Contagion effect among the United Kingdom, European Union, the United States, and Asia excluding Japan private sectors CDS Markets during Brexit shock

University of Amsterdam, Amsterdam Business School Master in International Finance

Master Thesis

Contagion effect among the United Kingdom, European Union, the United States, and Asia excluding Japan private sectors CDS Markets during Brexit shock

Min Jui Wu August 2017

2

Abstract

Globalization makes companies’ business and financial transactions more connected and dependent, while it also prompts the credit risk contagion among different industries and markets especially when crisis happened. Numbers of literatures have covered contagion effects exist in banking system and a few has mentioned the risk contagion between American and European markets in almost most crisis periods. However, contagion should not be limited in banking or other financial sectors because of strong connections between other private sectors’ business operation and financial world. Moreover, more Asian multinational companies expand their business and set up the subsidiaries in the western markets nowadays which means the global credit markets are somehow linked together. Therefore, I extend the research on how private sectors in United States, Europe Union countries and Asia excluding Japan CDS markets connected and reacted to each other when the United Kingdom voted to leave the European Union (the “Brexit” shock) in the mid of 2016.

We applied comovement, volatility spillover and granger causality test concluding little contagion effect in global credit market proving Brexit is not Lehman moment proposed by Schiereck (2016). We also find mutual causality between the U.K. and EU credit markets supporting “No contagion, only interdependence” proposed by Forbes and Rigobon (2002).

3

Contents

1. Introduction ... 4

2. Literature Review ... 9

3. Methodology and Hypothesis ... 14

3.1 Adjusted Correlation ... 14

3.1.1 Hypothesis 1 ... 16

3.2 GJR GARCH ... 18

3.2.1 Hypothesis 2 ... 20

3.3 VAR and Granger Causality ... 20

3.3.1 Hypothesis 3 ... 22

4. Data and Empirical Result ... 23

4.1 Data Overview ... 23

4.2 Empirical Result ... 26

4.2.1 Adjusted Correlation ... 26

4.2.2 GJR GARCH ... 29

4.2.3 VAR and Granger Causality ... 35

5. Conclusion ... 38

4

1. Introduction

The United Kingdom (the U.K.) decided to leave European Union (EU) based on the referendum, 52% voted to leave, on 23 June 2016, which is also commonly called “Brexit”. The result was hard to predict, and out of ordinary expectation, up to 90% polls of Brexit rejection according to Bloomberg. This unexpected black swan effect (Taleb, 2007) cause the turmoil of the U.K. stock markets and credit risk markets especially having significant negative effect on financial institutes’ shares and its counterparty risk (Schiereck, Kiesel and Kolaric, 2016). We would like to see how the U.K. private sector CDS premia return and volatility increase spillover to EU, the U.S. and Asia excluding Japan private sector ones, testing contagion effect among these credit markets after British referendum.

The main reason that Brexit’s shock causing the upheavals to the U.K.’s financial market is political uncertainty. The uncertainty will affect a corporate’s share price, and further influence on its credit risk. In Brogaard (2015) and Baker (2012)’s studies, the government economic policy change (political uncertainty risk) is difficult to diversify and correlated to the cash flow and discount factors, which affect the asset price. Moreover, the effect on the asset price is not only a short-term impact, but an effect that remains significant for two years on stock markets. Woodford Investment Management (2017) has summarized and covered the different aspects of influence on economics caused policy uncertainties after Brexit, including immigration, trade and manufacture, financial service outflows, foreign investment decreasing, which will all bring the negative signals to the companies registered and operated in the U.K., showing the plunge of price in the stock market. Furthermore, corporates’ credit risk will then suffer from the poor share performance caused by the political uncertainty. According to Merton’s model (1974) and KMV model (1997), they calculated the corporates default possibility based on Black-Scholes pricing model, with share price and its volatility. The lower the corporate’s share price, the higher possibility that the firm’s value will be below the bankruptcy threshold (higher corporate’s default possibility), directly reflecting on rising CDS premium. Empirically, Anderson (2012) find the fact that 10% increase in political uncertainty, proxied with VIX, will bring about 3% of positive variation of CDS premia. These evidences above all show the high possibility that the political uncertainty, Brexit, will affect the U.K.’s credit market.

5 However, the question we want to study today is - Does the British political uncertainty only affect the U.K. CDS market, or the turmoil in the U.K.’s CDS premium will also spillover to the one in other regions. Different regions or countries’ credit market are correlated to each other during the crisis. This can be explained in fundamental or non-fundamental factors. Merton (1974), Jacobs and Oviedo (2009) take corporate share price value, volatility and face value of bond as fundamental factors evaluating the credit risk, while Anderson (2012) found that fundamental factors account for less than 20% of the increasing correlation among 150 corporate CDS premium between 2007 and 2009. Instead, the increasing correlation can be explained by the non-fundamental factors such as liquidity risk, especially in financial sectors. However, Longstaff (2005) argue that CDS premium volatility are not connected via liquidity risk in the crisis; also, many researchers like Norden and Weber (2009) found fundamental factors of 58 international companies’ stock market contagion is the reason of increasing correlation of their credit market in 2008 financial crisis. However, other than studying on the factors causing increasing correlation, we would like to see if the contagion effect exists in credit markets after Brexit referendum.

Contagion effect has been broadly discussed and studied through different channels (stock, bond, currency, credit risk, real estate market) in different financial shocks (1997 Asian Crisis, 2007 Subprime mortgage bubble, 2009 European Sovereign Debt Crisis, 2016 Black Swan from political risk), we will summarize these findings in the literature review section. Before our further studies, we need to define “contagion effect”. There are few researchers used “contagion”, which is mostly related to medical disease spread over different regions, as an economical and academic term before Asian Financial Crisis in 1997. The first researchers studying on the economic contagion are Calvo and Reinhart (1996), they defined the contagion as “the transmission of crisis from one specific country to others due to fundamental link or other premia channels like investors behaviors” (also called as “true contagion”). More specifically, Forbes and Rigobon (2002, p 2223) have clearly defined contagion as “a significant increase in the cross-market linkage during the period of turmoil.” or “how an initial country-specific shock was rapidly transmitted to markets of very different sizes and structures around the globe.” They also quantify contagion by seeing increasing correlation during financial crisis, which is also called as “shift contagion” in their study. According to the Work Bank, there are three definitions of contagion- the broad definition, the

6 restrictive definition and the most restrictive definition. The broadest definition of contagion is “general spillover effect cross countries, and transmission of shocks, either positive or negative shocks, and do not need to be a crisis.” The restrictive definition referred to “the transmission of shocks beyond any fundamental link among the economies.” The most restrictive definition of the contagion occurs “when the correlation of cross-countries’ assets significantly increase during the crisis periods.” In first part of my study, I will test contagion following Forbes and Rigobon (2001)’s “shift contagion” definition; this is the most common definition in several contagion researches such as Kaminsky (2003), Bae (2003), Coudert and Gex (2008). One more relevant thing we can find in Forbes and Rigobon (2002)’s study is that they found the fact that “shift contagion” effect disappear after adjustment for heteroschasticity and endogenous in the crisis in 1990s and conclude “no contagion, only interdependence” (will be explained in the literature review part). I will follow their methodology adjusting the private sectors CDS premium correlation between the U.K. and other regions to see whether the “shift contagion” exist after the British referendum.

Even though testing correlation increase between markets is a prevailing method to measure contagion, several researchers have criticized Forbes and Rigobon (2001)’s in different ways. Stijn Claessens (2001) has mentioned some misconceptions related to Forbes and Rigobon (2001)’s definition and calculation. He argued that even if we can’t see the significant increase statistically in cross-countries, it is ungrounded and arbitrary to say that there is no contagion in these countries. On the other hand, some economists even argue that employing correlation methodology to test contagion is too simple to see the propagation mechanisms which is mentioned in Masson (1997). Therefore, we test contagion with different methodology making the contagion test more robust. Although Forbes and Rigobon (2001) argue that the correlation test may be biased as correlation between two countries are conditional on their volatility (which is one of the reason they develop the adjusted correlation model to get rid of volatility inconsistent before and after crisis), several researches proved that volatility spillover is an important factor lead to contagion which cannot be overlooked or ruled out proposed by King and Wadhwani (1990). Kolb (2011) found the volatility of different companies’ CDS premia rose sharply average from 40 percent to 60 percent in the GM default crisis in 2005 and concluding the contagion in that crisis is stem from volatility changes. Therefore, we follow previous research (Nicolo and Kwast, 2002; Schroeder and Schueller, 2003)

7 applying GARCH model testing volatility spillover as this model allow us to capture both first and second moment, variance, of CDS premia change to measure contagion effect.

