Vous êtes sur la page 1sur 16

This article was downloaded by: [University of Liverpool] On: 16 January 2013, At: 22:04 Publisher: Routledge Informa

Ltd Registered in England and Wales Registered Number: 1072954 Registered office: Mortimer House, 37-41 Mortimer Street, London W1T 3JH, UK

Applied Financial Economics


Publication details, including instructions for authors and subscription information: http://www.tandfonline.com/loi/rafe20

Capital market integration: evidence from the G7 countries


David Morelli
a a

Kent Business School, University of Kent, Canterbury, Kent, CT2 7PE, UK Version of record first published: 15 Jun 2009.

To cite this article: David Morelli (2009): Capital market integration: evidence from the G7 countries, Applied Financial Economics, 19:13, 1043-1057 To link to this article: http://dx.doi.org/10.1080/09603100802167262

PLEASE SCROLL DOWN FOR ARTICLE Full terms and conditions of use: http://www.tandfonline.com/page/terms-and-conditions This article may be used for research, teaching, and private study purposes. Any substantial or systematic reproduction, redistribution, reselling, loan, sub-licensing, systematic supply, or distribution in any form to anyone is expressly forbidden. The publisher does not give any warranty express or implied or make any representation that the contents will be complete or accurate or up to date. The accuracy of any instructions, formulae, and drug doses should be independently verified with primary sources. The publisher shall not be liable for any loss, actions, claims, proceedings, demand, or costs or damages whatsoever or howsoever caused arising directly or indirectly in connection with or arising out of the use of this material.

Applied Financial Economics, 2009, 19, 10431057

Capital market integration: evidence from the G7 countries


David Morelli
Kent Business School, University of Kent, Canterbury, Kent, CT2 7PE, UK E-mail: D.A.Morelli@kent.ac.uk

Downloaded by [University of Liverpool] at 22:04 16 January 2013

This article examines whether the capital markets of the G7 countries are integrated. Capital market integration is examined under the joint hypothesis of an international multifactor asset pricing model. International factors are extracted from a world portfolio using both maximum likelihood analysis and principal component analysis. Results show that international common factors exist, some of which are priced and equal across some countries, however, the international pricing model does not hold for all G7 countries. The price of risk is not found to be the same across all countries and the hypothesis of full capital market integration is not supported.

I. Introduction The extent of integration between the world capital markets is clearly of importance in the finance world with respect to investment selection and financing decision. If the world capital markets are perfectly integrated then the same asset pricing relationship would exist for all countries, with the reward for risk being the same irrespective of which market one invests in. The absence of integration would imply that the risk return relationship differs across countries, which would lead to arbitrage opportunities in that investors could simply adjust their portfolio by investing in countries offering a greater return whilst maintaining the same level of risk. If the capital markets of different countries are integrated, the expected return of a security or portfolio of a particular country should be determined solely by its exposure to the worlds risk factor or factors, depending on whether one assumes a single or multifactor pricing model. Failure to show this would indicate that the relationship between risk

and return is explained by domestic and not worldwide factors. The existence of nonintegration across financial markets is most likely to be due to factors such as, market imperfections, the existence of differing rates of taxation or restrictions imposed by the markets or countries on the ownership of securities (Eun, 1985; Eun and Janakiramanan, 1986). Studies by Divecha et al. (1992), Michaud et al. (1996) and De Fusco et al. (1996) found the lack of integration across international markets was due primarily to barriers to international trade and investment, insufficient information on foreign securities and simply a bias by investors to home securities. Various studies have been conducted to test for international integration across various financial markets. Early studies focused on a single risk factor as a proxy for the market portfolio in an international capital asset pricing model. Solnik (1974) found evidence in support of integration between various European countries and the United States. Jorion and Schwartz (1986) on examining the

Applied Financial Economics ISSN 09603107 print/ISSN 14664305 online 2009 Taylor & Francis http://www.informaworld.com DOI: 10.1080/09603100802167262

1043

1044
integration of the Canadian stock market relative to a global North American market failed to find evidence of integration. Campbell and Hamao (1992) found evidence of common movement in expected excess returns across the United States and Japan. Chou and Lin (2002) found evidence in support of an international pricing model across 17 developed countries, which included all those within the G7. Empirical studies have also employed cointegration techniques to examine the interdependence between the world stock markets. Byers and Peel (1993) tested for multivariate cointegration between the stock markets of the US, UK, Japan, Germany and the Netherlands, and found, with the exception of the UK and Japan, no evidence to suggest that the international stock markets were cointegrated. Kanas (1998) found that the US stock market was not pairwise cointegrated with any of the major European stock markets. Empirical studies also focused on multifactor asset pricing models. Evidence against integration was found by Gultekin et al. (1989) examining the USA and Japanese stock markets, and also Korajczyk and Viallet (1989) examining the capital markets of the United States, Japan, France and the UK. Evidence supporting the hypothesis of capital market integration was found by Heston et al. (1995) examining the capital markets of Europe and the USA, Cheng (1998) examining the UK and US stock markets and also Swanson (2003) examining three major financial markets, namely Japan, Germany and the USA. A recent study by Vo and Daly (2005) found little evidence of integration between European equity markets and concluded that diversification benefits within Europe exist for US investors. This article examines whether the capital markets of the G7 countries, namely, Canada, France, Germany, Italy, Japan, the UK and the USA are integrated. Empirical tests of integration require an international asset pricing model. The use of an international asset pricing model assumes that the capital markets are integrated, for if they were not integrated the pricing model would not hold. Thus, the question of capital market integration is tested under the joint hypothesis of an international asset

D. Morelli
pricing model. In the empirical implementations in this article it is assumed that returns follow a k-factor structure, thus a multifactor international asset pricing model is adopted.1 Pricing securities on the basis of an international multifactor pricing models implies that the only priced risk should be the systematic risk relative to the world factors. With an international pricing model domestic systematic factors should be diversified away. The extent to which countries of the G7 share common factors is examined by extracting factors from a world portfolio consisting of a combined subsample of securities from each of the G7 countries. A global factor structure is obtained using two well-known methods of factor extraction, namely, principal component analysis and maximum likelihood analysis. Factor scores are then constructed using three commonly used methods: Thurston (1935) regression method, Bartlett (1937) and AndersonRubin (1956). The factor scores can then be used as proxies for the factors and used in subsequent tests to determine whether the international multifactor asset pricing model holds. Do the factors command a risk premium? Is the risk premium equal across countries thereby implying that the reward for risk is the same irrespective of which country one invests in? Such a finding is essential in order to show that the G7 capital markets are fully integrated.2 This article explicitly differentiates between security returns being correlated internationally, and financial integration. High international correlations are at least as much about correlated fundamentals as they are about integration, and this article investigates pricing in response to these putative correlated international factors. This article contributes to the existing literature on capital market integration, and as far as I am aware that the methodology adopted in this article has not been attempted in any of the existing literature. This article is organized as follows. The data and corresponding statistics is discussed in Section II. Section III discusses the international asset pricing model. Section IV discusses the two methods used to

Downloaded by [University of Liverpool] at 22:04 16 January 2013

Asset pricing models explain the relationships between security returns and a common factor or factors. Asset pricing models, whether single or multifactor, are based on the notion that security returns can be explained by systematic risk factors. The most well-known multifactor asset pricing model being the Arbitrage Pricing Theory of Ross (1976, 1977). 2 Market integration could be determined by examining the returns on two portfolios of securities from two different countries that are perfectly correlated. If perfect market integration exists the price of such securities should be exactly the same, since any disequilibrium in the price would lead to arbitrage opportunities upon which equilibrium would be quickly restored. Theoretically this makes perfect sense; however, in practice it is almost impossible to construct these perfectly correlated portfolios and thus virtually impossible to test.

