The Return Co-movements and Volatility Transmission of Japan – China – USA Stock Markets

(1)

The Return Co-movements and Volatility

Transmission of Japan – China – USA Stock Markets

University of Groningen

Faculty of Economics and Business

Master’s thesis, MSc Finance

Author: Yang Li

Student Number: s2313103

Address: Esdoornlaan 558, 9741ME Groningen, The Netherlands

Email: echo.liyang21@gmail.com

(2)

Abstract

Using the data of four stock indices (S&P 500 Composite, Nikkei 225, SHS A and SHS B) in period 2003-2013, I examine the co-movement and volatility transmission among the US, the Japanese and the Chinese stock market by applying the MA(1)-GARCH(1,1)-M model. In order to find the influence of policy adjustment and global financial crisis, data is divided into three sub-panels. According to the research evidence, I find spillover effects from the US stock market to the Japanese and the Chinese A and B Shares for both close-to-open and open-to-close returns. But for the open-to-close returns, the co-movement between US and the Chinese markets is not obviously strengthened during the crisis period.

JEL codes: C22, F21, G15

(3)

1. Introduction

The extent of international economic and financial integration has received much attention since the 1990s. Under this background, it would be interesting to see whether the emerging markets have also joined the globalization, and what kind of influence they could bring over to other markets. In this paper, I consider China as a representative of emerging markets. After joining the WTO, China has built a tighter relation with the world economy. According to the figure of GDP from IMF, UN and WBG, China has become the second largest economy since 2010. Meanwhile, the Chinese stock market, although still in developing stage, has also got substantial progress. By the end of year 2011, Shanghai Stock Exchange became one of the world top 10 stock exchanges, ranked by market cap and trade volume.

As shown in Figure 1 and Figure 2, both A and B Shares’ market cap fluctuated dramatically during the period 2003–2013. Although the market cap had a very short term increase in both Shares during 2003, it experienced a gentle decrease during the next three years. Then the market cap of A and B Shares rebounded after hitting rock bottom in middle 2006, and picking at 26849.727 and 141.859 billion USD respectively by the end of 2007. Since the global financial crisis begins in 2007, the market cap of both shares falling rapidly within one year. By the end of 2008, the market cap of A Shares evaporated nearly 50%, and B Shares’ dropped nearly to the historical lowest point since 2003. Although both shares slightly recovered after 2009, the growth rates were still low and fluctuated, and neither market shares has rich the level before the crisis. The performance of both shares is in accordance with the world stock market.

Figure 1. A Shares Market Cap (source: The People’s Bank of China, and Shanghai Stock

(4)

Figure 2. B Shares Market Cap (source: The People’s Bank of China, and Shanghai Stock

Exchange)

Meanwhile, the Chinese stock market is emerging all the time, not only in the size of market capital, but also in the system and corresponding regulations. There are two main reforms: the introduction of QFII (Qualified Foreign Institutional Investor), and the Split Share reform. At the beginning, A and B Shares are produced for Chinese local investor and foreign investor respectively. After the introduction of QFII in 2001, foreign investment in Mainland China A Shares officially begins. It can be viewed as a sign of connection between Chinese stock market and the global stock market. In the same year, domestic citizens were allowed to invest in B Shares via the secondary market. Despite the influence from other economic factors, the Split Share reform, which is completed by the end of 2006, may also contribute to the rebound. Before the split share reform, there was a large volume of non-tradable state-owned and legal person shares, and only about one-third of the shares in domestically listed firms could be treaded on the stock markets. In other words, this reform enlarged the scale of available investment scope, and made the Chinese stock market more open to both domestic and

global investors. As the policy changes might affect the macroeconomic conditions,

influences may transmit to capital market, and therefore affect the co-movement with other stock markets.

(5)

Figure 3. USD/RMB Monthly Average Exchange Rate (source: The People’s Bank of China)

Besides, during the last decade, the RMB appreciated over 20%. Figure 3 presents the steady decrease trend of USD/RMB monthly average exchange rate. Despite the investment interest, the expected appreciation of RMB can also be viewed as a source of return. Since the domestic exchange rate movement belongs to macroeconomic factors, it might also have influence on capital market. Although the relationship between exchange rate and stock markets’ co-movement has not been studied, some researchers do find relationship between stock indices and exchange rate (e.g. Tsai, 2012; Lin, 2012).

Since Chinese stock market is growing in high pace, it has received more attention. Huang et al (2000) find that, compared to Japan, stock price changes in the US stock market have more impact on the Chinese stock market during 1992-1997, but there was no co-integration relationship between the mainland Chinese stock market and others. Chow and Lawler (2003) study the co-movement of Shanghai and New York stock in 1992-2002, and document that there was positive serially correlations on volatility in both markets, but the volatility measure of the two markets are significantly negatively correlated. Taking the similar research period, Lin et al (2009) argues that there was a low correlation between the Mainland Chinese markets and Major western markets. In the more recent research, Chow et al (2011) find a steady increasing trend on the effect of the current stock return from New York to Shanghai after the 1997 Asian financial crisis. Besides, the effect of the current return for Shanghai on New York also becomes signify positive and increasing after 2002, when China entered WTO.

Although a lot of research has been done on the co-movement between the Chinese stock market and others, rare study clarifies the influence of Chinese stock market

(6)

reform or present clearly comparison between A and B Shares. Besides, the effect of exchange rate on stock market co-movement is rarely mentioned. What is more, using earlier data, those study results may not very applicable to describe the recent situation. Therefore, in order to have an in-depth look of the Chinese stock market and its co-movement with other majority stock markets, I select Japan and US as the representatives of developed stock markets in Asia and Western countries, and the following questions are addressed in this paper:

(1) Did the stock markets linkages strengthen due to Chinese stock market reform? If so, there would be significant changes on coefficients among different period. (2) Did the different structure and target shareholders of Chinese A and B Shares

affect the stock markets co-movements? As there is stepped up demands for the combination of A and B Shares, this analysis could provide some data support on this topic.

(3) Did geographical location be a source of influence on stock markets’ co-movements? I expect geographically and economically close countries should exhibit higher levels of market linkages due to the presence of similar investor groups and multi-listed companies.

(4) Did exchange rate influence the co-movement of stock markets? By comparing the two regression results which containing and not containing the exchange rate, the effect can be observed.

(7)

2. Literature review

According to previous studies, there are two main theoretical explanations for the stock markets’ co-movements. The first is economic fundamentals. Solnik (1974), Grauer et al (1976) and Stulz (1981) argue that, at the macroeconomic level, there are common fundamental variables that can influence several stock markets across different economies. Recent studies confirm this theory and clarify that macroeconomic factors such as business cycle fluctuations, the inflation environment, monetary policy stance and domestic exchange rate movements affect international stock correlations (Dumas et al, 2003; Yang et al, 2009; Kiviaho et al, 2012; Park, Park, 2014). The second theory is market contagion. Forbes and Rigobon (2002) define financial contagion as a significant increase in cross-market linkages after a shock to an individual country, and there is contagion only if two markets show a significant increase in co-movement during crisis periods compared to periods of stability. Madaleno and Pinho (2012) also argue that cross-market co-movements do not increase significantly after the shock, then any continued level of market correlation only suggests the interdependence between the two economies.

(8)

increasing co-movements in price returns, volatilities and correlations among the four stock markets, particularly between the US and Europe. Besides, they attribute the idiosyncratic behavior of the Japanese stock market to the persistent stagnation of the economy and the weak fundamentals over the 1990s. Ozdemir and Cakan (2007) use non-linear causality tests to examine the dynamic relationships among the stock market indices of the US, Japan, France and the UK, and find a strong bidirectional nonlinear causal relationship between the US and the other countries. For the emerging stock markets, Samarakoon (2011) develops a VAR model to study the daily index returns of 62 stock markets for the period 2000-2009, and argues that there exists important bi-directional and asymmetric interdependence and contagion in emerging markets, with important regional variations. Kenourgios et al (2011) investigate financial contagion in a multivariate time-varying asymmetric framework, focusing on four emerging equity markets (the BRIC countries) and two developed markets (US and UK) during five financial crises. They confirm a contagion effect from the crisis country to all others for each of the examined financial crises. Their results also suggest that emerging BRIC markets are more prone to financial contagion, while industry-specific turmoil has a larger impact than country-specific crises.

(9)

Singapore are highly integrated with the stock market in Japan. Such integration gets stronger over time since 1994.

A large number of studies also document that co-movement increases significantly during crisis periods. Meric et al (2008) use the principal components analysis (PCA) technique to study the co-movements of the national benchmark and sector index returns in the US, UK, German, French and Japanese stock markets during the bull- (1997-2000) and bear- (2000-2002) markets. They find high global diversification in a bull market, while the sectors of different countries tend to be more closely correlated in a bear market. Arouri et al (2010) employ a DCC-GARCH (Dynamic Conditional Correlation GARCH) model to analyze the time variations in the co-movements of six Latin American stock markets. Their findings echo other researchers: the co-movements are subject to various regime shifts; stock markets move much more together in times of crisis. The study of Kiviaho et al (2012) investigates the stock return co-movement of European frontier markets with the US market and the three largest developed markets in Europe (UK, Germany, and France) by applying a three-dimensional analysis of wavelet coherency. They conclude that, due to the influence of macroeconomic factors the level of co-movements of different stock market are diverse, but in general, the co-movement increases during the turbulent period of the global financial crisis of 2008/2009.

(10)

(Mierau and Mink, 2013).

Although exchange rate movements are viewed as one of the macroeconomic factors that influence the stock markets’ co-movements, direct research on this topic has been rare. The main body of previous related studies focuses on the relationship between stock price indices and exchange rates. However, the empirical evidence is mixed. For instance, Tsai (2012) adopts the Augmented Dickey-Fuller test and Phillips-Perron test to estimate the relationship between stock price indices and exchange rates among six Asian countries. He finds a negative relation between stock and foreign exchange markets which is more obvious when exchange rates are extremely high or low. Lin (2012) gets similar results, and the empirical results suggest that the co-movement between exchange rates and stock prices becomes stronger during crisis periods. Moreover, Ozair (2006) is unable to find any significant relationship between these two variables.

