have the gipsi settled down? breaks and multivariate stochastic volatility models for, and not...
Post on 18-Dec-2016
213 Views
Preview:
TRANSCRIPT
Journal of Banking & Finance 37 (2013) 3639–3653
Contents lists available at SciVerse ScienceDirect
Journal of Banking & Finance
journal homepage: www.elsevier .com/locate / jbf
Have the GIPSI settled down? Breaks and multivariate stochasticvolatility models for, and not against, the European financial integration
0378-4266/$ - see front matter � 2013 Elsevier B.V. All rights reserved.http://dx.doi.org/10.1016/j.jbankfin.2013.04.035
⇑ Corresponding author. Tel.: +44 (0)121 204 3338.E-mail addresses: b.t.gebka@ncl.ac.uk (B. Gebka), m.karoglou@aston.ac.uk (M.
Karoglou).1 Tel.: +44 (0)191 208 1578.
Bartosz Gebka a,1, Michail Karoglou b,⇑a Newcastle University Business School, University of Newcastle upon Tyne, 5 Barrack Road, Newcastle upon Tyne NE1 4SE, United Kingdomb Aston Business School, Aston University, Aston Triangle, Birmingham B4 7ET, United Kingdom
a r t i c l e i n f o a b s t r a c t
Article history:Received 31 October 2012Accepted 22 April 2013Available online 11 May 2013
JEL classification:C22C32C58F36G01G15
Keywords:European integrationEMUFinancial spilloversBreak testsStochastic volatility models
We investigate the integration of the European peripheral financial markets with Germany, France, andthe UK using a combination of tests for structural breaks and return correlations derived from severalmultivariate stochastic volatility models. Our findings suggest that financial integration intensified inanticipation of the Euro, further strengthened by the EMU inception, and amplified in response to the2007/2008 financial crisis. Hence, no evidence is found of decoupling of the equity markets in more trou-bled European countries from the core. Interestingly, the UK, despite staying outside the EMU, is notworse integrated with the GIPSI than Germany or France.
� 2013 Elsevier B.V. All rights reserved.
1. Introduction
As the financial meltdown of 2007/2008 translated into realeconomic problems, two distinctive groups of countries emergedwithin the Eurozone: those able to weather the shocks to theirfinancial systems and those failing or at best struggling to preventthe collapse of their banking systems and national creditworthi-ness without outside help, and which pose an indirect systemicrisk to the former. This latter group consists mainly of Greece, Italy,Portugal, Spain and Ireland, hence the acronym GIPSI, whichendeavours to replace the more pejorative PIIGS that was in com-mon use before mainly amongst financial market practitioners. Ourpaper investigates the time-varying nature of the financial integra-tion as measured by the respective stock-market correlations ofthe core equity markets in Europe with that part of the Europeanperiphery which is at the heart of the recent financial crisis.
Following the inception of the Economic and Monetary Union(EMU) and the introduction of the EURO in 1999 (2001 in Greece),several studies demonstrated a positive impact of the commoncurrency on stock market integration within the Eurozone: correla-tions between equity returns have generally been found to have in-creased (Kim et al., 2005; Baele and Inghelbrecht, 2009; Rua andNunes, 2009; Savva, 2009; Baele and Inghelbrecht, 2010; Büttnerand Hayo, 2011) and their volatility to have declined (Kim et al.,2005; Savva, 2009; Savva et al., 2009). However, it is actually quitefascinating that several key controversies regarding the nature ofthe European financial integration process remain unaddressed,three of which are within the scope of our study.
First, it is not clear that the increased integration was sustain-able; several studies show that some relationships between theEMU countries became stronger immediately after the introductionof the Euro but loosened up in the most recent periods (e.g., Yanget al., 2003; Bley, 2009). Second, the debate is still ongoing as towhether it was the inception of the EMU in 1999 or its anticipatedeffects prior to that date which triggered the European integration;many studies find market comovements to have intensified in themid-1990s already (e.g., Kim et al., 2005; Hardouvelis et al., 2006;
3640 B. Gebka, M. Karoglou / Journal of Banking & Finance 37 (2013) 3639–3653
Savva, 2009; Baele and Inghelbrecht, 2010; Asgharian and Noss-man, 2011), while some authors go even further to suggest thatby the end of that decade most, if not all, of the convergence wascompleted (Hardouvelis et al., 2006) and that EU membershipwas far more important than the EMU (Bekaert et al., 2012). Third,especially relevant to the UK and other EU members remaining out-side the Eurozone, and directly related to the previous point, theverdict is still out as to whether one has to adopt the Euro to benefitfrom increased integration with the EMU countries.2
Moreover, it has recently become quite evident that one cannottalk about the European integration without taking also into ac-count the impact of the financial crisis which started in 2007. Onthe one hand, as economic conditions across the EMU becamemore heterogeneous, one would expect that the national equitymarkets drifted apart, too. On the other hand, stock markets arelargely globalised in terms of the origin of both investors and listedcompanies, as well as with regard to the companies’ markets forboth inputs and outputs. Therefore, the idiosyncratic nationalevents might not weight too much on equity markets’ dynamics.Furthermore, correlations across markets are widely known to behigher in times of crises (e.g., King and Wadhwani, 1990; Lee andKim, 1993; Calvo and Reihart, 1996). As a result, the impact ofthe recent crisis on the integration of equity markets in Europe isquite ambiguous: there might have been a decoupling of the GIPSIfrom the rest of the Eurozone, or correlations across all EMU coun-tries could have increased.
With regard to the ongoing discussions about the ‘‘two-speedEurope’’ and the future of the Euro, the underlying questions ofour paper are: (i) whether or not, in terms of stock market integra-tion, the GIPSI are really the outcasts of the Eurozone; and (ii) ifthey are, when did they actually become as such. Only once an-swers to these questions have been established will the ‘hows’and ‘whys’ that are currently debated in the popular press and aca-demic literature have a solid platform upon which some meaning-ful interpretation vis-à-vis the real-world can be drawn.
To this aim, and in line with the main strand of the literature,we employ the correlations between index returns as an indicatorof stock market integration which in turn is expected to reflect thestate of the more general albeit vague notion of financial integra-tion. The popularity of this indicator can be attributed not onlyto the fact that it is key for the implementation of diversificationdecisions but also because it is intuitively straightforward to ac-cept that as the level of integration rises so will the correlation val-ues (e.g., Bekaert et al., 2009). However, given that changes incorrelations might not signal changes in the intensity of linkagesbetween countries but rather result from changes in either the vol-atility in one of the countries or in the common factor, or even fromthe (unaccounted for) simultaneity in causality (Forbes and Rigo-bon, 2002; Billio and Pelizzon, 2003; Rigobon, 2003; Corsettiet al., 2005), the use of raw correlations could well be severely mis-leading. For that reason, we propose a two-step framework inwhich (i) structural changes or severe albeit latent non-linearitiesthat manifest or, at the very least, can be captured as breaks inthe dynamics of stock market returns and/or their respective vola-tilities are first identified; and (ii) based on the joint subsamplesthat these breaks identify, several multivariate stochastic volatility(MSV) models, that may or may not allow for one-sided volatilityspillovers, are used to obtain estimates of the time-varying corre-lations. In this way, we robustify our results not only against thepresence of country-specific breaks in the mean and/or volatility
2 Yang et al. (2003) and Hardouvelis et al. (2006) find that the UK is left behindwith respect to integration, whereas Kim et al. (2005), Bartram et al. (2007) and Baeleand Inghelbrecht (2010) show in contrast an increased integration of London with theEurozone. On the other hand, Greece does not seem to have been affected by its lateEMU entry (Kim et al., 2005).
dynamics of the underlying returns but also to the possibility ofvolatility spillovers and daily correlation changes.
This paper’s contributions to the literature are as follows. Onmethodological grounds, we propose to combine a battery of testsfor structural breaks with the stochastic volatility (SV) models. Fur-thermore, unlike most of the existing studies, SV rather thanGARCH-type approach is employed to obtain correlations esti-mates. Empirically, we utilize a long sample to assess the timing,magnitude, and sustainability of the impact of Euro on financialintegration in Europe, and how the crisis of 2007–2008 affectedthis process, in an attempt to reconcile the conflicting findings re-ported in the existing literature. We also investigate whether thetroubled Eurozone periphery decoupled from the main financialmarkets as a result of divergence in economic and fiscalperformance.
To preview our main results, we find a positive impact of theEMU on financial market integration between GIPSI and the Euro-pean core, with correlations higher (and increasing) and their vol-atilities lower following the introduction of the EURO. This processof tightening links was initiated but not fully accomplished inanticipation of the benefits of the EMU, and did not fade off inthe latest period. The financial turmoil of 2007/2008 did not causea split between the European core and the GIPSI; rather, correla-tions between markets became stronger and most of their volatil-ities declined following the outbreak of the crisis. Hence, in termsof financial market integration, the GIPSI should not be seen asEuropean outcasts. However, there is substantial heterogeneity interms of financial integration dynamics across both the GIPSI andthe core countries. Lastly, membership in the EMU does not seemto be necessary for strengthening the links with the financial mar-kets of the Eurozone, as increased integration can also be observedfor the UK, a country outside the EMU.
The remainder of this paper is organised as follows. Section 2reviews briefly some literature on the impact of the EMU on finan-cial integration in Europe. Section 3 presents the data and Section 4describes the methodology. Section 5 discusses the results; and fi-nally Section 6 contains our concluding remarks.
2. Literature review
The impact of the EMU on financial market integration in Eur-ope has been, and continues to be, the subject of many studies.In general, measures of correlation in stock index returns betweenthe EMU countries, typically as implied by various conditionallyheteroskedatic (GARCH-type) structures, were shown to have in-creased (e.g., Kim et al., 2005; Baele and Inghelbrecht, 2009; Ruaand Nunes, 2009; Savva, 2009; Baele and Inghelbrecht, 2010; Bütt-ner and Hayo, 2011). Also, the EMU seems to have a ‘calming’ im-pact on the volatility of cross-market linkages, as volatilities incorrelations are reported to have declined in the post-euro era(e.g., Savva, 2009; Savva et al., 2009; Kim et al., 2005; with the lat-ter noting that this effect is most pronounced in Portugal, Ireland,the Netherlands, and Greece). Similar studies based on differentmeasures of integration have more or less painted the same picture(Fratzscher, 2002; Morana and Betratti, 2002; Yang et al., 2003;Bartram et al., 2007; Bley, 2009; Mylonidis and Kollias, 2010).
The question of whether the EMU resulted in increased inte-gration and whether this effect was permanent is however onlyone of the issues that are addressed by the relevant literature.Other issues include (i) the origin of integrating sources; (ii) thetiming of the integration process; and (iii) whether staying out-side the EMU hampers integration with the equity markets ofthe Eurozone.
