Equity market integration in emerging Balkan markets

Research in International Business and Finance 25 (2011) 296– 307

Contents lists available at ScienceDirect

Research in International Businessand Finance

journal homepage: www.elsevier.com/locate/r ibaf

Equity market integration in emerging Balkan markets

Dimitris Kenourgiosa,∗, Aristeidis Samitasb

a University of Athens, Faculty of Economics, Greeceb Department of Business Administration, Business School, University of the Aegean, Greece

a r t i c l e i n f o

Article history:Received 4 November 2008Received in revised form 18 February 2011Accepted 23 February 2011

Available online 4 March 2011

JEL classification:C22G15

Keywords:Emerging Balkan stock marketsCointegrationRegime switchingDynamic conditional correlations

a b s t r a c t

This paper examines long-run relationships among five Balkanemerging stock markets (Turkey, Romania, Bulgaria, Croatia,Serbia), the United States and three developed European markets(UK, Germany, Greece), during the period 2000–2009. Conven-tional, regime-switching cointegration tests and Monte Carlosimulation provide evidence in favour of a long-run cointegrat-ing relationship between the Balkan emerging markets within theregion and globally. Moreover, we apply the Asymmetric Gen-eralized Dynamic Conditional Correlation (AG-DCC) multivariateGARCH model of Cappiello et al. (2006), in order to capture theimpact of the 2007–2009 financial crisis on the time-varying cor-relation dynamics among the developed and the Balkan stockmarkets. Results show that stock market dependence is heightened,supporting the herding behaviour during the 2008 stock marketcrash period. Our findings have important implications for inter-national portfolio diversification and the effectiveness of domesticpolicies, as these emerging markets are exposed to external shocks.

© 2011 Elsevier B.V. All rights reserved.

1. Introduction

Recent developments in international financial markets and the consequent globalization haveattracted a number of studies focusing on international stock markets convergence. The literature onemerging stock markets supports that they became less segmented from the world markets elevating

∗ Corresponding author. Tel.: +30 210 3689449; fax: +30 210 3225542.E-mail address: [email protected] (D. Kenourgios).

0275-5319/$ – see front matter © 2011 Elsevier B.V. All rights reserved.doi:10.1016/j.ribaf.2011.02.004

dx.doi.org/10.1016/j.ribaf.2011.02.004

http://www.sciencedirect.com/science/journal/02755319

http://www.elsevier.com/locate/ribaf

mailto:[email protected]

dx.doi.org/10.1016/j.ribaf.2011.02.004

D. Kenourgios, A. Samitas / Research in International Business and Finance 25 (2011) 296– 307 297

the benefits of economic growth created via improvements in allocational efficiency and reductionsin macroeconomic volatility (e.g., Baele et al., 2004).

Cointegration methodology developed by Engle and Granger (1987) and Johansen (1988) has beenadopted in a numerous studies investigating the long-run relationships between international stockmarkets. These studies have traditionally focused on European and US mature equity markets, Asianand Latin American emerging markets and to a lesser extent on Central European stock markets,providing evidence on long-run interrelationships.1

However, the long-run stable equilibrium relationships conjectured by these techniques are notsuitable for modeling the dynamic process of stock market integration as it is incomplete and continuesto exhibit strong variations over time.2 Campos et al. (1996) and Gregory and Hansen (1996) find thatthe time-varying relations along with some structural breaks are possible to deteriorate the powerof cointegration tests, rejecting the null hypothesis of cointegration. Hence, their methodology takesinto account the possibility of instability in short-run relations. Furthermore, their research based onMonte Carlo simulation technique verifies that when a shift in parameters takes place, standard testfor cointegration lose power and provides false signals.

Despite the extensive use of Johansen and Gregory–Hansen cointegration tests on the examinationof international equity market integration, scholars have been also using generalized autoregressiveconditional heteroskedasticity (GARCH) family models to avoid the restriction of constant volatilityimplied by the aforementioned methodologies. A large body of literature applies several variants ofGARCH models to accommodate the possibilities of non-normalities and asymmetries in the vari-ance of returns (e.g., Bekaert et al., 2002; Baele, 2005). However, most of the GARCH family modelsassume that correlation coefficients are constant over the sample period, while their multivariatevariants suffer from the curse of dimensionality. Recently, two more advanced GARCH models havebeen developed. Engle (2002) develops a dynamic conditional correlation (DCC) GARCH model, whichallows conditional asymmetries in both volatilities and correlations in a time-varying framework.Furthermore, Cappiello et al. (2006) extend the original model of Engle (2002) and use the asym-metric generalized dynamic conditional correlation (AG-DCC) model to investigate asymmetries inconditional variances and correlation dynamics.

This paper examines the interrelationships among five Balkan emerging stock markets (Romania,Bulgaria, Serbia, Turkey and Croatia), the US and three developed European markets (UK, Germany,and Greece), during the period 2000–2009. Romania and Bulgaria have recently joined EU, whileTurkey and Croatia remain EU accession candidates, and Serbia struggles to emerge from isolationand start negotiations for future EU enrolment. Following the existing literature, we initially employa conventional empirical analysis to test the equity market integration hypothesis (Johansen andGregory–Hansen residual-based cointegration tests and Monte Carlo simulation technique to testsensitivity of the cointegration estimation). However, to avoid the limitations of this analysis andprovide more robust results, we also apply the AG-DCC model developed by Cappiello et al. (2006).We estimate time-varying asymmetric conditional correlations between each developed market andall other Balkan markets during stable and crisis periods in order to examine the impact that the2007–2009 financial crisis had on the integration of Balkan emerging equity markets. This multivariateprocess allows for series-specific news impact and smoothing parameters, permits conditional asym-metries in correlation dynamics and accounts for heteroskedasticity directly by estimating correlationcoefficients using standardized residuals. Moreover, this specification is well suited to investigate thepresence of asymmetric responses in conditional variances and correlations during periods of negativeshocks.