Capital market integration


Table 1. Summary statisticsa Countryb Canada France Germany Italy Japan UK USA World Mean (%) 0.794 0.918 0.942 1.042 1.101 1.117 1.081 0.961 SD (%) 5.103 7.302 7.930 8.932 6.619 7.036 5.729 4.892 Skewness 0.138 0.108 0.082 0.207 0.361 0.219 0.349 0.131 Kurtosis 1.12 0.87 1.33 0.71 1.16 0.68 0.95 1.07 KSc 0.476 0.694 0.401 0.732 0.431 0.703 0.657 0.491

1045

Notes: aStatistics provided over the total sample period. b Statistics for each country is based on a value weighted average of all 160 securities, and for the world portfolio a value weighted average of 350 securities (50 from each country). c p-value of the KolmogorovSmirnov test for normality of returns.

Downloaded by [University of Liverpool] at 22:04 16 January 2013

extract factors, maximum likelihood and principal component analysis. Section V discusses the methods by which the factor scores are estimated. Section VI discusses the empirical tests. Empirical results are presented in Section VII, and the conclusion is presented in Section VIII.

II. Data The data is collected from Datastream and consists of monthly security returns from each of the G7 countries over the period January 1990 to December 2000.3 A total of 160 securities for each country were selected with data covering the total period.4 All returns are calculated in terms of US dollars, and the monthly return on a 3-month US Treasury Bill is used as the risk free asset. The world portfolio used in this article consists of a combined subsample of 50 randomly selected securities from each of the seven countries (thus a value weighted average of 350 securities).5

Table 1 shows the monthly mean percentage return for each country in addition to the world portfolio calculated in terms of US dollars, along with the SD, skewness, kurtosis and the Kolmogorov Smirnov test for normality. The mean monthly returns range from 0.794% for Canada to 1.117% for the UK. With respect to the SD, which is a measurement of volatility, its value ranges from 4.892% for the world portfolio to 8.932% for Italy. Given that the world portfolio has the lowest SD this clearly shows the benefit of risk reduction by diversifying away unsystematic risk. With respect to examining the risk return relationship it is shown that the world portfolio offers a higher return than Canada, France and Germany for a lower SD, thus a rational investor would be advised to invest in the world portfolio as apposed to either of these three national markets given the better risk return relationship. With respect to skewness, with the exception of France and the USA, all other countries in addition to the world portfolio are positively skewed.

The use of monthly returns, as opposes to say daily, avoids the problems associated with thin trading, primarily causing biases when estimating the correlation matrices from which the factors are then extracted. 4 The use of factor analysis requires the sample selected to have simultaneous observations given that this is required to calculate the correlations. Given this requirement, only securities that have continuous data over the total period are selected. This naturally introduces a survival bias into the sample given that a number of firms will be excluded, for example, companies that have failed, or are newly listed, or those that have simply merged or been taken over. Such a survival bias is common to all empirical tests requiring the use of factor analysis, and increases with the length of the sample period. The sample size, 160 securities from each of the G7 countries, consists of securities from a number of different industry groups, thus representing a fair distribution of industries and should not be considered a sector specific sample. It is important for a reasonable number of securities to be contained in the sample. The sample size adopted in this article is believed to satisfy this condition. 5 Given the need to extract factors from the correlation matrix of security returns that constitutes the market portfolio, and also the need to estimate factor scores, the market portfolio adopted cannot consist of an index, but must consist of a sample of securities from all countries. The market portfolio consists of a subsample of securities from each country, for if the market portfolio had consisted of all securities, this would have resulted in a 1120 1120 correlation matrix (160 securities 7 countries), well beyond the software capabilities with respect to computing factor loadings.

1046
Table 2. Correlation matrix between all G7 countries portfolios and also the world portfolioa Canada France Germany Italy Japan UK USA World 0.201 0.354 0.168 0.182 0.254 0.337 0.244 France 0.374 0.291 0.206 0.282 0.261 0.287 Germany Italy Japan UK

D. Morelli

USA

0.363 0.154 0.308 0.274 0.271

0.148 0.325 0.334 0.257

0.176 0.263 0.221

0.392 0.284

0.331

Note: aSee footnote b from Table 1.

Downloaded by [University of Liverpool] at 22:04 16 January 2013

Kurtosis levels are not large, and the Kolmogorov Smirnov test for normality is not rejected for all G7 countries and the world portfolio.6 Table 2 shows the correlation matrix between all seven countries and also the world portfolio. Two important inferences can be made from Table 2, first that linear relationships exist between the G7 countries given that the correlations are far from zero, and secondly given that the correlations are also far from unity, the possibility exists for international diversification.7

Or in matrix notation: Rt Ft "t


0

III. International Asset Pricing Model Testing for integration across the G7 capital markets requires an international asset pricing model, given that one needs to determine whether the price of risk is the same across all countries. A valid international asset pricing model itself implies integration across the international capital markets. In this article, the international asset pricing model used is a multifactor model given by: Rit i1 F1t i2 F2t ik Fkt "it
6

where Rt is a 1 n row vector for n excess security returns at time t, Ft is a 1 k row vector of observations on the common factors at time t generated by using factor analysis, B is an n k matrix of coefficients or loadings on the k factors for each of the n securities, "t is an n 1 column vector of idiosyncratic terms for each security at time t. The return generating process is composed of two components, a common component and an idiosyncratic one. It is assumed that the idiosyncratic terms are independent and identically distributed as a joint multivariate normal distribution with mean zero E("t) 0, and covariance matrix D over time, cov("t"t0 )  2I D, which is diagonal and proportional to the identity matrix.8 In addition, it is assumed that the idiosyncratic terms and the factors are independent of each other, cov(Ft"t) 0. The assumption relating to the covariance matrix D of idiosyncratic terms implies that the idiosyncratic variances equate to one another thereby allowing the use of principal component analysis to estimate the return generating factor model shown by Equation 2. This assumption is not required when using maximum likelihood analysis, thus the

Factor analysis requires multivariate normal distribution of the security returns. Maximum likelihood factor analysis can be used when one assumes the data to be normally distributed, thereby allowing the use of significance tests on the factors extracted. It is difficult to test for multivariate normal distribution given the numbers involved; however, one can test for univariate normality given that this is required for multivariate normality. Table 1 reports the results for average returns for each country. (For individual securities the results are not reported due to large amounts of data though are available upon request). The requirement of multivariate normal distribution itself creates an additional bias in the sample to that already discussed in footnote 4. Those companies that have abnormally high or low returns during periods within the data period will not be included. Clearly by using monthly returns, any extremities in returns will tend to be smoothed out, though if daily or weekly data has been used the results with respect to the requirement of normality may not have been so favourable. 7 Table 2 reports, in some cases, low correlations between the returns of various countries. The correlation coefficient is a measure of linear association or linear dependence only and has no meaning for describing nonlinear relations. It is possible that low correlations, such as those shown for some countries in Table 2, may originate as a result of the hypothesis of linear correlation being false; however, tests performed to determine whether this could be the case clearly indicate that no nonlinear relationships exists. (The results from this test are available upon request). 8 The assumption with respect to the diagonal matrix implies that the idiosyncratic terms across the different securities are, on average, uncorrelated over time.