In the second section, we will discuss how the previous researchers’ views on financial contagion in different markets, especially focus on a few studies in CDS markets contagion effect. Besides, we will mention how previous research measured the contagion, and the correction on correlation coefficient due to the problem of bias. Moreover, we will introduce the GJR-GARCH model, Granger Causality model and related literatures. Furthermore, we will present related formula, adjustment of the models for measuring the contagion effect and also list out our hypothesis based on the literatures and models.

In last section, we perform our empirical study following three steps examination of comovement, asymmetric volatility spillover, and granger causality test. We found significant comovement (shift contagion) of the CDS premium in banking, other finance, consumer, manufacturing and TMT credit market during the Brexit uncertainty period and the evidence that stronger credit market correlation in financial sector among the U.K. EU and the U.S. one year after referendum. However, combining the result of volatility spillover test, we can only see the contagion still exist in financial sectors of EU and TMT market over the world one year after the vote. After introducing Granger causality test, we see only interdependence rather than contagion in British and EU other finance sector market during the Brexit uncertainty period.

There are several contributions of our study. The evidence based on our result support Forbes and Rigobon (2002)’s finding (no contagion, only interdependence) not only in other financial stock market of 1990s crisis but also valid in the in other financial sector credit market in 2016 Brexit shock. Moreover, most previous research found contagion from European to American banking CDS premia in 2009 European debt crisis. However, our research find contagion only show up in the short period after Brexit but disappear one year after Brexit shock in US credit market, which implies more resilient credit risk management system in the United States. Furthermore, rather than only use one methodology testing contagion effect, we conduct stricter test of the contagion with both shift contagion (comovement) on CDS premia and CDS premia volatility spillover effect from the U.K. to other credit markets. Last but not least, we consider the endogeneity issue in our testing

8 period and apply granger causality to test contagion when we see endogeneity suggested by Forbes and Rigobon (2002), and we are the first one introduce granger causality testing private-to-private contagion effect after the U.K. referendum.

9

2. Literature Review

In this section, we will review some previous literatures on contagion effect. Initially, we will briefly mention some papers studying contagion effect in different financial markets such as stock and bonds with their study methodology and results. Moreover, we would like to put our focus on the literatures related to the measurement of contagion we will use in our papers - adjusted correlation, GJR GARCH and Granger causality. Also, some literatures will be introduced because of their comments and improvement related to the methodology we use.

Several empirical studies focus on the contagion effects in financial markets for different crisis periods, and most of them proved the existence of contagion effect with different measurement. Dungey and Martin (2002) employed Latent factor model to test contagion effect during the crisis of Russian default and LTCM collapse in 1990s. The result showed that contagion effect account for 8 to 17 percentage of the volatility increase at the period, and both developed and developing countries’ bond markets suffered from the crisis. Longstaff (2010) studied the U.S. CDO and corporates’ bond market before 2008 financial crisis with correlation test, and proved contagion in these two markets. We can also see even more academic evidence of contagion effect on stock markets across different countries such as Baele and Inghelbrecht (2010) with dynamic factors model, Baur and Schulze with quantile regression, Fry McKibbin (2013) with co-skewness and Lagrange multiplier tests, Bodart and Candelon (2009) with frequency domain approach, and Grammatikos and Vermeulen (2012) with GARCH, Khalid and Rajaguru (2006) with multivariate GARCH, and Li (2013) with GJR-GARCH.

Among all methodologies measuring contagion in different financial markets, we would like to summarize four of them which is most prevailing and related to our study: Cointegration, regression, GARCH, across-countries correlation coefficient and regression.

Cointegration focus on the long-term relationship between two markets. We need to test the change in co-integrating vector between two markets. Longin and Solnik (1995) used this approach to test contagion in seven OECD countries from 1960 to 1990, and they find that co-integrating coefficient increase 0.36 between the U.S. stock market and the one of other countries during this period.

10 Goudert and Gex (2008) alsouse VECM model to see co-integration, spillover and lead-lag effect between CDS markets and stock markets. In our paper, however, we do not take co-integration approach into consideration. It is because that, as Stijn Claessens, Kristin J. Forbes (2001) mentioned, if two markets are co-integrated, they have the real linkage lasting overtime, the significant relationship between the countries may be built up on their economic trading partnership, rather than the shock; thus, we will not call it contagion based on our definition in previous part.

A few researchers use regression model to directly test the propaganda mechanism and channels in specific events and between particular countries or industries. Schiereck (2016) used the method to compare the U.K., EU, Non-EU banks’ CDS and stock market affected by the 2008 financial crisis and 2016 Brexit announcement. He found that the short-term banks’ share price plunged and CDS premia increase after Brexit announcement in 2016 is much less significant than the impacts of Lehman Brothers bankruptcy filing in 2008. Moreover, he found that the CDS premia increased mainly happened on EU banks; hence, he concluded that the rising counterparty risks for EU banks caused by political uncertainties has little contagion effect to banks in other region because of the enhancement and more resilient of financial system.

Among all contagion measuring approaches, testing contagion based on correlation coefficient changes is the most common-used one because it is the most intuitive and straight forward way. However, this methodology may also cause the misconception due to the exist of heteroscedasticity and endogeneity. Forbes and Rigobon (1999)’s study is the first one defining contagion as significantly increase of correlation between markets during the crisis. If two markets correlation is high not only in the unstable period, but also in tranquil time, then we can only conclude the “interdependence” rather than contagion. Moreover, they also mention that the correlation coefficient cannot be the accurate measurement of contagion. It is because that the volatility of the market increase during unstable period will also increase cross market correlation (see Boyer, 1997); therefore, they recommend adjust the correlation ruling out the impacts of different volatility (also called as heteroscedasticity) in tranquil and unstable period. They applied the adjusted correlation coefficient on 28 emerging countries’ stock index in 1987 U.S. stock market crisis, 1994 Mexican currency crisis, and 1997 Asian Financial Crisis. They found that the unadjusted correlation coefficients of half of samples have significantly increased after the crisis, while for the

11 rise of adjusted correlation coefficients after crisis was not significant. Hence, they concluded that the correlation coefficients adjusted for heteroscedasticity proved no contagion but interdependence in 1987,1994 and 1997 crisis. Moreover, in CDS markets, Goudert and Gex (2008) follow their approach and studied the contagion effects among European and American investment grade companies’ CDS premia during the General Motors (GM) and Ford downgraded crisis in 2005. They found correlation significant increase (contagion) during the first week after the GM and Ford’s default in all private sectors with unadjusted correlation coefficient, while only interdependence can be found in all sectors except finance, consumer and manufacturing sector after adjusting the correlation coefficient. In our paper, we do not compare the correlation difference before and after adjustment, we directly use the adjusted correlation methodology proposed by Forbes and Rigobon (2002) testing comovement of CDS premia in different private sectors after Brexit.

Forbes & Rigobon (2002)’s methodology only discuss the comovement of the assets index or return instead of volatility inconsistence before and after the crisis. However, correlation increase also caused by higher volatility spillover in the crisis period (Engle, 1990), so only test the comovemet during crisis in Forbes & Rigobon (2002)’s study seems not be sufficient to capture contagion. Furthermore, Groba (2013) test the mean and volatilty contagion within european countries CDS markets in European sovereign debt crisis with GARCH model. He found the volatility spillover from PIIGS to other countries is more significant and lead credit default swap return comovement. Therefore, in our study, we also want to test volatility spillover effect among provate sectors credit market after Brexit. In fact, before 2008, most researches study on contagion focus on the impacts of first moment, log return, while after Beirne and Caporale (2009) mentioning the importance of second moment impacts, volatility, more researchers using different kinds of variance model to capture volatility spillover during the crisis. They capture the volatility of mature and emerging markets in stable period and unstable period testing if the volatility in different period change significantly. If the volatility increase significantly after the crisis, then they conclude the volatility spillover and contagion exist. Their result suggests contagion of stock volatility spillovers from mature markets do impact the volatility of returns in emerging stock markets. Similar to this methodology, several researhes catch volatiliy of the underlying asset in defferent ways, and the most prevailing variance models are VAR, ARCH, and especially GARCH models. However,

12 theses variance econometric model cannot capture the volatility increase due to the negative shock (asymmetric volatility), which is important because negative past news will increase volatility more than positive news (Black, 1976), and we only want to see how previous negative information such as crisis can affect current volatility. Engle & Ng (1993) have studied on the variance asymmetry in Japanese stock market, and compared the goodness-of-fit of EGARCH, VGARCH, NGARCH, and GJR GARCH models. They found that GJR-GARCH is the best model in terms of goodness-of-fit to test asymmetric volatility conditional on previous bad news (crisis). Fornari & Mele (1997) also use the model to find the fact that the asymmetric volatility existed among the U.K., the U.S., Singapore, Italy, and Japan’s stock market. Therefore, we also apply GJR GARCH to procced our study on volatility spillover effect between the U.K private sectors CDS market to the one of other region. Only we see significant comovement of CDS premia return and volatility spillover from the U.K. to other region credit market can we conclude the contagion exist.