1 See, for example, Francis and Leachman (1998), Phylaktis (1999), Manning (2002), Chen et al. (2002), Chaudhuri and Wu(2003), Voronkova (2004), Syriopoulos (2006), and Samitas and Kenourgios (2007).

2 Masih and Masih (2002) provide evidence that the Johansen procedure is heavily dependent on time series that exhibit fewoutlying observations. Voronkova (2004) adopts the Gregory and Hansen test and finds several equilibrium relations omittedby the conventional cointegration testing procedures, showing that long-run relations do not cease after a structural changehas occurred. Kim et al. (2005) support that only the existence of an equilibrating process and not the driving forces behind thelong-run equilibrium are investigated in standard cointegration analyses.

298 D. Kenourgios, A. Samitas / Research in International Business and Finance 25 (2011) 296– 307

Our empirical analysis provides three main findings. First, there is a time-varying cointegratingrelationship between the Balkan emerging markets within the region and globally. Second, a regimeshift appeared for all Balkan markets in year 2002, implying their increased exposure to the globalenvironment. Third, dependence is heightened among the Balkan and the developed equity markets,supporting the herding behaviour during the 2008 stock market crash period. The analysis in thepaper has implications for international portfolio diversification, the smooth E.U. accession of Balkaneconomies and their long-term growth prospects.

Our analysis contributes to the literature in the following aspects. First, it applies a much widerbattery of tests than other previous studies, trying to shed further light on the presence of long-run linkages between emerging stock markets and their mature counterparts. Second, to the bestof our knowledge it employs the AG-DCC model that has not been used before to test the Balkanequity markets integration hypothesis using data for the 2007–2009 financial turmoil. Third it extendsprior but very limited research on this topic, which report interrelationships and linkages betweendeveloped and Balkan stock markets using conventional cointegration tests and vector error correctionmodels for a period until 2007 (Syriopoulos, 2011).

The structure of the paper is organized as follows: Section 2 provides an overview of the Balkaneconomic and financial environment. Section 3 analyzes the methodological issues. Section 4 presentsthe data. The empirical results are reported in Section 5. The final section contains the concludingremarks.

2. The Balkan stock markets and economies

Over the last decade, impressive changes have occurred in Balkans; from the conflicts and economiccollapse to the break up of traditional trade within the region. Since 2000, the Balkan economies arethrough a transitory phase of structural adjustment towards a market oriented economic system.Nevertheless, during all these years, the Balkan region displays robust growth rates, expanding morerapidly than the EU average.

Table 1 presents key financial statistics for the Balkan region. Average growth measured as a per-centage of GDP exceeded 4% in the Balkan region. Romania, Croatia, Bulgaria and Turkey are among thetop performers. Inflation continues to drop to a single-digit annual rate throughout the region. Since2004, inflation rates converge to the EU average. The simultaneous growth increase and inflationdecrease appreciated capital inflows of foreign direct and capital investments.

The Balkan stock markets have a brief history compared to the mature markets of Europe and US.These markets started trading in the mid-1980s to mid-1990s with a small number of stocks, manyof which were illiquid. During 2000–2006, stock prices in the Balkan markets increased over 70%in dollar terms (on average), compared to the 15% of MSCI world market return. Among the Balkanstock markets, Turkey, Romania, Bulgaria and Croatia are considered the most developed, in terms ofcapitalization, turnover and market return.

Despite the robust growth rates, the Balkan equity markets remain small in terms of capitalization,turnover and liquidity compared to the developed European markets. At the end of 2006, all thesemarkets together amounted to less than 7% of the German stock market. The market capitalization issmall and the trading activity is low with value traded as a share of market capitalization averaging lessthan 20%. Institutional investors are small in the Balkan economies, with the exception of the Turkishstock exchange (ISE) in which they own above 50% of the free float shares. However, prospects forthese markets seem to be prosperous due to the vast restructuring effort in public and private sectors,the expectation to join EU, the increased value of institutional investors’ assets and improvements ininvestor protection.

3. Methodology

3.1. Johansen and Gregory–Hansen cointegration tests

We test for cointegration among the developed and the Balkan emerging mar-kets using the Johansen Maximum Likelihood Procedure (Johansen, 1988). The


Table 1Financial statistics for the Balkan region.