Capital market integration


idiosyncratic variances can differ from each other; however the assumption that returns follow a multivariate normal distribution is required instead. The pricing relationship as shown by Equation 2 is examined. The factors represent world factors, and given that they are unobservable, they are extracted and estimated from the world portfolio using the statistical technique of factor analysis. Factor analysis begins with a T n matrix of excess security returns, the relationship between these returns is shown by the correlation matrix, and factor analysis attempts to simplify this matrix such that it can be explained in terms of a small number of underlying common factors (factor analysis is discussed in more detail in the next section). Tests are conducted, as discussed in Section VI, testing the validity of the k-factor international pricing model and integration of the capital markets across the G7 countries.

1047
Principal component analysis and maximum likelihood analysis differ with respect to the assumption that is made regarding the amount of the unit variance of each variable (security return) which is to appear in the common factors, referred to as the communality. It is the figure placed in the diagonal of the correlation matrix that determines this, given that the diagonal value in the correlation matrix represents the total amount of variance of a variable distributed among the common factors. With principal component analysis, unity (the number one) is entered in the diagonal of the matrix, thus all the variance of the variable is explained by the factors. There is no unique factor in the model as all the variance of a variable is treated as being common. Given this, if the number of factors extracted equalled the number of variables, 100% of the variance of all the variables would be accounted for.9 Clearly a criteria needs to be selected so as to determine how many factors to extract that represent the correlation matrix. The Kaisers criterion is a method that is often adopted. When applying the Kaisers criterion the eigenvalue is examined given that this represents the total variance explained by each factor. Given that factors with eigenvalues less than one are no better than individual variables, only those factors with eigenvalues greater than one are extracted, as they are looked upon as common factors.10 Clearly it would be beneficial if one could separate the common from the unique variance given the importance of the common variance. In order to do this, one would require, before commencing, some knowledge regarding the communality of a variable, and place this value in the diagonal of the correlation matrix, thus allowing for unique variance to be built into the model. This is the principle underlying maximum likelihood factor analysis and represents the fundamental difference between the two methods. One needs to determine what value to enter in the diagonal of the correlation matrix. Unlike with principal component analysis the initial communalities have to be less than one, given that we are concerned only with the common variance. The multiple R2 from the regression equation that predicts

Downloaded by [University of Liverpool] at 22:04 16 January 2013

IV. Factor Analysis Factor analysis has been around for many years, first developed by Spearman (1904) using mathematical models in a study of human ability. Factor analysis is simply a statistical technique that can be used to identify a small number of common factors that explain the relationship between a number of interrelated variables, in this case security returns. The correlation matrix shows the relationship between the security returns, thus the objective is simply to reproduce the correlation matrix with a small number of common factors. There are two commonly used methods of extracting factors from a sample correlation matrix, namely principal component analysis and maximum likelihood analysis (Lawley and Maxwell, 1971). Their difference lies with the amount of variance of the variable that is to be explained. In this article, both methods are used to extract factors from the correlation matrix of returns from each of the G7 countries and also the world portfolio.
9

It would be fruitless to extract as many factors as there are variables given the aim of factor analysis is to explain a correlation matrix in terms of a few underlying factors. 10 There do exist other criteria, in addition to eigenvalues greater than one, to determine how many factors to correctly retain. These include, the criteria of substantive importance, the Scree-test, and also the criteria of interpretability and invariance. The criteria of substantive importance simply sets a criteria at which one would consider a factor to be substantively important. So if one considers important factors as those that explain a minimum of 5% of the variance of the variables, the criteria set would be 5%. The percentage of the variance of the variables explained by the factors is simply a percentage version of the eigenvalue. The Scree-test involves plotting the eigenvalue against each corresponding factor. The number of factors to retain is represented by the point where the eigenvalues begin to level off (so-called Scree as it represents the rubble at the foot of a mountain). This criterion is very subjective as one can find more than one break in the graph. The criteria of interpretability and invariance attempts to combine various rules and accepts decisions that are supported by a number of criteria. This method is extremely illusive. Given the above, it is for these reasons why the eigenvalue criteria is adopted.

1048
that variable from all other variables is used as the initial estimate of the communality of a variable. The Chi-square goodness-of-fit statistics for the adequacy of the model is used to determine the number of factors to extract. Factors are simply added one at a time until the Chi-square statistics no longer shows significance at the 10% level.11 The main difference between principal component analysis and maximum likelihood analysis, in terms of the extracted factors, is that, because with principal component analysis the communality (the diagonal element in the correlation matrix) is equal to one, common and unique variance is not separated out. The extracted factors therefore capture all the variance of the variables, because it is the variables that determine the factors, and that the variables consist of both common and unique variance. With maximum likelihood analysis, because the communality is not equal to one, common and unique variance is separated out, thereby resulting in common factors.12 When dealing with large samples, such as the world portfolio, the methods used to extract factors can result in the extraction of a large number of factors.13 With principal component analysis, using the Keiser criterion can result in many factors having eigenvalues that are greater than one, though at the same time being very close to one, which is not ideal. With respect to maximum likelihood factor analysis, with large sample sizes the goodness-of-fit statistics can cause small discrepancies in fit to be statistically significant which in turn will result in a larger number of factors then is required being extracted. A large factor model would be of no benefit when examining capital market integration given that it would not

D. Morelli
focus on risk premia for the common factors between countries. Given these problems with large sample sizes, the number of factors extracted from the world portfolio is restricted to a fixed number based on two criteria, firstly the average number of factors extracted from each of the G7 countries, and secondly, examination of the eigenvalue of the average number of factors 1, so as to determine how much additional variance of the variables is being explained by this additional factor (this is discussed further in Section VII). With respect to the first of these criteria, the factors extracted from each of the G7 countries are not extracted from a single portfolio consisting of all 160 securities, due to the problem discussed earlier relating to the extraction of factors from large samples, instead the securities of each country are randomly divided into four equal portfolios of 40 securities and factor analysis conducted on each group. The average of the number of factors extracted, across all the portfolios of all the G7 countries, is adopted with respect to these criteria in terms of determining the number of factors to be used to explain the world portfolio. Restricting the number of factors is necessary due to this positive relationship that exists between the number of factors and portfolio size.

