Int. Fin. Markets, Inst.

and Money 20 (2010) 363375

Contents lists available at ScienceDirect

Journal of International Financial Markets, Institutions & Money

j o ur na l ho me pa ge : w w w . e l s e v i e r . c o m / l o c a t e / i n t f i n

European capital market integration: An empirical study based on a European asset pricing model
David Morelli
Kent Business School, University of Kent, Canterbury, Kent CT2 7PE, UK

a r t i c l e

i n f o

a b s t r a c t
This paper investigates the integration between the capital markets of 15 European countries, all of which are members of the European Union. Integration is tested under the joint hypothesis of a European multifactor asset pricing model. A European portfolio is constructed from which common factors are extracted using maximum likelihood factor analysis. Empirical tests are undertaken to determine whether these European factors are not only priced, but also equally priced across the European capital markets. The results show that a number of common factors are extracted from the European portfolio and a degree of capital market integration is shown to exist across the European capital markets. 2010 Elsevier B.V. All rights reserved.

Article history: Received 4 December 2009 Accepted 25 March 2010 Available online 1 April 2010 JEL classication: G12 G15 Keywords: European capital markets Integration Factor analysis Pricing model

1. Introduction This paper examines whether the capital markets of the European countries that form the European Union are integrated. Over the years there has been a continuing process of integration within the European Union. Events such as the harmonisation of monetary and scal policy, none more so than the introduction of the Euro, have seen the capital markets of the countries of the European Union become more integrated. From the point of view of investors, looking to create international portfolios by investing in different European markets, so as to benet primarily from international diversication by reducing country specic systematic risk, greater capital market integration will reduce, and may even eventually remove such benets. Perfect integration across European capital markets would imply that these capital markets share the same riskreturn relationship, thus securities would be

E-mail address: D.A.Morelli@kent.ac.uk. 1042-4431/$ see front matter 2010 Elsevier B.V. All rights reserved. doi:10.1016/j.intn.2010.03.007

364

D. Morelli / Int. Fin. Markets, Inst. and Money 20 (2010) 363375

priced according to the same asset pricing model. To test for integration, in terms of examining the riskreturn relationship between countries, an asset pricing model is required. The pricing model adopted in this paper assumes that securities are priced according to a European multifactor asset pricing model. Assuming this, then integration across the European capital markets would imply that a securitys expected return should be directly related to its sensitivity to European risk factors. Different methodologies have been adopted in studying capital market integration. One such methodology is the use of multivariate cointegration techniques. Corhay et al. (1993) found evidence of one cointegrating vector on examining ve European capital markets. A study by Chung and Lui (1994) found two cointegration vectors on examining the capital markets of the US, Japan, Taiwan, Hong Kong, Singapore and South Korea. A more recent study by Chen et al. (2002) found evidence of one integrating vector on examining the South American capital markets of Argentina, Brazil, Chile, Columbia, Mexico and Venezuela. Pascal (2003) examining long-run comovements in the UK, French and German capital markets found no evidence of an increasing number of cointegrating vectors. Asset pricing models, both single and multifactor, have been applied so as to examine capital market integration. Single factor models such as the international CAPM examines whether security risk can be explained by the covariance of national returns with an international portfolio. The results from testing for integration using a single risk factor model have been somewhat mixed. Solnik (1977) found evidence of a degree of integration between US and European countries, and Stehle (1977) showed that the pricing of US securities was signicantly related to a global market portfolio. Studies by Stulz (1981) and Alder and Dumans (1983) provided evidence in support of an international CAPM, whereas Jorion and Schwartz (1985) however found little evidence of integration between the Canadian and US markets. Empirical studies to date adopting a multifactor asset pricing model to examine integration across various capital markets have also produced mixed results. Studies by Gultekin et al. (1989) examining the stock markets of the US and Japan, Korajczyk and Viallet (1989) examining the stock markets of the US, Japan, France and the UK, and Vo and Daly (2005) on examining the European equity markets, all failed to nd any strong evidence of integration across these markets. Studies however by Heston et al. (1995) on the capital markets of Europe and the USA, Cheng (1998) examining the capital markets of the UK and USA, and Swanson (2003) examining Japan, Germany and the USA, all produced evidence in support of integration across these markets. In this paper integration between the European capital markets is examined under the context of a European multifactor asset pricing model. Applying a European pricing model itself implies that the capital markets of Europe are integrated, thus the joint hypothesis problem exists. The application of a European multifactor asset pricing model assumes that returns follow a k-factor structure.1 The k-factor structure represents a number of common factors that explain the underlying correlations between security returns across different markets. Clearly, the greater the correlation the greater the integration. Various studies have examined correlations between different markets in an attempt to identify integration across global markets. Studies by Daly (2003) found, on examining the Asian markets, increased correlation after the stock market crash of 1997. Adjaoute and Danthine (2002) and Hardouvelis et al. (1999) found evidence of correlation between the European markets. This paper examines capital market integration across 15 European countries all of which form part of the European Union. The countries include: Austria, Belgium, Denmark, Finland, France, Germany, Greece, Ireland, Italy, Luxembourg, Portugal, Spain, Sweden, The Netherlands and the UK. The question of integration is examined by testing a European multifactor pricing model. The analysis involves extracting common factors from a European portfolio which is made up of a combined subsample of securities from each of the European countries. The technique of maximum likelihood factor analysis is adopted to extract common factors so as to determine the European factor structure. Once the factor structure is known, factor scores are subsequently estimated based on the methods of; Thurston (1935), Bartlett (1937) and Anderson and Rubin (1956), and are then adopted to test the validity of the European multifactor asset pricing model. Are the European factors priced, in the sense that there is a risk premium associated with them, and is this risk premium the same across all European countries.