Countries Year GDP Inflation FDIa Market return

Turkey 2000 3.4% 34.2% 1.2 −45.32%2001 −7.5% 54% 2 −29.1%2002 7.9% 21.6% 0.5 −44.46%2003 5.8% 8.6% 0.5 75.67%2004 9% 8.2% 0.7 27.88%2005 7.4% 7.6% 2.4 84.9%2006 7.1% 8% 4.7 −12.68%2007 5% 9% 20.1 8.76%2008 1% 10% 15.4 −17.6

Romania 2000 5.7% 14.5% 2 −36.82%2001 5.7% 12.5% 2.5 −27.55%2002 5.1% 6.3% 2.1 −57.08%2003 5.2% 5% 2.4 29.71%2004 8.4% 3.4% 6.1 73.56%2005 4.1% 3.2% 6.5 64.32%2006 7.5% 3.8% 8 47.8%2007 6% 5% 9.6 27.9%2008 9% 8% 13.6 −48.2%

Bulgaria 2000 2.5% 7.8% 5.9 −29.63%2001 4% 7.4% 6.1 −32.9%2002 4.8% 5.9% 6 −68.73%2003 4.5% 2.3% 10.4 19.24%2004 5.6% 6.1% 11.8 35.6%2005 5.5% 5% 10.9 58.94%2006 6.1% 7.3% 16.6 16.89%2007 6% 8% 11.4 9.37%2008 6% 12% 8.4 −28.5%

Croatia 2000 2.9% 7.6% 6 −4%2001 3.4% 7.1% 8.3 −37.18%2002 3.6% 6% 8.6 −73.86%2003 5.7% 4.2% 10.9 29.03%2004 7.2% 3% 11.2 45.55%2005 8.3% 2.5% 11.4 69.47%2006 6.3% 3% 13.6 42.26%2007 5% 3% 4.7 33.0%2008 2% 6% 4.6 −47.8%

Serbia 2000 2.6% 11.7% 0.9 −20.72%2001 5.1% 10.2% 1.6 −38.29%2002 4.5% 8.3% 3 −44.33%2003 2.4% 6.5% 6.7 16.98%2004 9.3% 5.1% 3.9 31%2005 6.8% 4.9% 5.9 35.02%2006 4.7% 3% 5.5 37.51%2007 8% 6% 2.5 27.8%2008 1% 13% 2.7 −52.6%

IMF.a Foreign direct investments (FDI) in million Euros.

Johansen cointegration procedure firstly specifies the following unrestricted N-variableVAR:

xt = � +k∑

i=1

˘ixt−i + εt (1)

where x′t = [f ′

t , s′t], � is a vector of intercepts terms and εt is a vector of error terms.

Johansen (1988) and Johansen and Juselius (1990) re-parameterized Eq. (1) in the


form:

�xt = � +k−1∑

i=1

�i�xt−i + ˘xt−k + εt (2)

Eq. (2) is now a VAR re-parameterized in error correction form, where ̆ = −(˘ − ˘1 − · · · − ˘k)represents the long response matrix. Writing this matrix as ̆ = ˛ˇ′, then the linear combinations ˇ′xt−kwill be I(0) in the existing of cointegration, with ̨ being the adjustment coefficients, and the matrix ˘will be of reduced rank. The Johansen approach can be used to test for cointegration by assessing therank (r) of the matrix ˘ . If r = 0 then all the variables are I(1) and there are no cointegrating vectors.If 0 < r < N there will be r cointegrating vectors. Last, if r = N then all of the variables are I(0) and, giventhat any linear combinations of stationary variables will also be stationary, there are N cointegratingvectors.

In order to allow for the possibility of a structural change, we use the Zivot and Andrews (1992)sequential test for a unit root with the alternative hypothesis of stationarity and a single structuralchange in the deterministic trend. Then, we apply the Gregory–Hansen test which states the nullhypothesis of no cointegration against the alternative of cointegration with one structural break.Gregory and Hansen suggest three alternative model specifications.

Model 1 accommodates the changes in parameters of the cointegration vector under the alternative.A level shift model allows for a change only in the intercept.

Y1t = �1 + �2ϕt� + ˛′y2t + εt, t = 1, . . . , n (3)

Model 2 accommodates a trend in the data, while also restricts the changes to shifts in the level.

Y1t = �1 + �2ϕt� + ˇt + ˛′y2t + εt, t = 1, . . . , n (4)

Model 3 is the most general specification that allows for changes both in the intercept and in theslope of the cointegration vector.

Y1t = �1 + �2ϕt� + ˛′1t + ˛′

2y2tϕt� + εt, t = 1, . . . , n (5)

The dummy variable which captures the structural change is represented as

ϕt� = 0, t ≤ [n�] and ϕt� = 1, t ≥ [n�]

where � ∈ (0, 1) is a relative timing of the change point with the break point changing over the previousinterval.

3.2. Monte Carlo simulation

Monte Carlo simulation technique is applied to test if cointegration produces false signals. We focuson the two regime model with fixed switching transition probabilities ˛′

1 = ˛′2 = 0 (=ˇ1 = ˇ2 = 0). For

each Monte Carlo experiment, 400 simulated series are generated (scenario generation process) fromModel 3 (Eq. (5)), replicated under the null hypothesis of a random walk (no cointegration) with i.i.d.normal innovations. We consider two portfolios; one for the Balkan stock markets and the other forthe mature markets measured both in daily average absolute return and evaluate the performance ofmaximum likelihood estimation when the true model is the endogenous switching model in Eq. (5).

