The Effect of the Financial Crisis on the Relative Importance of Country versus Industry Effects in Euro-zone Equity Returns

(1)

The Effect of the Financial Crisis on the Relative Importance of

Country versus Industry Effects in Euro-zone Equity Returns

Klaas van Bork 10444017 14/8/2018

Master Thesis Finance, Asset Management Faculty of Economics and Business (FEB) University of Amsterdam

Supervisor: Jeroen Jansen

(2)

2

Abstract

The relative importance of country and industry effects on international equity returns has been a point of discussion over the past decades. Questions in this debate concern whether country or industry effects are more important, as well as what may drive changes in the relative importance of these two. In this paper we focus on the financial crisis and with that the level of equity market integration as a potential moderator of the relative importance of country versus industry effects. The research in this paper is conducted on ten countries and ten industries in the European Monetary Union (EMU) over the 1999-2018 period. We hypothesized that industry effects are dominant in explaining international equity returns before the start of the financial crisis in 2008, but that country effects dominate international equity returns after the start of the financial crisis. To test the relative importance of country versus industry effects a variance ratio is constructed using style analysis. The non-parametric bootstrap, non-parametric bootstrap and the permutation test are used to test for differences in the variance ratio before and after the start of the financial crisis. Our findings suggest that industry effects are more important than country effects in explaining international equity returns in the EMU before and after the start of the financial crisis. In the five years after the start of the financial crisis industry effects even gain in relative importance. This increase of relative importance of industry effects however seems to have been temporarily. We did not find the expected opposite effect for the level of equity market integration. However, exploratively zooming in on the relationship between the level of equity market integration and the relative importance of country versus industry effects, we did find a negative relationship over time. These explorative findings suggest that there may be a relationship between equity market integration and country versus industry effects but that we did not find it in our data. Further research could possibly confirm this relationship.

Statement of Originality

This document is written by Klaas van Bork who declares to take full responsibility for the contents of this document. I declare that the text and the work presented in this document are 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)

3

I Introduction

Research on equity returns has become increasingly important over the past decades as the amount and scale of equity investments has increased rapidly over this period. Research on this matter mainly focusses on the strategic aspect of equity investment. Harry Markowitz (1952) introduced Modern Portfolio Theory, which is a mathematical framework that captures investment strategy, based on the risk-return profile, assuming investors are risk averse. A key insight of Modern Portfolio Theory is that the risk-return profile of a single asset should be assessed by its contribution to the risk-return profile of a portfolio. In other words, a substantial part of the risk of a single asset can be eliminated by holding the asset in a well-diversified portfolio, simply indicating that strategic diversification can eliminate risk. This fueled the debate on strategic diversification.

Country and industry diversification

Academic literature on this topic focusses on two popular diversification strategies. Geographic diversification and industry diversification. With geographical diversification, often referred to as country diversification, the assets in a portfolio are diversified over different geographical regions (e.g., Eiling, Gerard, & De Roon, 2011; Heston & Rouwenhorst, 1994; Moerman, 2008). This way, a (political, economic, etc.) shock in the returns of an asset in one geographical region can be balanced out by assets located in other (not affected) regions. Industry diversification is the industry equivalent of this strategy, in which the assets in a portfolio are diversified over different industries. Whether to diversify by country or by industry depends on the influence of country versus industry effects on the stability of returns. If a shock in a specific country (country effect) has a large impact on the returns on an asset in that country, diversification by country should be most effective in stabilizing portfolio returns. Over the past decades a lot of research focused on the question whether country effects or industry effects have a bigger impact on international equity returns. Literature on the question whether country or industry effects are more important is conflicting (see Table 1). An explanation for these conflicting results may be that the influence of these factors can vary over time due to the fact that international industrial structure does not remain constant over a very long time (Roll, 1992). A recent paper by Eiling et al. (2011) suggests that, with the introduction of the Euro in 1999, there has been a shift from country effects to industry effects having a dominant influence on equity returns for the Euro-area. This might indicate that investors should now diversify more by industry to eliminate risk.

(5)

5

Financial crisis

The financial crisis started with the global collapse of financial markets in the last quarter of 2008. A key event was the bankruptcy of investment bank Lehman Brothers Holdings Inc., mainly due to the bank’s activity in the US subprime mortgage market. The fall of Lehman Brothers led to a major loss of trust in the financial system and in financial institutions. Due to, among others, a sharp drop in international trade, the financial crisis sparked a global economic crisis (Milesi-Ferretti & Tille, 2011).

In Europe, the financial and economic crisis developed into what is known as the sovereign debt crisis. In 2009 the European sovereign debt crisis was fueled by a number of European countries that reported larger deficits than expected. For example, that year, the newly elected Greek government forecasted a budget deficit of more than 12%. In addition, deficits of previous years were also revised to be higher than reported. These developments, in combination with a deteriorating banking sector led to a sharp increase of the annual spread on ten-year sovereign bond yields between Germany and countries such as Greece, Ireland, Portugal, Spain and Italy (Lane, 2012). The global financial crisis caused a significant decrease in cross-border financial flows in Europe. Investors re-evaluated international risk exposure and preferred a domestic allocation of capital. This was especially troublesome for countries that relied more heavily on external funding (Milesi-Ferretti & Tille, 2011).

The financial crisis caused a loss of trust in the European financial system. Would there have been a European sovereign debt crisis if European countries still owned their own currencies, if they were still able to make their own policies and act without having to follow rules set by the European Union? The belief that the European Union only created the problems instead of preventing them, became increasingly popular over the past decade. Since the financial crisis Euroscepticism has increased significantly. The trust in the European Union among European citizens has declined strongly since a peak shortly before the financial crisis (European Commission, 2015). The popularity of Eurosceptic political parties, such as UKIP (United Kingdom), Front National (France), Fidesz (Hungary), PVV (Netherlands) and MoVimento 5 Stelle and Lega Nord (Italy), has risen strongly since 2008. Most illustrative is the referendum held in the UK in 2016 on the continued membership in the European Union, which was voted against by 51.9% of the voters leading to the UK leaving the European Union (Brexit).

(6)

6

Research question

As described above, the financial crisis, resulting in an economic crisis and in the European sovereign debt crisis, severely harmed the trust in the financial system, financial institutions, as well as the trust in a European Union. With that, the sovereign debt crisis led to a sharp increase of the annual spread on ten-year sovereign bond yields in the European Monetary Union (EMU). The financial crisis caused a significant decrease in cross-border financial flows in Europe (Milesi-Ferretti and Tille, 2011). An important question is to what extent the financial crisis caused a dispersion between and disintegration of financial markets in the EMU. In the following we will investigate whether the financial crisis moderated the relative importance of country and industry specific factors in international equity returns in the European Monetary Union. More specifically, the research question answered in this paper is:

Did country effects become more important than industry effects in explaining international equity returns in the European Monetary Union (EMU) as a result of the financial crisis?

Method and data

Most of the prior research on the topic use the multi-factor approach introduced by Heston and Rouwenhorst (1994). However, recently this method was criticized by among others Brooks and Del Negro (2002) for being too restrictive. One of the reasons is the assumption that a company in the dataset can only represent one country and one industry, which for example means that all companies listed in the US are constrained to one “US” country factor. With multinationals and cross listed companies this assumption is too restrictive and is strongly rejected by the data presented by Brooks and Del Negro (2002).