With respect to (i) Bekaert et al. (2009) find that increasingcorrelations across Europe are more likely due to the increasing
B. Gebka, M. Karoglou / Journal of Banking & Finance 37 (2013) 3639–3653 3641
integration of individual countries with the global equity marketrather than to intra-European integration (also Baele and Inghelbr-echt, 2010). However, Hardouvelis et al. (2006) argue that Euro-pean integration in the run-up period to the EMU was Eurozone-specific and independent from the world-wide integration of finan-cial markets.
With respect to (ii) Hardouvelis et al. (2006) show that integra-tion started for most countries around mid 1990s and that, by thetime the EMU came into being, this process was almost complete.Similar results of intensifying integration prior to the introductionof the Euro are reported by Kim et al. (2005), Savva (2009), Baeleand Inghelbrecht (2010) and Asgharian and Nossman (2011). How-ever, Pascual (2003) does not find evidence of changes in the de-gree of financial integration for the UK and German stockmarkets, but observes the French market to have become moreintegrated with the other two ones.
Finally, with respect to (iii) Hardouvelis et al. (2006) found thatthe UK was the only country which did not experience increasingintegration in the pre-EMU decade while Yang et al. (2003) takethe argument even further and claim that the UK has become lessintegrated with the Eurozone following the introduction of theEuro. In contrast, Bartram et al. (2007) observe that the UK marketexperienced an increase in equity comovements with the Eurozonecountries following the introduction of the Euro and Kim et al.(2005) and Baele and Inghelbrecht (2010) documented an increas-ing integration predating the introduction of Euro for the Europeancountries not joining the EMU as well. In a similar vein, Greece’sintegration was found not to be negatively affected by its late entryto the Eurozone, and Bekaert et al. (2012) suggest that integrationis fostered by membership in the EU, with the Euro having minimaleffect on reduction of differences in value of financial assets acrossparticipating countries. Lastly, Bekaert et al. (2009) state that theincreasing similarities in market movements across Europe shouldbe attributed to integration of European markets with the globalmarket rather than within Europe.
3. Data
We employ daily data for national indices as constructed by theDatastream for the following countries: GIPSI, being Greece, Italy,Portugal, Spain and Ireland, and core being the three major EUequity markets namely France, Germany, and the UK. The dataspans the period 01/01/1990 and 23/11/2010, yielding 5454 obser-vations for each series.3
At this point it is worth mentioning that we have paid particularattention to the selection of the end date of the sample for threereasons. First, because the European debt crisis, that is oftenviewed to have started at the start of 2009, is still on-going.
3 Several studies have argued that contemporaneous correlations between marketscalculated using daily closing prices will be downward biased if those markets havedifferent trading hours, i.e., they do not fully overlap, resulting in asynchronous data(Kahya, 1997; Burns et al., 1998; Martens and Poon, 2001). As we study the Europeanmarkets, this issue is expected to cause minor issues, if any; all but two countries(Greece and the UK) are located in the same time-zone and their opening hoursoverlap extensively. To obtain fully synchronous data, information on intraday priceswould be necessary for all markets (Martens and Poon, 2001; Kleimeier et al., 2008)but such information is not readily available for all markets and the entire time periodinvestigated here. Other approaches for data synchronisation are problematic, asdiscussed in Martens and Poon: firstly, adjustment methods by Riskmetrics™ andBurns et al. (1998) add noise to the data and are sensitive to model specifications anddata types; secondly, using weekly data might reduce the biases but it causes adecrease in the sample size and, hence, in efficiency of estimates and low frequencydata cannot capture daily return dynamics; lastly, studies using open-to-close andclose-to-open returns (e.g., Hamao et al., 1990; Koutmos and Booth, 1995) cannotdistinguish between contemporaneous and lagged interdependencies, and the resultsare reported to be similar to those obtained from close-to-close returns. Hence, weuse daily closing prices in our analysis.
Therefore, modelling the anticipated projection to the stock marketdynamics of the still on-going and changing sovereign crisis wouldinevitably lead to serious misspecification errors that would natu-rally affect our integration measure estimates. Second, because thesovereign crisis started not long after the 2007–2008 global finan-cial crisis. Given that our results need to be robust even when theunderlying economies are in turmoil, unless as mentioned abovesuch periods are still on-going, we had to include in our dataset en-ough recent observations so that our econometric methods canidentify the end of the crisis that is over and the beginning of thecrisis that is still on-going. And third, because of the long observedlead-lag effects that arise every time an economic event manifestsitself in the dynamics of stock returns.
For these reasons, we have selected the end data of our sampleto be exactly seven months after the 23rd of April 2010, when theGreek government requested officially its first loan from the EUand the IMF. Admittedly, the exact timing of the sovereign crisisis rather debatable,4 not the least because it is also highly unlikelythat its impact on EU economies was both instantaneous andsimultaneous.5 This is why we have chosen a quite conservativeapproach and have included enough observations after the ‘official’start of the sovereign crisis. Hence, our sample ends in November2010.
Table 1 presents the basic statistical features of the index re-turns employed. Overall, the underlying markets do not seem todiffer from each other in terms of mean returns, although the GIPSIseem to be more volatile than the core markets. The only exceptionis Portugal which recorded the lowest volatility amongst all mar-kets. The well-documented negative skewness and leptokurtosisare also present in the series, yielding the emblematic non-normal-ity of stock market returns.
4. Methodology
This section describes our approach to robustify the inferencewith respect to how correlations have evolved over time. To thisend, we argue that financial time series will necessarily experienceeither structural changes that manifest themselves as structuralbreaks in the dynamics of the underlying series, or substantial, al-beit latent, nonlinearities, the effect of which for the purposes ofour study can be proxied satisfactorily by the identification ofone or more appropriate breakpoints. Therefore, we hypothesizethat the impact of such phenomena on correlation measurementscan be controlled to a large extend by a combination of a suitablebreak identification method and conditional mean and volatilitydynamics structures.
With the above in mind, we approach the modelling of stockmarket integration in two steps. The first step, as described inthe first part of this section, involves a data-driven procedure forfinding multiple breaks in returns and their underlying volatilities.In this way, we robustify our subsequent modelling and analysis tothe presence of structural changes and/or severe non-linearitieswhile remaining agnostic as to whether the breaks are endoge-nously or exogenously generated.6 Then, the second step drawson the dates of the identified country-specific breaks to generate
4 For example, many argue that it started around January 2009 when Irelandnationalized the Anglo-Irish Bank.
5 For example, while Italy and Spain were struggling but were less likely to defaultor be in need of bailouts, Greece was attempting to accommodate the bailoutconditions into their fiscal policy (in October 2010, the Greek government announcednew, tougher, austerity measures for its 2011 draft budget), Ireland agreed on the EU/IMF bailout and Portuguese parliament passed a new austerity budget in November2010.
6 Regime-switching specifications are typically used for identifying the so-calledendogenous breaks while the official dates of major economic events are typicallyused for identifying the so-called exogenous breaks.
Table 1Data overview.
France Germany Greece Ireland Italy Portugal Spain UK
Mean 0.02% 0.02% 0.03% 0.01% 0.01% 0.01% 0.02% 0.02%Std. Dev. 1.23% 1.19% 1.69% 1.27% 1.31% 0.98% 1.25% 1.06%Skewness �0.1 0.1 0.2 �0.6 �0.1 �0.3 �0.2 �0.2Kurtosis 8.2 14.1 8.9 11.3 7.7 15.1 8.8 10.1Jarque-Bera 6266 28,061 7797 15,946 5093 33,488 7615 11,528
3642 B. Gebka, M. Karoglou / Journal of Banking & Finance 37 (2013) 3639–3653
break-free subperiods for each pair of the countries investigated andto draw inference based on various correlation measures for eachpair and each subperiod. In this way, we robustify our analysis bothto the possibility of daily variations of the correlation measures andto the possibility of volatility spillovers by gradually relaxing themodelling assumptions we make. The second part of this section isdedicated to describing the MSV models that we use to obtain corre-lation estimates beyond the standard correlation statistics. Finally,we test for the equality of the derived correlation values of contigu-ous segments using (i) the t-statistic based on the Fisher-trans-formed correlation coefficients for the correlation means, which isapproximately normally distributed; and (ii) the Levene’s (1960)and the Brown and Forsythe’s (1974) statistics for the correlationvariances.
7 It is worth mentioning that we assume the stock markets we investigate to be atleast weakly informationally efficient at daily frequency – and therefore we impose arestriction that lagged returns have no impact on current returns. Otherwise, dailyreturns would be easily predictable and excess profits systematically achievable,which is hardly a realistic assumption for the established markets studied here. Ourresults would be valid even in the presence of statistically significant lagged effects, aslong as their magnitude is not substantial.
4.1. Identification of breaks
We identify the breaks of each series by adopting the ‘Nominat-ing-Awarding’ two-stage procedure that is explained in detail inKaroglou (2010). The main advantage of the procedure is that itmakes possible to combine the results of two different batteriesof tests in order to robustify the adoption of a specific breakdate.We refer the interested reader to the aforementioned paper formore details. Here, we just list the tests that we have employedin each stage for the respective battery of tests. It is also worthmentioning that for the detection of multiple breaks we haveadopted the algorithm of Karoglou (2010) which enforces theexisting breaks to be detected in a time-orderly fashion. In otherwords, the first break proposed by the algorithm is also the earliestbreak in the series, the second break proposed is the second earli-est break, and so forth. This is particularly important when transi-tional periods exist in which case a simple binary divisionprocedure is likely to produce more breaks in the interim period.
For the ‘Nominating breakdates’ stage, we have used the follow-ing tests: I&T (Inclan and Tiao, 1994); SAC1 (the first test of Sansóet al., 2003); SACBT
2 ; SACQS2 ; SACVH
2 (the second test of Sansó et al.,2003, with the Bartlett and Quadratic Spectral kernel and the New-ey-West (1994) automatic procedure for the bandwidth selection,and with the Vector Autoregressive Heteroskedasticity and Auto-correlation Consistent or VARHAC kernel of den Haan and Levin,1998, which bypasses the bandwidth selection issue, correspond-ingly); K&LBT, K&LQS, K&LVH (the refined by Andreou and Ghysels,2002 version of the Kokoszka and Leipus, 2000, test using theaforementioned kernels). Note that at this stage we are not muchconcerned with detecting more breaks than those that actually ex-ist because whichever is not an actual break-point will be pickedup in the ‘Awarding breakdates’ stage that comes next.
For the ‘Awarding breakdates’ stage, we have used the methodsdesigned to test for the equality of means and variances of differentsamples, which in this case are two contiguous segments. Thesetests constitute a different approach to the tests described in the‘Nominating breakdates’ stage in that they test for the equality ofmeans and variances without encompassing the time-seriesdimension of the data. They include: for the equality of meansthe standard t-test and the Satterthwaite-Welch t-test which is
more robust when the corresponding variances are different, andfor the equality of variances the standard F-test, the Siegel-Tukeytest with continuity correction (Siegel and Tukey, 1960, and She-skin, 1997), the adjusted Bartlett test (see Sokal and Rohlf, 1995,and Judge et al., 1985), the Levene test (1960) and the Brown-Forsythe test (1974).
