Developed stock market integration : recent change in the importance of global factors for international stock returns

(1)

Master Thesis

Developed stock market integration: Recent change in the

importance of global factors for international stock returns

MSc Finance: Asset Management

Name: Klemen Iskra Student number: 11366125 Date: June 2018

(2)

Statement of Originality

This document is written by Student Klemen Iskra 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

(3)

Abstract

In this thesis, we try to answer the question, whether equity market integration among developed nations has slowed down or perhaps even reversed in the years following the financial crisis of 2008. We look at the performance of local, regional and global versions of our five-factor pricing model in different sub-periods. The results show that regional and even more so global models have performed relatively worse when compared to the local versions in the last few years after the crisis. Developed countries of Europe and North America have become somewhat less regionally and globally integrated, while the nations of the Asia Pacific region have at least seen a slowdown in stock market integration.

(4)

1 Introduction

For the better part of the last century foreign investment among the world’s financial markets has been steadily increasing as Karolyi and Stulz (2003) point out. This was a consequence of higher levels of market integration due to financial liberalization and economic integration among those markets (Bekaert et al., 2009). In the case of the European Union, it almost came to the blurring of the national borders and investment restrictions (Carrieri et al., 2004). One of the consequences of such integration is the inevitably higher correlation among equity returns (Goetzmann et al., 2005). These changes in correlation have huge implications for the benefits of international diversification and asset allocation decisions, as shown by Grubel (1968) among others. The higher the correlations among different asset classes, the smaller are the diversification benefits of including them in the same portfolio. Consequently, this topic has gotten a lot of attention from the international finance academics in recent years, especially at the turn of the century. One of the main reasons the financial crisis of 2007-2009 has spread globally so quickly is due to high degrees of financial market integration and correlation. This phenomenon of higher market co-movements during crises is known as contagion. Bekaert et al. (2014) find evidence for this, as they conclude the financial crisis spread through trade and financial linkages to other countries, especially those with weaker economic fundamentals. There are clear signs though that this trend towards a higher degree of financial integration has reversed in the last decade. Most recent papers, like Akbari, Ng and Solnik (2018) and Bekaert, Harvey, Kiguel and Wang (2016), already found some evidence that shows a decrease of globalization and integration. Somewhat more isolationist trade policies of the Trump administration in the US and the announcement of Brexit in the UK could also have an impact on the trend of integration. The purpose of this thesis is to examine how the trend in developed equity market integration has changed in the recent decade. We want to see whether the process of equity market integration has slowed down or even reversed in the developed world due to some notable economic and political developments, like the Eurozone crisis and the regulations changes that came after.

More specifically, we perform our analysis for 21 developed markets from Europe, North America and Asia, for the period between 1998 and 2017. Following the methodology of Hao,

(6)

Introduction

Karolyi and Kho (2011) we test the performance of different versions our own pricing model, namely local, regional and global, which contain the correspondingly aggregated factors. Better performance of global models would mean a higher importance of global factors as determinants of stock returns and a relatively more globally integrated market. In these models, we use pre-constructed monthly observations of factor mimicking portfolios of returns of individual countries and regions. These are sorted based on the four factors of interest, namely the book-to market ratio, the earnings yield, cash-flow-to-price ratio and the dividend yield. The analysis is further divided into four 5-year sub-periods in order to assess how the relative performance of individual versions of the models has changed over time. There are 2 sub-periods from 1998 to 2007 in the decade before the start of the Great Recession, one during the height of the crisis 2008-2012 and one after its resolution 2013-2017. To assess the performance of a particular model we look at the absolute pricing error (constant or alpha), its significance or the t-statistic and the adjusted R-squared, which measure how well the variation of the dependent variable is explained by the model. In each sub-period, these values are averaged across countries in the region in order to draw conclusions for the whole region of countries.

Several additions and contributions are made in this paper relative to Hao, Karolyi and Kho (2011) and other literature on the topic of equity market integration. The pricing model methodology is applied to the latest sample of developed country returns, including the most recent years after the crisis. The analysis is divided into 4 sub-periods in order to determine the development of different models’ performance through time. The results of the last 5-year period are especially crucial, as no paper so far has explicitly focused on investigating the stock market integration in that period specifically. Besides local and global versions, we also test regional model performance to assess to what extent countries are integrated within their respective regions. Additionally, the specific five-factor model used is unique and has not been tested before, although the factors used have been proven to be significant determinants of stock returns. This study has potential to be relevant for the international finance research community and global investors. A slow down or the reversal of market integration, can have important implications for the geographical diversification benefits of international investors, since correlation changes are a consequence of the change in integration. The relative change in importance of local, regional and global factors would also force factor investors to reconsider

(7)

Introduction

their investment decisions. We base our study on only four factors, but they have been quite relevant, so investors should keep in mind that factor relevance can decrease over time.

