The influence that a common currency and market conditions have on economic integration : A cross-quantilogram and DCC-EGARCH approach

(1)

Linköping University | Department of Management and Engineering Master’s thesis, 30 credits| Master’s programme Spring 2018| LIU-IEI-FIL-A--18/02863--SE

The influence that a common

currency and market conditions have

on economic integration

-A cross-quantilogram and DCC-EGARCH approach

Sebastian Lindman Tom Tuvhag

Linköping University SE-581 83 Linköping, Sweden +46 013 28 10 00|www.liu.se

(2)

I English title:

The influence that a common currency and market conditions have on economic integration: A cross-quantilogram and DCC-EGARCH approach

Authors: Sebastian Lindman sebli385@student.liu.se Tom Tuvhag tomtu462@student.liu.se Supervisor: Gazi Salah Uddin

Publication type: Master’s Thesis in Economics

Master’s Programme in Economics at Linköping University Advanced level, 30 credits

Spring semester 2018

ISRN Number: LIU-IEI-FIL-A--18/02863--SE Linköping University

Department of Management and Engineering (IEI) www.liu.se

(3)

II

Abstract

Countries participating in a common currency area increase their integration within the area. This paper investigates the impact common currency areas have for economic integration with economies of different characteristic outside the area. Results for a common currency group compares to a sovereign currency group. The common currency group consists of three countries who have adopted the euro, while the sovereign currency group consist of three European countries with sovereign currencies. The level of economic integration is examined towards three different economies; European drivers, global markets and emerging markets. The period ranges from 1993M01 to 2017M09 and includes industrial production indices and stock market indices. Economic integration is studied through a DCC-EGARCH model, on both aggregated and time-dependent level, which yield correlations. In comparison to previous studies, this paper also applies a cross-quantilogram method to examine the impact of different market conditions have on the correlations.

Higher correlations for the common currency group than for the sovereign currency group do exist with the European drivers and the global countries. With the emerging markets such pattern is not found, instead low correlations are mainly examined. Besides the correlation with the emerging countries, the results indicate membership in a common currency area, in this case the EMU, to increase the economic integration. Overall, highest levels of correlation are found with the European drivers, followed by the US as a global economy, corresponding with the importance of homogeneity for high economic integration. Due to no conclusive change in correlations during the euro implementation, membership in a common currency area per se does not increase economic integration. However, a common currency area with a strong currency do along with other characteristics influence the economic integration. We find evidence that market regimes have an impact on economic integration. Adverse market conditions overall seem to influence the integration in a higher degree than normal or good conditions. The results indicate that the adverse conditions increase the economic integration, this is in particularly seen for the common currency countries correlation with the European drivers and the US.

Keywords: Economic integration; Business cycle synchronization; Correlation; Market condition;

(4)

III

Acknowledgements

We would first like to thank our supervisor, Gazi Salah Uddin for great support throughout our work with this thesis. It has been a pleasure working with you. Also, a great thank you to our seminar group and opponents for constructive comments and ideas. To Axel Hedström for obtaining the R-code, making our cross-quantilogram estimations possible, and to Bo Sjö for your econometric guidance.

(5)

IV

1 Introduction

Is the membership in a common currency union more beneficial for economic integration than using a sovereign currency? By entering a common currency area, the possibility for a sovereign central bank disappear. However, a common central bank and its monetary policy can be viewed to influence the integration through synchronized business cycles. The common currency union can also reduce the impact of adverse shocks through risk-sharing (Mundell, 1973). Illustrated by the global financial crisis in 2008, the countries adopting the euro increased their business cycle synchronization which provided an effective stabilization instrument (Bekiros, et al., 2015). An increased economic integration within the eurozone and the European Union, EU, supported countries like Greece but simultaneously raised questions of the willingness to participate in the union. The United Kingdom, UK, composes the greatest sceptic, declaring referendum of leaving the EU with the result of Brexit. The UK’s exit might jeopardize the economy’s access to one of its most significant markets, with 44 percent of their export and 53 percent of their import (House of Commons Parliament, 2016). One of EU’s fundamental keystones, the idea of a single market with free movement of goods, services, capital, and labor for approximately 500 million consumers (European Union, 2018), is therefore threatened.

The countries within Europe have since the 1950’s developed closer economic ties which today are expressed in a single internal market. The creation of the common market aims to ensure the free movement of capital, goods, labor, and services (European Commission, 2018). Because of the globalization, we live in a highly connected world characterized as a global common market today. The EU benefits from globalization as it negotiates trade agreements with countries and regions around the world for its members, e.g., the Transatlantic Trade and Investment Partnership between the EU and the United States of America, US. These sorts of trade agreements are negotiated to eliminate customs duties, remove or reduce customs tariffs or provide a general framework for bilateral economic relationships (European Commission, 2018). The trade agreements are therefore enhancing the trade and integration, connecting the EU to other countries and regions. The continued global integration hinge on the ability of countries to adjust, adapt and encourage globalization. A cost of global integration is when countries give up on the autonomy in the form of monetary policy since increased global integration demands parties to agree to more equivalent game rules (Ku and Yoo, 2013).

As previously pointed out, there are several advantages and disadvantages of high economic integration. To better understand how countries, economies and markets behave and are affected in different prospects such as recessions, booms, fluctuations and economic uncertainties it is essential to first examine and understand the level of co-movement of business cycles and dependence in economic integration. Since the concept of economic integration is comprehensive, with several definitions in previous studies and theoretical works, we will refer to economic integration as the transformation in welfare occurring in an area when it and one or more areas reduce their regulations against each other (Balassa, 1962) throughout this paper. The transformation and thus higher integration can be achieved by increasing trade in goods or by increasing flows of ideas (Rivera-Batiz and Romer, 1991). By flows of ideas, we include factor mobility such as capital, goods, labor and services. If two countries are highly economically integrated, either in production or the financial market, we interpret the countries to have synchronized business cycles. Thus, making economic integration and business cycle synchronization cooperative concepts.

Previous studies in the field of business cycle synchronization have generally focused on three areas: (1) the short- and long-term relationship with the Vector Autoregressive- (VAR), Vector Error Correction- (VECM), and Autoregressive Distributed Lag (ADL) models. (2) Generalized Autoregressive Heteroskedasticity- (GARCH) types of Dynamic Conditional Correlation (DCC) models examining the correlation in business cycle synchronization. (3) Determining the dependence, directionality or causality in a time- or frequency scale. Furthermore, previous studies have had a focus on investigating the level of economic integration within a common currency area. A frequently studied currency area is the Economic and Monetary Union of EU, EMU, which seems to have had a significant impact towards an increased business cycle synchronization for the member countries. The theoretical framework often concludes from

(7)

2 optimum currency area theory, which declare different characteristic for a currency area to be optimum (Mundell, 1961). Among the benefits of being part of a common currency area is the increased risk-sharing within the area (Mundell, 1973). Due to unexpected changes and increased uncertainty followed by the globalization and today’s economic context, recessions such as the global financial crisis in 2008 will most likely occur again. On the other hand, a country being part of a common currency area loses its autonomy to regulate monetary policy. The outcome of an optimum common currency area is increased business cycles co-movement within the area (Mundell, 1961). However, how the countries as individuals and groups integrate with countries of different characteristics outside the currency area has no obvious conclusion from the optimum currency area theory.