This paper follows a method demonstrated by Eiling et al. (2011) to investigate the research question. In this method a variance ratio is constructed, measuring the relative importance of country versus industry effects in explaining international equity returns. More specifically, international equity returns will be regressed on country returns and on industry returns in two separate regression models, both resulting in one coefficient for the explained variance in the model. The ratio of these country and industry specific explained variances reflects the relative importance of country versus industry effects. The regression models are estimated in a returns-based style analysis (Sharpe, 1992).

(7)

7

The dataset will consist of monthly returns of ten country indices and ten industry indices. The time frame for which the data is collected is from 1999 until 2018.

First, in section II, the existing literature will be reviewed ending with the main hypotheses tested in this paper as well as a table summarizing the reported results of previous research. In section III the research method used in this paper is discussed in more detail, after which an overview of the data will be presented. Thereafter, in section IV the empirical results are discussed. Section V discusses some concluding remarks based on the reported results. Finally, section VI presents some relevant points for discussion.

(8)

8

II Literature Review

Over the past decades, the literature on the country-industry debate, regarding the factors of influence on international equity returns, has led to a broad understanding on this topic. Roll (1992) showed that stock markets reflect the idiosyncrasies of the country's industrial structure, contributing to the importance of industry effects. Heston and Rauwenhorst (1994) found evidence in a sample period from 1978 through 1992 suggesting that, for twelve European countries that are in many respects similar in terms of economic policies, country effects have a dominant influence on stock returns. Using the Dow Jones World Stock Index, these findings were later replicated by Griffin and Karolyi (1998).

Later research focused more on the globalization and integration of capital markets and their influence on the importance of country and industry effects. Cavaglia, Brightman, and Aked (2000) presented evidence that with globalization industry effects became more and more important and may now even be more influential than country effects. Brooks and Del Negro (2004), however, came to the conclusion that the rise of industry effects is a temporary phenomenon. They argue that the increase in importance of industry effects is a result of the IT-hype, which is temporary. Bekaerd et al. (2009) found results that were in line with the conclusion of Brooks and Del Negro (2004). Their findings show the continuing importance of country-specific factors, suggesting that the benefits of international diversification have persisted despite globalization. Their findings suggest that despite globalization, country-specific factors still have a dominant influence on equity returns.

Moerman (2008) and Eiling et al. (2011) focus on the role of the establishment of a monetary union in Europe, with respect to globalization and the integration of financial markets. In contrast to the results of Brooks and Del Negro (2004) who concluded a dominant effect of country effects, Moerman (2008) shows that industry effects dominate throughout a sample from 1995 to 2004. So, while Brooks and Del Negro proposed the IT-hype as an explanation for a temporary increase in the importance of industry effects, Moerman argues that the IT sector cannot explain away the importance of the industry effects. Moerman demonstrates that correcting for the IT sector does not influence the results and thus concludes that industry effects have a dominant influence on European equity returns.

Eiling et al. (2011), examined the importance of country and industry effects on equity returns for the Euro-area over the 1990-2008 period. Their findings suggest that in the pre-Euro period (1990-1998) country effects are dominant, while after the

(9)

9

introduction of the Euro, industry effects are more important. According to Eiling et al., this shift from country effects to industry effects is driven primarily by countries that were least integrated in the Economic and Monetary Union (EMU) and world markets in the early 1990s and for which the EMU convergence process led to rapid strengthening of linkages with the core Euro-zone. Moreover, they present evidence that for countries with less volatile currencies and strong economic links with Germany before 1990, the industry effects prevail over the whole sample period. Contrarily, for countries with high exchange rate volatility and weaker economic links with Germany before 1990, the country effects dominate before the introduction of the Euro, while the industry effects become more important thereafter.

When looking at the findings over the past decades (see Table 1), an important conclusion to draw is that the importance of country versus industry effects does not seem to be constant over time, but is rather changing. Some argue that these changes are temporarily and probably due to specific events (Brooks & Del Negro, 2004). Others argue that the importance of industry effects is increasing due to globalization and that the importance of industry effects is a sign of the integration of international equity markets (Cavaglia et al., 2000; Moerman, 2008; Eiling et al., 2011).

In addition, we see that the previous studies have different samples. Most of the samples consist of a relatively small amount of developed countries (global, Table 1), which is comparable to a sample of Euro-zone countries. One important difference is that the Euro-zone countries form a monetary union (Economic and Monetary Union, EMU), which might be an indication that the equity markets in the EMU are more internationally integrated. With the recent financial crisis in the Euro area this level of integration may have been affected due to the increased financial risk and even the probability of default for certain EMU countries. In the current paper we contribute to this line of research by investigating whether the financial crisis changed the effects of country and industry factors, with an eye on the underlying level of integration of financial markets. We do this by following similar methods as Eiling et al. (2011) in their study about the effect of the introduction of the euro on the effects of country and industry factors. Table 1 provides a summary of the findings on the influence of country versus industry effects, as discussed above.

(10)

10 Table 1: Former findings

Author(s) Region Year Sample Period Explained

Variable Stated Effect Roll 24 Countries (Global) 1992 1988 - 1991 Equity returns Industry-specific Heston & Rouwenhorst 12 Countries (Europe) 1994 1978 - 1992 Equity returns Country-specific

Griffin & Karolyi 25 Countries (Global)

1998 1991 - 1995 Equity

returns

Country-specific

Cavaglia et al. 21 Countries (Global) 2000 1986 - 1999 Equity returns Industry-specific (due to globalization) Brooks & Del Negro 42 Countries (Global) 2004 1985 - 2002 Equity returns Country-specific Moerman 11 Countries (EMU-zone) 2008 1995 - 2004 Equity returns Industry-specific (since adoption of the Euro in 1999) Bekaert et al. 23 Countries

(Global) 2009 1980 - 2005 Equity returns Country-specific (despite globalization)

Eiling et al. 11 Countries (EMU-zone) 2011 1990 - 2008 Equity returns Industry-specific (since adoption of the Euro in 1999) Financial integration

The importance of country versus industry effects seems to be changing over time. Since 2000, most of the literature suggests either an increase in the importance of industry effects relative to country effects or industry effects already prevailing over country effects. Globalization and the integration of international equity markets are often proposed as possible explanations for the increase in the importance of industry effects. For instance, Eiling et al. (2011) argue that one of the main drivers of the importance of country versus industry effects for countries in the EMU is the integration of equity markets. The introduction of the Euro as a common currency improved financial integration, causing industry effects to become more dominant over time.

Bekaert and Harvey (1995) argue that the integration of capital markets is indeed varying over time. A point brought forward by Bekaert and Harvey is that the level of financial integration is hard to measure and dependent on one’s definition of financial integration. They note two commonly used measures for financial integration. Firstly, the correlation of local market returns with world returns. However, they argue that

(11)

11

when a country’s industry mix differs from the world’s industry mix, the correlation can be low while the country may in fact be perfectly integrated. The second measure uses investment restrictions as an indicator of integration. A flaw of this measure is that some restrictions have a bigger impact than other restrictions. Also, some restrictions are not binding to investors and they may be able to access the national market in other ways (Bekaert & Harvey, 1995).