Another trend to analyze contagion effect is applying Granger Causality Test, which can not only see the linkage between two markets, but also see how the risk spillover from one market to the other. Therefore, the reason why Granger Causality has been widely used in contagion test is that we can see the contagion direction rather than merely observing contagion phenomenon between two markets (Marta and Simon, 2014). Another important reason researchers like to apply granger causality to test contagion is that they may have no idea where is the exact source of the crisis or endogeneity problem occur in our samples. Forbes and Rigobon (2002) suggest use granger causality methodology to test contagion if the crisis source is not defined and endogeneity. In our study, we cannot make sure the correlation of CDS premia increase between the U.K. and other regions in our defined unstable period is only caused by Brexit shock. Also, we cannot make sure if the turmoil in other regions credit market will influence back to the U.K. ones. For example, Trump’s victory brings uncertainty of financial regulation in the United States also raise American banks’ CDS premia in the end of 2016 (Britta and Florian, 2016). Therefore, neither can we confirm whether the comovement of raising CDS premia between the United States and the U.K. is caused by the Brexit shock or American election shock (crisis source cannot be well defined), nor can we make sure if American election shock will spill over to the British credit market (endogeneity). Hence, in our study, we also have to test contagion effect with granger causality. In fact, several researchers have proved contagion exist with granger causality test in public-to-public and

private-13 to-public CDS premia. Kalbaska and Gatkowski (2012) applied granger causality test on PIIGS, France, U.K. and Germany CDS market, and found that Greeks CDS premia changes granger caused CDS premia of other Southern European countries, which suggested Greek CDS markets is the core problem of the 2009 Eurozone sovereign contagion, proving the public-to-public contagion in credit market. Alter and Schuler (2011) also use VAR and granger causality test studying the contagion between public and private sector. They test causality of seven sovereign CDS of EU countries and the one of EU banks, proving banking CDS premia upheaval spillover to the sovereign ones before governments’ bailout, and confirming private-to-public contagion. However, no research applies granger causality test on private-to-private CDS premia contagion effect; therefore, we will make the contribution testing the contagion and spillover direction with granger causality test.

14

3. Methodology and Hypothesis

In our study, we will test the contagion effect based on the comovement test with adjusted correlation and volatility spillover test with GARCH (Chiou, Wu and Wang, 2015) between British private sectors CDS market and other region CDS market (European Union private sectors, the United States private sectors and Asia excluding Japan private sectors). After the shock of Brexit, the political uncertainty impact on the U.K. CDS premia and spillover to other countries in EU (Belke, 2017). Therefore, we believe the rising of U.K. CDS premia and volatility may also be contagious to other countries outside EU. We expect contagion could be found in several sectors especially in banking and other finance sectors as financial market is closely connected and CDS in financial market is more sensitive to liquidity (non-fundamental factors) mentioned in Anderson (2012), so liquidity shortage in the global funding market during crisis will affect the credit market all over the world (Darolles, 2012) (Siriwardane,2015); therefore, we believe contagion can be found in banking and other finance sector CDS market globally. For those sectors which are more isolated and confined to local regulation and markets, we believe there is no contagion effect after Brexit shock. To test contagion, and being stricter on our study, we can only conclude contagion effect exist when we both find the evidence of comovement and volatility spillover with adjusted correlation test and GJR GARCH test respectively. After confirming the contagious indexes, we then test Granger casualty to see the impact direction. During the time in middle to the end of 2016, Brexit is the major shock affecting the financial markets; therefore, we expect the U.K. CDS market granger cause others and no reverse causality.

3.1 Adjusted Correlation

Testing correlation of two markets during tranquil and unstable period separately to see whether the correlation increase significantly in unstable period. This methodology is based on the theory proposed by Forbes & Rigobon (2002). In their study, they define shift contagion as “a significant increase in cross-market linkages after a shock to one country (or group of countries).” They mention that even though we see high degree of comovement in unstable period, we cannot conclude contagion as it is possible that the high degree of comovement may also happen in tranquil period due to daily business, transaction and economical linkage between markets, which is defined

15 as interdependence rather than shift contagion in their study. Follow their definition and methodology testing shift contagion, we then test the correlation between markets as follows. The correlation coefficient (conditional correlation) is

𝜌𝑡𝑐 = 𝜎𝑖𝑗

𝜎𝑖𝜎𝑗

(1)

where 𝜌_𝑡𝑐 _{is correlation coefficient conditional on market volatility, i represent for the U.K. private}

sectors credit market and j represent for the ones in EU, the U.S., or Asia excluding Japan. However, Forbes & Rigobon (2002) mentioned that the conditional correlation would be upward biased due to heteroscedasticity and endogenous. According to them, we have to firstly calculate the conditional correlation and δ, the relative increase in conditional correlation in the country started the crisis, and calculate the unconditional correlation coefficient, which shows in terms of following formulas, 𝜌_𝑡𝑢𝑛 = 𝜌𝑡𝑐 √1+𝛿𝑡[1−(𝜌𝑡𝑐)2] , 𝛿𝑡 = 𝜎_𝑖𝑖𝑐𝑟𝑖𝑠𝑖𝑠 𝜎_𝑗𝑗𝑡𝑟𝑎𝑛𝑞𝑢𝑖𝑙 -1 (2)

where 𝜌_𝑡𝑢𝑛 is unconditional coefficient which is adjusted to eliminate heteroscedasticity and endogeneity problem; 𝜎_𝑖𝑖𝑐𝑟𝑖𝑠𝑖𝑠 is the variance during uncertainty or crisis and 𝜎_𝑖𝑖𝑡𝑟𝑎𝑛𝑞𝑢𝑖𝑙 is the variance during the tranquil period. In Forbes & Rigobon (2002)’s study, the 𝜌_𝑡𝑐 _{will increase}

during the crisis period while the adjusted correlation coefficient, 𝜌_𝑡𝑢𝑛_{, will stay stable even in the}

period with high volatility if two countries have no shift contagion effect (interdependence).

In our paper, we first calculate the unadjusted (conditional) correlation coefficient in tranquil period (𝜌₀𝑐) and Brexit period (𝜌₁𝑐) respectively, then we adjust the correlation coefficient based on formula (2) into adjusted (unconditional) ones, 𝜌₀𝑢𝑛 and 𝜌₁𝑢𝑛. If adjusted (unconditional) correlation coefficient in the Brexit period significantly larger than the one in the tranquil period, we then claim the shift contagion effect (comovement) exist. We will use Fisher Z test as our test statistics to see whether the adjusted (unconditional) correlation increase significantly.

𝐻₀ ∶ 𝑍₁ ≤ 𝑍₀; 𝐻₁ ∶ 𝑍₁ > 𝑍₀ 𝑍₁ = 1 2ln ( 1+𝜌1𝑢𝑛 1−𝜌₁𝑢𝑛), 𝑍0 = 1 2ln ( 1+𝜌0𝑢𝑛 1−𝜌₀𝑢𝑛);

16 E (𝑍₁-𝑍₀) = 𝑍₁-𝑍₀, Var (𝑍₁-𝑍₀) = 1 𝑛1−3 + 1 𝑛0−3; Test Statistic Z = 𝐸(𝑍1−𝑍0) √𝑉𝑎𝑟(𝑍1−𝑍0)

where 𝜌₁𝑢𝑛 is adjusted (unconditional) correlation after Brexit shock, 𝜌₀𝑢𝑛 in tranquil period, 𝑛₁ is total samples after Brexit shock and 𝑛0 is total samples before Brexit uncertainty.