1 Multifactor asset pricing models assume that the return on a security can be explained by common systematic risk factors (see the Arbitrage Pricing Theory of Ross (1976, 1977)).

D. Morelli / Int. Fin. Markets, Inst. and Money 20 (2010) 363375

365

The same risk premium would indicate the same riskreturn relationship for all European countries, implying full integration across the European capital markets. Countries of the European Union have over recent years become more integrated primarily as a result of economic and monetary union, and this paper contributes to the existing literature on capital market integration by examining the European correlation structure, so as to investigate pricing in order to test for integration across all 15 European countries. The main ndings of this paper are that common European factors do exist, some of which are priced, and priced equally across some of the European countries. Full integration is not found, in that the risk premium associated with the factors is not found to be the same for all countries, implying that diversication benets exist for investors looking to invest across the European capital markets. 2. Data This study uses monthly security returns over a 13-year period, January 1995December 2007. For each of the 15 European countries 100 securities are selected having continuous data over the total time period. Returns are calculated in US Dollars.2 To convert the returns in to excess returns, the one month US T-Bill rate is used to proxy for the risk-free rate. The need to select securities having continuous data is a requirement of factor analysis, given the need to calculate correlations which requires simultaneous observations. This clearly introduces a survival bias in to the sample as it excludes those companies that have merged, been taken over, failed and those that are new listings.3 The European portfolio adopted in this study comprises of a combined subsample of 25 randomly selected securities from each of the 15 European countries, thereby consisting of 375 securities. The return on the European portfolio is a value-weighted average of all 375 securities. The selection of an equal number of securities from each country ensures that no single country or group of countries dominate the European portfolio, reducing the problem of extracting country specic factors from the European portfolio. The data is obtained from Datastream. Table 1 reports the mean, standard deviation, skewness, kurtosis and KolmogorovSmirnov test for normality for the monthly returns for all 15 countries and the European portfolio over the total time period. It can be seen that Sweden and The Netherlands have the highest return whilst Greece and Portugal offer the lowest. Greece has the highest volatility as measured by the standard deviation and Austria the lowest. The volatility of the European portfolio is lower than any of the 15 European countries. The European portfolio can be seen as a more efcient portfolio for any risk averse investor compared to an investment in country portfolios; Austria, Finland, Greece, Ireland, Italy, Portugal and Spain, given that it offers a superior riskreturn trade off. The skewness statistic shows that for all countries apart from Germany and Sweden the returns tend to be positively skewed, and that the kurtosis levels are not high. The KolmogorovSmirnov test for normality clearly shows that for all countries, including the European portfolio, the returns do not depart from normality. One of the requirements in order to use factor analysis is for the security returns to be multivariate normally distributed. Maximum likelihood factor analysis can be adopted if the data is normally distributed, for the assumption regarding normality is required in order to apply signicance tests when attempting to determine the number of factors in the k-factor model. To test for multivariate normality is complex due to the innite number of linear combinations of variables for normality, however given that univariate normality is required for multivariate normality, one can test for the former. The requirement for normality does introduce an additional bias in the selection of securities given that those securities with extreme observations are excluded.4

2 Of the 15 European countries, as of yet only 3 have not adopted the Euro as their currency; Denmark, Sweden and United Kingdom, furthermore for the remaining countries the Euro was only introduced in 1999. The sample thus contains a degree of exchange rate risk. Due to different currencies throughout the time period of this study all returns are calculated in US Dollars. 3 Such a survival bias is common to all empirical studies adopting factor analysis. The greater the time period of the study the greater the bias. The time period of this paper extends over a period of 13 years and resultantly the survival bias is not as strong as those studies which extend over greater time periods. 4 Table 1 reports the results from the KolmogorovSmirnov test for normality for the average returns for each country over the total time period. Test are conducted on each security (due to the large sample size the results are not reported for each

366

D. Morelli / Int. Fin. Markets, Inst. and Money 20 (2010) 363375

Table 1 Summary statistics over the total time period. Mean Austria Belgium Denmark Finland France Germany Greece Ireland Italy Luxembourg Portugal Spain Sweden The Netherlands UK Europe 0.976 1.165 1.196 0.967 1.078 1.097 0.912 0.982 0.994 1.065 0.914 0.945 1.243 1.214 1.095 1.03 S.D. 5.143 5.781 5.821 6.835 7.214 6.821 7.983 6.023 7.832 5.724 6.012 5.945 6.876 5.723 6.893 4.472 Skewness 0.231 0.416 0.38 0.43 0.21 -0.18 0.43 0.34 0.278 0.243 0.40 0.372 -0.412 0.12 0.253 0.217 Kurtosis 0.33 1.73 1.42 1.31 0.72 0.41 1.31 0.64 0.79 0.53 1.21 0.97 1.19 0.35 0.71 1.15 KS 0.892 0.185 0.454 0.521 0.663 0.741 0.583 0.412 0.257 0.512 0.246 0.378 0.312 0.892 0.713 0.602

Summary statistics for each European country are based on a value-weighted portfolio return of all 100 securities in the sample, and for the European portfolio on a value-weighted average of the 375 securities.