Later research focussed on creating a clear definition of financial integration to be able to present a fitting measure. According to Adam, Jappelli, Menichini, Padula, and Pagano (2002) a financial market is integrated when the law of one price holds. They argue that different financial markets, such as the bond market, equity market or credit market, have different levels of financial integration, which should be measured separately. In this paper the focus is on financial integration of equity markets. In the case of financial integration of equity markets, the law of one price implies that stocks that create identical cashflows should yield identical returns, regardless of the country the stocks are issued in (Adam et al., 2002). This in fact closely relates to the country versus industry debate. Financial integration of equity markets in the European Union should imply that stock market returns become more correlated among countries (Adam et al., 2002). The extent to which industry factors have a dominant influence on international equity returns (and therefore the lack of influence of country factors) can thus be seen as a measure of financial integration.

Baele, Ferrando, Hördahl, Krylova, and Monnet (2004) agree with Adam et al. on the fact that different financial markets have different levels of financial integration, which should be measured separately. They define financial integration as a situation where all market participants face the same set of rules when dealing with financial instruments or services, are treated equally in the market and have equal access to financial instruments or services (Baele et al., 2004). They present several measures for equity market integration. Country versus industry effects is one of their measures for the integration of equity markets.

Problematic is the fact that these indirect methods of measuring financial integration actually measure a consequential effect. That is, if country versus industry effects is used as a measure for financial integration, it cannot be compared to the relative importance of country and industry effects itself. In that case it is not possible to test whether financial integration has a causal relationship with country versus industry effects. This underlines the importance of the definition of financial integration and the corresponding measure.

(12)

12

Beale et al. (2004) also present a news based measure where the level of integration of equity markets is measured based on the co-movement of returns of a country with the region of interest (world, EMU, etc.). Eiling et al. (2011) adopt this measure to be able to test whether the importance of country versus industry effects is related to the level of integration of equity markets. Their findings suggest that the level of integration of equity markets is consistent with the relative importance of country versus industry effects.

Hypotheses

To answer the research question of this paper, different aspects of the question will be tested. To test the relative importance of country versus industry effects the methodology presented by Eiling et al. (2011) will be used. Returns-based style analysis, will be used to estimate country and industry specific variance. Then, a variance ratio will be constructed, measuring the relative importance of country versus industry effects.

As the research question implies, the event of interest in this paper is the financial crisis with its consequences for the European Monetary Union. The effect of the financial crisis on the relative importance of country versus industry effects has several aspects. The sovereign debt crisis led to a sharp increase of the annual spread on ten-year sovereign bond yields between Germany and countries such as Greece, Ireland, Portugal, Spain and Italy. Also, the financial crisis caused a significant decrease in cross-border financial flows in Europe. Investors preferred a domestic allocation of capital (Milesi-Ferretti and Tille, 2011). Furthermore, the financial crisis fueled Euroscepticism and the rising popularity of Eurosceptic political parties. Concluding, there are strong indications that the financial crisis causes segmentation of European financial markets. With the disintegration of equity markets we expect country effects to be more important than industry effects in explaining international equity returns in the European Union as a result of the financial crisis. Based on previous literature we expect industry effects to be more important than country effects in explaining international equity returns in the European Monetary Union before the financial crisis. The global collapse of financial markets in the last quarter of 2008 (01-10-2008) will be marked as the start of the financial crisis. This translates into the following hypotheses.

(H1) Industry effects are more important than country effects in explaining international equity returns in the European Monetary Union before the financial crisis.

(13)

13

(H2) Country effects are more important than industry effects in explaining international equity returns in the European Monetary Union after the start of the financial crisis. (H3) The level of equity market integration is lower after the start of the financial crisis than before the start of the financial crisis.

(14)

14

III Methodology

In this section the research method is explained. First, the statistical model that is used to analyze country and industry effects is presented. Then, a brief description of the data is given. The section concludes with a summary of the data statistics.

Style analysis

Style analysis, brought forward by Sharpe (1992), is an important tool for understanding and being able to measure the performance of a financial portfolio for which the composition is unknown. A style is referring to the composition of a financial portfolio. Different styles may be Small cap stocks versus Mid cap or Large cap stocks or Value stocks versus growth stocks. But styles may also imply stocks from different regions or different industries. As the composition of a financial portfolio is often only known by the portfolio manager, style analysis aims to provide a tool to measure and compare the performance of a portfolio with an unknown composition with respect to different portfolios. To do so, Sharpe (1992) used a simple factor model, with the Capital Asset Pricing Model (CAPM) as its backbone:

𝑟𝑖,𝑡 = 𝛽𝑖,1𝐹1+ 𝛽𝑖,2𝐹2+ 𝛽𝑖,3𝐹3+ ⋯ + 𝛽𝑖,𝑛𝐹𝑛+ 𝜀𝑖 (1)

Where:

𝑟𝑖,𝑡 = return of portfolio i over month t

𝐹𝑛 = factor of influence

𝛽𝑖,𝑛 = corresponding coefficient

𝜀𝑖 = disturbance term

Withlimited information on a specific financial portfolio’s composition, return based style analysis uses historical returns to construct a portfolio of indices that best replicates the historical returns of the target portfolio (Otten & Bams, 2001). In this factor model (equation (1)) each of the factors refers to an asset class or portfolio with a specific style. The coefficients represent the sensitivity of a funds or financial portfolio’s returns to these styles (Sharpe, 1992).

To analyze the style of a financial portfolio, a least squares constrained regression model is used to create a replicating portfolio (Sharpe, 1992). The constraints, posed on the regression βs from equation (1), are that they must be nonnegative and together must sum up to 1. The intuition behind these constraints is that the estimated coefficients can be interpreted as composition quotas of a replicating portfolio (∑𝑁𝑗=1𝛽𝑖,𝑗= 1) and to reflect the short-selling constraint most fund managers are subject to (𝛽_𝑖,𝑗 ≥ 0) (Sharpe,

(15)

15

1992). Dealing with these constraints implies a mathematical optimization problem. Quadratic programming provides a tool to solve the optimization problem to be able to use style analysis (Sharpe, 1992).

To measure the performance of a fund or financial portfolio, the returns are compared to those of the replicating portfolio provided by the style regressions. A fund manager influences the selection process. A fund manager determines the style of a fund and then chooses a selection of stocks within the style. The style is therefore not part of the manager’s skill. Only the selection process of stocks within the fund is. If the style regression portfolio is able to perfectly replicate the fund’s performance, the fund did not outperform the replicating portfolio and the fund’s performance can be dedicated to the style and not to the selection of specific stocks by the fund manager. Looking at Equation (1) the variance of a funds return left unexplained by the styles (factors), is captured by the residuals (𝜀_𝑖). Therefore, (1 − 𝑅2_{) of equation (1) can be interpreted as} the part of the variance of a funds return left unexplained by the styles and thus explained by the managers selection of specific stocks.

Industry versus country effects

In the case of country versus industry effects, return based style analysis actually provides a tool to investigate whether the variance in country returns can be replicated by the returns of a portfolio of industry indices or whether the variance in industry returns can be replicated by the returns of a portfolio of country indices.

To examine the importance of country versus industry effects, the variance of country-specific returns will be compared to the variance of industry-specific returns. To estimate these two variances, returns-based style analysis will be used (Sharpe, 1992). With the following model for country- and industry-specific returns the variances will be estimated using style regression (Eiling et al., 2011).