3. Methodology

In this paper, I will study the co-movement of stock markets in Shanghai, Tokyo and New York in separated time periods. As the Chinese stock market is developing all the time, the system and corresponding regulations are also emerging. The changes might affect the influence of the macroeconomic conditions on the capital market, and therefore affect the co-movement with other stock markets. This research takes the 1st August 2003 as the starting point, after which both Mainland Chinese and foreign investors have equal investment opportunities in the Chinese stock market.1 I use 31st October 2006 as the break point to divide the whole research period into two subsamples, since it is the date that “split share reform” was completed. Taking this brick point in this research aims to test the influence of regulation on stock markets co-movements. Besides, as mentioned previously, a lot of researchers argue that market crashes and pressure enhance the co-movement of stock markets. It is interesting to see whether this conclusion is applicable to these three stock markets. Hence, I take the 2007 global economic crisis as a third subsample period to test the influence of market pressure on co-movement. However, there is no specific date on the beginning and ending of this crisis. The mainstream thinks the crisis started since middle 2007 and ended around 2009,

(11)

while some researchers consider the influence of the crisis still exists. My observation of the A and B shares’ market cap is in accordance with the mainstream point of view. Besides, taking this sub research period aims to take an in-depth look of the core crisis

period. Therefore, I plan to set the third subsample period from 1st July 2007 to 31st

December 2009. By dividing the time period properly, this paper can better reveal the real

situation of the co-movement of the three stock markets.

In order to study the co-movement of returns and volatility between stock markets, a model that can reflect and capture the changes of stock market returns and volatilities is needed. Hence, an ARCH type model could be a proper choice. Since there are many models developed from the ARCH model, it is necessary to have a general overview of the models before the model selection.

The ARCH(q) model developed by Engle (1982) assumes a time-varying variance. It describes how the variance of the errors evolves. With the conditional variance𝜎_𝑡2_{, the}

simplest representation of this model ARCH(1) would be written 𝑦𝑡 = 𝛽1+ ∑𝑛𝑖=2𝛽𝑖𝑥𝑖𝑡+ 𝜖𝑡, 𝜖𝑡 ∼ 𝑁(0, 𝜎𝑡2)

𝜎_𝑡2 = 𝛼₀+ 𝛼₁𝜖_𝑡−12

where 𝑦_𝑡 denotes the stock return in one market, and the 𝑥s are the factors that could influence or interpret the stock return. There are three limitations on ARCH(q) models. First, there is no specific answer on how to decide on the number of lags of squared residuals. Second, the number of lags of squared errors might be very large if required to capture all dependences in the conditional variance. Third, “everything else equal, the more parameters there are in the conditional variance equation, the more likely it is that one or more of them will have negative estimated values” (Brooks, 2008, p392). These limitations restrict the usage of ARCH(q) model.

Bollerslev (1986) and Taylor (1986) developed the GARCH model by allowing the conditional variance to be dependent upon previous own lags. The conditional variance of GARCH (1,1) becomes

𝜎_𝑡2 _{= 𝛼}

0+ 𝛼1𝜖𝑡−12 + 𝛽𝜎𝑡−12

(12)

can only be avoided by placing artificial constraints on the model coefficients to force them to be non-negative; second, GARCH models cannot account for leverage effects;2 finally, the model does not allow for any direct feedback between the conditional variance and the conditional mean (Brooks, 2008). Due to these limitations, a huge number of extensions and variants have been proposed.

It is supposed that investors should be rewarded for taking additional risk by obtaining a higher return. Engle et al (1987) developed the GARCH-M model, which allows the conditional variance of asset returns to enter into the conditional mean equation, and the conditional variance is defined in the same way as the GARCH (1, 1) model. The GARCH-M (1, 1) has the form

𝑦𝑡= 𝜇 + 𝛿𝜎𝑡2 + 𝜖𝑡, 𝜖𝑡 ∼ 𝑁 (0, 𝜎𝑡2)

𝜎_𝑡2 _{= 𝛼}

0+ 𝛼1𝜖𝑡−12 + 𝛽𝜎𝑡−12

When coefficient 𝛿 is positive and statistically significant, it represents that the increased risk leads to a rise in the mean return. According to Brooks (2008: p. 410), in some empirical applications, the conditional variance term 𝜎_𝑡2 appears to be lagged 𝜎_𝑡−12 , or in square root form 𝜎𝑡−1.

GARCH models can be used to model the volatility of a series over time. In order to account for other important features of financial series at the same time, other models could be combined with GARCH models as a more complex hybrid model. Bollerslev (1986) and French et al (1987) extend the GARCH (1, 1)-M model to allow the first-order moving average MA (1) to be another explanatory variable of the conditional mean return. The equation then becomes

𝑦𝑡 = 𝜇 + 𝛿𝜎𝑡2+ 𝛾𝜖𝑡−1+ 𝜖𝑡, 𝜖𝑡∼ 𝑁 (0, 𝜎𝑡2)

Hamao et al (1990) find that daily stock returns measured from close-to-open and open-to-close can be approximated by the MA(1)-GARCH(1,1)-M model, and deploy this model in their study. Some later researches also consider GARCH(1,1) or EGARCH(1,1) could properly and extensively study the volatility of stock market (Lamoureux & Lastrapes, 1993; Scheicher, 2001). Therefore, in the following section,

2_{GARCH models enforce a symmetric response of volatility to positive and negative shocks. However, a}

(13)

I will employ the MA(1)-GARCH(1,1)-M model, and then test the volatility spillover effect and co-movements by introducing the squared residuals and returns of previous foreign stock market to the domestic market’s conditional variance and conditional means of returns respectively. In Hamao’s method, he breaks the stock markets’ daily close-to-close return into their close-to-open and open-to-close components. The spillover effects from foreign markets on the conditional means of the close-to-open return “reflects the effects on opening prices in the domestic market”; while spillover effects on conditional means of the open-to-close return “reflect effects on prices in the domestic market after the opening of trading” (Hamao et al, 1990, p 282). Hence, the foreign market’s influence on opening price and trading price after the opening can both be observed and evaluated. What is more, in their research, a dummy variable for the trading day following a weekend or holiday is included in both the conditional mean and variance equations, which aims to capture potential ‘day of the week’ effects. The dummy variable is necessary to be included into the model, because if any of such calendar phenol are present in the data but ignored by the model-building process, the result is likely to be a misspecified model (Brooks, 2008: Chapter 9).

In this paper, I follow the approach of Hamao et al (1990) to test the co-movement of three stock markets. The different type of return can be shown as:

𝑅_𝑖,𝑡𝐶𝐶 _{= ln(𝑝}

𝑖,𝑡𝐶) − 𝑙𝑛(𝑝𝑖,𝑡−1𝐶 )

where 𝑅_𝑡𝐶𝐶denotes the continuously compounded close-close return on day t, 𝑝_𝑡𝐶 denotes the closing price on day t, i denotes the country’s ISO 3166-1 alpha-2 codes (CN – China, JP – Japan, US – USA).

𝑅_𝑖,𝑡𝐶𝑂 _{= ln(𝑝}

𝑖,𝑡𝑂) − 𝑙𝑛(𝑝𝑖,𝑡−1𝐶 )

where 𝑅_𝑡𝐶𝑂denotes the continuously compounded close-open return on day t, 𝑝_𝑡𝑂 denotes the opening price on day t.

(14)

First, I employ the MA(1)-GARCH(1,1)-M model as Model I to test the spillover effects in open-close stock returns.

𝑅_𝑖,𝑡𝑂𝐶 _{= 𝛼 + 𝛽𝜎}

𝑖,𝑡2 + 𝛿𝐷𝑖,𝑡 + 𝛾𝜖𝑖,𝑡−1+ 𝜖𝑖,𝑡

𝜎_𝑖,𝑡2 = 𝑎 + 𝑏𝜎_{𝑖,𝑡−1}2 + 𝑐𝜖_{𝑖,𝑡−1}2 + 𝑑𝐷𝑖,𝑡

where 𝜎2 represents the conditional variance of the stock index return R, and D represents a dummy variable, which takes the value of 1 on a day following weekends and holidays, or 0 otherwise, i denotes the country code.

Next, I introduce one exogenous variable, which capture the potential volatility spillover effect from the previously open foreign market’s open-to-close return, into the domestic stock market’s open-to-close conditional variance. Model II:

𝑅_𝑖.𝑡𝑂𝐶 _{= 𝛼 + 𝛽𝜎}

𝑖,𝑡2 + 𝛿𝐷𝑖,𝑡 + 𝛾𝜖𝑖,𝑡−1+ 𝜖𝑖,𝑡

𝜎_𝑖,𝑡2 _{= 𝑎 + 𝑏𝜎}

𝑖,𝑡−12 + 𝑐𝜖𝑖,𝑡−12 + 𝑑𝐷𝑖,𝑡+ 𝑓𝜖𝑗,𝑡2,𝑂𝐶

where 𝜖_𝑗,𝑡2,𝑂𝐶 denotes the most recent squared residuals derived from Model I applied to the open-close return of the previously open foreign markets. The j is the country code that represents the foreign market. Due to the time difference of the three stock markets, the specific selections on the date of the foreign market’s squared residual are different. Figure 4 shows the trading hours of different stock exchanges.

For the Japanese stock market: 𝑅_{𝐽𝑃,𝑡}𝑂𝐶 _{= 𝛼 + 𝛽𝜎}

𝐽𝑃,𝑡2 + 𝛿𝐷𝐽𝑃,𝑡+ 𝛾𝜖𝐽𝑃,𝑡−1+ 𝜖𝐽𝑃,𝑡

𝜎_{𝐽𝑃,𝑡}2 _{= 𝑎 + 𝑏𝜎}

𝐽𝑃,𝑡−12 + 𝑐𝜖𝐽𝑃,𝑡−12 + 𝑑𝐷𝐽𝑃,𝑡 + 𝑓𝜖𝐶𝑁,𝑡−12,𝑂𝐶

or 𝜎_{𝐽𝑃,𝑡}2 = 𝑎 + 𝑏𝜎_{𝐽𝑃,𝑡−1}2 + 𝑐𝜖_{𝐽𝑃,𝑡−1}2 + 𝑑𝐷_{𝐽𝑃,𝑡}+ 𝑓𝜖_{𝑈𝑆,𝑡−1}2,𝑂𝐶 For the Chinese stock market:

𝑅_{𝐶𝑁,𝑡}𝑂𝐶 _{= 𝛼 + 𝛽𝜎}

(15)

𝜎_{𝐶𝑁,𝑡}2 _{= 𝑎 + 𝑏𝜎}

𝐶𝑁,𝑡−12 + 𝑐𝜖𝐶𝑁,𝑡−12 + 𝑑𝐷𝐶𝑁,𝑡+ 𝑓𝜖𝐽𝑃,𝑡−12,𝑂𝐶

or 𝜎_{𝐶𝑁,𝑡}2 = 𝑎 + 𝑏𝜎_{𝐶𝑁,𝑡−1}2 + 𝑐𝜖_{𝐶𝑁,𝑡−1}2 + 𝑑𝐷𝐶𝑁,𝑡+ 𝑓𝜖𝑈𝑆,𝑡−12,𝑂𝐶

Although the Japanese stock market opens earlier than the Chinese stock market, the opening hours of the two markets are almost totally overlapped. Hence, I take the previous day’s open-to-close return of the Japanese market. The other overlap problems in the following models are handled in the same way.