The correlation between the returns of all 15 European countries and also the European portfolio is shown in Table 2. What is evident to see from Table 2 is that the European countries exhibit a degree of integration given that the correlations are far from zero, implying a linear relationship between these countries. Furthermore it can be seen that perfect correlation does not exist, implying that benets may exist for international diversication. 3. European multifactor asset pricing model As previously mentioned, testing for capital market integration across the European capital markets is performed under the joint hypothesis of a European asset pricing model. The multifactor European asset pricing model adopted and tested in this paper is given by: Rt = F t + t (1)

where Rt represents 1 n row vector of excess security returns at time t, n represents the number of securities, is a n k matrix of coefcients on the k-factors for each of the n securities, Ft is a 1 k row vector of common factors at time t generated from factor analysis, t is a n 1 column vector of idiosyncratic terms for each of the n securities at time t. The idiosyncratic terms are assumed to be independent of the factors, cov(Ft t ) = 0, and identically distributed as a joint multivariate normal distribution with mean zero E( t ) = 0, and covariance matrix D over time, cov(t t ) = 2 I = D. The covariance matrix D is assumed to be diagonal and proportional to the identity matrix, I. The factors as shown in Eq. (1) represent European factors extracted and estimated from a European portfolio adopting maximum likelihood factor analysis. Factor analysis simply involves attempting to extract a small number of common factors from a large number of interrelated variables, namely the excess security returns. The relationship between the excess security returns is shown by a correlation matrix and factor analysis explains this matrix using underlying common factors. A key advantage of adopting maximum likelihood analysis is that it allows the variance of the excess security returns to be separated out into their common and unique components resulting in the extraction of common factors. Maximum likelihood factor analysis not only provides a theoretical reasoning for the estimation process but also allows the use of statistical tests, namely the Chi-square goodness of t statistic,

security though are available upon request), and for each European country the securities selected comply with the assumption of normality.

D. Morelli / Int. Fin. Markets, Inst. and Money 20 (2010) 363375

Table 2 Correlation matrix between portfolio returns of European countriesa . Austria Belgium Denmark Finland France Germany Greece Ireland Italy Luxembourg Portugal Spain Sweden The Netherlands UK Europe
a

Belgium 0.301 0.231 0.287 0.319 0.327 0.216 0.187 0.325 0.213 0.276 0.219 0.308 0.317 0.275

Denmark

Finland

France

Germany

Greece

Ireland

Italy

Luxembourg

Portugal

Spain

Sweden

The Netherlands

UK

0.315 0.189 0.303 0.266 0.284 0.187 0.243 0.318 0.327 0.203 0.168 0.269 0.176 0.216 0.206

0.328 0.216 0.305 0.275 0.187 0.218 0.264 0.327 0.175 0.194 0.215 0.228 0.217

0.227 0.313 0.214 0.228 0.327 0.338 0.187 0.221 0.285 0.319 0.169 0.221

0.314 0.323 0.311 0.328 0.331 0.216 0.238 0.304 0.227 0.318 0.297

0.201 0.235 0.342 0.318 0.238 0.227 0.197 0.314 0.332 0.301

0.251 0.217 0.186 0.198 0.230 0.218 0.206 0.278 0.197

0.258 0.275 0.253 0.289 0.301 0.286 0.341 0.261

0.162 0.189 0.231 0.210 0.231 0.279 0.172

0.231 0.301 0.175 0.198 0.205 0.221

0.301 0.234 0.208 0.271 0.238

0.198 0.221 0.243 0.205

0.210 0.246 0.261

0.186 0.198

0.207

Portfolio returns for each European country is a value-weighted average of all 100 securities, and the European portfolio a value-weighted average of 375 securities (25 from each country).

367

368

D. Morelli / Int. Fin. Markets, Inst. and Money 20 (2010) 363375

to determine the number of factors to extract.5 Having extracted the common factors a structure then exists for the k-factor European pricing model. The use of factor analysis on large samples can cause problems given that a positive relationship exists between the number of factors extracted and the size of the sample. Extracting many factors would be of little use when testing for capital market integration as a large factor model would concentrate more on risk premia associated with country specic factors as opposed to risk premia associated with common factors between the European countries. To overcome this problem, a restriction is placed on the total number of factors that can be extracted from the European portfolio. The restriction is based upon the average number of factors extracted from each European country, and also on the eigenvalue of the average number of factors +1. The eigenvalue represents the total amount of variance of the excess security returns within each of the portfolios that is explained by the common factors extracted. The eigenvalue of the average number of factors + 1 is examined in order to determine whether the amount of additional variance of the excess security returns within the portfolio explained by this additional factor is signicant. In order to determine the average number of factors extracted from each European country, given the problems associated with using large samples in factor analysis, factor analysis is conducted on four randomly divided equal size portfolios of 25 securities for each European country. From each of these portfolios factors are extracted, and it is the average number of factors extracted from each of these portfolios, from each of the European countries, that determines the factor structure that will be applied to the European multifactor pricing model.6 Once the factor structure for the European portfolio has been determined, the factor scores are then estimated. Due to the indeterminacy problem associated with constructing factor scores, the factor scores are estimated according to three different criteria.7 The criteria are that the estimated factors and the true factors should display a high degree of correlation, they should also be univocal, and also orthogonal. Unfortunately, not one estimator satises all three criteria, resultantly three commonly adopted methods are used, namely; Anderson and Rubin (1956), Bartlett (1937) and Thurstons (1935) regression method, as shown by Eqs. (2), (3) and (4), respectively. F = R U2 B(B U2 SU2 B) F = R U2 B(B U2 B) F = R (S
1 1 1/2

(2) (3) (4)

B)

where F is the T k matrix of factor scores, R is a T n matrix of excess security returns for the European portfolio, n represents the number of variables, k the number of factors, B is a n k matrix of factor loadings, T is the time period, U is a n n diagonal matrix of unique variances, S is a n n sample correlation matrix of excess security returns.

