AMSTERDAM BUSINESS SCHOOL
Development of Stock Market Integration and Contagion: Evidence
from China and its Main Trading Partners
MSc Finance: Asset Management Track Master Thesis
Author: Xiaoou Wang (11272848)
Supervisor: Dr. L. (Liang) Zou
Statement of Originality
This document is written by Xiaoou Wang, who takes full responsibility for the contents of this document.
I declare that the text and the work presented in this document are original and that no source other than those mentioned in the text and its references have been used in creating it.
The Faculty of Economics and Business is responsible solely for the supervision of completion of the work, not for the contents.
Abstract
Based on previous studies, emerging markets attract capital from developed markets for portfolio diversity with higher returns and volatilities. The aim of this thesis is to evaluate the integration and contagion in a country-level between the biggest emerging market, China, and its main trading partners including United States, United Kingdom, Germany, Netherlands, Australia, Japan, and South Korea. A latent factor/ GARCH model, based on the trade linkage, is applied for baseline estimation. Three most recently crisis periods are studied, as the Global Financial Crisis, European Debt Crisis, and China Market Turbulence. The estimation results show that South Korea has integration with China during the Global Financial Crisis, but not for other periods. Furthermore, the connection between China and Unites States markets is getting stronger after the Global Financial Crisis. For results of contagion, Australia has contagion with China during the European Debt Crisis, and United States has contagion with China after the Global Financial Crisis. Moreover, Australia, Japan, and South Korea have high comovements with China but no evidence of contagion. Other markets have no contagion or have negative correlations with China.
Table of Contents
1. Introduction ... 1
2. Literature review ... 5
2.1 Transmission of Financial Markets Shocks ...5
2.2 Measurement of Contagion ...7
3. Methodology ... 10
3.1 Estimation Model ... 10
3.2 Tests of Market Integration... 13
3.3 Tests of Market Contagion ... 13
3.4 Estimation Method and Model Selection ... 14
4. Data and Summary Statistics ... 15
5. Empirical Results ... 20
5.1 Estimation Results of Integration ... 20
5.2 Estimation Results of Contagion ... 31
6. Robustness Check ... 39
6.1 Estimation Results of Integration ... 39
6.2 Estimation Results of Contagion ... 44
7. Conclusion and Discussion ... 49
7.1 Conclusion ... 49
7.2 Discussion... 50
References ... 51
1
1. Introduction
As the biggest emerging market, China is full of critical questions from international investors since its little-known capital market and unclearly connection with the global market. As an emerging market, why China’s market matters? Since the 1980s, China’s economic reform has made itself one of the global leading economic powers and changed the game of global economy. China’s merchandise trade had increased from less than 1% of the total worldwide merchandise to more than 24% in 20151. China has been tagged as the “World Factory” since its strong connection with
the rest of global economy through trade. At the meanwhile, China has developed a huge capital market, as the second largest stock market after the United States. Moreover, foreign holdings of China’s financial assets worth about U.S.$2 trillion, more than any other emerging markets. On the other hand, China’s financial market capitalization into world equity market is unmatched with its economy into the global economy, as Bekaert (1995), and Claessens and Forbes (2013) present.
China and other emerging markets attract capital from developed markets for several reasons, such as high GDP growth rates, booming of the middle class, and abundance of natural resources. When investing in emerging markets, investors based on information of developed markets face different perspective as to investing in the developed market. As Kearney (2012) finds, even though facing some important and specific risks, for example, high political risk, large current account deficits, and high unemployment rate, investors still expect higher returns to cover these risks. On one hand, understanding integration and contagion between emerging markets and developed markets is important to investors to determine the cost of capital and to make investment and asset allocation decision when investing in emerging markets. On the other hand, how emerging markets, like China, connect with global financial markets and developed markets attracts researchers’ attention, for example, Baele and Inghelbrecht (2009), Bekaert and Harvey (2013), and Beirne et al, (2013).
Integration and contagion are the two majority directions in academic research to measure the connection between developed markets or between developed markets and
1 WTO International Trade and Market Access Data. Available on:
2
emerging markets. Correlations between developed markets and emerging markets have increased as the result of globalization, but study about market integration and contagion of emerging markets are still incomplete (Bekaert and Harvey 2014). Specific for China, the financial market is most studied about spillover effect rather than integration and contagion as for other emerging markets. The degree of integration is necessary to explain the link between financial markets. As Bekaert and Harvey (1995, 1997) indicate that the financial market risk is hard to quantify if one market is not fully integrated into the global financial markets, but the global capital markets are not always more integrated at country-level. Besides integration, how shocks of financial markets transmit across the borders are also important to determine asset prices and asset allocation. Ever since King and Wadhwani (1990), transmission of shocks across borders has been studied by countless international finance articles. Numbers of studies on contagion have addressed that the transmission of volatility across developed markets or from developed markets to emerging markets, which could not be explained by economic fundamentals. However, the studies about transmission between developed markets and emerging markets reached variety, often conflict, conclusions, particular for studies on emerging markets. The dynamical correlations between stock markets supposed to have changed during and after the economic recession periods, such as the Global Financial Crisis in 2007 - 2008, and recent depreciation of China’s economic growth. Investors have long been wary of sneezes from United States, knowing they can give the world a cold. As a dynamic change of financial market integration and contagion, probability from China now rivals that from United States, that a new question arises: if China sneezes, will the world has a cold? Therefore, the aim of this study is to evaluate whether the integration and contagion between China stock and other developed markets have changed.
Before answering the research question, defining the meaning of contagion is necessary. During crisis periods, volatilities of financial markets tend to increase so dramatically that increasing correlations are not sufficient to indicate contagion. In this study, the definition of contagion follows the way of Bekaert et al. (2014) that contagion is narrowly defined as unexplained increases in factor loadings and residual correlations2.
3
Real linkage, such as trade linkage, is one of the most discussed sources of contagion by previous researchers. In this study, the real linkage is being investigated as the source of contagion, which is measured by merchandise trade. Chen and Zhang (1997) find that the correlations of stock returns are related to the merchandise trade among markets. Different with previous studies focus on (geographic) regional merchandise trade (for example, Bekaert, Harvey, and Ng, 2005), a group of markets based on the trading connection with China is been chosen for this study. Based on the merchandise trading data in the end of 2015, the main trading partners of China include United States, Germany, United Kingdom, Netherlands, Australia, Japan, and South Korea. By the end of 2015, the total value of merchandise trade between China and above markets was near U.S.$1.7 trillion, occupying more than 10% of the global merchandise trade3. These markets together hold more than 46% of world merchandise
trade. The trade between these markets not only hold a large piece of cake in global trade but also play a major role in merchandise trade between each other.
To predict market returns and volatilities based on merchandise trade, a time-varying two-factor CAPM model, based on Bekaert, Harvey, and Ng (2005), is applied in this study. The two factors are the stock market returns of China and a constructed equity portfolio which consist of the rest of the sample. The contagion of financial market will be supported if the markets’ idiosyncratic shocks show significant correlations. Integration, which could be observed by the two-factor model, plays a critical role before test contagion. The volatility model is related to Bekaert and Harvey (1997) that volatilities of stock market return follow the univariate generalized autoregressive conditional heteroscedasticity (GARCH). The framework of models and variables will be introduced in detail in following sections.
The estimation results of the latent factor/GARCH model give two major conclusions. First, there is no evidence of integration between China and its trading partners during the whole sample period. Only South Korea has integration with China during Global Financial Crisis, but not for other periods. Australia, Japan, and South
and disadvantages, see Forbes (2012). The definition is not accepted universally, broadly defined as the transmission of crises across countries implied by common shocks is generally accepted.
3 Hong Kong SAR is also an important trading partner with mainland China. But since the merchandise
trading at Hong Kong works more like a transfer center between China and other markets, different with merchandise trading commonly used, it is being excluded from the sample.
4
Korea show co-movement with China. Furthermore, the connection between China and Unites States is getting stronger after the Global Financial Crisis. Second, for results of contagion, Australia has contagion with China during the European Debt Crisis, and the United States has contagion with China after the Global Financial Crisis. Moreover, Australia, Japan, and South Korea have high co-movement with China but no evidence of contagion. Other markets have no contagion or have negative correlations with China.