Downloaded by [University of Liverpool] at 22:04 16 January 2013

V. Factor Scores Having estimated the factor structure for the world portfolio, the next step is to estimate the factor scores.14 Factor scores are constructed from a linear

11

Other methods of extracting risk factors do exists, such as, principal factor analysis, minimum residual factor analysis, image analysis and alpha factor analysis. These are different methods of common factor analysis. These methods are similar to the maximum likelihood analysis in that they separate the common from unique variance. Given this family of techniques that extract common factors, this article only adopts maximum likelihood analysis given its key advantage in that it adopts statistical tests for the significance of the factors extracted, which is the most satisfactory solution from solely a statistical viewpoint. 12 Given that with principal component analysis the factors are derived from the actual correlation matrix of the variables, the factors extracted are termed real factors, also referred to as components. With maximum likelihood analysis, because the communalities are estimated (is <1) the common factors extracted are hypothetical. The common factors can capture the correlation among the variables without being totally defined by the variables. Maximum likelihood is one of the different methods of common factor analysis, and it is important to appreciate that principal component analysis and common factor analysis (where maximum likelihood is one such method) are completely different techniques for the reasons given. 13 Studies by Kryzanowski and To (1983) and Diacogiannis (1986) provide evidence of a positive relationship between the number of factors and portfolio size. 14 Factor scores are required as the explanatory variables in testing the international asset pricing model given by Equation 1 in Section III. If the world portfolio had contained all the securities of the G7 countries used in this study, and factors had been extracted from the correlation matrix of the return series of such a portfolio, the factor loadings would have represented the coefficients from Equation 1, and thus there would have been no need to estimate the factor scores. As mentioned in footnote 4, the use of all securities in the world portfolio would cause problems when attempting to compute the factor loadings due to software limitations when dealing with such a large correlation matrix. Furthermore such a large portfolio would not be ideal due to the problems previously discussed. Given that all securities are used to test the international asset pricing model, and not just those used to construct the world portfolio, factor scores are thus calculated.

Capital market integration


combination of the observed variables (excess security returns R) as shown by Equation 3. Fj wj1 R1 wj2 R2 wjn Rn 3

1049
three methods, the factor scores are estimated as follows: Anderson Rubin F R0 U2 BB0 U2 SU2 B1=2 5 Bartlett Thurston F R0 U2 BB0 U2 B1 F R S B
0 1

Downloaded by [University of Liverpool] at 22:04 16 January 2013

where Fj represents the j-th factor, wji represents the factor score coefficient for the j-th factor and the i-th variable, where i 1, . . . , n (n representing the number of securities in the world portfolio, namely 350). When maximum likelihood analysis is used to extract factors, it is not possible to exactly identify the common factors from the variables due to the fact that each variable consists of a factor component, as given by its communality, and an idiosyncratic component, which represents its uniqueness. In terms of common and unique factors, this can be expressed as shown by Equation 4. Ri i1 F1 i2 F2 ik Fk Ui 4

6 7

where F represents the common factors, U represents the unique factor, unique to the i-th variable. Given that the factor scores are a linear combination of the observed variables, as shown by Equation 3, there thus exists a degree of indeterminacy when constructing the factor scores. The problem with constructing factor scores is that there does not exist a unique method. Given the indeterminacy problem that exists when constructing factor scores, three criteria are applied to the estimated factor scores. First, the estimated factors should be highly correlated with the true factors. Second, the estimated factors should be univocal, in that they should be highly correlated with the corresponding true factors and no other factors, and finally, the estimated factors should be orthogonal. Given that there does not exist one estimator that can satisfy all the above criteria, three well-known estimating methods are adopted, namely: AndersonRubin (1956), Bartlett (1937) and Thurstons (1935) regression method.15 AndersonRubin is orthogonal but not univocal, the Bartlett method is univocal but not orthogonal and the Thurston method does not meet the orthogonal nor univocal criteria though is superior with respect to the estimated factors correlating highly with the true factors. For each of the
15

where F represents a T k matrix of factor scores, R is a T n matrix of observed variables (excess security returns for the world portfolio), B is an n k factor loading matrix, U is an n n diagonal matrix of unique variances, S is an n n sample correlation matrix of observed variables, T represents the time period, n and k the number of variables and factors, respectively. When principal component analysis is used, because it is an exact mathematical transformation of the variables, and as a result the problem of indeterminacy discussed earlier does not apply, all three methods, AndersonRubin, Bartlett and Thurstons regression method will result in identical exact factor scores and not estimates.16 With maximum likelihood analysis, all three methods result in different factor scores.

VI. Tests for Integration of the G7 Capital Markets Once the factor scores are estimated for the world portfolio tests can be carried out to determine whether individual country security returns are correctly priced by the world factors. The pricing model as given by Equation 1 states that there exists a linear pricing relationship between the expected excess return and the k world factors. In order to test this, firstly a time-series regression of all individual security returns from each country on the world factors is performed. This is carried out on a country-by-country basis. Rit i i1 F1t i2 F2t ik Fkt "it 8

There is another method that can be adopted to construct factor scores, namely, least square criterion. As the least square criterion is similar to Thurstons (1935) regression method, it is not adopted in this article. One could also construct factorbased scores, which involves the simple summation of variables having large factor loadings. This method considers only variables with a factor loading above a given value, and a factor-based score is created from these chosen variables. This approach utilizes information from factor analysis, namely the factor loadings, to create the factor-based scores. As this method creates more factor-based scores than factor scores, it is not adopted in this article. 16 Recall that with principal component analysis, the factors account for all the variance in the correlation matrix, there is no separate unique variance, thus Equation 4 would not include the unique factor U.

1050
where Rit represents security is excess return at time t (i 1, . . . , n, where n 160 securities), Fjt represent the factor scores at time t (j 1, . . . , k factors), i and ik represents the parameters to be estimated with ij representing the sensitivity of the i-th security to the j-th factor where j 1, . . . , k, "it, the idiosyncratic term for i-th security. The resulting time series regression results in estimates of an n 1 vector of Is, an n k matrix of ijs, and D which is simply an n n unbiased matrix of the covariance matrix of idiosyncratic terms. One then performs, for each country, a crosssectional regression of excess security return against the s estimated from the time series regression. Given that the data sample used consists of monthly observations spanning 11 years, a total of 132 crosssectional regressions are conducted for each country (one cross-sectional regression for each month t). The cross-sectional regression is shown as follows: Ri 0 1 i1 2 i2 k ik Or in matrix notation; R X& 10 9

D. Morelli
employed to determine the significance of individual risk premia, and also to test whether the intercept term is zero. As previously discussed, integration of the capital markets requires an international pricing model to price risk. The above tests would determine if the international pricing model holds, in that individual country security returns are properly priced by the world risk factors. This itself would imply integration; however, it does not automatically imply that the capital markets of the G7 are fully integrated. For full integration to exist the price of risk must be the same across all countries. With respect to determining whether capital markets are fully integrated, an additional hypothesis is tested, namely, the hypothesis that the risk premia for corresponding factors are the same across all the G7 countries, thereby indicating that risks are priced equally across countries. To test this a paired t-test is conducted between time series estimates of risk premia for corresponding factors between groups of two countries. This is performed across all countries.18 A further test can be performed with respect to the intercept term. As previously discussed, the intercept term is tested to determine whether it equals zero for each individual country. Additionally, one can test a joint hypothesis that the intercept terms across all G7 countries are zero. The exact F-test of Gibbons et al. (1989) is employed to test this hypothesis.

