The relative importance of country effects vs industry effects in European equity returns

(1)

The relative importance of country effects vs industry

effects in European equity returns

Author : Mathijs Huijbregts Student number : 10541438

Program : MSc Finance

Specialization : Asset Management Supervisor : Dr. Zou

(2)

Statement or Originality

This document is written by Mathijs Huijbregts, who declares to take full responsibility for the contents of this document.

I declare that the text and the work presented in this document is original and that no sources other than those mentioned in the text and its references have been used in creating it.

The Faculty of Economics and Business is responsible solely for the supervision of completion of the work, not for the contents.

(3)

Abstract

In this paper the relative importance of country effects over industry effects in European equity returns is studied. A return-based style analysis is used in this study. The full sample period is 2008-2015. This sample period is divided into a crisis period and a post-crisis period. By using these sub periods, the change in the relative

importance of country effects over industry effects in European equity returns can be studied. This study finds that the financial crisis causes a significant increase in the average country-specific variance in European equity returns. This causes a

significant increase in the variance ratios as well, which are calculated by the average country-specific variance over the average industry-specific variance. Mainly due to an increased financial market integration in Europe since the adoption of the Euro, there was a gaining relative importance of industry effects over country effects in European equity returns before the financial crisis. However, this study finds that the financial crisis causes a reversal in this gaining relative importance of industry effects over country effects in European equity returns. The increase in the simple- and exclusive variance ratios is significant at a 1% level for both the equally weighted- and the value weighted approach. These results support the main hypothesis that there was a reversal in the gaining relative importance of industry effects over country effects in European equity returns. In this paper, there is a separation between Euro countries and non-Euro countries. The results show as well that the variance ratios are larger for non-Euro countries compared to Euro countries. This is due to a larger amount of country effects in the returns of the non-Euro country portfolios compared to Euro countries. The financial markets of non-Euro countries are less integrated in the European market. For example, this is due to the use of their own currency.

(4)

1. Introduction

Since 1970 there has been a debate about the importance of country effects and industry effects in equity returns. The interesting part of the research in this field, is that previous research found contrary results. Eiling et al. (2012) studied country effects vs industry effects in equity returns. Their study was based on data from 1990 till 2008. By using this

timeframe, they want to study the adoption of the Euro as a single currency in Europe. Their argument for this study was that the adoption of the Euro has caused large effect on the monetary and economic environment and processes in Europe. The introduction of the Euro caused a standardizing in pricing in financial markets and reduced the transaction cost of investors (Hardouvelis et al., 2006). Eiling et al. (2012) found after the adoption of the Euro in increase in the relative importance of industry effects over country effects in equity returns. The main reason for this was the increasing financial market integration between the Euro countries had increased. Due to an increasing financial market integration, the average country-specific variance decreased and the average industry-specific variance across countries increased. Eiling et al. (2012) mention this as an increase in the link between different national equity markets. However, Chou et al. (2014) and Bekaert et al. (2014) found results in contrast with the results of Eiling et al. (2012). Bekaert et al. (2014) found that local factors became more importantly again after the financial crisis. Their argument for this was that the financial crisis has caused a change in the process of an increasing financial market integration. However, they studied a Global sample. Therefore, we will see in this study if their results hold as well for European countries. Chou et al. (2014) found results that were in line with these results.

Hwang et al. (2014) came up with some possible reasons for these increasing importance of country effects in stock returns. First, that there could be an increase in segmentation between different economic markets and countries, due to the different ways of how these countries react on economic shocks. In terms of fiscal and monetary policy, countries could have different ways and reasons to react to these fundamental economic shocks. The second reason, maybe even more important, is the existing home bias for investors. In times of financial uncertainty, Investors tend to invest more than needed in their home country. This is due to that investors are better informed and more optimistic

(6)

about their home country. Furthermore, investors put more weight in certain international markets. This could be markets in countries with a common language or culture. Or just countries that are close in distance to the home country of the investor.

Thus, the relative importance of industry effects vs country effects in European equity returns is still an open debate. Especially, because the recent financial crisis took place and had a large impact on the European financial markets. Consequently, it is good to revise this relative importance in European equity returns. This can give interesting insights for the portfolio management industry. To show the trend in the relative importance over time. The full sample period, which is 2008-2015, will be slit up in 2 smaller timeframes. These timeframes will be the crisis period and the post-crisis period. By doing this, we could see if the financial crisis had an impact on the relative importance. Therefore, the research question will be: Has the financial crisis caused a change in the gaining relative importance of industry effects over country effects in European equity returns?

A wide used method to study country effects vs industry effects in equity returns is the decomposition method of Heston and Rouwenhorst (1994). This method makes use of industry- and country dummy variables for all the countries and industries in the respective sample. This methodology requires data from individual stocks instead of equity index returns. Though, Bekeart and Harvey (2017) showed in their study that this approach has some weaknesses. They showed that market betas are time varying. However, in the study of Heston and Rouwenhorst (1994), they assume that all assets have a unit exposure to the global market for all assets. Therefore, in this study will we used an alternative method. This method is called return-based style analysis. This method is used to see if returns of a specific country can be mimicked by a replicated European industry portfolio and vice versa. If country returns are important in explaining European industry equity returns, then country returns will perform slightly well in explaining European industry equity returns. While replicating portfolios of European industries, which have been set up with the indices of local industries, would not explain as much of the variance in the country returns. Pure industry effects and pure country effects will be extracted as well. This will be done by excluding the overlapping industry or country from the benchmark portfolio’s. This methodology will be explained in more detail in the methodology section.

The outline for the coming parts in this thesis is as follows. In the second part of this study the related literature in this field can be found, as well as the hypotheses for this

(7)

study. Results from previous studies within this field will be given there as well. In the third part, the data description will be presented. In the fourth part, the methodology is explained in more detail. After this, the results are given in part five. In the end, the conclusions based on the results will be formed there. Furthermore, some possibilities for future research will be explained at the end.

(8)

2. Related Literature

Country- and industry effects have been studied a lot during the last few decades. For investors, it is interesting to know, because they want to know on which diversification strategy they should focus. Thus, should they focus their diversification strategy on

diversifying their portfolio more across countries or across industries. In this section, first the previous results of some important studies in this field will be summarized. Secondly, the hypotheses for this study will be formed and explained in more detail.

2.1 Previous research on country effects vs industry effects in equity returns

In this part, there can be found a summary of previous results in the field of country effects vs industry effects in equity returns.

Equity returns are driven by multiple factors. These factors can be divided into two main groups. These two main groups are country factors, also called local factors, and industry factors, also called regional- or global factors. The topic of industry effects vs country effects in equity returns is already a debate for a few decades. Therefore, there is already some existing literature in this field. For example, already in 1994, Heston and Rouwenhorst (1994) published an article in this field. In this field, their article was very important. They came up with a new decomposition method. By using this model, they could separate country effects from industry-driven sources of return variation and vice versa. At the time this article was published, the financial market integration process was still in a very early stage. Therefore, it is not surprisingly that they found that industry effects are much smaller compared to country effects in equity returns. Their sample consists of 829 firms that were in the MSCI index of 12 different European countries. These firms were distributed over 7 different industries. They used all the firms in the indices, therefore you could say that they studied the effect of country indices and industry indices. Normally, country indices are more volatile than industry indices. Following Heston and Rouwenhorst (1994), the main reason for this is that the countries are less diversified across industries than industries are diversified across countries. However, their results were contradicting with this argument. They came up with the conclusion that countries explain more in the variation of European industry returns compared to how much industries explain in the variation of country returns. Thus, that country effects were more important than industry effects in European equity returns.