The remainder of the thesis is structured as follow. Section 2 presents the literature review of studies on transmission of financial markets shocks, especially on integration and contagion. Furthermore, methodologies of measuring contagion are also being reviewed.Section 3 introduces the methodology used in this study. Section 4 presents the data and summary statistics. Section 5 shows the primary results. Section 6 uses financial sector as a robustness check. The last section presents conclusion and discussion of this study.
5
2. Literature review
This section reviews the relevant researches on contagion and provides the academic background for this thesis. This section starts with a relevant literature review on the transmission of shocks between financial markets. Next, empirical models applied to measure contagion are addressed.
2.1 Transmission of Financial Markets Shocks
Countless papers have studied about why and how shocks transmit across borders. Initially, King and Wadhwani (1990) investigate why most stock markets fell together during the 1989 global stock market crash, regardless widely differences in economic circumstances. They find with the increase of volatility, correlations between markets increase significantly. Latterly, the 1992 European exchange rate mechanism (ERM) crisis boosts the empirical analysis on the transmission of shocks which continuous hot after the Global Financial Crisis and recently European Debt Crisis. Longin and Solnik (1995) find that during a volatile period, the cross-country correlation increases more than a pre-crisis period. Erb, Harvey, and Viskanta (1994), and Santis and Gerard (1997) find correlations across markets are in different directions when markets go up or down. While Longin and Solnik (2001), Ang and Bekaert (2002), Ang and Chen (2002), and Das and Uppal (2004) find higher cross-markets correlations during bear periods, comparing to bull ones. Most studies define transmission of shocks as interdependence, spillover, or contagion. While opinions on the definition of contagion differ, there is a broad consensus among empirical studies that define contagion refers to an unexpected transmission of shocks. Thus, contagion should be differentiated from interdependence and spillover across asset pricing markets. As Forbes and Rigobon (2002) distinct that interdependence is a high level of market co-movement during all periods, while contagion as a significant increase in cross-markets correlations after a shock.
After several global or regional financial market crises, growing researcher study on drivers of the transmission of crises across markets in both firm-level and country- level. Literature, like Tong and Wei (2010), Diebold and Yilmaz (2012), and Almeida et al. (2012) focus on drivers of shock transmission during the Global Financial Crisis across sectors and financial markets within the United States. Tong and Wei (2010) explain how the mortgage crisis of bank sector influences upon nonfinancial firms in
6
2007. Diebold and Yilmaz (2012) find stock, bond, currency, and commodities markets do have cross-market excess correlation when the global financial crisis began.
Some researchers focus on the contribution of globalization in market contagion, like financial linkages and trade linkages (Bekaert et al., 2014). Allen and Gale (2000) and Eichengreen et al. (2012) and some other researchers try to explain contagion through the financial linkages. Allen and Gale (2000) find that the possibility of contagion depends more on inter-regional claims of the banking system. They also claim the liquidity shock in financial sectors could spread out to the entire economy since contagion between financial sectors and real sectors during the global financial crisis. Eichengreen et al. (2012) question why a small issue in a corner of U.S. financial market could affect the entire global financial markets. They find an effect arises from common factors to the global banking system that steadily increases after the collapse of Lehman Brothers. Rose and Spiegel (2010, 2011) indicate that a country holding shock-suffering securities is exposed through a financial channel, but a country that exports to a volatility country is exposed through a real channel. Salgado, Caramazza, and Ricci (2000) investigate the lending relation between “ground zero”4 country and major lenders and find that countries suffering crises relied more on the capital lender for funding than non-crisis countries. Kaminsky and Reinhart (2000) identify that bank lending channel, liquidity channel, and trade channel could transmit shocks.
On the other side, some researchers show the economic linkage or saying, real linkage, would explain the contagion across markets as well. Eichengreen and Rose (1999) find strong support for the importance of trade mechanism to contagion. They conclude that their findings lend support to trade linkage rather than macro economy similarities that could transmit shocks. Glick and Rose (1999) develop this framework to a further complete research. Instead of using aggregate country-level market data, Forbes (2000) uses the firm-level information to study the importance of trade count on the transmission of crises. This study concludes that both direct trade effect and competition between exporters are important channels of transmission of crises. Forbes (2002, 2004) find that international trade linkages spread one country's stock market crisis to others’ equity markets, especially for markets with similar third-country export profiles. Mendoza and Quadrini (2010), Fratzscher (2012), and Brière, Chapelle, and
4
7
Szafarz (2012) figure out similar conclusions that the increased vulnerability to crises that come from both financial and economic integration. Van Rijckeghem and Weder (2000) argue that financial linkages are critical to explaining shock transmission, but they also conclude that trade and financial linkages are strongly correlated.
Some other channels, such as information asymmetries and herding behavior of investors, could also possible to explain the transmission of crises. Information asymmetries may reduce cross-border capital flow and arise home bias, as Brennan and Cao (1997), Albuquerque, Bauer, and Schneider (2009) introduced. The geographic distance from a market to the crisis market makes domestic and foreign investors hold different opinions to news, may create contagion across borders, as Portes and Rey (2005), Daude and Fratzscher (2008), and Dumas, Lewis, and Osambela (2016) find. Investor’s risk aversion and liquidity requirement could also cause contagion. Bekaert et al. (2011) and Baker, Wurgler, and Yuan (2012) suggest that international assets prices are sensitive to investor’s risk aversion. Brunnermeier and Pedersen (2009), and Adrian and Shin (2010) reach similar conclusions that illiquidity may cause crisis across borders.
Furthermore, numbers of articles study of financial sectors or credit shocks of firm-level contagion during the global financial crisis. For example, Tong and Wei (2011) study 4000 firms in 24 emerging markets and find that the companies in emerging markets intrinsically dependent on external financing suffer more than others. Beltratti and Stulz (2012) use global large banks to investigate whether volatility in the cross-section of stock returns is more relative to other external markets, rather than internal factors like governance or monetary policy. Calomiris, Love, and Peria (2012) show that stock market holds pressure from global demand shocks, credit supply shocks, and selling pressures, which have significant negative effects on stock returns.
2.2 Measurement of Contagion
The following part reviews the five most used methods to measure contagion, including probability analysis, cross-market correlations, VAR models, latent factor/GARCH models, and extreme value analysis. Most studies measure the contagion of stock markets since available of high-frequency data and a broad
cross-8 section of markets.
Eichengreen, Rose, and Wyplosz (1996), De Gregorio and Valdes (2001), and Forbes (2002) use probability analysis to evaluate the presence and importance of contagion by test whether a crisis happened in a ground zero market would affect another market’s likelihood. These studies find extensive evidence that the probability of crisis occurred in market increases if there is another crisis elsewhere, especially for countries in the same geographic region. Forbes and Warnock (2012) extend this methodology to test the contagion by explaining sudden jump in capital flows. Constancio (2012) uses the similar way to test the default probabilities derived from CDS to explain contagion across financial markets.
King and Wadhwani (1990), and Bordo and Murshid (2001), for example, investigate contagion by test whether cross-markets correlations of stock returns increased significantly after a shock. Following their definition of contagion, these studies find significant evidence of contagion as co-movements of markets increase significantly during most crises. However, Forbes and Rigobon (2002) critique of this method that increased volatility during crises causes upward biases to the correlation coefficient. After adjustment of this heteroskedasticity in asset pricing, it leads no evidence of contagion. Furthermore, as Forbes (2012) summaries, even if this issue is solved, using correlation coefficients to test contagion is still under challenge because of endogeneity, omitted variables, and common shocks across markets.
Vector Autoregression (VAR) is a popular methodology applied to investigate contagion, for instance, Forbes and Rigobon (2002), Favero and Giavazzi (2002), and Constancio (2012). The VAR model evaluates the predicted errors through predicts stock market returns with controlling of global information factors and country-specific information factors.Contagion is measured with impulse-response function predicting the impact of an unexpected shock from one market to others. This methodology is adjusted for the heteroskedasticity in returns and generates more significant evidence of contagion. However, Bekaert, Harvey, and Ng (2005) imply that the bias correction for correlations does not work for the common shocks across markets as well.
Instead of using VAR model, Hamao et al. (1990), Bekaert, Harvey, and Ng (2005), Dungey et al. (2010), and Bekaert et al. (2011) use latent factor/GARCH model to
9
investigate contagion across markets and countries. In most of these papers define contagion as the “excess correlation”, after controlling for fundamental factors. The use of latent factor and GARCH model allow return variances change across regimes, which could not be avoided by VAR model. These papers focus on estimating cross-market co-movements in the second moments of asset prices, instead of measuring directly on asset prices. Using this methodology, researchers find evidence of contagion but not for all crises.