For each simulation, we generate the vector of exogenous explanatory variables as xt = [1, x∗t ]

where x∗t ∼ i.i.d.N(0, 2), and fix the vector of regime switching parameters to ˇ1 = (ˇ0,1, ˇ1,1)′ = (1.0,

1.0)′, ˇ2 = (ˇ0,2, ˇ1,2) = (−1.0, −1.0)′, �1 = 0.33, �2 = 0.67.3 We consider three different sets of transitionprobabilities corresponding to moderate persistence (p11 = 0.7, p22 = 0.7), high persistence (p11 = 0.9,p22 = 0.9), and differential persistence (p11 = 0.7, p22 = 0.9). We also consider three different values for

3 We identify switching regimes based on two parameters. First, the beta (ˇ), where ˇ1 is the movement of the first country’sstock index and ˇ2 is the movement of the second country’s stock index. So, if ˇ1 (ˇ2) is 1.0 or −1.0 then we identify a switchingregime. Second, the sigma, where �1 must be 0.33 or 0.67 and �2 must be 0.67 or 0.33 so that �1 + �2 = 1.0.


corresponding to high correlation ( = 0.9), moderate correlation ( = 0.5), and zero correlation ( = 0).We record the outcome of 5% nominal size Wald and Likelihood ratio tests of the null hypothesis p = 0.For those cases where p /= 0, we use size adjusted critical values taken from Monte Carlo simulationgenerated with p = 0, to measure the power of the tests. The levels adopted in our research has beenused because these allow control for many parameters affecting the size distortion of the LR and Waldtest statistics, such as the speed of adjustment, the correlation between the innovations, and the signalto noise parameters.

3.3. AG-DCC GARCH model

Volatilities and correlations measured from historical data may miss changes in risk. Therefore,Cappiello et al. (2006) investigate properties of international equity returns generalizing the DCC-GARCH model of Engle (2002) by introducing two modifications: asset-specific correlation evolutionparameters and conditional asymmetries in correlation dynamics.4

In this paper, we estimate univariate volatility applying a GARCH (1, 1) model, and the standardisedresiduals are used to estimate the correlation parameters. The evolution of the correlation in thestandard DCC model (Engle, 2002) is given by:

Qt = (1 − a − b)P̄ + aεt−1ε′t−1 + bQt−1 (6)

Pt = Q ∗−1t QtQ

∗−1t (7)

where P̄ = E[εtε′t] and a and b are scalars such that a + b < 1. Q ∗

t = [q∗iit

] = [√

qiit] is a diagonal matrixwith the square root of the ith diagonal element of Qt on its ith diagonal position. As long as Qt ispositive definite, Q ∗

t is a matrix which guarantees Pt = Q ∗−1t QtQ

∗−1t is a correlation matrix with ones

on the diagonal and every other element ≤1 in absolute value. The model described by Eqs. (6) and(7), however, does not allow for asset-specific news and smoothing parameters or asymmetries.

Cappiello et al. (2006) modify the correlation evolution equation as:

Qt = (P̄ − A′P̄A − B′P̄B − G′N̄G) + A′εt−1ε′t−1A + G′nt−1n′

t−1G + B′Qt−1B (8)

where A, B and G are k × k parameter matrices, nt = I[εt ≺ 0] ◦ εt (I[·] is a k × 1 indicator function whichtakes on value 1 if the argument is true and 0 otherwise, while “◦” indicates the Hadamard product) andN̄ = E[ntn′

t]. Eq. (8) is the AG-DCC model. In order the Qt to be positive definite for all possible realiza-tions, the intercept, P̄ − A′P̄A − B′P̄B − G′N̄G must be positive semi-definite and the initial covariancematrix Q0 positive definite.

4. The data

The data used in this study consist of the daily closing prices (in logs) in five Balkan stock markets,US, UK, Germany and Greece. The stock market indices of interest are SOFIX of Bulgaria, VANGUARD ofRomania, CROBEX of Croatia, ISE NATIONAL 50 of Turkey, BELEX of Serbia, S&P500 of US, FTSE 100 ofUK, Xetra DAX of Germany and the Athens General Index (ASE) of Greece. All Balkan daily closing pricesare adjusted to inflation. The inclusion of UK, Germany and US is important as reasonable proxies forthe mature European and North-American stock markets, respectively, while Greece is also chosenfor its traditional trading and cultural ties with the Balkan countries.

All the national stock indices are selected to guarantee representativeness of the domestic marketsexamined in this study. Furthermore, following studies on other emerging stock markets, we useindices expressed in local currencies (e.g., Voronkova, 2004). This restricts their changes to movementsin the security prices and avoids distorting the cointegration analysis results with devaluations of theexchange rates that took place in the region. However, the findings are basically the same when thedata are converted in euros.

4 They support that the introduction of fixed exchange rate regimes led to increased correlation among stock markets andnear-perfect correlation among government bond returns within the EMU countries. They also document a structural break incorrelation in 1999 using equity indices from European, Australasia and American countries.


Table 2Descriptive statistics.

Stock markets Min Max Standard deviation Skewness Kurtosis

Turkey −0.21 0.18 0.023 −0.19 8.04Romania −0.17 0.14 0.014 −0.17 6.93Bulgaria −0.19 0.21 0.020 −0.19 7.87Croatia −0.15 0.14 0.022 −0.13 7.16Serbia −0.19 0.21 0.011 −0.24 8.39

This table reports summary statistics for the five Balkan equity index returns (at daily frequency) during the period 2000–2009.All of the indices are expressed in local currencies.