(9)

Eiling et al. (2012) studied country effects, industry effects and currency effects in stock returns. Their sample consists of 7 developed stock markets. These countries are also called the G7 countries. The timeframe of their sample was 1975-2011. Currency effects and industry effects had the largest impact on international stock returns, following Eiling et al. (2012). By using the smaller timeframes, they observed a change in this relative importance between these 3 effects. They found an increase in the importance of industry effects over country effects in stock returns. These results are line with the increasing market integration during their timeframe. In their paper, they mentioned that several other papers found the same results. Although, that these results were mainly caused by the IT sector. When excluding the IT-sector, the country effects should dominate the industry effects again.

Campa and Fernandes (2006) found results which were in line with the results Eiling et al. (2012) found. They support the papers that mention an increase in the relative

importance of industry effects over country effects in the recent decade. The main reason for this is financial market integration, mentioned in other papers as well. Catao and

Timmerman (2010) supported these results with their study. They showed that the industry factors increased significantly relative to the country factors since the late 1990’s. They found that around 2000, the contribution to the total risk of equity returns was 30% for the industry factors. This was only 10% in the full sample period 1973-2002. This shows the significant increase of the exposure to industry factors in the end of this timeframe.

Hwang et al. (2014) conducted a study with five ASEAN countries. They selected 227 different equities across 10 different industries in 5 different countries. They used selected firms instead of indices. They found an increasing relative importance of industry effects over country effects, although that country effects still explain more variation in stock returns compared to industry effects. This was between 2002 and 2010. Their results are in line with the results from Catao and Timmerman (2010). Although, the increasing relative importance of industry effects did stop for a while, following Hwang et al. (2014). They showed graphical that there was an increasing trend of country effects, starting between 2007 and 2009.

Chou et al. (2014) published an article where they compared EU Euro-countries to EU non-Euro-countries. They found that before the financial crisis, in the year of the adoption of the Euro, there was an increase in industry effects in stock returns. This is a logical result, because of the increasing financial market integration. This increasing financial market

(10)

integration was mainly due to the adoption of the Euro as a general currency in some European countries. During the financial crisis, the importance of industry effects decreased significantly relative to country effects. The evidence for this is given by the p-values, which are all 0.000 for the Euro-countries and the PIGGS countries. These are the p-values for the null hypothesis that the variance ratios are equal in the crisis period and the post-crisis period. Thus, this is evidence for a significant difference between the variance ratios in these two periods. However, only for the non-PIGGS countries that do not use the Euro, the p-values are not significant. Therefore, the financial crisis did not cause a significant change in the variance ratios of these countries. They found that before the crisis period, the average country-specific variance was 36% lower than the average industry-specific variance, but this was only 6% during the crisis period. Like in the study of Chou et al. (2014), the weights of the countries and industries are constrained to be non-negative in this study. Therefore, the change in the variance ratios that is and will be observed is robust to a situation where short-sales are not permitted.

Boamah et al. (2017) studied country effects and industry effects in African stock returns. They found that country effects dominating industry effects in African stock returns. The reason for this could be that African financial markets are less integrated compared to other financial markets.

In the study of Bekaert et al. (2009), they used a global sample. They showed evidence in their paper that there is again an increase in the relative importance of country effects over industry effects. They conclude that the increase in the relative exposure of industry effects over country effects in equity returns was temporary. Their argument for this was that the process of financial integration has not yet caused changes of the

correlations among international equity markets that are permanent. These results are in contrast with the results found by Eiling et al. (2012). Although, they both studied a global sample. This contrast emphasizes again the importance of this question, because

researchers keep on founding contrary results in this field. The timeframe used in this study will contain more recent years compared to the other studies in this field.

(11)

2.2 Hypotheses

The hypotheses will be mainly based on the existing results from the previous studies. Because Bekaert is an important researcher in this field, his results will be considered as important. Chou et al. (2014) their results were highly significant. These results will be considered as important as well. Especially because they also studied the factor reversal that would be eventually caused by the financial crisis. Therefore, the main hypothesis will be: There is a change in the gaining relative importance of industry effects over country effects in stock returns. This will be tested by dividing the timeframe used in this study in a crisis period and a post-crisis period. Therefore, I will be able to show the change in the relative importance of country effects vs industry effects in European equity returns around the financial crisis.

H0 : Variance Ratio crisis period = Variance Ratio post-crisis period

H1 : Variance Ratio crisis period < Variance Ratio post-crisis period

The main hypothesis is described in a different form above. This hypothesis will be tested for the simple variance ratios and the exclusive variance ratios. There will be used two different approaches to calculate both these kind of variance ratios. This will be done by using a value weighted approach and an equally weighted approach. These weights are the weights of the countries and industries in the Euro-wide portfolio.

In the analysis, there is made a separation between Euro-countries (group 1) and non-Euro countries (group 2). This separation is based on the study of Chou et al. (2014) and Eiling et al. (2012). Both these studies made this separation as well. Based on their previous results an expectation for these 2 groups of countries will be formed. The expectation will be that the variances ratios for group 2 countries are larger compared to the variances ratios of the group 1 countries. This expectation is based on the further state of the process of financial market integration in which the group 1 countries are compared to the group 2 countries. Mainly due to a main currency.

H0 : Variance Ratio Group 1 full period = Variance Ratio Group 2 full period

(12)

3. Data

3.1 Sample selection and data collection

In this study, a European sample is used. The sample consists of 12 countries and 10

industries. Local industry indices for the 10 different industries across these 12 countries are obtained from DataStream. The selection of the countries is based on the countries on which the MSCI Europe index is based. This index contains 15 European countries. The countries Ireland, Austria and Greece are excluded from the sample. This is due to a lack in the availability of the data for these three countries. For all these local industry indices, weekly returns and market values are obtained from DataStream. For the returns, the RI is used. This are total return indices with dividends reinvested. DataStream itself has all the

exchange rates. Therefore, by using a function in DataStream, the returns and market values for the non-Euro countries are converted in Euro returns and market values. By aggregating the local industry indices, the country portfolios and European industry portfolios are composed. The summary statistics for these country portfolios and European industry portfolios are in table 1. Table 1 consists of Panel A, Panel B, Panel C and Panel D. In Panel A and Panel B are the summary statistics for the full period. To see the impact of the crisis, the full period is divided in a crisis period and a post-crisis period in Panel C. The crisis period used in this study is based on the paper of Chou et al. (2014). They did a comparable study in Europe as well. They used 2008:1-2011:9 as crisis period. The period 2011:10-2015:12 is used as the post-crisis period. In Panel D, the results of the Wald tests are reported. These Wald tests are performed over the results in Panel A and Panel B.