3.1.1 Hypothesis 1

According to recent studies introducing Forbes and Rigobon’s methodology on credit market contagion issue - Markose (2012), Andenmatten (2011), Longstaff (2010), Coudert and Gex (2008), and Belke (2017) - contagion effect exists in credit derivative markets of the U.S., European Union and some Asia countries in 2007 to 2009 financial crisis, 2009 to 2012 European debt crisis and 2017 Brexit vote. However, most papers study on the CDS premia of country or corporate-level; only Coudert and Gex (2008) cover the contagion effect sector by sector in 2005 GM and Ford default crisis. In our study, we also want to test the contagion effect on private sector level to see which industry is the main sector we can see contagion after Brexit. Therefore, we test shift contagion of seven sectors - Bank, Consumer, Energy, Manufacturing, Other Finance, Service and TMT. According to Goudert and Gex’s (2008) study on contagion effect of financial crisis in different private sectors, they have found that contagion existed in all sectors before adjustment of correlation coefficient, while contagion only significant existed in consumer, financial and manufacturing sectors after adjusting the correlation coefficient. This indicates that the correlation increase can only be seen in these sectors because of fundamental and policy linkage changes rather than volatility increase (see Boyer, 1999). Due to Brexit, companies with international business such as finance and consumer sectors have decided to make the operation decision, moving headquarter in London, laying off employees based in the U.K. etc, to deal with political uncertainty. Therefore, we expect to see the significant correlation change due to the major corporates or industry’s policy changes in banking (Hypothesis 1), other finance (Hypothesis 1-2) and consumer (Hypothesis 1-3) sector rather than the increasing correlation caused by rising volatility. For the increasing correlation of manufacturing sector mentioned in Goudert and Gex’s study, we believe it is related specific industry shock, GM and Ford’s default, so we expect no shift contagion can be found this time in Brexit uncertainty. As long as the Z test of any one markets

17 (EU, the U.S., AXJ) significantly higher after Brexit uncertainty period, we then reject null hypothesis and conclude the existence of shit contagion.

Null Hypothesis 1-1 𝐻₁₋₁𝑏𝑎𝑛𝑘∶ 𝑍₁ ≤ 𝑍₀; Null Hypothesis 1-2 𝐻₁₋₂𝑜𝑡ℎ𝑒𝑟 𝑓𝑖𝑛 ∶ 𝑍1 ≤ 𝑍0; Null Hypothesis 1-3 𝐻₁₋₃𝑐𝑠𝑚∶ 𝑍1 ≤ 𝑍0;

Moreover, different from Goudert and Gex’s (2008) study where CDS market’s respond to GM default crisis was really prompt, Brexit shock’s impact should be long lasting instead of dramatic and short-term influence into other countries. Policy and Economic uncertainty spread over to financial and non-financial private sectors, international companies are now planning to move out headquarters from the U.K. to Europe continent because the U.K. loss the role as the gate to EU markets. International capital also being hesitate on the investment to the U.K. According to London School of Economics, the foreign direct investment estimated to decrease 22% over the next 10 years after 2016 Brexit referendum, which may lead to capital shortage from outside to the U.K.’s corporates. Without new funds, this implies the potential inability to payout the old debt, and limits corporates’ growth momentum, increasing the long-term credit risk for the U.K., even spill over the risk to the rest of the world in the following years. Therefore, we also believe the contagion effect will exist for at least one year after the referendum.

𝐻₀ ∶ 𝑍₂ ≤ 𝑍₀; 𝐻₁ ∶ 𝑍₂ > 𝑍₀ 𝑍₁ = 1 2ln ( 1+𝜌1𝑢𝑛 1−𝜌₁𝑢𝑛), 𝑍0 = 1 2ln ( 1+𝜌0𝑢𝑛 1−𝜌₀𝑢𝑛);

where 𝜌₂𝑢𝑛 is adjusted (unconditional) correlation one year after Brexit shock, 𝜌₀𝑢𝑛 in tranquil period, 𝑛₂ is total samples one year after Brexit shock and 𝑛₀ is total samples in tranquil period.

18

3.2 GJR GARCH

Even though Forbes & Rigobon (2002)’s methodology take heteroscedasticity and endogeneity into consideration, while they only discuss the comovement of the assets instead of volatility inconsistence before and after the crisis. We cannot overlook the correlation increase caused by higher volatility in the crisis period (Boyer, 1999). Moreover, other than testing shift contagion, the extensive literature also test on volatility spillover on financial markets (Engle, 1990). In fact, credit market is more sensitive and connected in volatility than in CDS premia (Groba, 2013). He showed significant cross-broder CDS premia volatility spillover from distressed EU countries to others in European countries during European debt crisis, volatility spillover even leading default swap return contagion. Therefore, we believe volatility spillover is still essential when we want to check if contagion exist in markets. Most studies capture volatility spillover with GARCH models. They proposed to use GARCH model to adjust the correlation coefficient to make the test more reliable and objective. However, in GARCH model, where current variance is composed of only previous conditional variance and square of residual, has been criticized for non-accountability of the asymmetric reaction of negative shock for volatility nowadays. Asymmetric reaction is important issue in our study if we want to test volatility spillover during crisis period, because negative past news will increase volatility more than positive news, which is so called as leverage effect (Black, 1976).

TGARCH (GJR-GARCH, 1993) or EGARCH (1991) can be appropriate models to deal with leverage effect, while Engle and Ng (1993) proved that EGARCH model will overestimate the volatility, and GJR GARCH is the model with better good of fitness to deal with volatility asymmetry or leverage effect. Therefore, we followed Engle and Ng’s study and introduce dummy variable as proxy of bad news, capturing how previous conditional variance caused by bad news can affect the current variance (Glosten, Jagannathan and Runkle, GJR GARCH, 1993)

If 𝑦_𝑡 represent for the return of the any financial assets in period t, then

𝑦𝑡= 𝑎0+ 𝜀𝑡 (3)

where 𝜀_𝑡 is the residual. If 𝜀_𝑡−1< 0, it means that the previous period return is less than 𝑎₀, which bring the bad news to the investors. In the contrast, if 𝜀_𝑡−1≥ 0, then at least this is not a

19 bad news from previous period. We extend our previous bad news to k period lag, then GJR GARCH formula is presented as follows:

𝜎𝑡2 = 𝜔0+ ∑ 𝛼𝑖𝜀𝑡−𝑖2 𝑞 𝑖=1 + ∑ 𝛾𝑘𝜀𝑡−𝑘2 𝐷𝑡−𝑘 𝑟 𝑘=1 + ∑ 𝛽𝑗𝜎𝑡−𝑗2 𝑝 𝑗=1 (4)

where 𝜀_𝑡~𝑁(0, 𝜎_𝑡2) , and 𝐷_𝑡 is dummy variable for market information, 𝛾_𝑘 is estimated coefficient represent for asymmetric volatility.

𝐷_𝑡−𝑘 = { 1 𝑖𝑓 𝜀𝑡−𝑘 < 0 (𝑏𝑎𝑑 𝑛𝑒𝑤𝑠)

0 𝑜𝑡ℎ𝑒𝑟𝑤𝑖𝑠𝑒 , 𝑓𝑜𝑟 𝑘 = 1,2, … 𝑟.

In our study, we would like to apply Z test to see if estimated asymmetric volatility, coefficient γ, of CDS premia in different private sectors of the U.S., EU, or Asia excluding after Brexit shock rise significantly. As mentioned in Chiou and Wu (2015)’s study, crisis or shock (bad news) will cause significantly lifting volatility to other markets; therefore, other than test on the correlation of the CDS premia, we will also see rising asymmetric volatility from the U.K. CDS markets will spillover to other markets volatility after the Brexit referendum announcement.

𝐻₀ ∶ γ₁ ≤ γ₀; 𝐻₁ ∶ γ₁ > γ₀ 𝐻0 ∶ γ2 ≤ γ0; 𝐻1 ∶ γ2 > γ0

where γ₀ is asymmetric volatility in tranquil period, γ₁ is asymmetric volatility in Brexit period, γ₂ is asymmetric volatility after Brexit.