𝑟𝑖,𝑡𝑐𝑡𝑟= 𝛼𝑖+ ∑ 𝛽𝑖,𝑗𝑟𝑗,𝑡𝑖𝑛𝑑 𝑁 𝑗=1 + 𝜀𝑖,𝑡𝑐𝑡𝑟 (2) Where: 𝛼𝑖 = intercept

𝑟𝑖,𝑡𝑐𝑡𝑟 = return of country i over month t

𝑟𝑗,𝑡𝑖𝑛𝑑 = return of industry j over month t

𝛽𝑖,𝑗 = corresponding coefficient

(16)

16

The basic idea of this method is to see whether industry-portfolios are better at replicating country return patterns or vice versa. If a portfolio of industry indices is perfectly replicating a country’s return pattern, the variance of that country’s returns is entirely explained by industry return variance and the R² of the regression will be 1. This indicates that there is no variance in the country’s return left unexplained by the variance in industry returns which can be interpreted as the country having no effect on its own return variation and thus a lack of country effects. The other way around, if a portfolio of industries is not able to replicate a country’s return pattern, the part of the country’s return variance which was left unexplained by the return variance of the industry indices will be high and the R² will be low. Put differently, (1 − 𝑅𝑐𝑡𝑟,𝑖2 ) is a measure of country specific variance for country i, where 𝑅_{𝑐𝑡𝑟,𝑖}2 _{is the R² of the style regression in equation} (2). The same method is used to measure industry specific variance (1 − 𝑅𝑖𝑛𝑑,𝑗2 ) for industry j, by replicating industry returns with a portfolio of country returns. As mentioned before, replicating returns with a portfolio of returns imposes two restrictions on the β of the style regression. The sum of the βs must be 1 and each individual β must be nonnegative. This implies that a positive weight (𝛽𝑖,𝑗≥ 0) portfolio (∑𝑁𝑗=1𝛽𝑖,𝑗 = 1) is created. With ten countries and ten industries, the style regression in equation (2) should produce ten country-specific variances and ten industry-specific variances. The weighted average country-specific variance is then calculated as:

𝑉𝐴𝑅𝑐𝑡𝑟= ∑ 𝑤𝑖𝑐𝑡𝑟 𝑁

𝑗=1

(1 − 𝑅_{𝑐𝑡𝑟,𝑖}2 ) (3)

Where:

𝑉𝐴𝑅𝑐𝑡𝑟 = country specific variance

𝑤𝑖𝑐𝑡𝑟 = weight of country i.

𝑅𝑐𝑡𝑟,𝑖2 = R² of the style regression in equation (1)

The weights corresponding to the different countries are based on their market cap within the EMU portfolio as a whole (value weighted). Again, the weighted average specific variance is obtained identical to that of countries, with industry-specific variances (1 − 𝑅𝑖𝑛𝑑,𝑗2 ) and weights based on the market cap of each industry within the EMU portfolio as a whole (𝑤_𝑗𝑖𝑛𝑑_{). This weighted average country- or} industry-specific variance is a measure of country- or industry effects. To measure their relative importance, a variance ratio is constructed. The average country specific variance is divided by the average industry-specific variance:

𝑉𝑎𝑟𝑖𝑎𝑛𝑐𝑒 𝑟𝑎𝑡𝑖𝑜 =𝑉𝐴𝑅𝑐𝑡𝑟 𝑉𝐴𝑅𝑖𝑛𝑑

(17)

17

The variance ratio gives us an interpretation of the importance of country versus industry effects. If the weighted average country specific variance is lower than the weighted average industry specific variance (𝑉𝐴𝑅_𝑐𝑡𝑟< 𝑉𝐴𝑅𝑖𝑛𝑑), the variance ratio will be between 0 and 1, which indicates industry effects are more important. A ratio above 1 (𝑉𝐴𝑅_𝑐𝑡𝑟> 𝑉𝐴𝑅𝑖𝑛𝑑) indicates country effects are more important. A ratio of 1 (𝑉𝐴𝑅_𝑐𝑡𝑟 = 𝑉𝐴𝑅𝑖𝑛𝑑) indicates that country and industry effects are equally important (Eiling et al., 2011).

To test for changes in the variance ratio, the style regressions are performed on the following four different subsamples. The Pre-crisis subsample, consisting of all observations dating from before the start of the financial crisis (1-2-1999 until 1-9-2008). The Post-crisis subsample, consisting of all observations dating from after the start of the financial crisis (1-10-2008 until 1-5-2018). The 5-year pre-crisis subsample and the 5-year post-crisis subsample, consisting of observations from the 5 years prior to the start of the financial crisis (1-10-2003 until 1-9-2008) and observations from the 5 years directly after the start of the financial crisis (1-10-2008 until 1-9-2013) respectively.

Integration of equity markets

The level of integration of equity markets will be tested by comparing country return variance to the world market return variance. More specifically, the share of country return variance which can be explained by international (world) factors. In order to achieve this, the following regression is used:

𝑟_𝑖,𝑡𝑐𝑡𝑟= 𝛼𝑖+ 𝛽𝑖𝑟𝑊,𝑡+ 𝜀𝑖,𝑡 (5)

Where:

𝛼𝑖 = intercept

𝑟𝑖,𝑡𝑐𝑡𝑟 = return of country i over month t

𝑟𝑊,𝑡 = world market return over month t

𝛽𝑖 = corresponding coefficient

𝜀𝑖,𝑡 = disturbance term

From this regression the R² can be used as a measure of financial integration within the world market. The R² of equation (5) is the part of the variance in country returns, explained by the variance in the world market returns. The weighted average R² over the countries functions as a measure of financial integration of equity markets in the EMU.

(18)

18

Test for differences in variance ratios

In the previous section we defined the variance ratio as a parameter resulting from the style analysis that reflects the replicating abilities of country versus industry. This section sets out three methods for testing whether the variance ratio differs before and after the crisis. To see whether the variance ratio has changed over time due to the financial crisis, the variance ratio from the period before the start of the financial crisis is compared to the variance ratio from the period after the start of the financial crisis. To test whether the two observed ratios are different we estimate a confidence interval for the observed variance ratios. When the confidence interval of the variance ratio before the crisis and after the crisis do not overlap, we reject the null hypothesis that the variance ratio is the same and that the observed difference is merely due to sampling variability. The following three techniques construct a sampling distribution of either the variance ratio (parametric and non-parametric bootstrap) or of the difference between the variance ratios (permutation test) to test the null hypothesis 𝐻₀: Variance ratio before crisis = Variance ratio after crisis. Throughout this paper the difference in variance ratios is tested based on a 95% confidence interval (α = 0.05).

Test 1: Non-parametric bootstrap

The first test for differences in variance ratios is the non-parametric bootstrap analysis. The method relies on the sample to function as the population, which is unknown. The following steps demonstrates non-parametric bootstrapping as a resampling technique to create a sampling distribution and construct a confidence interval (which is called a Bootstrap Percentile confidence interval; Efron & Tibshirani, 1994) for the observed variance ratios:

1. Divide the observations in the dataset into two groups:

Dataset 1 – all observations from the period before the start of the financial crisis Dataset 2 – all observations from the period after the start of the financial crisis For each of the datasets conduct the following steps separately:

2. Randomly draw n observations from the dataset with replacement creating bootstrap sample b, where n is equal to the number of observations in the corresponding Dataset 1 or 2.