5 Other methods of factor extraction exist such as, minimum residual factor analysis, image analysis, alpha factor analysis. These common factor analysis methods separate the common from unique variance of the variables and in that sense are similar to maximum likelihood analysis, however the method of determining the number of factors to extract is more subjective, unlike maximum likelihood analysis which adopts statistical tests. It is for this reason why maximum likelihood analysis is adopted in favour of these other methods of factor extraction. An alternative to these common factor analysis methods used to extract factor is principal component analysis. Principal component analysis is simply a mathematical transformation of the data. The factors extracted do not separate out the common from unique variance, and given that it is the common variance that is of interest, for this reason this method of factor extraction is not applied. 6 This approach is necessary given the problems with using large data samples when using factor analysis. Given that maximum likelihood analysis adopts the Chi-square goodness of t test statistic to determine the number of factors to extract, applying such a test to large sample can result in small discrepancies in t showing signicance, which in turn would result in a larger factor model. The European market portfolio consists of 375 securities, the application of factor analysis to such a large data sample would clearly result in such problems. This is overcome by restricting the numbers of factors extracted from the European portfolio based on the criteria discussed. 7 This indeterminacy problem exists because factor scores are not unique, primarily due to the fact that each excess security return contains a factor component and idiosyncratic (unique) component. Factor scores are constructed from a linear combination of the excess security returns, thus the factor scores will consist of two components; a deterministic linear combination of the excess security returns and a random vector orthogonal to the excess security returns.

D. Morelli / Int. Fin. Markets, Inst. and Money 20 (2010) 363375

369

4. Empirical tests The pricing model shown by Eq. (1) implies a linear pricing relationship between the expected excess returns and the k European factors. To test the validity of this pricing relationship, a timeseries regression is undertaken on a country by country basis of all individual security returns from each European country on the European factors extracted from the European portfolio. Rt = + F t + t (5)

The time-series regression results in estimates of a n 1 vector of s, a n k matrix of s, and D, a n n unbiased matrix of the covariance matrix of idiosyncratic terms. Having estimated s from time-series regression given by Eq. (5), for each country a cross-sectional regression is then performed of excess security return against the s as follows: R=X (6)

where R represents a n 1 vector of monthly excess security returns, n representing the number of securities, X represents a n (k + 1) matrix, the rst column being a vector of ones and the subsequent k columns a n 1 vector of s estimated from Eq. (5), is a (k + 1) 1 vector of risk premia estimated from a generalised least square regression, where = (X D1 X)1 X D1 R. Are the returns from individual European countries explained by the k-factor generating model? Various tests can be applied to test the validity of the pricing relationship shown by Eq. (6). With respect to the intercept term, given that excess security returns are used, this should equal zero. A simple t-test can be adopted to test, for each country, the hypothesis Ho : 0 = 0 against the alternative H1 : 1 = 0. / Furthermore, one can apply the exact F-test of Gibbons et al. (1989) to test the joint restriction that the intercept term is equal to zero across all the 15 European countries, which one would expect to nd if these markets were integrated. The Chi-square test is adopted for each European country to test if the vector of risk premia is statistically signicant.8 The hypothesis tested is simply, Ho : 1 = 2 = = k = 0 against the alternative H1 : 1 = 2 = = k = 0. With respect to the signicance of individual risk / / premia, the hypothesis tested is simply, Ho : 1 = 0, 2 = 0, . . . k = 0 against the alternative H1 : 1 = 0, / / 2 = 0 . . . k = 0. The hypothesis is tested using the simple t-test. Such tests establish whether the European pricing model is a valid pricing model, in that the tests determine whether the European risk factors price European countries security returns, which would imply integration across the European countries. This would not necessarily imply full integration between the European capital markets, as full integration requires the same price of risk across all the European countries. It is therefore necessary to test whether the risk premia for corresponding factors equate across all the European countries. This is tested across all European countries using a paired t-test. A paired t-test is performed between the time-series estimates of risk premia for corresponding factors between the groups of two countries. 5. Results As discussed in Section 3 the number of factors extracted from the European portfolio is determined by the average number of factors across the European countries and examination of the eigenvalue of the k + 1 factor extracted from the European portfolio. Table 3 reports, for each European country, the number of factors extracted from each of the 4 portfolios each consisting of 25 securities, in addition to the eigenvalue given as a percentage of the total factor model. It can be seen that for each country the number of factors extracted is not the same, implying that the k-factor return generating model is not the same across all the European countries. Some of the factors extracted will clearly be country specic factors, which inturn explains why the k-factor model is not unique across countries. The nding of country specic factors is expected, however given that country specic factors are not common to all countries, as the name suggests, they will most probably not be captured by the

8 The test statistic is T k W1 k = 2 where k is simply a vector of average risk premia, W represents a covariance matrix of time-series estimates of risk premia. The test statistic is 2 distributed with k degrees of freedom.

370

D. Morelli / Int. Fin. Markets, Inst. and Money 20 (2010) 363375

Table 3 Number of factors extracted from the European countries. Austria Portfolio 1 Portfolio 2 Portfolio 3 Portfolio 4 5 (52.6%) 5 (54.8%) 6 (55.2%) 5 (51.4%) Italy Portfolio 1 Portfolio 2 Portfolio 3 Portfolio 4 6 (58.3%) 6 (55.9%) 5 (57.2%) 7 (60.3%) Belgium 7 (58.6%) 6 (55.9%) 6 (52.1%) 7 (57.8%) Denmark 6 (60.3%) 5 (58.1%) 5 (62.1%) 4 (60.3%) Finland 5 (47.2%) 6 (45.6%) 6 (52.2%) 6 (50.1%) Spain 6 (43.2%) 7 (46.8%) 7 (50.7%) 6 (48.4%) France 5 (47.3%) 4 (45.6%) 6 (48.2%) 6 (53.1%) Germany 6 (61.2%) 5 (59.3%) 5 (63.6%) 6 (60.4%) Greece 5 (48.2%) 5 (50.6%) 6 (46.4%) 7 (53.4%) The Netherlands 7 (54.8%) 7 (51.9%) 6 (48.6%) 6 (46.2%) Ireland 6 (40.5%) 7 (43.7%) 7 (46.4%) 6 (44.7%) UK 7 (51.2%) 6 (54.7%) 7 (50.2%) 6 (56.7%)

Luxembourg 6 (53.4%) 7 (57.3%) 6 (51.9%) 7 (53.8%)

Portugal 6 (49.5%) 7 (47.2%) 7 (44.6%) 5 (43.8%)

Sweden 7 (47.5%) 7 (49.2%) 6 (44.7%) 6 (46.3%)

Table reports the number of factors extracted using maximum likelihood analysis from each of the 4 portfolios consist of 25 securities for each of the European countries. The eigenvalue, expressed as a percentage, of the factor model is also shown in parenthesis.