It is worth noting that there can be unaccounted-for breaks thatare due to the disturbance covariance matrix of the system ofequations (see Bataa et al., forthcoming). As our break detectionapproach is based on time-series characteristics of individual ser-ies, this could lead to segmenting our series too few times, i.e., tooverlooking some breaks which actually took place. The main im-pact of these undetected breaks would be (i) an averaging of theunderlying correlation estimates across non-separated subsam-ples, and (ii) a corresponding increase in correlations’ variability.The effect of (i) would have affected our inference only if the differ-ences between the affected correlation estimates were substantial.However, it is unlikely that such a dramatic change in the covari-ance matrix of the systems of equations would have had only anegligible impact on the volatility dynamics of each series, so asnot to be detected by the break tests adopted here. Hence, wecan be reasonably confident that our approach captures all signif-icant breaks. Furthermore, the effect of (ii) would be reflected inthe time-varying correlation estimates, as depicted in Fig. 2; there,however, we observe that the variability is typically quite small,and that in most cases it declines over time. Consequently, we con-clude that our results are rather unlikely to be biased due to exis-tence of undetected breaks.
4.2. Correlation measures
We obtain correlation measures using the standard correlationcoefficient as a benchmark as well as the MSV models that are de-scribed and compared in Yu and Meyer (2006). In particular, ourresults focus primarily on a Granger-Caused MSV specificationand on a Dynamic Correlation MSV (DC-MSV) specification usingthe Fisher transformation suggestion of Tsay (2002) and Christod-oulakis and Satchell (2002) in the MARCH framework.7
Let at time t, for t = 1, 2 . . . T, the (mean-centred) log-returns bedenoted rt = (r1t, r2t). Then, for et = (e1t, e2t), ht = (h1t, h2t),m = (m1, m2), vt = (v1t, v2t), Xt = diag(exp(ht/2)), the GC-MSV canbe written as:
rt ¼Xt �et; et�iid Nð0;ReÞ;Re¼1 qe
qe 1
� �;
ht ¼mþUðht�1�mÞþvt; U¼u11 u12
u21 u22
� �; vt�iid Nð0;diagðr2
g1;r2
g2ÞÞ;
ð1Þ
B. Gebka, M. Karoglou / Journal of Banking & Finance 37 (2013) 3639–3653 3643
for h0 = m, and u12 = 0 to allow for one-sided only Granger causalityin volatility between the two assets – specifically the volatility ofthe second asset is Granger caused by the volatility of the first asset,thus allowing for volatility spillovers. Consequently, the main fea-ture of this model is that both the returns and volatilities are depen-dent while the conditional correlation is scalar.
Using similar notation, the DC-MSV can be written as:
rt ¼ Xt � et ; et jXt �iid Nð0;Re;tÞ; Re;t ¼1 qt
qt 1
� �;
ht ¼mþ diagð/11;/22Þðht�1 �mÞ þ vt ; vt �iid Nð0;diagðrg1;rg2ÞÞ;
qt ¼ w0 þ wðqt�1 � w0Þ þ rqut; ut �iid Nð0;1Þ; qt ¼expðqtÞ � 1expðqtÞ þ 1
ð2Þ
with h0 = m, q0 = w0. The main feature of this model is that it allowsboth volatilities and correlation coefficients to evolve over time byrestricting volatility spillovers.8,9
5. Results
This section presents our results and is split into seven parts.The first part overviews the outcome of the Nominating-Awardingprocedure. The second, third and fourth parts describe the differ-ences amongst the correlation measures. Finally, the remainingparts discuss what the results mean with respect to stock marketinterdependence and by extension financial integration.
5.1. The identified breaks
Tables 2 and 3 report the results from the Nominating-Award-ing procedure and Table 4 overviews the basic statistical propertiesof each identified segment.
It is rather obvious that the underlying stochastic processesexhibit complex dynamics. For all countries, there is ample evi-dence that at least three breaks are necessary to capture substan-tial changes in the mean and primarily volatility dynamics. It isalso interesting to note that a break around the period of thebankruptcy of the Lehman Brothers is identified in all series, add-ing to the pool of evidence about the magnitude of the economicimpact of the event. Our subsequent analysis is based on thebreak-free subsamples as defined by the breaks identified foreach pair.
5.2. Evidence from correlation estimates
We start the analysis of correlations between the GIPSI and theEuropean core financial markets by eyeballing the time series plotsof correlations for each country pair. Fig. 1 shows raw correlationscomputed for each pair and subperiod separately.
In general, there appears to be a general raise in correlations be-tween the GIPSI and the core markets over time. However, this pro-cess is not monotonic; for instance, the integration of thePortuguese stock market with the core weakened in the mid-2000s, and all GIPSI experienced drops in correlations with the corein the post 2007 era. Also, the (anticipated) introduction of theEuro seems to have had a positive effect on correlations between
8 It is worth noting that the two MSV models are not actually nested in one another– and to our knowledge there is neither a known analytical solution to a nestedspecification, nor a tractable numerical one so far. Hence, we employ both approachesto capture different features of financial integration, i.e., time-varying correlationsand volatility spillovers.
9 It would be particularly interesting to extend our models to include more marketsbut unfortunately the curse of dimensionality for such specifications makes itpractically impossible at present.
the European core and the periphery as most correlations in-creased in the late-1990s.
Fig. 2 presents the time-varying correlations for each pair andsubperiod separately. Overall, the results are in line with those inFig. 1; there is a general trend upwards, with a significant in-crease in the late 1990s and a non-monotonic behaviour through-out the sample period. Furthermore, these time-varyingcorrelations demonstrate the apparent decline in the later subpe-riods of the volatility of correlation coefficients, maybe with theexception of Greece. The same conclusions can be drawn whenallowing for volatility spillovers from the core to the periphery,as shown in Fig. 3.
5.3. Evidence from MSV models with time-varying correlations (DC-MSV)
The test results for the equality of the mean and variance corre-lation values of each segment derived from the MSV model with-out volatility spillovers (Fig. 2) are reported in Tables 5 and 6respectively.
For Greece, all correlation changes with the UK are significant atleast at 10% level, with breaks in each of these countries able totrigger these significant changes in linkages. For correlations withFrance and Germany, fewer changes are significant and they aretriggered almost exclusively by domestic breaks in core countries.Significant correlation changes seem to occur in earlier subperiods,until early-to-mid 2008. For Ireland, correlation breaks tend to beinsignificant, with the significant ones initiated by breaks in eitherof the countries in each pair, or common breaks in case of the UK.Significant correlation changes tend to occur in earlier subperiods,until 24/07/2007. In case of Italy, a similar pattern is observed. ForPortugal, breaks tend to cause more significant changes in correla-tions with France and Germany than vice versa, but there arehardly any significant changes in the links with the UK. For thelinks with Germany significant correlation changes took place untilDecember 2008, and its relationship with France has been time-varying throughout the sample period. On the other hand, the re-cent financial turmoil does not seem to have decoupled the Portu-guese market from the UK one, as the last correlation change wasobserved as far back as 2002. In case of Spain, a similar pattern toItaly and Ireland emerges: initiation of breaks in correlations orig-inates in each of the countries of any pair, and later subperiodstend to have insignificant correlation changes, starting from July2007 for the UK and December 2008 for Germany (links withFrance seem to be more time-varying).
In summary, Spain, Italy and Ireland seem to break-cause asmany changes in correlations with the core countries as the latterones do, whereas changes to correlations with Greece (Portugal)seem to be caused by domestic breaks in the core countries (inPortugal). Latest subperiods (since 2007–2008) tend to show few-er cases of significant changes in correlations than the earlierones, indicating that the individual breaks in price behaviour ofnational indices observed in the most recent period did not trans-late into further decoupling of the GIPSI from the financial core ofEurope. Lastly, for each pair the volatility of correlations (‘S.D.’)changes substantially across all subperiods, with Levene (1960)and Brown and Forsythe (1974) test statistics indicating signifi-cant differences in the correlation volatilities (‘W0’ and ‘W50/W10’, respectively).
Looking across all country pairs, in the most recent subperiodsthe GIPSI country that is most integrated with the core ones ap-pears to be Italy, followed by Spain, Portugal and Ireland in order.Greece is, and has been on average throughout the entire sampleperiod, an outlier in this group, with correlations roughly half thesize of those for Italy.
Table 2Nominated Breakdates.