For the US stock market: 𝑅_{𝑈𝑆,𝑡}𝑂𝐶 _{= 𝛼 + 𝛽𝜎}

𝑈𝑆,𝑡2 + 𝛿𝐷𝑈𝑆,𝑡+ 𝛾𝜖𝑈𝑆,𝑡−1+ 𝜖𝑈𝑆,𝑡

𝜎_{𝑈𝑆,𝑡}2 _{= 𝑎 + 𝑏𝜎}

𝑈𝑆,𝑡−12 + 𝑐𝜖𝑈𝑆,𝑡−12 + 𝑑𝐷𝑈𝑆,𝑡+ 𝑓𝜖𝐶𝑁,𝑡2,𝑂𝐶

or 𝜎_{𝑈𝑆,𝑡}2 = 𝑎 + 𝑏𝜎_{𝑈𝑆,𝑡−1}2 + 𝑐𝜖_{𝑈𝑆,𝑡−1}2 + 𝑑𝐷_{𝑈𝑆,𝑡}+ 𝑓𝜖_{𝐽𝑃,𝑡}2,𝑂𝐶

Figure 4. Trading time of the Japanese, the Chinese and the US stock markets.

Then I extend Model II by adding one more exogenous variable – the open-close return of the previously open foreign markets, to estimate the spillover effects on the conditional means of stock returns. The Model III will be:

(16)

𝜎_𝑖,𝑡2 _{= 𝑎 + 𝑏𝜎}

𝑖,𝑡−12 + 𝑐𝜖𝑖,𝑡−12 + 𝑑𝐷𝑖,𝑡+ 𝑓𝑋𝑗,𝑡𝑂𝐶

where 𝑅_𝑗,𝑡𝑂𝐶 denotes the open-close return of the previously open foreign market. For the Japanese stock market:

𝑅_{𝐽𝑃,𝑡}𝑂𝐶 _{= 𝛼 + 𝛽𝜎} 𝐽𝑃,𝑡2 + 𝛿𝐷𝐽𝑃,𝑡+ 𝜙𝑅𝐶𝑁,𝑡−1𝑂𝐶 + 𝛾𝜖𝐽𝑃,𝑡−1+ 𝜖𝐽𝑃,𝑡 𝜎_{𝐽𝑃,𝑡}2 _{= 𝑎 + 𝑏𝜎} 𝐽𝑃,𝑡−12 + 𝑐𝜖𝐽𝑃,𝑡−12 + 𝑑𝐷𝐽𝑃,𝑡 + 𝑓𝜖𝐶𝑁,𝑡−12,𝑂𝐶 or 𝑅_{𝐽𝑃,𝑡}𝑂𝐶 = 𝛼 + 𝛽𝜎_{𝐽𝑃,𝑡}2 + 𝛿𝐷_{𝐽𝑃,𝑡}+ 𝜙𝑅_{𝑈𝑆,𝑡−1}𝑂𝐶 + 𝛾𝜖_{𝐽𝑃,𝑡−1}+ 𝜖_{𝐽𝑃,𝑡} 𝜎_{𝐽𝑃,𝑡}2 _{= 𝑎 + 𝑏𝜎} 𝐽𝑃,𝑡−12 + 𝑐𝜖𝐽𝑃,𝑡−12 + 𝑑𝐷𝐽𝑃,𝑡 + 𝑓𝜖𝑈𝑆,𝑡−12,𝑂𝐶

For the Chinese stock market: 𝑅_{𝐶𝑁,𝑡}𝑂𝐶 _{= 𝛼 + 𝛽𝜎} 𝐶𝑁,𝑡2 + 𝛿𝐷𝐶𝑁,𝑡+ 𝜙𝑅𝐽𝑃,𝑡−1+ 𝛾𝜖𝐶𝑁,𝑡−1+ 𝜖𝐶𝑁,𝑡 𝜎_{𝐶𝑁,𝑡}2 _{= 𝑎 + 𝑏𝜎} 𝐶𝑁,𝑡−12 + 𝑐𝜖𝐶𝑁,𝑡−12 + 𝑑𝐷𝐶𝑁,𝑡+ 𝑓𝜖𝐽𝑃,𝑡−12,𝑂𝐶 or 𝑅_{𝐶𝑁,𝑡}𝑂𝐶 = 𝛼 + 𝛽𝜎_{𝐶𝑁,𝑡}2 + 𝛿𝐷𝐶𝑁,𝑡+ 𝜙𝑅𝑈𝑆,𝑡−1+ 𝛾𝜖𝐶𝑁,𝑡−1+ 𝜖𝐶𝑁,𝑡 𝜎_{𝐶𝑁,𝑡}2 _{= 𝑎 + 𝑏𝜎} 𝐶𝑁,𝑡−12 + 𝑐𝜖𝐶𝑁,𝑡−12 + 𝑑𝐷𝐶𝑁,𝑡+ 𝑓𝜖𝑈𝑆,𝑡−12,𝑂𝐶

For the US stock market: 𝑅_{𝑈𝑆,𝑡}𝑂𝐶 _{= 𝛼 + 𝛽𝜎} 𝑈𝑆,𝑡2 + 𝛿𝐷𝑈𝑆,𝑡+ 𝜙𝑅𝐶𝑁,𝑡𝑂𝐶 + 𝛾𝜖𝑈𝑆,𝑡−1+ 𝜖𝑈𝑆,𝑡 𝜎_{𝑈𝑆,𝑡}2 _{= 𝑎 + 𝑏𝜎} 𝑈𝑆,𝑡−12 + 𝑐𝜖𝑈𝑆,𝑡−12 + 𝑑𝐷𝑈𝑆,𝑡+ 𝑓𝜖𝐶𝑁,𝑡2,𝑂𝐶 or 𝑅_{𝑈𝑆,𝑡}𝑂𝐶 = 𝛼 + 𝛽𝜎_{𝑈𝑆,𝑡}2 + 𝛿𝐷_{𝑈𝑆,𝑡}+ 𝜙𝑅_{𝐽𝑃,𝑡}𝑂𝐶 + 𝛾𝜖_{𝑈𝑆,𝑡−1}+ 𝜖_{𝑈𝑆,𝑡} 𝜎_{𝑈𝑆,𝑡}2 _{= 𝑎 + 𝑏𝜎} 𝑈𝑆,𝑡−12 + 𝑐𝜖𝑈𝑆,𝑡−12 + 𝑑𝐷𝑈𝑆,𝑡+ 𝑓𝜖𝐽𝑃,𝑡2,𝑂𝐶

(17)

domestic changes (both financial or non-financial) happened in this period. In other words, the open-close period contains more information than the close-open one. Therefore, I expect that there are potential volatility spillover effects from the previously open foreign markets’ open-close return to the domestic stock market’s close-open conditional variance. Since the domestic close-open period contains less information, and acquires information mainly from foreign markets’ open-close period, the overlap of two periods is acceptable. In order to do the test, I rerun the Model II with local close-to-open return. Hence, the Model IV will be:

𝑅_𝑖.𝑡𝐶𝑂 = 𝛼 + 𝛽𝜎_𝑖,𝑡2 + 𝛿𝐷𝑖,𝑡+ 𝛾𝜖𝑖,𝑡−1+ 𝜖𝑖,𝑡

𝜎_𝑖,𝑡2 _{= 𝑎 + 𝑏𝜎}

𝑖,𝑡−12 + 𝑐𝜖𝑖,𝑡−12 + 𝑑𝐷𝑖,𝑡+ 𝑓𝜖𝑗,𝑡2,𝑂𝐶

where 𝑅_𝑖,𝑡𝐶𝑂 is the domestic close-open return, 𝜎𝑡2is the conditional variance of 𝑅𝑡,

and 𝜖_𝑗,𝑡2,𝑂𝐶 is the most recent squared residual derived from Model I applied to the open-close return of the previously open foreign market.

For the Japanese stock market: 𝑅_{𝐽𝑃,𝑡}𝐶𝑂 _{= 𝛼 + 𝛽𝜎}

𝐽𝑃,𝑡2 + 𝛿𝐷𝐽𝑃,𝑡+ 𝛾𝜖𝐽𝑃,𝑡−1+ 𝜖𝐽𝑃,𝑡

𝜎_{𝐽𝑃,𝑡}2 _{= 𝑎 + 𝑏𝜎}

𝐽𝑃,𝑡−12 + 𝑐𝜖𝐽𝑃,𝑡−12 + 𝑑𝐷𝐽𝑃,𝑡 + 𝑓𝜖𝐶𝑁,𝑡−12,𝑂𝐶

or 𝜎_{𝐽𝑃,𝑡}2 = 𝑎 + 𝑏𝜎_{𝐽𝑃,𝑡−1}2 + 𝑐𝜖_{𝐽𝑃,𝑡−1}2 + 𝑑𝐷_{𝐽𝑃,𝑡}+ 𝑓𝜖_{𝑈𝑆,𝑡−1}2,𝑂𝐶 For the Chinese stock market:

𝑅_{𝐶𝑁,𝑡}𝐶𝑂 _{= 𝛼 + 𝛽𝜎}

𝐶𝑁,𝑡2 + 𝛿𝐷𝐶𝑁,𝑡+ 𝛾𝜖𝐶𝑁,𝑡−1+ 𝜖𝐶𝑁,𝑡

𝜎_{𝐶𝑁,𝑡}2 _{= 𝑎 + 𝑏𝜎}

𝐶𝑁,𝑡−12 + 𝑐𝜖𝐶𝑁,𝑡−12 + 𝑑𝐷𝐶𝑁,𝑡+ 𝑓𝜖𝐽𝑃,𝑡−12,𝑂𝐶

or 𝜎_{𝐶𝑁,𝑡}2 = 𝑎 + 𝑏𝜎_𝐶𝐶2 _,𝑡−1+ 𝑐𝜖_{𝐶𝑁,𝑡−1}2 + 𝑑𝐷_{𝐶𝑁,𝑡}+ 𝑓𝜖_{𝑈𝑆,𝑡−1}2,𝑂𝐶 For the US stock market:

