Brexit referendum and equity market integration in the Euro area

(1)

Brexit referendum and equity market

integration in the Euro area

Braulio Gonz´

alez Varela

MSc Thesis

University of Amsterdam, Amsterdam Business School

MSc Finance, Banking and Regulation

(2)

Statement of Originality

This document is written by Braulio Gonzalez Varela who declares to take full responsibility for the contents of this document.

I declare that the text and the work presented in this document is original and that no sources 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 su-pervision of completion of the work, not for the contents.

Acknowledgments

I would like to express my sincere gratitude to my supervisor, Ms. Derya G¨uler, for her advise and remarks throughout the process of writing my MSc thesis.

(3)

Abstract

In March 2017, the United Kingdom started a withdrawal process from the European Union, following the results of a non-binding referendum in June 2016. The current research investigates if stock markets in the Euro area and the United Kingdom are less integrated during the period follow-ing the approval process of the membership referendum and before the exit negotiations. Using cointegration of time series and a set of time-varying measures of equity market integration, I find evidence that Euro area and United Kingdom stock markets are less integrated following the start of the referendum process. Segmentation between Euro area and United Kingdom markets is stronger that segmentation within Euro area markets. Cointe-gration tests further suggest an structural break between UK and the Euro area equity markets. The political uncertainty around this period could be associated with the segmentation of markets during the period of study, and have further importance for research and policy discussion.

Keywords: Brexit, Equity Market Integration, EMU, UK, Cointegration, Structural Break

(4)

List of Tables

4.1 Description of the data sample . . . 17

4.2 Summary statistics: Equity price indexes . . . 18

4.3 Summary statistics: Equity index returns . . . 19

6.1 Cointegration of equity price indexes: Individual countries and broader Euro area . . . 33

6.2 Cointegration of equity price indexes: Country pairs . . . 34

8.1 Summary table: Correlation of returns . . . 46

8.2 Summary table: Conditional betas . . . 46 8.3 Summary table: Index of international diversification benefits 47

(7)

List of Figures

4.1 Distribution of country return variances . . . 20 6.1 Correlation of returns between Euro area and UK stock markets 35 6.2 Correlation of returns within Euro area stock markets . . . 36 6.3 Conditional betas against World market . . . 37 6.4 Index of international diversification benefits . . . 38 8.1 Historical index of international diversification benefits . . . . 43 8.2 Correlation of returns within Euro area stock markets . . . 44 8.3 Correlation of returns between Euro area and UK stock markets 45

(8)

1.

Introduction

Integration of financial markets is an essential idea behind the creation of the Economic and Monetary Union (EMU) of the European Union (EU). In March 2017, the United Kingdom started a withdrawal process from the European Union, following the results of a non-binding referendum in June 2016. Following the decision of the UK to exit the EU this research assesses the impact on equity markets integration, which is part of the broader sense of financial integration. More precisely, the main question of this research is if stock markets in the Euro area and the United Kingdom are less integrated during the period after the approval of the membership referendum process and before the exit negotiations take place.

Behind the creation of the Euro area there is a rationale that via a higher degree of financial integration, potential output and risk diversification would increase among its member states, as pointed by Santis and G´erard (2006). The referendum process constitutes an opportunity to analyze how markets react to political uncertainty, and to further analyze if this political uncer-tainty increases equity market segmentation. In particular, I investigate how the impact of the referendum process differs between two different groups, (i) within Euro area stock markets and between (ii) the Euro area stock market and UK stock markets.

The exit of an EU country member is also of interest within the field of equity market integration. This work adds to this line of research by pro-viding evidence of the effects political uncertainty towards less integration in the region. Researchers such as Bekaert et al. (2013), Cappiello et al. (2010) and Hardouvelis et al. (2007), documented that higher levels of fi-nancial integration within European markets, compared to stock markets in other regions, are in part due to political efforts to achieve higher levels of in-tegration(for example the creation of a monetary union, the European single market, among others). The decision of the UK to exit the EU is a case study to assess if political uncertainty around the exit negotiations (that could im-pact free movement of labor, goods, services and capital across borders), have an impact on equity market integration.

Although price based measures are predominant on the literature, there does not seem to be a standard methodology to measure equity market in-tegration. For instance Billio et al. (2017), presents one of few studies that contrast and compare the effectiveness of a set of different methodologies. The current research is innovative in the sense that it applies a combination

(9)

of various methodologies from the literature, and compares the results to assess the impact of the referendum process on equity market integration.

I use cointegration of time series, cross-country correlations, conditional correlations estimated with a DCC GARCH, and conditional betas to gauge the impact of the referendum in stock market integration. I use a collection of daily MSCI equity indexes, starting January 2012 and ending May 2017, for a selection of Euro area countries and the UK. In addition, I use an international diversification benefit index as a robustness check to contrast the conclusions of the other methodologies. As argued by Christoffersen et al. (2012) and Billio et al. (2017), a consequence of higher stock market integration is that the benefits of international diversification diminish.

I find evidence of lower levels of equity market integration after the refer-endum process, but the magnitude and nature differs within Euro area stock markets and between Euro area and UK stock markets. While I find evidence of cointegration with an structural break between Euro area and UK stock markets, I do not find substantial evidence of any structural break within Euro area stock markets. Time-varying measures of integration, correlation and conditional correlation of returns, point to lower levels of integration in both the Euro area an UK stock markets, with a stronger impact in the integration between the Euro area and UK than within Euro area markets.

This research is structured as follows, chapter 2 contains the literature review and previous results on equity market integration in European mar-kets, chapter 3 briefly explains the context of the UK membership referen-dum, chapter 4 describes the sample and contains descriptive statistics on the sample, chapter 5 contains the research question and hypotheses, along with the empirical methodology used in this research, chapter 6 presents the results of the research and finally chapter 7 summarize the conclusions of the current work and recommendations for further research.

(10)

2.

Literature Review

This section contains a review of relevant literature on equity market integration and a summary of previous empirical findings on European and Euro area stock market integration. The first part provides the definition and measures relevant for the research. The second part describes factors associated with integration or segmentation in stock markets. Finally, the third part describes previous documented trends in financial integration for European and Euro area stock markets.

2.1 Definition and measurement of equity

mar-ket integration

Financial markets are integrated when assets with equal expected cash flows, maturities and bearing a similar risk command the same rate of return, as discussed by Kearney and Lucey (2004). In particular, stock markets are integrated when investors seeking higher returns end in equalizing the ex-pected yield of returns across different stock markets. This definition of stock market integration is closely related to comovements of returns across different markets, and how changes in these comovements imply changes in integration.

There are difficulties in translating the previous definition of financial in-tegration into an operational measure, as Kearney and Lucey (2004) explain. These difficulties arise in how to measure if markets have an homogeneous ex-posure to risk and how to measure expected returns when we observe realized returns instead. Following these difficulties in translating the definition into a measure, Billio et al. (2017) explain there does not seem to be a consensus in how to measure equity market integration.

Despite this divergence in how to measure stock market integration, most of the recent studies can be classified as price-based, as opposed to quantity-based measures (such as volumes of traded stock or foreign ownership of equity), and regulatory measures. Price-based measures have a clear inter-pretation from the definition of financial integration, as opposed to quantity-based or regulatory measures. In addition, the availability of high frequency price data and the ease at which indicators can be updated, further contribute to make price-based measures predominant over the reviewed literature Billio et al. (2017).