We can only conclude contagion exist if two methodology show significant increase on testing statistics, which means only can we see contagion effect in Brexit shock when we see both shift contagion (comovement) and asymmetric volatility rise (volatility spillover).

20 3.2.1 Hypothesis 2

According to previous researches, during crisis, contagion effect obviously exists in financial sectors, even sometimes the financial private sectors credit volatility will have impacts on sovereign credit market (Dieckmann and Plank, 2011). Therefore, we believe we can find contagion effect in banking and other finance system in EU (Hypothesis 2-1) and The U.S. market (Hypothesis 2-2) both in Brexit and one year after referendum. However, for most Asian financial corporates, their main operations are focus on the regional markets instead of global business such as westerns financial institutes do. Therefore, we do not think the CDS premia in Asia financial sector will comove with the one in the U.K. during Brexit. We also believe volatility increase of CDS premia in the U.K. financial market is confined to Europe, at most to USA, region, little turbulence will be transmitted to Asia. Therefore, we believe only interdependence rather than contagion can be found in Asia financial credit market during Brexit uncertainty period (Hypothesis 2-3). We expect the both comovement and volatility spillover will show up in consumer (Hypothesis 2-4) and TMT sector (Hypothesis 2-5) in the short run at least in one of the three regions, because the telecom, media, internet of things and consumer products trading is connected together due to business globalization and the advent of global on-line shopping. while in the long run the effect will be vanished, because Therefore, we may only see interdependence. For energy sector, most utility system in different countries are highly regulated and managed by local government and state-owned corporates; thus, we believe there is no contagion effect from Brexit shock (Hypothesis 2-6).

3.3 VAR and Granger Causality

After screening out the markets with significant contagion effect by applying adjusted correlation method and GJR GARCH method, we would like to see what is the causality between those contagious markets.

We need to construct the Vector Autoregression (VAR) model to choose the appropriate length of lag (with AIC and SIC methodology) and proceed Granger Causality Test. The constructing framework is presented as follow:

21 We have to make sure the sample we used are stationary data; otherwise, the spurious regression may occur (Granger and Newbold, 1974). Unit Root Test is the basic and important methodology to test if the variables are stationary or integrated. We will conduct Augmented Dickey-Fuller (ADF) test to test the null hypothesis that the DGP has unit root (integrated). If we can not significantly reject the null hypothesis, then we need to take the first order difference data as our study samples.

After confirming our series is stationary, we need to select the optimal lag length for VAR model. There are some models widely employed to examine the model’s goodness-of-fit and find the optimal lagged period such as Akaike information criterion (AIC) and Schwartz Bayesian information criterion (SBC). The smaller the figures of AIC and SBC are, the better forecastability of the model. In our study, we apply AIC as our test of goodness-of-fi as the results of AIC will be more consistent compared to SBC with different series samples.

We choose VAR rather than other multivariate time series model such as VECM for Granger causality which proposed by Coudert and Gex (2008) because of following advantages. First, we have no clue in advance if there is long-term cointegration among our sample series, and VAR can be used when the cointegration structure is unknown. Second, we do not know beforehand about which variables are endogenous and which are exogenous. Thus, we do not have to restrict Brexit shock as exogenous variable, limiting the examination of contagion. Furthermore, Brooks (2014) compared out-of-sample forecasting accuracy among ECM, ECM-COC, ARIMA, VAR, RMSE models and concluded that VAR provided better forecast accuracy. Therefore, some empirical researchers prefer using VAR as their model to conduct structure analysis and forecasting including Granger Causality Test, Impulse Response Function and Forecast Error Variance Decomposition.

22 Granger (1969) proposed the intuitive concept of forecasting ability with VAR model, rather than implying the true causality. If a group of variables (X1) can be used to predict another group of variables (X2), then X1 granger cause X2. If X2 fail to granger cause X1 then the lagged tems of X2 can not explain X1, which can be dipicted as the following model.

(X1𝑡 X2𝑡) = ( c1 c2𝑡) + ( 𝛼₁₁1 0 𝛼₂₁1 _𝛼 221 ) (X1𝑡−1 X2𝑡−1) + ⋯ + ( 𝛼₁₁𝑝 0 𝛼₂₁𝑝 𝛼₂₁𝑝 ) ( X1𝑡−𝑝 X2𝑡−𝑝)+( ε1𝑡 ε2𝑡) (5)

Similarly, if X1 failes to explain X2, then all the coefficients of lagged value of X1will be zero.

(X1𝑡 X2𝑡) = ( c1 c2𝑡) + ( 𝛼₁₁1 𝛼₁₂1 0 𝛼₂₂1 ) ( X_1𝑡−1 X2𝑡−1) + ⋯ + ( 𝛼₁₁𝑝 𝛼₁₂𝑝 0 𝛼₂₁𝑝 ) ( X_1𝑡−𝑝 X2𝑡−𝑝)+( ε1𝑡 ε2𝑡) (6) 3.3.1 Hypothesis 3

In our study, variables (indexes) for Granger Causality Test are selected after the adjusted correlation method and GJR GARCH Method; hence, we expect that CDS premia of U.K. industry sector i (i represent for contagious sectors) will granger cause CDS premia of other regions industry sector i. (Hypothesis 3-1) in Brexit uncertainty period and one year after referendum period. Moreover, according to Kalbaska and Gatkowski (2012), after August 2008, the U.K. fail granger cause a big distress in the Eurozone in normal period. Therefore, we assume that U.K. industry sector i CDS market will fail granger cause other region industry sector i CDS market in tranquil period.

Null Hypothesis 3-1

𝐻₃₋₁1 ,2 : U.K. i --/-> Other Region i, in Breixt period and one year after referendum

Null Hypothesis 3-2

23

4. Data and Empirical Result

4.1 Data Overview

Our paper focus on CDS premia study in the U.K., EU, the U.S. and Asia excluding Japan markets. To be more specific, we break down overall CDS market into six private sectors- Bank, Consumer, Energy, Manufacturing, Other Finance, Service and TMT CDS indexes. We would like to study on how the rising uncertainty in the U.K. sectors’ counterparty risk caused by Brexit influencing the sectors’ credit risk in different regions. Thomson Reuters iTraxx database provides complete data of the U.K. sovereign CDS premia and private sectors’ CDS indexes in four regions. Our sample contains daily data on 852 CDS premia index in different sectors. In order to construct the research model, we select the most liquid five-year on-the-run CDS index from Datastream and Thomson Reuters from 6/23/2014 to 6/23/2017, and separate the samples into three period - Tranquil and Brexit uncertainty period and One year after the referendum.

In order to define the exact beginning and ending dates for each period, we need to examine the movement and volatility of the U.K. sovereign CDS premia and the volatility of overall private sectors’ credit market index.

Chart 1: The U.K. 5-year CDS premia, in basis points.

15 20 25 30 35 40 45 50

I II III IV I II III IV I II III IV I II III

2014 2015 2016 2017

Source: Datastream Brexit Referendum Result 6/23/2016

24 Breixt and against-Brexit disputes started from the David Cameron’s victory in 2015 U.K. general election. In accordace with his commiment, he proposed to host the United Kingdom European Union membership referendum no later than 2017. Queen's Speech in 2015 and her biographer, Robert Lacey’s report on June 22 2016- The Queen has been canvassing opinion on the EU debate by asking dinner companions: "Give me three good reasons why Britain should be part of Europe."- increasing the uncertainty of the issue and the posibility of Breixt, which reflecting directly on ascending sovererign CDS premia before the referendum as presented in Chart 1.

Furthermore, most crisis period definition methodologies are analyzing the volatility difference points. Therefore, we firstly plot the volatility of CDS premia to capture volatility clusters, and perform the Chow breakpoint test to see the data structure change points.