𝑅_{𝑈𝑆,𝑡}𝐶𝑂 _{= 𝛼 + 𝛽𝜎}

(18)

𝜎_{𝑈𝑆,𝑡}2 _{= 𝑎 + 𝑏𝜎}

or 𝜎_{𝑈𝑆,𝑡}2 = 𝑎 + 𝑏𝜎_{𝑈𝑆,𝑡−1}2 + 𝑐𝜖_{𝑈𝑆,𝑡−1}2 + 𝑑𝐷_{𝑈𝑆,𝑡}+ 𝑓𝜖_{𝐽𝑃,𝑡}2,𝑂𝐶

Furthermore, I build the Model V to estimate the spillover effects on the conditional means from the previous opening foreign markets’ open-close return to domestic market’s close-open conditional mean return. Model V:

𝑅_𝑖,𝑡𝐶𝑂 _{= 𝛼 + 𝛽𝜎}

𝑡2+ 𝛿𝐷𝑡+ 𝜙𝑅𝑗,𝑡𝑂𝐶+ 𝛾𝜖𝑖,𝑡−1+ 𝜖𝑖,𝑡

𝜎_𝑖,𝑡2 = 𝑎 + 𝑏𝜎_{𝑖,𝑡−1}2 + 𝑐𝜖_{𝑖,𝑡−1}2 + 𝑑𝐷𝑖,𝑡+ 𝑓𝜖𝑗,𝑡𝑂𝐶

where 𝑅_𝑖,𝑡𝐶𝑂 denotes the open-close return of the previously open foreign market. For the Japanese stock market:

𝑅_{𝐽𝑃,𝑡}𝐶𝑂 _{= 𝛼 + 𝛽𝜎} 𝐽𝑃,𝑡2 + 𝛿𝐷𝐽𝑃,𝑡+ 𝜙𝑅𝐶𝑁,𝑡−1𝑂𝐶 + 𝛾𝜖𝐽𝑃,𝑡−1+ 𝜖𝐽𝑃,𝑡 𝜎_{𝐽𝑃,𝑡}2 _{= 𝑎 + 𝑏𝜎} 𝐽𝑃,𝑡−12 + 𝑐𝜖𝐽𝑃,𝑡−12 + 𝑑𝐷𝐽𝑃,𝑡 + 𝑓𝜖𝐶𝑁,𝑡−12,𝑂𝐶 or 𝑅_{𝐽𝑃,𝑡}𝐶𝑂 = 𝛼 + 𝛽𝜎_{𝐽𝑃,𝑡}2 + 𝛿𝐷_{𝐽𝑃,𝑡}+ 𝜙𝑅_{𝑈𝑆,𝑡−1}𝑂𝐶 + 𝛾𝜖_{𝐽𝑃,𝑡−1}+ 𝜖_{𝐽𝑃,𝑡} 𝜎_{𝐽𝑃,𝑡}2 _{= 𝑎 + 𝑏𝜎} 𝐽𝑃,𝑡−12 + 𝑐𝜖𝐽𝑃,𝑡−12 + 𝑑𝐷𝐽𝑃,𝑡 + 𝑓𝜖𝑈𝑆,𝑡−12,𝑂𝐶

For the Chinese stock market:

𝑅_{𝐶𝑁,𝑡}𝐶𝑂 = 𝛼 + 𝛽𝜎𝐶𝑁,𝑡2 + 𝛿𝐷𝐶𝑁,𝑡+ 𝜙𝑅𝐽𝑃,𝑡−1+ 𝛾𝜖𝐶𝑁,𝑡−1+ 𝜖𝐶𝑁,𝑡 𝜎_{𝐶𝑁,𝑡}2 _{= 𝑎 + 𝑏𝜎} 𝐶𝑁,𝑡−12 + 𝑐𝜖𝐶𝑁,𝑡−12 + 𝑑𝐷𝐶𝑁,𝑡+ 𝑓𝜖𝐽𝑃,𝑡−12,𝑂𝐶 or 𝑅_{𝐶𝑁,𝑡}𝐶𝑂 = 𝛼 + 𝛽𝜎_{𝐶𝑁,𝑡}2 + 𝛿𝐷_{𝐶𝑁,𝑡}+ 𝜙𝑅_{𝑈𝑆,𝑡−1}+ 𝛾𝜖_{𝐶𝑁,𝑡−1}+ 𝜖_{𝐶𝑁,𝑡} 𝜎_{𝐶𝑁,𝑡}2 _{= 𝑎 + 𝑏𝜎} 𝐶𝑁,𝑡−12 + 𝑐𝜖𝐶𝑁,𝑡−12 + 𝑑𝐷𝐶𝑁,𝑡+ 𝑓𝜖𝑈𝑆,𝑡−12,𝑂𝐶

For the US stock market: 𝑅_{𝑈𝑆,𝑡}𝐶𝑂 _{= 𝛼 + 𝛽𝜎}

(19)

𝜎_{𝑈𝑆,𝑡}2 _{= 𝑎 + 𝑏𝜎}

or 𝑅_{𝑈𝑆,𝑡}𝐶𝑂 = 𝛼 + 𝛽𝜎_{𝑈𝑆,𝑡}2 + 𝛿𝐷_{𝑈𝑆,𝑡}+ 𝜙𝑅_{𝐽𝑃,𝑡}𝑂𝐶 + 𝛾𝜖_{𝑈𝑆,𝑡−1}+ 𝜖_{𝑈𝑆,𝑡} 𝜎_{𝑈𝑆,𝑡}2 _{= 𝑎 + 𝑏𝜎}

𝑈𝑆,𝑡−12 + 𝑐𝜖𝑈𝑆,𝑡−12 + 𝑑𝐷𝑈𝑆,𝑡+ 𝑓𝜖𝐽𝑃,𝑡2,𝑂𝐶

In order to test whether exchange rates have influence on stock markets co-movement and risk spillover, stock indexes in each market will be valued in domestic currency, and then exchanged into USD using daily average exchange rates.

4. Data analysis

4.1 Sample collection

In this article, I study the daily open and close data from three stock markets: New York, Tokyo and Shanghai, over a ten-year period (from 1st August 2003 to 31st December 2013). There is a breaking point at 31th October 2006, after which the split share reform of the Chinese stock market was almost completed, and a large proportion of state-owned and legal person shares became tradable in the market. Daily data is used in this short-term study, since daily data can provide a more detailed reflection on co-movements among different stock markets (Bekaert et al, 2009). Although using weekly data may avoid the estimated serial correlations, which usually happens on the first and last day of the week, adding dummy variables could also solve this problem. The descriptive statistic part will present a more detailed solution. All data are collected from Datastream and Yahoo Finance.

(20)

components are reviewed once a year. With the long history and strong comparability, it has become one of the most reliable indices, and commonly used by the mass media and scholars.

In the Shanghai Stock Exchange, I use the data of class A shares and class B shares. The A shares (RMB Common Stocks), traded in RMB, are issued by companies registered in Mainland China, subscribed and traded by organizations or individuals within the territory (excluding Taiwan, Hong Kong and Macao investors). Overseas individual investors and foreign institutional investors cannot take part in A share transactions. However, since July 2003, QFIIs are allowed to participate in more businesses (including A-share investment) under the guidance of the China Securities Regulatory Commission (CSRC). The B shares (Domestically Listed Foreign Investment Shares), traded in foreign currency, were only open to foreign investors initially.3 After 19th February 2001, the CSRC permitted the domestic citizens to invest in B shares via the secondary market. Hence, during the whole research period, both domestic and foreign investors have the equal rights to purchase and sell A and B share stocks. There are mainly two reasons to study Chinese stock market in separate share types. First, among the circulating shares, A and B shares are both issued and traded in Mainland China. There are also other types of shares like N-share, S-share and H-share, which are the stocks issued and traded in NYSE, Singapore and Hong Kong respectively. Chinese individual investors are restricted to invest in those stocks. Since this paper aims to study the co-movement of domestic stock markets, I take A and B shares as the research objects. Second, it allows us to compare and contrast the different effects of different institutional settings. As the clamour for A and B shares’ combination grew louder, it is interesting to see whether there is difference between both shares’ performance.

4.2 Descriptive statistics

As the stock market performance may also have influence on the later stock returns, the test of autoregressive is necessary. Therefore, I begin with the examination of the

3

(21)

serial correlation of the close-to-close, close-to-open and open-to-close returns of the three stock indices for the three sub-periods. Appendix 1 in appendix presents the examination results.

For the close-to-close return series, we can observe a negative correlation at lags 2, 5 and positive correlation at lag 10 for the Shanghai A Shares and New York market in three sample periods. The significant level is much higher in the second period than other periods. For the close-to-open returns, we can find negative correlation at lag 2 and positive correlation at lag 10 for all three markets in each sub-period. The correlation of lag 5 and 9 maintains positive in New York market, but negative in A Shares and Tokyo markets. For the open-to-close returns, there are negative correlation at lag 1 in all three markets, negative correlation at lag 2 in Shanghai and New York, and positive correlations at lags 5 & 9 in Tokyo. Shanghai two shares also have positive correlations at lags 9 & 10 in the second and third period. The above evidence potentially reflects the “day of the week” effect, and need to be captured in the model. Following the approach of Hamao et al (1990), a moving–average parameter MA(1) and a dummy variable, which equals 1 on the trading day following a weekend or holiday in both the conditional mean and variance, could be conjunct with GARCH-M model and thus capture this potential effect.

(22)

same direction during the third period. What is more, the correlations of Nikkei225 and S&P500 between themselves are much larger than their correlations with A or B Shares.

Table 1. Correlation coefficient

5. Results

5.1 Spillover effects on the conditional variance of open-to-close returns

As introduced in methodology, Model I is built to test whether MA(1)-GARCH(1,1)-M model is suitable to reflect and capture the changes of stock market returns and volatilities. Appendix 2 shows the results of the initial estimation of the model for the open-to-close returns series in the Chinese, the Japanese, and the US stock markets. In general, there is no serious model misspecification, as most of the Ljung-Box values for the first 12 normalized residuals or residuals squared are significant at conventional levels. However, the Ljung-Box(12) for normalized residuals on value of 22.609 for the A Shares in second period indicates the autocorrelation of residuals in 5% significant level. While the Ljung-Box(12) for normalized squared residuals on value of 34.501 and 32.885 in the second and third period means that, there is conditional heterorscedasticity. For all three stock markets, all the coefficients of kurtosis for the normalized residuals are slightly larger than 3.