Therefore, the purpose of this study is to investigate the impact a common currency union has on the economic integration. Compared for countries participating in the union and countries not participating, with major economies outside and within the market of the common currency union. Since the literature already has studied the impact within a common currency group, we aim to direct the light towards the integration with major global and emerging economies. Further, we aim to study the impact of the common currency union on the economic integration during different market regimes. We investigate how the relationship of economic integration affects our countries and groups depending on them experiencing adverse, normal or good market conditions, namely how the business cycle varies. Market conditions impact is essential to analyze due to the development of globalization and changing uncertainty in different market regimes, with the global financial crisis as an evident example. Therefore, we will apply cross-quantilogram analysis to identify if there are differences in the effects depending on market regimes. The cross-quantilogram analysis is a relatively new methodology originating from 2016 (Han, et al., 2016) and is used studying the directional predictability, in particular for financial time series hence its ability to capture the volatility.

Our starting point is Europe where we have selected three countries within the EMU; Finland, France and Italy which make up the common currency group. The sovereign currencies group consists of Hungary, Norway and Sweden. Our major economies include three groups, the European drivers which consists of the economic centre of Europe, Germany, respectively the financial centre of Europe, the UK. The Global group consists of the US and Japan. The last group in this study, the Emerging market group, consists of China and India. Considering our diverse sample, the limitation in previous studies and our choice of methodology, we will focus on the following research questions:

1. What impact has the common currency union had on the economic integration for countries adopting the Euro and for countries using a sovereign currency in Europe?

2. Are there different integration developments towards the three groups; European drivers, Global and Emerging countries?

3. How do different market conditions affect the level of economic integration?

Following the news of the UK leaving the EU, it is interesting to examine if a shift in economic integration has appeared for the UK, altering from the European market towards the global and emerging market. Thereby differencing the UK’s development from Germany’s. These results could be used to interpret if the UK is prepared for Brexit. However, since our primary focus is the impact of a common currency area on economic integration we look at Brexit as a secondary, but important analysis.

Deriving from the previous studies and our theoretical framework the following hypothesises apply. A high degree of integration should exist within the EU, due to its homogeneity with a common market and regulations. The countries which adopted the euro should display an even higher degree of integration since they share even more similarities. The connection between Europe and the US is inevitable due to their long history, with Transatlantic Trade and Investment Partnership as an example, and should display a stronger integration than for Europe with any other country or group. Following the recent development of the UK’s decision of leaving the EU, a trend of increased integration towards countries outside EU should be discovered for Brexit to be supported.

(8)

3 We apply time series analysis to investigate the correlation, which we interpret as the level of economic integration, between industrial production and stock markets for twelve countries. Exponential Generalized Autoregressive Heteroscedasticity (EGARCH), Dynamic Conditional Correlations (DCC) and cross-quantilogram are used as econometrical models. Our approach focuses on comparing the results from the models to be able to analyze the correlations on (1) an aggregated level, (2) with a time-dependent condition and (3) for different market regimes, to make conclusions regarding economic integration. This approach lets us evaluate the correlation both on static and dynamic levels. By breaking down the countries correlations into different market regimes a more complete picture can be generated. Monthly data is obtained from the Thompson Reuters DataStream, WorldBank database and the OECD database, covering the period 1993M01 to 2017M09. The results indicate that higher correlations for the common currency group than for the sovereign currency group with the European drivers and the global countries do exist. With the emerging countries such pattern is not found. Further, market regimes influence economic integration. In particular adverse market conditions increase economic integration.

Our contribution to the literature is to improve the knowledge of the impact the euro has had on economic integration with economies of different characteristics outside and within the European market. Further, our paper contributes to the literature via inclusion of market conditions influence on the level of economic integration. As far as we understand, this is the first study in this field conducting a cross-quantilogram method to examine the impact of different market conditions on the correlations, making our study a pioneer in the field of economic integration. Therefore, we intend to fill a gap in the existing literature by combining the aggregated and time-dependent approaches including the market regimes.

This paper proceeds with a presentation of the related literature in section 2. Section 3 presents the theoretical framework including economic integration and optimum currency area theory. Section 4 presents the methodology which consists of methods used in this paper such as EGARCH, DCC and cross-quantilogram. Section 5 discloses data and primary analysis where summary statistics, break- and stationarity tests are presented. In Section 6 the results are presented and a discussion is held, followed by conclusion and policy implications in section 7.

(9)

4

2 Related literature

The literature of business cycle synchronization has historically had a focus on Europe and the global developed countries such as the US and Japan. A highly researched question is the impact of common currency areas on business cycle synchronization. The question proceeds from Mundell’s (1961) work of optimum currency area theory. In Europe, this has been studied concerning the implementation of the euro as the common currency for EMU. By using different time series approaches several authors (e.g. Afonso and Sequeira, 2010; Lafuente and Ordóñez, 2009; Rua, 2010; Yang et al., 2003a) have found that EMU significantly increased economic integration between the member countries. During the last century, we have seen an increased number of studies examining economic integration in emerging markets, as for the Asia (e.g. Johnson and Soenen, 2002; Moneta and Rüffer, 2009; Komatsubara et al., 2017). Yet, the focus has remained on synchronization within the specific geographical or common currency area group. Variables such as gross domestic product (GDP), industrial production indices and inflation have been used for industrial sectors to examine business cycle synchronization. While for financial sectors, mainly financial variables such as equity or stock indices have been used. The usage of the exchange rate as the proxy variable for business cycle synchronization is in contrast rare, though Moneta and Rüffer (2009) have explained the exchange rate of Japan yen and US dollar to be influential for synchronization.

Since business cycles are studied over time, time series is the most frequently used econometric approach. The choice of time series model varies along the motivation for the chosen method. The VAR model enables the studying of causality and direction. GARCH is often used for financial data and enables the modelling of time-varying volatility. Wavelet analysis is the third method most used in the previous literature of economic integration. The benefit of Wavelet is that it distinguishes changes in the pattern of business cycle synchronization, through the inclusion of time and frequency domain (Bekiros et al., 2015). Apart from these methods of time series, others have also been used to study economic integration.

Looking at the European market, Afonso and Sequeira (2010) calculated the correlation for 27 EU countries’ individual cyclical GDP components and the EU, as a group, cyclical GDP components. The data was on an annual basis, running from 1970 to 2009. Notably, the synchronization increased for the sample over the entire period and reached a higher level after the euro implementation. A similar conclusion is conducted by Rua (2010) when using a wavelet approach to measure the degree of co-movement for the major euro countries, Germany, France, Italy and Spain. Since the mid-1990’s integration in GDP has increased and is possibly an effect of the establishment of EMU. Shifting focus to the financial market, Yang et al. (2003a) used VAR to conduct that the stock markets of EMU member economies have become more integrated and interdependent. Another interesting finding is the results regarding the UK stock market. Since the foundation of EMU, the level of integration has declined with the UK (Yang et al. 2003a). Laufente and Ordóñez (2009) go further as they state that long-run financial integration only appears among the Eurozone members and not with the UK. This long-run relationship between European stock markets however occurred before implementation of the euro and was examined through the dynamic conditional correlation multivariate GARCH model (DCC-MV-GARCH).