3. Perform the style regressions as discussed above on the 𝑏𝑡ℎ_{bootstrap sample,} resulting in 𝑉𝑎𝑟𝑖𝑎𝑛𝑐𝑒 𝑟𝑎𝑡𝑖𝑜𝑏₌𝑉𝐴𝑅𝑐𝑡𝑟𝑏

𝑉𝐴𝑅_𝑖𝑛𝑑𝑏 , corresponding to bootstrap sample b. 4. Repeat the process explained in steps 2 and 3 B times, where B denotes the

(19)

19

5. Create a sampling distribution using the B variance ratios and construct a Bootstrap Percentile confidence interval based on a chosen value for alpha (e.g., α = 0.05).

The test will result in a sampling distribution and a Bootstrap Percentile confidence interval for both the variance ratio before the start of the financial crisis and the variance ratio after the start of the financial crisis. To see whether the variance ratio before the start of the financial crisis differs from the variance ratio after the start of the financial crisis, we compare the confidence intervals. If the confidence intervals overlap the null hypothesis that there is no difference between the variance ratios cannot be rejected. If the confidence intervals do not overlap we reject the null hypothesis (Efron & Tibshirani 1994).

Test 2: Parametric bootstrap

The parametric bootstrap is in many ways similar to the non-parametric bootstrap. It differs in that it uses the parameters from the estimated model fitted on the data to generate new data. In the case of generating equity returns for a sample consisting of ten countries and ten industries and assuming the returns are multivariate normally distributed, the parameters of interest are the mean return per country/industry and their covariances. That is, the returns in the bootstrap samples should have the same mean and covariance structure (covariance matrix) as the original dataset. The parametric bootstrap process used in this paper is demonstrated by the following steps:

1. Divide the observations in the original dataset into two groups:

Dataset 1 – all observations from the period before the start of the financial crisis Dataset 2 – all observations from the period after the start of the financial crisis For each of the datasets conduct the following steps separately:

2. Estimate the mean return per country/industry and the covariance matrix from the data.

3. Simulate n observations with the same covariance structure and means as the original data by multiplying the Cholesky decomposition of the estimated covariance matrix with p variables ~𝛮(μ,1), creating bootstrap sample b, where μ is the mean of the variable in the original dataset and where n is equal to the number of observations in Dataset 1 or 2 correspondingly.

4. Perform the style regressions as discussed above on the 𝑏𝑡ℎ_{bootstrap sample,} resulting in 𝑉𝑎𝑟𝑖𝑎𝑛𝑐𝑒 𝑟𝑎𝑡𝑖𝑜𝑏₌𝑉𝐴𝑅𝑐𝑡𝑟𝑏

(20)

20

5. Repeat the process explained in steps 2 and 3 B times, where B denotes the number of bootstrap samples (e.g., B = 10.000).

6. Create a sampling distribution using the B variance ratios and construct a Bootstrap Percentile confidence interval based on a chosen value for alpha (e.g., α = 0.05).

Note that by multiplying the Cholesky decomposition of the covariance matrix with p variables that have variance one (~𝛮(μ,1)), the variance of the resulting variables is the variance of the original variables multiplied by one, so that random variables are obtained that have the same covariance structure and the same variances as the original variables. Similar to the non-parametric bootstrap the test results in a sampling distribution and a Bootstrap Percentile confidence interval for both the variance ratio from before the start of the financial crisis and the variance ratio from after the start of the financial crisis. The confidence intervals are compared to test for a difference in variance ratios. If the confidence intervals overlap, the null hypothesis that there is no difference between the variance ratios cannot be rejected. If the confidence intervals do not overlap we reject the null hypothesis (Efron & Tibshirani 1994).

Test 3: Permutation test

The third resampling technique is the permutation test. The permutation test is a non-parametric statistical significance test in which we assume under the null hypothesis that labels on the data are interchangeable. In the case of variance ratios we want to test whether there is a difference in the variance ratio from the period before the start of the financial crisis is compared to the variance ratio from the period after the start of the financial crisis. Here “before” and “after” the start of the financial crisis are the labels. Under the null hypothesis we assume that there is no difference between the variance ratios for the groups with these labels. To test this hypothesis we want to estimate a distribution and confidence interval under the null hypothesis. The permutation test is demonstrated in the following steps:

1. Label the observations corresponding to “before” and “after” the start of the financial crisis.

2. Create two samples, each consisting of half of the observations. That is, randomly draw n/2 observations without replacement creating permutation sample p1, where n is equal to the number of observations in the original dataset. The

(21)

21

remaining observations, also consisting of n/2 observations, form permutation sample p2.

3. Perform the style regressions as discussed above on the 𝑝1𝑡ℎ_{and 𝑝2}𝑡ℎ permutation sample, resulting in 𝑉𝑎𝑟𝑖𝑎𝑛𝑐𝑒 𝑟𝑎𝑡𝑖𝑜𝑝1₌𝑉𝐴𝑅𝑐𝑡𝑟𝑝1

𝑉𝐴𝑅_𝑖𝑛𝑑𝑝1 and 𝑉𝑎𝑟𝑖𝑎𝑛𝑐𝑒 𝑟𝑎𝑡𝑖𝑜𝑝2₌𝑉𝐴𝑅𝑐𝑡𝑟𝑝2

𝑉𝐴𝑅_𝑖𝑛𝑑𝑝2, corresponding to permutation sample p1 and p2. 4. Repeat the process explained in steps 2 and 3 P times, where P denotes the

number of permutation samples (e.g., P = 10.000).

5. Create a sampling distribution under the null hypothesis by subtracting p1 from p2 using the P variance ratios and construct a confidence interval based on a chosen value for alpha (e.g., α = 0.05).

Note that the distribution under the null hypothesis should be distributed around zero. The null hypothesis is rejected if the difference between the originally observed variance ratios is not within the constructed confidence interval. The p-value is given for the permutation test. For the bootstrap, the p-values are not reported. In the permutation test we use resampling to simulate the null process to then estimate a distribution under the null hypothesis. This is not the case for the non-parametric and parametric bootstrap (Efron & Tibshirani 1994). The R code that implements these functions can be found in the appendix (R Core Team, 2018).

Analyses in R

The analyses described above are all performed in the statistical software program R (R Core Team, 2018). For the style analyses, we adapted code that was made publicly available on GitHub by Systematicinvestor (url: https://github.com/systematicinvestor-/SIT/raw/master/sit.gz). The adapted R code for these constrained regressions is included in the appendix. For the parametric bootstrap test we used the function cov() to estimate the covariance matrix of the data, which uses maximum likelihood estimation. The bootstrap tests and the permutation test rely on Monte Carlo simulations that we performed in the statistical program R. The R code that implements the bootstrap tests and the permutation test to test for a difference in the variance ratios can be found in the appendix as well.

(22)

22

Data

Data on EMU industry indices and EMU country indices will be retrieved from Datastream for the period 1999-2018. The date of the first and last observation in the dataset are 1-2-1999 and 1-5-2018 respectively. The collected indices are Morgan Stanley Capital International (MSCI) equity indices. A benefit of using the MSCI indices is that the countries and industries overlap. For example the MSCI EMU industry indices consist only of assets from the original EMU countries. The EMU country and industry indices cover approximately 85% of the free float-adjusted market capitalization of the EMU. The data will be collected for the original EMU countries that adopted the Euro in 1999. The ten original EMU countries are Germany, France, Netherlands, Belgium, Italy, Ireland, Spain, Portugal, Austria and Finland. Note that Luxembourg was excluded from the dataset.

