Time-Varying Linkages of Economic Activities in China and the Stock Markets in ASEAN-5

This study attempts to investigate the evolution of dynamic linkages and volatility spillover between the five countries of the Association of Southeast Asian Nations (ASEAN-5) stock markets and China’s economic activities. By using the movements and structural breaks of the time-varying correlation and Granger causality test, a suitable destination for equity portfolio diversification can be determined among the studied markets. This study covers monthly data from January 1991 to March 2015. The DCC-MGARCH model shows that the studied countries are time-varying correlated, while the structural break observed by Bai and Perron test coincided with major economic shocks, policy changes and the establishment of regional trade policies. The VAR model Granger causality test observed no volatility spillover from Chinese economic activities to the ASEAN-5 stock markets, except for Malaysia and the Philippines. However, the ASEAN-5 stock markets’ volatility exerts a significant influence on China’s economy, except for Singapore’s stock market volatility. This study reveals that ASEAN-5 has gradually became the preferred destination for diversifying equity portfolios for investors in China.


Introduction
Over the past two decades, China has played a growing and significant role in global trade. By 2011, China became either the largest or second-largest trading partner to 78 countries, which accounted for almost 55 per cent of the global GDP. These countries include all major global economies -the EU, the U.S., the Association of Southeast Asian Nations (ASEAN), Japan and India. 1 Although China claimed to be immune from crises in Asia, particularly the Asian Financial crisis, the late 2007 financial crisis that originated in the U.S. has an apparent impact on the future prospects of Asian economies, including China. The Chinese GDP growth rate plunged from 14.2% in 2007 to 9.6% in 2008, and it has not experienced a double-digit growth rate since then. This condition has a varying impact on various industries and its trading partners as a reduction in investments and infrastructure spending causes downward pressure for China's demand. This has been a major concern for ASEAN-5 investors; the impact of China's economic slowdown could be felt around the world. 2 It is also important to note that China has experienced This paper contributes to the existing literature by addressing the chicken-and-egg paradox that existed in the relationship between stock market volatility and economic movement volatility. This study also attempts to fill the gap in investigating the interaction between emerging economies caused by the ASEAN-5 stock market performance and Chinese economic growth in a time-varying manner and provides some noteworthy insights into the aspect of the volatility spillover direction across stock markets and economic activity movements in a multi-country context. We hope that our findings may shed some light on international asset diversification and management in ASEAN-5 and China, serving as a constructive guide to investors in determining the appropriate portfolio channel for equity market diversification.

Time-Varying Linkages of Economic
The remainder of this study is structured as follows: Section 2 examines the portfolio diversification literature, and Section 3 presents the methodology and de-scribes the data. The results are reported in Section 4, and the last section concludes the study.