While the UK is a significant economy and financial actor in Europe (Lane and Milesi-Ferretti, 2008), Germany together with France constitute the core of the EMU. The two core countries are also shown to have the highest economic integration among countries in EU according to Aguiar-Conraria and Soares (2011). Their sample consists of EU-12 and EU-15 and therefore includes both core and periphery countries. The proxy variable is composed of industrial production indices, and a robustness test is conducted through Monte Carlo estimations for the significant synchronizations from the wavelet analysis. The influence of Germany on European business cycle is also verified by Artis and Zhang (1997). By including global countries such as the US, Japan and Canada, Artis and Zhang (1997) find that the European countries have shifted business cycle linkages from the US towards Germany since the formation of the European Exchange Rate Mechanism. The UK was, however, not comprised in the shift towards Germany. The sample period extended from 1961 to 1993 and thus excludes the euro period. The hypothesis of Germany or any core country as an attractor for European co-movement is opposed by Camacho et al. (2006) when using four different approaches, VAR-based, spectral-based, dummy and comprehensive

(10)

5 approach. By using industrial production indices for the euro countries, Camacho et al. (2006) show the existence of an attractor is shown to be negligible. However, the existence of higher integration in business cycles for the founder members of the EMU than for new members is shown.

Both in present time and a century ago the geographical proximity has made an impact on economic integration. In Bergman and Jonung’s (2011) paper, the Scandinavian countries increased their synchronization during the Scandinavian Monetary Union, existing in the years around the 19th_{and 20}th

century. Aguiar-Conraria and Soares (2011) also find that geographically close countries have more synchronized business cycles. Overall, Rose and Engel (2002) verify the findings of higher synchronized business cycles for the member countries of a currency union than for countries with sovereign currencies. With a sample not only referring to EMU they find that the countries inside a currency union experience more trade and less volatile real exchange rates.

The European market has grown and become a dominant financial market according to Fratzscher (2002). Fratzscher (2002) analyzes European equity markets between 1986 and 2000 using a GARCH model with three key results. (1) European equity markets have become highly integrated since 1996, (2) the euro area market has taken over from the US as the dominant market in Europe. Lastly, (3) the high integration within European equity markets is mainly explained by the development of EMU and the elimination of exchange rate volatility. A greater impact for European financial market and a shift from the US towards Europe is further examined by Diebold and Yilmaz (2014) using VAR and a developed connectedness framework. Focusing on the integration between the US and Europe, Pérez-Rodríguez (2006) finds a high correlation for the European currencies, the euro and British pound, against the US dollar by using a DCC-GARCH model. Further, a strong connection is shown to exist between the European Central Bank reference rate and the US traded spot rates according to the study. A high connectedness could partially be explained by Wang and Wen (2007) who analyze inflation co-movements. According to their study, when the US experience high inflation after an output boom, Europe will also experience high inflation. Their results are an argument to connect inflation and GDP as a measurement for synchronization.

In contrast to the economies in Europe, which can mostly be characterized as developed countries, the Asian market consists of a higher share of developing countries. Several studies have shown high integration for the emerging and developing economies within Asia (e.g. Johansson, 2011; He and Liao, 2012; Komatsubara et al., 2017). Since 2007 the integration has increased significantly, and the global financial crisis in 2008 could potentially explain this (Komatsubara, et al., 2017). This argument is supported by a second study from Yang et al. (2003b) showing strengthened integration during the Asian financial crisis in 1997. Also, during and in the aftermath of the global financial crisis in 2008 a high level of co-movement was shown for China and Asia-Pacific economies according to Berdiev and Chang (2015). The authors used a wavelet approach and measured synchronization in growth cycles, with real GDP as the main variable. Additional to China, Japan is along with the US also synchronized with the Asia-Pacific countries. It is further shown that the level of business cycle synchronization fluctuates across frequencies and over time. GDP, industrial production indices, equity and stock indices, and inflation are common proxy variables used to measure and analyze business cycle synchronization. In contrast to most of the previous literature, Moneta and Rüffer (2009) explain the exchange rate of Japanese yen and US dollar to be influential for economic integration. According to Flood and Rose (2010), the mentioned variables from previous literature together with exchange rate are argued to be key proxy variables when analyzing integration, bilateral or multilateral, between economies. Additionally, Flood and Rose (2010) explain the importance of using inflation targets. Countries who target inflation are slightly more synchronized with foreign cycles. Sethapramote (2015) used the Association of Southeast Asian Nations, ASEAN, countries and measured static and dynamic correlations in several macroeconomic and policy variables. A main finding was the evidence of synchronization in the key variables. Further, the study showed that the influence of trade integration on business cycle synchronization was greatest within ASEAN. The financial integration was found most important between ASEAN and the US. One study with an expanded sample is Loyaza et al. (2001). Except for the European economies and East Asia, they also include Latin America. They found

(11)

6 significant co-movement within both Europe and East Asia, on short- and long-run, but not for Latin America.

Eickmeier and Breitung (2006), Park and Shin (2009), Quah and Crowely (2012) and Sethapramote (2015) are examples of studies which includes several variables conducting different types of integration. However, only Park and Shin (2009) studied a sample consisting of countries from Europe, Asia and America. Their research question was to analyze East Asia’s business cycle since 1990 and to examine whether the cycle has strengthened within the group and decoupling from EU and the US. This makes the study one of few in the field looking at cross-regional integration.

In summary, the previous literature has had a focus on synchronization within a specific geographical area or a common currency area, with Europe and the EMU as one relevant example. Variables as GDP, industrial production and inflation have been used along with financial variables as equity and stock indices. Among the time series methods, VAR, GARCH and Wavelet analysis seem to be most commonly used in previous studies. Overall, a significant impact of EMU towards an increased business cycle synchronization has been shown, with higher economic integration for the euro countries than for countries with sovereign currencies. Characteristics such as being a core country also affect the integration positive among countries in EU. For the Asian market, financial crises appear to strengthen integration. Few studies have examined cross-regional integration. Therefore, we see a research gap in the previous literature.

(12)

7

3 Theoretical framework

3.1 Economic integration

Economic integration is the direction of economic theory which analyzes the effects of integration between economies. The first studies in the field of economic integration used a static effect analysis to assess the impact of integration on welfare. The static effect analyzes the effects of trade creation and trade diversion (Appleyard and Field, Jr, 2016). According to Viner (1950), using the static effect analysis, customs unions and free trade areas do not always enhance welfare, and sometimes even lower the global economic welfare (Hosny, 2013; Marinov, 2014). Viner’s (1950) study was the incipient work to distinguish between the advantages and disadvantages of economic integration and expresses the incentives that countries must encourage integration only if the benefits exceed the costs, or in other words when integration leads to more trade promotion than trade reduction.