One of the benefits of working with a dataset that consist of EMU countries is that the influence of currency risk on country and industry effects can be neglected. With cross-border capital flows, different exchange rates may affect returns. However, with all EMU countries using one single currency, exchange rate risk does not play a role in cross border capital flows and therefore does not affect returns.

The industries are based on the “Sector” classification in Global Industry Classification Standard (GICS), which is an industry classification system developed in 1999 by MSCI and Standard & Poor's (S&P). These sectors are Energy, Materials, Industrials, Consumer Discretionary, Consumer Staples, Financials, Health Care, Information Technology, Telecom Services, Utilities and Real Estate. However, the Real Estate sector was added as a sector in 2015 while it was positioned under the financial sector as an “Industry Group” before 2015. In the EMU MSCI sector indices Real Estate is still positioned under Financials and therefore excluded from the dataset leaving us with ten industries. The data are monthly total return indices with dividends reinvested from January 1999 until May 2018 adding up to a total of 233 months.

For the weighted average country and industry specific variances, the weights are based on the market cap of each of the countries and industries with respect to the market cap of the EMU as a whole. Due to limited access to information on the composition of MSCI indices, these weights are based on information published by MSCI per March 2018.

To measure the financial integration of equity markets data on world market equity returns are gathered. The MSCI World ex EMU index covers the world market and excludes the EMU market. This causes the EMU dataset and the World dataset to be non-overlapping. This is required to avoid biased results when regressing country returns on

(23)

23

world market returns. Data on the MSCI World ex EMU index consist of monthly total returns with dividends reinvested and is also retrieved from Datastream for the period 1999-2018. All returns on the MSCI indices used in this research are Euro denominated returns. Preferably, when measuring the integration of equity markets, the returns of the EMU countries would be compared to local returns denominated in local currencies to eliminate the currency factor. Using Euro denominated returns causes the exchange rates of local currencies with the Euro to influence the rate of return. However, to obtain all returns in local currencies is a complex task. In this research we chose to use the MSCI World ex EMU index denominated in Euros. In the following an overview of the summary statistics is presented.

(24)

24

Table 1: Summary Statistics of Country and Industry Portfolio Returns for the EMU

The table shows the summary statistics for the returns on ten country indices and ten industry indices. The monthly returns are collected from Datastream over the 1999-2018 period, reporting a total of 232 observations per country/industry. Mean(%) shows the average gross return per country/industry per month over the whole sample period. Std.(%) shows the corresponding standard deviation of the returns over the sample period. Weight is the percentage share of the market cap within the total market cap of countries/industries per March 2018. Corr ctr and Corr ind are the average correlations of the returns of country/industry portfolios with the returns of other country portfolios or industry portfolios respectively. The row “Average” shows the average value of the Mean(%), Std.(%), Corr ctr and Corr ind.

Obs. Mean Std. Weight Corr ctr Corr ind

Industries (%) (%) (% 2018) Energy 232 0.72 5.48 5.72 0.55 0.51 Materials 232 0.85 6.41 8.59 0.70 0.65 Industrials 232 0.88 6.10 14.75 0.76 0.69 Consumer Discretionary 232 0.66 6.35 14.24 0.73 0.68 Financials 232 0.35 7.41 21.98 0.75 0.65 Health Care 232 0.60 4.61 7.57 0.48 0.47 Consumer Staples 232 0.57 4.18 9.75 0.59 0.56 Information Technology 232 0.60 8.75 8.43 0.64 0.57 Telecom Services 232 0.33 6.65 3.87 0.53 0.45 Utilities 232 0.48 5.31 5.11 0.64 0.58 Average 0.60 6.13 0.64 0.58 Countries Italy 232 0.30 5.76 7.73 0.71 0.68 Germany 232 0.60 6.12 29.74 0.73 0.74 France 232 0.57 5.08 33.23 0.77 0.77 Spain 232 0.54 5.94 9.64 0.69 0.65 Belgium 232 0.42 5.53 3.13 0.66 0.62 Netherlands 232 0.56 5.37 10.82 0.74 0.73 Austria 232 0.64 6.58 0.78 0.61 0.55 Portugal 232 0.09 5.33 0.46 0.58 0.54 Finland 232 0.81 8.88 3.10 0.55 0.57 Ireland 232 0.07 6.34 1.38 0.56 0.51 Average 0.46 6.09 0.66 0.64 Summary statistics

In the previous paragraph the data was presented. This paragraph reports the summary statistics of the data sample. The summary statistics for the whole sample period are presented in Table 1 after which they are discussed. The summary statistics for the pre-crisis and post-pre-crisis sample period are also briefly discussed in this paragraph. These summary statistics are reported in Table A1 and A2 in the Appendix.

Table 1 shows the summary statistics of country and industry portfolio returns for the whole sample period. From the total returns, the monthly gross returns were calculated. This leaves us with 232 observations per country and per industry. The highest average return per month over the sample period (1999-2018) is reported by Industrials (0.88%), closely followed by Materials (0.85%). The lowest average returns per month are reported by Financials (0.35%) and Telecom Services (0.33%). Among countries, Finland (0.81%) had the highest average return per month over the sample

(25)

25

period. The lowest average returns per month are reported by Ireland (0.07%) followed by Portugal (0.09%).

The weights are based on the share of the market cap of each of the industry indices and country indices in the total market cap of the MSCI EMU index. Comparing the weights of industries and countries per 30 April 2018, the weights of industries are more evenly distributed. Financials is the industry with the largest market cap. Financials has a 21.98% share in the total market cap of the MSCI EMU index. Other industries with a large share in the market cap of the MSCI EMU index are Industrials (14.75%) and Consumer Discretionary (14.24%). Telecom Services has the smallest market cap, with a share of 3.87%. Among countries the weights have a larger distribution, ranging from 0.46% (Portugal) to 33.23% (France). France and Germany (29.74%) are the largest contributors to the total market cap of the MSCI EMU index. Notable is the fact that some of the larger firms in these countries, for example Total S.A. (France, Energy), SAP SE (Germany, Information Technology) and Allianz Group (Germany, Financials) have a larger market cap, with a share of 3.05%, 2.18% and 2.12% respectively, than some of the countries within the MSCI EMU index.

Table A1 and A2 in the Appendix show the same summary statistics as Table 1, for the period before the start of the financial crisis (01-10-2008) and after the start of the financial crisis respectively. The average return for countries, as well as for industries, is higher after the start of the financial crisis than before the start. This might seem counterintuitive, but with both subperiods ranging almost ten years, the effect of the financial crisis might be nullified by the recovery over this period in terms of returns.

Ireland is the only country with a negative average return over the period before the financial crisis. Finland and Austria both show an average monthly return of just over 1%. However, these countries make up a relatively small part of the total MSCI EMU index. France and Germany, together accounting for approximately two thirds of the total market cap of the MSCI EMU index, both have average monthly returns of just under 0.5% before the start of the financial crisis. After the start of the financial crisis, France and Germany have average monthly returns of 0.68% and 0.74% respectively.

Comparing standard deviations, or volatilities in the case of equity indices, mentionable is the high standard deviation of Information Technology before the start of the financial crisis and the high standard deviation of Financials after the start of the financial crisis. These relatively high standard deviations are possibly caused by the IT-bubble, affecting the IT stocks and the financial crisis, affecting financial stocks.