Downloaded by [University of Liverpool] at 22:04 16 January 2013

where R is an n 1 vector of monthly excess security returns for n securities, X is an n (k 1) matrix, the first column of which is a vector of ones and the next k columns representing an n 1 vector of systematic risks estimated previously from the time series regression given by Equation 8, & is a (k 1) 1 vector of risk premia to be estimated. Estimation of & is obtained by employing a generalized least square regression procedure, where & (X0 D1X)1X0 D1R. The above regressions are carried out for each country. Thus for each country, 160 time-series regressions are performed and 132 cross-sectional regressions, resulting in the estimation of 132 k risk premia, the mean of these risk premia is then calculated upon which test are conducted. Does the k-world factor generating model explain the returns from individual G7 countries? In order to determine whether the pricing relationship as given by Equation 10 holds, for each individual country the Chi-square test is employed to determine the significance of the vector of risk premia.17 The t-test is
17

VII. Results Table 3 shows the number of factors extracted from each of the four portfolios (each consisting of 40 securities as explained in Section IV) for each individual country, and also the eigenvalue expressed as a percentage of the total factor model, using both maximum likelihood and principal component analysis.19 Depending on the method used to extract the factors, the number of factors extracted differs for each of the G7 countries. If one accepts a k factor return generating model, then the existence of differing number of factors indicates that this return generating model is not unique across the

The test involves testing the null hypothesis that l1 l2 lk 0 against the alternative 60. The test statistic is given as T lk W1 l0 k 2, where W is the covariance matrix of the time series estimates of risk premia, lk is a vector of mean risk premia. The test statistic is 2 with k degrees of freedom. 18 Testing, for example, whether the price of risk in the UK is the same as for the USA for international factor 1, l1UK l1USA, is tested adopting a paired t-test with null hypothesis l1UK l1USA 0 against the alternative 6 0. Clearly for full integration the price of risk must be the same across all countries. 19 The purpose of this is to compare the factor structure across the G7 countries and to use this information to impose a global factor structure based on the world portfolio. One could have conducted inter-battery factor analysis between two countries for all countries in an attempt to extract common factors, or also canonical correlation to determine whether countries share common factors; however, both of these methods have been conducted in previous research and thus are not attempted here.

Capital market integration


Table 3. Number of factors extracted from G7 countries and world portfolioa Canada Panal A: Based on maximum Portfolio 1 6 (48.32%) Portfolio 2 7 (49.04%) Portfolio 3 5 (46.08%) Portfolio 4 5 (47.91%) France Germany (49.72%) (55.75%) (54.93%) (55.98%) (56.21%) (60.37%) (57.64%) (58.96%) Italy 6 5 7 6 8 6 9 8 (48.04%) (47.21%) (52.82%) (53.84%) (57.42%) (52.05%) (55.41%) (56.92%) Japan 6 5 5 7 6 6 6 7 (48.37%) (43.54%) (45.47%) (51.52%) (52.07%) (48.56%) (50.53%) (53.98%) UK 8 9 9 8 9 10 10 9 (61.03%) (62.84%) (59.23%) (58.31%) (63.43%) (66.2%) (63.72%) (60.62%) USA 5 7 6 6 6 8 6 7

1051

likelihood analysis 6 (53.45%) 5 8 (56.93%) 7 7 (56.52%) 6 7 (55.61%) 7

(54.72%) (56.03%) (52.05%) (57.83%) (58.21%) (62.73%) (55.48%) (61.38%)

Panal A: Based on principal component analysis Portfolio 1 7 (52.37%) 8 (61.32%) 7 Portfolio 2 7 (51.03%) 9 (63.81%) 9 Portfolio 3 6 (49.21%) 9 (60.52%) 6 Portfolio 4 6 (48.21%) 8 (58.71%) 8

Notes: aThis table shows the number of factors extracted and also the eigenvalue of the factor model expressed as a percentage in parenthesis. For each of the G7 countries, factor analysis is conducted on each of their four portfolios each consisting of 40 securities.

Downloaded by [University of Liverpool] at 22:04 16 January 2013

G7 countries. A possible reason for this is due to the existence of country-specific factors. Such factors are unlikely to be captured by the world portfolio as they are not common to all countries.20 The average number of factors extracted across the G7 countries is seven, based on maximum likelihood factor analysis and eight based on principal component analysis. Thus, maximum likelihood analysis and principal component analysis is performed on the world portfolio, restricting the number of factors to this amount.21 Based on the maximum likelihood analysis, the eigenvalue of the factor structure of the world portfolio is found to be 43.61% and with principal component analysis, 46.28%. The existence of common factors extracted from the world portfolio does not indicate that the markets are integrated; however, it is of importance given that one can only test whether the price of risk is the same across countries if common sources of risk exists. Table 4 shows the results from testing the pricing relationship as given by Equation 10, where the factors are extracted using principal component analysis. The table reports, with respect to each country, the average coefficients and corresponding

t-statistics testing whether the intercept is zero and whether the risk premium for each factor is priced, and also the results of the joint 2 test for the vector of risk premia. When examining individual risk premia associated with the factors for each country, we can see that even though the majority of the factors are not priced, a number of factors are priced. Factor 1 is priced across four countries, namely Canada, Japan, Germany and the USA. Factor 2 is priced across the UK and USA, and factor 4 across Canada, Italy, Japan and the USA. What is clear to see, however, is that the same factors are not priced across all countries. With respect to the 2 test, the results show that for three countries, Canada, Japan and the USA, the hypothesis that the risk premia vector is not statistically significant is rejected. It is important to recognize that this test is biased in favour of the null hypothesis (Type II error) given that all the factors are included in this test, the majority of which are not significant. One could simply repeat the test with a reduced number of factors, however, this would create a bias against the null hypothesis. By incorporating all the factors the power of the test is clearly reduced. Table 4 also

20

It is possible that some of the factors extracted from the world portfolio may turn out to be common to a particular country within that portfolio, given that the factors attempt to explain the correlation matrix which itself contains correlations between security returns from the same country. This would be more likely with a large factor structure, which has been avoided in this article (see discussions in Section IV). 21 The restriction is due to the problems discussed in Section IV relating to large samples. On examining the eigenvalue of the K 1 factor for the world portfolio, in other words the 8th and 9th factor, this is found to be 1.28 and 1.19 for maximum likelihood and principal component analysis, respectively, which in percentage terms represents 0.37 and 0.34% of the variation in the returns of the world portfolio, thus is of minor importance. As the number of factors increase, the estimated factor loadings of high-order models will contain more noise than information. Given this, the SEs may well be of the same magnitude as the actual coefficients, which, if one was attempting to predict the price of risk, may result in very unstable predictions.