Some researchers, for example, Longin and Solnik (2001), Bae, Karolyi, and Stulz. (2003), Hartman et al. (2004), Boyer et al. (2006), and Boyson, Stahel, and Stulz (2010) use multivariate extreme value theory to test whether tails of returns are correlated across markets. This approach is based on analysis the periods when a particular group of variables’ realization values significantly exceed expect values, either in absolute values or compare to returns’ distributions. Another similar way focuses on the periods when showing significant “jump” in asset prices, as Aït-Sahalia, Cacho-Diaz, and Laeven (2010), and Pukthuanthong and Roll (2014) do. Papers using these approaches find evidence of contagion during some crises, still, not all crises.
As Rigobon (2002), Dungey et al. (2010), and Forbes (2012) discuss, each of these methodologies have their advantages and disadvantages in measuring contagion. The choice of methodology is based on the definition of contagion used in the study. Specific for this thesis, since the contagion is defined as extortionate comovement that could not explain by fundamentals, a latent factor/GARCH model is be chosen. Previous studies on international market contagion mainly focus on the United States or global market index respect to other stock markets. To answer the research question, China market is used as benchmark market respect to other markets based on trade linkages.
Another important point about methodology is the application of the asymmetric model. As computed by Longin and Solnik (2001), Ang and Bekaert (2002), and Ang and Chen (2002), correlations between market returns are especially stronger during market turmoil, suggesting that contagion may be asymmetric.
10
3. Methodology
Consistent with the study of Bekaert, Harvey, and Ng (2005), a latent factor/GARCH model is used in this thesis. Instead of using U.S. stock market as the benchmark, China’s stock market is being used to answer the research question. The other external factor is a constructed portfolio from the rest of sample, excluding China market and the market under examination.
3.1 Estimation Model
Let 𝑅𝑖,𝑡 be the daily return on the national equity index of country 𝑖 , which obtained as the first difference of the natural log of the price indices of the eight stock markets:
𝑅𝑖,𝑡 = [log(𝑃𝑖,𝑡) − log(𝑃𝑖,𝑡−1)] ∗ 100 (1)
where 𝑃𝑖,𝑡 is the market price index for market 𝑖 at time 𝑡. Returns are based on U.S. dollar terms and adjusted for U.S. short-term risk free rate. By this adjustment, it is convenient to compare the results and could be observed directly for international investors5. Following is the full model:
𝑹𝒊,𝒕 = 𝛿𝑖′𝒁𝒊,𝒕−𝟏+ 𝛽𝑖,𝑡−1𝐶𝑁 𝜇𝐶𝑁,𝑡−1+ 𝛽𝑖,𝑡−1𝑃 𝜇𝑃,𝑡−1+ 𝛽𝑖,𝑡−1𝐶𝑁 𝑒𝐶𝑁,𝑡+ 𝛽𝑖,𝑡−1𝑃 𝑒𝑃,𝑡+ 𝑒𝑖,𝑡 (2) 𝑒𝑖,𝑡|𝛀𝑡−1~𝑵(0, 𝜎𝑖,𝑡2) (3) 𝜎𝑖,𝑡2 = 𝑎
𝑖+ 𝑏𝑖𝜎𝑖,𝑡−12 + 𝑐𝑖𝑒𝑖,𝑡−12 + 𝑑𝑖𝜂𝑖,𝑡−12 (4)
where 𝜇𝐶𝑁,𝑡−1 is the conditional excess return of China. 𝜇𝑃,𝑡−1 is the conditional excess return of a constructed portfolio from the sample markets. The portfolio excludes China market and the market 𝑖 under estimation. Excess returns are based on the information available at time 𝑡 − 1. 𝑒𝑖,𝑡 is idiosyncratic shock of market 𝑖, including China and other markets. 𝜂𝑖,𝑡 is the negative return shock of any market 𝑖, measured by 𝜂𝑖,𝑡 = min {0, 𝑒𝑖,𝑡} . Ω𝑡−1 is the information sector which includes all the information available at time 𝑡 − 1. 𝒁𝒊,𝒕−𝟏 is the information variables vector used to
5
U.S. dollar is used as pricing currency here since it’s a common used as international foreign reservation for each country and as price of mostly international assets. It’s more acceptable by international investors.
11
estimate the expected return of market 𝑖 . For China market, 𝒁𝑪𝑵,𝒕−𝟏 includes the spread of overnight and 3-month money market interest rates, financial depth6 , and
yield of 10-years maturity government bond. For individual markets, the local information vector 𝒁𝒊,𝒕−𝟏 includes dividend yield of market index and financial depth of market 𝑖 at time 𝑡 − 1. Different with former researches that using regional market index, the local information factors vector 𝒁𝑷,𝒕−𝟏 of the constructed portfolios
includes market value weight dividend yield and GDP weight financial depth. Eqq. (4) measures the GARCH process with asymmetric effect in the conditional variance, based on the assumption that the variance of idiosyncratic return shock of market 𝑖 follows the univariate generalized autoregressive conditional heteroscedasticity (GARCH) as Glosten, Jagannathan, and Runkle (1993) and Zakoian (1994) present.
The sensitives of equity market 𝑖 to the external information are measured by the parameters 𝛽𝑖,𝑡−1𝐶𝑁 and 𝛽
𝑖,𝑡−1𝑃 . These time-varying risk exposures based on trade are
measured as following:
𝛽𝑖,𝑡−1𝐶𝑁 = 𝑝1,𝑖′ 𝑿𝑖,𝑡−1𝐶𝑁 + 𝑞𝑖′𝑿𝑖,𝑡−1𝑊 ×𝑤𝐶𝑁,𝑡−1 (5) 𝛽𝑖,𝑡−1𝑃 = 𝑝
2,𝑖′ 𝑿𝑖,𝑡−1𝑃 + 𝑞𝑖′𝑿𝑖,𝑡−1𝑊 ×(1 − 𝑤𝐶𝑁,𝑡−1) (6)
where 𝑤𝐶𝑁,𝑡−1 denotes stock market value of China, relative to the whole sample market value at time 𝑡 − 1 . 𝑿𝑖,𝑡−1𝐶𝑁 , 𝑿
𝑖,𝑡−1
𝑃 and 𝑿 𝑖,𝑡−1
𝑊 are trade information
instruments. 𝑿𝑖,𝑡−1𝐶𝑁 measures the sum of export to and import from China divided by
the total exports and imports of market 𝑖 with the whole sample markets. 𝑿𝑖,𝑡−1𝑃 is
used to measure the trade value of market 𝑖 with the rest of sample except China, divided by the sum of the total export and import of market 𝑖 with the whole sample. 𝑿𝑖,𝑡−1𝑊 is the total trade size of market 𝑖 with the rest of whole sample markets, as percentage of market 𝑖’s GDP. China market model is a special case for eqq.(2) - (6). For China market, 𝑝1,𝐶𝑁 = 𝑝2,𝐶𝑁 = 𝑞𝐶𝑁 = 0 , which leads the risk factors 𝛽𝑖,𝑡−1𝐶𝑁 =
𝛽𝑖,𝑡−1𝑃 = 0.
In eqq. (5) – (6), 𝛽𝑖,𝑡−1𝐶𝑁 is used to measures the risk exposure respect to China on
6 Financial depth is defined as the private credit from all sector as percentage of GDP, which follows
the definition from World Bank. A discussion is available at: http://www.worldbank.org/en/publication/gfdr/background/financial-depth.