(11)

Within price-based measures we can further distinguish various method-ologies derived on the classification by Billio et al. (2017). These groups include (i) cross-country correlations, (ii) conditional correlation (GARCH models),(iii) asset pricing models or (iv) cointegration models. The current research covers one methodology in each group, the technical details can be found on 5.2.1 for (i), 5.2.2 for (ii), 5.2.3 for (iii) and 5.2.4 for (iv). While the technical details are described in the methodology, I discuss the rationale of each group in the current section.

The use of correlation of returns as a measure of equity market integration derives from the given definition of integration. Cappiello et al. (2010) argue that comovements of returns between equity markets are consistent with greater financial integration and economic interdependence.

However, the risk in each stock market, reflected by the volatility, is not constant over time and differs between countries, as argued by Roll (1992). Therefore the use of correlation could be an imperfect measure of integration when there is heteroskedasticity in the distribution of returns. Forbes and Rigobon (2002) argue that high correlation coefficients during crisis could be bias owing to high volatility. Some authors developed tools to estimate cor-relations controlling for heteroskedasticity. For example, the use of GARCH models to estimate conditional correlation of returns across markets in the presence of heteroskedasticity. One of these models is the DCC GARCH by Engle (2002). Other studies using addressing heteroskedasticity to analyze equity market integration can be found on Virk and Javed (2017), Forbes and Rigobon (2002) and Christoffersen et al. (2012).

Another approach to measure equity integration is with the use of asset pricing models. Under the traditional capital asset pricing model (CAPM), the required rate of return of an specific portfolio ri _{can be priced based on}

the relation E[ri] = rf + βiE[rm − rf_{], where r}f _{is the return offered by}

a risk free asset, E[rm _{− r}f_{] is a risk premium, or the excess return of the}

market portfolio over the risk free rate, and the β is a measure of sensitivity of the portfolio returns to the market portfolio returns. Under asset pricing models, integration can be assess in terms of which market portfolio (rm₎

explain better the returns of the original portfolio (ri_{), whether a local or a}

global portfolio is a better choice for the market portfolio suggest market are markets segmentation at a local level or integrated. Further works based on this approach can be found on Hau (2011) and Eiling et al. (2012). However for the purpose of this research I analyze equity integration by assessing beta convergence as suggested by Billio et al. (2017) and discussed in section 5.2.3. Under this approach the convergence of country betas against a common market, such as a global market, can be used to assess integration.

(12)

if there is integration between stock markets. This technique tests for non-spurious correlation between index prices across stock markets. Cointegra-tion implies that two time series evolve over time towards a long-term equi-librium, allowing for divergences in the short term, the formal definition can be found on section 5.2.4.

Although cointegration can be used to assess if there is a long-term equi-librium between two equity market price index, some authors argue that it is unusual to find such cointegrating relations in efficient markets. As argued by Richards (1995), economic theory suggests that cointegration is unlikely to be observed in efficient markets. Under efficient markets prices should reflect all available information at a given time, and therefore should not be predictable. If there is a deviation from the long-term trend between cointegrated price indexes, there is predictability, since the prices would ad-just towards the long-term equilibrium in the future. However, the use of Gregory and Hansen (1996) cointegration test with a regime break could cap-ture cointegrating relations that traditional test fails to identify, as argued by Voronkova (2004).

The different measures discussed have both advantages and limitations in the context of the current research. Depending on the technique chosen, measures can be static or time-varying. For example cointegration allows testing for a long-term relation between stock prices in a period of study, this is, it will reject or fail to reject the hypothesis of cointegration. Other measures such moving correlations propose a time-varying measure of the process of integration, and allow to analyze how the process evolves over time. The limitation in most time-varying measures is that there does not seem to be a formal test to assess the statistical significance of a deviation in a trend. For example, if higher levels of correlation can be taken as an increase of integration at a given level of significance.

Given the divergence in measures and limitations of the commented method-ologies to identify integration, I consider that a combination provides an in-tegrated approach to add robustness to the conclusions. The technical spec-ifications of the measures are described in further detail in the methodology of the research.

2.2 Determinants of equity market

integra-tion

This section summarizes the main factors associated with equity market in-tegration that could be relevant under the current research question.

(13)

Since the material consequences of Brexit have not yet materialized, po-litical uncertainty around the outcomes of the pending negotiations might affect market integration in the region, following the argument from Pastor and Veronesi (2013). Stock markets could become less integrated reflecting the political incertitude around the future negotiations on the exit conditions for the UK.

Within the literature we can further distinguish another factor associated with integration among stock markets, which corresponds to real business cycle synchronization, Asgharian et al. (2013). This linkage between equity market integration and real business cycle is important, given the economic linkages in the Euro area and the efforts in the region to achieve higher level of economic integration, as discussed further in section 3.1.

2.2.1 Political uncertainty

Virk and Javed (2017) document how the political dimension of the EU Greece relation played an important role in the integration of these equity markets after the financial crisis and near discussions of a Greek government default, or potential Grexit. They argue that although Italy and Spain faced similar distress in the aftermath of the financial crisis, in terms of higher levels of debt, ”the political fallout between Greece and European Commission (EC) has resulted in different integration structures”. The Greek case is important in the current research given some similarities with the current referendum, in particular, how markets reacted to an hypothetical exit of Greece from the EU.

When there is heterogeneity in the possible outcomes of a political deci-sion, Pastor and Veronesi (2013) finds that the risk premium increases for stocks. The increase in risk, owing to the uncertainty of a political decision, implies higher levels of segmentation. Pastor and Veronesi (2013) develop a model for how stock prices respond to political news. They find further empirical evidence supporting that political uncertainty increases volatility and correlation across stock, especially when the economy is weak.

Bekaert et al. (2011) find that a country political risk profile plays an im-portant role in explaining equity market segmentation, in addition to country regulation towards foreign capital flows and investment. This idea of politi-cal uncertainty increasing the risk in equity markets and therefore increasing segmentation, is also present in the work of Kelly and Veronesi (2016). Their study provides evidence of how political uncertainty around summits and elections is priced in the option market.

The relevance of political uncertainty in the current research derives from the nature of the discussion between UK and EU representatives, and for the

(14)

pending negotiations on the exit conditions of the UK. The period of research includes the decision of the UK to exit the EU, but excludes the period of the actual negotiations. Therefore the reaction of markets on this period is built on expectations of the possible outcomes. The dimension of the discus-sion is also relevant, with political representatives taking different stances on regulation pertaining the free movement of goods, labor and capital across the union as explained in chapter 3.

2.2.2 Real cycle synchronization

Previous findings provide a range of possible factors underlying integration, that include industrial structure (Roll (1992)), monetary integration (Walti (2011)), bilateral trade (Forbes and Chinn (2014)), geographical proximity (Asgharian et al. (2013)), among others. However there is no consensus on a single factor explaining equity market integration, but rather a more general process within the real sector that points to synchronization of the real business cycle.

One of the first studies seeking to explain the behavior in equity market stock is done by (Roll, 1992). The research tries to explain why country stock indexes show disperse behavior in their respective volatilities. The divergence in country volatility can be attributed to the construction of the index, the influence of exchange rates and each country industrial structure.

Asgharian et al. (2013) compare a selection of economic linkages, finding that bilateral trade is the most important factor explaining equity market integration. The authors examine in addition exchange rate volatility, di-vergence in inflation expectations, interest rate differentials, bilateral FDI and geographical distance of the exchanges. A possible explanation for the finding given by the authors is the synchronization of the real business cycle associated implied by higher levels of direct trade.