SHA O-C SHB O-C S&P O-C NK O-C

S&P O-C (-1) 0.038298 0.038421 NK O-C 0.126625 S&P O-C (-1) 0.046384 NK O-C (-1) 0.036477 0.028084 SHA O-C 0.003849 SHA O-C (-1) -0.003300 SHB O-C -0.002793 SHB O-C (-1) 0.001020

S&P O-C (-1) -0.026378 0.03578935 NK O-C 0.146697 S&P O-C (-1) 0.225237 NK O-C (-1) -0.148143 -0.1090891 SHA O-C 0.041520 SHA O-C (-1) -0.048948 SHB O-C 0.044747 SHB O-C (-1) -0.061221

S&P O-C (-1) 0.002629 0.04789673 NK O-C 0.137815 S&P O-C (-1) 0.337362 NK O-C (-1) -0.154841 -0.1309104 SHA O-C 0.015371 SHA O-C (-1) -0.076090 SHB O-C 0.022321 SHB O-C (-1) -0.109123 (-1) denotes one-period lag

Panel A: Sample period: August 1, 2003 -- October 31, 2006

Panel B: Sample period: November 1, 2006 -- December 31, 2013

(23)

Which means, the residual distributions are leptokurtic kurtosis, and shows a thin waist. It is noteworthy that the conditional variance do not has signify effect on the conditional mean in all three markets during the whole research period, except for B Shares in period one. Meanwhile, we can also observe that all the conditional variances are significantly influenced by its 1 period lag and the lag 1 squared error term.

(24)

Table 2. Volatility spillovers estimated from a GARCH model using open-to-close returns of the domestic market and one foreign market (in local currency)

Local country Source of influence Number of obs. Log-likelihood

Coeff. Prob. Coeff. Prob. Coeff Prob. Coeff. Prob. Coeff Prob. Coeff. Prob. Coeff Prob. Coeff. Prob. Coeff. Prob. Coeff Prob. -0.000690 0.5740 -0.006158 0.3143 -0.000027 0.9846 -0.012892 0.0000 -0.000841 0.7257 -0.008524 0.0173 -0.013292 0.0003 -0.000159 0.9388 -0.000838 0.6905 0.041852 0.6316 0.073880 0.6277 0.539518 0.2581 -0.003745 0.9825 0.924836 0.0000 0.071481 0.7333 0.600186 0.0128 0.923934 0.0005 0.084588 0.7973 0.189673 0.5718 -6.454450 0.6331 0.000527 0.4738 0.000293 0.8409 0.000719 0.3532 -0.001258 0.5266 0.000349 0.7587 -0.002634 0.0454 -0.001859 0.1516 0.000741 0.2920 0.000747 0.2912 0.000144 0.9490 -0.077660 0.0611 0.004658 0.9492 -0.081361 0.0509 0.003627 0.9615 -0.079899 0.0554 0.039462 0.3764 0.012587 0.7827 -0.044935 0.2903 -0.042400 0.3162 -0.007317 0.8492 a 0.000000 0.8594 0.000067 0.0050 -0.000001 0.4104 0.000121 0.0005 0.000004 0.1336 0.000033 0.0001 0.000042 0.0000 0.000002 0.2652 0.000004 0.1347 0.000028 0.0328 b 0.926211 0.0000 0.593639 0.0003 0.937812 0.0000 0.578169 0.0000 0.936354 0.0000 0.802993 0.0000 0.780329 0.0000 0.865784 0.0000 0.856254 0.0000 0.343373 0.2817 c 0.062585 0.0001 0.147554 0.0596 0.047145 0.0009 0.136210 0.0518 0.045710 0.0004 0.095788 0.0000 0.072896 0.0004 0.058697 0.0120 0.059399 0.0170 0.006088 0.6364 d 0.000006 0.4226 -0.000043 0.1563 0.000002 0.8038 -0.000091 0.0087 -0.000020 0.0498 -0.000016 0.3942 0.000008 0.6457 0.000002 0.7626 0.000003 0.6655 -0.000002 0.5775 -0.002629 0.4049 -0.000005 0.0000 0.000002 0.0464 -0.000016 0.0000 0.000003 0.1323 -0.000005 0.0809 -0.000012 0.0000 0.000001 0.3239 -0.000001 0.3478 0.000000 0.5983 Coefficient of skewness for normalized residuals Coefficient of kurtosis for normalized residuals Ljung-Box(12) for normalized residuals Ljung-Box(12) for normalized squared residuals

China SHS A China SHS B U.S

15.955 5.3593 13.134 12.189 67.582 11.62 13.64 7.0585 7.4562 8.9484 6.6485 9.3471 7.2721 7.5504 From U.S 0.134796 3.649924 4.209754 3.305326 4.115588 3.764271 7.638641 5.461496 2.939838 2.937275 3.806631 -0.346671 -0.173268 -0.28246 0.115575 0.213462 0.096597 0.060017 -0.101956 -0.11038 707 1966.729 From U.S 709 2006.832 2577.757 707 2068.407 709 2149.876 5.6657 34.633 10.38 8.9124 7.5941 From China SHS B 710 2577.882 7.4562 Panel A: Sample period: August 1, 2003 -- October 31, 2006

From China SHS B 707 2253.739 From Japan 707 2350.738 695 2324.855 From Japan From China SHS A From U.S

(25)

Table 2. Continue

Coeff. Prob. Coeff. Prob. Coeff Prob. Coeff. Prob. Coeff Prob. Coeff. Prob. Coeff Prob. Coeff. Prob. Coeff. Prob. Coeff Prob. -0.000746 0.3476 -0.002381 0.4235 -0.002113 0.6395 -0.001195 0.3454 -0.009093 0.0420 0.000578 0.6240 0.001446 0.2039 0.000273 0.6620 -0.002512 0.2770 0.000554 0.8517 0.106286 0.2314 0.186220 0.3633 0.138204 0.5932 0.146021 0.1081 0.499971 0.0154 0.066057 0.4148 0.006292 0.9351 0.051701 0.4377 0.199396 0.1922 0.012970 0.9309 -0.001707 0.0038 -0.000933 0.4085 -0.001594 0.3141 0.000308 0.7177 -0.000137 0.9310 -0.000569 0.5420 -0.000096 0.9142 -0.000434 0.4824 -0.000328 0.7679 -0.001574 0.3178 -0.080436 0.0037 0.003940 0.9274 0.004331 0.9397 -0.053831 0.0431 0.003066 0.9444 -0.008306 0.7867 0.008833 0.7624 -0.073731 0.0168 0.002758 0.9397 0.004501 0.9274 a 0.000004 0.0005 0.000075 0.0000 0.000119 0.0000 0.000000 0.8043 0.000209 0.0000 0.000005 0.2161 -0.000002 0.3750 0.000004 0.0005 0.000096 0.0000 0.000156 0.0000 b 0.843985 0.0000 0.589519 0.0000 0.590200 0.0000 0.948475 0.0000 0.561989 0.0000 0.766878 0.0000 0.814971 0.0000 0.889172 0.0000 0.577116 0.0000 0.590836 0.0000 c 0.121877 0.0000 0.147626 0.0003 0.148203 0.0007 0.043836 0.0000 0.131554 0.0001 0.186316 0.0000 0.134789 0.0000 0.099970 0.0000 0.148490 0.0000 0.148827 0.0000 d -0.000001 0.8238 -0.000048 0.0177 -0.000085 0.0000 0.000009 0.4058 -0.000128 0.0000 0.000081 0.0000 0.000019 0.1087 -0.000010 0.0282 -0.000145 0.0000 -0.000236 0.0000 0.000000 0.4375 -0.000004 0.0000 -0.000007 0.0654 0.000001 0.2612 -0.000023 0.0000 0.000002 0.3590 0.000019 0.0000 0.000000 0.8994 -0.000004 0.0000 -0.000005 0.0000 Coefficient of skewness for normalized residuals Coefficient of kurtosis for normalized residuals Ljung-Box(12) for normalized residuals Ljung-Box(12) for normalized squared residuals

Japan China SHS A China SHS B U.S

17.696 22.348 20.823 19.502 13.433 20.189 36.925 11.775 234.44 19.483 13.078 11.889 7.2084 334.12 -0.407435 4.384612 5.073768 20.78135 4.380618 13.31329 6.453515 6.273595 3.890661 15.15704 7.835783 14.051 52.503 27.52 -0.441901 -0.436608 0.593681 12.909 30.57 18.035 4875.959

From Japan From U.S 1502 1550 -0.3384 -0.364031 -1.790079 -0.214325 0.533723 -0.537992 4871.381 4583.764 1502 1502 1556 4657.518 4010.426 4142.67 From Japan 1502 4212.392 From U.S 1550 4075.139 1557 From China SHS A From China SHS B From Japan

1551 4590.965 From China SHS A From China SHS B From U.S

(26)

Coeff. Prob. Coeff. Prob. Coeff Prob. Coeff. Prob. Coeff Prob. Coeff. Prob. Coeff Prob. Coeff. Prob. Coeff. Prob. Coeff Prob. -0.002842 0.0990 -0.003013 0.0623 -0.001705 0.2336 -0.002178 0.6830 -0.001950 0.6781 0.001653 0.6170 0.001164 0.7417 -0.004203 0.0435 -0.004180 0.0449 -0.000805 0.6700 0.191001 0.1692 0.207889 0.1160 0.113593 0.3718 0.148774 0.5860 0.162566 0.4950 0.007525 0.9631 0.015675 0.9277 0.297072 0.0520 0.295881 0.0525 0.065540 0.6084 -0.000939 0.4335 -0.001345 0.2509 -0.001150 0.3707 0.001108 0.6196 0.000501 0.8165 -0.001147 0.5932 -0.001856 0.3719 0.000128 0.9392 0.000127 0.9400 0.000228 0.8920 -0.112219 0.0431 -0.111874 0.0415 -0.104168 0.0405 -0.086516 0.0635 -0.089047 0.0543 0.008271 0.8791 0.020813 0.6452 -0.122009 0.0141 -0.120341 0.0152 -0.113151 0.0355 a 0.000013 0.0003 0.000005 0.2908 -0.000007 0.0125 0.000017 0.4295 -0.000010 0.3655 0.000036 0.0392 -0.000012 0.2838 -0.000001 0.8205 0.000000 0.9529 0.000007 0.2046 b 0.812651 0.0000 0.848057 0.0000 0.905092 0.0000 0.831637 0.0000 0.901527 0.0000 0.666581 0.0000 0.867952 0.0000 0.885475 0.0000 0.885343 0.0000 0.870589 0.0000 c 0.139258 0.0001 0.107798 0.0000 0.066868 0.0000 0.082804 0.0084 0.053774 0.0098 0.258413 0.0000 0.076983 0.0000 0.091238 0.0007 0.092591 0.0008 0.109833 0.0000 d -0.000036 0.0052 -0.000020 0.1911 0.000017 0.1644 0.000079 0.1704 0.000087 0.0698 0.000121 0.0447 0.000011 0.8332 0.000030 0.2713 0.000028 0.3027 0.000005 0.8545 0.000004 0.1181 0.000008 0.0029 0.000009 0.0000 0.000006 0.4702 0.000014 0.0526 0.000000 0.9971 0.000039 0.0000 0.000002 0.6434 0.000000 0.9674 -0.000002 0.6460 Coefficient of skewness for normalized residuals Coefficient of kurtosis for normalized residuals Ljung-Box(12) for normalized residuals Ljung-Box(12) for normalized squared residuals