3.2 Summary statistics

Table 1 - Summary statistics for European country and European industry portfolio returns This table reports summary statistics for the weekly returns on European country portfolios and European industry portfolios. The countries are divided into two groups in Panel A. In group 1 are the Euro-countries and in group 2 are the non-Euro countries. In the first column are the names of the respective countries given. In the second column, the mean returns for all countries are given in percentages. Third, the standard deviations of these returns are given. Corr (ctr), in the fourth column, is the average correlation of between the returns of that country with all the returns of all other countries. Corr (ind) is the average correlation of the return of that country with the return of all other industries. In the last column, the weight of the respective countries in the European portfolio are given. The averages for all

(13)

these statistics for all different countries are given as well. Thus, this is the European

average. In Panel B are the same statistics given as in Panel A, except that these statistics are for the industry portfolios instead of the country portfolios. To see the change over time, the full period is divided into two subsample periods in Panel C. The first period is the crisis period. This period is from January 2008 till September 2011. The second period is the post-crisis period. This period is from October 2011 till December 2015. Panel D reports the p-values from the Wald tests. These are performed on the mean returns for group 1, the mean returns for group 2 and the mean European industry returns.

Panel A: Returns on country indices – Full timespan Mean

(%)

Std. (%)

Corr (ctr) Corr (ind) Weight (%) Herf. Group 1 – Euro countries Belgium 0.15 3.08 0.70 0.75 3.64 0.18 Finland 0.04 3.46 0.71 0.78 1.72 0.06 France 0.07 3.10 0.77 0.86 17.14 0.03 Germany 0.10 3.13 0.73 0.83 16.84 0.05 Italy 0.01 3.52 0.74 0.79 6.75 0.10 Netherlands 0.07 3.24 0.76 0.83 5.55 0.09 Portugal -0.08 3.12 0.67 0.71 0.70 0.06 Spain Average Group 2 – Non-Euro countries Norway 0.03 0.05 0.00 3.55 3.28 4.08 0.68 0.72 0.66 0.73 0.79 0.72 7.52 7.48 3.26 0.11 0.09 0.21 Sweden 0.18 3.74 0.72 0.79 5.57 0.12 Switzerland 0.19 2.46 0.68 0.79 11.49 0.14 United Kingdom 0.15 2.13 -0,04 -0.04 19.81 0.04 Average 0.13 3.58 0.51 0.57 10.03 0.13

Panel B: Returns on European industry indices – Full timespan Mean

(%)

Std. (%)

Corr (ctr) Corr (ind) Weight (%)

Herf.

Oil & Gas -0.02 3.60 0.70 0.75 8.89 0.25

Basic Materials 0.00 4.15 0.72 0.76 8.06 0.17 Industrials 0.11 3.19 0.79 0.82 12.96 0.09 Consumer Goods 0.21 2.69 0.72 0.78 16.45 0.09 Healthcare 0.23 2.52 0.58 0.66 8.33 0.23 Consumer Services Telecom 0.14 0.12 2.75 2.77 0.76 0.65 0.82 0.72 7.66 5.24 0.15 0.09 Utilities -0.04 2.96 0.70 0.75 6.52 0.09 Financials Technology 0.02 0.13 3.89 3.12 0.77 0.73 0.78 0.79 22.71 3.81 0.08 0.11 Average 0.09 3.16 0.71 0.77 10.06 0.13

(14)

Panel D: Results of the Wald tests for the mean returns in panel A and Panel B Panel A – Group 1 – Euro countries

H0: All country means are zero (0.003)

H0: All country means are equal (0.465)

Panel A – Group 2 – Non-Euro countries

H0: All country means are zero (0.013)

H0: All country means are equal (0.301)

Panel B – Group 2 – European industries

H0: All industry means are zero (0.000)

H0: All industry means are equal (0.196)

The analysis is extended by dividing the countries into two groups. Due to a different economic and monetary policy, the financial crisis will probably have different effects on these two groups of countries. By using the separation between these two groups of countries, we can shed more light on the differences in the change in the relative importance of country effects and industry effects in European equity returns. Panel A shows that the mean return of country portfolios in non-Euro countries is higher than the mean return of country portfolios in Euro countries in the full period. This could be an indicator that the Euro countries suffered more from the financial crisis compared to the

Panel C: Summary statistics subsample periods Mean

(%)

Std. (%)

Corr (ctr) Corr (ind) Herf.

Group 1 – Euro-countries Crisis period -0.22 3.90 0.75 0.80 0.0873 Post-Crisis period 0.29 2.56 0.68 0.76 0.0988 Group 2 – Non-Euro countries Crisis period 0.02 3.25 0.51 0.56 0.1338 Post-Crisis period 0.24 1.78 0.49 0.56 0.1237 European industries Crisis period -0.12 3.90 0.72 0.77 0.1028 Post-Crisis period 0.27 2.31 0.69 0.75 0.1071

(15)

non-Euro countries. The average correlation between the Euro countries and the other countries is higher than the average correlation between the non-Euro countries and the other countries, i.e. 0.72 relative to 0.51. This is mainly caused by the common currency the Euro countries are using. These Euro countries are more financially integrated.

Although, both group 1 and group 2 have a higher average correlation with industries than with the other countries. A reason for this could be that there is one overlapping local industry index in both an industry portfolio and a country portfolio. Another reason could be that the European industries are just more financially integrated among each other

compared to the European countries. Next to the two columns with correlations, is the column with the weights from all countries in the European portfolio. United Kingdom has the largest weight across the European countries in this sample. This large weight is mainly caused by the large financials sector in the UK. After the UK, Germany and France have the largest weight. In the last column in Panel A is the Normalized Herfindahl index given. This index in Panel A measures the concentration of industries in country portfolios. The Herfindahl index is calculated by the summation of the squared local industry weights. H is the Herfindahl index and M is the weight of the local industry index. This is also called the unadjusted Herfindahl Index. This index is calculated by using equation (2):

𝐻𝑒𝑟𝑓𝑖𝑛𝑑𝑎𝑙 𝐼𝑛𝑑𝑒𝑥 = ∑ 𝑀2₍₂₎

Although, table 1 reports the Normalized Herfindahl Indices. These are used to account for the smaller amount of countries and industries in this sample. H is the Herfindahl Index and N is the number of countries (industries) in the industry (country) portfolio. This index is also called the adjusted Herfindahl index. These are calculated by using equation (3):

𝑁𝑜𝑟𝑚𝑎𝑙𝑖𝑧𝑒𝑑 𝐻𝑒𝑟𝑓𝑖𝑛𝑑𝑎ℎ𝑙 𝐼𝑛𝑑𝑒𝑥 = [𝐻 −_𝑁1] / [1 −_𝑁1] (3)

The Norwegian country index is the most concentrated country index. This is mainly caused by a very large Oil & Gas industry in Norway. This is confirmed by the results in Panel B as well. Because the Normalized Herfindahl index from the Oil & Gas industry is very high, i.e. 0.25. This is due to the large Oil & Gas industry in Norway as well. The Normalized Herfindahl index from the Health Care industry is relatively high as well, i.e. 0.23. Thus, these two industries are the most concentrated industries across Europe. In other words, the Oil & Gas and Health Care industry are the most geographical concentrated industries across Europe.