Downloaded by [University of Liverpool] at 22:04 16 January 2013

1052

Table 4. Results from testing the pricing relationship as given by Equation 10 where factors are extracted using principal component analysis l1 l2 l3 l4 l5 l6 l7 l8 2 15.91** 10.32 10.23 8.64 13.92* 9.02 16.03** 13.51*

l0

Canada

France

Germany

Italy

Japan

UK

USA

All G7 countries

0.0072 (1.02) 0.0033 (0.79) 0.0109* (1.98) 0.0064 (0.57) 0.0028 (1.03) 0.0041 (0.63) 0.0015 (0.52) 0.0027 (0.71)

0.0325** (4.17) 0.0063 (0.84) 0.0288** (3.46) 0.0019 (0.37) 0.0216** (3.03) 0.0080 (1.07) 0.0421** (4.92) 0.0357** (3.57)

0.0069 (0.85) 0.0051 (0.72)) 0.0013 (0.22) 0.0015 (0.27) 0.0021 (0.93) 0.039** (3.37) 0.0237** (3.21) 0.0105 (1.21)

0.0114 (1.41) 0.0194* (1.82) 0.0029 (0.32) 0.0032 (0.51) 0.0013 (0.62) 0.001 (0.38) 0.0032 (0.63) 0.00371 (0.53)

0.0254** (2.29) 0.0069 (0.85) 0.0016 (0.24) 0.0198* (1.84) 0.0308** (3.91) 0.0017 (0.42) 0.0181* (1.71) 0.0217** (2.36)

0.0089 (1.12) 0.0132 (1.56) 0.0072 (1.09) 0.0021 (0.43) 0.0009 (0.59) 0.0092 (1.16) 0.0041 (0.83) 0.0081 (1.02)

0.0014 (0.39) 0.0022 (0.58) 0.0049 (0.56) 0.001 (0.25) 0.0039 (1.27) 0.0030 (0.56) 0.0021 (0.51) 0.0032 (0.51)

0.00218 (0.45) 0.0015 (0.47) 0.0018 (0.25) 0.0009 (0.17) 0.0016 (0.69) 0.0023 (0.0.47) 0.0019 (0.47) 0.0017 (0.32)

0.0011 (0.32) 0.004 (0.68) 0.0027 (0.31) 0.0016 (0.30) 0.0025 (0.97) 0.0017 (0.38) 0.0015 (0.32) 0.0024 (0.46)

Notes: Average intercept and risk premium for each country is shown along with corresponding t-statistic in parenthesis. The 2 test at the end of each row testing whether the vector of risk premia is significantly different from zero. The critical 2 value with 8 d.f. is given as 15.507 at the 5% level and 13.361 at the 10% level. * and ** indicate significance at the 10 and 5% levels, respectively.

D. Morelli

Capital market integration


shows that when the cross-sectional regression is performed across all G7 countries, as opposed to each individual country, two factors are found to be significant and the risk premia vector is found to be statistically significant, though only at the 10% level. With respect to the intercept term, it is found that for all countries except Germany, the null hypothesis that the intercept is zero is not rejected. Table 5 reports the results from the same pricing relationship as shown by Equation 10, but where the factors are extracted using maximum likelihood analysis and the factor scores estimated using AndersonRubin, Bartlett and Thurstons regression method. The results are very similar across all three methods of factor score estimation. Again factor 1 shows significance across Canada, Germany, Japan and the USA, with factor 2 being priced across France, UK and USA, and factor 3 priced across Canada and Italy. All countries have at least one priced factor, though again this is not the same factor. The 2 test results are similar to those found in Table 4 in that the same three countries, Canada, Japan and the USA, reject the hypothesis that the risk premia vector is not statistically significant. When the cross-sectional regression is performed across all G7 countries, two factors are found to be significant and the risk premia vector is found to be statistically significant, again only at the 10% level. The 2 statistic is the same irrespective as to which method is used given that the total amount of variance of the variables explained by the common factors stays the same across all three methods. The amount of variance explained by individual common factors can change, though overall the total variance explained stays the same. The coefficient with respect to the intercept term stays the same across all three methods as this is not affected by the method of factor score estimation, and for all countries one fails to reject the hypothesis that the intercept term is zero. Irrespective as to which method is used to extract the factors or estimate the factor scores, it can be seen that the international asset pricing model does not hold for all the G7 countries. For four out of the seven countries, France, Germany, Italy and the UK, the model does not hold. This itself implies the absence of integration throughout all the G7 countries. Only when the cross-sectional tests are performed across all G7 countries, as opposed to individually, is the risk premia vector found to be statistically significant, thus implying a degree of integration existing between these markets, though clearly this is influenced by the

1053
strong cross-sectional results for Canada, Japan and the USA. Table 6 reports the F-statistic testing the joint restriction that all intercept terms across the G7 countries are zero. The results clearly show that irrespective as to which method is adopted regarding factor extraction and factor score estimates, the hypothesis that the intercept of all G7 countries is zero cannot be rejected. Table 7 summarizes the results from testing whether the risk premia are the same across all seven countries.22 Results are shown for both principal component and maximum likelihood analysis. Factor 1 is found to have the same risk premia across Canada, Germany and the USA, and thus the price of risk is the same for this factor across these countries, implying a degree of integration between these three stock markets. Factor 1 is the most important factor given it explains the highest proportion of the total variance of the world portfolio. The summarized results show that some countries clearly have the same price of risk, but this cannot be said across all countries. The results are not surprising following the results shown in Tables 4 and 5, given that capital market integration can only be tested under the joint hypothesis of an international asset pricing model, which clearly does not hold across all G7 countries. Where factors are extracted using maximum likelihood analysis, those countries having the same risk premia are very similar, and in many cases the same, across the three different methods of factor score estimation. For example, for factor 2, which accounts for the second largest proportion of total variance of the world portfolio, the UK and USA have the same risk premia across all three methods of factor score estimation, thus implying a degree of integration. The integration that exists between some of the G7 countries is dependant upon the technique used to extract the factors, for again, using factor 2 as an example, when using principal component analysis, Canada, France, Germany and Italy were found to have the same price of risk, and not the UK and USA which was the case with maximum likelihood analysis. However, irrespective as to the technique adopted, Table 7 clearly shows that the price of risk is not the same across all the G7 countries and thereby implying that the capital markets of the G7 are not fully internationally integrated.

Downloaded by [University of Liverpool] at 22:04 16 January 2013

22

The results are summarized due to the vast amounts of data. Full statistical results are available upon request.