6.0045 13.316 16.702 15.049 9.3408 10.076 4.3923 20.445 13.219 13.521 26.268 16.093 16.679 9.9183 11.335 8.5885 16.135 10.961 8.3036 8.5766 -0.388445 6.749077 6.460975 4.564689 3.302024 3.508917 4.918859 5.334874 3.135885 3.133311 3.401457 -0.7983 -0.75241 -0.32309 -0.359431 -0.346678 -0.261451 -0.364579 -0.314635 -0.312535 From China SHS B 542 1526.248 From China SHS A 542 1526.379 From Japan From U.S

527 541 1262.595 From China SHS A 527 From U.S 541 1383.213 1301.881 From U.S 537 1596.049 From Japan 527 1305.989 From China SHS B 527 1554.335

Panel C: Sample period: July 1, 2007 -- December 31, 2009

(27)

From the results in table 2, we can find that the sensitivity of A Shares to other stock markets’ volatility do increases after the Split Share reform, but it is still less active than B Shares. Appendix 3 presents the results in USD. No significant difference can be observed compared with table 2.

5.2 Spillover effects on the conditional means of open-to-close returns

(28)

Table 3. Mean and volatility spillovers estimated from a GARCH model using open-to-close returns (in local currency)

Local country Source of influence

Number of obs. 710

Log-likelihood 2577.918

Coeff. Prob. Coeff. Prob. Coeff Prob. Coeff. Prob. Coeff Prob. Coeff. Prob. Coeff Prob. Coeff. Prob. Coeff. Prob. Coeff Prob. -0.000688 0.5753 -0.006154 0.3283 -0.000044 0.9749 -0.001946 0.5419 -0.001122 0.6455 -0.008024 0.0215 -0.013162 0.0003 -0.000132 0.9489 -0.000890 0.6720 0.005391 0.0003 0.073484 0.6297 0.539224 0.2712 -0.003106 0.9857 0.181046 0.4895 0.091540 0.6685 0.569709 0.0161 0.912766 0.0005 0.080409 0.8061 0.197668 0.5550 -0.820412 0.0003 0.000532 0.4698 0.000292 0.8432 0.000740 0.3418 -0.000192 0.8658 0.000422 0.7121 -0.002752 0.0351 -0.001735 0.1745 0.000741 0.2903 0.000753 0.2878 0.000697 0.3451 0.002786 0.9057 -0.000897 0.9805 0.023072 0.6559 0.045229 0.3358 0.086621 0.2087 0.089105 0.1031 0.103559 0.1756 0.007146 0.7089 0.004146 0.7791 0.097855 0.0002 -0.078104 0.0624 0.004588 0.9499 -0.084044 0.0460 -0.061301 0.1556 -0.080474 0.0552 0.039034 0.3831 0.017500 0.7014 -0.045755 0.2833 -0.043151 0.3081 -0.023159 0.5414 a 0.000000 0.8598 0.000068 0.0054 -0.000001 0.4192 0.000016 0.0018 0.000004 0.1019 0.000036 0.0000 0.000042 0.0000 0.000002 0.2606 0.000004 0.1299 0.000039 0.0162 b 0.926329 0.0000 0.592490 0.0003 0.938353 0.0000 0.938307 0.0000 0.932697 0.0000 0.795784 0.0000 0.779004 0.0000 0.866528 0.0000 0.856994 0.0000 0.122590 0.7618 c 0.062522 0.0001 0.147072 0.0784 0.046832 0.0010 0.020737 0.0784 0.048454 0.0003 0.100652 0.0000 0.073566 0.0004 0.058791 0.0116 0.059126 0.0171 0.000877 0.9735 d 0.000006 0.4290 -0.000043 0.1588 0.000002 0.7970 -0.000033 0.0268 -0.000021 0.0466 -0.000023 0.2264 0.000006 0.7281 0.000002 0.7929 0.000003 0.6893 -0.000003 0.4871 -0.002596 0.4128 -0.000005 0.0000 0.000002 0.0580 -0.000004 0.0217 0.000003 0.1793 -0.000006 0.0326 -0.000012 0.0000 0.000001 0.3168 -0.000001 0.3251 -0.000003 0.0000 Coefficient of skewness for normalized residuals Coefficient of kurtosis for normalized residuals Ljung-Box(12) for normalized residuals Ljung-Box(12) for normalized squared residuals

From China SHS B 5.7036 34.759 10.419 6.664 7.2344 16.611 5.4316 12.819 11.931 11.653 13.58 7.2136 6.0567 9.0456 5.7655 9.9448 7.3405 7.5542 3.646438 4.202086 3.296318 4.162855 3.752756 7.716414 5.437921 2.946792 2.941516 -0.346618 -0.171672 -0.280366 0.151291 0.223469 0.139267 0.047817 -0.100728 -0.109127 707 2253.896 709 2127.814 2150.72 -0.113433 3.031598 7.6367 18.898 Panel A: Sample period: August 1, 2003 -- October 31, 2006

From China SHS A 707 2350.745 From U.S 695 2324.963

From China SHS A From Japan

710 696

2577.82 2529.882

From Japan From U.S 707

From Japan From U.S

707 709

(29)

Coeff. Prob. Coeff. Prob. Coeff Prob. Coeff. Prob. Coeff Prob. Coeff. Prob. Coeff Prob. Coeff. Prob. Coeff. Prob. Coeff Prob. -0.000730 0.3608 -0.000693 0.3854 -0.000917 0.2448 -0.000898 0.4807 -0.008984 0.0572 0.001701 0.8321 0.001379 0.2252 -0.002547 0.6296 -0.002487 0.2889 0.000207 0.7491 0.105339 0.2379 0.102847 0.2499 0.100943 0.2424 0.127736 0.1653 0.493795 0.0236 -0.010431 0.9720 0.008130 0.9162 0.196918 0.4841 0.195651 0.2068 0.047309 0.4758 -0.001685 0.0043 -0.001679 0.0043 -0.001592 0.0057 0.000130 0.8766 -0.000135 0.9316 -0.001195 0.5787 -0.000090 0.9198 -0.000974 0.5807 -0.000279 0.8011 -0.000485 0.4735 -0.013647 0.3245 -0.018766 0.1094 0.086213 0.0001 -0.149359 0.0000 -0.032154 0.4604 -0.189425 0.0118 0.095440 0.0182 0.032272 0.3650 0.027271 0.0595 0.133616 0.0000 -0.075797 0.0067 -0.076366 0.0060 -0.125332 0.0000 -0.035911 0.1846 0.002923 0.9465 0.005952 0.9015 0.004715 0.8719 0.003720 0.9477 0.002239 0.9518 -0.075351 0.0205 a 0.000004 0.2379 0.000004 0.0017 0.000003 0.0122 0.000000 0.8212 0.000209 0.0000 0.000296 0.0000 -0.000002 0.4078 0.000139 0.0001 0.000096 0.0000 0.000003 0.0044 b 0.844334 0.0043 0.845138 0.0000 0.841765 0.0000 0.951862 0.0000 0.558933 0.0000 0.556747 0.0000 0.816110 0.0000 0.583476 0.0000 0.574194 0.0000 0.869319 0.0000 c 0.121268 0.3608 0.120931 0.0000 0.121163 0.0000 0.041676 0.0000 0.129065 0.0001 0.132957 0.0001 0.133227 0.0000 0.145864 0.0053 0.148167 0.0000 0.111392 0.0000 d -0.000001 0.3245 -0.000001 0.7863 -0.000008 0.0983 0.000005 0.6603 -0.000125 0.0000 -0.000162 0.0084 0.000018 0.1210 -0.000081 0.0105 -0.000145 0.0000 -0.000001 0.8076 0.000000 0.0067 0.000000 0.7955 0.000003 0.0017 0.000000 0.6579 -0.000023 0.0000 -0.000020 0.0000 0.000019 0.0000 -0.000010 0.0017 -0.000004 0.0000 0.000000 0.4407 Coefficient of skewness for normalized residuals Coefficient of kurtosis for normalized residuals Ljung-Box(12) for normalized residuals Ljung-Box(12) for normalized squared residuals 11.561 10.976 3.1086 16.809 2.023 139.15 12.843 11.616 19.207 33.866 23.38 -0.584386 4.425297 4.511406 7.56715 4.315307 46.77612 6.396293 6.329925 5.887858 14.75 14.687

From U.S 1502 1556 4871.827 4997.147 7.434065 4.638042 -0.34578 -0.356951 -0.654403 -0.190883 2.421892 -0.518695 -0.426934 From China SHS B 1502 4872.55 1502 1550 3748.773 4145.494 1551 1557 4465.851 4882.332 From China SHS B 1551 4620.215 -0.253113 -0.021628 28.567 18.213 370.64 1000.34

From China SHS A From Japan From Japan 16.504 23.212 From U.S 14.517 4220.646 4021.328 From China SHS A

Panel B: Sample period: November 1, 2006 -- December 31, 2013 From Japan From U.S

(30)