The sample covers a period from January 2000 till February 2009.5 Because most markets are oper-ating in the same time zone, the problem of non-overlapping trading hours does not arise (except theUS). Price changes in Europe are reflected the next day because trading on European Stock exchangesis over by the time trading on the American market commences. Also, when a stock exchange is closeddue to a national holiday, we use the previous day closing prices.

In order to examine the impact of 2008 global financial crisis on the integration of Balkan emerg-ing equity markets, we apply the AG-DCC model, taking into account the 2008 stock market crashas crisis period (26/09/2008 to end of period) and the period before (1/10/2002 to 25/09/2008) asstable period (excluding the dot-com bubble collapse from 10/03/2000 till 27/09/2002).6 Then, wecompare the difference in returns, actual and asymmetric volatilities and correlations between stableand crisis periods. Therefore, we split the estimation procedure into subgroups in order to estimateunconditional and conditional average correlations between each developed market and all otherBalkan emerging markets.

5. Empirical results

Table 2 reports the descriptive statistics for the Balkan stock market returns that are of primeinterest to international portfolios. All stock price series show leptokurtosis and there is evidenceof negative skewness. Significant kurtosis and negative skewness (long left tail) indicate rejection ofnormality in stock return distributions.

5.1. Johansen and Gregory–Hansen cointegration test results

Since the null hypothesis of a unit root cannot be rejected, a vector error correction model isestimated to consider the nine series jointly.7 Three alternative model versions are compared: (a) amodel with constant restricted to the cointegration space; (b) a model with unrestricted constant tothe cointegration space; and (c) a model with a linear trend in the cointegration sector in order tosearch for the presence or the absence of cointegrating relationships in the data set.8

The Johansen cointegration test results are presented in Table 3. The findings indicate the presenceof one cointegrating vector in our sample, in all three versions of the model. Thus, the null hypothesisthat the Balkan emerging markets, US, Germany, UK and Greece are not cointegrated (r = 0) againstthe alternative of one cointegrating vector (r ≤ 1) is rejected, since the max and trace statistics exceedthe critical value at the 5% significance level.

5 Constraints on data availability for most Balkan emerging markets before year 2000 and “market failures” (small capital-ization, infrequent trading, and low volume) result to focusing our analysis during the period 2000–2009.

6 A sensitivity analysis shows that period definition (tranquil and turbulent period) do not affect the central results.7 ADF and KPSS unit root tests results indicate that all series can be described as I(1) processes. Furthermore, the breakpoints,

as identified by the Zivot-Andrews test, are statistically significant and vary across countries. All test results are available up onrequest.

8 The order that the stock indices enter the VAR model is based on their capitalization. The optimal lag structure was chosenby the Akaike Information Criterion minimization and the absence of autocorrelation in the VAR residuals; one lag for the levelsof variables was included.


Table 3Johansen cointegration test results (2000–2009).

Null Eigenvalues trace test Critical values at 95%

Model 1 Model 2 Model 3 Model 1 Model 2 Model 3 Model 1 Model 2 Model 3

r = 0 0.0212 0.0209 0.0202 107.85 105.91 109.23 103.00 102.70 103.58r ≤ 1 0.0136 0.0145 0.0151 63.49 61.05 67.67 73.19 69.04 75.34

Null max test Critical values at 95%

Model 1 Model 2 Model 3 Model 1 Model 2 Model 3

r = 0 40.25 39.08 40.83 39.90 38.74 40.68r ≤ 1 19.36 19.78 21.34 32.55 31.78 32.58

Model 1: model with a constant restricted to the cointegrating space. Model 2: model with unrestricted constant. Model 3:model with a linear trend in the cointegrating vector. Critical values are obtained from Osterwald-Lenum (1992).

Table 4 reports the Gregory–Hansen cointegration test results. The pairwise test results showthat the null hypothesis of no cointegration among the Balkan equity markets is rejected. Moreover,Gregory–Hansen test using Model 3, which allows for changes both in the intercept and in the slope ofthe cointegration vector, produces the stronger cointegrating relation among the Balkan stock markets.

Table 4Gregory–Hansen cointegration test results among equity markets (2000–2009).

Stock indices Break point Philips Zt

Model 1 Model 2 Model 3 Model 1 Model 2 Model 3

Panel A: Cointergation results among Balkan stock marketsISE NATIONAL 50 – VANGUARD 0.52 0.52 0.53 −4.85* −4.85* −4.88*ISE NATIONAL 50 – SOFIX 0.40 0.40 0.44 −4.78* −4.78* −4.80*ISE NATIONAL 50 – CROBEX 0.39 0.39 0.40 −4.77* −4.77* −4.79*ISE NATIONAL 50 – BELEX 0.30 0.30 0.33 −4.71* −4.72* −4.75*VANGUARD – SOFIX 0.48 0.49 0.51 −4.81* −4.82* −4.83*VANGUARD – CROBEX 0.38 0.38 0.44 −4.78* −4.79* −4.80*VANGUARD – BELEX 0.33 0.32 0.31 −4.69* −4.69* −4.70*SOFIX – CROBEX 0.34 0.39 0.41 −4.68* −4.68* −4.69*SOFIX – BELEX 0.34 0.33 0.39 −4.67* −4.67 −4.70*CROBEX – BELEX 0.37 0.37 0.39 −4.70* −4.70* −4.72*