12
individual market 𝑖’s expected return. 𝛽𝑖,𝑡−1𝑃 is to measure the risk exposure respect
to the rest of the sample on the market 𝑖 . Both 𝛽𝑖,𝑡−1𝐶𝑁 and 𝛽
𝑖,𝑡−1𝑃 are based on the
merchandise trade relationship as eqq.(5) – (6) show. The unanticipated part of the market return is a result of shocks from the local market and from the two external factors originating from China and the constructed portfolio. The return residuals of market 𝑖 could be explained as:
𝜀𝑖,𝑡 = 𝛽𝑖,𝑡−1𝐶𝑁 𝑒𝐶𝑁,𝑡+ 𝛽𝑖,𝑡−1𝑃 𝑒𝑃,𝑡+ 𝑒𝑖,𝑡 (7)
The variance and covariance of the pricing errors could be expressed in the following way, as Bekaert, Harvey, Ng (2005) proved:
ℎ𝑖,𝑡 = 𝐸[𝜀𝑖,𝑡2 |𝛀𝑡−1] = (𝛽𝑖,𝑡−1𝐶𝑁 )2𝜎 𝐶𝑁,𝑡2 + (𝛽𝑖,𝑡−1𝑃 )2𝜎𝑃,𝑡2 + 𝜎𝑖,𝑡2 (8) ℎ𝑃,𝑡 = (𝛽𝑃,𝑡−1𝐶𝑁 )2𝜎 𝐶𝑁,𝑡2 + 𝜎𝑃,𝑡2 (9) ℎ𝑖,𝑡𝐶𝑁 = 𝐸[𝜀 𝑖,𝑡𝜀𝐶𝑁,𝑡|𝛀𝑡−1] = 𝛽𝑖,𝑡−1𝐶𝑁 𝜎𝐶𝑁,𝑡2 (10) ℎ𝑖,𝑡𝑃 = 𝐸[𝜀 𝑖,𝑡𝜀𝑃,𝑡|𝛀𝑡−1] = 𝛽𝑖,𝑡−1𝐶𝑁 𝛽𝑃,𝑡−1𝐶𝑁 𝜎𝐶𝑁,𝑡2 + 𝛽𝑖,𝑡−1𝑃 𝜎𝑃,𝑡2 (11)
Eqq.(8) – (9) show that variances of market 𝑖 and constructed portfolio are positively correlated with the conditional variance of the China market. Eqq.(10) and (11) measure the covariance of market 𝑖, with China and the constructing portfolio.
The correlations between market 𝑖 with China and constructed portfolio could be measured as:
𝜌𝑖,𝑡𝐶𝑁 = 𝛽𝑖,𝑡−1𝐶𝑁 𝜎𝐶𝑁,𝑡
√ℎ𝑖,𝑡 (12)
𝜌𝑖,𝑡𝑃𝑜𝑟𝑡 = 𝛽𝑖,𝑡−1𝐶𝑁 𝛽𝑃,𝑡−1𝐶𝑁 𝜎𝐶𝑁,𝑡2 +𝛽𝑖,𝑡−1𝑃 𝜎𝑃,𝑡2
√ℎ𝑖,𝑡ℎ𝑃,𝑡 (13) where the conditional variance of portfolio return ℎ𝑃,𝑡 follows eqq.(9). To explain the
relativity proportion of conditional return variance from China market or the constructed portfolio, variance ratios are introduced as:
𝑉𝑅𝑖,𝑡𝐶𝑁 = (𝛽𝑖,𝑡−1𝐶𝑁 )2𝜎𝐶𝑁,𝑡2
13
𝑉𝑅𝑖,𝑡𝑃 = (𝛽𝑖,𝑡−1𝑃 )2𝜎𝑃,𝑡2
ℎ𝑖,𝑡 (15) When risk exposures from merchandise trade increase, these variances also increase. Consistent with other measurements used in this study, their performance during the crisis is interested that whether the correlations across markets change.
3.2 Tests of Market Integration
Eqq. (2) - (6) are the general two-factor model that could be used to test the market integration. If the two-factor model holds, it means the expected return of market 𝑖 could be explained sufficiently by these two external risk factors. The integration is proved. More specifically, the integration is test under the constraint that 𝒑𝟐,𝒊 = 𝟎 (𝒑𝟏,𝒊 = 𝟎), and 𝒒𝒊= 𝟎 with China (constructed portfolios) as benchmark market.
This implies that market 𝑖 is fully integrated with China’s market (constructed portfolio). Two tests could be interpreted as 1) 𝜹𝒊 = 𝒑𝟐,𝒊 = 𝒒𝒊 = 𝟎 to test the market
integration between China and market 𝑖; 2) 𝜹𝒊 = 𝒑𝟏,𝒊 = 𝒒𝒊= 𝟎 to test the market
integration between constructed portfolio and market 𝑖.
3.3 Tests of Market Contagion
Based on the definition of contagion used in this study, contagion could be measured as the correlation of the model’s idiosyncratic shock. A time-series cross-sectional model is used to estimate the contagion over the particular crisis period:
𝑒̂𝑖,𝑡 = 𝑤𝑖 + 𝑣𝑖,𝑡𝑒̂𝑔,𝑡+ 𝜇𝑖,𝑡 (14)
𝑣𝑖,𝑡 = 𝑣0+ 𝑣1𝐷𝑖,𝑡 (15)
where 𝑒̂𝑖,𝑡 and 𝑒̂𝑔,𝑡 are the estimated idiosyncratic return shocks of market 𝑖 and China market, or construed. Two cases are under studied: 𝑒̂𝑔,𝑡 = 𝑒̂𝐶𝑁,𝑡, and 𝑒̂𝑔,𝑡 = 𝑒̂𝑝,𝑡. 𝐷𝑖,𝑡 is a dummy variable that identifies crisis periods for three recently financial market crises: the Global Financial Crisis, European Debt Crisis, and China Market Turbulence. In this study, the market contagion will be test by whether 𝑣0 = 𝑣1 = 0
for overall contagion, and whether 𝑣1 is significantly different from zero for contribution of a particular period to market contagion.
14
3.4 Estimation Method and Model Selection
The joint multivariate likelihood function for markets’ returns is estimated in three stages. In the first stage, the univariate model of China’s market is under estimation. Based on eqq. (2) - (6), conditional on information sector 𝛀𝑡−1, China’s market return
depends only on 𝜹𝑪𝑵, 𝑎𝐶𝑁, 𝑏𝐶𝑁, 𝑐𝐶𝑁, and 𝑑𝐶𝑁. In the second stage, the constructed portfolios are estimated. Except regional local factor 𝛀𝑡−1, the constructed portfolios are also conditional on 𝑅𝐶𝑁,𝑡 . Thus, the constructed portfolios are based on risk exposures: 𝜹𝒑𝒐𝒓𝒕, 𝑎𝑝𝑜𝑟𝑡, 𝑏𝑝𝑜𝑟𝑡, 𝑐𝑝𝑜𝑟𝑡, 𝑑𝑖 and 𝒑𝟏,𝒑𝒐𝒓𝒕. These risk exposures could be obtained by maximizing the univariate likelihood for the constructed portfolio’s returns. In the third stage, for the individual markets, returns are based on estimations of China and constructed portfolio, which depend on 𝜹𝒊, 𝑎𝑖, 𝑏𝑖, 𝑐𝑖, 𝑑𝑖, 𝒑𝟏,𝒑𝒐𝒓𝒕, 𝒑𝟐,𝒑𝒐𝒓𝒕 and
𝒒𝒊. In the third stage, since different portfolios are used, the individual markets are
estimated one by one based on eqq. (2) - (6).
Based on the study of Zivot (2009), because of asymmetric and symmetric GARCH models generating different conditional variances, a likelihood ratio test is introduced to decide which model to use for China, constructed portfolios, and individual markets. The asymmetric model would be picked if the likelihood ratio test could reject the null hypothesis at a 10% level, otherwise, the symmetric model is selected.
15
4. Data and Summary Statistics
Eight stock markets are studied in the empirical analysis of daily returns starts from January 3rd, 2003 to December 31st, 2016. The sample data consist stock markets of China and its seven principal trading partners, including United States, United Kingdom, Germany, Netherlands, Australia, Japan, and South Korea. One index for each country is used, as Shanghai Composite Index (SECI) used for China, S&P 500 used for United States, FTSE100 index used for United Kingdom, DAX 30 Performance index used for Germany, AEX index used for Netherlands, S&P/ASX200 index used for Australia, TOPIX used for Japan, and KOSPI used for South Korea. All the data are available through DataStream. The market returns are calculated based on eqq.(1) and converted into U.S. dollar term, as to compare returns on the same basement and intuitively to international investors. The risk-free rate is measured by the U.S. 3-month treasury constant maturity rate and subtracted from U.S. dollar term returns of markets. Another important adjustment is the use of rolling average two-days average of returns, since the effect of cross time-zone of markets in the simple7. Besides using
individual markets data, several portfolios are constructed to control shocks from the rest of sample markets rather than China. Each one represents a portfolio which has a daily return weighted on market value, excluding China market and market 𝑖 under estimation.