IT ASC1 ASCBT2 ASCQS
2 ASCVH2
KLBT KLQS KLVH LMT Adopted
France4/1/1997 4/1/1997 7/31/1998 4/1/1997 7/31/1998 7/31/1998 4/1/1997 7/31/1998 4/1/1997 4/1/19971/15/2008 1/15/2008 – 1/15/2008 – – 1/15/2008 – 1/15/2008 7/31/19984/3/2009 4/3/2009 – 4/3/2009 – – 4/3/2009 – 4/3/2009 1/15/20084/27/2010 – – – – – – – – 4/3/20095/28/2010 – – – – – – – – –9/2/2010 – – – – – – – – –
Germany8/27/1991 8/27/1991 8/27/1991 8/27/1991 8/27/1991 8/27/1991 8/27/1991 8/27/1991 8/27/1991 8/27/19917/21/1997 7/21/1997 7/21/1997 7/21/1997 7/21/1997 7/21/1997 7/21/1997 7/21/1997 7/21/1997 7/21/19979/4/2008 9/4/2008 – 9/4/2008 – – 9/4/2008 – 9/4/2008 9/4/20084/3/2009 4/3/2009 – 4/3/2009 – – 4/3/2009 – 4/3/2009 4/3/200912/2/2009 12/2/2009 – 12/2/2009 – – 12/2/2009 – 12/2/2009 12/2/20097/8/2010 – – – – – – – – –
Greece5/29/1991 5/29/1991 5/29/1991 5/29/1991 5/29/1991 5/29/1991 5/29/1991 5/29/1991 5/29/1991 5/29/19916/23/2008 6/23/2008 6/23/2008 6/23/2008 6/23/2008 6/23/2008 6/23/2008 6/23/2008 6/23/2008 6/23/200811/21/2008 11/21/2008 – 11/21/2008 – – 11/21/2008 – 11/21/2008 11/21/200811/20/2009 – – – – – – – – –6/15/2010 – – – – – – – – –
Ireland10/28/1997 10/28/1997 10/28/1997 10/28/1997 10/28/1997 10/28/1997 10/28/1997 10/28/1997 10/28/1997 10/28/19977/24/2007 7/24/2007 7/24/2007 7/24/2007 7/24/2007 7/24/2007 7/24/2007 7/24/2007 7/24/2007 7/24/20075/20/2009 5/20/2009 5/20/2009 5/20/2009 – 5/20/2009 5/20/2009 – 5/20/2009 5/20/20094/23/2010 – – – – – – – – –5/28/2010 – – – – – – – – –
Italy4/8/2003 4/8/2003 4/8/2003 4/8/2003 4/8/2003 4/8/2003 4/8/2003 4/8/2003 4/8/2003 4/8/20031/15/2008 1/15/2008 1/15/2008 1/15/2008 1/15/2008 1/15/2008 1/15/2008 1/15/2008 1/15/2008 1/15/20084/3/2009 4/3/2009 – 4/3/2009 – – 4/3/2009 – 4/3/2009 4/3/20094/27/2010 – – – – – – – – –7/8/2010 – – – – – – – – –
Portugal11/5/2002 11/5/2002 11/5/2002 11/5/2002 11/5/2002 11/5/2002 11/5/2002 11/5/2002 11/5/2002 11/5/20028/8/2007 8/8/2007 8/8/2007 8/8/2007 8/8/2007 8/8/2007 8/8/2007 8/8/2007 8/8/2007 8/8/200712/9/2008 12/9/2008 – 12/9/2008 – – 12/9/2008 – 12/9/2008 12/9/20084/16/2010 – – – – – – – – –5/28/2010 – – – – – – – – –8/12/2010 – – – – – – – – –
Spain4/8/2003 4/8/2003 4/8/2003 4/8/2003 4/8/2003 4/8/2003 4/8/2003 4/8/2003 4/8/2003 4/8/20031/15/2008 1/15/2008 1/15/2008 1/15/2008 1/15/2008 1/15/2008 1/15/2008 1/15/2008 1/15/2008 1/15/200812/9/2008 12/9/2008 – 12/9/2008 – – 12/9/2008 – 12/9/2008 12/9/20084/27/2010 – – – – – – – – –7/8/2010 – – – – – – – – –
UK10/22/1997 10/22/1997 10/22/1997 10/22/1997 10/22/1997 10/22/1997 10/22/1997 10/22/1997 10/22/1997 10/22/19977/24/2007 7/24/2007 7/24/2007 7/24/2007 7/24/2007 7/24/2007 7/24/2007 7/24/2007 7/24/2007 7/24/20074/6/2009 4/6/2009 4/6/2009 4/6/2009 – 4/6/2009 4/6/2009 – 4/6/2009 4/6/2009
Note: the significance level is 5% for values in italics and 1% otherwise.
3644 B. Gebka, M. Karoglou / Journal of Banking & Finance 37 (2013) 3639–3653
Looking onto the periphery from the core’s perspective, Francehas got the least stable relations with the GIPSI markets, with only37.5% of changes in correlations being insignificant, followed bythe UK (40.74%) and Germany (48.72%). This may well indicate thatFrance and the UK are driven by factors idiosyncratic to the pan-European capital market (either purely domestic or global factorswhich affect other European countries to lesser extent) at least toa higher degree than Germany. Germany, on the other hand, is fea-turing the fewest cases of significant changes in correlations withthe GIPSI, i.e. the fewest instances of decoupling and consequentlya more stable relationship with the periphery. However, the aver-age correlation of Germany over all sub-periods and with all theGIPSI markets is lower that the correlations for France and theUK: the European periphery’s integration with the German marketseems to be less intensive albeit more stable than with the remain-ing two core countries.
5.4. Evidence from MSV models with volatility spillovers (GC-MSV)
When examining correlations from the MSV models with vola-tility spillovers, the results are roughly the same (Table 6). ForGreece, its correlations with Germany are largely stable over time,while those with France tend to change after domestic breaks inFrance; the Greece-UK correlations change due to domestic breaksin either country. For Ireland, many correlation breaks remaininsignificant, with the significant ones initiated by breaks in Ire-land for France, in either country for Germany, or common breaksfor the UK. For Italy, breaks in correlations originate in both coun-tries for the pairs Italy-France and Italy-Germany, with those withthe UK initiated only by the UK. Breaks in correlation of Portugalwith other countries are the least frequent for linkages with theUK, and tend to be preceded by domestic breaks on either marketof each pair. Lastly, in case of Spain, breaks in correlations can orig-
Table 3Testing the equality of means and variances of contiguous segments.
Segments (obs in 1st, obs in 2nd) t-Test Satterthwaite-Welch t-test*
F-test Siegel-Tukey Bartlett Levene Brown-Forsythe
France 1 & 2 (1891,348) �1.98** �1.68* 1.62*** 3.66*** 37.96*** 24.54*** 23.88***
France 2 & 3 (348,2467) 1.63 1.73* 1.19** 0.2 4.22** 0.82 0.89France 3 & 4 (2467,318) 2.34** 1.54 3.08*** 8.05*** 235.49*** 121.30*** 121.27***
France 4 & 5 (318,427) �1.92* �1.79* 2.96*** 5.42*** 107.18*** 44.05*** 44.05***
Ireland 1 & 2 (2041,2540) 0.42 0.43 1.50*** 6.31*** 90.60*** 48.70*** 48.91***
Ireland 2 & 3 (2540,476) 3.81*** 2.30** 5.23*** 14.92*** 753.55*** 469.11*** 466.54***
Ireland 3 & 4 (476,394) �1.62 �1.69* 2.60*** 5.48*** 91.88*** 46.54*** 46.23***
Germany 1 & 2 (431,1539) �1.55 �1.13 3.37*** 7.76*** 299.14*** 116.69*** 116.11***
Germany 2 & 3 (1539,2903) 1.24 1.46 3.32*** 14.30*** 616.14*** 265.73*** 263.05***
Germany 3 & 4 (2903,151) 2.23** 1.1 5.09*** 8.33*** 314.55*** 181.85*** 181.30***
Germany 4 & 5 (151,173) �1.47 �1.41 4.34*** 4.65*** 81.68*** 30.08*** 30.04***
Germany 5 & 6 (173,254) 0.49 0.47 1.75*** 4.37*** 16.25*** 17.00*** 17.12***
Greece 1 & 2 (367,4453) 2.56** 1.47 3.90*** 10.19*** 461.49*** 269.10*** 261.13***
Greece 2 & 3 (4453,109) 4.17*** 2.16** 4.06*** 6.32*** 167.02*** 110.67*** 108.43***
Greece 3 & 4 (109,522) �2.28** �1.76* 2.29*** 2.25** 36.28*** 18.74*** 18.12***
Italy 1 & 2 (3461,1245) �0.73 �0.95 3.35*** 16.53*** 539.47*** 269.85*** 270.08***
Italy 2 & 3 (1245,318) 3.49*** 2.08** 9.27*** 12.84*** 831.67*** 355.04*** 351.93***
Italy 3 & 4 (318,427) �1.96** �1.85* 2.45*** 3.44*** 73.92*** 26.66*** 26.33***
Portugal 1 & 2 (3351,1241) �2.46** �3.03*** 2.62*** 6.92*** 356.06*** 86.44*** 86.10***
Portugal 2 & 3 (1241,349) 4.96*** 3.00*** 10.51*** 13.41*** 968.92*** 337.62*** 333.65***
Portugal 3 & 4 (349,510) �2.54** �2.37** 2.10*** 2.41** 58.32*** 16.88*** 16.59***
Spain 1 & 2 (3461,1245) �1.1 �1.35 2.52*** 12.40*** 331.89*** 159.41*** 160.45***
Spain 2 & 3 (1245,235) 2.97*** 1.64 8.11*** 11.15*** 624.08*** 297.86*** 296.39***
Spain 3 & 4 (235,510) �1.26 �1.1 2.11*** 2.99*** 48.02*** 20.37*** 20.38***
UK 1 & 2 (2037,2544) 0.95 0.99 2.12*** 8.19*** 302.19*** 125.32*** 125.27***
UK 2 & 3 (2544,444) 2.01** 1.29 3.83*** 11.10*** 450.56*** 237.65*** 236.00***
UK 3 & 4 (444,426) �1.77* �1.79* 3.06*** 5.88*** 128.08*** 50.62*** 50.12***
* Denotes significance at 1% level, respectively.** Denotes significance at 5% level, respectively.*** Denotes significance at 1% level, respectively.
Table 4Descriptive statistics for the segmented series.
Series Obs. Mean (%) Std. Dev. (%) Skewness Kurtosis Jarque-Bera
France seg. 1 1891 0.02 0.91 �0.24 6.76 1130.3���
France seg. 2 348 0.13 1.16 �0.06 5.54 93.7���
France seg. 3 2467 0.02 1.26 �0.18 5.74 784.7���
France seg. 4 318 �0.18 2.21 0.30 6.24 143.7���
France seg. 5 427 0.07 1.29 0.10 6.47 214.6���
Italy seg. 1 3461 0.01 1.34 �0.14 5.43 862.6���
Italy seg. 2 1245 0.04 0.73 �0.78 5.29 399.1���
Italy seg. 3 318 �0.22 2.24 0.28 6.38 155.1���
Italy seg. 4 427 0.04 1.43 0.12 6.83 262.4���
Greece seg. 1 367 0.26 2.89 0.53 6.44 198.3���
Greece seg. 2 4453 0.03 1.46 �0.06 7.71 4114.4���
Greece seg. 3 109 �0.58 2.95 �0.02 4.27 7.4��
Greece seg. 4 522 �0.06 1.95 0.07 4.02 23���
Ireland seg. 1 2041 0.05 0.90 �0.46 12.31 7445.1���
Ireland seg. 2 2540 0.03 1.11 �0.42 6.66 1496.2���
Ireland seg. 3 476 �0.24 2.53 �0.26 5.40 119.6���
Ireland seg. 4 394 <0.001 1.57 �0.26 4.67 50.6���
Germany seg. 1 431 �0.02 1.26 �0.85 12.88 1803.7���
Germany seg. 2 1539 0.05 0.68 �0.25 5.46 404.7���
Germany seg. 3 2903 0.01 1.25 �0.37 5.75 980.8���
Germany seg. 4 151 �0.24 2.81 1.50 10.57 417.2���
Germany seg. 5 173 0.11 1.35 �0.38 3.25 4.5Germany seg. 6 254 0.05 1.02 �0.36 4.40 26.4���
Portugal seg. 1 3351 0.01 0.92 �0.39 10.77 8517.9���
Portugal seg. 2 1241 0.07 0.57 �0.08 5.95 452.5���
Portugal seg. 3 349 �0.23 1.84 �0.02 9.25 567.9���
Portugal seg. 4 510 0.04 1.27 0.33 9.69 959.2���
Spain seg. 1 3461 0.02 1.24 �0.28 6.12 1452.5���
Spain seg. 2 1245 0.06 0.78 �0.65 5.35 376.4���
Spain seg. 3 235 �0.18 2.23 0.21 5.95 86.8���
Spain seg. 4 510 0.00 1.54 0.41 10.23 1125.5���
UK seg. 1 2037 0.04 0.71 0.25 6.58 1110���
UK seg. 2 2544 0.01 1.03 �0.26 5.65 775.4���
UK seg. 3 444 �0.11 2.02 0.02 6.33 204.6���
UK seg. 4 426 0.09 1.16 �0.11 4.14 23.9���
B. Gebka, M. Karoglou / Journal of Banking & Finance 37 (2013) 3639–3653 3645
Fig. 1. Correlation coefficient for the jointly segmented series.