Lipsey (1957) argued that Viner’s (1950) analysis was concentrated on the production side only, neglecting the consumption effect. This consumption effect arises because members of the union will increase their consumption of each other’s products while reducing their consumption of products from countries outside the union (Hosny, 2013; Marinov, 2014). Also, the static method used by Viner (1950) has further drawbacks and may not be sufficient for the analysis of trade creation and trade diversion, which is argued by Balassa (1962). Balassa (1962) concluded that static analysis was not enough to analyze the effects of economic integration on welfare and hence introduced a dynamic effect analysis of economic integration. The dynamic effect analysis implies that the welfare effects of economic integration could be an economic reason behind the creation of economic cooperation between countries. Balassa (1962) defined the dynamic effect of integration as “large-scale economies, technological change, as well as the impact of integration on market structure and competition, productivity growth, risk and uncertainty, and investment activity”. Marinov (2014) summarized the dynamic effects of economic integration as economic growth, sustainable increase in demand, increase in investments, strengthened production, increased specialisation and production efficiency, utilization of resources and area. Schiff and Winters (1998) compiled the definition of the dynamic effects of economic integration as all events that affect the rate of medium- and long-term economic growth of the members in the integration agreement. Since the introduction of the dynamic effect analysis, this is the most commonly used method in the field of economic integration.

Abolishment of discrimination within an area influence the level of economic integration (Balassa, 1962). This discrimination can be interpreted by the abolition of trade barriers, e.g. on goods and services among the area’s countries which will increase economic integration (Balassa, 1962). Less discrimination and hence higher degree of economic integration can merely be a result for the members of a trade alliance (Baldwin, 2008), for which few or no trade barriers exist. From the work of Balassa (1962) and Baldwin (2008), economic integration can be interpreted as the transformation in welfare occurring in an area when it and one or more areas reduce their regulations against each other. The decrease in regulation ought to increase the co-movement between the areas which in turn will increase the welfare among the participating areas in the partnership. The increased economic integration due to closer economic ties can also foster increased co-movement in various macroeconomic variables including linkages in financial, production and trade sectors (Dées and Zorell, 2011). The transformation and thus closer integration can be achieved by increasing trade in goods or by increasing flows of ideas (Rivera-Batiz and Romer, 1991). The theoretical argument behind this reasoning is that long-run correlation and cohesion relate to co-integration, countries with linkages in terms of co-movement will increase their integration via business cycle synchronization (Croux, et al., 2001).

Brada and Mendez (1988) illuminate that increasing economic integration assumes to raise the investments and reduce risk in countries affected by the integration. Foreign direct investments along with private sector participation and market allocation of resources are factors to influence economic integration (Lawrence, 1996). According to Hosny (2013), the rise in investments and reduction in risk is because “larger markets will raise the expected return on investments and reduce uncertainty by enabling firms to lower their cost as a result of increased economies of scale, and a bigger pool of customers”. Development of the financial

(13)

8 markets due to economic integration can be viewed as financial integration. Financial integration is the process in which countries’ financial markets become more integrated (Xing and Abbott, 2007). Further, Xing and Abott (2007) describe the development process through the elimination of barriers for financial institutions and then conclude that it may imply linking banking, equity and other types of financial markets together. Baele et al. (2004a) define financial integration via the law of one price, which states that in integrated markets assets generating the same cash flows will be priced equivalently irrespective of on which market the transaction takes place. According to Jain and Bhanumurthy (2005), the financial integration entails an increased flow of foreign capital and a tendency for financial asset price to return to converge. Financial integration may take many forms, Kalemli-Ozcan et al. (2014) mention forms such as bank lending, portfolio investment and foreign direct investments.

Overall, financial integration describes how the financial markets are connected. Baele et al. (2004b) characterize a market to be completely integrated, with a specific set of financial instruments and/or services, if all potential market actors have the same conforming characteristics: (1) Actors front a single set of rules when they use the financial instruments and/or services. (2) Actors have equal access to the financial instruments and/or services. (3) Actors are treated equally at the market. Markets which do not fulfil these requirements can, therefore, be viewed as segmented. According to Aggarwal et al. (2010) a segmented market is the result of obstructions, e.g. high transaction costs, for the integration.

3.2 Optimum currency area theory

A self-administrated domestic central bank using flexible exchange rate has the economic strength to counter events that otherwise could lead to a financial setback or depression, and thus govern the national monetary policy (Mundell, 1961). When entering a common currency area, the possibility for a sovereign central bank disappears, which can be seen as a disadvantage. Easily viewed, the advantages of entering and being part of a common currency area must outweigh the disadvantages. The theory of optimum currency areas starts with Mundell’s work in 1961, with developing thoughts by McKinnon (1963), Kenen (1969) and Mundell (1973). Optimum currency areas divide between those with a single currency and those with several currencies, all with a fixed exchange rate (Mundell, 1961). Mundell (1961) separates these different areas as interregional and international, for which different adjustment process’ are needed to be optimal. One of the adjustments that differ is the structure of central banks included. For the international currency area with additional currencies and central banks, cooperation between the central banks is essential (Mundell, 1961). In his work, Mundell (1961) refers to diverse possibilities for the different areas when dealing with inflation and unemployment. Though, the success of dealing with inflation and unemployment does not depend on the type of currency area. Instead, Mundell (1961) appoint the domain of the currency area as essential to target these policy tasks. Overall, smaller and more homogeneous areas are optimal and preferable over larger and heterogeneous areas according to Mundell (1961).

Homogeneity can, for instance, be expressed through the political context in the area. Western Europe with its common market exemplifies a homogeneous region that Mundell (1961) brings up in his discussion of possible optimum currency areas. It is essential that the currency area builds on a region to conduct stabilization (Mundell, 1961). According to McKinnon (1963), the size of the currency area and the level of openness within the currency area are essential variables for the degree of efficiency. Economies with extensive exports and imports toward each other, do benefit from a currency area due to less currency fluctuations (McKinnon, 1963). In opposition to Mundell (1961), McKinnon (1963) argues that an optimum currency area consists of a large area because of the idea of diversification among additional and major economies. Diversification within the currency area is the most crucial element according to Kenen (1969). A highly diversified production within the currency area softens negative consequences, such as higher unemployment, from external shocks (Kenen, 1969).

One of the most influential elements discussed in the paper of Mundell (1961) is the presence of factor mobility. A currency area should establish if the region has internal factor mobility. On the other hand, if internal factor immobility exists, the currency area with flexible exchange rates cannot be expected to

(14)

9 perform as a stabilization function and an optimum currency area. An optimum currency area does for similar reasons not sustain if factors are mobile with countries outside the region (Mundell, 1961).

In 1973, Mundell developed the thoughts of optimum currency areas, the focus shifted to explaining risk-sharing within the currency area. If one country or region within the common currency area suffers from an adverse shock, the country or region is argued to better negotiate from the shock through risk-sharing with its common currency countries. The argument builds on mobility within the currency area as the member countries influence each other’s output and thus share possible losses. It is possible to view the solution to this issue as the common currency area using joint reserves which are allocated to strengthen the most vulnerable country through the recession. If the countries have separate currencies with flexible rates, the lack of a common currency area will make the country bear the loss alone, as a currency area and its members cannot serve as a shock absorber (Mundell, 1973). If mobility exists between two countries and they are in fact dependent on each other, Mundell (1973) motivates a common currency as an automatic and equal risk sharer of fluctuations in the economy. In the long-run, both countries in a currency area benefits, even if the well-being is temporarily lowered in country A to counter an adverse shock that has hit country B. A global currency area with a single currency is, despite a substantial reserve, not sustainable due to increased transaction costs to retain the area efficiently (Mundell, 1973).

(15)

10

4 Methodology

In this paper, we apply time series econometric models to investigate the correlation between our two variables and twelve countries. Time series it the most applicable method used to study economic integration as seen in the previous literature.