The stationary of stock market returns is tested by Augmented Dickey-Fuller (ADF) test. Given the choice of estimation model, it is necessary to test the ARCH effect with Lagrange Multiplier test, and GARCH effect with Ljung-Box test (Bauwens, Laurent, and Rombouts, 2006). The Lagrange test is used for the ARCH effect with the null hypothesis of no ARCH effect. Ljung-Box tests the null hypothesis that the stock market return has no GARCH effect or saying, there is no serial correlation in the standardized residuals. If both null hypotheses are rejected, the returns have appropriate GARCH effects. The normal distributions of the returns are tested by the Jarque-Bera test, with the null hypothesis that the skewness and excess kurtosis are both equal to zero. Table 1 shows the summary statistics of the indices’ daily returns. The mean
7 Under this adjustment, the bias from different trading time zone may still affect the returns. Longer
estimation period of return, like weekly return or monthly return, works better than daily base, but it will also weak the shocks of markets. Discussion see Forbes and Rigobon (2002).
16
values are shown in percentage to more visible. The sample stock market indexes show a similar level of mean return and volatility. Most indices show leptokurtic distributions. The index of kurtosis indicates that returns have heavier tails than normal distributions. Based on the Jarque-Bera test result rejects the null hypothesis that skewness and excess kurtosis are zero, indicates that distributions are not normality. The ADF tests, under 1% confidential level, reject the null hypothesis that returns have no unit root. Furthermore, both null hypotheses that no ARCH or GARCH effect are rejected significantly, based on Lagrange Multiplier test and Ljung-Box test respectively.
Table 1
Summary Statistic of Stock Market Indexes Returns
The table reports the summary statistics for the sample markets indices from January 2003 to December 2016. The mean return is measured as a percentage. The ADF tests are applied to test the null hypothesis of the unit root of the returns; LM-ARCH is the Lagrange Multiplier test for the null hypothesis of no ARCH effect (with a 5th order); Ljung-Box test (Q test) is used to test the GARCH effect, under the null hypothesis of no GARCH effect (with a 5th order); Jarque-Bera (J-B) test are used to test the normality of the return, with null hypothesis of both the skewness and excess kurtosis being zero. *, **, and *** indicate significance at 10%, 5%, and 1%, respectively.
United States United Kingdom Germany Netherlands
Observations 3653 3653 3653 3653 Mean Return (%) 0.0335 0.0242 0.0389 -0.133 St.D 0.783 0.955 1.114 1.091 Skewness -0.5518 -0.4213 -0.2430 -0.3151 Kurtosis 11.3574 11.7851 7.8401 8.8255 ADF -38.6877*** -36.1949*** -35.5336*** -35.1686*** LM-ARCH 984.2198*** 1230.9713*** 968.8812*** 1041.7060*** Ljung-Box 680.3668*** 863.9240*** 880.8793*** 912.4840*** J-B 28781.8085*** 3077.9977*** 4479.9903*** 19568.5017***
Australia Japan South Korea China
Observations 3653 3653 3653 3653 Mean Return (%) -0.1139 -0.1366 -0.1176 -0.138 St.D 1.080 0.954 1.197 1.162 Skewness -0.2724 -0.2305 -0.1239 -0.4149 Kurtosis 8.9654 8.7717 19.0985 7.1093 ADF -32.9335*** -38.9154*** -31.2056*** -35.0954*** LM-ARCH 832.6779*** 1344.9327*** 1258.8765*** 803.2644*** Ljung-Box 1089.9471*** 700.4969*** 1260.5670*** 949.1367*** J-B 1586.2370*** 6855.9268*** 11040.9128*** 2434.6056***
17
Table 2
Sample and Sub-Sample Periods
In this study, the Global Financial Crisis defined as a period starts from August 7th, 2007, when the liquidity crisis showed its head, till March 15th, 2009. The European Debt Crisis period starts from December 8th, 2009, as a downgrade of Greece to BBB+ from A- by Fitch, till December 21st, 2011, when ECB announced the refinancing operation. The China Market Turbulence period starts from June 12th, 2015 and ends on February 15th, 2016.
Period Observations Crisis
Full Sample Jan.3, 2003 to Dec.31, 2016 3653 Subsample 1 Jan.2, 2003 to Aug.6, 2007 1199
Subsample 2 Aug.7, 2007 to Mar.16, 2009 420 Global Financial Crisis Subsample 3 Mar.17, 2009 to Dec.7, 2009 190
Subsample 4 Dec.8, 2009 to Dec.21, 2012 794 European Debt Crisis Subsample 5 Dec.21, 2012 to Jun.11, 2015 644
Subsample 6 Jun.12, 2015 to Feb.15, 2016 177 China Market Turbulence Subsample 7 Feb.16, 2016 to Dec.31, 2016 229
Based on recently three financial market crises, the whole sample is spilled to seven subsamples: The Global Financial Crisis period from August 7th, 2007, when the liquidity crisis showed its head, till March 15th, 2009. The European Debt Crisis period from December 8th, 2009, as a downgrade of Greece to BBB+ from A- by Fitch, till December 21st, 2011, when European Central Bank announced the refinancing operation. The China Market Turbulence period starts from June 12th, 2015, to February 15th, 20168. Table 2 reports the subsample periods used in this study.
Table 3 shows the unconditional correlations of market indices of the full sample period and three crisis subsamples. China’s market has clearly increase correlations with markets of its trading partners, especially with United States, Australia, Japan, and South Korea. During the European Debt Crisis, correlations between markets of China and European are also stronger than the former Global Financial Crisis. During the China Market Turbulence, the correlations between China, United States, and Japan significantly differ from the ones in previous crises. On the other side, the rest of markets have lower correlations with China’s market during this period, even for Australia and South Korea which have strong economic connections with China.
8 The Brexit is not considered in this study. Even though the volatility of financial markets indeed
increased, especially for European markets. It is still in process which makes it technically impossible to identify the end point.
18
Table 3
Cross-Correlations of the Whole Sample and Subsamples
This table represents the unconditional correlations of returns between markets in the sample. Panel A shows the whole sample period starts from January 3rd, 2003 to December 31st, 2016. Panel B reports the sample period of Global Financial Crisis starts from August 7th, 2007 to March 16th, 2009. Panel C shows the correlations during the European Debt Crisis, from December 8th, 2009 to December 21st, 2012. Panel D reports for period starts from June 12th, 2015 to February 15th, 2016, during the China Market Turbulence.
Panel A: Whole Sample
US UK Germany Netherlands Australia Japan S. Korea China
US 1 UK 0.7128 1 Germany 0.7316 0.8432 1 Netherlands 0.7151 0.8841 0.8989 1 Australia 0.5712 0.6925 0.6300 0.6742 1 Japan 0.2238 0.3793 0.3536 0.4008 0.5418 1 S. Korea 0.3388 0.4081 0.4091 0.4217 0.5558 0.5265 1 China 0.1414 0.1935 0.1688 0.1915 0.2709 0.2213 0.2350 1
Panel B: Global Financial Crisis
US UK Germany Netherlands Australia Japan S. Korea China
US 1 UK 0.7492 1 Germany 0.7684 0.8859 1 Netherlands 0.7515 0.9268 0.9072 1 Australia 0.6332 0.7966 0.7698 0.7902 1 Japan 0.1518 0.4807 0.4113 0.4606 0.6239 1 S. Korea 0.3330 0.4409 0.4599 0.4468 0.6190 0.6137 1 China 0.0860 0.2087 0.2082 0.2335 0.3014 0.2700 0.2713 1
Panel C: European Debt Crisis
US UK Germany Netherlands Australia Japan S. Korea China
US 1 UK 0.8305 1 Germany 0.8189 0.8966 1 Netherlands 0.8202 0.9239 0.9529 1 Australia 0.7422 0.7802 0.7230 0.7505 1 Japan 0.2972 0.3523 0.3718 0.3589 0.5074 1 S. Korea 0.3893 0.4630 0.4021 0.4179 0.6002 0.5350 1 China 0.2630 0.3183 0.2813 0.2743 0.4116 0.2685 0.3419 1
Panel D: China Market Turbulence
US UK Germany Netherlands Australia Japan S. Korea China
US 1
UK 0.7404 1
19
Table 3 - Continued
US UK Germany Netherlands Australia Japan S. Korea China
Germany 0.5428 0.8230 1 Netherlands 0.6048 0.9115 0.9056 1 Australia 0.6062 0.6935 0.5475 0.5969 1 Japan 0.3682 0.4252 0.2970 0.3857 0.5478 1 S. Korea 0.5039 0.5758 0.5122 0.5480 0.6375 0.5313 1 China 0.3667 0.3041 0.2402 0.2604 0.3230 0.3826 0.3403 1
20
5. Empirical Results
This section presents empirical results of this study. The sample includes stock market returns of China and its several major trading partners. The sample period starts from January 1st, 2003 and ends on December 31st, 2016. Before estimate individual markets, several portfolios are being constructed, corresponding to market 𝑖 under estimation. The portfolio includes the rest of the sample except China and market 𝑖. At time 𝑡 , the market return of a constructed portfolio, corresponding to market 𝑖 , is constructed following: 𝑅𝑝𝑜𝑟𝑡/𝑖 = ∑ 𝑤𝑗𝑅𝑗 𝑗 ≠ 𝑖 / ∑ 𝑤𝑗 𝑗 ≠ 𝑖
where 𝑗 indicates the market from the rest of sample, except market 𝑖 and China. 𝑤𝑗 indicates the stock market value of market 𝑗 from the rest of sample markets. For example, Portfolio – US represents the constructed portfolio of the whole sample, but excludes China and U.S.