Fig. 2. Stochastic correlation coefficient for the jointly segmented series (no volatility spillovers).
3646 B. Gebka, M. Karoglou / Journal of Banking & Finance 37 (2013) 3639–3653
inate in either country of any pair. For Germany and the UK, thelatter subperiods tend to show more insignificant correlationchanges, implying more stable linkages between the biggest Euro-pean equity markets and the GIPSI since around 2008. As this re-sult holds regardless of whether the volatility spillovers from thecore to periphery are explicitly modelled or not, it indicates thatbehaviour of correlations over time cannot be contributed to uni-lateral changes in volatility in these core countries. For France, itslinkages with Greece, Spain and Ireland changed significantly aslate as April–May 2009, showing decoupling from the former twoand a return to high integration level with the latter.
In the most recent sub-period, the cross-country pattern of inte-gration between the GIPSI and the core countries remains the sameirrespective of whether volatility spillovers are modelled or not.Specifically, Italy shows highest values of the correlation coefficient,
followed by Spain, Portugal and Ireland. Greece again is, and hasbeen on average throughout the sample period, an outlier in thisgroup, with correlations roughly half the value of those for Italy.
When it comes to the volatility of links between countries,France shows the largest fraction of changes in correlations beingsignificant (only 34.37% insignificant), whereas Germany and theUK tend to have more stable linkages with the GIPSI (46.15% and48.15% of insignificant correlation changes, respectively). Interest-ingly, being outside the Euro-zone does not seem to be a dominantfactor when it comes to the stability of linkages with the GIPSI: themost outstanding core country in terms of the frequency of signif-icant changes in correlations is France, not the UK. When it comesto the average correlation across the whole sample period, againFrance and the UK show higher values of correlations with theGIPSI than Germany. Hence, the integration of the German equity
Fig. 3. Constant correlation coefficient for the jointly segmented series (with volatility spillovers).
B. Gebka, M. Karoglou / Journal of Banking & Finance 37 (2013) 3639–3653 3647
market with the European periphery seems to be the most stablealbeit the least advanced one, whereas the French equity marketenjoys high but volatile correlations and the UK seems to be highlyand consistently integrated with the GIPSI.
5.5. Stability of links: a discussion
The stability of linkages between the GIPSI and Germany couldpoint to reliance on different channels of financial spillovers and,consequently, indicate a different type of integration than that be-tween the GIPSI and the UK and/or France. Specifically, if countriesare linked to each other via real economic factors (foreign trade orFDI), the intensity of these links is likely to be less volatile in theshort run. To a certain extent, financial integration due to cross-border bank lending, especially long term one, is also unlikely tochange frequently in its intensity. These two channels of linkagesbetween countries, the real economic activities and long term banklending, could be the dominant feature of the integration of theGIPSI with Germany, resulting in correlations of equity priceswhich do not change as abruptly and frequently as those betweenthe GIPSI and France or the UK. On the other hand, if the relation-ships between the periphery and the UK or France are primarilygoverned by shock transmission channels related to the capitalmarkets rather than economic or long-term banking activities(liquidity shocks, international investors’ herding, etc.), this wouldlead to a more volatile behaviour of correlations across time, i.e.similar to the pattern we observe.
Apparently, the importance of real economic and banking link-ages for Germany could explain why this country was eager topush for further fiscal integration within the EU: as tougher criteriaon public finances are expected to improve the soundness of theperiphery’s national budgets and their economic performance,hence moderating the magnitude and frequency of shocks originat-ing in the GIPSI markets and which can potentially affect the Ger-man financial market as well as the banking sector and in turn thewhole economy. In contrast, shocks originating in the peripheryand transmitted to the remaining core countries may mostly affecttheir equity markets but not so much the real economies or thecredit policy of their banks.
5.6. The impact of the EMU and the 2007 crisis
Next, we investigate whether the stability of integration, as cap-tured by the volatility of correlation coefficients, changed due tothe EMU and the 2007 crisis. The test results are reported inTable 7. When analysing the impact of the EMU, we use two alter-native dates, denoted 1 and 2, for when the Euro was introduced inthe initial eleven countries (on 01/01/1999) and in Greece (on 01/01/2001). Accordingly, in testing the impact of the 2007 crisis onEuropean integration, the pre-crisis (but post-Euro) period startsin 1999 or 2001.
After the introduction of the Euro, evidence from a longer post-euro sample than those utilized by previous studies shows that themean correlations indeed changed and remained significantlyhigher than in the pre-euro era, for all pairs of countries consid-ered. This is indicated by positive and significant values of the teststatistics for the differences in correlations (‘Stat’). Hence, therewas a significant raise in integration of the GIPSI with the coreEuropean equity markets around the time the EMU came into exis-tence. This result also holds for the UK, a country outside theEurozone. As for the volatility of correlations (‘SD CORR’), for Italy,Spain and Portugal they are significantly lower in the post-euroera, as indicated by the significant values of the statistics (W0,W10 and W50). For Greece and Ireland, they are lower only withrespect to their linkages with Germany. In contrast, following theinception of EMU the volatility of correlations between Greeceand Ireland on the one hand, and France and the UK on the other,actually increased. This corresponds to the finding reported aboveof more stable integration of Germany, as compared to France andthe UK, with the GIPSI. All these findings are virtually unaffected bythe date we select to denote the inception of EMU, 1999 or 2001.This suggests that each of these effects was not limited to the1999–2001 period but rather continued to hold after 2001.
To assess whether the most recent trends in correlationchanges, both in correlation levels and volatilities, are driven bythe recent financial crisis, we test for changes in correlation meansand volatilities between the post-euro-pre-crisis period and thepost-euro-post-crisis periods. The timing of the crisis outbreak ischosen to be 1st August, 2007. The evidence (Table 7) shows thatthe integration of the core equity markets with the GIPSI was an
Table 5Results from MSV models without volatility spillovers (DC-MSV) from the core to the periphery.
Note: ‘Obs’ denotes the number of observations in each period, ‘Break from’ indicates the country in which the break in return time series took place. ‘Mean’ stands for the mean correlation coefficient in a given subsample, and‘Stats’ denotes the test statistic for the equality of correlation coefficients, which is a t-statistic based on Fisher-transformed correlation coefficients and approximately follows normal distribution, with ‘P-value’ being thecorresponding p-value. ‘S.D.’ stands for the standard deviation of time-varying correlation coefficients. ‘W0’ denotes the value of the Levene’s (1960) statistic for the equality of variances of correlation coefficients which is robustunder non-normality, and ‘W50’ and ‘W10’ stand for the Brown and Forsythe’s (1974) statistics which replace the mean in Levene’s formula with alternative location estimators: the median (W50) and the 10% trimmed mean(W10). All three statistics are significant at 1%.
3648B.G
ebka,M.K
aroglou/Journal
ofBanking
&Finance
37(2013)
3639–3653
Table 6Results from MSV models with volatility spillovers (GC-MSV) from the core to the periphery.
Pair Period Obs Break from Mean Stat P-value
Pair Period Obs Break from Mean Stat P-value
Pair Period Obs Break from Mean Stat P-value
France–Greece
02/01/1990–29/05/1991
367 0.189 Germany–Greece
02/01/1990–29/05/1991
367 0.247 UK–Greece
02/01/1990–29/05/1991
367 0.179
30/05/1991–01/04/1997
1524 Greece 0.088 �1.776 0.076 30/05/1991–27/08/1991
64 Greece 0.204 �0.334 0.739 30/05/1991–22/10/1997
1670 Greece �0.916 �30.279 <0.1%
02/04/1997–31/07/1998
348 France 0.262 3.032 0.002 28/08/1991–21/07/1997
1539 Germany 0.053 �1.207 0.227 23/10/1997–24/07/2007
2544 UK 0.387 62.642 <0.1%
03/08/1998–15/01/2008
2467 France 0.488 4.646 <0.1% 22/07/1997–23/06/2008
2850 Germany 0.408 12.020 <0.1% 25/07/2007–23/06/2008
239 UK 0.676 6.123 <0.1%
16/01/2008–23/06/2008
114 France 0.693 3.338 0.001 24/06/2008–04/09/2008
53 Greece 0.616 2.055 0.040 24/06/2008–21/11/2008
109 Greece 0.800 2.398 0.017
24/06/2008–21/11/2008
109 Greece 0.746 0.826 0.409 05/09/2008–21/11/2008
56 Germany 0.648 0.278 0.781 24/11/2008–06/04/2009
96 Greece 0.656 �2.240 0.025
24/11/2008–03/04/2009