4.1 Stationarity

Econometric models using a time series approach are dependent on stationarity for the model to produce persistence and robust results and thus avoid spurious regression. A stationary process is a stochastic process whose unconditional joint probability distribution does not vary over time. If a time series is a non-stationary process, it contains a unit root process which indicates a problem with time dependency among the variables (Verbeek, 2012). The unit root tests, Augmented Dickey-Fuller (1979) and Philipps-Perron (1988), are conducted to ensure that the variables are stationary before proceeding to the mean equation. The Augmented Dickey-Fuller test (ADF) is an augmented version of the Dickey-Fuller test. The null hypothesis is that a series has a unit root and can be tested against the alternative hypothesis that a series is stationary. The test is carried out considering two different models, using a constant, equation (1), or a constant and trend, equation (2). The ADF test whether 𝛿 = 0, if 𝛿 ≠ 0 we can reject the null hypothesis that the series has a unit root. In equation (1) and (2), ∆ is the first difference operator, 𝛾 is a coefficient, 𝑡̅ is the linear time trend, 𝛿 is the test coefficient, y_t-1 is the lagged dependent variable in logarithm form, ∆𝑌𝑡−𝑖 is the lagged dependent in first difference, and 𝜀 is the residual.

∆𝑦_𝑡 = 𝛾 + 𝛿𝑦_𝑡−1+ ∑ 𝛼_𝑖∆𝑦_𝑡−𝑖 𝑘 𝑖=1 + 𝜀_𝑡 (1) ∆𝑦_𝑡 = 𝛾 + 𝛽_𝑡̅+ 𝛿𝑦_𝑡−1+ ∑ 𝛼_𝑖∆𝑦_𝑡−𝑖 𝑘 𝑖=1 + 𝜀_𝑡 (2)

While the ADF solves the problem of non-white noise residuals by adding lags to the dependent variable, Philipps-Perron test (PP) apply a non-parametric correction of the test statistic so the Dickey-Fuller distribution can be used without white noise. PP apply a t-statistic which creates a robustness test that corrects for autocorrelation and heteroscedasticity. The null hypothesis is that a series has a unit root and can be tested against the alternative hypothesis that a series is stationary. The test is carried out considering two different models, using a constant, equation (3), or a constant and trend, equation (4). The PP test whether 𝛾 = 0, if 𝛾 ≠ 0 we can reject the null hypothesis that the series has a unit root. In equation (3) and (4), ∆ is the first difference operator, 𝜑 is a constant, 𝑡̅ is the linear time trend, 𝛾 is the test coefficient, y_t-1 is the lagged dependent variable in logarithm form, ∆𝑌𝑡−𝑗 is the lagged dependent in first difference,

and 𝜀 is the residual.

∆𝑦_𝑡 = 𝜑 + 𝛾𝑦_𝑡−1+ ∑(𝛿∆𝑦𝑡−𝑗) + 𝜀𝑡 𝑝 𝑗=1 (3) ∆𝑦_𝑡 = 𝜑 + 𝛾𝑦_𝑡−1+ 𝛽_𝑡̅+ ∑(𝛿∆𝑦𝑡−𝑗) + 𝜀𝑡 𝑝 𝑗=1 (4)

(16)

11 The Kwiatkowski-Philipps-Schmidt-Schin test (KPSS) propose an additional test for stationarity where stationarity around a deterministic trend, linear or mean, is examined. The null hypothesis for the test is that the series is stationary, and the alternative hypothesis is that the series has a unit root. This specification to circumvent the problem that the unit root often has low power (Verbeek, 2012). The KPSS test statistic is based on a one-sided Lagrange multiplier statistic expressed in equation (5). Where 𝑆𝑡2 is the partial sum

process of the residuals, 𝜎̂𝜀2 is the estimate of the long run variance of the residuals. (Kwiatkowski, et al.,

1992). 𝜂 = ∑𝑆𝑡2 𝜎̂_𝜀2 𝑛 𝑡=1 (5)

4.2 Generalized Autoregressive Conditional Heteroscedasticity model

The GARCH model developed by Bollerslev (1986) is a generalization of the autoregressive conditional heteroscedasticity (ARCH) model developed by Engle (1982). ARCH means that the variance of a process changes systemically over time, the ARCH model is used to model this time-varying volatility in time series data. Several economic time series, in particular financial time series (Verbeek, 2012), assumes a volatility clustering pattern where large shocks, e.g. increased demand, tend to be followed by a large shift in either direction and small shocks tend to be followed by small shocks (Verbeek, 2012). This clustering pattern creates a problem with heteroskedasticity and variance varying over time which the ARCH- and GARCH models consider.

A long lag structure in the ARCH process can be substituted with lagged dependent variables to create a shorter process. In its general form, the GARCH (p, q) model can be written as equation (6), where 𝛼(𝐿) and 𝛽(𝐿) are lag polynomials, 𝜎𝑡2 is the residual variance, 𝜛 is the constant residual variance, and 𝜀𝑡 is the

unconditional variance.

𝜎_𝑡2_{= 𝜛 + 𝛼(𝐿)𝜀}

𝑡−12 + 𝛽(𝐿)𝜀𝑡−12 (6)

The basic GARCH model is the univariate GARCH (1, 1) model in which the conditional variance matrix is calculated from a long-run average variance rate and lagged terms. The GARCH (1, 1) has three unknown parameters to estimate. To achieve non-negativity of 𝜎𝑡2 it requires that 𝜛, 𝛼 and 𝛽 are non-negative.

Stationarity requires that 𝛼 + 𝛽 < 1, if the value of 𝛼 + 𝛽 is close to 1 the persistence in volatility is high.

𝜎𝑡2= 𝜛 + 𝛼𝜀𝑡−12 + 𝛽𝜀𝑡−12 (7)

Economic time series often react differently to adverse shocks than to positive ones. Adverse shocks often have more impact on the future volatility than the positive shocks (Verbeek, 2012). The EGARCH model developed by Nelson (1991) is a development from the univariate GARCH model which allows for asymmetric effects from negative and positive shocks. The GARCH model assumes that 𝜎𝑡2 must be

non-negative with a probability of one to be the conditional variance of 𝜀𝑡 given the information of 𝑡. To ensure

this the GARCH model composes 𝜎𝑡2 as a linear combination of positive random variables. The EGARCH

model adopts another natural function to ensure that 𝜎𝑡2 remains non-negative, by composing ln (𝜎𝑡2) as

linear in some functions of time. To accommodate the asymmetric relationship between return and volatility the value of 𝑔(𝑧𝑡) must be a function of both the magnitude and sign of 𝑧𝑡. If 𝑔(𝑧𝑡) is a linear combination

of 𝑧𝑡 and |𝑧𝑡|, 𝜎𝑡2 is given “well-behaved moments" (Nelson, 1991). This linear combination of 𝑔(𝑧𝑡) is

(17)

12 ln(𝜎_𝑡2_{) =∝} 𝑡+ ∑ 𝛽𝑘𝑔(𝑧𝑡−𝑘) ∞ 𝑘=1 (8) 𝑔(𝑧_𝑡) ≡ 𝜃𝑧𝑡+ 𝛾[|𝑧𝑡| − 𝐸(𝑧𝑡)] (9)

The infinite moving average process described in equation (8) can be expressed in simpler terms, equation (10), (11) and (12). Where α0 is a constant, α𝑡is a shock of a return at time t, 𝜀𝑡 is the standardised shock,

𝜎_𝑡 is the square root of the volatility, 𝜎𝑡2 is the volatility, 𝐵 is the lag operator, 𝑔(𝜀𝑡) is a linear combination

of 𝜀𝑡 and is an independent and identically distributed random sequence with mean zero and variance one.