Further, Forbes and Chinn (2014) study a series of economic linkages that could explain cross country transmission of shocks across stock mar-kets. Among these factors, bilateral trade flow is a significant characteristic present in markets that are affected by transmission of shocks. Instead Walti (2011) argues that monetary integration leads to higher levels of stock market integration.

Previous studies point towards synchronization of real markets. For ex-ample, the importance of direct bilateral trade can be seen as evidence of syn-chronization of real business cycles and alignment of the fundamentals that determine the dynamics in real markets. The process of monetary union also eliminates volatility in real terms. In particular for the Euro area, the crite-ria that country members have to satisfy imply synchronization of

(15)

macroe-conomic variables and fiscal policies as well. The link between equity market integration and real business cycle synchronization is important, considering the pending negotiations that could affect policies pertaining free movement of labor, goods, services and capital between UK and the EU, as discussed in section 3.1.

2.3 Previous findings

This section list the main findings documented on previous empirical studies on equity market integration in European and Euro area equity markets.

2.3.1 Integration in European and Euro area equity

markets

Previous studies document a trend towards a higher degree of equity mar-ket integration, for both the countries in the Economic and Monetary Area (EMU) and the countries members of the European Union (EU). Despite this general trend towards a higher degree of integration, the magnitude, stability and channels remain unclear.

Bekaert et al. (2013) argue that EU membership is associated with an increase in stock market integration. The authors study integration in EU stock markets through convergence in industry yields across different coun-tries for the period ranging from 1990 to 2007. The analysis is robust when extending the period to 2012 to incorporate the EU sovereign crisis. Al-though they isolate different possible channels behind this finding, (such as foreign direct investment, regulation, financial development and divergence in real interest rates) they find that none of them capture the whole effect of EU membership on integration.

Contrasting the previous finding, Cappiello et al. (2010) argue that there is an increase in comovements of returns in EMU stock markets after the introduction of the single currency. This result is robust when controlling for changes in global trends, and covers the period of 1987 to 2008 for a selection of countries in the Euro area.

Hardouvelis et al. (2007) also argue that integration in European markets was influenced by introduction of the single currency. They find a reduction in the cost of equity among countries who decided to adopt the Euro. When repeating the analysis on Denmark, Sweden and the UK the reduction on the cost of equity is statistically insignificant and smaller than the reduction in EMU countries. Their conclusions further support that the introduction of the common currency played an important role in the convergence process in

(16)

the period from 1992 to 1998. Similar results towards higher integration in the EMU after the adoption of the common currency are pointed by Eiling et al. (2012), in the period from 1990 to 2008. The last study finds that the integration process differs between country groups, where convergence is stronger for the members least integrated to the EMU and world markets in the 90s.

For the period 1990-2013, Virk and Javed (2017) document a convergence in EMU and EU non-EMU equity markets in the post Euro period, and point to stabilization of this trend after the implementation of the common currency. They further point that the divergence in integration patterns is characterized better by small versus large markets, instead of Euro versus non-Euro markets. An important trend is that documented after the financial crisis between Greece and the EU, where the correlation between Greece and the benchmark German market diminished by almost half near 2013. Authors argue that European markets reacted as a way of isolating risk towards a possible Greek government default and a possible Grexit.

The work of Sehgal et al. (2016) analyzes the degree of integration of EMU stock markets around the global financial crisis and the sovereign debt crisis, for the period from 2002 to 2013. Using a combination of integration measures (beta convergence, sigma convergence, variance ratio, asymmetric DCC, dynamic cointegration, market synchronization and common compo-nents approach), they find that although there is a higher degree of inte-gration among EMU countries, the process is heterogeneous and dependent on the size of the economy. Countries are categorized by their GDP, where Germany, France, Italy, Spain, The Netherlands and Belgium show a higher degree of convergence. They document similar findings related to the Greek crisis, pointing towards segmentation of the stock market as a way to isolate from the perceived risk.

Although previous research points towards a higher degree of stock market integration in European markets, the integration process is different between groups of countries and dependent on the linkages of each country with the global economy. Further, the evidence is not conclusive in terms of isolating a single factor able to explain the integration in European stock markets. While some authors point to monetary union, others make reference to EU membership or the importance of synchronization of real markets and bilat-eral trade. An alternative interpretation is that these different factors are intertwined and are not mutually exclusive. It is possible that the process of integration implies an interaction of these different explanations documented in the current section.

Although previous research suggest stock markets are becoming more integrated with the World market over time, such as Hardouvelis et al. (2007),

(17)

Hau (2011), and Kearney and Lucey (2004), some evidence suggest that equity market integration in European and Euro area markets cannot be fully attributed to global patterns. In a study for US, UK and seven European countries for the period from 1974 to 2001, Fraser and Oyefeso (2005) find the presence of a common long-term convergence trend for these three stock markets. The US and UK in particular, show short-term deviation from this trend and showed relatively more divergence than European markets.

Finally, there seem to be a consensus on higher level of equity market integration for European and Euro area markets over time, with some di-vergence in the case of Greece in the aftermath of the financial crisis and the Greek crisis, as documented by Virk and Javed (2017) and Sehgal et al. (2016). Different factors could be associated with the levels of integration in the region, ranging from EU membership, monetary integration and eco-nomic linkages between countries. The current study asks to which extent the results of the referendum hinder the integration trend within the Euro area.

(18)

3.

Context

This section gives a brief summary of the historical and regulatory con-text of the research. The first section describe the historical concon-text of the Euro area (EMU) and its relevance in the context of equity market integra-tion, while the second part contains key facts about the UK membership referendum of 2016.

3.1 Economic and Monetary Union of the

Eu-ropean Union

The creation of the monetary union goes back to 1991 with the signing of the Maastricht Treaty. The process of a single currency implied the replace-ment of the monetary base of the initial members Austria, Belgium, Finland, France, Germany, Greece, Ireland, Italy, Luxembourg, the Netherlands, Por-tugal and Spain and the members that joined later Cyprus, Estonia, Latvia, Lithuania, Malta, Slovakia and Slovenia. 1 In addition to a single currency, establishing the union involved a process of monetary policy and fiscal coor-dination.

Euro are countries have to satisfy a set of convergence criteria. The can-didate members of the EMU need to show convergence in terms of price stability, sound public finances, sustainable public finances, durability of convergence and exchange rate stability 2. The convergence criteria that the members of the Euro have to satisfy could influence real business cycle synchronization and risk convergence. Both real cycle synchronization and risk convergence are factors associated with stock market integration, and documented in the literature to have a role in the equity market integration of the zone.

In addition, the Euro area countries and the remaining EU members are part of a broader market for goods, services, labor and capital, known as the European Single Market. 3 _{The European single market grants the freedom}

of movement of goods, services, capital and labor within its borders. This is

1_{”What Is the Economic and Monetary Union? (EMU).” European Commission. N.p.,}

10 Feb. 2017. Web. 25 June 2017.

2_{”Convergence Criteria for Joining.” European Commission. N.p., 25 Jan. 2017. Web.}

25 June 2017.

3_{”How Economic and Monetary Union Works.” European Commission. N.p., 22 Feb.}

(19)

also an important component that enables real business cycle synchronization in the region.