Panel B: Cointergation results between the developed and the Balkan stock marketsISE NATIONAL 50 – S&P 0.32 0.30 0.34 −4.72* −4.70* −4.73*ISE NATIONAL 50 – FTSE 100 0.33 0.31 0.36 −4.74* −4.74* −4.75*ISE NATIONAL 50– DAX 0.39 0.39 0.40 −4.79* −4.79* −4.80*ISE NATIONAL 50 – ASE G.I. 0.36 0.33 0.36 −4.78* −4.73* −4.79*VANGUARD – S&P 0.30 0.31 0.33 −4.69* −4.69* −4.70*VANGUARD – FTSE 100 0.32 0.32 0.34 −4.70* −4.71* −4.72*VANGUARD – DAX 0.38 0.37 0.40 −4.76* −4.76* −4.78*VANGUARD – ASE G.I. 0.40 0.39 0.41 −4.79* −4.80* −4.81*SOFIX – S&P 0.30 0.31 0.33 −4.69* −4.68* −4.70*SOFIX – FTSE 100 0.31 0.31 0.33 −4.69* −4.68* −4.70*SOFIX – DAX 0.36 0.35 0.38 −4.75* −4.75* −4.76*SOFIX – ASE GI 0.39 0.39 0.41 −4.78* −4.77* −4.79*CROBEX – S&P 0.30 0.30 0.31 −4.69* −4.68* −4.70*CROBEX – FTSE 100 0.31 0.31 0.32 −4.69* −4.70* −4.71*CROBEX – DAX 0.32 0.32 0.36 −4.71* −4.71* −4.76*CROBEX – ASE G.I. 0.33 0.34 0.34 −4.71* −4.71* −4.72*BELEX – S&P 0.27 0.28 0.27 −4.62* −4.64* −4.62*BELEX – FTSE 100 0.28 0.28 0.27 −4.62* −4.64* −4.63*BELEX – DAX 0.30 0.29 0.31 −4.62* −4.65* −4.69*BELEX – ASE G.I. 0.31 0.38 0.36 −4.69* −4.72* −4.75*

The correspondence between stock markets and indices is: SOFIX: Bulgaria; VANGUARD: Romania; CROBEX: Croatia; ISENATIONAL 50: Turkey; BELEX: Serbia. The optimal lag length of the VAR is chosen on the basis of Schwarz Information Criterion.The 5% critical value of Zt is −4.61. ‘*’ indicates rejection of the null hypothesis at the 5% level.


Regime Probability

0

0,05

0,1

0,15

0,2

0,25

0,3

0,35

0,4

0,45

2000

2001

2002

2003

2004

2005

2006

2007

2008

2009

Date

Pro

bab

ilit

y

ISE 50

VANGUARD

CROBEX

SOFIX

BELEX

Fig. 1. Regime probability in the Balkan region (2000–2009). The figure shows the probability to occur a structural breakbetween the Balkan markets and US S&P500, using Gregory–Hansen Model 3. When the value approaches one (1), the probabilityis higher. On the other hand, when the value approaches zero (0), the probability is very low. The correspondence between stockmarkets and indices is: SOFIX: Bulgaria; VANGUARD: Romania; CROBEX: Croatia; ISE NATIONAL 50: Turkey; BELEX: Serbia.

The results also reject the null hypothesis of no cointegration among the Balkan and the developedstock markets. This implies that a structural change is present in the pattern of long-run stock marketco-movements. Moreover, it seems that the Athens Stock Exchange (ASE) plays a critical and leadingrole in the Balkan region among the developed stock markets, since it shares the stronger cointegratingrelationship with most of the Balkan markets (Romania, Bulgaria, and Serbia). This is due to theirhistorical neighboring, the extensive trade and the strong financial penetration of Greek firms in theregion.

Fig. 1 shows the probability to occur a structural break between the Balkan markets and the USS&P500, using Gregory and Hansen Model 3.9 A structural break seems to appear for all Balkan marketsin year 2002, where developed stock markets indices declined significantly due to the Internet bubblecollapse (2000–2002). Increased regime probabilities also observed in years 2007 and 2008, mostlydue to the emergence of the global financial crisis, while Turkey exhibit the higher structural breakamong all other Balkan markets in 2009. These findings imply the increased exposure of the Balkanemerging markets to the global environment.

5.2. Monte Carlo simulation results

Table 5 reports Monte Carlo simulation results between developed and Balkan stock markets. Inour analysis, we investigate maximum likelihood estimation of the endogenous switching Model 3,since it produces the stronger cointegrated results using Gregory and Hansen test. Particularly, wereport the mean of the maximum likelihood point estimates (ˇ1, ˇ2), the mean of the standard errorsfor these parameter estimates (�1, �2), and the rejection rate of the test (p = 0).

Our findings indicate that the null hypothesis of no cointegration is rejected at conventional sig-nificance levels, supporting the results obtained from Gregory and Hansen test. The estimation biasthat occurs from the endogenous state variable is close to their true value. The mean estimates of �1and �2 are also biased downwards. The estimates of ˇ1,1 and ˇ1,2 are close to their true values. Sincethe estimates are close to their true values and the LR test is not rejected, we provide endogenousestimators with very accurate estimates. Indeed, for all values of the transition probabilities p, themean parameter estimates are nearly identical to their true values.