(Jane and Ding, 2009).

ln(𝜎_𝑡2_{) = 𝛼} 0+ (1 + 𝜑₁𝐿 + ⋯ + 𝜑_𝑞𝐿𝑞₎ (1 − ∆₁𝐿 − ⋯ − ∆_𝑝𝐿𝑝₎𝑔(𝜀𝑡−1) (10) 𝑎_𝑡 = 𝜎_𝑡𝜀_𝑡 (11) 𝑔(𝜀_𝑡) = 𝜃𝜀_𝑡+ 𝛾[|𝜀_𝑡| − 𝐸(𝜀_𝑡)] (12)

4.3 Dynamic Conditional Correlation model

The DCC-GARCH model is a generalization of Bollerslev’s (1990) constant conditional correlation (CCC)– GARCH model. In the CCC-GARCH the conditional correlation matrix is constant over time. Engle (2002) proposed a model where the conditional correlations matrix is time-dependent, the DCC-model. When generalizing the assumption of constant correlation over time and instead make it time-dependent, a more efficient method of modelling correlations emerges (Pérez-Rodríguez, 2006). Hence, the DCC-GARCH model makes it possible to examine the development of integration over time, and how it affects volatility pattern of shocks. For equation (13) 𝑟𝑡 is the vector of variables, 𝐻𝑡 is the conditional covariance matrix of

𝑟𝑡 and 𝑧𝑡 is a random vector of errors. In equation (14) 𝐷𝑡 is the dynamic standard deviation and 𝑅𝑡 is the

conditional correlation matrix equation.

𝑟𝑡 = 𝜇 + 𝑎𝑟𝑡−1+ 𝐻𝑡0.5𝑧𝑡 (13)

𝐻𝑡 = 𝐷𝑡𝑅𝑡𝐷𝑡 (14)

The DCC-GARCH is a two-step model estimating the parameters of a nonlinear combination of a univariate and a multivariate GARCH model. The first step is to estimate the conditional variance for each of the series. Given the parameters from the first step, the conditional correlation is estimated. The conditional correlation is calculated through equation (15) as 𝜌𝑖,𝑗,𝑡, for series i and j in time t and 𝑞𝑖. 𝑞𝑖,𝑗,𝑡

is the conditional covariance between the standardized residuals. 𝑞𝑖,𝑖,𝑡 respective 𝑞𝑗,𝑗,𝑡 is corresponding

series standardized residuals. The DCC model includes conditions that make the process covariance stationarity and the variance-covariance matrix positive determined at all times.

𝜌_{𝑖,𝑗,𝑡} = 𝑞𝑖,𝑗,𝑡

(18)

13

4.4 Goodness-of-fit

Goodness-of-fit tests are performed to test whether the chosen GARCH model is suitable. First, the squared Ljung-Box test (Ljung and Box, 1978), equation (16), tests whether the univariate GARCH model exhibit white noise in the squared residual. Under the null hypothesis, the univariate GARCH model does not exhibit a lack of fit and white noise is reported in the squared residual. The test is illustrated in equation (17). 𝑄(𝑚) = 𝑇(𝑇 + 2) ∑ 𝜌̂12 𝑇 − 𝑘 𝑚 𝑘=1 (16)

Second, the squared Li-McLeod test (Li and McLeod, 1981), equation (17), tests whether the chosen multivariate GARCH model, namely our DCC-GARCH is appropriate. Under the null hypothesis, the multivariate GARCH model is adequate.

𝑄_𝑚∗ _{= 𝑄} 𝑚+

𝑘2_{𝑚(𝑚 + 1)}

2𝑛 (17)

The specification of GARCH-types is based on the log-likelihood ratio function (Neyman and Pearson, 1933), equation (18), and Akaike’s Information Criterion (Akaike, 1974), equation (19). The specification of GARCH-types of models is based on the log-likelihood ratio function, equation (18), and Akaike’s Information Criterion, equation (19). T denotes the sample size, ℎ𝑡 is the lagged variance, 𝜀𝑡2 is the squared

conditional variance, 𝜎̂𝑒2 is the estimated residual variance and 𝑘 is the number of lags.

𝐿𝑜𝑔 𝐿 = 𝑇 2log 2𝜋 − 𝑇 2 ∑(𝑙𝑜𝑔ℎ𝑡) 𝑇 𝑡=1 +𝜀𝑡 2 ℎ_𝑡 (18) 𝐴𝐼𝐶 = 𝑙𝑜𝑔𝜎̂𝑒2+ 2𝑘 𝑇 (19)

4.5 Cross-quantilogram

Quantilogram was introduced by Linton and Whang (2007) to measure the predictability in different sectors of the distribution in a stationary time series based on the correlogram of “quantile hits” (Han, et al., 2016). Linton and Whang (2007) applied quantilogram to test directional predictability and to examine the null hypothesis that a given time series has no directional predictability. The test for predictability is conducted via comparing the quantilogram to a pointwise confidence interval. Han et al. (2016) highlight several advantages for directional predictability for quantilogram compared to other tests. Among the advantages, the method is based on “quantile hits” which does not require moment conditions like the ordinary correlogram and it is applicable for series with strong tailedness.

Han et al. (2016) developed the univariate quantilogram framework by Linton and Whang (2007) to a multivariate setting to measure the quantile dependence between two stationary time series. The cross-quantilogram applies conditional quantiles to measure the directional dependence between two time-series after parsimoniously controlling for the information at the prediction. Moreover, the applied distribution is asymptotic to be valid uniformly over a range of quantiles.

The cross-quantilogram captures the serial dependence between the series at different conditional quantiles, expressed in equation (20) while equation (21) expresses the quantilogram for a sample. The

(19)

cross-14 quantilogram is a measure of the serial dependence between two events {𝑦1𝑡 ≤ 𝑞1,𝑡(𝜏1)} and {𝑦2,𝑡−𝑘 ≤

𝑞𝑤,𝑡−𝑘(𝜏2)} for any pair of 𝜏. 𝜏 denotes the range of quantiles we are interested in evaluating for the

directional predictability. The quantile of 𝑦𝑖,𝑡 is 𝑞𝑖(𝛼𝑖) = inf(𝜐: 𝐹𝑖(𝜐) ≥ 𝛼𝑖) for 𝛼𝑖 ∈ (0, 1). The

indicator function is expressed by 1[∙] and {1[𝑦𝑖𝑡 ≤ 𝑞𝑖,𝑡(∙)]} is called the “quantile hit” process for i = 1,

2. The cross-quantilogram is defined as the cross-correlation of the “quantile hit” process for 𝑘 = 0, ±1, ±2, … , ∞, where 𝜓𝑎(𝑢) ≡ 1[𝑢 < 0] − 𝑎 (Han, et al., 2016).