3.2 United Kingdom membership referendum

The European Union Referendum Act is a document calling for a non-binding referendum on whether the United Kingdom should remain part of the Eu-ropean Union. The act received Royal Assent on 17 December 2015, and was effectively enforced on February 2016 when the then UK Prime Minister David Cameron announced the referendum date to be Thursday 23 June of 2016. 4

The results of the referendum were announced on June 24, when the country voted to leave the EU, with 51.89% of the voters opted to leave the EU. In March 29, 2017 the current UK Prime Minister, Theresa May, triggers article 50 of the Treaty of Lisbon, which officially starts the separation process from the EU and calls for negotiations on the exit conditions. 5

The negotiations of the exit conditions started on June 19, 2017. These conditions have a considerable impact on the regulations related to the free-dom of movements for goods, services, capital and labor across the single market and the UK 6_{. The political climate is characterized as uncertain}7_,

given the economic impact of these negotiations and opposing views between EU and UK representatives.

As argued by Pastor and Veronesi (2013), political uncertainty has an economic cost on markets, especially when there is such high dispersion in the possible political outcomes and their repercussions in the economy. The referendum itself present the opportunity to analyze how markets reacts to political uncertainty, and to further investigate and evaluate the influence on equity market integration.

4_{”Referendum on the UK’s Membership of the European Union.” Electoral}

Commis-sion. N.p., n.d. Web. 25 June 2017.

5_{”Article 50 and Negotiations with the EU.” Article 50 and Negotiations with the EU}

- GOV.UK. UK Government, n.d. Web. 25 June 2017.

6_{”Why Brexit Is Grim News for the World Economy.” The Economist. The Economist}

Newspaper, 24 June 2016. Web. 25 June 2017

7_{Rachman, Gideon. ”The Great Unknows of Brexit.” Financial Times, 13 Mar. 2017.}

(20)

4.

Data

This section describes the time series used in the current research. It is comprised of two parts, the first part describes the sample selection and the second part contains descriptive statistics of the data.

4.1 Equity indexes

The main inputs for the analysis are equity price indexes from Morgan Stan-ley Capital International (MSCI), provided by DataStream. Each series cor-responds to daily closing prices denominated in Euros, for a selection of countries indexes, the Euro area and the World index. The selected coun-tries from the Euro area include Belgium, France, Germany, Netherlands, Spain, Italy and United Kingdom.

Table 4.1. Description of the data sample

Selected MSCI Indexes

Frequency Daily closing indexes from January 1st, 2012 to May 31, 2017. Ex-cluding the period corresponding to EU sovereign debt crisis (July 2011 December 2011), in order to provide an appropriate time window for the structural break test described in the section for the research design. Currency Euro

Countries Belgium, France, Germany, Italy, Netherlands, Spain and the United Kingdom.

Regions EMU, EMU excl. Country, World and World excl. EMU Observations 1,413 trading days.

Description of data sample from Morgan Stanley Capital International (MSCI equity indexes).

The selection of the Euro countries is based on the weight these stock markets represent on the MSCI Euro index. Belgium, France, Germany, Netherlands, Spain and Italy together represent over 90% of the weight in the MSCI Euro index 1 and therefore are can be seen as a representative selection of the Euro area stock market. The EMU index instead, covers 85% of the free float market capitalization in the Euro area according to the methodology by MSCI.

(21)

The period for the analysis goes from January 1st, 2012 to May 31, 2017. The rationale for this sample period derives from the limitations of the Gre-gory and Hansen (1996) cointegration test, described in section 5.2.4 of the methodology, this test allows to detect one structural break. Billio et. al (2016) argues that the period of the EU sovereign debt crisis is highly volatile, in particular the period from July 2011 to December 2012. For this reason the research period starts at January 2012, to avoid the possibility of the test capturing an structural break around a different event other than the UK membership referendum.

4.2 Descriptive statistics

This section presents summary statistics for the selected MSCI indexes. Ta-bles 4.2 contains a collection of descriptive statistics for the equity indexes and table 4.3 contains a collection of descriptive statistics for the indexes returns. Returns are obtained by taking the first difference of the natural logarithm of the indexes. 2

Table 4.2. Summary statistics: Equity price indexes

Index Start Minimum Maximum Arithmetic

Mean Standard Deviation Geometric Mean ADF Unit Root Test Belgium 580.5 569.1 1,347.1 1,004.6 219.1 979.1 -2.9 France 1,118.3 1,034.0 1,938.2 1,494.0 205.7 1,479.2 -3.0 Germany 579.8 553.2 1,087.7 827.3 121.5 818.0 -2.7 Italy 572.9 464.3 847.5 660.8 88.3 655.0 -2.2 Netherlands 814.4 746.8 1,641.4 1,180.4 226.2 1,157.9 -3.0 Spain 776.7 527.9 1,097.6 849.0 127.4 839.1 -2.1 United Kingdom 80.0 77.7 118.1 95.6 8.3 95.3 -2.9 EMU 733.4 657.8 1,244.2 970.2 136.0 960.3 -2.6 EMU + UK 77.9 72.7 124.5 99.3 11.7 98.6 -2.7

EMU excl. Belgium 76.5 68.6 129.1 101.2 14.3 100.1 -2.6 EMU excl. France 70.4 62.6 122.0 94.1 13.9 93.0 -2.5 EMU excl. Germany 70.8 62.0 117.9 92.1 13.2 91.1 -2.7 EMU excl. Italy 79.7 72.3 139.0 107.9 15.9 106.7 -2.7 EMU excl. Netherlands 77.2 69.5 130.7 102.4 14.4 101.4 -2.6 EMU excl. Spain 72.9 67.5 127.7 99.9 14.6 98.7 -2.8 World excl. EMU 96.8 96.8 191.3 141.5 27.0 138.9 -3.1 Note: Summary statistics for Morgan Stanley Capital International (MSCI) equity indexes. The ADF test for the presence of an unitary root. Statistical significance for the ADF test is given at the 1%, 5%, and 10% level is indicated by ***, **, and *, respectively.

For table 4.2, the collection of Euro area indexes (EMU excl. Country) has a relatively homogeneous starting level. The indexes for individual countries

2_{For index y}i_{, returns are given by r}i t= ln _yi t yi t−1 .

(22)

are more heterogeneous in terms of the starting closing price for the sample period.

Tables 4.2 and 4.3 include an augmented Dickey Fuller (ADF) test for the presence of an unitary root. The test has null hypothesis of unitary root, versus the alternative hypothesis of stationary of the time series.3 _Section

5.2.4 contains further technical details on the test.

For the indexes in table 4.2 the ADF test fails to reject the null hypothesis, pointing to the presence of an unitary roots for all series. The corresponding test presented in table 4.3 for the returns of the indexes, reject the null hy-pothesis at a 1% level of significance for all the series, pointing to stationary of returns. Both conclusions imply that returns are stationary, which is a nec-essary condition for cointegration between stock index as further explained in section 5.2.4.