(16)

The financials industry is the least geographical concentrated industry in Europe. The Normalized Herfindahl index from this industry is 0.08. The Oil & Gas and Health Care industries have Normalized Herfindahl indices that are about three times as large as the Normalized Herfindahl index from the financials industry. The financials industry is the least concentrated industry across Europe. The financials industry has the largest weight in the European portfolio as well. Thus, it is the biggest sector in Europe. The average standard deviation from the industry portfolios is 3.16%. This is higher than the average standard deviation from the average country portfolios, which is 2.58%. Panel A and Panel B show that this is in line with the higher correlation that the industry portfolios have.

Panel C shows a separation of the full period in a crisis period and in a post-crisis period. The mean returns of the Euro-countries are -0.22% in the crisis period. Although, the non-Euro countries have a mean return of 0.02% during the crisis period. From this can be concluded that the Euro-countries suffered more from the financial crisis compared to the non-Euro countries. Whereas the recovery of the Euro-countries was better compared to the recovery of the non-Euro countries. The mean returns in the Euro-countries increased with 0.51 after the financial crisis. The mean return of the non-Euro countries increased with 0.22. The average standard deviations of both the Euro countries and the non-Euro countries decreased after the crisis. The returns from both these groups of countries became less volatile after the crisis period. Thus, the financial crisis was a period of higher uncertainty and higher volatility. This is reflected in a higher standard deviation. The correlations in the crisis period are higher than the post-crisis period as well. The change in the Normalized Herfindahl index is different for the two groups of countries. For group 1, the index increased from 0.0873 to 0.0988. This means that this the industries in this group of

countries became more concentrated. The opposite holds for the second group of countries. For group 2, the index decreased from 0.1338 to 0.1237. Thus, in these countries the

industries became less concentrated after the crisis. The concentration of the countries in the industry indices increased after the crisis.

Panel D shows the results of the Wald tests. The p-values are only significant at a 5% and 10% level for the hypotheses where all country means are equal to zero, i.e. 0.003, 0.013 and 0.000. We cannot conclude that all country- or industry means are zero, because the p-values are not significant, i.e. 0.465, 0.301 and 0.196.

(17)

4. Methodology

As mentioned before in the introduction, a wide number of studies in this field uses the decomposition method from Heston and Rouwenhorst (1994). In this study, I perform a return-based style analysis. This analysis tests the relative importance of country effects and industry effects in equity returns by comparing average country-specific- and average industry-specific returns’ variances. This analysis is used in the paper of Eiling et al. (2012) as well. However, Sharpe (1988, 1992) use it as first. In this model, there is a null hypothesis that states that a country return is a portfolio of local industry returns. The return of the country portfolio should be perfectly replicated by the local industry returns. Thus, this means that there is not a country-specific effect in the returns of the country portfolio. The opposite should hold as well for the industry return. This industry return, which is European-wide, should be replicated by a portfolio of local industry returns from the same industry across the European countries. This method gives us a tool to set up and study these replicating portfolios.

4.1 Replicating portfolios and simple style regressions

The sample contains 12 countries and 10 industries. Therefore, there are 120 local industry returns across T weekly observations. The analysis is based on this dataset. As mentioned in the data part. The market values of the local industry indices are obtained as well. By using these market values, weights of these local industry indices in the total country and total industry can be constructed. With the weights and the returns of the local industry indices, European value weighted industry portfolios and value weighted country portfolios can be constructed. Thus, referring to the list of returns of these portfolios. Country I’s replicating portfolio returns in terms of European industry returns is estimated in the following regression:

𝑟

_𝑖,𝑡𝑐𝑡𝑟

= ∝

_𝑖

+ ∑

𝑁_𝑗=1



_𝑖,𝑗

𝑟

_𝑗,𝑡𝑖𝑛𝑑

+



_𝑖,𝑡𝑐𝑡𝑟

(4)

s.t. ∑

𝑁



_𝑖,𝑗

(18)

In the regression above, 𝑟𝑖,𝑡𝑐𝑡𝑟 is the return of country i in month t. And 𝑟𝑗,𝑡𝑖𝑛𝑑 is the return of

industry j at time t. The coefficient from the industry j that is relatively large in country i will be relatively large compared to the coefficients of other industries in country i. By imposing the restriction for the regressions, which ensures that all the industry coefficients for country I will add up to 1, the implication is made that a positive weighted portfolio will be constructed. The error term, _𝑖,𝑡𝑐𝑡𝑟_{, can be interpreted as the part of the return that is not}

explained by the industry replicating portfolio. Thus, this can be viewed as the country-specific return for country i. If this error term is 0, so there is no country-country-specific return and no country-specific variance, we can argue that the country return can be replicated using a portfolio of industry returns. The style’s regression R2_{will be equal to 1. From this, there can}

be argued that the variance of the error term can be viewed as a measure of the country specific variance or the industry-specific variance. However, if a country return is more volatile, the variance of the error term will be higher as well. Therefore, the variance of the error term will be scaled by the variance of country i’s return. Thus, (1 - 𝑅𝑐𝑡𝑟,𝑖2 ) will measure

the country-specific variance of country i. The 𝑅𝑐𝑡𝑟,𝑖2 is the R2 of the style regression in

equation (4) where the return of country i is the dependent variable. 4.2 Average country-specific- and industry-specific variances

For each regression that will be performed with equation (4), there will be an estimated R2

as well. By using these R2_{’s, a country-specific variance can be calculated. The average}

country-specific variance will be calculated with equation (5):

𝑉𝐴𝑅

_{𝑐𝑡𝑟_𝑠𝑝𝑒𝑐𝑖𝑓𝑖𝑐}

= ∑

𝐾_𝑖=1

𝑤

_𝑖𝑐𝑡𝑟

(1 − 𝑅

_{𝑐𝑡𝑟,𝑖}2

)

(5)

The same holds for the average industry-specific variance. This will be calculated by using equation (6):

𝑉𝐴𝑅

_{𝑖𝑛𝑑_𝑠𝑝𝑒𝑐𝑖𝑓𝑖𝑐}

= ∑

𝐾_𝑖=1

𝑤

_𝑖𝑖𝑛𝑑

(1 − 𝑅

_{𝑖𝑛𝑑,𝑖}2

)

(6)

4.3 Exclusive style regressions and variance ratios

(19)

reason for this is that industries are not equally distributed across the different countries. This will be solved by excluding an overlapping country or industry from the replicating portfolio.

In equation (5), 𝑤𝑖𝑐𝑡𝑟 is the weight of country i in the European portfolio. This weights

are the average over the sample period for which the regressions are performed. In equation (6), 𝑤𝑖𝑖𝑛𝑑 is the weight of industry j in the European portfolio. This is again the

average over the sample period.

In this study, the equally weighted market weights are used as well. Eiling et al. (2012) did this as well.

Thus, up to now the average industry-specific variance and the average country-specific variances are calculated. Although, to see the relative importance between country effects and industry effects, equation (7) is used. Equation (7) gives a number which

indicates the relative importance.