European portfolio.9 From Table 3 one can determine that the average number of factors extracted across all the European countries is six. In terms of examining the eigenvalue of the k + 1 factor, namely the 7th factor, maximum likelihood analysis is performed on the European portfolio and analysis of the eigenvalue of the 7th factor if found to show no statistical signicance. Resultantly, factor analysis is performed on the European portfolio where the number of factors extracted is restricted to six.10 For the European portfolio, the eigenvalue of the factor structure is 48.23%, thus almost half of the variance of the excess security returns that constitute the European portfolio can be explained by the six common factors. The nding of common factors imply that sources of common risk exist, however in order to be able to show that the European capital markets are integrated, these common sources of risk must be priced across the individual European countries, and priced equally to show full integration. Table 4 shows the results from the cross-section regression equation (6). The results show a strong similarity across all three methods of factor score estimation. Factor 1 shows signicance across Belgium, Denmark, Finland, France, Greece, Italy, Spain, Sweden, and the UK. Factor 2 is priced across Austria, Belgium, Germany, Ireland, Luxembourg, Portugal, Spain, The Netherlands and the UK. Factor 3 is priced across Austria and Germany and Factor 4 is priced across France and Sweden. Thus, for these factors for these countries the null hypothesis of no riskreturn relationship is rejected, implying priced European factors. What is evidently clear is that for all countries a minimum of at least one factor is priced, though not always the same factor. The Chi-square test shows that for, Belgium, Finland, Germany, Spain, Sweden and the UK, the null hypothesis that the risk premia vector is not statistically signicant is rejected. The cross-sectional regression across all the European countries results in factors one and two showing statistical signicance, along with a statistically signicant risk premia vector. The results also show that the Chi-square statistic does not change according to the method of factor score estimation (Thurston, 1935; Bartlett, 1937; Anderson and Rubin, 1956). Dependent upon the method used to estimate the factor scores, the amount of variance of the security returns explained by individual factors can change, however the total amount of variance explained by all the common factors does not change. The results with respect to testing the hypothesis relating to the intercept term show that for all European countries, with the exception of Germany (at the 10% signicance level), one fails to reject

9 Clearly the possibility does exist that country specic factors may be extracted from a European portfolio. The factors capture the correlations between the security returns, and if strong correlations exist between the returns within specic countries this may be captured by a factor. Such occurrences would be more probable with large factor structures, a problem that has been avoided in this study given the restrictions applied when determining the factor structure of the European portfolio. 10 Restricting the factor model to six factors is due to the problems associated with large samples. The eigenvalue of the K + 1 factor (7th factor) for the European portfolio is 1.16, which equates to only 0.31% of the variation in the returns of the European portfolio.

Table 4 Results from testing the signicance of the factor risk premia.
0 1 2 3 4 5 6 2

Austria aA b c

0.0032 (0.52) 0.0032 (0.52) 0.0032 (0.52)

0.0048 (0.61) 0.0058 (0.69) 0.0051 (0.54) 0.0413** (3.62) 0.0389** (3.91) 0.0408** (4.08) 0.0410** (3.49) 0.0362** (3.57) 0.0397** (3.31) 0.0304** (3.59) 0.0341** (3.41) 0.0301** (3.68) 0.0323** (3.43) 0.0338** (3.61) 0.0309** (3.17) 0.0046 (0.49) 0.0059 (0.44) 0.0051 (0.57) 0.0387** (3.51) 0.0361** (3.38) 0.0396** (3.55) 0.0071 (1.04) 0.0083 (0.95) 0.0079 (1.01) 0.0323** (3.29) 0.0301** (2.31) 0.0318** (3.22)

0.0338** (3.14) 0.0371** (2.92) 0.0354** (3.25) 0.0326** (3.49) 0.0308** (3.79) 0.0351** (3.62) 0.0068 (0.39) 0.0073 (0.32) 0.0087 (0.47) 0.0045 (0.51) 0.0053 (0.43) 0.0038 (0.41) 0.0095 (0.96) 0.0084 (0.81) 0.0108 (1.13) 0.0375** (3.96) 0.0396** (4.15) 0.0352** (3.74) 0.0051 (0.57) 0.0063 (0.52) 0.0045 (0.62) 0.0305** (3.14) 0.0326** (3.07) 0.0289** (2.87) 0.0049 (0.64) 0.0055 (0.71) 0.0041 (0.59)

0.0025** (0.34) 0.0029** (0.28) 0.0021** (0.39) 0.0049 (0.57) 0.0044 (0.61) 0.0057 (0.62) 0.0031 (0.37) 0.0029 (0.41) 0.0042 (0.34) 0.0032 (0.29) 0.0051 (0.33) 0.0044 (0.28) 0.0026 (0.42) 0.0041 (0.39) 0.0029 (0.18) 0.0217** (2.51) 0.0237** (2.69) 0.0221** (2.27) 0.0041 (0.52) 0.0057 (0.59) 0.0046 (0.51) 0.0093 (1.21) 0.0091 (1.24) 0.0083 (0.96) 0.0074 (0.81) 0.0068 (0.72) 0.0087 (0.92)