95 Greece 0.700 �0.697 0.486 24/11/2008–03/04/2009
95 Greece 0.596 �0.501 0.617 07/04/2009–23/11/2010
426 UK 0.547 �1.525 0.127
06/04/2009–23/11/2010
427 France 0.565 �2.001 0.045 06/04/2009–02/12/2009
173 Germany 0.493 �1.146 0.252
03/12/2009–23/11/2010
254 Germany 0.507 0.189 0.850
France–Ireland
02/01/1990–01/04/1997
1891 0.439 Germany–Ireland
02/01/1990–27/08/1991
431 0.337 UK–Ireland
02/01/1990–22/10/1997
2037 0.435
02/04/1997–28/10/1997
150 France 0.446 0.104 0.917 28/08/1991–21/07/1997
1539 Germany 0.424 1.871 0.061 29/10/1997–24/07/2007
2541 both 0.600 7.630 <0.1%
29/10/1997–31/07/1998
198 Ireland 0.539 1.143 0.253 22/07/1997–28/10/1997
71 Germany 0.943 10.813 <0.1% 25/07/2007–06/04/2009
444 both 0.774 6.542 <0.1%
03/08/1998–24/07/2007
2342 France 0.569 0.582 0.560 29/10/1997–24/07/2007
2540 Ireland 0.538 �9.670 <0.1% 07/04/2009–20/50/2009
32 UK 0.705 �0.834 0.404
25/07/2007–15/01/2008
125 Ireland 0.787 4.542 <0.1% 25/07/2007–04/09/2008
292 Ireland 0.729 5.255 <0.1% 21/05/2009–23/11/2010
394 UK 0.793 1.094 0.274
16/01/2008–03/04/2009
318 France 0.787 0.000 1.000 05/09/2008–03/04/2009
151 Germany 0.582 �2.602 0.009
06/04/2009–20/05/2009
33 France 0.660 �1.474 0.140 06/04/2009–20/05/2009
33 Germany 0.619 0.300 0.764
21/05/2009–23/11/2010
394 Ireland 0.806 1.779 0.075 21/05/2009–02/12/2009
140 Ireland 0.744 1.223 0.221
03/12/2009–23/11/2010
254 Germany 0.749 0.101 0.920
France–Italy
02/01/1990–01/04/1997
1891 0.297 Germany–Italy
02/01/1990–27/08/1991
431 0.462 UK–Italy
02/01/1990–22/10/1997
2037 0.408
02/04/1997–31/07/1998
348 France 0.719 10.274 <0.1% 28/28/1991–21/07/1997
1539 Germany 0.295 �3.584 <0.1% 23/10/1997–08/04/2003
1424 UK 0.824 21.249 <0.1%
03/08/1998–08/04/2003
1222 France 0.933 12.824 <0.1% 22/07/1997–08/04/2003
1491 Germany 0.798 21.673 <0.1% 09/04/2003–24/07/2007
1120 Italy 0.829 0.446 0.655
09/04/2003–16/01/2008
1246 Italy 0.891 �6.436 <0.1% 09/04/2003–15/05/2008
1245 Italy 0.839 3.306 0.001 25/07/2007–15/01/2008
125 UK 0.915 3.919 <0.1%
17/01/2008–03/04/2009
319 both 0.938 4.666 <0.1% 16/01/2008–04/09/2008
167 Italy 0.876 1.695 0.090 16/01/2008–03/04/2009
318 Italy 0.895 �1.032 0.302
06/04/2009–23/11/2010
427 both 0.945 0.891 0.373 05/09/2008–06/04/2009
152 Germany 0.861 �0.542 0.588 06/04/2009–23/11/2010
426 both 0.895 �0.020 0.984
07/04/2009–02/12/2009
173 both 0.905 1.818 0.069
03/12/2009–23/11/2010
254 Germany 0.892 �0.686 0.493
France–Portugal
02/01/1990–01/04/1997
1891 0.166 Germany–Portugal
02/01/1990–27/08/1991
431 0.148 UK–Portugal
02/01/1990–22/10/1997
2037 0.157
02/04/1997–31/07/1998
348 France 0.575 8.369 <0.1% 28/08/1991–21/07/1997
1539 Germany 0.214 1.240 0.215 23/10/1997–05/11/2002
1314 UK 0.578 14.135 <0.1%
03/08/1998–05/11/2002
1112 France 0.677 2.750 0.006 22/07/1997–05/11/2002
1381 Germany 0.639 14.541 <0.1% 06/11/2002–24/07/2007
1230 Portugal 0.698 5.141 <0.1%
06/11/2002–08/08/2007
1241 Portugal 0.576 �4.068 <0.1% 06/11/2002–08/08/2007
1241 Portugal 0.501 �5.252 <0.1% 25/07/2007–08/08/2007
11 UK 0.729 0.208 0.835
09/08/2007–15/01/2008
114 Portugal 0.732 2.832 0.005 09/08/2007–04/09/2008
281 Portugal 0.720 5.408 <0.1% 09/08/2007–09/12/2008
349 Portugal 0.800 0.565 0.572
16/01/2008–09/12/2008
235 France 0.834 2.334 0.020 05/09/2008–09/12/2008
68 Germany 0.767 0.780 0.436 10/12/2008–06/04/2009
84 Portugal 0.747 �1.086 0.277
(continued on next page)
B.Gebka,M
.Karoglou
/Journalof
Banking&
Finance37
(2013)3639–
36533649
Tabl
e6
(con
tinu
ed)
Pair
Peri
odO
bsB
reak
from
Mea
nSt
atP- va
lue
Pair
Peri
odO
bsB
reak
from
Mea
nSt
atP- va
lue
Pair
Peri
odO
bsB
reak
from
Mea
nSt
atP- va
lue
10/1
2/20
08–
03/0
4/20
0983
Port
uga
l0.
757
�1.
644
0.10
010
/12/
2008
–03
/04/
2009
83Po
rtu
gal
0.65
1�
1.44
60.
148
07/0
4/20
09–
23/1
1/20
1042
6U
K0.
792
0.93
00.
353
06/0
4/20
09–
23/1
1/20
1042
7Fr
ance
0.82
21.
433
0.15
206
/04/
2009
–02
/12/
2009
173
Ger
man
y0.
778
1.97
40.
048
03/1
2/20
09–
23/1
1/20
1025
4G
erm
any
0.74
7�
0.74
10.
459
Fran
ce–
Spai
n02
/01/
1990
–01
/04/
1991
1891
0.62
9G
erm
any–
Spai
n02
/01/
1990
–27
/08/
1991
431
0.53
4U
K–
Spai
n02
/01/
1990
–22
/10/
1997
2037
0.54
1
02/0
4/19
91–
31/0
7/19
9834
8Fr
ance
0.74
13.
662
<0.1
%28
/08/
1991
–21
/07/
1997
1539
Ger
man
y0.
408
�2.
980
0.00
323
/10/
1997
–08
/04/
2003
1424
UK
0.84
518
.305
<0.1
%
03/0
8/19
98–
08/0
4/20
0312
22Fr
ance
0.83
84.
274
<0.1
%22
/07/
1997
–08
/04/
2003
1491
Ger
man
y0.
777
16.5
97<0
.1%
09/0
4/20
03–
24/0
7/20
0711
20Sp
ain
0.81
2�
2.66
80.
008
09/0
4/20
03–
16/0
1/20
0812
46Sp
ain
0.88
74.
776
<0.1
%09
/04/
2003
–15
/01/
2008
1245
Spai
n0.
837
4.52
6<0
.1%
25/0
7/20
07–
15/0
1/20
0812
5U
K0.
889
3.03
00.
002
17/0
1/20
08–
09/1
2/20
0823
5bo
th0.
938
4.44
0<0
.1%
16/0
1/20
08–
04/0
9/20
0816
7Sp
ain
0.89
72.
999
0.00
316
/01/
2008
–09
/12/
2008
235
Spai
n0.
896
0.30
20.
763
10/1
2/20
08–
03/0
4/20
0983
Spai
n0.
932
�0.
350
0.72
605
/09/
2008
–09
/12/
2008
68G
erm
any
0.80
1�
2.48
10.
013
10/1
2/20
08–
06/0
4/20
0984
Spai
n0.
857
�1.
333
0.18
2
06/0
4/20
09–
23/1
1/20
1042
7Fr
ance
0.89
6�
1.89
20.
059
10/1
2/20
08–
03/0
4/20
0983
Spai
n0.
848
0.89
50.
371
07/0
4/20
09–
23/1
1/20
1042
6U
K0.
864
0.21
30.
832
06/0
4/20
09–
02/1
2/20
0917
3G
erm
any
0.86
90.
619
0.53
6
03/1
2/20
09–
23/1
1/20
1025
4G
erm
any
0.82
5�
1.58
80.
112
Not
e:‘O
bs’d
enot
esth
en
um
ber
ofob
serv
atio
ns
inea
chpe
riod
,‘B
reak
from
’in
dica
tes
the
cou
ntr
yin
wh
ich
the
brea
kin
retu
rnti
me
seri
esto
okpl
ace.
‘Mea
n’s
tan
dsfo
rth
em
ean
corr
elat
ion
coef
fici
ent
ina
give
nsu
bsam
ple,
and
‘Sta
ts’
den
otes
the
test
stat
isti
cfo
rth
eeq
ual
ity
ofco
rrel
atio
nco
effi
cien
ts,
wh
ich
isa
t-st
atis
tic
base
don
Fish
er-t
ran
sfor
med
corr
elat
ion
coef
fici
ents
and
appr
oxim
atel
yfo
llow
sn
orm
aldi
stri
buti
on,
wit
h‘P
-val
ue’
bein
gth
eco
rres
pon
din
gp-
valu
e.
3650 B. Gebka, M. Karoglou / Journal of Banking & Finance 37 (2013) 3639–3653
ongoing process after the introduction of the euro, and becomestronger after the crisis broke out in 2007, compared the post-europeriod of early 2000s. All correlations are found to be statisticallyhigher in the post-crisis period regardless of which pair and yearfor the EMU is considered. The volatility of pair-wise correlationsseems to have been lower in the post-crisis era for the GIPSI, withthe exception of Portugal which experienced higher volatility incorrelations with the core European equity markets from August2007 onwards. The decline in correlation volatilities throughoutthe 2000s, despite the outbreak of a financial crisis affecting differ-ent countries in different ways and to different degrees, highlightsthe ongoing integration of the periphery with the core of theEuropean equity markets.
5.7. European or global integration?
To assess whether the comovements between European mar-kets, and their changes over time, are not entirely due to therespective market’s integration with the global stock market (assuggested by Bekaert et al., 2009) but rather due to the intra-Euro-pean integration, we compare on the one hand the evolution ofcorrelations between European countries and on the other the evo-lution of their individual correlations with the global market (asimilar approach was adopted by Savva and Aslanidis, 2010),which we proxy by the S&P500 and the MCSI WORLD Index. Theresults are not reported in detail to conserve space, but the mainfindings are as follows. Firstly, correlations with the global marketare systematically lower than those with other European markets,implying that a substantial part of intra-European comovements isnot due to the dependence of the European markets on the globalone, but rather to financial integration within Europe. Secondly,changes in correlations are higher for those measured for Europeanpairs than those featuring the global market, which shows that theprocess of increasing European integration was genuine and not anartefact of increasing European dependence on the global market.This holds true even though the intra-European correlations werehigher than those with the world from the start of our sample;hence in principle there was less scope for them to rise over time.All these results hold regardless of whether the model allows forvolatility spillovers or not. Lastly, the volatility of correlations withthe global market, especially if the latter is proxied by the bench-mark US series, seems to have increased over time, which is theopposite of our finding of decreased volatility for intra-Europeanpairs. Hence, increasing intra-European integration, as evident inlower volatilities of market comovements, occurred despite, andnot due to, the volatile nature of the impact of the world stock mar-ket on Europe.
6. Summary and concluding remarks
In general, there are five conclusions that our results support.First, the evidence presented in this paper supports the notion
that the EMU has had a positive effect on intra-European financialintegration as proxied by the stock market comovements, sincepairwise correlations between the peripheral and core equity mar-kets are significantly higher and the volatilities of correlations arelower in the post-Euro era. This finding supports the views of Kimet al. (2005), Baele and Inghelbrecht (2009), Rua and Nunes (2009),Savva (2009), Baele and Inghelbrecht (2010), Büttner and Hayo(2011), for mean correlations and Kim et al. (2005), Savva (2009)and Savva et al. (2009), for volatilities.