𝜌_𝜏(𝑘) = 𝐸[𝜓𝜏1(𝑦1𝑡− 𝑞𝑞,𝑡(𝑇1)𝜓𝜏2(𝑦2,𝑡−𝑘− 𝑞2,𝑡−𝑘(𝜏2))] √𝐸[𝜓_𝜏12 _(𝑦

1𝑡− 𝑞𝑞,𝑡(𝜏1)√𝐸[𝜓𝜏22 (𝑦2,𝑡−𝑘− 𝑞2,𝑡−𝑘(𝜏2))

(20)

The cross-quantilogram in equation (20) captures serial dependence between two series at different quantile levels, taking 𝛼 = (𝛼1, 𝛼2) = (𝛼𝐹𝐼𝑁, 𝛼𝐺𝐸𝑅) as an example 𝜌𝛼(1) measures the cross-correlation between

the Finnish market being above or below quantile 𝑞𝐹𝐼𝑁(𝛼𝐹𝐼𝑁) at time t and the German market being

above or below quantile 𝑞𝐺𝐸𝑅(𝛼𝐺𝐸𝑅) at time t-1. Therefore, 𝜌𝛼(1) = 0 implies that whether the German

market is above or below quantile 𝑞𝐺𝐸𝑅(𝛼𝐺𝐸𝑅) at time t, it does not help to predict on average whether the

Finnish market will be above or below the quantile 𝑞𝐹𝐼𝑁(𝛼𝐹𝐼𝑁) on the next event in time. In contrast, if

𝜌_𝛼(1) ≠ 0 there exist a one event direction predictability from Germany to Finland at 𝛼 = (𝛼𝐹𝐼𝑁, 𝛼𝐺𝐸𝑅).

Han et al. (2016) consider a linear quantile regression model proposed by Koenker and Basset (1978) where 𝑞_𝑖,𝑡(𝜏𝑖) = 𝑥𝑖𝑡𝑇𝛽𝑖(𝜏𝑖) with a 𝑑𝑖∗ 1 vector of the unknown parameters 𝛽𝑖(𝜏𝑖) for i = 1, 2. Equation (21)

expresses a minimization problem which estimates the unknown parameters where 𝜌(𝑢) ≡ 𝑢(𝑎 − 1[𝑢 < 0]), 𝛽̂(𝜏) ≡ [𝛽̂1(𝜏1)𝑇, 𝛽̂2(𝜏2)𝑇]𝑇 and 𝑞̂𝑖,𝑡(𝜏𝑖) = 𝑥𝑖𝑡𝑇𝛽̂𝑖(𝜏𝑖) for i = 1, 2. For observations

{(𝑦_𝑡, 𝑥_𝑡)}_𝑡=1𝑇 _{the sample counterpart of cross-quantilogram, equation (21), is constructed from the}

estimated conditional quantile functions.

𝛽̂_𝑖(𝜏_𝑖) = arg min 𝛽𝑖𝜖 ℝ𝑑𝑖 ∑ 𝜚_𝑇 𝑖(𝑦𝑖𝑡−𝑥𝑖𝑡𝑇𝛽𝑖) 𝑇 𝑡=1 (21) 𝜌̂_𝛼𝑘 = ∑ 𝜓𝛼(𝑦𝑡− 𝑇−𝑘 𝑡=1 𝜇̂𝛼)𝜓𝛼(𝑦𝑡−𝑘− 𝜇̂𝛼) √∑ 𝜓_𝛼2_(𝑦 𝑡− 𝑇−𝑘 𝑡=1 𝜇̂𝛼) √∑𝑇−𝑘𝑡=1 𝜓𝛼2(𝑦𝑡−𝑘−𝜇̂𝛼) , 𝑘 = 1,2, … , 𝑇 − 1 ₍₂₂₎

The cross-quantilogram method results in two output, quantile cross-correlation heatmap and rolling windows for different market regimes. In this paper, the results are specified with a lag length of one, bootstrap iterations of 500 and a significance level of 0.05. A lag length of one indicates that we measure the correlations between two months. The selection of bootstrap is evaluated with iterations between 100 and 1000. Based on robustness, where the results are consistent over time, 500 iterations are selected. The significance level of 0.05 is selected based on standard econometric arguments to prevent rejecting a correct null hypothesis.

The quantile cross-correlations heatmap is based on calculating the average quantile cross-correlations between countries. The interpretation of the quantiles is that lower quantiles, e.g. quantile 0.05, displays adverse market conditions, normal market conditions are represented by quantiles in the middle, e.g. 0.5 and higher quantiles presents good market conditions, e.g. 0.95. The interpretation of this results is based

(20)

15 on a scale from dark red via white to dark blue, where dark red illustrates high positive correlation and dark blue represents high negative correlation. Interpreting market conditions influence is essential due to the development of globalization and changing uncertainty in different market regimes.

The rolling window estimation is based on the average quantile cross-correlation during the window and the time-dependent change from one window to another. Hence, the rolling window cross-quantilogram model makes it possible to examine the development of integration over time when controlling for the impact of market conditions. The rolling window size is selected to 48 months, the argument for this selection is based on political term periods which often extends 48 months. The rolling window figures also include confidence intervals, the interpretation of the result is that when the blue line is inside the confidence intervals the change is statistically significant. When the line is outside the confidence interval, the result is not statistically significant.

4.6 Methodological criticism

The choice of using EGARCH as our specified GARCH model instead of, e.g. GJR-GARCH (Glosten, Jagannathan and Runkle, 1993) is that the EGARCH estimations generate values for a positive respectively an adverse shock, referred to as Theta-values later in this paper. Therefore, the EGARCH, unlike other GARCH models, enable us to analyze the predictions of how the series affects by economic shocks. Regarding the cross-quantilogram estimations, criticism can be expressed in the specifications of the lag length, bootstrap iterations, significance level and rolling window size, which have an impact on the results. Other specifications could be used. However, we use economic arguments to set up the specifications as explained in section 4.5.

Choosing to fit the same specification of the EGARCH model for both the univariate and the multivariate estimations does not generate the best model individually for the estimations, which can be viewed as a limitation for the GARCH approach. Hence α + βcan be above one for the estimations, indicating non-mean reverting series. The choice of using the same orders for the univariate and multivariate estimations yield a possibility to directly compare the results between the different groups, namely European drivers, global and emerging markets. This without making allowances for the degree of impact in different orders. Further, the choice to limit the sample of autoregressive and moving average (ARMA) process up to (5, 5), as explained in chapter 6.1, could induce to miss the best fitted ARMA-order. The same drop-out can also exist by restricting the GARCH-process, also described in section 6.1. The limitation of the ARMA- and GARCH-orders has been considered by comparing previous studies in the field.

The methodology applied in this paper could consist of other methods or additional methods to provide a more comprehensive picture of the studied questions, e.g. a Wavelet approach could have enriched this paper results with the directionality of the correlations. The Wavelet approach could further enlighten which country leading the business cycle and hence determine the driver of the economic integration as Bekiros et al. (2015) succeed in their study. A VAR and/or VECM approach could also have contributed to a more comprehensive analysis displaying the causality as seen in previous literature. However, we find the approach with DCC-EGARCH and cross-quantilogram the most efficient way to fulfil the purpose and limit the scope of this paper. Thus, we can examine the economic integration between our groups over time and regarding shocks and market conditions. The latter is unique for the cross-quantilogram analysis.