Table 4.3. Summary statistics: Equity index returns

Returns Start Minimum Maximum Arithmetic

Mean 1 Standard Deviation 1 Geometric Mean 1 ADF Unit Root Test Belgium -0.07% -5.21% 4.05% 13.86% 17.24% 12.37% -27.0*** France 0.72% -7.99% 4.48% 9.38% 18.07% 7.75% -27.6*** Germany 1.44% -6.67% 4.70% 10.26% 17.93% 8.65% -27.0*** Italy 1.16% -13.60% 6.19% 3.81% 24.84% 0.71% -27.6*** Netherlands 0.17% -6.08% 4.05% 12.27% 16.40% 10.92% -27.1*** Spain 0.20% -13.95% 6.23% 5.02% 22.87% 2.38% -27.1*** United Kingdom 2.30% -9.38% 4.96% 4.29% 16.81% 2.87% -27.2*** EMU 0.81% -8.25% 4.64% 8.68% 18.25% 7.02% -27.4*** EMU + UK 1.51% -8.48% 4.26% 7.18% 16.54% 5.81% -27.2*** World excl. EMU 1.16% -6.20% 3.84% 11.34% 12.74% 10.53% -25.5*** EMU excl. Belgium 0.88% -8.07% 4.51% 8.92% 17.87% 7.32% -27.3*** EMU excl. France 0.91% -7.84% 4.49% 9.02% 17.77% 7.43% -27.2*** EMU excl. Germany 0.61% -8.39% 4.62% 8.66% 18.04% 7.03% -27.4*** EMU excl. Italy 0.82% -7.47% 4.34% 9.57% 17.38% 8.05% -27.3*** EMU excl. Netherlands 0.91% -8.12% 4.56% 8.81% 17.98% 7.19% -27.3*** EMU excl. Spain 0.93% -7.21% 4.42% 9.64% 17.51% 8.10% -27.3*** Note: Summary statistics for selected Morgan Stanley Capital International (MSCI) indexes. The ADF test for the presence of an unitary root. Statistical significance for the ADF test is given at the 1%, 5%, and 10% level is indicated by ***, **, and *, respectively.

1 Annualized returns, using 252 trading days.

From table 4.3 the average daily return annualized by 252 trading days, varies between 4.29% and 13.86% measured by an arithmetic mean, or be-tween 0.71% and 12.37% measured by a geometric mean. Standard deviation annualized by 252 trading days is in the range of 16-25%.

For EMU, EMU + UK and World indexes, the average daily return annu-alized by 252 trading days varies between 7.18% and 11.34% as measured by

3_{For a time series y}

t, and error term t with mean zero and constant variance, ythas

(23)

the arithmetic mean, or between 5.81% and 10.53% as measured by the geo-metric mean. The annualized standard deviation is in the range of 12-18%. The annualized returns of the Euro area indexes (EMU excl. country) are more homogeneous with returns close in the range of 8.5-9.5% and standard deviation in the range of 17.5-18%.

Figure 4.1 contains the historical distribution of variances for a selec-tion of MSCI country indexes, including Belgium, France, Germany, Italy, Netherlands, Spain and the United Kingdom. The figure contains moving variances calculated on subperiods of half-year realized returns, in order to calculate the daily variance, annualized based on 252 trading days.

0% 2% 4% 6% 8% 10% 12% 14% 16% 18% 20% Jan -12 Ap r-12 Ju l-12 Oct -12 Jan -13 Ap r-13 Ju l-13 Oct -13 Jan -14 Ap r-14 Ju l-14 Oct -14 Jan -15 Ap r-15 Ju l-15 Oct -15 Jan -16 Ap r-16 Ju l-16 Oct -16 Jan -17 Ap r-17 V ariance [25% 75%] [ Min Max ] Average Median

Figure 4.1. Distribution of country return variances

In figure 4.1 describes the disribution of variances for returns of the coun-try price indexes. There is an initial high dispersion in councoun-try volatility around July 2012 to January 2013, with return variances ranging from 2% to 12%. The dispersion of variances diminishes from January 2013 until the beginning of 2016, with the presence of high positive skewness during late 2014 until mid 2015. From December 2015 there is a sudden increase in the dispersion of return variances that persists until January 2017, with variances ranging from around 3% to 15%. The increase in volatility dispersion around

(24)

December 2015 coincides with the month when the referendum process re-ceived the last legislative approval. During the period from December 2015 until the end of the study variance drops gradually, except a sudden drop near between December 2016 and January 2017, from around 5% to 2%. The distribution further suggests a positive skewness during the period of study, as the median remains below the average during the complete sample.

(25)

5.

Research Design

This section describes the empirical methodology used in this research. The first part of the section explains the research question and the hypotheses derived from the literature and previous findings. The second part describes the different methodologies used to assess the hypotheses.

5.1 Stock market integration after the

refer-endum

The objective of this research is to identify if stock markets in the Euro area and UK are less integrated following the results of the membership referendum of 2016. In particular we are interested in two trends. The first trend corresponds to the level of integration within stock markets in the Euro area and the second correspond to the level of integration between the Euro area and UK equity markets.

The relevant literature suggest that political uncertainty affecting an economy can be priced in the stock markets and can decrease the level of integration, Bekaert et al. (2011). As mentioned in the literature review, the Greek crisis provides documented evidence on how European stock mar-kets became less integrated with the Greek market, as they faced a possible Grexit, Virk and Javed (2017). Following the concerns on a Greek default and a debate about a possible exit of the country as a member of the Euro-pean Union, equity markets showed a detachment from Greek risk, effectively reducing integration with the EMU market. This translated into a decrease of equity price correlations between Greece and other EU stock markets.

The pending negotiations on the exit conditions could add further in-certitude to the markets, as the outcome of these negotiations could affect regulations on free movement of capital, goods and labor (factors impacting real business cycle synchronization). As documented in the literature review, real cycle synchronization between countries is associated with higher levels of equity market integration.

Based on the trends documented by Virk and Javed (2017) between Eu-ropean stock markets and Greece, in addition to the political uncertainty around the outcome of pending exit negotiations, this research test two main hypotheses around the integration of the equity markets,

(26)

follows the UK membership referendum and before the exit negotiations. Hypothesis 2: EMU stock markets are less integrated with the UK in the period that follows the UK membership referendum and before the exit negotiations.

5.2 Methodology

Given there is no consensus in how to measure integration between financial markets, we use a representative methodology of each of the main groups of the price-based measures described by Billio et al. (2017). The measures are (i) cross-country correlation (section 5.2.1), (ii), conditional correlations (DCC GARCH; section 5.2.2), (iii) conditional betas (section 5.2.3) and (iv) cointegration (section 5.2.4) and (v) international diversification benefit in section 5.2.5.

The rationale in selecting different methodologies is to avoid the limita-tions of each measure by having additional support to the conclusions reached by one particular method. The next sections describe the measures chosen from each of this groups, highlights further advantages and drawbacks of each.

5.2.1 Cross-country correlations

This methodology use comovements of daily stock returns between different markets to capture financial integration. A higher degree of correlation in equity returns is consistent with greater integration and economic interde-pendence. As argued by Cappiello et al. (2010), integrated markets show higher comovements in returns as prices for similar financial assets move proportionally following the law of one price.

To capture the integration dynamic in the markets I calculate a time-varying measure using moving correlations of returns across markets. Cor-relations of daily returns are calculated on overlapping subperiods of half a year (152 trading days). The use of a time-varying measure is of special interest in order to analyze how the level of integration changed since the date of the referendum.

In order to assess both hypotheses, I separate the integration trends in two groups, first within Euro countries and second between Euro countries and the UK. To do so we first calculate per trading day the average correlation between Belgium, France, Italy, Netherlands and Spain against Germany. Calculating the correlation between each Euro country and Germany implies that the German market is treated as benchmark, in a similar way that Virk

(27)

and Javed (2017). In addition, we calculate at each trading day the average correlation of returns between the selection of Euro countries and the United Kingdom.