𝑉𝑎𝑟𝑖𝑎𝑛𝑐𝑒 𝑅𝑎𝑡𝑖𝑜 =

𝑉𝐴𝑅𝑐𝑡𝑟_𝑠𝑝𝑒𝑐𝑖𝑓𝑖𝑐

𝑉𝐴𝑅𝑖𝑛𝑑_𝑠𝑝𝑒𝑐𝑖𝑓𝑖𝑐

(7)

The null hypothesis is that there are no country effects. This means that you can replicate a country return simply by using the industry returns. Under this null hypothesis the ratio is 0. The opposite would that that there are no industry effects. The ratio will go to infinity then. If the variance ratio is 1, the country effects and industry effects are equally important. If the variance ratio is above 1, country effects are dominating industry effects. If the ratio is below 1, the industry effects are larger than the country effects. The variance ratio is the main measure to see the relative importance of country effects vs industry effects in European equity returns.

After the variance ratios are obtained, the significance of these variance ratios is tested. For this, asymptotic standard errors are needed. Details about the calculation of these asymptotic standard errors can be found in the Appendix of Eiling et al. (2012).

As mentioned earlier, all the portfolios are based on the same local industry indices. Therefore, when you regress a certain industry on a certain country. There is always one overlapping local industry index. Therefore, country portfolios and industry portfolios are computed again, although without the overlapping local industry index.

(20)

5. Results

In this section, the results of this study are presented. In the first paragraph, the full set of countries is observed. The coefficients of all the simple- and exclusive style regressions are presented here. Ultimately, the relative importance between country effects and industry effects will be studied by analysing the variance ratios. These variance ratios will be studied in the second paragraph of this section. The final conclusions will be based on the changes in these variance ratios.

5.1 Results of the Simple- and Exclusive style regressions

In this first paragraph the results of the full sample of countries are given. By using the local industry indices in Europe, the replicating country- and industry portfolios are composed. The results of these replicating portfolios are in this paragraph. In table 2 are the results of the regressions of all the replicating portfolios that have been composed presented. All results of the regressions in table 2 are performed over the full period and for all twelve countries used in this study. For each of the countries and industries, there is a simple style regression and an exclusive style regression. For a specific country or industry, the

dependent variable is the same for the simple style regression and the exclusive style regression. However, the difference between these two kind of regressions can be found in the independent variables. In the exclusive style regressions, the overlapping industry or country is removed from all independent variables, which are called the benchmark

portfolios as well. In other words, there are created exclusive portfolios. These portfolios are regressed on the same dependent variable as in the simple style regressions. For example, for the first exclusive style regression of all replicating industry portfolios on the replicating country portfolio of Belgium, the local industry sector in Belgium is removed from all replicating industry portfolios. Thus, for example, there is replicated a European Oil & Gas portfolio without the local Belgium Oil & Gas index. This has been done for all countries and industries. Thus, for the exclusive regressions where the country portfolios will be the dependent variable, there are replicated 120 exclusive industry portfolios. The same holds for the exclusive style regressions with the replicating industry portfolios as the dependent variable. For these exclusive style regressions, there are replicated 120 exclusive country

(21)

portfolios. Thus, in Panel A, the country returns from the dependent variable are removed from the industry portfolio benchmark returns, which are the independent variables. Table 2 – Style analysis for European industries and countries

In this table are the coefficients of the simple style regressions and the exclusive style regressions. The restriction for all these regressions is that the coefficients should be

positive and that they sum to one within each regression. In panel A are the country styles in terms of the industry benchmarks and in panel B are the industry styles in terms of the country benchmarks. In the columns “Sample” are the coefficients of the simple style regressions and in the columns “Excl.” are the coefficients of the exclusive style regressions. For each pair of the “Sample” and “Excl.” regressions there is given a rank correlation coefficient. This is the Spearman rank correlation, which indicates the connection between the simple- and exclusive regression. At the bottom of all the regression coefficients are given the estimates of the country- and industry specific variances. *, ** and *** means significance at a 1%, 5% and 10% respectively. Below the variances are the respective asymptotic standard errors presented. Both the variances and the asymptotic standard errors are given in percentages.

Panel A: Country styles in terms of industry benchmark Belgium Sample Excl. Finland Sample Excl. France Sample Excl. Germany Sample Excl. Intercept 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00

Oil & Gas 0.00 0.00 0.04 0.06 0.03 0.01 0.00 0.03

Basic Materials 0.00 0.22 0.11 0.12 0.00 0.16 0.15 0.12 Industrials 0.16 0.22 0.14 0.31 0.32 0.10 0.42 0.44 Consumer Goods 0.36 0.24 0.00 0.00 0.07 0.00 0.13 0.00 Health Care 0.07 0.08 0.00 0.00 0.00 0.00 0.00 0.10 Consumer Svs. 0.10 0.16 0.00 0.10 0.00 0.00 0.00 0.00 Telecom Utilities Financials Technology Rank correlation (1 - 𝑅_𝑐𝑡𝑟2 ₎ Standard error (%) 0.00 0.01 0.30 0.00 0.82*** 24% 2.05 0.00 0.01 0.28 0.00 25% 2.14 0.00 0.01 0.18 0.52 0.83*** 16% 1.45 0.00 0.00 0.23 0.18 20% 1.76 0.04 0.26 0.23 0.06 0.52 5% 0.45 0.08 0.15 0.41 0.09 9% 0.81 0.06 0.09 0.10 0.05 0.57* 9% 0.88 0.04 0.08 0.19 0.00 16% 1.47 _______________________________________________________________________________

(22)

Panel A (continued) _______________________________________________________________________________ Italy Sample Excl. Netherl. Sample Excl. Norway Sample Excl. Portugal Sample Excl. Intercept 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00

Oil & Gas 0.05 0.02 0.00 0.00 0.64 0.54 0.10 0.10

Basic Materials 0.00 0.03 0.20 0.19 0.30 0.41 0.00 0.00 Industrials 0.04 0.01 0.20 0.32 0.04 0.00 0.19 0.22 Consumer Goods 0.00 0.00 0.06 0.00 0.00 0.00 0.00 0.00 Health Care 0.00 0.00 0.08 0.08 0.00 0.00 0.03 0.04 Consumer Svs. 0.00 0.00 0.11 0.08 0.00 0.00 0.14 0.10 Telecom Utilities Financials Technology Rank correlation (1 - 𝑅𝑐𝑡𝑟2 ) Standard error (%) 0.00 0.37 0.50 0.04 0.84*** 12% 1.10 0.00 0.31 0.57 0.06 16% 1.42 0.02 0.08 0.23 0.04 0.91*** 9% 0.88 0.00 0.09 0.24 0.01 11% 0.99 0.00 0.00 0.02 0.00 0.85*** 19% 1.71 0.00 0.00 0.05 0.00 24% 2.01 0.19 0.22 0.12 0.00 0.97*** 34% 2.75 0.19 0.21 0.13 0.00 35% 2.77 _______________________________________________________________________________ Panel A (continued) _______________________________________________________________________________ Spain Sample Excl. Sweden Sample Excl. Switz. Sample Excl. UK Sample Excl. Intercept 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00