(21)

16

5 Data and preliminary analysis

The employed dataset contains indices for seasonally adjusted industrial production and stock markets. The indices are available at a monthly frequency which enables a sufficient number of observations for empirical analysis. The investigated period spans from January 1993 to September 2017 and consists of observations from the last day every month. The starting point synchronizes with the implementation of the Maastricht Treaty by the EU which was essential for the establishment of the EMU in 1999 (Bekiros et al., 2015). The selection of countries is based on our purpose and contains twelve countries divided into five sub-groups based on their perspective; common currency, sovereign currency, European drivers, global and emerging markets. The common currency group consists of Finland, France and Italy, this selection is because all countries belong to the EMU as well as that the selection is diversely seen to geographic distribution. Finland is located in Northern Europe, while France and Italy are located in the and southern parts of Europe. Further, France and Italy can be viewed as to core countries, while Finland has the characteristic of a peripheral country in Europe, which may have implications for the level of economic integration (Aguiar-Conraria and Soares, 2011). For the sovereign currency group, we have selected Hungary, Norway and Sweden. These countries have sovereign currencies inside Europe. Norway and Sweden belong to Northern Europe, while Hungary is a country in Eastern Europe. Overall, the diversifications allow for different conditions for the candidates.

Germany and the UK compose the European drivers. Our approach assumes Germany as the primary driver of the European economic activity (Bekiros, et al., 2015) and the UK as the driver of the financial activity in Europe (Lane and Milesi-Ferretti, 2008). In the global group, we have selected the US and Japan due to their characteristics as the financial centres. The US is a world actor and the leading financial driver in the world. Whereas Japan is the leading financial driver on the Asian continent and the Japanese yen is also considered the third most traded currency in the world (Kim, et al., 2013). Both countries are often included as anchor countries in previous studies examining business cycle synchronization (e.g. Hartmann et al., 2003; Berdiev and Chang, 2015; Sethaprmote, 2015). Among the emerging countries, we have selected China and India. This selection is based on the view that they are the most promising emerging economies with a high average growth rate (Herd and Dougherty, 2007). According to Berdiev and Chang (2015), China has in the recent decades increasingly influenced the economic structure of the world.

The industry as a sector has shown to be an influential driver to business cycle synchronization (Afonso and Furceri, 2009), and have frequently been used as a proxy variable in the previous literature (e.g. Artis and Zhang, 1997; Camacho et al., 2006; Aguiar-Conraria and Soares, 2011). Industrial production is used as a measurement for the supply side of the economy. In this context, a preferable variable is GDP since it measures both the demand and the supply side. GDP is also used as the proxy variable for business cycle synchronization in several studies (e.g. Moneta and Rüffer, 2009; Afonso and Sequiera, 2010; Berdiev and Chang, 2015). Though as argued by Camacho et al. (2006), GDP series presents several disadvantages. Mainly the series for GDP are conducted yearly, leading to a narrow sample without magnitude of volatility in the series. Since we aim to analyze economic integration from the forming of the EMU, monthly series are preferable and are as well available for the stock market.

After examining several alternative data sources, industrial production indices are collected from the Organisation for Economic Co-operation and Development, OECD, and the World Bank database. This dual selection depends on the diverse availability of data from the sources. OECD provides industrial production indices for all countries except China and India while the World Bank complements with the data for China and India. The definition of industrial production index coincides between the two sources, as it measures the volume of production output for industry sectors. The World Bank refer the industry sectors to manufacturing, mining and utilities, while OECD includes the total industrial production. The two sources use different base years, OECD uses 2010 while the World Bank uses 2005. To be able to compare and analyze the full sample, we have transformed the data extracted from the World Bank and set the baseline to 2010.

(22)

17 The globalization has led to a significant increase of financial transactions among countries. Capital flows freely over a global market, where the highest return moderates the direction (Büttner and Hayo, 2011). Business cycle synchronization can contribute to financial stability internationally and influence the degree of capital market integration across countries over time (Lafuente and Ordóñez, 2009). Coincidentally, Büttner and Hayo (2011) explain financial integration to be vulnerable to financial crises, thus making the examination of financial markets important. Financial integration tends, besides increasing the international correlation in consumption and GDP, to affect the relationship between output growth and volatility (Lafuente and Ordóñez, 2009).

Thompson Reuters Datastream provides a broad range of financial data and long time periods; hence the stock market indices are collected from this source. The data selection for stock markets is based on the index that reflects the country-specific stock markets as well as the index that contains the entire sample period. For some countries, several indices can be viewed as an acceptable proxy for the stock market. In those cases, we selected the index that appears to have the broadest coverage of stocks within the market. The stock market indices are collected as the price index for the entire sample. For some countries, return index are also available. The return index could be argued as more favourable since it contains reinvested dividends, which better reflects the market development. The price index focuses on solely a price development; hence it will be affected more by noise than the return index. After an investor buys their shares, their interest in the price is reducing in favour of how the company is performing. Depending on the performance of the company it will, in turn, generate dividends which will lower the price index value while the return index will not be affected by dividends. Since return index, doesn’t exist for the entire sample, we use price index. It enables us to make more certain comparisons and analyzes.

All stock market indices are extracted in their local currency and transformed into a common currency. We have selected US dollars as the common currency due to its view as a world currency. Since our sample focuses on the European market, with EMU as our common currency area, one could argue that the transformation is a better fit to the euro. However, the euro as a currency was first implemented in 1999 while our time series expand from 1993, and hence the US dollar is chosen. The transformation to US dollar was conducted via Thompson Reuters Datastream exchange rate conversion function. This transformation implies the influence of exchange rate fluctuations on the indices. However, to be able to compare each countries development in industrial production and the stock market, they need to be expressed in a mutual currency. As argued we find it better to use US dollars than the euro, the choice of euro would have given the sovereign currency countries a disadvantage against the common currency countries towards Germany. Hence the exchange rate fluctuation would only affect the sovereign currency countries.

Table 1 illustrates the first difference of seasonally adjusted industrial production indices and stock market indices for the individual countries during the period 1993M1-2017M9. The first difference denotes the average change in industrial production and the return for the stock market. In comparison to the entire sample, the industrial production and the stock market in China and Hungary are more volatile measured by the standard deviation. For the industrial production, Norway also expresses relatively high standard deviation. Skewness is negative for almost all series which indicates a left-tail of the distribution. The exceptions are India and Norway for industrial production. Further for industrial production, Panel A, China and Japan distinguish with high absolute values, interestingly the same country pair has the lowest absolute values when looking at Panel B, the stock market. Kurtosis indicates a non-normal tailedness of the distribution for almost all series, the exceptions are France in Panel A and Italy and Japan in Panel B. These three countries have Kurtosis values close to the normal distribution. A high Kurtosis and a negative skewness indicate volatility in the series which strengthens the argument of using the GARCH approach. The Jarque-Bera (JB) test confirms that most indices are not normally distributed, France in Panel A and Japan in Panel B exhibits a t-value below the critical value and are hence normally distributed. Non-normal distribution does not cause a problem for the estimation itself, but the estimations tend to be less efficient. The ARCH (12) indicates that the series are affected by heteroskedasticity and cluster volatility which further strengthens the argument to use the GARCH model for the estimations. Nonetheless, for industrial

The influence that a common currency and market conditions have on economic integration : A cross-quantilogram and DCC-EGARCH approach