Corr ctr and Corr ind (Table 1) are the average correlations of the returns of country/industry portfolios with the returns of other country portfolios or industry

(26)

26

portfolios respectively. The average cross-country correlation over the whole sample period is 0.66. The average cross-industry correlation over the whole sample period is 0.58. The average country correlation being higher than the average cross-industry correlation means that returns show more comovement among countries than among industries. This might be a first indication that diversifying among industries eliminates more risk than diversifying among countries over the whole sample period. Before the start of the financial crisis the average cross-country correlation is 0.62 and the average cross-industry correlation is 0.53 (see Table A1 in Appendix). After the start of the financial crisis the average cross-country and cross-industry correlations are 0.72 and 0.66 respectively (see Table A2 in Appendix). Notable about these correlations is that the average cross-country and cross-industry correlation both increased over time during the sample period. This may be a sign of increased integration of equity markets in the EMU area over the sample period. In the next section the results are reported and discussed.

(27)

27

IV Results

In the previous section the method was introduced and an overview of the data and its statistics was presented. In this section the results are presented and discussed. First, the style regression coefficients and variance ratios are discussed. With that, the non-parametric bootstrap, non-parametric bootstrap and permutation test results are reported. Thereafter, the level of equity market integration is discussed and compared to the results of the style analysis.

Pre-crisis versus post-crisis

Table 2 shows the style regression results for the sample period before the start of the financial crisis. In Panel A the results are shown for the style regressions where the MSCI country returns were regressed on the MSCI industry returns. In Panel B the results are shown for the style regressions where the MSCI industry returns were regressed on the MSCI industry returns. The value weighted average R² in Panel A and in Panel B are 0.89 and 0.71 respectively. The equally weighted average R² in Panel A and in Panel B are 0.74 and 0.64 respectively. These results suggest that a replicating portfolio of industry indices is better in explaining country specific variance in returns than a replicating portfolio of country indices is at explaining industry specific variance in returns. This is translated in the value of the observed variance ratio.

Table 3 shows the variance ratios and their distribution properties as estimated with the non-parametric and parametric bootstrap and the permutation test. The value weighted variance ratio before the start of the financial crisis is 0.37. A variance ratio between 0 and 1 indicates that the return variation replicating abilities of a portfolio of industry indices is better than that of a portfolio of country indices. That is, industry effects are more important than country effects. The confidence intervals as estimated with the non-parametric and parametric bootstrap are 95% confidence interval (CI) [0.290, 0.444] and 95% CI [0.299, 0.439] respectively. Both confidence intervals are entirely between 0 and 1. Industry effects are significantly more important than country effects. These results are in line with the first hypothesis (H1) that industry effects are more important than country effects in explaining international equity returns in the European Monetary Union before the start of the financial crisis.

Looking at the style regression coefficients in Panel A the financial sector on average reports the largest coefficients, suggesting it is important in replicating country returns. With the financial sector being the largest sector in the sample based on market cap this follows our predictions.

(28)

28 Table 2: Pre-crisis Style Regression Coefficients

The table shows the style regression coefficients for the sample period before the start of the financial crisis. Panel A reports the style regression coefficients where the returns of the MSCI country indices are regressed on the returns of the MSCI industry indices. Panel B reports the style regression coefficients where the returns of the MSCI industry indices are regressed on the returns of the MSCI country indices. The style regression coefficients are constrained to be nonnegative and sum up to one to replicate a positive weight portfolio. The last row in Panel A and Panel B reports the R² for each of the style regressions.

Panel A: Country styles Pre-crisis

Ita Ger Fra Spa Bel Ned Aus Por Fin Ire

intercept 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.01 -0.01 Energy 0.17 0.00 0.14 0.02 0.00 0.18 0.28 0.00 0.09 0.05 Materials 0.00 0.17 0.00 0.00 0.12 0.05 0.23 0.00 0.00 0.24 Industrials 0.18 0.15 0.13 0.06 0.00 0.08 0.02 0.25 0.00 0.03 ConsDiscr 0.00 0.23 0.13 0.05 0.00 0.15 0.02 0.00 0.00 0.00 Financials 0.25 0.18 0.17 0.37 0.35 0.28 0.00 0.00 0.00 0.12 HealthCare 0.00 0.00 0.13 0.00 0.06 0.00 0.08 0.11 0.00 0.12 ConsStaple 0.10 0.00 0.07 0.05 0.36 0.22 0.19 0.13 0.01 0.19 InfoTech 0.00 0.05 0.07 0.00 0.00 0.05 0.00 0.00 0.88 0.00 Telecom 0.20 0.14 0.11 0.27 0.00 0.00 0.00 0.21 0.02 0.10 Utilities 0.11 0.07 0.05 0.19 0.11 0.00 0.20 0.30 0.00 0.14 R² 0.78 0.94 0.96 0.79 0.77 0.91 0.43 0.55 0.84 0.43

Panel B: Industry styles Pre-crisis

Energy Mat Indus ConsD Finan Health ConsS Techno Telcom Utilities

intercept 0.00 0.01 0.01 0.00 0.00 0.00 0.00 0.00 0.00 0.00 Italy 0.00 0.00 0.14 0.01 0.01 0.00 0.00 0.00 0.00 0.10 Germany 0.00 0.41 0.57 0.62 0.33 0.00 0.00 0.28 0.15 0.00 France 0.09 0.00 0.00 0.01 0.00 0.38 0.07 0.00 0.00 0.27 Spain 0.00 0.00 0.01 0.09 0.09 0.00 0.00 0.00 0.44 0.04 Belgium 0.00 0.12 0.00 0.00 0.28 0.32 0.41 0.00 0.00 0.18 Netherlands 0.49 0.28 0.15 0.24 0.28 0.00 0.33 0.00 0.00 0.00 Austria 0.42 0.16 0.00 0.00 0.00 0.11 0.12 0.00 0.00 0.16 Portugal 0.00 0.00 0.10 0.00 0.00 0.12 0.07 0.00 0.13 0.21 Finland 0.00 0.00 0.03 0.03 0.00 0.00 0.00 0.72 0.29 0.00 Ireland 0.00 0.03 0.00 0.00 0.00 0.06 0.00 0.00 0.00 0.05 R² 0.42 0.75 0.87 0.88 0.87 0.08 0.53 0.88 0.64 0.50

The coefficients in Panel B show that Germany and the Netherlands on average report the largest coefficients, implicating they are important in replicating industry returns.

Table 4 shows the style regression results for the sample period after the start of the financial crisis. Looking at the style regression coefficients in Panel A and Panel B, the results are similar to style regression results for the sample period before the start of the financial crisis. The financial sector on average reports the largest coefficients, suggesting it is important in replicating country returns. Germany and France on average report the largest coefficients in portfolios replicating industry returns.

(29)

29 Table 3: Variance Ratio

The table shows the variance ratio for each of the tested sample periods. Panel A reports the “Value weighted” and “Equally weighted” variance ratio for the sample period before the start of the financial crisis (Pre-crisis) and the sample period after the start of the financial crisis (Post-crisis). Panel B reports the “Value weighted” and “Equally weighted” variance ratio for the 5-year sample period before the start of the financial crisis (5-year Pre-crisis) and the 5-year sample period after the start of the financial crisis (5-year Post-crisis). The 95% confidence intervals (CI) for the non-parametric bootstrap, parametric bootstrap and the permutation test are reported within brackets. The row “p-value” reports the p-values for the permutation test. For “Value weighted” the weights in the variance ratio corresponding to the different countries or industries are based on their market cap within the EMU portfolio as a whole. For “Equally weighted” each of the countries and industries have an equal weight in the variance ratio. The row “Difference” reports the difference between the Pre-crisis variance ratio and the Post-crisis variance ratio.