Oil & Gas 0.00 0.00 0.00 0.00 0.00 0.03 0.00 0.18

Basic Materials 0.00 0.00 0.05 0.18 0.00 0.02 0.00 0.00 Industrials 0.00 0.03 0.76 0.58 0.11 0.00 0.00 0.00 Consumer Goods 0.00 0.00 0.00 0.00 0.11 0.04 0.18 0.00 Health Care 0.00 0.00 0.00 0.00 0.58 0.48 0.53 0.34 Consumer Svs. 0.00 0.00 0.00 0.04 0.07 0.13 0.18 0.49 Telecom Utilities Financials Technology Rank correlation (1 - 𝑅_𝑐𝑡𝑟2 ₎ Standard error (%) 0.12 0.36 0.52 0.00 0.85*** 21% 1.79 0.02 0.36 0.59 0.00 27% 2.29 0.00 0.00 0.14 0.06 0.74*** 15% 1.36 0.00 0.00 0.20 0.00 19% 1.67 0.00 0.01 0.12 0.00 0.52 13% 1.17 0.05 0.06 0.14 0.06 23% 1.94 0.11 0.00 0.00 0.00 0.55* 17% 2.96 0.00 0.00 0.00 0.00 39% 1.50 ___________________________________________________________________________

(23)

Panel B: Industry styles in terms of country benchmark OilGas Smpl. Excl. BasMats Smpl. Excl. Indus. Smpl. Excl. ConsGds Smpl. Excl. HealthC Smpl. Excl. Intercept 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 Belgium 0.00 0.00 0.00 0.00 0.00 0.00 0.15 0.02 0.00 0.00 Finland 0.00 0.13 0.06 0.00 0.11 0.06 0.00 0.00 0.00 0.00 France 0.13 0.00 0.00 0.11 0.14 0.00 0.00 0.00 0.00 0.00 Germany 0.15 0.14 0.30 0.25 0.24 0.20 0.51 0.30 0.00 0.13 Italy 0.09 0.00 0.00 0.00 0.00 0.06 0.00 0.00 0.00 0.00 Netherlands 0.00 0.13 0.21 0.00 0.20 0.27 0.00 0.00 0.00 0.00 Norway Portugal Spain Sweden Switzerland United Kingdom Rank correlation (1 - 𝑅_𝑖𝑛𝑑2 ₎ Standard error (%) 0.52 0.03 0.00 0.00 0.05 0.02 0.11 17% 1.5 0.19 0.00 0.03 0.06 0.00 0.00 29% 2.41 0.36 0.00 0.00 0.08 0.00 0.00 0.66** 17% 1.48 0.38 0.00 0.00 0.26 0.00 0.00 22% 1.88 0.03 0.02 0.00 0.25 0.00 0.01 0.69*** 5% 0.51 0.08 0.00 0.00 0.19 0.00 0.14 10% 0.89 0.00 0.00 0.00 0.00 0.25 0.10 0.78*** 14% 1.30 0.00 0.00 0.02 0.07 0.21 0.38 22% 1.90 0.00 0.00 0.00 0.00 0.89 0.11 0.70*** 25% 2.09 0.00 0.02 0.00 0.00 0.77 0.07 44% 3.22 Panel B (continued) ConSv Smpl. Excl. Telecom Smpl. Excl. Utilities Smpl. Excl. Financ. Smpl. Excl. Technol. Smpl. Excl. Intercept 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 Belgium 0.06 0.10 0.00 0.00 0.00 0.00 0.07 0.00 0.00 0.00 Finland 0.08 0.07 0.00 0.00 0.00 0.00 0.05 0.41 0.39 0.11 France 0.20 0.16 0.13 0.06 0.37 0.09 0.04 0.39 0.07 0.24 Germany 0.05 0.01 0.15 0.08 0.00 0.00 0.00 0.11 0.18 0.13 Italy 0.00 0.00 0.00 0.00 0.19 0.31 0.31 0.00 0.00 0.04 Netherlands 0.13 0.02 0.00 0.00 0.00 0.00 0.15 0.00 0.12 0.05 Norway Portugal Spain Sweden Switzerland United Kingdom Rank correlation (1 - 𝑅_𝑖𝑛𝑑2 ₎ Standard error (%) 0.00 0.04 0.02 0.15 0.15 0.12 0.82** 11% 1.23 0.00 0.05 0.03 0.10 0.11 0.34 14% 1.16 0.00 0.08 0.15 0.00 0.34 0.14 0.94*** 32% 2.58 0.00 0.11 0.16 0.00 0.42 0.17 35% 2.76 0.00 0.02 0.13 0.00 0.17 0.11 0.83*** 22% 1.88 0.02 0.00 0.17 0.00 0.29 0.12 26% 2.22 0.00 0.00 0.20 0.18 0.00 0.00 -0.22 12% 1.08 0.08 0.00 0.00 0.00 0.00 0.00 26% 2.20 0.00 0.00 0.00 0.08 0.10 0.06 0.70*** 15% 1.34 0.00 0.00 0.00 0.14 0.08 0.20 20% 1.72 _______________________________________________________________________________________

(24)

The coefficients of the simple style regressions and the exclusive style regressions are compared. Some benchmarks have large weights (coefficients) in the replicating portfolio. Although, this could have two reasons. The first reason could be that that there is a large overlapping component between the replicated industry portfolio and the replicated country portfolio. The second reason could be that the benchmark portfolio can well mimic the returns of the dependent replicated portfolio. Thus, by comparing the coefficients of these simple- and exclusive regressions, the reason behind some large weights in replicating country and industry portfolios is explained in more detail.

The coefficient from the Technology sector in Finland decreased significantly in the exclusive style regression, i.e. from 0.52 to 0.18. The industrials sector in France decreased from 0.32 in the simple style regression to 0.10 in the exclusive style regression. Because the weights/coefficients decrease significantly in the exclusive style regression with respect to the simple style regression, there can be argued that both these results are evidence of large overlapping elements.

If the coefficients of the same industry are all relatively high in the regression in panel A, we can argue that this industry is relatively homogeneous across all countries in Europe. The industry Financials is an example of this. All coefficients are relatively high in both the simple style regressions and the exclusive style regressions. This means that the Financials sector in Europe is relatively important for mimicking returns of country portfolios.

Furthermore, the coefficients do not differ that much in the exclusive regression with respect to the simple regression. This can be observed from the Spearman’s rank correlation coefficient. This measures the correlation between the list of coefficients in the simple regression and the list of coefficients in the exclusive regression. This has been done for each country and for each industry. Except for France, Germany, Switzerland and the UK, all countries have a positive significant rank correlation coefficient at a 1% level. The rank correlation of UK and Germany is significant at a 10% level. For France and Switzerland, they are positive, although they are not significant at all. Thus, the coefficients from the simple- and exclusive regressions have not a significant correlation among each other. Large overlapping components between these two country portfolios and the ten replicated industry portfolios could be a reason for this. Furthermore, rank correlation coefficients are strong if they are between 0.80 and 1.00. Indeed, most of the rank correlation coefficients are in this range. Thus, this is evidence for a strong correlation between these groups of

(25)

coefficients. For the industry styles with the country returns used as a benchmark, all rank correlation coefficients are positive except for the Financials sector. This is due to a change in the coefficients of the replicating country portfolios in the exclusive style regressions.

Below the rank correlation coefficient, table 2 reports the (1 – R2_{). For each country}

and industry, the (1 – R2_{) is higher for the exclusive style regression compared to the simple}

style regression for a specific country of industry. This is due to a higher R2_{in the simple}

regressions than in the exclusive regression. The reason for this is that the simple style regressions include the overlapping components as well. Therefore, the explanatory power of the simple style regression is always larger than the explanatory power of the exclusive style regression for a certain country or industry.