5.1 Estimation Results of Integration
A. Estimation of China and Portfolio Models
Table 4 details the estimations of China and constructed portfolio model. The result of China market strongly rejects the null hypothesis of no asymmetric GARCH. Wald Test I shows that test of the local information factors rejects the null hypothesis that 𝜹𝑪𝑵 = 𝟎, which implies that the local information factors have significant explanatory
power of China’s market return. Table 4 also reports the estimation of portfolios. Based on Likelihood Ratio Test, all the portfolios are following asymmetric GARCH. Wald Test of local information factors is also being tested as same as China. Chosen local factors have statistic significant effects on market returns for almost all the markets, except for Japan which has a p-value 0.1639. Table 4 also reports test (Wald Test II) for the significant level of portfolio variables on trade exposures to 𝛽̂𝑝,𝑡−1𝐶𝑁 respect with
China. As Wald Test II reports, all the portfolios are significant at 1%, which indicates that the trade linkage is a channel of markets co-movements, whether trade linkage a channel of integration or contagion need to be test for individual markets.
21
Table 4
The China and Portfolio Market Return Model of the Whole Sample The table reports the estimate of the following model:
𝑹𝒑,𝒕= 𝛿𝑝′𝒁𝒑,𝒕−𝟏+ 𝛽𝑝,𝑡−1𝐶𝑁 𝜇𝐶𝑁,𝑡−1+ 𝛽𝑝,𝑡−1𝐶𝑁 𝑒𝐶𝑁,𝑡+ 𝑒𝑝,𝑡
𝑒𝑝,𝑡|𝛀𝑡−1~𝑵(0, 𝜎𝑝,𝑡2 )
𝜎𝑝,𝑡2 = 𝑎𝑝+ 𝑏𝑝𝜎2𝑝,𝑡−1+ 𝑐𝑝𝑒𝑝,𝑡−12 + 𝑑𝑝𝜂𝑝,𝑡−1 2
𝜂𝑖,𝑡= min {0, 𝑒𝑖,𝑡}
The portfolio represents a constructed portfolio consist of market value weight returns of the rest of sample, excluding China and market 𝑖 underestimation. For example, Portfolio – US represents the constructed portfolio of the whole sample but excludes China market and U.S. market. The model selection between asymmetric and symmetric GARCH is based on the Likelihood Ratio test under 10% statistics significant level. The p-value of the Likelihood ratio test is reported. Two hypotheses are tested. Wald Test I is a test of the significance of local information factors. Wald Test II test the significant level of the portfolio variables on the trade exposure to 𝛽̂𝑝,𝑡−1𝐶𝑁 . Sample means and standard deviation of 𝛽̂𝑝,𝑡−1𝐶𝑁 , conditional correlation between China and constructed portfolio (𝜌̂𝑝,𝑡−1𝐶𝑁 ) and the variance ratio of conditional
variances of the portfolio account for China are reported. The definitions of 𝛽̂𝑝,𝑡−1𝐶𝑁 , 𝜌̂𝑝,𝑡−1𝐶𝑁 , and 𝑉𝑅̂𝑝,𝑡−1𝐶𝑁 are following the eqq. (5), (12) and (14).
Market
Model (Asym/Sym) Wald Test (p-value) 𝛽̂𝑝,𝑡−1𝐶𝑁 𝜌̂𝑝,𝑡−1𝐶𝑁 𝑉𝑅̂𝑝,𝑡−1
𝐶𝑁
Asym/Sym p-value I II Mean Std. Dev Mean Std. Dev Mean Std. Dev
China Asym 0.0015 0.0046 Portfolio - US Asym <0.0001 <0.0001 <0.0001 0.1276 0.0439 0.2352 0.1280 0.0717 0.0773 Portfolio - UK Asym <0.0001 <0.0001 <0.0001 0.0771 0.0265 0.1758 0.0999 0.0409 0.0466 Portfolio - DE Asym <0.0001 <0.0001 <0.0001 0.0744 0.0166 0.1741 0.0862 0.0377 0.0377 Portfolio - NL Asym <0.0001 0.0006 <0.0001 0.0777 0.0267 0.1781 0.1016 0.0420 0.0483 Portfolio - AU Asym <0.0001 0.0014 <0.0001 0.0758 0.0266 0.1730 0.0990 0.0397 0.0456 Portfolio - JP Asym <0.0001 0.1639 <0.0001 0.0742 0.0277 0.1648 0.0998 0.0371 0.0456 Portfolio - KO Asym <0.0001 0.0001 <0.0001 0.0746 0.0257 0.1682 0.0976 0.0378 0.0444
22
In Table 4, sample means of conditional betas (𝛽̂𝑝,𝑡−1𝐶𝑁 ) and correlations of portfolios
with China (𝜌̂𝑝,𝑡−1𝐶𝑁 ) are also reported. Portfolio excluding U.S. has the largest trade
risk exposure with China markets with 𝛽̂𝑝,𝑡−1𝐶𝑁 equal to 0.1276. Other portfolios show
a relative smaller 𝛽̂𝑝,𝑡−1𝐶𝑁 around 0.07 and lower standard deviations. The conditional
correlation 𝜌̂𝑝,𝑡−1𝐶𝑁 of portfolio excluding U.S. shows the similar result with 𝛽̂ 𝑝,𝑡−1𝐶𝑁 .
When portfolios excluding Japan or South Korea, the 𝛽̂𝑝,𝑡−1𝐶𝑁 and 𝜌̂
𝑝,𝑡−1𝐶𝑁 show lower
level comparing to other portfolios but more observable in the conditional correlations 𝜌̂𝑝,𝑡−1𝐶𝑁 , with 0.1648 and 0.1682 respectively. For markets in the sample, both China and United States are important trading partners. It disperses the trade exposure with China of the constructed portfolio including U.S. As Chambet and Gibson (2008) find, markets with concentrated trade structure have more integrated financial markets. The portfolio without U.S. has an undiversified trade structure comparing to other portfolios. Moreover, Japan and Korea have a strong connection with both China and United States on trade. In the last columns of Table 4, variance ratios are report to explain the weight of shock from China. Portfolio excluding U.S. has more than 7% of the conditional return variance could be explained by China market’s shock. When portfolios include U.S. market, the variance ratios get lower to less than 5%. This is not a surprise comparing to estimations of 𝛽̂𝑝,𝑡−1𝐶𝑁 and 𝜌̂
𝑝,𝑡−1𝐶𝑁 .