Chart 2 The U.K. Sovereign CDS return

-.20 -.15 -.10 -.05 .00 .05 .10 .15

I II III IV I II III IV I II III IV I II III

2014 2015 2016 2017 -.3 -.2 -.1 .0 .1 .2 .3 .4

I II III IV I II III IV I II III IV I II III 2014 2015 2016 2017

Source: Datastream

In chart 2, we can clearly see the fact that the U.K. public CDS market was more volatile starting from the beginning of 2016. In the end of second season (referendum date), the volatility moved up to the highest level in the past three years. However, it is too arbitrary to conclude that the first half of 2016 to be our Brexit uncertainty period. We would also like to examine the volatility of overall private sector credit market in the U.K., which shows longer uncertainty period starting from May 2015, which is consistent with 2015 Cameron’s victory on May 7th_{. Moreover, we want}

to test the correlation difference before and after the referendum date; hence, we extend the referendum impact to three months after the voting date. After Chow breakpoint test, we can find the breakpoints on May 7th 2015 and September 23th 2016, which defined as our Brexit uncertainty period. Furthermore, we take one year before the uncertainty period and one year after referendum

25 to see the relationship changes due to the referendum and how long does this would impact on the rest of the world. Our test periods are listed as following:

1. Tranquil Period: From June 23th 2014 to May 6th 2015 (228 days)

2. Brexit Uncertainty Period: From May 7th 2015 to September 23th 2016 (362 days) 3. One Year After Brexit Period: From June 24th 2016 to June 23th 2017 (262 days)

Table 1 Global CDSs Index Return and Volatility

Source: Datastream

In Table 1, we can see the U.K. sovereign CDS premia has average higher positive return in the Brexit Uncertainty Period compared to the tranquil and post Brexit period. It may be good news

26 for the credit market speculators, while it also brings some concerns to their debt holder. As we know that CDS premia is the protection and compensation to credit derivative buyers; therefore, if CDS premia higher, it also means the higher default possibility, which is a bad signal to global financial stability. We can also find the market is in turmoil in Brexit uncertainty period with 1.21% higher volatility of the U.K. sovereign CDS, and decrease to 1.87% after the uncertainty.

Breaking down into the various private sectors, all CDS premia show mean return higher, higher default chance, in the Brexit uncertainty period except TMT market in EU market; the volatility also increased in different private sectors except Bank CDS index in Asia excluding Japan; Energy CDS index in the U.K., the U.S. and Asia excluding Japan; Other Finance CDS index in the U.K. and the U.S. ; Service CDS Index in the U.S.

4.2 Empirical Result

4.2.1 Adjusted Correlation

Our study follows Forbes and Rigobon’s (2002) methodology eliminating problems of heteroscedasticity and endogeneity by comparing the significance of adjusted (unconditional) correlation increase, also called as “shift contagion” or “comovement”, between tranquil period (6/23/2014-5/6/2015), and Brexit uncertainty period (5/7/2015-9/23/2016), as well as tranquil period and one year after referendum (9/24/2016-6/23/2017).

In Table 2, we can find that among 17 indexes of different private sectors (Bank, Consumer, Energy, Manufacturing, Other Finance, Service and TMT) in EU, the U.S. and Asia excluding Japan CDS markets, 10 indexes show no shift contagion (comovement) during Brexit uncertainty period. Breakdown to sectors, we find some more interesting evidence which may not support the results of previous research with the identical methodology. Goudert and Gex’s (2008) found shift contagion in bank, other finance, consumer and manufacturing sectors during GM financial distress crisis, and several researchers we mentioned in literatures review part also proved the contagion in Lehman Brothers bankruptcy crisis and European sovereign debt crisis especially in banking credit markets. In our statistic result, we can find significant increasing correlation between the U.K. banking CDS premia and EU banking CDS premia in Brexit uncertainty period (rejecting null

27 hypothesis 1-1), while not significant in the U.S. and Asia excluding Japan CDS premia. Moreover, our results are in line with other research result as well, after May 7 2015, David Cameron won the United Kingdom general election of 2015, the uncertainty of Brexit reflect on the U.K. other finance sectors CDS premia, and also significantly comove with other finance sector credit market of other regions (rejecting our null hypothesis 1-2).

Table 2 Comovement Test between tranquil period and Brexit uncertainty period

Significant at * 10%, **5%,***1%

The shift contagion evidence can only be found between the U.K. and the U.S. market in consumer sectors, the adjusted correlation rises from negative to 0.5640 in the U.S., which supports

Period (sample number) Date

Sector Market Volatility (%) Volatility (%) Z

Bank EU 0.0545 10.4770 0.2331 28.8996 2.1511** US -0.0078 3.6054 0.0904 9.2197 1.1585 AXJ 0.0787 4.2407 0.1997 11.3751 1.4521 Consumer EU 0.4648 3.3925 0.0770 3.8690 -5.0132*** US -0.1093 6.4836 0.5640 12.2111 8.8015*** AXJ Energy EU 0.9763 10.0959 0.9308 38.9633 -6.4316*** US 0.9385 60.1931 0.8040 85.5938 -7.24*** AXJ -0.6862 5.9673 0.8065 6.7039 23.0231*** Manufacturing EU 0.3722 15.4310 0.4352 43.8370 0.8857 US -0.0803 9.6905 0.2403 55.3129 3.8292*** AXJ 0.3677 5.1098 0.4508 7.2867 1.1757 Other Finance EU -0.1504 9.5934 0.1701 84.5007 3.8034*** US 0.1478 12.9345 0.5483 46.4319 5.4927*** AXJ 0.0865 4.3956 0.3450 9.4919 3.2112*** Service EU 0.6311 10.6953 0.5398 23.9078 -1.6391 US 0.3171 41.6202 0.6368 24.8935 4.9917*** AXJ 0.2653 26.6731 0.3088 60.6671 0.5587 TMT EU -0.0117 8.3813 -0.1149 27.2192 -1.2197 US -0.1756 8.9228 0.6179 20.6776 10.5735*** AXJ -0.0046 8.9850 0.5310 22.4466 7.011*** Tranquil (228) Brexit (362) 6/23/2014 - 5/6/2015 5/7/2015 - 9/23/2016 𝜌𝑢𝑛 𝜌𝑢𝑛

28 Biemond’s study (2017) that during the Brexit uncertainty period, the consuming products international trading has significantly decreased which implies that Brexit brings uncertainty to the daily trading business operation indirectly increasing the counterparty risk of the corporates all over the world (reject null hypothesis 1-3). Nevertheless, we found the adjusted correlation significantly decrease between the U.K. and EU consumer sector credit market. In other sectors, we can find the increasing adjusted correlation in energy sector between the U.K. and Asia excluding Japan, manufacturing sector between the U.K. and the U.S. service sector between the U.K. and the U.S., TMT sectors between the U.K. and the U.S. as well as the U.K. and Asia excluding Japan credit market.

Similarly, we perform the same methodology testing shift contagion effect (comovement) between tranquil and one year after Brexit referendum. According to our study result presented in Table 3 surprisingly, we have found that the shift contagion effect appears in most of the indices. The adjusted correlation significantly increases in banking, consumer, other finance and TMT sectors between the U.K. and rest of the world. In energy and manufacturing sectors, we can also find the significant comovement between the U.K. and Asia excluding Japan market. Only in service sector can we find the decreasing adjusted correlation among different markets one year after the referendum. This may imply that the Brexit impacts on different private sectors of other regions may be long lasting.

29

Table 3 Comovement Test between tranquil period and One year after referendum

Significant at * 10%, **5%,***1%

4.2.2 GJR GARCH

Other than testing the comovement, we apply the methodology also testing volatility spillover effect to make sure contagion exist. One way to measure the volatility spillover during crisis is building up GJR model to capture asymmetric volatility movement (Chiou, Wu, Wang, 2015). Before modelling, we need to make sure if CDS series fit the GARCH model. The descriptive statistics including standard error, skewness, kurtosis, JB normality test and LB-Q and Q square test and ADF unit root test results for the whole periods are displayed in Table 4.