Literature Review
Studies in the international diversification of assets have generated interest among academics and investors in ASEAN-5 because the ASEAN-5 stock markets are becoming more financially integrated over time (Candelon, Piplack, & Straetmans, 2008;Lim, 2009;Majid et al., 2009;Plummer & Click, 2005). It is vital to understand the concept of portfolio management prior to the discussion on the allocation of assets in international markets.
The modern portfolio theory was proposed by Markowitz (1952). According to this theory, risk-averse investors optimize the return on their investments by diversifying the portfolio of assets with negative correlations. This means that low and negative correlations among assets will offer gains in diversification (Hassan & Naka, 1996;Laopodis, 2005). High correlations, high volatility spillover and highly synchronized movements among the stock markets implied limited opportunity for risk diversification among the correlated and synchronized assets (Bekaert & Harvey, 1997). However, the rise in correlation among financial markets does not necessarily increase volatility in the domestic market (Bekaert & Harvey, 1997). It may be concluded that a high level of market integration reduces the possibility of risk sharing in the asset portfolio among these markets.
One should note that market integration shows a dynamic trend and changes over time (Gupta and Donleavy, 2009). The level of interdependency between countries changes gradually over time with trade and financial liberalization, the formation and implementation of regional trade arrangements, and information technology breakthroughs (Guillaumin, 2009;Harvey, 2000;Hooy & Goh, 2010;Laopodis, 2005). Aside from this, a regime change in the financial market of an emerging economy is prone to stimulation by the rise of market integration between countries (Bekaert & Harvey, 2002).
The diversification of assets is, in fact, a risk management method that intends to minimize the risk of the selected portfolio while maximizing the return of the portfolio (Vo & Daly, 2005). This can be done by combining a broad selection of investments with Time-Varying Linkages of Economic Activities in China and the Stock Markets in ASEAN-5 different levels of risk within a portfolio. To optimize portfolio investment, investors are advised to invest in various markets, in different industries, and in different regional blocks because the risk exposure tends to vary regionally. This implies that investors should aim to diversify in markets with low correlation. Even if the correlation rises over time and the value of the correlation is still relatively small over time, it is still possible to gain risk-adjusted returns by diversifying in that market (Bekaert, Harvey, & Lumsdaine, 2002;Forbes & Rigobon, 2002;Gupta and Donleavy, 2009).
Investors can benefit by diversifying their portfolio of assets internationally over time, but Kasa (1992) highlighted that previous studies may generate spurious conclusions based on the benefits gained from international diversification if researchers refer solely to the correlation. According to Kasa (1992), countries that share a common stochastic movement will increase their overall portfolio risk by diversifying into equities among these markets. Hence, financial markets that share a common stochastic trend are said to be cointegrated or financially integrated, which implies that the benefits of international diversification are limited in the long run (Laopodis, 2005). Nevertheless, a rational investor should not totally rule out cointegrated assets from the choices when diversifying his portfolio because there is still space for short-run gains in these cointegrated markets (Marashdeh, 2006;Masih & Masih, 1999;Syriopoulos, 2011). Benefits gained in the short run by international investors are possible by taking advantage of arbitrage mispricing that occurs in financial markets because the law of one price does not hold in the short run (Marashdeh, 2006;Syriopoulos, 2011). This point is especially practical when cointegration occurred among countries with low correlation (Phengpis & Swanson, 2006).
The existence of a low correlation between cointegrated economies potentially limits the benefits of assets portfolio diversification between these markets in the long run, but the opportunity for gains in the short run still exists (Bekaert & Harvey, 1997). If the two integrated financial markets are less correlated during a financial crisis, the collapse of the stock market in the first country will not cause the stock market in another country to collapse, as well. Hence, the portfolio diversification by the first country into the other country will help minimize the total losses during a financial crisis in the first country.
Several methods have been used to assess the potential gains in international portfolio diversification, namely, the bivariate and multivariate correlation in a GARCH framework, the cointegration analysis, VAR, VECM, Granger causality and cycle synchronization based on the stock market turning points (Guillaumin, 2009;Huang, Yang, & Hu, 2000;Pagan and Harding, 2002;Sohel Azad, 2009;Syriopoulos, 2011;Vo & Daly, 2005). These procedures can be used to assess the possibility of international portfolio diversification for the long and short horizon. It is hypothesized that if two stock markets are cointegrated, the gains from diversifying in these two markets are limited in the long horizon (Candelon et al., 2008;Sohel Azad, 2009).
However, the possibility for risk sharing among these markets in the short horizon still exists. The short horizon portfolio diversification can be assessed through the VECM and Granger causality (Gilmore & McManus, 2002).