There are potential drawbacks to using correlations to measure financial integration. The first is that the volatility in stock markets increases more by negative shocks than by positive shocks of the same magnitude, as argued by Christoffersen et al. (2012). In addition, correlation itself increases when we observe high variance, even if this increase is not owing to integration but heteroskedasticity, as argued by Billio et al. (2017).

To overcome these difficulties, authors proposed heteroskedasticity cor-rection such as the GARCH DCC model explained in further detail in the next section. Further works using correlations in equity returns can be found in Forbes and Rigobon (2002) and Billio et al. (2017).

5.2.2 Dynamic conditional correlation (DCC GARCH)

The dynamic conditional correlation GARCH (DCC GARCH) is a multivari-ate tool to model time varying correlations of time series that is robust to heteroskedasticity. This correction for heteroskedasticity is especially impor-tant in this research considering the dispersion in return variances in figure 4.1, in chapter 4.

First we proceed to explain the basis of the model departing from the simple GARCH, then further expand to the DCC GARCH described in Engel (2009).

Consider a return series, ri

t = µi + t, with µi the expected return for

index i, and t an error term generated by a random process with mean zero.

The error term can be further characterized by t = δtσt, where δt follows a

normal distribution.

The GARCH model assumes a process for the volatility of the returns in order to calculate the conditional volatility. For example a GARCH(1, 1), is given by,

σ_t2 = α + β2_t−1+ γσ2_t−1 (5.1)

Equation 5.1. GARCH(1,1) process

Where the (1, 1) refers to the number of lagged values of the error and volatility terms respectively. The estimation of the parameters, η, α, β and γ is done by maximizing the log-likelihood function. 1_{. This estimation is}

done with the use of an statistical software.

1_{For an observed sample r}

(28)

param-In the first step of GARCH DCC in this research, the conditional volatility of each index return, σi

t, is estimated using the GARCH(1,1) specification in

equation 5.1, given the evidence of unitary roots in the indexes.

If Dt is a diagonal matrix with σit in its diagonal and zero elsewhere, we

can define the set of standardized residuals as vt = Dt−1t. Further denote

ˆ

R = 1

TΣ

T

t=1vtvt0.

The second step of the model consist in estimating the parameters α, β, from the following equation,

Qt= ˆR + α(vt−1vt−1− ˆR) + β(Qt−1− ˆR) (5.2)

Equation 5.2. GARCH DCC model

Here the matrix Qt, contains the conditional correlations between asset

returns ri _{and r}j _{at time t.}

Equation 5.2 can be rewritten in a similar form to the GARCH(1,1) in equation 5.1,

Qt = A + α(vt−1vt−1) + βQt−1 (5.3)

Equation 5.3. GARCH DCC model II

The representation of the GARCH DCC in equation 5.3 allows to see the model as a multivariable generalization of the univariate process in equation 5.2. This model is a tool to analyze and forecast conditional correlations, covariances and variances of returns. The estimation of the GARCH DCC model is done with an statistical software.

In the current research we use the estimation of the conditional correlation matrix, Qt, to analyze the changes in the level of integration formulated in the

research question. We employ a GARCH(1,1) in the first step, and include the index returns for Belgium, France, Germany, Italy, Netherlands, Spain and United Kingdom.

To assess the integration within EMU markets, we calculate at time t the average conditional correlation between Germany and the remaining Euro countries. In order to asses the integration of Euro area markets with the UK, we use the same procedure, but averaging the conditional correlations of the individual EMU countries and the UK. These conditional correlations eters θ = (µ, α, β, γ) with a join density function f (r1, r2, ..., rn | θ) the log-maximization

problem is given by maximize

(29)

measure the comovemente of prices across stock markets, controlling for het-eroskedasticity.

Although the GARCH DCC addresses heteroskedasticity and provides a dynamic measure of integration, the model is aimed at forecasting condi-tional volatilities. Therefore it does not provide a formal test to assess the statistical significance of a change in a trend. This lack of formal test is present in the literature, and further justifies the use of several methodolo-gies. Other studies using GARCH models include the work of Baele et al. (2010), Christoffersen et al. (2012), Sehgal et al. (2016) and Virk and Javed (2017),.

5.2.3 Conditional time-varying beta

Under the traditional CAPM model the expected returns of an specific mar-ket ri are given by E[ri] = rf + βiE[rm− rf_{], where r}f _{is the return offered}

by a risk free asset, E[rm_{− r}f_{] is the risk premium or the excess return of the}

market over the risk free rate, and the β is a measure of the volatility of a financial asset compared to the market. The β is usually calculated by based on historical returns, for example by a regression of the form ri _{= α+βr}m₊

t.

Billio et al. (2017) argue that the use of betas can be used to estimate financial integration. The beta is a measure of sensitivity of the returns in serie ri _{to the market returns r}m_{. The authors use a different model than}

the traditional CAPM, based on conditional variances and covariances, Et−1[rti] =

Covt−1(rti, rmt )

V ar(rm

t )

Et−1[rtm] (5.4)

Equation 5.4. Conditional time-varying beta

In equation 5.4 the conditional betas are given by the term βi =

Covt−1(rit, rtm)

V ar(rm

t )

. In the original study the authors use a multivariate specification of a GARCH model to estimate the conditional covariances and variance in the model above. In this research I use the GARCH DCC from section 5.2.2 to es-timate the covariances and variances in the conditional time-varying beta model.

To estimate these conditional variances and covariances between ri_t and market returns rm

t we estimate a series of bi-variate GARCH-DCC models.

We use two bivariate DCC GARCH models, one that includes the EMU equity index and the World equity index, and another that includes UK and the World index. In this way the conditional betas give an assessment of

(30)

both Euro and UK stock market against the world market. The rationale of using the conditional betas against Global markets is to have a perspective of how the risk in stock markets for the UK and the Euro area evolve compared to Global markets.

For further references in the use of other asset pricing models to as-sess financial integration please refer to Hau (2011), Eiling et al. (2012) and (Bekaert et al., 2014).

5.2.4 Cointegration and structural breaks

To explain cointegraion consider a two time series yt and xt. We say that

the series yt is stationary, when the joint probability of different observations

(yk+1, ..., yk+t), does not depend on t. Stationary implies that the mean and

variance of yt are constant over time, and we call yt integrated of order zero

or I(0). If yt is not stationary but the series ∆yt is, we call yt integrated of

order one or I(1).

Cointegration refers to time series that share a common long-term trend, which implies that a linear combination of the series is stationary. For an error term tstationary and with mean zero, we say ytand xtare cointegrated

when,

yt = β + αxt+ t (5.5)

Equation 5.5. Cointegration, Engle and Granger (1987)

The standard method for testing the null hypothesis of no cointegration in 5.5 is based on the distribution of the residuals t. One of this tests is the

two-step method developed by Engle and Granger (1987). Equation 5.5 is estimated using OLS, then and a unitary root test is applied on the fitted residuals ( ˆt) using the critical values determined by MacKinnon (1990,2010).

Under the null hypothesis of no cointegration between ytand xt, the residuals

are non-stationary. The EG statistic correspond to the t-ratio associated with H0 : t−1= 0 in following regression,

∆t= µ + αt−1+ ηt (5.6)

Equation 5.6. Cointegration test, Engle and Granger (1987) MacKinnon (1990,2010)

The t-ratio in equation 5.6 is also used to test for unitary roots. The difference between the unitary root test (ADF), and the cointegration test

(31)

(EG) are the distributions under the null hypothesis. For the EG test for cointegration, the critical values are provided by MacKinnon (1990,2010). The null hypothesis of no Cointegration is rejected in favor of the alternative hypothesis of cointegration when the t-statistic EG is smaller than the critical values.