Conditional betas and correlations between China and constructed portfolios are the basements of further cross-sectional and time-series tests of individual markets during crisis periods. The changing of conditional betas or correlations during different crisis periods is reported in Table A1 (Appendix). The Panel A of Table A1 reports the estimation for the Global Financial Crisis. Different with estimation for the whole sample, a symmetric GARCH model is used for China market, which fails to follow an asymmetric GARCH model. The Wald Test rejects the null hypothesis that local information factors have no explanatory power to China’s market return. Same with the previous estimation, portfolios show high confidential levels of asymmetry. An important finding is that all the portfolio could not reject the null hypothesis 𝜹𝑪𝑵 = 0,
which shows returns could not be explained by the local information during the Global Financial Crisis. As 𝛽̂𝑝,𝑡−1𝐶𝑁 show, exposures with China are in similar results with
estimation of the whole sample, around 5%. Correlations with China and variances ratios are lower than the estimation of whole sample period. When portfolio including
23
U.S. market, there are weak conditional correlations with China. Same for the variance ratios, only 1% - 2% of variance could be explained by China’s market, when portfolio including U.S. Since during the Global Finance Crisis, most of volatilities are triggered by Unites States. Moreover, China shows a weak connection with the rest of sample, which means China is independence with other markets. This connection need further analysis by test on individual markets and contagion with China.
For results of European Debt Crisis period in Panel B, based on the likelihood ratio test (p-value = 0.8967), the symmetric model is selected for China market. Different with Global Financial Crisis and the whole sample, the local information sector has no explanatory power to China’s market return. For constructed portfolios, all portfolios have asymmetric GARCH. There is no evidence the market returns could be explained by local information factors for portfolios either. Moreover, the trade exposure shows a strong rejection of 𝛽̂𝑝,𝑡−1𝐶𝑁 = 0. Comparing to estimations of whole sample and Global Financial Crisis period, the risk exposures respect to China are higher, especially for portfolio exclude United States. 𝛽̂𝑝,𝑡−1𝐶𝑁 of the portfolio without United States has the highest 𝛽̂𝑝,𝑡−1𝐶𝑁 with 0.3283. When considering United States market, portfolios show risk exposures around 0.2 to China market. Similar results happen to correlations and variance ratios.
Estimation results of China Market Turbulence are reported in Panel C of Table A1. Based on the model selection test, asymmetric GARCH model is the choice for China. Different with European Debt Crisis, during China Market Turbulence, the local information factors reject the null hypothesis that 𝜹𝑪𝑵= 0 . For six out of seven portfolios, asymmetric GARCH model is chosen, and local information factors do not explain the market return in a statistic meaning, as 𝜹𝑷= 0. The exception is the portfolio without Unites States, it does not follow the asymmetric GARCH model, and local information has explanation power to market return. Similar with former analysis, it could be a result of integration between United States and other markets. However, 𝛽̂𝑝,𝑡−1𝐶𝑁 , the risk exposure respects to China market is still significant at 1% level for all portfolio. 𝛽̂𝑝,𝑡−1𝐶𝑁 are not as high as ones in European Debt Crisis, but do significantly increase from Global Financial Crisis. Betas of portfolios including United States drop from 0.2 to 0.08, but higher than 0.06 in Global Financial Crisis. Without potential impact from United States, more than 10% of market shock could be influenced by
24
China market’s shock. When United States is included, only 4% of market shock could be explained by China market shock.
B. Estimation of Individual Markets
The individual markets are based on three factors: a set of local information factors, a corresponding constructed portfolio factor, and a factor exposure to China. Same with the test on China’s market and portfolios, the asymmetric GARCH model is chosen for all the individual markets after likelihood test. The Wald Test is used to test the null hypothesis of parameters, including 𝜹𝒊, 𝒑𝟏,𝒊, 𝒑𝟐,𝒊, and 𝒒𝒊, are equal to zero. If the null hypothesis that 𝜹𝒊 = 𝟎 could not be rejected, these lagged information factors should not be considered in the model to explain the expected return of market 𝑖. Thus, the asset pricing model is able to capture volatility of conditional return thorough time-varying risk exposure.
B.1.Whole Sample Period
Estimation for the entire sample period is reported in Table A2. The null hypothesis that local information factors are unrelated with pricing errors is rejected by all markets in different significance level, which implies local factors are important to the pricing error to all the market models. Whether the beta to China is influenced by the merchandise trade with China is the previous step to study integration and contagion, so whether trade linkage important is tested by whether 𝒑𝟏,𝒊= 𝟎. Four out of seven markets, Germany, Netherlands, Australia, Japan and South Korea, show a significant rejection of the null hypothesis that 𝒑𝟏,𝒊 = 𝟎, which indicate that trade with China affects, 𝛽̂𝑖,𝑡−1𝐶𝑁 , the conditional betas respect to China significantly. But trade with China is not an important to risk exposure regard to United States and United Kingdom, comparing to other markets. Considering about trade linkage to risk exposure between market 𝑖 and the rest of sample, the portfolio benchmark is being tested by whether 𝒑𝟐,𝒊 = 𝟎. Trade with rest of the portfolio is effective to most individual markets at a significant level, except for United States. The total trade size as GDP affect to betas of China and portfolio is being tested by whether 𝒒𝒊 = 𝟎. This test is to figure out how the trade depth in the economic would influence the exposures to China and corresponding portfolio. Four out of seven markets reject the null hypothesis. But the null hypothesis could not be reject for Germany, Netherlands and Japan at 5% level.
25
Even as common known that economic of Germany, Netherlands, and Japan depend heavily on trade, the trade size as GDP has no explanatory power to the risk exposures (betas) to China and the rest of the sample markets.
Table A2 also reports the average and standard errors of betas, correlations, and variance ratios. The betas and correlations, respect to China, are on small levels for most markets. More specific, conditional beta 𝛽̂𝑖,𝑡−1𝐶𝑁 between China and United States is only 0.0257 and with high volatility. Markets in Europe, United Kingdom, Germany, and Netherlands have similar small betas, 0.0462, 0.0524, and 0.0571 respectively. Consistent with former research, Australia, Japan and South Korea show higher risk exposures with China 𝛽̂𝑖,𝑡−1𝐶𝑁 , as they are in the same geographic region with China. For most markets, risk exposure with the rest of sample is much larger than the one with China, as 𝛽̂𝑖,𝑡−1𝑃 show. Based on the estimation of betas and correlations, the fraction of the return shock variance explained by China is tiny, around 0.3% to 0.6% for United States and European markets and 1.8% to 2.7% for Australia, Japan, and South Korea. The explanatory power of shock from portfolio accounts for 18% to 42% of total shock for United States, United Kingdom, Germany, Netherlands, and Australia. But for Japan and South Korea, the fraction of shock from China and portfolios don’t have noticeable differences. For the whole sample period, no evidence of integration between China market and its main trading partners is proved.
B.2. Global Financial Crisis
Table 5 reports the estimations of three crisis periods. The estimation method is same as the previous one used for the whole sample period. In Panel A of Table 5, the results for the Global Financial Crisis period is reported. Six out of seven markets in the sample fail to reject the null hypothesis that 𝜹𝒊 = 0 at 5% level, which implies that
market returns fail to explained by local information factors. Australia is the only one could reject the null hypothesis at a 5% level. When checking the trade exposure to beta respect to China, only Germany and Australia reject the null hypothesis that 𝒑𝟏,𝒊 = 0 with p-value smaller than 5%. None of the rest markets could reject the this null hypothesis. Three out of seven markets show significantly influenced by trade with other markets at a 10% level. United States, United Kingdom, Netherlands and South Korea do not show significant rejections of the null hypothesis. Only Germany and
26
Australia reject the null hypothesis 𝒒𝒊 = 𝟎 at 5% level, which implies that total
trade size as percentage of GDP is important to explain the risk exposures respect to China and the rest of sample.