We also find that the Wald test is oversized, with rejection rate as high as 11%. However, the LR testhas roughly correct size (6.5%) and appears to be fairly reliable for the endogenous switching Model3. Overall, Monte Carlo maximum likelihood estimates confirm a cointegration relationship betweenBalkan and mature stock markets.

9 A structural break is also appeared in year 2002 among the Balkan markets and FTSE-100, Xetra DAX and ASE General Index,respectively.


Table 5Monte Carlo simulation results among the Balkan and the developed stock markets (2000–2009).

ˇ0,1 ˇ0,2 ˇ1,1 ˇ1,2 �1 �2

p11 = 0.7, p22 = 0.7Endog. Estimator 0.97 (0.05) −0.98 (0.08) 1.00 (0.02) −0.99 (0.03) 0.32 (0.05) 0.66 (0.04)Wald Test 11.0%LR Test 6.5%



Sample for T = 400. Each cell contains the mean of the maximum likelihood point estimates and the mean of the standarderrors of these estimates from the Monte Carlo experiment. Endogenous estimator refers to the maximum likelihood estimatorallowing the state process to be endogenous, so that p ∈ (−1, 1). The standard error is used to calculate confidence intervals forour true sample. Wald and LR tests’ significance at 5% level.

Table 6Unconditional average correlations among equity index returns.

Developed markets Emerging markets 2008 stock marketcrash (26/09/2008to end of period)

Stable period(1/10/2002 to25/09/2008)

S&P ISE NATIONAL 50 0.482* 0.374*VANGUARD 0.275* 0.197*SOFIX 0.263* 0.183*CROBEX. 0.285* 0.201*BELEX 0.198* 0.142*

FTSE 100 ISE NATIONAL 50 0.473* 0.365*VANGUARD 0.269* 0.189*SOFIX 0.209* 0.171CROBEX 0.267* 0.199*BELEX 0.167* 0.135*

DAX ISE NATIONAL 50 0.397* 0.320*VANGUARD 0.251* 0.199SOFIX 0.242* 0.179*CROBEX 0.430* 0.345*BELEX 0.153* 0.102*

ASE G.I. ISE NATIONAL 50 0.399* 0.327*VANGUARD 0.244* 0.201*SOFIX 0.191* 0.179*CROBEX 0.125* 0.105BELEX 0.183* 0.111*

This table reports unconditional correlations of equity index returns during crisis and stable periods. We estimate the uncon-ditional average correlations between each developed market and all other emerging markets. T-test statistics are of tests ofequivalent correlation with the developed country index return between the turmoil period and stable period. The rejectionof the null hypothesis against the one-sided alternative that during the turmoil period correlation increases is tested at the5% significance levels denoted by ‘*’. All of the indices are expressed in local currencies. The correspondence between stockmarkets and indices is: SOFIX: Bulgaria; VANGUARD: Romania; CROBEX: Croatia; ISE NATIONAL 50: Turkey; BELEX: Serbia.

5.3. Estimates of the AG-DCC model

Table 6 summarises information about the distribution of the unconditional correlations betweenthe each developed stock market and all others Balkan emerging markets during stable and crisisperiod. Results show that unconditional correlations increase significantly during the 2008 stockmarket crash period.


Table 7Conditional average correlations among equity index returns.

Developed markets Emerging markets 2008 stock marketcrash (26/09/2008to end of period)

Stable period(1/10/2002 to25/09/2008)

S&P ISE NATIONAL 50 0.503* 0.408*VANGUARD 0.298* 0.233*SOFIX 0.286* 0.199*CROBEX. 0.299* 0.256*BELEX 0.213* 0.156*

FTSE 100 ISE NATIONAL 50 0.499* 0.395*VANGUARD 0.283* 0.206SOFIX 0.253* 0.191CROBEX 0.279* 0.212*BELEX 0.197* 0.145*

DAX ISE NATIONAL 50 0.431* 0.389*VANGUARD 0.289* 0.232SOFIX 0.262* 0.199*CROBEX 0.498* 0.395*BELEX 0.203* 0.162*

ASE G.I. ISE NATIONAL 50 0.415* 0.376*VANGUARD 0.276* 0.226SOFIX 0.212* 0.187*CROBEX 0.265* 0.183BELEX 0.189* 0.131

This table reports conditional correlations during crisis and stable periods. Estimates are obtained using the AG-DCC model. Weestimate the conditional average correlations between each developed market and all other emerging markets. T-test statisticsare of tests of equivalent correlation with the developed country index return between the turmoil period and stable period. Therejection of the null hypothesis against the one-sided alternative that during the turmoil period conditional correlation increasesis tested at the 5% significance levels denoted by ‘*’. All of the indices are expressed in local currencies. The correspondencebetween stock markets and indices is: SOFIX: Bulgaria; VANGUARD: Romania; CROBEX: Croatia; ISE NATIONAL 50: Turkey;BELEX: Serbia.

Table 7 shows non-parametrically the presence of asymmetries in conditional second moments.Results indicate that average correlations between each developed market and all Balkan markets arehigher during the turbulent than the stable period. Moreover, conditional correlations are substantiallygreater than unconditional correlations, supporting the presence of asymmetric responses to negativeshocks. The higher correlations appear among US, UK, Germany and all Balkan countries, implying theimpact that the 2007–2009 financial crisis had to the regional equity integration. Overall, this paperprovides evidence that dependence is heightened among the Balkan and the developed equity markets,and supports the herding behaviour during the 2008 stock market crash period.