In our main analysis of the pricing models, we found strong results for the European and North American region, while the results were somewhat weaker for the countries of the Asia Pacific region. More specifically, individual countries have become less financially integrated with other countries of their respective region in the last five years after the crisis. Additionally, regions themselves have seen a decrease in global stock market integration in the same period, as local models improved and regional models deteriorated significantly less when compared to the global versions. As mentioned, these results were less obvious for the Asia Pacific region, though we can say the equity market integration process has at least slowed down for those countries. These findings are in line with some of the most recent literature on the reversal of financial and economic integration. Akbari, Carreri and Malkhozov (2017) have found reversals in global market integration in the past years, as well as reduced correlations. Akbari, Ng and Solnik (2018) also confirm a slowdown in financial integration, by analyzing funding barriers and regulation changes, during the same period. Finally, Bekaert, Harvey, Kiguel and Wang (2016) mention recent protectionist policies as one of the reasons for a slowdown in the globalization of financial markets of developed nations.

The remainder of this paper is organized as follows. Section 2 presents the relevant literature on the topic of equity market integration. Section 3 explains in detail the methodology of the main analysis. Section 4 describes the data, its construction and presents the summary statistics of factors and factor mimicking portfolios of returns. In section 5, we present the results of the main analysis and discuss their significance and implications. In section 6, we perform a robustness check and conclude the study in Section 7.

(8)

Literature review

2 Literature review

There is an abundance of academic research and literature on the topic of market integration and international stock return co-movements. Different approaches have been used by the international finance community when trying to quantify the levels of integration in either emerging markets, developed markets or both. In this literature review, we focus on the studies published from the 1990s onward, as they mostly assume partially integrated markets or that degree of integration changes through time. A lot of studies before that assume markets are segmented, even for the developed world. To be fair, they were more segmented in the past, but they quickly got more integrated rapidly in the past decades, as shown by most studies in this review. This literature section is divided into five sub-sections based on the methodological approach used in the studies. These include, in order, correlation examination, industry versus country factor analysis and global versus local factor model comparison. The fourth sub-section is concerned with the most recent papers that point to relatively more developed financial market segmentation in recent years. The final sub-section presents the two main hypotheses.

2.1 Historical correlations

The first group of studies includes those that deal with examining equity market correlation and stock price co-movements. Even though the methodology of these papers is pretty straightforward, they produce important ramifications when it comes to international diversification benefits.

Goetzmann, Li and Rouwenhorst (2005) examine the correlation of global stock market returns for the longest possible period of 150 years. They found that the national stock return correlations change considerably through time and are higher in periods of greater market integration. At the time of their study correlations were the highest ever, besides the period of the Great Depression. High correlations during crises mean that diversification benefits are limited when they are most needed, so investors suffer twofold. Globalization of equity markets is basically a double edged sword when it comes to international diversification. On one hand investors have a larger opportunity set of investments, but on the other hand, they have fewer diversification benefits due to increased integration. Consequently, including emerging market stocks in the portfolio has become increasingly more popular among international investors. Despite being individually

(9)

Literature review

riskier, stocks from developing countries do not seem to represent the constant fraction of risk, as measured by variance, in a particular portfolio.

Baur (2009) examines the bond-bond, stock-stock and stock-bond correlations among 8 developed countries in order to reassess the diversification benefits and try to explain the frequent portfolio rebalancing. He also finds a positive trend in stock market co-movements among these markets for the period between 1989 and 2009. Several other papers like Bekaert, Hodrick and Zhang (2009) and Longin and Solnik (1995) among others, have shown that higher degrees of regional market integration and globalization also cause stock return correlation, which reduces the benefits of international diversification. However, simple correlations as a measure of integration have been criticized by the research community (Pukthuanthong and Roll (2009)), since a country could be very integrated into the global economy, but exhibit negatively correlated returns due to its specific industry composition. In other words, high return correlation is not necessarily a sign of integration, but it is usually a consequence of high levels of market integration.

2.2 Country vs. industry factors

The second group of studies deals with the question, how well stock return variation is explained by industry structure as compared to country specific factors. This is important as it explains whether investors can better diversify over industries rather than over countries. There used to be two sides to the debate, one supporting the idea of strict industry diversification, while the other stressing the dominance of country factors and the idea of geographical diversification. Nowadays, most studies in this section find the importance of both industry and country factors, though country-wide diversification still appears to be weakly dominant.

Carrieri, Errunza and Sarkissian (2004) study market integration at the industry level, as they think country-level diversification has become relatively less important in recent decades. They show that when a country has most of her industries integrated, it is more likely to be globally integrated. However, country-level integration does not necessarily prohibit industry-level segmentation. A country with a high degree of stock market integration can still have some fully segmented industries, as well as, a more isolated country can have some individual industries that are more globally integrated. Although it is suggested to diversify across countries and industries,

(10)

Literature review

from the fact that the data is from the G7 countries from the 1990s when developed markets experienced a rapid rise in market integration and return correlation, which subsequently meant fewer cross-country diversification gains.

Griffin and Karolyi (1998) focus on investigating the importance of country versus industry factors for diversification benefits of the global portfolio. The variation of a certain index’s return is assumed to be driven by industry or country-specific effects. Returns are therefore regressed on the country and industry dummy variables. They find that industrial structure does not explain a great deal of variation, unlike the country factor that has proven to be quite crucial in explaining the variation of country index returns. It is worth mentioning that the analysis was performed on data spanning just from 1992 to 1995, so it would be interesting to see if the industry structure is also less relevant for years before and after.