For example, the (1-R2_{) for the simple style regression on the replicating country}

portfolio of Belgium is 24%. This means that the replicated industry portfolios explain 76% of the variation in the Belgium index returns. Thus, 24% of the variation in the Belgium index returns is not captured by the replicated industry portfolios. The 24% is called the country-specific variance of Belgium. If the country-country-specific variance is relatively high, for instance for Portugal or for the exclusive regression of the UK, there can be argued that the variation of the returns of the replicated country portfolio contains relatively large country effects. This is because a relatively large part of the country-specific variance cannot be captured by the benchmark industry portfolios. Moreover, the countries France, Germany and the

Netherlands have the lowest country-specific variances among all countries, i.e. 5%, 9% and 9%. This is due to the large extend of financial market integration of these three economies in the European economy. In that sense, Portugal is a relatively surprisingly result. They have the highest country-specific variances among all countries except for the exclusive

regression of the UK. Although, Portugal are using the Euro, they still have higher country-specific variances than countries not using the Euro as their main currency.

Furthermore, Belgium has the highest Herfindahl index among the Euro countries in group 1. Thus, the industry composition in Belgium is slightly concentrated, which results in relatively large country effects compared to other countries that have a smaller industry

concentration.

Table 2 reports the the asymptotic standard errors in percentages at the bottom of each panel. The asymptotic standard errors are calculated by using the Appendix B of the paper of Eiling et al. (2012). These standard errors show how accurate the estimation is. The

(26)

country- and industry-specific variances are all statistical significant different from 0. This means that a country portfolio cannot be viewed as a portfolio of different industries and that an industry portfolio cannot be seen a portfolio of countries. This holds especially for countries that have higher country-specific variances. These countries have less linkages to the other European countries.

If the portfolios used as benchmark portfolios do not contain parts of the test asset, i.e. the dependent portfolio, like in the exclusive style regression. Then it is more difficult to mimic the test portfolio. This makes sense, because the identical overlapping components between the test asset and the benchmark portfolios are removed from the benchmark portfolios in the exclusive regression. All the industry-specific variances are at least 10 standard errors away from 0, thus highly statistical significant as well. Therefore, there can be concluded that there are significant industry effects as well. Furthermore, all industry-specific variances are below 30%, except for the exclusive regression of HealthCare and both the simple- and exclusive regressions of the Telecom industry.

Table 2 shows only results for the full period. However, in table 3 the analysis is extended with results for the sub-periods as well. Ultimately, we want to show how the variance ratios, which are depending on the variances, change over time. Table 3 reports the country- and industry-specific variances for the group of all countries in the full period, the crisis period and the post-crisis period. In more detail, table 3 reports the value-weighted averages and the equally-weighted averages. Below each average is the corresponding asymptotic standard error given. All results in table 3, except for the p-value, are given in percentages. To calculate the value weighted averages, the average weights of the countries and industries in the total European portfolio are used. These average weights are based on the market values for all local industry portfolios in the Euro-wide portfolio.

For the value weighted approach, the average country-specific variance is 14% and the average industry-specific variance is 16%. These variances given are scaled by the total country- and industry variance. This means that when the full period is observed, the industry effects are more important than the country effects in European equity returns. These results are in line with the results of Chou et al. (2014). These results can be assigned to an increased financial market integration in Europe. Although, the results of the crisis period and the post-crisis period show that the variances have changed over time. The average country-specific variance has grown from 12% in the crisis period to 20% in the

(27)

post-crisis period. This is a relatively large increase of the average country-specific variance. The average industry-specific variance has decreased from 16% to 15%. For the equally weighted approach, the country specific variance increased from 15% to 20% and the

industry-specific variance was staying almost equal at 17%. The main conclusion drawn from panel A is that the country-specific variance has increased in both the equally weighted- and value weighted approach. In the next section, which is about the variance ratios, all these results are explained in more detail. Because the variance ratios are based on all results shown in this section. The relatively small standard errors show that all country-specific- and industry-specific variances are significant different from 0. Thus, there could be argued that country effects and industry effects are indeed present in this period.

Exact the same analysis is performed for the sub-sample periods. This has been done for the crisis period and for the post-crisis period, to see the change in the country- and industry specific variances over time. The p-values show if the changes in the average variances over time are significant or not. All changes in the average country-specific variances are significant at a 5% level. The largest p-value among the average

country-specific variances is for the equally weighted approach in the exclusive style regressions. This p-value is 0.048. Thus, there can be argued that the crisis did cause a significant change in the average country-specific variance among the European countries in our sample. This in contrast to the changes of the average industry-specific variances. The p-values for these changes are all above 0.7. Thus, these changes are all far from significant. Therefore, the crisis had more effect on the country-specific variances compared to the European industry-specific variances. Further on, there is shown if the changes in the country-industry-specific variances causes significant changes in the variance ratios. Furthermore, table 3 shows that all

variances given are at least more than 2 standard errors away from 0. Therefore, for all variances, the null hypothesis that there are no country-specific and industry-specific variances can be rejected. This is proof that during our sample period country effects and industry effects are present.

In both the equally weighted approach and the value weighted approach, the average country-specific variances are increasing. Therefore, they show the same pattern. However, the increase in the value-weighted approach is larger than the increase in the equally-weighted approach. For example, observing the increase in the simple style regressions, the value-weighted approach shows an increase from 12% to 20% and the

(28)

equally weighted approach shows an increase from 15% to 20%. This could be due to the UK in our sample. This country has a relative large weight in the sample and has a relatively large increase in the country-specific variance. Thus, the equally-weighted approach diminishes this increased effect that is caused by the UK.

Table 3 – Average industry-specific and country-specific variances for all countries In this table are variances and asymptotic standard errors between brackets. All these numbers or for all 12 countries in the sample. These variances and asymptotic standard errors are reported for the value weighted portfolios and for the equally weighted

portfolios. The column “Full period” is for the full period in the sample. The column called “Crisis” is the first period during the crisis, this is from January 2008 till September 2011. The “Post-Crisis” period is from October 2011 till December 2015. A country-specific variance is estimated by (1 – R2) for a specific regression with a country portfolio as the dependent

variable. The same holds for the industry-specific variance. The value weighted average variances are calculated by taking a value-weighted average of the country-specific or industry specific variances. The same holds for the equally-weighted variances. The p-value is reported in the last column. This is the p-value for a test of H0: the average country- or

industry-specific variance is equal in the crisis period and in the post-crisis period. This would mean that there is no significant change in the average country- or industry-specific

variance.