Period (sample number) Date

Sector Market Volatility (%) Volatility (%) Z

Bank EU 0.0545 10.4770 0.7304 18.5783 9.6014*** US -0.0078 3.6054 0.6401 10.4368 8.4068*** AXJ 0.0787 4.2407 0.8726 9.2408 13.882*** Consumer EU 0.4648 3.3925 0.6656 2.6094 3.2857*** US -0.1093 6.4836 0.4825 5.9000 6.9791*** AXJ - - - - -Energy EU 0.9763 10.0959 -0.9506 11.3245 -44.4319*** US 0.9385 60.1931 -0.9734 42.6924 -42.552*** AXJ -0.6862 5.9673 0.9643 7.3491 31.2192*** Manufacturing EU 0.3722 15.4310 -0.1649 66.8529 -6.1156*** US -0.0803 9.6905 0.8813 17.2399 16.0449*** AXJ 0.3677 5.1098 0.5563 8.3869 2.6523*** Other Finance EU -0.1504 9.5934 0.5671 9.4853 8.7212*** US 0.1478 12.9345 0.4291 20.6135 3.4007*** AXJ 0.0865 4.3956 0.6166 12.0308 6.9431*** Service EU 0.6311 10.6953 0.4219 10.7834 -3.2175*** US 0.3171 41.6202 0.1077 23.6962 -2.4165*** AXJ 0.2653 26.6731 -0.2485 66.9469 -5.7672*** TMT EU -0.0117 8.3813 0.3394 7.8209 4.0069*** US -0.1756 8.9228 -0.1662 17.2656 7.011*** AXJ -0.0046 8.9850 0.4140 22.2020 4.8834***

Tranquil (228) One Year after Brexit (262) 6/23/2014 - 5/6/2015 6/23/2016 - 6/23/2017

30

Table 4 Descriptive statistics for whole periods

Significant at * 10%, **5%,***1%

In table 4, after we take first order difference, we can see no unit root in our data according to ADF test; therefore, we have stationary series for further testing. The standard deviation represents for the volatility and market risk of first order difference CDS premia. The risk is higher in the U.K. and EU banking credit markets which implies more unstable of in Europe area compared to the rest of the world. For other finance and manufacturing credit market, EU countries are relatively risker and more volatile than others. Most of the series have leptokurtic distribution, all significantly higher than 3, kurtosis of normal distribution; the result is in line with JB-normality test. We then apply Ljun-Box Q statistics testing it if there is auto correlation of the series; the results are all significantly reject the null hypothesis that there is autocorrelation of our samples; therefore, we can use these first order series to test the asymmetric volatility spillover with GARCH model.

Sector Market S.D. Skewness Kurtosis Jarque-Bera Q ADF

Bank UK 9.750533 0.009154 28.90783 21926.39*** 6.1454*** 2.6275*** -21.45284*** EU 8.537723 -0.24068 24.12995 14592.41*** 21.614*** 21.614*** -19.67351*** US 2.523795 -1.384995 29.05662 22429.6*** 15.453*** 4.8124*** -32.9927*** AXJ 2.254341 0.436341 7.225673 608.1843*** 22.187*** 10.335*** -25.52594*** Consumer UK 2.335214 0.535408 14.94991 4702.268*** 3.8046*** 2.9219*** -33.14975*** EU 2.02269 -0.299267 42.55961 51133.83*** 21.802*** 0.2309*** -24.06936*** US 1.751041 0.107452 6.965731 515.2581*** 19.29*** 7.9318*** -36.10397*** AXJ - - - -Energy UK 2.650856 0.526744 253.2957 2046536*** 11.304*** 0.1738*** -43.78241*** EU 3.415129 1.622847 82.75147 208113.9*** 11.242*** 3.6234*** -31.93587*** US 9.4434 0.075595 25.51891 16566.06*** 10.296*** 0.424*** -12.70356*** AXJ 3.848338 0.296285 21.39433 11064.29*** 23.961 5.9412*** -25.97354*** Manufacturing UK 3.564488 1.762284 100.723 312365.6*** 18.824*** 5.0002*** -14.92553*** EU 16.07721 -0.373518 55.16084 88896.19*** 6.6524*** 0.5857*** -45.97436*** US 5.011321 0.008259 38.41107 40962.18*** 17.155*** 9.6477*** -29.42125*** AXJ 2.509809 -0.233972 11.11574 2158.752*** 18.913*** 25.102 -24.5836*** Other Finance UK 3.2062 0.867449 215.1912 1470918*** 13.419*** 0.5569*** -27.30261*** EU 15.49491 0.360187 53.50997 83358.02*** 14.815*** 0.1059*** -17.75999*** US 6.493759 -0.15071 13.05401 3305.014*** 44.005 3.8103*** -21.36791*** AXJ 2.585507 0.003846 12.96981 3246.973*** 34.164*** 8.4735*** -17.58759*** Service UK 1.518361 1.612777 27.13243 19364.09*** 9.6454*** 1.041*** -26.90702*** EU 2.520966 3.194892 77.25456 181449.3*** 15.13*** 0.9441*** -33.57436*** US 6.816027 -0.528861 61.74691 112775.7*** 8.7367*** 1.8642*** -17.21911*** AXJ 20.37375 -0.573509 43.30522 53110.33*** 28.507*** 44.331 -25.96364*** TMT UK 2.726207 3.446039 56.4295 94805.59*** 6.7197*** 1.0088*** -30.20679*** EU 5.821289 0.701921 55.77159 91035.85*** 6.2573*** 2.871*** -27.51362*** US 3.041547 3.576519 124.9754 487685.9*** 8.6534*** 3.4974*** -25.26297*** AXJ 8.482923 -0.293943 15.06032 4762.7*** 20.761*** 9.4921*** -27.84275*** 2₍₁₂₎

31 In our study, we compare the coefficient 𝛾 in different periods, the coefficient 𝛾 represent for the scale of asymmetric volatility. If one market’s coefficient increases in Brexit uncertainty period compared to the normal periods, then it’s asymmetric volatility increase, which also means the shock (bad news) in the U.K. spillover to that market.

In table 5, we can find 𝛾 is significantly positive in bank, service and other finance sector for all regions in Brexit uncertainty period. The positive coefficient means the bad news in previous period will increase volatility today, also known as leverage effect. In the contrast, the negative coefficient means that the volatility decreases after the bad news happen, showing in energy, manufacturing and TMT sector during Brexit period, so we will say the increasing volatility of the sector in U.K. does not spillover to these markets with negative 𝛾; hence, volatility spillover does not exist.

Table 5 GJR GARCH model between tranquil period and Brexit uncertainty period

Significant at * 10%, **5%,***1%

In table 6, we do not see obvious leverage effect in some sectors in one year after Brexit CDS markets as their coefficient 𝛾 are negative. Positive coefficient can only be found in credit

Period Sector Market ω 𝛼 𝛽 𝛾 ω 𝛼 𝛽 𝛾 Bank EU -0.043581* 0.086638 1.031014*** -0.182049** -0.341495*** 0.037624*** 1.00497*** 0.076994*** US 0.243878*** 0.19518*** 0.827531*** -0.274628*** 0.484087*** 0.470792*** 0.624371*** 0.162498* AXJ 0.151661 0.012313 0.887712 0.17144 0.146845*** 0.021311*** 0.985654*** 0.080318*** Consumer EU 0.187843*** 0.340231*** 0.733809*** -0.236*** 1.463792*** 0.071644*** 0.577328*** -0.089327*** US 0.116026*** 0.111483*** 0.937616*** -0.133862*** 0.76917*** 0.207417*** 0.493064*** 0.245585* AXJ Energy EU 1.017289*** 0.603709*** 0.319567*** -0.337225* 1.73067*** 2.413483*** 0.168557*** -0.118648*** US 5.101214*** 0.244458*** 0.849001*** -0.194902*** 0.206007*** 0.18398*** 0.836306*** -0.03959* AXJ 1.107916*** 0.156355*** 0.800594*** 0.020194 9.649449*** 0.704638*** -0.017727* -0.525339*** Manufacturing EU 0.554958*** 0.024025 0.775755*** 0.124651 21.75236*** 0.040484 0.685844*** 0.354616*** US 0.142744*** 0.052312*** 0.914755*** -0.096446*** 0.224513*** 0.272061*** 0.855527*** -0.088567 AXJ -0.014917*** 0.102096*** 0.987782*** -0.132952*** 3.167787*** 0.27859*** 0.311421*** -0.108618* Other Finance EU 40.9641 0.143225** 0.588806 -0.253212** -0.018756 0.464683*** 0.863384*** 0.284681*** US 13.77363*** -0.124824*** 0.370408*** 0.000166** 9.865204*** -0.071608*** 0.374803*** 0.893002*** AXJ 0.972907*** 0.308798*** 0.443648*** 0.050074 1.099478*** 0.341797*** 0.568061*** -0.166349* Service EU 0.443169*** 0.259274** 0.624004*** -0.016623 4.666669*** 0.18391*** -0.009927 0.789112*** US -0.004478 6.364207*** 0.579835*** -0.04819*** 0.123043*** -0.017636*** 0.927568*** 0.186249*** AXJ 102.8895*** 0.099952 -0.045004 0.279717 286.0178*** 0.625837*** 0.117328*** 0.482694* TMT EU 1.539761*** 0.575807*** 0.325529*** -0.151272*** 3.1381*** 0.806678*** 0.29505*** 0.505822 US 0.920657*** 0.127611*** 0.767967*** -0.151681*** 1.009889*** 1.461528*** 0.271087*** -0.123141*** AXJ -0.001422 0.000241 0.024242*** -0.017051*** 1.54134*** 0.394931*** 0.783068*** -0.233839*** Tranquil Brexit

32 derivative market of the U.S. in energy, other finance, TMT sectors; Asia excluding Japan in service and TMT. The sectors in EU countries are more vulnerable compared to other regions; we can see positive asymmetric volatility in EU bank, manufacturing, other finance, and TMT sector which shows the fact that the EU credit market may still affected by the bad news of Brexit.