If two time series are cointegration we can estimate an error correction model of the form, ∆yt= β1∆xt+ β2t−1ˆ + ηt. Which implies that deviation

from a long term-trend captured by t−1ˆ = yt − ˆβ − ˆαxt have predictive

power in explaining ∆yt. Richards (1995) argue that this predictability is

not consistent with efficient market theory and therefore finding correlation between stock markets would be unusual under efficient markets.

Although cointegration can be difficult to observe across equity markets, Voronkova (2004) argues that the Gregory and Hansen (1996) cointegration test with a regime break could capture cointegrating relations that traditional tests fails to identify.

The methodology by Gregory and Hansen (1996) allows to test for a structural break of unknown timing at T . Denote IT

t , a dummy variable

that takes the value of 1 for t > T and 0 otherwise. We can describe a cointegrating relation with an structural break by,

yt = β + β0ItT + αxt+ α0xtItT + t (5.7)

Equation 5.7. Cointegration with regime shift, Gregory and Hansen (1996)

Where the t term is stationary with mean zero. In this case β + β0

represent the level after the break, and α + α0 represents the slope after

the structural break. Gregory and Hansen (1996) build and ADFG statistic

to test for cointegration with a regime break of unknown timing. The null hypothesis of no cointegration with regime break is rejected for the alternative hypothesis of cointegration with a regime break if the ADFG statistic is

smaller than the critical value at a given level of significance.

In order to construct the ADFG, we calculate the EG statistic described

for 5.6 using the estimated error terms on 5.7 for a certain value of T . This step is repeated for possible values of T , the ADFG is the lowest of those

values. Denote EG(τ ) the EG statistic that correspond to T = τ . Then the authors create an ADFG statistic given by equation 5.8.

This is, the ADFG statistic is the lowest bound of the set containing the

EG stationarity test for cointegration, for all possible values of T in 5.7. The test allows tracking back T, and give and approximate date of the break.

(32)

ADF = inf

τ ∈1,..,T EG(τ ) (5.8)

Equation 5.8. ADF statistic for cointegration test with a struc-tural break, Gregory and Hansen (1996)

The strategy to test the hypotheses is to apply a cointegration test with a regime break to the EMU and UK equity indexes, in addition to the coin-tegration tests. If the ADFG test statistic fall below the critical values, that

would provide statistical evidence in favor of an structural break in an in-tegration trend, as described in 5.7. This implies that prices across equity markets have non-spurious correlation given by a long-term trend, but there is a structural break.

In order to avoid the possibility of other structural break and considering the test is only able to detect one break, the analysis is limited to the period following the EU Sovereign Debt Crisis. Billio et al. (2017) cites this period, as likely to show high correlations biased by on observed high volatility. As this effect is likely to have a wide impact on the integration trends in the region, this analysis do not include this period. The other metrics will be calculated in the same period for consistency.

5.2.5 Index of international diversification benefits

Following the argument in Christoffersen et al. (2012), an increase in equity market integration should be accompanied by a decrease in the benefits of international portfolio diversification. An index based on international diver-sification benefits captures the trend towards market segmentation, where an increase implies a higher degree of equity segmentation across equity markets. This index is calculated for two different groups. The first group corre-sponds to Euro area countries and contains the equity indexes of Belgium, France, Germany, Italy, Netherlands and Spain. The second group contains the same group of Euro area equity indexes and in addition to the UK.

For a given group of countries I calculate the weights of a minimum variance portfolio over half a year sub-periods. In this portfolio the assets corresponds to the country indexes. Denote by wtthe weights of the portfolio

on each asset, we can describe the variance of this portfolio by, σp,t =

q

w0_tΣtwt (5.9)

Where wt are the weights attached to each security, and Σt is the

(33)

matrix Σt can be decomposed as DtΦtDt, where Φt contains ones on the

di-agonal and correlations elsewhere and Dt contains the individual volatilities

in the diagonal, and zero elsewhere. In the case of no diversification benefits these correlations are effectively zeros, and the matrix Φt becomes and

iden-tity matrix I. We can describe the variance of a portfolio with uncorrelated asset returns by,

¯ σp,t= q w0_tDtIDtwt= w 0 tσt (5.10)

With σt, a vector of the individual variances of each index. Based on this,

the international diversification index can be calculated as,

DIt= ¯ σp,t− σp,t ¯ σp,t = 1 − √ w0Σtw w0_tσt (5.11)

Equation 5.11. International Diversification Benefit, Billio et al. (2017).

This index measures the benefits from correlations between assets returns, versus a scenario of no correlation between the assets returns. An increase in the index implies a higher degree of equity segmentation across equity markets.

5.3 Limitations

There are advantages and drawbacks from the different measures. While the time-varying methodologies provide a measure of how integration changes over time (cross-country correlations, conditional correlations, and condi-tional betas), the literature does not provide a formal statistical test to de-tect an structural break, in the sense of a traditional F-test or t-test to assess the statistical significance of a deviation in a trend. Testing for cointegration gives an statistical test that could capture non-spurious correlations between prices in equity markets, but it does not capture how integration evolves over time. In some cases cointegration could fail to capture the existence of equity market integration as argued by Richards (1995). However under the presence of an structural break, the Gregory and Hansen (1996) could effectively capture these relations, as argued by Voronkova (2004). The set of measures complement one another, cointegration allows applying a formal test to detect an structural break. Correlations provide a time-varying mea-sure of integration across markets. The conditional betas, calculated against

(34)

the World market, allow analyzing the risk of the Euro area and UK against a different benchmark. Finally using the international diversification benefit in gives additional robustness to the conclusions of other measures.

(35)

6.

Results

This section contains the main results of the research. I compare the results of the cointegration tests and the different measures of financial inte-gration described in chapter 5. I finally compare these results to the diversi-fication benefit index to add robustness to the conclusions.

6.1 Cointegration in Euro area and UK

eq-uity price indexes

Table 6.1 contains the results of the Engle and Granger (1987) cointegration test and the Gregory and Hansen (1996) cointegration test in presence of an structural break. The EG statistic in 6.1 test the alternative hypothesis of cointegration between an individual country equity price index and a broader Euro area equity index that excludes that particular country. The selection of countries includes Belgium, France, Italy, Netherlands, Spain, Germany and the United Kingdom. The ADFG test the alternative hypothesis of

cointegration with an structural break between the same group of individual countries and the wider Euro area equity price index.

The ADFG test in table 6.1 provides statistical evidence of cointegration

with a structural break between Euro area and UK equity prices at a 1% level of significance. The approximate date of the regime break determined by the Gregory and Hansen (1996) test is December 21 of 2015, that coincides with the month when the House of Lords of United Kingdom approved the European Union Referendum Act 2015, and given royal assent on December 17, 2015 1. Although the date is not precise the month of the break could be taken as a proxy for the start of the referendum process.

The result of the ADFG test, between UK and the Euro area, suggests

there is a change in the long-term equilibrium that determines how equity returns commove across markets. The timing of the break, on December 2015 is also indicative of the start of the referendum process.