6. Concluding remarks

This study investigates long-run relationships among five Balkan emerging stock markets, the USand three developed European markets (UK, Germany and Greece) during the period 2000–2009.Johansen and Gregory–Hansen cointegration tests and Monte Carlo simulation provide evidence infavor of equity market integration within the region and globally. Moreover, the correlation dynam-ics obtained by the AG-DCC GARCH model show that dependence is heightened among Balkan anddeveloped equity markets, supporting the herding behaviour during the 2008 stock market crashperiod.

Results indicate that both domestic and external factors affect the Balkan stock markets, shapingtheir long-run equilibrium. The pivotal influence of “foreign factors” on the Balkan emerging stockmarkets could be attributed to their growing exposure to foreign capital flows, the financial liber-alization, the common regional economic prospects for full EU enrolment, the increased growth offoreign ownership and the trading behaviour of institutional investors in these markets during theturbulent period of 2008–2009.


The presence of long-run relationships suggests that the Balkan emerging markets have becomeincreasingly integrated with the world markets. This implies that international diversification bene-fits are quite limited in Balkan stock markets. Investors who diversify their portfolios across Balkanstock markets should expect rather short-run modest portfolios gains, given the volatile behaviourof portfolio returns to market shocks. Finally, the equity market integration in the Balkan region hassignificant impact on the domestic macroeconomic policies pursued, as emerging markets becomeless immunized to external shocks.

References

Baele, L., Ferrando, A., Hordahl, P., Krylova, E., Monnet, C., 2004. Measuring financial integration in the Euro area. Occasion Paper14, European Central Bank.

Baele, L., 2005. Volatility spillover effects in European equity markets. J. Finan. Quant. Anal. 40, 371–401.Bekaert, G., Harvey, C.R., Lumsdaine, R.L., 2002. Dating the integration of world equity markets. J. Finan. Econ. 65, 203–247.Campos, J., Ericcson, N.R., Hendry, D.F., 1996. Cointegration tests in the presence of structural breaks. J. Econ. 70, 187–220.Cappiello, L., Engle, R.H., Sheppard, K., 2006. Asymmetric dynamics in the correlations of global equity and bond returns. J. Finan.

Econ. 4, 537–572.Chaudhuri, K., Wu, Y., 2003. Random walk versus breaking trend in stock prices: evidence from emerging market. J. Bank. Finan.

27, 575–592.Chen, G., Firth, M., Rui, O., 2002. Stock market linkages: evidence from Latin America. J. Bank. Finan. 26, 1113–1141.Engle, R.F., 2002. Dynamic conditional correlation – a simple class of multivariate GARCH models. J. Bus. Econ. Stat. 20, 339–350.Engle, R.F., Granger, C.W., 1987. Cointegration and error correction: Representation, estimation and testing. Econometrica 55,

251–276.Francis, B.B., Leachman, L.L., 1998. Superexogeneity and the dynamic linkages among international equity markets. J. Int.

Monetary Fund 17, 475–494.Gregory, A.W., Hansen, B.E., 1996. Residual based tests for cointegration in the models with regime shifts. J. Econ. 70, 99–126.Johansen, S., 1988. Statistical analysis of cointegration vectors. J. Econ. Dyn. Control 12 (2), 231–254.Johansen, S., Juselius, K., 1990. Maximum likelihood estimation and inference on cointegration with application to the demand

for money. Oxford Bull. Econ. Stat. 52, 169–210.Kim, S.J., Moshirian, F., Wu, E., 2005. Dynamic stock market integration driven by the European Monetary Union: an empirical

analysis. J. Bank. Finan. 29, 2475–2502.Manning, N., 2002. Common trends and convergence? South East Asian equity markets 1998–1999. J. Int. Money Finan. 19,

183–202.Masih, A., Masih, R., 2002. Propagative causal price transmission among international stock markets: evidence from the pre-

and post-globalisation period. Global Finan. J. 13, 63–91.Osterwald-Lenum, M., 1992. A note with quartiles as the asymptotic distribution of the maximum likelihood cointegration rank

test statistics. Oxford Bull. Econ. Stat. 54 (3), 461–472.Phylaktis, K., 1999. Capital market integration in the pacific basin region: an impulse response analysis. J. Int. Money Finan. 18,

267–287.Samitas, A., Kenourgios, D., 2007. Macroeconomic factors’ influence on “new” European countries stock returns: the case of four

transition economies. Int. J. Finan. Serv. Manage. 2 (1–2), 34–49.Syriopoulos, Th., 2006. Risk and return implications from investing in emerging European stock markets. J. Int. Finan. Markets

Inst. Money 16, 283–299.Syriopoulos, Th., 2011. Financial integration and portfolio investments to emerging Balkan equity markets. J. Multinational

Finan. Manage. 21, 40–54.Voronkova, S., 2004. Equity market integration in Central European emerging markets: a cointegration analysis with shifting

regimes. Int. Rev. Finan. Anal. 12, 633–647.Zivot, E.D., Andrews, W.K., 1992. Further evidence on the great crash, the oil-price shock and the unit root hypothesis. J. Bus.

Econ. Stat. 10, 251–270.

Documents

Equity market integration in emerging Balkan markets