0.0064 (0.71) 0.0076 (0.75) 0.0061 (0.68) 0.0084 (0.76) 0.0091 (0.71) 0.0076 (0.62) 0.0121 (1.19) 0.0093 (0.85) 0.0105 (1.13) 0.0042 (0.51) 0.0058 (0.68) 0.0053 (0.61) 0.0224** (2.21) 0.0241** (2.35) 0.0208** (2.08) 0.0095 (1.04) 0.0073 (0.86) 0.0091 (0.99) 0.0069 (0.86) 0.0078 (0.89) 0.0073 (0.76) 0.0051 (0.64) 0.0059 (0.69) 0.0042 (0.52) 0.0042 (0.62) 0.0038 (0.61) 0.0051 (0.73)

0.0025 (0.33) 0.0029 (0.30) 0.0031 (0.27) 0.0039 (0.39) 0.0046 (0.33) 0.0031 (0.42) 0.0063 (0.52) 0.0072 (0.70) 0.0084 (0.67) 0.0071 (0.82) 0.0054 (0.68) 0.0063 (0.66) 0.0046 (0.52) 0.0061 (0.63) 0.0042 (0.55) 0.0084 (0.93) 0.0077 (0.91) 0.0081 (1.03) 0.0043 (0.49) 0.0049 (0.54) 0.0052 (0.53) 0.0059 (0.77) 0.0052 (0.79) 0.0063 (0.83) 0.0086 (0.92) 0.0092 (1.06) 0.0073 (0.88)

0.0058 (0.72) 0.0061 (0.65) 0.0053 (0.71) 0.0027 (0.31) 0.0022 (0.39) 0.0038 (0.41) 0.0021 (0.19) 0.0037 (0.25) 0.0029 (0.23) 0.0019 (0.28) 0.0031 (0.36) 0.0028 (0.41) 0.0019 (0.25) 0.0.0024 (0.19) 0.0016 (0.18) 0.0031 (0.43) 0.0035 (0.40) 0.0027 (0.37) 0.0019 (0.21) 0.0031 (0.29) 0.0026 (0.37) 0.0042 (0.31) 0.0033 (0.26) 0.0038 (0.34) 0.0023 (0.31) 0.0021 (0.30) 0.0017 (0.22)

8.04 8.04 8.04 13.61** 13.61** 13.61** 9.57 9.57 9.57 12.79** 12.79** 12.79** 10.31 10.31 10.31 14.02** 14.02** 14.02** 8.92 8.92 8.92 7.12 7.12 7.12 9.26 9.26 9.26

Belgium a 0.0025 (0.41) b 0.0025 (0.41) c 0.0025 (0.41) Denmark a 0.0028 (0.73) b 0.0028 (0.73) c 0.0028 (0.73) Finland a b c France a b c 0.0053 (0.64) 0.0053 (0.64) 0.0053 (0.64) 0.0042 (0.92) 0.0042 (0.92) 0.0042 (0.92)

D. Morelli / Int. Fin. Markets, Inst. and Money 20 (2010) 363375

Germany a 0.0098 (1.68)* b 0.0098 (1.68)* c 0.0098 (1.68)* Greece a b c 0.0061 (0.93) 0.0061 (0.93) 0.0061 (0.93)

Ireland a 0.0045 (0.59) b 0.0045 (0.59) c 0.0045 (0.59) Italy a1 b c 0.0044 (0.78) 0.0044 (0.78) 0.0044 (0.78)

371

372

Table 4 Continued
0 1 2 3 4 5 6 2

Luxembourg a 0.0059 (0.92) b 0.0059 (0.92) c 0.0059 (0.92) Portugal a 0.0028 (0.49) b 0.0028 (0.49) c 0.0028 (0.49) Spain a b c 0.0088 (1.24) 0.0088 (1.24) 0.0088 (1.24)

0.0064 (0.59) 0.0077 (0.69) 0.0061 (0.52) 0.0093 (1.31) 0.0089 (1.22) 0.0096 (1.29) 0.0312** (3.37) 0.0336** (3.59) 0.0341** (3.54) 0.0363** (3.48) 0.0345** (3.31) 0.0335** (3.23) 0.0086 (0.96) 0.0093 (1.17) 0.0078 (0.98) 0.0373** (3.69) 0.0369** (3.78) 0.0395** (4.08) 0.0351** (3.47) 0.0331** (3.18) 0.0319** (2.93)

0.0275** (2.67) 0.0296** (2.73) 0.0271** (2.81) 0.0303** (3.43) 0.0321** (3.36) 0.0313** (3.51) 0.0276** (2.95) 0.0285** (3.01) 0.0267** (2.88) 0.0087 (0.93) 0.0081 (0.88) 0.0096 (1.06) 0.0375** (3.86) 0.0391** (4.13) 0.0382** (3.78) 0.0248** (2.68) 0.0285** (3.96) 0.0277** (2.61) 0.0276** (2.11) 0.0284** (2.17) 0.0268** (2.05)

0.0049 (0.61) 0.0055 (0.64) 0.0048 (0.54) 0.0019 (0.25) 0.0022 (0.28) 0.0016 (0.23) 0.0165 (1.59) 0.0153 (1.43) 0.0161 (1.48) 0.0121 (1.24) 0.0132 (1.51) 0.0109 (1.30) 0.0076 (0.88) 0.0081 (0.97) 0.006 (0.74) 0.0053 (0.59) 0.0047 (0.61) 0.0038 (0.46) 0.0048 (0.67) 0.0057 (0.81) 0.0051 (0.75)