Downloaded by [University of Liverpool] at 22:04 16 January 2013

Table 5. Results from testing the pricing relationship as given by Equation 10 where factors are extracted using maximum likelihood analysis l1 0.0286** 0.0243** 0.0273** (0.72) (0.87) (0.83) (3.28) (3.64) (3.81) (0.44) 0.0031 (0.41) 0.0042 (0.51) 0.0026 (3.03) 0.0051 (3.31) 0.0054 (3.29) 0.0056 (0.58) (0.52) (0.61) (3.84) (4.01) (3.88) (3.03) (3.51) (3.42) 0.0241** 0.0272** 0.0231** (2.84) (3.35) (3.02) 0.0052 0.0073 0.0058 0.0301** 0.0324** 0.0294** (3.07) (3.58) (3.19) 0.0062 0.0051 0.0041 0.0319** 0.0328** 0.0331** (4.21) 0.0021 (4.02) 0.0034 (3.98) 0.0031 (0.26) (0.28) (0.29) 0.0041 0.0032 0.0035 (0.72) 0.0009 (0.79) 0.0014 (0.79) 0.0016 (0.12) (0.23) (0.26) 0.0341** 0.0361** 0.0352** (4.42) (4.56) (4.43) (0.42) (0.56) (0.31) 0.0219* 0.0227* 0.0231* (1.78) (1.95) (1.91) 0.0061 0.0075 0.0064 (0.85) 0.0042 (0.91) 0.0036 (0.85) 0.0044 0.0063 0.0071 0.0075 0.0024 0.0027 0.0023 (0.18) (0.20) (0.18) 0.0041 0.0053 0.0037 (0.48) 0.0012 (0.58) 0.0035 (0.51) 0.0032 (0.11) (0.41) (0.43) 0.0103 0.0087 0.0112 (1.28) (0.72) (1.23) (0.54) (0.47) (0.49) 0.0198** 0.0216** 0.0231** (1.98) (2.14) (2.11) 0.0031 0.0038 0.0027 (0.26) (0.34) (0.37) 0.0094 0.0104 0.0086 (0.96) 0.0079 (1.03) 0.0102 (0.94) 0.0086 (0.80) (1.03) (0.85) 0.0054 0.0073 0.0079 0.0031 0.0022 0.0021 0.0028 0.0046 0.0053 (3.18) 0.0098 (2.73) 0.0133 (3.02) 0.0109 (0.86) (1.02) (0.94) 0.0201** 0.0231** 0.0195** (1.93) 0.0142 (2.25) 0.0131 (1.84) 0.0157 (1.23) (1.06) (1.49) 0.0032 0.0049 0.0046 (0.41) (0.68) (0.62) 0.0043 0.0038 0.0033 l2 l3 l4 l5 l6 l7 2

l0

Canadaa

1 0.0063 (0.96) 2 0.0063 (0.96) 3 0.0063 (0.96)

(0.62) 0.0019 (0.23) 14.93** (0.53) 0.0027 (0.36) 14.93** (0.46) 0.0022 (0.31) 14.93** (0.56) (0.71) (0.76) 0.0041 0.0063 0.0061 (0.39) (0.54) (0.58) (0.33) 0.0033 (0.38) (0.18) 0.0026 (0.22) (0.15) 0.0029 (0.36) (0.25) 0.0018 (0.26) (0.62) 0.0023 (0.31) (0.71) 0.0029 (0.37) 8.62 8.62 8.62 9.14 9.14 9.14 7.23 7.23 7.23

France

1 0.0021 (0.29) 0.0076 2 0.0021 (0.29) 0.0086 3 0.0021 (0.29) 0.0081 0.0313** 0.0331** 0.0327** 0.0029 0.0033 0.0037 0.0239** 0.0248** 0.0241** 0.0059 0.0067 0.0063 0.0392** 0.0408** 0.0372** 0.0324** 0.0364** 0.0338**

Germany

1 0.0086 (1.32) 2 0.0086 (1.32) 3 0.0086 (1.32)

Italy

1 0.0071 (0.46) 2 0.0071 (0.46) 3 0.0071 (0.46)

Japan

1 0.0042 (1.41) 2 0.0042 (1.41) 3 0.0042 (1.41)

(0.91) 0.0013 (0.21) (0.89) 0.0009 (0.14) (0.96) 0.0011 (0.22) (0.46) 0.0121 (1.36) (0.29) 0.0132 (1.49) (0.33) 0.0141* (1.69) (1.21) (1.42) (1.27) 0.0013 0.0026 0.0042 (0.21) (0.38) (0.56) 0.0049 0.0041 0.0052 0.0043 0.0031 0.0052 (0.53) (0.38) (0.52)

0.0029 0.0041 0.0055 0.0037 0.0051 0.0092

(0.39) 12.84* (0.62) 12.84* (0.79) 12.84* (0.25) (0.49) (0.92) 8.39 8.39 8.39 (0.53) 0.0031 (0.47) 15.32** (0.39) 0.0018 (0.20) 15.32** (0.61) 0.0021 (0.26) 15.32**

UK

1 0.0031 (0.41) 2 0.0031 (0.41) 3 0.0031 (0.41)

USA

1 0.0029 (0.89) 2 0.0029 (0.89) 3 0.0029 (0.89)

(0.63) 0.0104 (0.67) 0.0131 (0.53) 0.0121 (0.72) (0.91) (0.86) 0.0073 0.0083 0.0071

All G7 1 0.0038 (0.54) countries 2 0.0038 (0.54) 3 0.0038 (0.54)

(1.31) 0.0084 (1.42) 0.0071 (1.09) 0.0063

(0.92) (0.88) (0.86)

0.0021 0.0029 0.0018

(0.25) 0.0038 (0.46) 12.38* (0.32) 0.0031 (0.42) 12.38* (0.20) 0.0041 (0.67) 12.38*

Notes: a1, 2 and 3 represent the AndersonRubin, Bartlett and Thurston methodology, respectively. Average intercept and risk premium for each country is shown along with corresponding t-statistic in parenthesis. The 2 test at the end of each row testing whether the vector of risk premia is significantly different from zero. The critical 2 value with 7 d.f. is given as 14.067 at the 5% level and 12.017 at the 10% level. * and ** indicate significance at the 10 and 5% levels, respectively.

Capital market integration


Table 6. The F-statistic testing the joint restriction that all the intercepts across the G7 countries are zero F-test Factors extracted using principal component analysis Factors extracted using maximum likelihood analysis AndersonRubin Bartlett Thurston 0.826

1055
component analysis and maximum likelihood factor analysis, the results obtained do not differ significantly upon the technique adopted. The number of factors extracted from the world portfolio differed only slightly, seven based on Maximum likelihood and eight for principal component analysis. Of the various criteria to adopt to determine the number of factors to extract (see footnote 10), the chi-square goodness-of-fit statistics and Kaisers criterion were applied. Clearly, different criteria can result in different numbers of factors extracted, however, what is important is not so much the number of factors extracted but whether the factors are priced, and more importantly, in terms of integration, whether they are priced equally across all G7 countries. Although the results show different numbers of priced factors (Tables 4 and 5) and different countries having a similar price of risk according to the technique used to extract the factors (Table 7), irrespective of the technique adopted that the price of risk is not found to be the same across all the G7 countries. The factor scores proxy for the true risk factors. The use of proxies can have an influence on the overall results depending upon the accuracy of the proxy. Three different methods were adopted to estimate the factor scores AndersonRubin, Bartlett and Thurston. Of these methods, Thurstons method is superior in terms of producing estimated factors which correlate highly with the true factors. Given this, the results are found to be similar across all three methods. Clearly, when extracting factors using maximum likelihood analysis the factor scores are estimates, unlike principal component analysis which produces exact factor scores, however, the results are still similar in that the price of risk is still found not to be the same across all G7 countries. Thus the conclusion from this article in terms of the question of full integration is similar irrespective of the technique adopted to extract the factors and estimate factor scores and thus implying that the conclusion drawn is not sensitive to the different methods and techniques adopted. To test for integration requires a valid international asset pricing model, which in turn is only a valid model if the markets are integrated, thereby resulting in a joint hypothesis problem. The results provide evidence against this joint hypothesis. Although a degree of integration is shown to exist between some of the G7 countries, one would have to conclude that based on an international multifactor asset pricing model, the hypothesis of full integration between all the G7 countries does not hold.