Risk exposures to China and the rest of markets (𝛽̂𝑖,𝑡−1𝐶𝑁 and 𝛽̂𝑖,𝑡−1𝑃 ), correlations (𝜌̂𝑖,𝑡𝐶𝑁 and 𝜌̂𝑖,𝑡𝑃 ) are also be considered for subsamples. As 𝛽̂𝑖,𝑡−1𝐶𝑁 and 𝛽̂𝑖,𝑡−1𝑃 show, United States has a negative risk exposure to China and relative high risk exposure to other markets. This is not a surprise since the Global Financial Crisis started from United States and then impacted to global financial systems. During most time of Global Financial Crisis period, China market has a different trend with the rest of markets. United Kingdom, Germany, and Netherlands only show a high exposure to portfolios, with betas to portfolios ranging from 0.9376 to 1.1168, not to China. Furthermore, Japan and South Korea have higher exposures to China market than others do, with 𝛽̂𝑖,𝑡−1𝐶𝑁 equal to 0.1608 and 0.2206 respectively. One exception is Australia, which shows high level of risk exposures to both China (𝛽̂𝑖,𝑡−1𝐶𝑁 = 0.2015) and portfolio (𝛽̂𝑖,𝑡−1𝑃 = 1.1067). Estimation results of 𝜌̂𝑖,𝑡𝐶𝑁 and 𝜌̂𝑖,𝑡𝑃 tell similar conclusion with
betas. Japan and South Korea show similar level of correlations with China and portfolios. Comparing to the rest of sample markets, Australia has strong correlations with both China and portfolio. For United States, United Kingdom, Germany, and Netherlands, tiny part of market shock could be explained by shock from China, as variance ratios show.
B.3. European Debt Crisis
Panel B reports the estimation for the European Debt Crisis. Only Japan and South Korea could reject 𝜹𝒊= 𝟎 under 5% and 10% respectively. For United Kingdom, Germany, and Netherlands, this result is not surprise since as European markets, they bear the brunt. Moreover, since both United States and Australia have strong economic connection with European markets, they suffer from the concern of Euro and stability of European Union as well. Considering the trade exposures to betas respect to China during this period, only South Korea provides a significant p-value (0.0181) to reject the null hypothesis that 𝒑𝟏,𝒊= 𝟎. The estimation for the rest of sample proves that, during European Debt Crisis, the trade with China has no significantly explanatory to risk exposures with China. The tests of significance of the total trade size as percentage
27
Table 5
Implied Statistics of Individual Markets for Subsamples The following model is estimated for the individual market:
𝑹𝒊,𝒕= 𝛿𝑖′𝒁𝒊,𝒕−𝟏+ 𝛽𝑖,𝑡−1 𝐶𝑁 𝜇̂ 𝐶𝑁,𝑡−1+ 𝛽𝑖,𝑡−1 𝑃 𝜇̂ 𝑃,𝑡−1+ 𝛽𝑖,𝑡−1 𝐶𝑁 𝑒̂ 𝐶𝑁,𝑡+ 𝛽𝑖,𝑡−1 𝑃 𝑒̂ 𝑃,𝑡+ 𝑒𝑖,𝑡 𝑒𝑖,𝑡|𝛀𝑡−1~𝑵(0, 𝜎𝑖,𝑡2) 𝜎𝑝,𝑡2 = 𝑎𝑝+ 𝑏𝑝𝜎2𝑝,𝑡−1+ 𝑐𝑝𝑒𝑝,𝑡−12 + 𝑑𝑝𝜂𝑝,𝑡−1 2 𝜂𝑖,𝑡= min {0, 𝑒𝑖,𝑡}
where 𝜇̂𝐶𝑁,𝑡−1 and 𝑒̂𝐶𝑁,𝑡 are the expected conditional excess return and residual of China’s markets, 𝜇̂𝑝,𝑡−1 and 𝑒̂𝑝,𝑡 are the expected conditional excess return and residual
of portfolio excluding market China’s markets 𝑖, 𝑒𝑖,𝑡 is the idiosyncratic shock of market 𝑖. The Wald Test is used to test the null hypothesis of parameters is equal to zero,
including 𝜹𝒊, 𝒑𝟏,𝒊, 𝒑𝟐,𝒊, and 𝒒𝒊. The p-values of the Wald Test are reported. Sample average and standard deviation of beta parameters (𝛽̂𝑖,𝑡−1𝐶𝑁 and 𝛽̂𝑖,𝑡−1𝑃 ), correlations with
China and portfolio excluding market 𝑖 (𝜌̂𝑖,𝑡𝐶𝑁 and 𝜌̂
𝑖,𝑡𝑃) , and variance ratios accounting by China and portfolio (𝑉𝑅̂𝑖,𝑡𝐶𝑁 and 𝑉𝑅̂𝑖,𝑡𝑃) are reported. Standard deviation is reported
in parentheses. Panel A shows estimation for the Global Financial Crisis period, from Aug.7, 2007 to Mar.16, 2009. Panel B reports the result of European Debt Crisis, starting from Dec.8, 2009 to Dec.21, 2012. Panel C presents the China’s Market Turbulence, which starts from Jun.12, 2015 to Feb.15, 2016. For South Korea, since it does not either asymmetric or symmetric GARCH model, it is not reported in Panel C.
Market
Wald Test (p-value)
𝛽̂𝑖,𝑡−1𝐶𝑁 𝛽̂𝑖,𝑡−1𝑃 𝜌̂𝑖,𝑡𝐶𝑁 𝜌̂𝑖,𝑡𝑃 𝑉𝑅̂𝑖,𝑡 𝐶𝑁
𝑉𝑅̂𝑖,𝑡 𝑃
𝜹𝒊 𝒑𝟏,𝒊 𝒑𝟐,𝒊 𝒒𝒊
Panel A. Global Financial Crisis
United States 0.6776 0.8672 0.2733 0.8026 -0.0156 0.5293 -0.0224 0.3556 0.0007 0.1500
(0.0057) (0.0781) (0.0126) (0.0793) (0.0008) (0.0618)
United Kingdom 0.2779 0.1502 0.1371 0.4545 0.0760 1.0677 0.0999 0.6534 0.0120 0.4256
(0.0160) (0.0370) (0.0451) (0.0983) (0.0120) (0.1206)
28
Table 5 - Continued
Market
Wald Test (p-value)
𝛽̂𝑖,𝑡−1𝐶𝑁 𝛽̂𝑖,𝑡−1𝑃 𝜌̂𝑖,𝑡𝐶𝑁 𝜌̂𝑖,𝑡𝑃 𝑉𝑅̂𝑖,𝑡 𝐶𝑁 𝑉𝑅̂𝑖,𝑡 𝑃 𝜹𝒊 𝒑𝟏,𝒊 𝒑𝟐,𝒊 𝒒𝒊 Germany 0.1215 0.0104 0.0668 0.0410 0.0586 0.9376 0.0894 0.6677 0.0129 0.4499 (0.0376) (0.1347) (0.0700) (0.1149) (0.0179) (0.1456) Netherlands 0.0907 0.9029 0.1364 0.4421 0.0987 1.1168 0.1248 0.6792 0.0178 0.4572 (0.0168) (0.0603) (0.0470) (0.1089) (0.0138) (0.1369) Australia 0.0287 0.0000 0.0005 0.0165 0.2015 1.1067 0.2030 0.5629 0.0489 0.3040 (0.0667) (0.0535) (0.0879) (0.1048) (0.0420) (0.1175) Japan 0.1361 0.3535 0.0481 0.8729 0.1608 0.3035 0.1908 0.1966 0.0414 0.0332 (0.0083) (0.0743) (0.0707) (0.0536) (0.0325) (0.0178) South Korea 0.7901 0.4326 0.2119 0.8135 0.2206 0.7051 0.2081 0.3436 0.0494 0.1079 (0.0126) (0.0184) (0.0779) (0.0826) (0.0386) (0.0518)
Panel B. European Debt Crisis
United States 0.5666 0.1194 0.4198 0.0764 0.0307 0.6830 0.0407 0.5095 0.0029 0.3316 （0.0267） （0.0469） (0.0355) (0.1319) (0.0040) (0.1448) United Kingdom 0.3459 0.1258 0.0000 0.0273 0.1013 1.0989 0.1160 0.6384 0.0162 0.4228 （0.0386） (0.0273) (0.0528) (0.1018) (0.0152) (0.1276) Germany 0.5302 0.1298 0.0526 0.0762 0.0604 1.4976 0.0474 0.6256 0.0037 0.4492 （0.0488） (0.1824) (0.0378) (0.1242) (0.0043) (0.1641) Netherlands 0.1871 0.2427 0.0008 0.1629 0.0393 1.3697 0.0360 0.6432 0.0019 0.4778 （0.0284） (0.0352) (0.0250) (0.1214) (0.0023) (0.1601) Australia 0.8981 0.4132 0.0000 0.6935 0.2585 1.0708 0.2524 0.6001 0.0730 0.3058 （0.0215） (0.0898) (0.0964) (0.1258) (0.0613) (0.1221) （Continued）