Data Description
The sample period in this study ranges from January  Table 1 shows that the mean growth rate of China's industrial production index was negative but almost close to zero for China. The standard deviation for economic activity in China was below 5.5 percent, which implies that the risk of economic turbulence was low in China, indicating a stable economy over time. The Jarque-Bera test statistics strongly reject the hypothesis of normality, which implies that all return series are not normally distributed. The Engle (1982) ARCH test rejects the null hypothesis of no ARCH effects for all stock return series. The results justified the choice of the GARCH model. The autocorrelation patterns in the return series was examined using the Ljung-Box test. The Q statistics test results are significant at lag 12 for all series except for the stock market index for the Philippines at lag 1. The stock market index for Thailand is insignificant for Q statistics. The first order autocorrelations are low; hence, the first-order autoregressive process, AR(1), needs to be included in the mean equation of the GARCH (1,1) model (Arouri & Nguyen, 2009;Hamilton & Gang, 1996). 7
Time-Varying Linkages of Economic Activities in China and the Stock Markets in ASEAN-5 weaknesses of other MGARCH models (Engle, 2000).
Hence, this multivariate model is superior to other methods and produces the most accurate estimators (Engle, 2000).
The DCC-MGARCH model is used in this study to evaluate changes in economic interdependency as it can capture the degree of volatility correlation changes between two countries in a time-varying manner.
The extent of market integration is indicated by the time-varying conditional correlation (dynamic conditional correlation), whereas its trend can be gauged by the patterns of conditional correlation movements. Markets are said to become more integrated when conditional correlations increase over time (Yu, Fung, & Tam, 2010). Conditional correlations can also infer whether two economies are decoupling or recoupling over time. For example, two economies are viewed as decoupling when the conditional correlation falls over time. A high conditional correlation between two stock markets can also indicate less opportunity for diversification.
The DCC-MGARCH model (Engle, 2002) generates the volatility correlation between two markets either directly through its conditional variance or indirectly through its conditional covariances. The conditional covariances generated show the time-variant crossmarket volatility correlation or the dynamic conditional correlation.
The DCC-MGARCH model is a dynamic model with the time-varying means, variances and covariances of the n return series, t i r , for country i at time t with the equations as follows: Each individual country's return series follows an autoregressive process (Arouri and Nguyen, 2009;Hamilton and Gang, 1996). The conditional mean equation of the return series for country i is written as µ i ,t = λ 0 + λ 1 r i ,t−1 , where λ 0 is a constant term, and λ 1 is the coefficient of the lagged return for each market.
The conditional variance-covariance matrix for the DCC-MGARCH model is written as H t = D t R t D t , where H t is also denoted as the conditional correlation estimator, R t of the DCC model is a (n x n) time-varying conditional correlation matrix, and the conditional variance must be unity, while D t is a (n x n) diagonal matrix of time-varying standard deviations of the returns in the mean equation from the univariate where τ i is a constant term for the conditional variance equation for country i, α i is the ARCH effect of the return series, while β i is the GARCH effect of the return series. A positive coefficient of β i implies volatility clustering and persistence in positive changes in stock market indices. A sum of α i and β i that is less than 1 indicates the stationarity of the GARCH model and shows that the volatility shock is time-decaying and mean-reverting (Bollerslev, 1986). 9 The R t matrix contains the coefficients of the conditional correlation, which is given as where Q t is a (n x n) conditional covariance matrix that is a symmetric and positive definite, given by Q denotes the unconditional covariances of the standardized errors matrix, which is a (n x n) symmetric positive definite matrix, and ε t = (ε 1,t ,...,ε n,t )' is the standardized residual terms. 10 If (α dcc + β dcc ) < 1, the model is mean reverting, but when (α dcc + β dcc ) = 1, the model is said to be integrated.
Engel (2002) suggested the estimation of the timevarying conditional correlations (q ij,t ) for any two return series included in R t by the following GARCH (1, 1) process.
where ρ ij is the unconditional correlation between ε i ,t and ε j ,t , and q ij ,t is the mean value of q ij ,t . The average variance is unity and the conditional correlation esti- can be rewritten as follows: The significance of α dcc and β dcc implies that the estimators obtained in the DCC-MGARCH are dynamic and time-varying. α dcc measures the short-run volatility impact, which means that the persistency of the standardized residuals from the previous period. β dcc measures the lingering effect of a shock impact on the conditional correlations, which is the persistence of the conditional correlation process.
In this paper, the conditional correlation between the stock return series of country i and the industrial growth rate series of country j at time t is referred to as [R t, ] ij = ρ ij ,t , where ρ ij ,t indicates the direction and strength of the correlation and measures the degree of covariance between two indices in relation to the market's individual variances (Savva, 2009). A positive ρ ij ,t implies that the correlation between the return series is rising and moving in the same direction and vice versa.