0.747 0.747 0.747

VIII. Conclusion Over the years, the economic and financial systems of the G7 countries have increasingly become more integrated due to the expansion in areas such as trade, services and financial assets. This article distinguishes between financial integration and security returns being correlated internationally, focusing on the latter in testing for capital market integration. This article provides empirical investigation of an international asset pricing model in an attempt to determine whether the capital markets of the G7 countries are integrated. Capital market integration is tested using a multifactor pricing model, where common factors are extracted from a world portfolio made up from a combined subset of securities from each of the G7 countries. Two well-known methods are adopted to extract factors, namely, maximum likelihood factor analysis and principal component analysis, where the number of factors extracted is based upon the chi-square goodness-of-fit statistics and Kaisers criterion, respectively. In order to imply that the capital markets of the G7 are fully integrated, the price of risk must be the same across all countries and thus the same factor must have the same risk premium for all countries. The cross-sectional results as shown in Tables 4 and 5 show that in terms of the international asset pricing model, for each country at least one, and in some cases more than one, risk premia is found to be priced, however this does not relate to the same factor across all countries. On examining the price of risk, as shown in Table 7, it is found to be the same for some factors for some of the countries, thereby implying a degree of integration, however, is not found to be the same for all countries. Clearly, the results obtained and conclusions drawn are based upon the methodology adopted, specifically in terms of the techniques and criteria adopted to extract factors and estimate the factor scores. Given the two different techniques adopted in this article to extract factors, namely, principal

Downloaded by [University of Liverpool] at 22:04 16 January 2013

Downloaded by [University of Liverpool] at 22:04 16 January 2013

Table 7. Summary of results from testing whether the risk premia are the same across all G7 countries Countries where risk premia were found to be the same Factors extracted using

Risk premia Principal component analysis Maximum likelihood analysis

Risk premia the same across all seven countriesa

l1

No

l2

No

AndersonRubin Canada, Germany and USA UK and USA

Bartlett Canada, Germany and USA UK and USA

Thurston Canada, Germany and USA UK and USA

l3 Italy and USA Canada and UK Germany and USA Italy, UK and USA

No

USA and Canada, Germany and Japan Canada and France, Germany and Italy Germany, Italy and USA

l4 l5 l6

No No No

Canada and Italy, Germany and USA Canada, UK and USA Canada, Germany and Italy

l7

No

Canada and Italy, Germany and USA Italy and France Canada and France, UK and USA Canada, Germany and Italy

Canada and Italy, Germany and USA Italy and France Canada and France, UK and USA Italy and Germany

l8

No

Notes: aThe results reported show that no individual risk premia was found to be the same across all seven countries. This was the case irrespective of whether the factors were extracted using maximum likelihood or principal component analysis, or whether AndersonRubin, Bartlett or Thurstons methodology was used to estimate the factor scores.

Capital market integration


References
Anderson, T. W. and Rubin, H. (1956) Statistical inference in factor analysis, in Proceedings of the 3rd Berkeley Symposium on Mathematical Statistics and Probabilities, Vol. 5, University of California Press, Berkeley, pp. 11150. Bartlett, M. (1937) The statistical conception of mental factors, British Journal of Psychology, 28, 97104. Byers, J. D. and Peel, D. A. (1993) Some evidence of interdependence of national stock markets and the gains from international portfolio diversification, Applied Financial Economics, 3, 23942. Campbell, J. Y. and Hamao, Y. (1992) Predictable stock returns in the United States and Japan: a study of long-term capital market integration, Journal of Finance, 47, 4370. Cheng, A. C. S. (1998) International correlation structure of financial market movement the evidence from the UK and US, Applied Financial Economics, 8, 112. Chou, P. H. and Lin, M. C. (2002) Tests of international asset pricing model with and without a riskless asset, Applied Financial Economics, 12, 87383. De Fusco, R. A., Geppert, J. M. and Tsetsekos, G. P. (1996) Long run diversification potential in emerging stock markets, Financial Review, 31, 34363. Diacogiannis, G. P. (1986) Arbitrage pricing model: a critical examination of its empirical applicability for the London stock exchange, Journal of Business Finance and Accounting, 13, 489504. Divecha, A. B., Drach, J. and Stefek, D. (1992) Emerging markets: a quantitative perspective, Journal of Portfolio Management, 19, 4150. Eun, C. S. (1985) A model of international asset pricing under imperfect commodity arbitrage, Journal of Economic Dynamics and Control, 9, 27390. Eun, C. S. and Janakiramanan, S. (1986) A model of international asset pricing with a constraint on the foreign equity ownership, Journal of Finance, 41, 897914. Gibbons, M. R., Ross, S. A. and Shaken, J. (1989) A test of efficiency of a given portfolio, Econometrica, 57, 112152. Gultekin, N., Bulent, M. N. and Penati, A. (1989) Capital controls and international capital market

1057
segmentation: the evidence from Japanese and American stock markets, Journal of Finance, 44, 84969. Heston, L. H., Rouwenhorst, K. G. and Wessels, R. E. (1995) The structure of international stock returns and the integration of capital markets, Journal of Empirical Finance, 2, 17397. Jorion, P. and Schwartz, E. (1986) Integration versus segmentation in the Canadian stock market, Journal of Finance, 41, 60316. Kanas, A. (1998) Linkages between the US and European equity markets: further evidence from cointegration tests, Applied Financial Economics, 8, 60714. Korajczyk, R. A. and Viallet, C. J. (1989) An empirical investigation of international asset pricing, Review of Financial Studies, 2, 55385. Kryzanowski, L. and To, M. C. (1983) General factor models and the structure of security returns, Journal of Finance and Quantitative Analysis, 18, 3152. Lawley, D. N. and Maxwell, A. E. (1971) Factor Analysis as a Statistical Method, 2nd edn, Butterworths and Co Ltd, London. Michaud, R. O., Bergstrom, G. L., Frashure, R. D. and Tajbakhsh, S. (1996) Twenty years of international equity investing, Journal of Portfolio Management, 23, 922. Ross, R. (1976) The arbitrage theory of capital asset pricing, Journal of Economic Theory, 13, 34160. Ross, R. (1977) Return risk and arbitrage, in Risk and Return in Finance (Eds) I. Friend and J. Bicksler, Ballinger, Cambridge. Solnik, B. (1974) The international pricing of risk: an empirical investigation of the world capital structure, Journal of Finance, 29, 4854. Spearman, C. (1904) General intelligence: objectively determined and measured, American Journal of Psychology, 15, 20192. Swanson, P. E. (2003) The interrelatedness of global equity markets, money markets, and foreign exchange markets, International Review of Financial Analysis, 12, 13555. Thurston, L. (1935) The Vectors of Mind, University of Chicago Press, Chicago. Vo, X. V. and Daly, K. J. (2005) European equity market integration implications for US investors, Research in International Business and Finance, 19, 15570.

Downloaded by [University of Liverpool] at 22:04 16 January 2013

Vous aimerez peut-être aussi