang and Wong (2008) (bivariate BEKK) and Antoniou and Pescetto (2007) (trivariate BEKK). In the real estate literature, only one published study by Cotter and Stevenson (2006) use a bivariate symmetric VAR-BEKK-GARCH model. 3 Brunzel and Vogelsang (2005) develop a time trend test that retains the good size properties of the “t-ps” test and also has better power (both asymptotically and in finite samples). It is called the “t-dan” test as it uses a “Daniel Kernel” to non-parametrically estimate the error variance needed in the test. 4 We follow the stock market literature that suggests a GARCH (1,1)-BEKK specification is appropriate for the conditional variance lag length. 5 Before we assess whether the correlation series display trending behavior, we need to examine whether the estimated correlation series display stochastic or deterministic trends. Results from the autocorrelation and ADF unit root tests indicate that the conditional correlation series are stationary and shocks only have a temporary impact. 33