Panel A: Simple style regressions – All countries

Value weighted Full period Crisis Post-Crisis P-value

VARctr_spec st. error VARind_spec st. error 14% (1%) 16% (1%) 12% (1%) 16% (2%) 20% (2%) 15% (2%) 0.000 0.726

Equally weighted Full period Crisis Post-Crisis P-value

(29)

________________________________________________________________

Panel B: Exclusive style regressions – All countries

VARctr_spec 22% 18% 24% 0.048 st. error VARind_spec st. error (2%) 24% (3%) (2%) 24% (2%) (3%) 23% (2%) 0.761

(30)

5.2 Variance ratios

Table 4 reports the variance ratios. These variances ratios are calculated by using equation (7). This is the average country-specific variance over the average industry-specific variance. Panel A reports the simple variance ratios. These ratios are based on the simple style

regressions. The variance ratio over the full sample period and for all countries is 0.89. This means that overall industry effects dominate country effects in European equity returns during the full sample period. More precise, it means that during the full sample period, the average country-specific variance is 11% smaller than the average industry-specific variance. Although, when the sample of all countries is divided in Euro countries and

non-countries, the results look different. The variance ratio for the group 1 non-countries, the Euro-countries, is 0.68. Thus, the industry effects dominate even more in this group. This is due to a common currency and more integrated financial markets of these countries. In contrast to group 1, group 2 has a variance ratio of 1.54. Country effects dominating over industry effects within this group of countries. These countries all have their own currency and have more financially segmented markets in contrast to the financial markets of the countries in group 1.

Panel B in table 4 reports the exclusive variance ratios. For the sample containing all countries, these variance ratios lead to the same conclusion as the simple variance ratios. The exclusive variance ratio is 0.95, which is slightly higher than the simple variance ratio of 0.89. This means that the exclusion of overlapping components caused a larger decrease the average industry-specific variance compared to the average country-specific variance. In the value weighted exclusive regressions, the variance ratios are increasing as well. This holds for the whole sample of all countries, group 1 and group 2. Although, the increase of the exclusive variance ratios in panel B are slightly smaller compared to the increase of the simple variance ratios in panel A. The p-value for the change in the simple variance ratios and exclusive variance ratios in both the value weighted- and the equally weighted approach are rounded to 0.000. This means that, from the crisis period to the post-crisis period, there is a significant increase in all these variance ratios. This supports the statement of an

increase in the relative importance of country effects over industry effects in European equity returns. Furthermore, the changes in the variance ratios from the crisis period to the

(31)

post-crisis period in absolute terms are larger for the group 1 countries than for the group 2 countries. Thus, the crisis has had more impact on the variance ratios for the Euro countries compared to the non-Euro countries.

For the equally weighted variance ratios, the previous statement still holds. However, the increase of both the simple- and exclusive variance ratios in the equally weighted

approach is smaller compared to the increase of the variance ratios in the value weighted variance approach.

In almost all cases, the variances ratios are at least 2 standard errors from one. Except for some of the variance ratios. If this is the case and the variance ratio is not significant different from 1, we can argue that the average country-specific variance is not significant different from the average industry-specific variance. Some of the variance ratios in this analysis are not significant different from 1. For example, in the crisis period, the equally weighted simple variance ratio of the group 2 countries is not significant different from 1. This variance ratio is 0.99 and the standard error is 0.03. The same holds for the exclusive variance ratio of group 2 in the crisis period. The same variance ratios in the value weighted approach are much larger, respectively 1.41 and 1.50. Thus, the average country-specific variance is relatively larger given that these ratios are above 1. This could be due to a large impact of the UK on the average country-specific effects, which is caused by a relatively large weight of the UK in our sample.

All variance ratios in table 4 are significant different from 0. In other words, they are at least two standard errors away from 0. This means that in all cases country-specific effects are present.

Table 4 – Variance Ratios: the relative importance of country effects over industry effects This table reports average variance ratios for three groups of countries. These groups are all countries together, the Euro-countries (Group 1) and the non-Euro-countries (Group 2). These variances ratios have been calculated for the full sample period, the crisis period (January 2008 – September 2011) and the post-crisis period (October 2011 – December 2015). The variance ratio is calculated by dividing the country-specific variance by the industry-specific variance, as mentioned before in the methodology. The country-specific variance is estimated as the (1-R2) of a style regression of country returns on industry

benchmark returns. For the industry-specific variance this will be done vice versa. This ratio measures the relative importance between these country effects and industry effects. If the variance ratio becomes very large, country effects are dominating. If the variance ratio goes to 0, the country effects are negligible and industry effects are dominating. If the variance

(32)

ratio is 1, both effects will have the same level of importance. The value-weighted variance ratios are based on the average weights of countries and industries in the European

portfolio. For the equally-weighted variances ratios, all industries will have the same weight and all countries will have the same weight. Thus, the country weights are (1/number of countries) and the industry weights are (1/number of industries). Below each variance ratio and between parentheses, the asymptotic standard errors are reported. The last column in this table reports p-values. These p-values are under the null hypothesis that the variance ratio in the crisis period and the post-crisis period is equal. In panel A are the simple variance ratios, these are based on the variances from the simple style regressions. Panel B reports the exclusive variance ratio’s; these are based on variances of the exclusive style regressions. In the exclusive style regression, the overlapping component from the test asset is removed from the benchmark portfolios.

Panel A: Simple Variance Ratios

All countries 0.89 0.77 1.35 0.000 Group 1 (0.02) 0.65 (0.03) 0.61 (0.02) 0.89 0.000 (0.03) (0.02) (0.03) Group 2 1.54 (0.03) 1.41 (0.04) 2.03 (0.03) 0.000

All countries 0.94 0.88 1.17 0.000 Group 1 (0.03) 0.89 (0.03) (0.05) 0.85 (0.02) (0.04) 1.11 (0.04) 0.000 Group 2 1.08 (0.03) 0.99 (0.03) 1.28 (0.04) 0.000

(33)

Panel B: Exclusive Variance Ratios

(34)

5.3 Robustness check

In the main analysis, the simple- and exclusive variance ratios are already calculated with two different approaches, the value weighted approach and the equally weighted approach. The equally weighted approach can be viewed as a robustness check for the value weighted approach. The value weighted approach is the most accurate approach, because this is based on the actual market weights of the respective industries and countries in the Euro-wide portfolio. To check if countries with relative large or small weights have a significant impact on the significance of our results, the equally weighted approach is performed as well.

Furthermore, the results of the simple variance ratios are most important to look at, however the exclusive variance ratios are calculated as well. This could be viewed as a robustness check for the simple variance ratios as well. By comparing the results of the exclusive variance ratios and the simple variance ratios, the impact of the overlapping components in a country- and industry portfolio can be observed.

However, before the conclusion will be drawn, based on the results in the previous section, the robustness of the results is tested by one more robustness check. The

robustness check is performed by excluding the UK from the sample. This decision is based on the relatively large weight of the UK in our sample and the relatively large amount of country-specific variance contained by the UK is a reason as well. The aim is to observe if the conclusions based on the variance ratios in the previous section still holds after excluding the UK from the sample.

To evaluate the robustness of the results in the previous section. The results of table 5 should be compared to the results in table 4. This is done by comparing the p-values. In table 4, all p-values are highly significant at a 1% significance level. They are all rounded to 0.000. However, in table 5, the p-value for the change in the simple variance ratio in the value weighted approach is 0.026. This change is not significant at a 1% level. We could therefore argue that the increase in the average country-specific variance is not that large when excluding the UK from the sample. The change is still significant at a 5% and 10% level, however not at a 1% level. This was a result that we could have expected, however we can still conclude that these results are robust, because the significance at 5% level is still there.

The relative importance of country effects vs industry effects in European equity returns