DCC-MGARCH Log Likelihood Estimation
This study uses the Quasi-maximum likelihood (QML) estimation proposed by Bollerslev and Wooldridge (1992) for the estimation of the DCC-MGARCH model. 11 The log-likelihood of the H t estimator can be written in the following forms: The log-likelihood function consists of the volatility component, and correlation component can be expressed as L (θ ,Φ) = L v (θ )+ L c (θ ,Φ) , where θ is the parameter for D t and Φ is the additional parameter in R t . The volatility component function can be written as while the correlation component is written as

Two-Stage Estimation of the DCC-MGARCH Model
The DCC-MGARCH model estimation uses a two-

Bai and Perron Multiple Structural Break Test
The BP multiple structural break test is applied to sta-  Gregory and Hansen (1996) cointegration test with structural breaks is not applied in this research because it can only capture one break in the intercept and/or slope coefficient in the cointegrating relationships. The regime switching method is another commonly used approach in dating structural breaks, but it is relatively less powerful than the BP method (Bai & Perron, 2003).
The Bai and Perron (BP) multiple structural break test has been adopted to estimate the number of break dates that happened naturally in the conditional correlations (Arouri & Nguyen, 2009;Nguyen & Bellalah, 2008). 12 It is vital to identify the date of the structural changes in the financial market, which is induced by the advancement in market integration (Bekaert & Harvey, 2002). This study followed the approach of Bekaert et al. (2002), which allows the structural break dates to narrate what is actually happening in the markets.
According to this method, the structural changes occurring naturally in the dynamic conditional cor-Time-Varying Linkages of Economic Activities in China and the Stock Markets in ASEAN-5 relation series at time t are denoted as y t = η j x t + ε t , where t = T j-1 +1,…,T j (j = 1,…, m+1), η j is the coefficient of the covariate vectors, and t x . ε t is the error term at time t.

Vector Autoregression Model -Granger Causality Tests
The two-step estimation allows this study to utilize the conditional variances generated in step 1 to more accurately reveal the direction of the volatility spillover among the ASEAN-5 stock markets and Chinese economic activities. This strategy differs from previous studies because the estimation is based on the stocks' return to determine the volatility spillover effect. 13 The results of the direction of spillover among the studied countries will aid the ASEAN-5 and investors in China in determining the direction of the investment portfolio.
Causality tests can be used to locate the direction of the volatility transmission and the degree to which one economy's stock market or economic volatility impacts other markets or volatilities (Gupta & Guidi, 2012;Huang et al., 2000;Savva, 2009

ASEAN-5 Stock Market Volatility and China Industrial Production Volatility
The univariate GARCH parameters, which are estimated from step 1 of the DCC-MGARCH model, are reported in Table 2. The impact of the previous disturbances on the conditional variance denoted by the ARCH coefficient (α) is significant for all selected countries in this study. All GARCH coefficients (β) are highly significant, which implied persistency in the positive changes in the volatility of the stock index except for China. The sum of (α + β) are all less than one for the ASEAN-5 stock markets as well as for China's economic performance, which indicates the persistence of disturbance over time.
The robust tests for the model standard residual are provided in The statistically insignificant results of these two tests indicate the absence of serial correlation and ARCHeffects in the conditional means and variances. The results of these two tests justified the employment of the DCC-GARCH specification for the selected countries, thus passing the misspecification tests. Table 2 also reports the diagnostics of the standardized residuals. The skewness is positive for most stock return indices except for JSE and SSI. CH showed a high excess kurtosis of more than three. Thus, the level of risk for CH is low as the past returns yield a leptokurtic distribution. 17 The JB test for normality strongly rejected the null hypothesis, which means that the indices do not follow a normal distribution. Hence, this suggests that the excess kurtosis in the residuals of the return indices were not fully eliminated by the conditionally normal GARCH process. Consequently, the diagnostics test justified the use of QML procedures in the estimation of the DCC-MGARCH model.