0.0045 (0.65) 0.0056 (0.67) 0.0051 (0.58) 0.0047 (0.58) 0.0055 (0.67) 0.0041 (0.54) 0.0051 (0.67) 0.0043 (0.61) 0.0062 (0.85) 0.0152** (1.98) 0.0141** (1.79) 0.0161** (2.07) 0.0109 (1.21) 0.0124 (1.37) 0.0096 (0.99) 0.0039 (0.51) 0.0046 (0.48) 0.0052 (0.59) 0.0055 (0.81) 0.0047 (0.74) 0.0059 (0.80)

0.0032 (0.42) 0.0041 (0.49) 0.0048 (0.44) 0.0075 (0.96) 0.0067 (0.88) 0.0082 (0.98) 0.0128 (1.43) 0.0108 (1.29) 0.0132 (1.44) 0.0042 (0.75) 0.0061 (0.87) 0.0047 (0.72) 0.0086 (1.01) 0.0098 (1.27) 0.0083 (1.11) 0.0056 (0.68) 0.0043 (0.55) 0.0052 (0.71) 0.0108 (1.46) 0.0115 (1.41) 0.0095 (1.17)

0.0028 (0.36) 0.0039 (0.44) 0.0047 (0.51) 0.0018 (0.26) 0.0029 (0.38) 0.0016 (0.27) 0.0042 (0.59) 0.0046 (0.69) 0.0037 (0.51) 0.0039 (0.58) 0.0025 (0.36) 0.0033 (0.43) 0.0025 (0.39) 0.0037 (0.46) 0.0031 (0.38) 0.0017 (0.29) 0.0026 (0.38) 0.0016 (0.30) 0.0032 (0.49) 0.0024 (0.41) 0.0028 (0.42)

7.84 7.84 7.84 8.67 8.67 8.67 13.14** 13.14** 13.14** 11.21* 11.21* 11.21* 7.12 7.12 7.12 14.21** 14.21** 14.21** 10.91* 10.91* 10.91*

D. Morelli / Int. Fin. Markets, Inst. and Money 20 (2010) 363375

Sweden a 0.0071 (0.94) b 0.0071 (0.94) c 0.0071 (0.94) The Netherlands a 0.0049 (0.58) b 0.0049 (0.58) c 0.0049 (0.58) UK a b c 0.0065 (1.05) 0.0065 (1.05) 0.0065 (1.05)

All European countries a 0.0048 (0.68) b 0.0048 (0.68) c 0.0048 (0.68)

For each country the average intercept and risk premium is reported with corresponding t-statistic shown in parenthesis. The 2 -test statistic testing the null hypothesis of whether the vector of risk premia is equal to zero is also shown. The critical 2 -value with 6 d.f. is 12.592 and 10.645 at the 5% and 10% level respectively. A a, b, c represents the Anderson-Rubin, Bartlett and Thurston factor score estimation, respectively.
* **

Signicance at 10% level. Signicance at 5% level.

D. Morelli / Int. Fin. Markets, Inst. and Money 20 (2010) 363375 Table 5 European countries having the same price of risk. Risk premia Price of risk the same across all seven countries Countries showing the same price of risk

373

Factor scores estimated using Anderson-Rubin

1

Bartlett Belgium and German, Finland and France and Spain and Sweden, Greece and UK Austria and Germany and The Netherlands, Ireland and Portugal, Spain and UK Austria and Denmark, Spain and Sweden, Belgium and France and UK Belgium and Denmark, Austria and Germany, Finland, and Luxembourg and Portugal Denmark and Germany and Portugal, Belgium and Greece and Ireland and Portugal and UK Belgium and France and Italy and Sweden and UK

Thurston Belgium and Denmark and Greece and Spain and Sweden, German and Austria Austria and Belgium and Germany, Italy and Finland, Ireland and Luxembourg and Spain and UK Denmark and Finland and Greece, Ireland and Italy, Greece and Luxemburg Belgium and Greece, Denmark and Italy and Luxembourg and UK Austria and Belgium, Finland and Ireland, Luxembourg and Sweden and UK Denmark and Germany and Ireland andThe Netherlands

No

Belgium and Denmark and Greece, Italy and France and Spain Austria and Belgium and The Netherlands, Finland and Italy, Ireland and Portugal Denmark and Finland, Belgium and Greece and Luxembourg and UK Finland and Italy and Luxembourg and Portugal and UK France and Greece and Sweden, German and Italy and The Netherlands Denmark and Finland and France and Greece and Italy and Portugal and UK

No

No

No

No

No

The table reports those European countries in which the price of risk (risk premia) is found to be the same. Results are shown for all three methods of factor score estimation; Anderson-Rubin, Bartlett or Thurstons methodology.

the null hypothesis that 0 = 0. In terms of testing the joint restriction that the intercept term across all countries are zero, application of the exact F-test of Gibbons et al. (1989) produces a F-statistic of 1.03, thus failing to reject the null hypothesis that the intercept term is zero across all countries.11 Despite that fact that some of the factors are priced, and for some countries the vector of risk premia is statistically signicant, the European pricing model is not found to be a valid model across all the European countries. This is the case irrespective of whether the Anderson and Rubin (1956), Bartlett (1937) and Thurstons (1935) regression method is used to estimate the factor scores. The results from Table 4 show that for 9 out of the 15 countries the European multifactor asset pricing model does not hold, thus integration across all the 15 European countries is not evident to see. On performing cross-sectional tests across all European countries, rather than individually, the null hypothesis of an insignicant risk premia vector is rejected, from which one can conclude a degree of European market integration. However the rejection of the null hypothesis of an insignicant risk premia vector is inuenced by the strong cross-sectional results for Belgium, Finland, Germany, Spain, Sweden and the UK. In terms of determining whether the price of risk is the same across all 15 European countries, summarised results are reported in Table 5. It is clear to see that some factors have the same risk premia across a number of countries, implying a degree of integration between these capital markets. From analysing Factor 1, given that this out of all the factors is the most important factor, as it explains the largest proportion of the total variance of the European portfolio, it can be seen that a number of European countries do have the same price of risk. Although it is found that a degree of integration is

11 The coefcients of the intercepts are not sensitive to the method of factor score estimations. The method of factor score estimation only effects the proportion of variance of the variables explained by the factors and as a result will not inuence the intercept term.