Coeff. Prob. Coeff. Prob. Coeff Prob. Coeff. Prob. Coeff Prob. Coeff. Prob. Coeff Prob. Coeff. Prob. Coeff. Prob. Coeff Prob. -0.002840 0.1010 -0.003036 0.0628 -0.002106 0.1474 -0.001189 0.8351 -0.001246 0.8081 0.001632 0.6362 0.001456 0.6792 -0.004149 0.0471 -0.004100 0.0479 -0.001190 0.5190 0.190607 0.1723 0.210207 0.1158 0.146926 0.2581 0.093998 0.7483 0.109850 0.6749 0.007267 0.9657 0.003225 0.9851 0.292907 0.0569 0.288803 0.0568 0.089246 0.4779 -0.000914 0.4461 -0.001325 0.2578 -0.001092 0.4049 0.001149 0.5971 0.000456 0.8321 -0.001164 0.5822 -0.001717 0.4067 0.000140 0.9338 0.000229 0.8936 0.000362 0.8300 -0.009148 0.7284 -0.006596 0.7930 0.210642 0.0000 -0.157134 0.0134 0.035600 0.5809 -0.097652 0.1307 0.102696 0.1776 0.008863 0.7840 0.012423 0.5949 0.174751 0.0014 -0.109198 0.0493 -0.110921 0.0433 -0.146800 0.0031 -0.071810 0.1296 -0.069090 0.1427 0.017082 0.7544 0.015427 0.7341 -0.122763 0.0139 -0.120810 0.0151 -0.160420 0.0040 a 0.000013 0.0003 0.000005 0.2888 -0.000007 0.0057 0.000010 0.5918 -0.000008 0.4772 0.000035 0.0280 -0.000010 0.3776 -0.000001 0.8225 0.000000 0.9331 0.000005 0.3430 b 0.813341 0.0000 0.847888 0.0000 0.922524 0.0000 0.863907 0.0000 0.905940 0.0000 0.679154 0.0000 0.862843 0.0000 0.885465 0.0000 0.885451 0.0000 0.868543 0.0000 c 0.137740 0.0001 0.107359 0.0000 0.049504 0.0004 0.065299 0.0110 0.041956 0.0295 0.243033 0.0000 0.081057 0.0000 0.091442 0.0007 0.093118 0.0008 0.107689 0.0001 d -0.000037 0.0051 -0.000020 0.1996 0.000015 0.2073 0.000065 0.2443 0.000051 0.2615 0.000113 0.0541 0.000003 0.9518 0.000030 0.2826 0.000028 0.3147 0.000004 0.8763 0.000004 0.1101 0.000008 0.0032 0.000009 0.0000 0.000008 0.2663 0.000021 0.0052 0.000001 0.9580 0.000039 0.0000 0.000002 0.6413 0.000000 0.9815 0.000001 0.8347 Coefficient of skewness for normalized residuals Coefficient of kurtosis for normalized residuals Ljung-Box(12) for normalized residuals Ljung-Box(12) for normalized squared residuals 5.412 13.211 16.541 14.213 9.8342 9.3856 4.1509 20.269 13.301 13.297 28.812 16.137 16.747 11.087 8.3016 7.4603 15.201 10.179 8.3968 8.8082 -0.369503 6.725203 6.442108 4.497984 3.361506 3.409043 4.933505 5.335601 3.136627 3.135799 3.443044 -0.799427 -0.752454 -0.234337 -0.373929 -0.4195 -0.274486 -0.3366 -0.312841 -0.305645

From Japan From U.S

527 541

Japan China SHS A China SHS B

From China SHS A From U.S

527 537 542

1526.358 From China SHS A

U.S From China SHS B

Panel C: Sample period: July 1, 2007 -- December 31, 2009

From Japan

From Japan From U.S From China SHS B

(31)

5.3 Spillover effects on the conditional variance of close-to-open returns

(32)

Table 4. Volatility spillovers estimated from a GARCH model using close-to-open returns of the domestic market and one foreign market (in local currency)

Coeff. Prob. Coeff. Prob. Coeff Prob. Coeff. Prob. Coeff Prob. Coeff. Prob. Coeff Prob. Coeff. Prob. Coeff. Prob. Coeff Prob. 0.001829 0.2127 0.001271 0.3630 0.001719 0.0135 -0.000417 0.2589 -0.000226 0.3671 0.001301 0.0000 0.000064 0.8699 -0.000142 0.5602 -0.000369 0.3920 0.000054 0.8492 -0.194828 0.4821 -0.086254 0.7445 -0.201322 0.1692 0.106740 0.4729 0.046123 0.7458 -0.577225 0.0000 -0.029802 0.8669 0.233659 0.5620 0.590056 0.3884 -0.117077 0.8223 -0.000365 0.4602 -0.000413 0.4058 -0.000017 0.9743 -0.000455 0.0932 -0.000294 0.2676 -0.000185 0.3085 -0.000398 0.0695 0.000056 0.4218 0.000092 0.3043 -0.000030 0.7747 0.033131 0.4091 0.033785 0.4061 -0.019370 0.5895 0.005979 0.9343 0.059467 0.2746 0.135608 0.0000 0.198342 0.0000 0.018509 0.7267 -0.003331 0.9458 0.027750 0.5104 a 0.000002 0.2596 0.000003 0.1729 0.000009 0.0001 0.000002 0.0000 0.000000 0.0008 0.000000 0.0000 0.000000 0.1828 0.000000 0.0000 0.000000 0.0000 0.000000 0.0000 b 0.780407 0.0000 0.774338 0.0000 0.088298 0.2744 0.599392 0.0000 0.372439 0.0000 1.004494 0.0000 0.807430 0.0000 0.079733 0.0981 0.508990 0.0000 0.243305 0.0039 c 0.109989 0.0095 0.116873 0.0087 0.091069 0.0790 0.148238 0.0001 0.152536 0.0001 -0.006379 0.0002 0.123945 0.0000 0.127578 0.0001 0.037340 0.0090 0.048543 0.0037 d 0.000003 0.3879 0.000004 0.3135 0.000004 0.2899 0.000008 0.0000 0.000002 0.0000 0.000001 0.0000 0.000001 0.0068 0.000000 0.0204 0.000000 0.0000 0.000000 0.0000 0.002575 0.5276 0.000000 0.6855 0.000015 0.0000 0.000000 0.0014 0.000002 0.0000 0.000000 0.8141 0.000001 0.0000 0.000000 0.0000 0.000000 0.0293 0.000000 0.0018 Coefficient of skewness for normalized residuals Coefficient of kurtosis for normalized residuals Ljung-Box(12) for normalized residuals Ljung-Box(12) for normalized squared residuals

13.214 14.94 38.11 1.1669 3.589 16.303 3.7997 7.6398 6.0191 17.153 16.909 7.6702 10.681 8.8076 57.455 78.574 18.765 17.273 2.544266 2.554264 2.463703 22.85216 17.62844 9.476922 8.985475 16.97305 19.54196 -0.142417 -0.156033 -0.050043 0.730868 0.797577 0.353091 -0.172681 1.79589 2.042959 4226.195 0.169811 9.998813 6.2401 19.137 Panel A: Sample period: August 1, 2003 -- October 31, 2006

From China SHS A From Japan From China SHS B

707 2690.147

From Japan From U.S

(33)

Coeff. Prob. Coeff. Prob. Coeff Prob. Coeff. Prob. Coeff Prob. Coeff. Prob. Coeff Prob. Coeff. Prob. Coeff. Prob. Coeff Prob. 0.000555 0.6007 0.000380 0.7409 0.002451 0.0016 -0.000836 0.0037 -0.000592 0.0490 -0.000396 0.1164 -0.000073 0.7607 0.001257 0.1710 0.000695 0.0283 0.000070 0.0007 -0.041091 0.7595 -0.019274 0.8941 -0.282918 0.0108 -0.021215 0.7267 -0.045328 0.4595 -0.076301 0.1954 -0.191129 0.0011 -0.348567 0.0460 -0.348567 0.0000 -0.239207 0.0000 0.000555 0.2447 0.000520 0.2734 0.001211 0.0410 0.000022 0.9367 0.000460 0.1154 -0.000105 0.5990 0.000539 0.0209 0.000222 0.6613 0.000189 0.4638 -0.000042 0.1771 -0.029553 0.2913 -0.029185 0.2935 -0.072012 0.0017 0.039138 0.1400 0.022161 0.4614 0.122395 0.0001 0.087267 0.0032 0.004998 0.9267 0.004990 0.8993 -0.050120 0.0992 a 0.000003 0.0118 0.000002 0.0688 0.000027 0.0000 -0.000001 0.0008 0.000000 0.0005 0.000000 0.4442 -0.000001 0.0000 0.000007 0.0000 0.000003 0.0000 0.000000 0.2225 b 0.905194 0.0000 0.914663 0.0000 -0.008556 0.7475 0.895472 0.0000 0.709676 0.0000 0.748457 0.0000 0.701253 0.0000 0.599927 0.0000 0.600048 0.0000 0.767380 0.0000 c 0.071655 0.0001 0.066199 0.0001 0.086120 0.0046 0.105165 0.0000 0.237529 0.0000 0.268981 0.0000 0.208362 0.0000 0.150017 0.0000 0.150366 0.0000 0.250097 0.0000 d -0.000001 0.8392 -0.000003 0.6005 0.000021 0.0007 0.000004 0.0013 0.000006 0.0000 0.000001 0.2974 0.000005 0.0000 -0.000003 0.0064 -0.000003 0.0000 0.000000 0.0000 -0.000001 0.0511 0.000000 0.4463 0.000036 0.0000 0.000000 0.0000 0.000003 0.0000 0.000002 0.0000 0.000003 0.0000 -0.000001 0.0000 0.000000 0.0000 0.000000 0.0000 Coefficient of skewness for normalized residuals Coefficient of kurtosis for normalized residuals Ljung-Box(12) for normalized residuals Ljung-Box(12) for normalized squared residuals

80.34 62.003 5.2768 17.344 16.474 231.12 6.0535 3.7127 4.2807 5.0672 271.64 261.77 2.399174 2.390448 2.443953 6.633123 9.132049 14.9155 14.11709 38.18219 32.97139 From China SHS B 1550 7397.53 4.5549 1.066452 0.020963 18.11806 1502 1550 5792.11 6039.502 -0.571364 0.471812 8677.357 1556 -0.264063 13.202 14.746 1550 5625.142 5805.662 1502 5106.606 -0.180835 -0.190144 0.009076 -0.731999 0.041839 9.792 10.146 18.464 13.412 15.703

From China SHS A From Japan From Japan

From China SHS B From Japan From U.S

From China SHS A From U.S From U.S

1550 6900.796

1502 1556

5108.18 5390.501

(34)