Conditional Volatilities
The monthly volatility of the ASEAN-5 stock markets and the industrial production growth rate for China, as measured by the conditional variances, are depicted in  Remarks: Lag length: Automatic is based on SIC, MAXLAG = 14. Lag length is given in parentheses. ***, **, * indicate significance at 1%, 5% and 10%, respectively. UniGARCH for the A5-US model will be used as a comparison reference for all models. Only minor and insignificant differences exist in the results for all models.    Table 3. To have a better picture of inter-regional correlation among ASEAN-5 stock markets with economic performance for China, the plots for their time-varying conditional correlations are illustrated in Figure 2. The DCCs were maintained at a level below 0.3 during the study period. At the beginning of the study period until 1994, most of the DCCs remained at a negative level, which implies that the ASEAN-5 stock markets and Chinese economies activities were inversely related.

The DCC-MGARCH -Time-Varying Patterns for ASEAN-5 Stock Markets with the Chinese Economic Activities
However, the DCC was increasing during pre-1994.   Remarks: ***, **, * indicate significance at 1%, 5% and 10%, respectively To summarize, from the analysis of the time variant correlation, the estimated conditional correlations between stock markets and economic activities were relatively low with a few peaks and plunges. A weak correlation of below 30% between these two regions shows that they were less integrated. To further confirm the observed movements of DCC in Figure 2, it is necessary to identify the possible structural changes that occurred during that time to determine the actual structural break dates. To achieve this, the BP multiple structural break test was adopted.

Perspective of Structural Change
The was at its peak and the NASDAQ index was over 5000 points (Yu, 2004). This devastating stock market crash swirled in the impacted economies for approximately two years. The internet bubble in China expanded until the market became overheated during the meltdown of the internet bubble in the U.S. because the spillover of the U.S. internet boom to China was slower than the internet burst in the U.S. (Girardin & Liu, 2007).  Overall, the BP multiple structural break dates show that some of the breaks corresponded with the events in the countries, and some are related to global or regional events. This indicates that the countries examined experienced not only cross-country spillover effects but also the spillover effect from external shocks that affect both countries' correlation. Hence, it is worthwhile to further analyze the direction of the cross-country spillover effects between the ASEAN-5 stock market volatility and the Chinese economic activity volatility. This is important because the results will show whether the Chinese economic activity volatility influences the ASEAN-5 stock markets or vice versa. The findings will aid investors in deciding their future investment directions for portfolios.

Volatility Spillover between ASEAN-5 Stock Markets and China's Economic Activities
During the period of study, the Chinese economic activities were Granger-caused by the ASEAN-5 stock markets except for the Singapore stock return volatility. However, the influence of China's economic movements on ASEAN-5 stock markets' volatility was insignificant over the period of study except with regard to Malaysia and the Philippines' stock markets. The findings of this study are consistent with Valadkhani and Chen (2014), who found that the stock volatility of one country or region does influence the economic volatilities in another country or region. Similar conclusions were observed in studies by Schwert (1989;1990), Hamilton and Gang (1996) and Ibrahim (2010 Tsouma's (2009) conclusion that the growth of the industrial production index influences stock returns in the emerging market.
As a conclusion, Singapore is a preferred destination for investors in China to diversify their asset portfolio and vice versa.

Conclusion
The purpose of this study is to explore the dynamic linkages between Chinese economic movement and ASEAN-5 stock markets using the time-variant framework and Granger causality tests. The aim is to identify the direction of the volatility spillover among the studied markets. This study also intends to detect the structural break dates observed between the ASEAN-5 stock markets and the Chinese economy, which are related to important facts and economic events. Last, the study attempts to review the equity diversification portfolio  Remarks: ***, **, * indicate significance at 1%, 5% and 10%, respectively. Null hypothesis: The excluded variable does not Granger-cause the dependent variable. The best lag selection is automatic based on lag length criteria (AIC). For the full sample, the best lag is 12.
Time-Varying Linkages of Economic Activities in China and the Stock Markets in ASEAN-5 destination of China with ASEAN-5. Taken together, this study aims to uncover the possible deepening in the future regional economic integration in Asia. These findings offered important implications for investors, portfolio managers, policymakers, and leaders in ASEAN and China. Hence, it is wise for ASEAN leaders to pursue market-based integration with China rather than an institutional-based integration. It is also not recommended for ASEAN to push forward with economic integration with the Eurozone due to its current debt crisis.