374

D. Morelli / Int. Fin. Markets, Inst. and Money 20 (2010) 363375

evident across some of the European countries, given the same price of risk, this is clearly not the case across all countries. Such ndings indicate the absence of full European capital market integration. The results are not surprising given that capital market integration can only be tested under the joint hypothesis of an European asset pricing model, which clearly does not hold across all 15 European countries. 6. Conclusion Countries of the European Union have, over the years, become more integrated as a result of growth in international trade, in services and nancial assets, and for a number of countries as a result of monetary union. Capital market integration implies that individual capital markets move in a similar way, and as a result of this have high correlations, which in turn implies reduced benets from international portfolio diversication. This paper examines the covariance structure of a European portfolio made up from a subset of securities from each of the 15 European countries, in an attempt to determine whether these capital markets are integrated. Integration of the capital markets is tested under the joint hypothesis of a European multifactor asset pricing model. Results show that common factors exist, some of which are priced. For some countries the price of risk is found to be the same, implying that their capital markets are closely related, however this is not the case for all 15 European countries of the European Union. A degree of integration is found to exist between some European countries, however based on the empirical analysis from testing a European multifactor asset pricing model it is evident that the hypothesis of full integration across all the European countries is not shown to hold. Such ndings imply that to an international portfolio investor there are benets of diversication from investing across the European capital markets, though between some of the European countries this benet has been slightly reduced due to the integration of the these capital markets. References
Adjaoute, K., Danthine, J.P., 2002. European nancial integration and equity returns: a theory based assessment. In: Gaspar, V., Hartmann, P., Sleijpen, O. (Eds.), The Transformation of the European Financial System. European Central Bank, Frankfurt, pp. 185245. Alder, M., Dumans, B., 1983. International portfolio selection and corporation nance: a synthesis. Journal of Finance 38, 925984. Anderson, T.W., Rubin, H., 1956. Statistical inference in factor analysis. In: Proceedings of the 3rd Berkeley Symposium on Mathematical Statistics and Probabilities, vol. 5, pp. 111150. Bartlett, M., 1937. The statistical conception of mental factors. British Journal of Psychology 28, 97104. Chen, G.M., Firth, M., Meng Rui, O., 2002. Stock market linkages: evidence from Latin America. Journal of Banking and Finance 26, 11131141. Cheng, A.C.S., 1998. International correlation structure of nancial market movementthe evidence from the UK and US. Applied Financial Economics 8, 112. Chung, P.J., Lui, D.J., 1994. Common stochastic trends in Pacic Rim stock markets. Quarterly Review of Economics and Finance 34, 241259. Corhay, A., Tourani Rad, A., Urbain, J.P., 1993. Common stochastic trends in European stock markets. Economics Letters 42, 385390. Daly, K.J., 2003. Southeast Asian stock market linkages, evidence from pre-and-post October. ASEAN Economic Bulletin 20 (1), 7385. Gibbons, M.R., Ross, S.A., Shanken, J., 1989. A test of the efcient of a given portfolio. Econometrica 57, 11211152. Gultekin, N., Bulent, M.N., Penati, A., 1989. Capital controls and international capital market segmentation: the evidence from Japanese and American stock markets. Journal of Finance 44, 849869. Hardouvelis, G.A., Malliaropulos, D., Priestley, R., 1999. EMU and European stock market integration. CEPR Discussion Paper Series No. 2124. London. Heston, L.H., Rouwenhorst, K.G., Wessels, R.E., 1995. The structure of international stock returns and the integration of capital markets. Journal of Empirical Finance 2, 173197. Jorion, P., Schwartz, E., 1985. Integration versus segmentation in the Canadian stock market. Journal of Finance 41, 603 614. Korajczyk, R.A., Viallet, C.J., 1989. An empirical investigation of international asset pricing. Review of Financial Studies 2, 553585. Pascal, A.G., 2003. Assessing European stock markets cointegration. Economic Letters 78, 197203. Ross, R., 1976. The arbitrage theory of capital asset pricing. Journal of Economic Theory 13, 341360. Ross, R., 1977. Return risk and arbitrage. In: Friend, I., Bicksler, J. (Eds.), Risk and Return in Finance. Cambridge, Ballinger. Solnik, B., 1977. Testing international asset pricing: some pessimistic views. Journal of Finance 32, 500524.

D. Morelli / Int. Fin. Markets, Inst. and Money 20 (2010) 363375

375

Stehle, R., 1977. An empirical test of the alternative hypothesis of national and international pricing of risky assets. Journal of Finance 32, 493502. Stulz, R., 1981. A model of International asset pricing. Journal of Financial Economics 9, 383406. Swanson, P.E., 2003. The interrelatedness of Global equity markets, money markets, and foreign exchange markets. International Review of Financial Analysis 12, 135155. Thurston, L, 1935. The Vectors of Mind. University of Chicago Press, Chicago. Vo, X.V., Daly, K.J., 2005. European equity market integrationimplications for US investors. Research in International Business and Finance 19, 155170.