Second, we find that this effect was not short-lived, as sug-gested by Yang et al. (2003) and primarily by Bley (2009), butwas sustained and intensified over time, albeit not in a monotonicfashion: when analysing the post-Euro era, the subperiod starting
Table 7Changes in correlations due to EMU and the financial crisis of 2007-8.
Period Observations Mean (%) Stat S.D. (%) W0 W50 W10 Observations Mean (%) Stat S.D. (%) W0 W50 W10 Observations Mean (%) Stat S.D. (%) W0 W50 W10
France–Greece Germany–Greece UK–GreeceEMU 1 (before, after) (2348, 3103) (14.3, 47.1) 13.4 (12.9, 17.3) 250.8 355.1 325.4 (2348, 3103) (14.5, 44.5) 12.1 (17.8, 16.2) 41.4 17.8 28 (2348, 3103) (13.8, 46.6) 13.4 (13, 15.8) 169.8 324.5 301.9EMU 2 (before, after) (2869, 2582) (16, 51.8) 15.2 (13.1, 14.1) 5.51⁄⁄ 37.9 18.8 (2869, 2582) (15.6, 49.3) 14.1 (17.2, 12) 473.5 333.1 416.4 (2869, 2582) (16.1, 50.6) 14.6 (13.2, 13.7) 1.18X 19.7 3.40⁄
crisis 1 (before, after) (2238, 865) (40.6, 63.8) 8.1 (15.3, 8.9) 223.4 180.4 211.9 (2238, 865) (39.8, 56.5) 5.5 (16.1, 8.6) 350.1 315.2 328.5 (2238, 865) (40.2, 62.9) 7.8 (12.5, 11.1) 2.57X 1.79X 2.45X
crisis 2 (before, after) (1717, 865) (45.7, 63.8) 6.2 (12.3, 8.9) 30.8 29.8 30.1 (1717, 865) (45.6, 56.5) 3.5 (11.8, 8.6) 104.2 98.2 101.9 (1717, 865) (44.4, 62.9) 6.3 (10.4, 11.1) 48.7 45.9 47.9
France–Ireland Germany–Ireland UK–IrelandEMU 1 (before, after) (2348, 3103) (42.6, 62.8) 10.3 (9.7, 16.4) 987.3 498.3 793.6 (2348, 3103) (42.9, 58.4) 7.7 (14.8, 14.5) 0.50X 1.11X 0.96X (2348, 3103) (45.5, 64.3) 10 (7.1, 12.5) 970.8 761 873.7EMU 2 (before, after) (2869, 2582) (41.5, 68.1) 14.4 (9.7, 11.8) 122.5 58.1 80.6 (2869, 2582) (42.4, 62.2) 10.2 (15.1, 10.8) 132.1 122.4 124.1 (2869, 2582) (45.7, 67.8) 12.2 (7.4, 10) 297.7 236 259.8crisis 1 (before, after) (2238, 865) (56.9, 77.9) 9.9 (15.7, 2.9) 1955.8 1509.9 1824 (2238, 865) (53.4, 71.4) 7.5 (14, 4.2) 531.7 521.8 523.3 (2238, 865) (59.9, 75.8) 7.5 (12, 3.4) 1016.7 978.8 1003.7crisis 2 (before, after) (1717, 865) (63.2, 77.9) 7.2 (11.6, 2.9) 1181.6 786.9 1045.3 (1717, 865) (57.5, 71.4) 5.7 (10.2, 4.2) 676.2 693.1 695.4 (1717, 865) (63.9, 75.8) 5.6 (9.8, 3.4) 767.6 733 757.3
France–Italy Germany–Italy UK–ItalyEMU 1 (before, after) (2348, 3103) (50.1, 88.3) 30.6 (18.2, 6) 2244 2234 2273.3 (2348, 3103) (37.9, 82.7) 28.5 (21.1, 8.6) 1185.5 927 1080 (2348, 3103) (45.2, 81.3) 23.7 (16.3, 9.1) 514.5 501.4 511EMU 2 (before, after) (2869, 2582) (55.2, 90.4) 32.1 (19.8, 3.6) 4147.6 4116.9 4189.9 (2869, 2582) (43.7, 85.2) 29.3 (23.2, 5.3) 3501.5 2380.5 3390.7 (2869, 2582) (49.3, 84) 25.1 (17.2, 6.9) 1683.2 1588.8 1705.9crisis 1 (before, after) (2238, 865) (86.3, 93.6) 10 (5.7, 2.2) 468.1 314.7 368.5 (2238, 865) (80.7, 87.7) 6.1 (8.8, 5.7) 231.6 182.1 197.6 (2238, 865) (77.8, 90.2) 11.1 (8.3, 2.5) 949.3 808.2 917crisis 2 (before, after) (1717, 865) (88.7, 93.6) 7 (3.1, 2.2) 94.9 91.1 91.3 (1717, 865) (84, 87.7) 3.4 (4.6, 5.7) 33.4 45.9 45.2 (1717, 865) (80.9, 90.2) 8.7 (6.2, 2.5) 503.3 540.5 530.7
France–Portugal Germany–Portugal UK–PortugalEMU 1 (before, after) (2348, 3103) (24.2, 65.9) 19.9 (20.3, 10.1) 879.2 431.4 597.8 (2348, 3103) (26.6, 60.6) 15.7 (19.6, 9.7) 676.7 506.7 598.4 (2348, 3103) (21.3, 61.3) 18.1 (22, 11) 1367.1 1082.5 1309.3EMU 2 (before, after) (2869, 2582) (31.8, 65.9) 17 (24.4, 11) 3062.2 934.1 2648.5 (2869, 2582) (32.5, 60.9) 13.6 (22, 9.9) 1724.5 1037.4 1657.5 (2869, 2582) (27.8, 62.2) 16.3 (24.1, 11.8) 2457.2 1712.8 2507.8crisis 1 (before, after) (2238, 865) (60.7, 79.4) 9.4 (5.4, 6.2) 2.45X 0.67X 1.15X (2238, 865) (56.4, 71.6) 6.5 (6.9, 6.9) 2.94⁄ 0.04X 0.58X (2238, 865) (54.8, 77.8) 10.6 (3.4, 4.9) 143.4 141.4 145.1crisis 2 (before, after) (1717, 865) (59.1, 79.4) 9.6 (5.2, 6.2) 24.7 21.6 21.6 (1717, 865) (55.5, 71.6) 6.5 (6.1, 6.9) 11.8 23 20.3 (1717, 865) (54.3, 77.8) 10.4 (3.4, 4.9) 137 137.1 141
France–Spain Germany–Spain UK–SpainEMU 1 (before, after) (2348, 3103) (60.5, 86.6) 22.5 (14.4, 5.9) 1647.7 1453.7 1602.3 (2348, 3103) (47.3, 81.6) 23 (13.5, 7.3) 714.7 452.6 564.1 (2348, 3103) (56.6, 79.5) 16.3 (9.7, 6.4) 356.3 325 357.8EMU 2 (before, after) (2869, 2582) (63.5, 88.5) 24 (14.6, 4) 2524.6 2157.1 2319 (2869, 2582) (51.7, 83.6) 23.4 (15.8, 5.1) 3515.8 1631.6 3050.5 (2869, 2582) (59.4, 81.1) 16.5 (10.6, 5.8) 882.2 867.6 872.3crisis 1 (before, after) (2238, 865) (85.1, 90.5) 6 (6, 3.5) 258.9 120.5 182.6 (2238, 865) (80, 85.5) 4.4 (7.3, 5.5) 179.5 106.3 129.2 (2238, 865) (76.9, 86.3) 7.1 (5.3, 2.9) 626.2 430.3 599.8crisis 2 (before, after) (1717, 865) (87.6, 90.5) 3.4 (3.9, 3.5) 1.62X 1.91X 4.23⁄⁄ (1717, 865) (82.6, 85.5) 2.41⁄⁄ (4.6, 5.5) 7 4.50⁄⁄ 4.40⁄⁄ (1717, 865) (78.5, 86.3) 5.9 (5.1, 2.9) 438.4 400.6 431.9
Note: ‘Observations’ denotes the number of observations in each period, ‘Mean’ stands for the mean correlation coefficient in a given subsample, and ‘Stats’ denotes the test statistic for the equality of correlation coefficients, whichis a t-statistic based on fisher-transformed correlation coefficients and approximately follows normal distribution. ‘S.D.’ stands for the standard deviation of time-varying correlation coefficients. ‘W0’ denotes the value of theLevene’s (1960) statistic for the equality of variances of correlation coefficients which is robust under non-normality, and ‘W50’ and ‘W10’ stand for the Brown and Forsythe’s (1974) statistics which replace the mean in Levene’sformula with alternative location estimators: the median (W50) and the 10% trimmed mean (W10). All statistics are significant at 1% level unless stated otherwise: ⁄⁄ stands for 5% and ⁄ for 10% significance level, X denotes lack ofsignificance (p-values higher than 10%). ‘Before EMU 1’/’After EMU 1’ (‘Before EMU 2’/’After EMU 2’) denote subperiods prior to and after 01/01/1999 (01/01/2001). ‘Before crisis 1’/’After crisis 1’ (‘Before crisis 2’/’After crisis 2’)denote periods prior to and after 08/01/2007, starting in the post EMU period 1, i.e., 01/01/1999 (post EMU period 2, i.e., 01/01/2001).
B.Gebka,M
.Karoglou
/Journalof
Banking&
Finance37
(2013)3639–
36533651
3652 B. Gebka, M. Karoglou / Journal of Banking & Finance 37 (2013) 3639–3653
at the outbreak of the financial crisis in 2007 shows higher, notlower, levels in integration, with higher correlations and lower vol-atilities for most country pairs considered. These advances in inte-gration evident after the EMU’s inception do not support the notionby Hardouvelis et al. (2006) that the integration process mainlytook place before and was mostly completed by 1999, driven bythe anticipated effects of the EMU.
Third, we find that the Euro could not have been the onlydriver of financial integration in Europe. This contributes to theliterature on another intensively debated issue, i.e., whether theintegration of equity markets in Europe is driven by the EMU,the effects of EU membership other than the Euro (Bekaertet al., 2012), or simply the strengthening of linkages betweenthe European markets and the global market (Bekaert et al.,2009; Baele and Inghelbrecht, 2010). Our findings also show thatthe integration of the UK, a core country outside the EMU, withthe GIPSI was not noticeably different from that of the Germanor French equity markets.