One more thing we can find in our result is that the 𝛾 of the U.S. and Asia excluding Japan bank sectors are positive in Brexit uncertainty period but negative one year after voting leave. The evidence shows the volatility increase led by the U.K. banking system fluctuation did impact the ones in USA and Asia, while the markets are immune from the shock in the long-term, showing the mechanism of banking system is stronger after several lessons since 2008 global financial crisis.

Table 6 GJR GARCH model between tranquil period and One year after referendum

Significant at * 10%, **5%,***1% Period Sector Market ω 𝛼 𝛽 𝛾 ω 𝛼 𝛽 𝛾 Bank EU -0.043581* 0.086638 1.031014*** -0.182049** -0.341495*** 0.037624*** 1.00497*** 0.05084* US 0.243878*** 0.19518*** 0.827531*** -0.274628*** 0.484087*** 0.470792*** 0.624371*** -0.967364*** AXJ 0.151661 0.012313 0.887712 0.17144 0.146845*** 0.021311*** 0.985654*** -0.1826*** Consumer EU 0.187843*** 0.340231*** 0.733809*** -0.236*** 1.463792*** 0.071644*** 0.577328*** -0.205687 US 0.116026*** 0.111483*** 0.937616*** -0.133862*** 0.76917*** 0.207417*** 0.493064*** -0.285343*** AXJ Energy EU 1.017289*** 0.603709*** 0.319567*** -0.337225* 1.73067*** 2.413483*** 0.168557*** -0.25286 US 5.101214*** 0.244458*** 0.849001*** -0.194902*** 0.206007*** 0.18398*** 0.836306*** 0.215843** AXJ 1.107916*** 0.156355*** 0.800594*** 0.020194 9.649449*** 0.704638*** -0.017727* -0.109794 Manufacturing EU 0.554958*** 0.024025 0.775755*** 0.124651 21.75236*** 0.040484 0.685844*** 0.998559*** US 0.142744*** 0.052312*** 0.914755*** -0.096446*** 0.224513*** 0.272061*** 0.855527*** -0.053789*** AXJ -0.014917*** 0.102096*** 0.987782*** -0.132952*** 3.167787*** 0.27859*** 0.311421*** -0.543415*** Other Finance EU 40.9641 0.143225** 0.588806 -0.253212** -0.018756 0.464683*** 0.863384*** 0.047654*** US 13.77363*** -0.124824*** 0.370408*** 0.000166** 9.865204*** -0.071608*** 0.374803*** 0.072625* AXJ 0.972907*** 0.308798*** 0.443648*** 0.050074 1.099478*** 0.341797*** 0.568061*** 0.083793 Service EU 0.443169*** 0.259274** 0.624004*** -0.016623 4.666669*** 0.18391*** -0.009927 -0.216179*** US -0.004478 6.364207*** 0.579835*** -0.04819*** 0.123043*** -0.017636*** 0.927568*** -0.654736** AXJ 102.8895*** 0.099952 -0.045004 0.279717 286.0178*** 0.625837*** 0.117328*** 0.081878*** TMT EU 1.539761*** 0.575807*** 0.325529*** -0.151272*** 3.1381*** 0.806678*** 0.29505*** 0.057472*** US 0.920657*** 0.127611*** 0.767967*** -0.151681*** 1.009889*** 1.461528*** 0.271087*** 0.501499*** AXJ -0.001422 0.000241 0.024242*** -0.017051*** 1.54134*** 0.394931*** 0.783068*** 0.156464**

33

Table 7 Spillover Test for whole periods

Significant at * 10%, **5%,***1%

Even though we found some positive asymmetric volatility in some indexes, we still need to perform the Z test to examine the fact that positive asymmetric volatility increases between tranquil and volatile periods. Only can we see the significant rise on asymmetric volatility can we prove volatility spillover effect exist. Table 7 shows the results of Z test for the 𝛾 coefficient in the whole testing periods. In Brexit uncertainty period, we can see EU and the U.S. banking and other finance sectors increase significantly (volatility spillover effect exist). Moreover, we can see strong volatility spillover effect in EU region, showing the fact that the politic and economic uncertainty did bring the impacts on most private sectors’ credit markets in EU, and these impacts last at least one year after the vote. In the contrast, Asia excluding Japan region suffer less compared to EU and the U.S. markets; the asymmetric volatility of banking system decreases significantly after the Brexit shock, other sectors’ volatility also decrease during the Brexit uncertainty period (except insignificantly increase in manufacturing sector and significantly increase in service sector). The results tell us that the business connection or trading partnership between Asia and the U.K. seems too weak to directly affect the credit risk of Asia’s corporates. Even though Asia’s credit risk of manufacturing and service sectors increase in Brexit period, we may not prove the impacts are

Period Tranquil Brexit Asymmetric

Sector Market𝛾 𝛾 Z-test Volatility

Bank EU -0.182049** 0.076994*** 3.072449*** Increase US -0.274628*** 0.162498* 5.24307*** Increase AXJ 0.17144 0.080318*** -1.089726*** Decrease Consumer EU -0.236*** -0.089327*** 1.775487* Increase US -0.133862*** 0.245585* 4.532318*** Increase AXJ Energy EU -0.337225* -0.118648*** 2.725519*** Increase US -0.194902*** -0.03959* 1.856033* Increase AXJ 0.020194 -0.525339*** -7.102061*** Decrease Manufacturing EU 0.124651 0.354616*** 2.886201*** Increase US -0.096446*** -0.088567 0.093463 Increase AXJ -0.132952*** -0.108618* 0.290436 Increase

Other Finance EU -0.253212** 0.284681*** 6.487308*** Increase US 0.000166** 0.893002*** 16.892891*** Increase AXJ 0.050074 -0.166349* -2.564123*** Decrease Service EU -0.016623 0.789112*** 12.768539*** Increase US -0.04819*** 0.186249*** 2.783467*** Increase AXJ 0.279717 0.482694* 2.81216*** Increase TMT EU -0.151272*** 0.505822 8.344679*** Increase US -0.151681*** -0.123141*** 0.342134 Increase AXJ -0.017051*** -0.233839*** -2.601391*** Decrease Period Tranquil One Year

after Brexit Asymmetric

Sector Market𝛾 𝛾 Z-test Volatility

Bank EU -0.182049** 0.05084* 2.578449*** Increase US -0.274628*** -0.967364*** -19.396163*** Decrease AXJ 0.17144 -0.1826*** -3.926306*** Decrease Consumer EU -0.236*** -0.205687 0.349707 Increase US -0.133862*** -0.285343*** -1.742691* Decrease AXJ Energy EU -0.337225* -0.25286 1.014902 Increase US -0.194902*** 0.215843** 4.5725909*** Increase AXJ 0.020194 -0.109794 -1.431242 Decrease Manufacturing EU 0.124651 0.998559*** 38.318416*** Increase US -0.096446*** -0.053789*** 0.470797 Increase AXJ -0.132952*** -0.543415*** -5.214782*** Decrease

Other Finance EU -0.253212** 0.047654*** 3.363522*** Increase

US 0.000166** 0.072625* 0.796485 Increase AXJ 0.050074 0.083793 0.371693 Increase Service EU -0.016623 -0.216179*** -2.2277*** Decrease US -0.04819*** -0.654736** -8.068498*** Decrease AXJ 0.279717 0.081878*** -2.252867*** Decrease TMT EU -0.151272*** 0.057472*** 2.304045** Increase US -0.151681*** 0.501499*** 7.726705*** Increase AXJ -0.017051*** 0.156464** 1.918187* Increase