Panel A Pre-crisis Post-crisis

Value weighted variance ratio 0.370 0.333

Non parametric bootstrap 95%CI [0.290, 0.444] [0.249, 0.431]

Parametric bootstrap 95%CI [0.299, 0.439] [0.271, 0.391]

Difference -0.037

Permutation test 95%CI under 𝐻0 [-0.112, 0.112]

p-value (0.536)

Equally weighted variance ratio 0.729 0.722

Difference -0.007

p-value (0.972)

Panel B 5-year Pre-crisis 5-year Post-crisis

Value weighted variance ratio 0.423 0.252

Difference -0.171*

p-value (0.026)

Equally weighted variance ratio 0.769 0.549

Difference -0.220

p-value (0.205)

* significant at α = 0.05

The value weighted average R² in Panel A and in Panel B are 0.93 and 0.78 respectively. The equally weighted average R² in Panel A and in Panel B are 0.80 and 0.72 respectively. These results again suggest that a replicating portfolio of industry indices is better in explaining country specific variance in returns than a replicating portfolio of country indices is at explaining industry specific variance in returns.

The value weighted variance ratio after the start of the financial crisis is 0.33. For the non-parametric and parametric bootstrap the estimated confidence intervals are 95% CI [0.249, 0.431] and 95% CI [0.271, 0.391] respectively (see Table 3).

(30)

30 Table 4: Post-crisis Style Regression Coefficients

The table shows the style regression coefficients for the sample period after the start of the financial crisis. Panel A reports the style regression coefficients where the returns of the MSCI country indices are regressed on the returns of the MSCI industry indices. Panel B reports the style regression coefficients where the returns of the MSCI industry indices are regressed on the returns of the MSCI country indices. The style regression coefficients are constrained to be nonnegative and sum up to one to replicate a positive weight portfolio. The last row in Panel A and Panel B reports the R² for each of the style regressions.

Panel A: Country styles Post-crisis

Ita Ger Fra Spa Bel Ned Aus Por Fin Ire

intercept 0.00 0.00 0.00 0.00 0.00 0.00 0.00 -0.01 0.00 -0.01 Energy 0.24 0.00 0.16 0.00 0.00 0.06 0.03 0.23 0.07 0.02 Materials 0.00 0.29 0.01 0.00 0.01 0.00 0.41 0.00 0.03 0.00 Industrials 0.00 0.05 0.24 0.00 0.11 0.37 0.04 0.04 0.00 0.45 ConsDiscr 0.01 0.32 0.06 0.00 0.00 0.00 0.00 0.00 0.01 0.00 Financials 0.50 0.08 0.18 0.49 0.13 0.10 0.38 0.08 0.06 0.07 HealthCare 0.00 0.14 0.10 0.00 0.00 0.05 0.00 0.00 0.00 0.11 ConsStaple 0.00 0.03 0.11 0.00 0.57 0.27 0.00 0.28 0.00 0.14 InfoTech 0.00 0.02 0.05 0.00 0.17 0.13 0.15 0.00 0.72 0.21 Telecom 0.10 0.00 0.08 0.35 0.00 0.03 0.00 0.12 0.00 0.00 Utilities 0.15 0.07 0.00 0.16 0.00 0.00 0.00 0.25 0.11 0.00 R² 0.93 0.96 0.97 0.90 0.74 0.89 0.74 0.54 0.83 0.49

Panel B: Industry styles Post-crisis

Energy Mat Indus ConsD Finan Health ConsS Techno Telcom Utilities

intercept 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 Italy 0.11 0.00 0.00 0.00 0.53 0.00 0.00 0.00 0.00 0.04 Germany 0.00 0.81 0.37 0.84 0.00 0.30 0.05 0.13 0.00 0.12 France 0.70 0.00 0.21 0.03 0.00 0.27 0.17 0.00 0.23 0.00 Spain 0.00 0.00 0.00 0.00 0.26 0.00 0.00 0.00 0.37 0.35 Belgium 0.00 0.00 0.00 0.00 0.00 0.18 0.36 0.03 0.01 0.08 Netherlands 0.00 0.00 0.22 0.00 0.00 0.10 0.23 0.29 0.16 0.00 Austria 0.00 0.18 0.08 0.04 0.21 0.00 0.00 0.01 0.00 0.00 Portugal 0.18 0.00 0.00 0.00 0.00 0.15 0.20 0.00 0.23 0.24 Finland 0.02 0.01 0.09 0.09 0.00 0.00 0.00 0.53 0.00 0.18 Ireland 0.00 0.00 0.03 0.00 0.00 0.00 0.00 0.01 0.00 0.00 R² 0.66 0.89 0.93 0.87 0.87 0.48 0.51 0.87 0.41 0.73

With the confidence intervals both being strictly below 1 these results suggest that after the start of the financial crisis industry effects are significantly more important in explaining equity return variation than country effects. These findings are not in line with the second hypothesis (H2) that country effects are more important than industry effects in explaining international equity returns in the European Monetary Union after the start of the financial crisis.

(31)

31 Figure 1: Estimated Distribution of the Variance Ratio

The figure shows the distribution of (the difference in) the variance ratio as estimated by the non-parametric bootstrap (Panel A), the parametric bootstrap (Panel B) and the permutation test (Panel C). Panel A and Panel B show the estimated distribution of the variance ratio before the start of the financial crisis in blue (Pre-crisis) and the estimated distribution of the variance ratio after the start of the financial crisis in red (Post-crisis). The constructed confidence intervals based on the estimated distributions are presented by the blue (Pre-crisis) and red (Post-crisis) dashed lines. Panel C shows the distribution of the difference in the pre-crisis and post-crisis variance ratio under the null hypothesis that there is no difference in the variance ratio before the start of the financial crisis and the variance ratio after the start of the financial crisis. The confidence interval as estimated by the permutation test is presented by the dashed red lines. The difference in the pre-crisis and post-crisis variance ratio observed in the data is presented by the solid red line.

The findings discussed above suggest that industry effects prevail over the whole sample period. However, the variance ratio may still have changed over time. The non-parametric and non-parametric bootstrap and the permutation test are three different techniques used in this paper to test for differences in the variance ratio. The estimated distribution of the variance ratio before the start of the financial crisis (0.37) is compared to that of the variance ratio after the start of the financial crisis (0.33). These estimated distributions are shown in Figure 1. Table 3 gives an overview of the results with the corresponding estimated confidence intervals. The 95% confidence interval, estimated with the non-parametric bootstrap, of the variance ratio before the start of the financial crisis is [0.290, 0.444], where after the start of the financial crisis the estimated 95% confidence interval for the variance ratio is [0.249, 0.431]. These confidence intervals overlap (Figure 1: Panel A). Therefore, based on the distribution estimated by the non-parametric bootstrap the null hypothesis, that there is no difference between the

The Effect of the Financial Crisis on the Relative Importance of Country versus Industry Effects in Euro-zone Equity Returns