Although this study, due to the limited cross-sectional dimen-sion of our sample, cannot empirically identify the forces whichhave driven integration and comovements of financial markets ofcore and peripheral Europe, a related literature on financial conta-gion and spillovers offers valuable insights into determinants ofmarkets’ co-dependence. For instance, Forbes (2012) points outthat linkages between countries through trade, banks, and portfo-lio investment, as well as reassessment of macroeconomic funda-mentals by investors, will all contribute to stronger linksbetween financial markets. Indeed, empirical evidence for the Euroarea in Forbes (2012) shows all these channels to be at work. Beka-ert et al. (2012) additionally observe the FDI flows, financial open-ness, financial development, and the EU-wide harmonization offinancial regulations to have influenced the integration amongfinancial markets in Europe. Hence, we would expect changes inthese variables over time to explain the overall upward trend inthe European integration, and cross-country differences in thesevariables to explain the observed heterogeneity across both GIPSIand the core countries.
Fourth, with respect to whether the GIPSI deserve their reputa-tion of the outcasts, as far as the equity markets are concerned ourfindings are relatively mixed. On the one hand, our results showthat their integration into the European core has been gettingstronger over the last two decades, a process strengthened whenthe Euro was introduced and further intensified following the out-break of the global financial crisis of 2007–2008. Also, the last timetheir correlations with the core changed significantly (around2007–2008), it was an increase in correlations for all countriesbut Portugal, suggesting an even tighter integration. Therefore,one could claim that the peripheral equity markets of Europe, asa group, do not deserve their ‘second-class citizen’ label.
On the other hand, one could argue that the GIPSI are not ahomogenous group: for instance, Italy has been showing levels ofcorrelations with the core double the size of the Greek ones. Inaddition, significant changes in correlations of Greece with the coremarkets have always been triggered by structural breaks on thelatter, whereas the remaining GIPSI countries are able to triggersignificant changes in integration levels with the core. Some resultsfor Portugal show that this market is also distinctive. Hence, someGIPSI are more decoupled from the European core than others.
However, it is also worth noting that ‘the core’ is not a homog-enous group, either: German equity market’s integration with theperiphery seems to be distinct, as it’s correlations with GIPSI arelower but more stable over time than those for France or the UK.Therefore, whether a peripheral equity market is to be seen as atroublemaker may depend on both the country in question andthe perspective from which it is viewed (London, Paris orFrankfurt).
Finally, from the point of view of an international investor,increasing correlations imply lower potential benefits of cross-country diversification, and correlations increasing in times of cri-ses means even more bad news, as diversification works worstwhen its needed most. However, our finding of declining volatilityin correlations, especially in periods of financial turmoil, can beviewed as good news for those who hold internationally diversifiedportfolios; it implies that the possibility of experiencing a very highcorrelation (which would wipe out all diversification benefits) isreduced, which, in turn, reduces the riskiness of holding an inter-nationally diversified portfolio.
Acknowledgements
We thank Jun Du, Haris Kotoulas and Kostas Mouratidis fortheir useful comments on earlier drafts of this paper as well asseminar participants of Newcastle University, seminar participantsof Aston University, and participants of the 7th ASSEE. Naturally,all remaining errors and omissions are our own.
References
Andreou, E., Ghysels, E., 2002. Multiple breaks in financial market volatilitydynamics. Journal of Applied Econometrics 17, 579–600.
Asgharian, H., Nossman, M., 2011. Risk contagion among international stockmarkets. Journal of International Money and Finance 30, 22–38.
Baele, L., Inghelbrecht, K., 2009. Time-varying integration and internationaldiversification strategies. Journal of Empirical Finance 16, 368–387.
Baele, L., Inghelbrecht, K., 2010. Time-varying integration, interdependence andcontagion. Journal of International Money and Finance 29, 791–818.
Bartram, S.M., Taylor, S.J., Wang, Y.-H., 2007. The euro and European financialmarket dependence. Journal of Banking and Finance 31, 1461–1481.
Bataa, E., Osborn, D.R., Sensier, M. van Dijk D., forthcoming. Structural breaks in theinternational dynamics of inflation. Review of Economics and Statistics.
Bekaert, G., Hodrick, R.J., Zhang, X., 2009. International stock return comovements.Journal of Finance 64, 2591–2626.
Bekaert, G., Harvey, C.R., Lundblad, C.T., Siegel, S., 2012. The European Union, theEuro, and Equity Market Integration, AFA 2012 Chicago Meetings Paper. SSRN:#1573308.
Billio, M., Pelizzon, L., 2003. Contagion and interdependence in stock markets: havethey been misdiagnosed? Journal of Economics and Business 55, 405–426.
Bley, J., 2009. European stock market integration: fact or fiction? Journal ofInternational Financial Markets, Institutions and Money 19, 759–776.
Brown, M.B., Forsythe, A.B., 1974. Robust test for the equality of variances. Journalof the American Statistical Association 69, 364–367.
Burns, P., Engle, R., Mezrich, J., 1998. Correlations and volatilities of asynchronousdata. Journal of Derivatives 5, 7–18.
Büttner, D., Hayo, B., 2011. Determinants of European stock market integration.Economic Systems 35, 574–585.
Calvo, S., Reinhart, C.M., 1996. Capital flows to latin america: is there evidence ofcontagion effects? In: Calvo, G., Goldstein, M., Hochreiter, E. (Eds.), PrivateCapital Flows to Emerging Markets. Institute for International Economics,Washington, DC.
Christodoulakis, G.A., Satchell, S.E., 2002. Correlated ARCH (CorrARCH): modellingthe time-varying conditional correlation between financial asset returns.European Journal of Operational Research 139, 351–370.
Corsetti, G., Pericoli, M., Sbracia, M., 2005. Some contagion, some interdependence:more pitfalls in tests of financial contagion. Journal of International Money andFinance 24, 1177–1199.
Den Haan, W.J., Levin, A., 1998. Vector Autoregressive Covariance MatrixEstimation. Manuscript, University of California, San Diego.
Forbes, K.J., 2012. The ‘‘Big C’’: Identifying Contagion. NBER Working Paper 18465.Forbes, K.J., Rigobon, R., 2002. No contagion, only interdependence: measuring stock
market comovements. Journal of Finance 57, 2223–2261.Fratzscher, M., 2002. Financial market integration in Europe: on the effects of EMU
on stock markets. International Journal of Finance and Economics 7, 165–193.Hamao, Y., Masulis, R.W., Ng, V., 1990. Correlations in price changes and volatility
across international stock markets. Review of Financial Studies 3, 281–307.Hardouvelis, G.A., Malliaropulos, D., Priestley, R., 2006. EMU and European stock
market integration. The Journal of Business 79, 365–392.Inclan, C., Tiao, G.C., 1994. Use of cumulative sums of squares for retrospective
detection of changes of variance. Journal of the American Statistical Association89, 913–923.
Judge, G., Griffiths, W., Hill, R., Lutkepohl, H., 1985. The Theory and Practise ofEconometrics. Wiley, New York.
Kahya, E., 1997. Correlation of returns in non-contemporaneous markets.Multinational Finance Journal 1, 123–135.
Karoglou, M., 2010. Breaking down the non-normality of daily stock returns.European Journal of Finance 16, 79–95.
B. Gebka, M. Karoglou / Journal of Banking & Finance 37 (2013) 3639–3653 3653
Kim, S.J., Moshirian, F., Wu, E., 2005. Dynamic stock market integration driven bythe European Monetary Union: an empirical analysis. Journal of Banking andFinance 29, 2475–2502.
King, M.A., Wadhwani, S., 1990. Transmission of volatility between stock markets.Review of Financial Studies 3, 5–33.
Kleimeier, S., Lehnert, T., Verschoor, W.F.C., 2008. Measuring financial contagionusing time-aligned data: the importance of the speed of transmission of shocks.Oxford Bulletin of Economics and Statistics 70, 493–508.
Kokoszka, P., Leipus, R., 2000. Change-point estimation in ARCH models. Bernoulli 6,513–539.
Koutmos, G., Booth, G.G., 1995. Asymmetric volatility transmission in internationalstock markets. Journal of International Money and Finance 14, 747–762.
Lee, S.B., Kim, K.J., 1993. Does the October 1987 crash strengthen the co-movementsamong national stock markets. Review of Financial Economics 3, 89–102.
Levene, H., 1960. Robust tests for equality of variances. In: Olkin, I., Ghurye, S.G.,Hoeffding, W., Madow, W.G., Mann, H.B. (Eds.), Contributions to Probability andStatistics: Essays in Honor of Harold Hotelling. Stanford University Press, MenloPark, CA.
Martens, M., Poon, S., 2001. Returns synchronization and daily correlation dynamicsbetween international stock markets. Journal of Banking and Finance 25, 1805–1827.
Morana, C., Beltratti, A., 2002. The effects of the introduction of the euro on thevolatility of European stock markets. Journal of Banking and Finance 26, 2047–2064.
Mylonidis, N., Kollias, C., 2010. Dynamic European stock market convergence:evidence from rolling cointegration analysis in the first euro-decade. Journal ofBanking and Finance 34, 2056–2064.
Newey, W., West, K., 1994. Automatic lag selection in covariance matrix estimation.Review of Economic Studies 61, 631–653.
Pascual, A.G., 2003. Assessing European stock markets (co)integration. EconomicsLetters 78, 197–203.
Rigobon, R., 2003. On the measurement of the international propagation of shocks:is the transmission stable? Journal of International Economics 61, 261–283.
Rua, A., Nunes, L.C., 2009. International comovement of stock market returns: awavelet analysis. Journal of Empirical Finance 16, 632–639.
Sansó, A., Aragó, V., Carrion, J.L., 2003. Testing for Changes in the UnconditionalVariance of Financial Time Series’. Department d’ Economia Aplicada WorkingPaper, Universitat de les illes Balears.
Savva, C.S., 2009. International stock markets interactions and conditionalcorrelations. Journal of International Financial Markets, Institutions andMoney 19, 645–661.
Savva, C.S., Aslanidis, N., 2010. Stock market integration between new EU memberstates and the Euro-zone. Empirical Economics 39, 337–351.
Savva, C.S., Osborn, D.R., Gill, L., 2009. Spillovers and correlations between US andmajor European stock markets: the role of the euro. Applied FinancialEconomics 19, 1595–1604.
Sheskin, D., 1997. Handbook of Parametric and Nonparametric StatisticalProcedures. CRC Press, Boca Raton, FL.
Siegel, S., Tukey, J., 1960. A nonparametric sum of ranks procedure for relativespread in unpaired samples. Journal of the American Statistical Association 55,429–445.
Sokal, R., Rohlf, F., 1995. Biometry: The Principles and Practise of Statistics inBiological Research, third ed. W.H. Freeman and Co, New York.
Tsay, R.S., 2002. Analysis of Financial Time Series. John Wiley & Sons, New York.Yang, J., Min, I., Li, Q., 2003. European stock market integration: does EMU matter?
Journal of Business Finance and Accounting 30, 1253–1276.Yu, J., Meyer, R., 2006. Multivariate stochastic volatility models: Bayesian
estimation and model comparison. Econometric Reviews 25, 361–384.
top related