Table 6.2 applies the same pair of Engle and Granger (1987) and Gregory and Hansen (1996) tests on pairs of Euro area and UK countries. Table 6.2 provides further evidence of cointegration with an structural break between United Kingdom and four of five of the six selected Euro area countries at least at a 5% level of significance, France, Germany, Spain and Belgium.

(36)

Table 6.1. Cointegration of equity price indexes: Individual countries and broader Euro area

Cointegration without structural break Cointegration with structural break EG ADFG Date United Kingdom -2.4 -5.7*** 12/21/2015 France -2.2 -3.9 12/30/2015 Germany -3.6** -4.2 08/3/2016 Italy -1.6 -4.1 1/19/2016 Netherlands -1.7 -3.4 1/07/2016 Spain -2.3 -4.3 1/19/2015 Belgium -1.6 -4.8* 1/05/2015

This table presents two sets of cointegration tests between individual country indexes and a broader Euro area index. The Euro area index excludes the country to which the test is applied. The ADF statistic tests the alternative hypothesis of cointegration with a regime break of unknown timing versus the null hypothesis of no cointegration with a regime break, as described by Gregory and Hansen (1996). The approximate date of the regime shift is determined by the test and is included in the table. The EG tests the alternative hypothesis of cointegration, versus the null hypothesis of no coin-tegration, using the Engle and Granger (1987) two-step method. The sample for both groups includes 1,413 trading days from January 1st, 2012 to May 31, 2017. Statistical significance at the 1%, 5%, and 10% level is indicated by ***, **, and *, respectively.

Puzzling in the case of Spain is that there is evidence for both cointegration with and without structural break, at 1% and 5% significance respectively. Another puzzling finding is the presence of cointegration between UK and Italy at 5% significance.

Further examining the results in table 6.1 there is evidence of cointegra-tion between Germany and the rest of the Euro area at 5% level of signif-icance. Examining 6.2, there is further evidence of cointegration between Germany and France at a 5% level of significance. These results are un-expected, as the literature mentions cointegrating relations between stock prices are uncommon, Richards (1995).

The structural break for Belgium and the Euro area in table 6.1 at 10% level of significance, does not seem robust considering there is no further evidence of an structural break in cointegration between Belgium and any other Euro country in table 6.2.

Further inspecting table 6.1 there does not seem to be a cointegrating relation for the remaining individual stock markets and the wide Euro area. This result does not imply that markets are segmented, as mention in the

(37)

Table 6.2. Cointegration of equity price indexes: Country pairs

Without structural break

United

Kingdom France Germany Italy Netherlands Spain

France -2.6 Germany -3.0 -3.5** Italy -3.8** -0.6 -1.4 Netherlands -2.7 -2.9 -2.6 -1.8 Spain -4.4*** -2.3 -2.9 -2.2 -1.5 Belgium -2.8 -1.8 -1.8 -2.0 -1.1 -1.9 Structural Break United

Kingdom France Germany Italy Netherlands Spain

France -5.4** Germany -6.1*** -4.1 Italy -4.6 -4.0 -4.0 Netherlands -6.1*** -4.6 -4.6 -4.0 Spain -5.2** -4.1 -4.5 -3.7 -3.7 Belgium -5.4** -2.4 -2.3 -3.6 -1.9 -3.0

This table presents two sets of cointegration tests between pairs of individual countries in the Euro area and United Kingdom. The statistic for cointegration without structural break is the ADF desribed by Gregory and Hansen (1996), with alternative hypothesis of cointegration with a regime break of unknown timing. The cointegration without structural break corresponds to Engle and Granger (1987) two-step method, with alternative hypothesis of cointegration without regime break. The sample for both groups includes 1,413 trading days from January 1st, 2012 to May 31, 2017. Statistical significance at the 1%, 5%, and 10% level is indicated by ***, **, and *, respectively.

literature, but could be a consequence of efficient markets. This drawback of using cointegration to measure financial integration when the null hypothe-sis is not rejected, is addressed with the additional set of measures in this research.

The time varying measures in the following section help capture how the process of integration evolves over the research period. These measures are used to complement the conclusions of the cointegration tests.

6.2 Time-varying measures of equity market

integration

Figure 6.1 plots the time-varying correlations between the Euro area countries and UK, described in sections 5.2.1 and 5.2.2.

(38)

20% 30% 40% 50% 60% 70% 80% 90% 100% Ma r-12 Jun -12 Sep -12 Dec -12 Ma r-13 Jun -13 Sep -13 Dec -13 Ma r-14 Jun -14 Sep -14 Dec -14 Ma r-15 Jun -15 Sep -15 Dec -15 Ma r-16 Jun -16 Sep -16 Dec -16 Ma r-17 In tegra tion meas ures Correlation Conditional Correlation

Figure 6.1. Correlation of returns between Euro area and UK stock markets

since the referendum, from an average of 87.4% in June 2016 to an average 44.8% in May 2017. This trend is an indicator of segmentation between the Euro area and UK stock markets. A similar trend was also documented in the case of Greece and EU markets, from correlation around 70% in the period previous to the financial crisis, which dropped to around 30% at the end of 2013. This finding resemble the documented trends by Virk and Javed (2017) and Sehgal et al. (2016), who point to how the correlation of returns between European markets and Greece decreased in the aftermath of the financial crisis and near the concerns of a government default. This reduction in comovements of returns between UK and the Euro area is also reflected in the conditional correlations estimated with the DCC GARCH. This conditional correlation goes from an average of 80.0% in June 2016 to 65.8% in May 2017.

The time varying correlations point to a reduction in financial integra-tion between Euro area and UK stock markets, following the period after the membership referendum and before the exit negotiations. Given the negotia-tion and therefore the consequences of Brexit are not material or foreseeable within the period of study, these findings suggest the market reaction is build on uncertainty or anticipation of the consequences of pending negotiations.

(39)

20% 30% 40% 50% 60% 70% 80% 90% 100% Ma r-12 Jun -12 Sep -12 Dec -12 Ma r-13 Jun -13 Sep -13 Dec -13 Ma r-14 Jun -14 Sep -14 Dec -14 Ma r-15 Jun -15 Sep -15 Dec -15 Ma r-16 Jun -16 Sep -16 Dec -16 Ma r-17 In tegra tion meas ures Correlation Conditional Correlation

Figure 6.2. Correlation of returns within Euro area stock markets

Figure 6.2 shows the time varying correlations within Euro area stock markets, measured by the average correlation of Euro country returns against Germany. The average correlation within Euro stock markets also decreases following the referendum from an average of 89.2% in June 2016 to 72.3% in May 2017. The conditional correlations for Euro are stocks decrease from an average 87.3% to a 84.4%.

Euro area markets are more integrated internally than with the UK at the beginning of the study. In addition, the decrease in return correlations within Euro area stocks markets after 2016, is lower in comparison to the reduction in correlations between the Euro area and UK stock markets for the same period. Comparing the levels of correlations in figures 6.2 and 6.1 for the period prior to 2016, Euro area stock markets have an average correlation of 83.7% and average conditional correlation 83.1%, while the average correlation of Euro area and the UK is around 73.1% and the average conditional correlation is around 73.9%.

Figure 6.3 plots the conditional betas from section 5.2.3. These betas measure the sensitivity of returns for Euro area countries and UK against the returns of the World index.

The average conditional beta for the Euro area is 1.03 for the sample period before 2016, while the average conditional beta for UK is 0.96 for the

Brexit referendum and equity market integration in the Euro area