Coeff. Prob. Coeff. Prob. Coeff Prob. Coeff. Prob. Coeff Prob. Coeff. Prob. Coeff Prob. Coeff. Prob. Coeff. Prob. Coeff Prob. 0.000608 0.8424 0.005538 0.5995 0.002976 0.0486 0.002141 0.2798 0.001122 0.3527 0.000497 0.7301 -0.000482 0.5284 0.001800 0.3065 0.000910 0.0042 0.000341 0.0000 -0.096683 0.7932 -0.623032 0.4765 -0.384067 0.0634 -0.426918 0.0921 -0.276221 0.1177 -0.217638 0.3327 -0.056961 0.6936 -0.440767 0.1292 -0.440715 0.0000 -0.290506 0.0000 -0.000303 0.7345 0.000299 0.8850 0.000057 0.9548 0.000722 0.4470 0.000080 0.9412 0.000938 0.2211 0.000127 0.8698 -0.000710 0.4454 0.000045 0.8578 -0.000229 0.0093 0.000571 0.9903 0.004999 0.9561 -0.039283 0.3222 0.001260 0.9795 -0.031491 0.5349 0.114359 0.0779 0.103902 0.0521 0.004929 0.9449 0.003988 0.9177 -0.074247 0.1539 a 0.000004 0.2684 0.000058 0.1729 0.000028 0.0003 0.000001 0.7674 0.000009 0.0014 0.000002 0.3306 0.000007 0.0000 0.000013 0.0001 0.000002 0.0000 0.000000 0.0000 b 0.897281 0.0000 0.596819 0.0782 0.027795 0.7987 0.872427 0.0000 0.380335 0.0000 0.728567 0.0000 0.183111 0.0000 0.599356 0.0000 0.605560 0.0000 0.746499 0.0000 c 0.050193 0.2737 0.148557 0.4693 0.104864 0.1303 0.074914 0.0000 0.143412 0.0001 0.111691 0.0001 0.228404 0.0000 0.150165 0.0008 0.163538 0.0000 0.280069 0.0000 d -0.000010 0.4857 -0.000041 0.0000 0.000010 0.3457 0.000006 0.3956 0.000022 0.0108 0.000020 0.0000 0.000014 0.0073 -0.000013 0.0016 -0.000002 0.0000 -0.000001 0.0000 0.000001 0.5654 -0.000004 0.2913 0.000033 0.0088 0.000002 0.0473 0.000021 0.0000 0.000004 0.0000 0.000019 0.0000 -0.000001 0.0020 0.000000 0.0000 0.000000 0.0003 Coefficient of skewness for normalized residuals Coefficient of kurtosis for normalized residuals Ljung-Box(12) for normalized residuals Ljung-Box(12) for normalized squared residuals

14.309 16.327 10.608 55.453 8.4818 9.6238 5.7176 19.956 101.34 147.09 10.785 3.3585 4.0593 5.0294 15.028 21.035 15.877 15.707 29.095 19.769 0.350354 1.930321 3.03237 1.992195 4.380542 4.531066 10.4692 6.11985 14.39162 10.36953 6.581434 -0.069853 -0.272448 0.02166 -0.486678 -0.327545 -0.823136 -0.469138 -0.944012 0.031314 From China SHS B 541 2435.974 2602.611 537 From Japan From China SHS A 541 2206.928 From Japan From U.S

527 541

1776.557 1858.694

From Japan From U.S

527 541

1871.522 2003.545 From China SHS A From U.S

527 537

1780.834 1844.114

From China SHS B 527 1714.327

(35)

5.4 Spillover effects on the conditional means of close-to-open returns

(36)

Table 5. Mean and volatility spillovers estimated from a GARCH model using close-to-open returns (in local currency)

Coeff. Prob. Coeff. Prob. Coeff Prob. Coeff. Prob. Coeff Prob. Coeff. Prob. Coeff Prob. Coeff. Prob. Coeff. Prob. Coeff Prob. 0.001838 0.2106 0.001387 0.3223 0.000513 0.5731 -0.000425 0.8144 0.000317 0.0100 0.000345 0.2318 0.000058 0.8822 -0.000107 0.6741 -0.000340 0.4353 0.000017 0.9570 -0.196453 0.4786 -0.110049 0.6780 -0.013441 0.9534 0.089654 0.8534 -0.261208 0.0000 -0.141602 0.2732 -0.022439 0.8991 0.175296 0.6751 0.545098 0.4299 -0.049846 0.9318 -0.000365 0.4615 -0.000388 0.4303 0.000181 0.6219 -0.000352 0.4131 -0.000281 0.0964 -0.000172 0.3766 -0.000427 0.0514 0.000055 0.4360 0.000098 0.2718 -0.000026 0.8263 0.003780 0.8193 -0.012133 0.3185 0.565164 0.0000 0.005797 0.7878 0.008456 0.4401 -0.004536 0.6370 -0.019357 0.2293 0.004638 0.0360 0.002697 0.1483 0.006805 0.0107 0.032809 0.4144 0.032749 0.4201 0.028290 0.5094 0.005007 0.9662 -0.037175 0.4288 0.163380 0.0001 0.198839 0.0000 0.018671 0.7295 -0.005449 0.9122 0.023837 0.5722 a 0.000002 0.2604 0.000003 0.1652 0.000001 0.1725 0.000006 0.0003 0.000000 0.3166 0.000000 0.0000 0.000000 0.1415 0.000000 0.0000 0.000000 0.0000 0.000000 0.0000 b 0.779122 0.0000 0.780586 0.0000 0.830237 0.0000 0.599871 0.0000 0.959683 0.0000 0.937161 0.0000 0.803506 0.0000 0.085153 0.0903 0.537263 0.0000 0.201086 0.0136 c 0.110235 0.0095 0.115609 0.0086 0.077319 0.0051 0.149944 0.0678 0.029842 0.0000 0.040672 0.0000 0.125535 0.0000 0.127202 0.0001 0.034159 0.0079 0.056093 0.0031 d 0.000004 0.3808 0.000003 0.4087 -0.000003 0.0697 0.000002 0.0840 0.000000 0.0000 -0.000001 0.0000 0.000001 0.0033 0.000000 0.0051 0.000000 0.0000 0.000000 0.0000 0.002616 0.5265 0.000000 0.7351 0.000001 0.0036 -0.000001 0.0000 0.000000 0.0000 0.000000 0.0083 0.000001 0.0000 0.000000 0.0000 0.000000 0.0367 0.000000 0.0608 Coefficient of skewness for normalized residuals Coefficient of kurtosis for normalized residuals Ljung-Box(12) for normalized residuals Ljung-Box(12) for normalized squared residuals U.S -0.14243 2.547986 17.22 12.981 -0.152581 2.53238 16.616 15.113 0.036616 3.422624 8.4713 8.7516 1.154403 1.898233 0.50195 -0.15567 25.85851 21.2242 10.98529 8.766912 10.258 24.617 96.724 From China SHS B 709 4227.863 2.051565 19.30214 16.893 6.098 From China SHS A Japan China SHS A

From Japan From U.S

707 709 3227.431 3246.268 China SHS B From China SHS B 707 2690.657 From U.S From China SHS A From U.S From Japan

77.161 1.0141 1.2423 3.1738 3.8702 4272.755 707 695 707 709 709 1.835493 0.164459 17.25615 9.816284 17.517 6.5371 7.0108 18.645

Panel A: Sample period: August 1, 2003 -- October 31, 2006

From Japan 695

(37)

Coeff. Prob. Coeff. Prob. Coeff Prob. Coeff. Prob. Coeff Prob. Coeff. Prob. Coeff Prob. Coeff. Prob. Coeff. Prob. Coeff Prob. 0.005595 0.0905 0.000376 0.7443 0.000173 0.7460 -0.000867 0.0021 -0.001098 0.0000 -0.000395 0.1039 -0.000168 0.5374 0.000594 0.0149 -0.000035 0.0033 0.000027 0.2672 -0.571511 0.0540 -0.018970 0.8961 -0.012371 0.8985 -0.014816 0.8024 -0.002425 0.9681 -0.079484 0.1572 -0.221845 0.0019 -0.359180 0.0000 0.107393 0.0000 -0.177081 0.0000 0.001002 0.2898 0.000519 0.2755 0.000819 0.0165 0.000036 0.8974 0.000553 0.0082 -0.000088 0.6441 0.000657 0.0045 0.000398 0.0858 -0.000107 0.0000 0.000030 0.3685 -0.006538 0.7250 0.000958 0.9240 0.484630 0.0000 0.048811 0.0004 0.259670 0.0000 0.056805 0.0000 0.181095 0.0000 0.013958 0.0063 0.000738 0.0563 0.020076 0.0000 0.004829 0.9158 -0.029426 0.2919 -0.068102 0.0085 0.030477 0.2618 0.026208 0.3854 0.121158 0.0002 0.083478 0.0046 0.004988 0.8915 -0.087407 0.0027 -0.029946 0.2674 a 0.000054 0.0003 0.000002 0.0692 0.000000 0.7709 -0.000001 0.0001 -0.000001 0.0002 0.000000 0.7268 0.000000 0.0084 0.000002 0.0000 0.000000 0.0030 0.000000 0.0000 b 0.593493 0.0000 0.914643 0.0000 0.927056 0.0000 0.890822 0.0000 0.884633 0.0000 0.741037 0.0000 0.731808 0.0000 0.600053 0.0000 0.814675 0.0000 0.730535 0.0000 c 0.147039 0.0225 0.066254 0.0001 0.061829 0.0000 0.110144 0.0000 0.120952 0.0000 0.282781 0.0000 0.201735 0.0000 0.150310 0.0000 0.259026 0.0000 0.265276 0.0000 d -0.000045 0.0000 -0.000003 0.6008 -0.000001 0.4463 0.000004 0.0003 0.000003 0.0003 0.000000 0.5205 0.000005 0.0000 0.000001 0.0000 0.000000 0.0000 0.000000 0.0000 -0.000006 0.0000 0.000000 0.4521 0.000001 0.0000 0.000000 0.0000 0.000000 0.0000 0.000002 0.0000 0.000002 0.0000 0.000000 0.0000 0.000000 0.0000 0.000000 0.0000 Coefficient of skewness for normalized residuals Coefficient of kurtosis for normalized residuals Ljung-Box(12) for normalized residuals Ljung-Box(12) for normalized squared residuals From China SHS B 1550 8708.232 -0.545707 16.57324 U.S From China SHS A 9.4554 21.89 1.8792 18.452 -0.694491 -0.190153 2.390095 10.14 16.573 17.01455 62.761 7.9576 9.1643 328.85 4.122 7.7586 7503.104 8611.92

From Japan From U.S 1502 1550 5631.933 6047.128

From Japan From U.S 1502 1550 5803.14 6176.276 4935.198 5770.887 0.186254 -0.582492 0.673243 6.571694 10.18958 0.505914 0.30324 9.154377 4.502617 Japan From China SHS B 1502 5106.611

From China SHS A From U.S

1502 1556

From Japan

1550 1556