Besides industry and country factors, Eiling, Gerard, Hillion and de Roon (2012) also test the relevance of currency risk factors for returns of the global portfolio. When the portfolio is static, not being rebalanced, none of the factors prove to be significantly better drivers of returns. However, when a portfolio is rebalanced every month and the first and second moments are allowed to be time-varying, industry and currency risk factors seen to be more determinant of these returns. One of their important conclusions for my thesis is that the higher importance of global over local factors implies a higher level of market integration, which appears to be true for the sample of seven largest developed markets in their study.

Bekaert, Hodrick and Zhang (2009) examine international stock co-movements for a similar period, for an even larger sample of developed markets. They offer further support for the dominance of country over the temporary importance of industry factors for international returns. They find a substantial upward trend in stock return correlations for European countries. Similar to our approach, they test linear models with global and local factors and determine that the Fama and French three factor model comes really close to the arbitrage pricing theory model in explaining the data. Finally, they find that a higher cross-country correlation among growth stocks as compared to small value stocks when comparing similar country portfolios. These results have important ramifications for our study, as it relies on the relevance of return models that include similar global and local versions of those factors.

(11)

Literature review

Baele and Inghelbrecht (2009) analyze the effect of rising market globalization on the relevance of diversification benefits across countries as compared to across industries. Like some other more contemporary studies they do not impose restrictions on static factor exposures. They find that higher levels of integration have caused the country and industry betas to converge, particularly in Europe. They conclude that despite the upward trend in market integration in recent decades, cross-country diversification is still more beneficial than cross-industry diversification.

Bekaert, Harvey, Lundblad and Siegel (2013) concentrate on economic and financial integration among the member countries of the EU. They do not use any model for analyzing equity returns or trends in correlation like other papers on this topic. Instead, they look at valuation ratios like the discount rate and earnings yield differentials across countries and industries as the measure. They find that EU membership reduced the spread of those rates but the adoption of Euro has not increased integration. Lastly, they include additional five years of post-2007 data into their analysis and find that the results are robust even after the financial crisis.

2.3 Global vs. local factor models

The third and most essential body of literature includes several studies that compare the relevance of global, local and/or regional factors as determinants of international stock returns, through several types of models. Most of them construct some sort of measure or establish significance criteria in order to interpret, whether the results of factor models offer supportive evidence for market integration and globalization.

Eiling and Gerard (2015) study the co-movement of emerging stock markets since the end of the Cold War era and the underlying macroeconomic fundamentals responsible. They found that correlations of equity returns were increasing within and among different regions, though at different speeds and through different macroeconomic channels. Eastern European and Asian markets seem to have become integrated the fastest, due to market liberalization/development and market openness, respectively. Emerging markets are also becoming more correlated with the rest of the world, so investors get fewer gains from diversification then they used to when including their stocks to the portfolio. Not using only correlations and similar to Pukthuanthong and Roll (2009), they develop a really intuitive way of measuring integration as a fraction of global risk

(12)

Literature review

versus total risk. The more a country’s return varies due to global factors, the higher the measure indicating a higher degree of integration within the region.

Bekaert and Harvey (1995) introduced a novel approach for the time that allowed them to analyze the degree of integration over time of developed as well as emerging markets. They apply time varying weights to the covariance and variance of country returns, which are used as proxies for global and country-specific risk respectively. The performances of the two extreme models of either complete global or local risk are observed over time. Nevertheless, they found significantly increased integration for only a handful of countries. The reason for such poor results could be the use of both emerging developed market data along with the fairly early observation period from 1969 to 1992. Carrieri, Errunza and Hogan (2007) follow up on the study by Bekaert and Harvey but use the GARCH-in-mean methodology to estimate the E-L (1985) international asset pricing model. Furthermore, they construct a sort of “integration index” to estimate the level of integration for several emerging markets towards during the last quarter of the 20th century. They conclude integration is increasing over time and also observe occasional reversals. They also warn about interpreting correlation of stock market indices with the global market as a degree of integration, since correlations are too low when compared to their calculated measure.

Hao, Karolyi and Kho (2011) take a comprehensive look at firm characteristics that could explain stock returns. Firstly, they look at the cross-sectional and time series equity return predictability of several firm characteristics. Out of all the different characteristics, value-based factors specifically the cash flow-to-price ratio (C/P) has proven to have the highest explanatory power for global stock returns. In time series analysis a global three factor model including the C/P ratio and momentum factor mimicking portfolios seems to perform the best, as it exhibits the lowest pricing errors and rejection rates of the GRS F-statistic. Furthermore, they examine the local, international and global versions of established factor models like the CAPM, Fama and French three factor model and their own HKK model, which includes the market-wide return in addition to the cash-flow-to-price ratio and momentum factors. They found that the purely global models performed the worst in terms of pricing errors, even more for segmented markets. Purely local models do exhibit lowest pricing errors and highest explanatory power. However, extending them by including international factors further decreases the pricing error, besides lowering the rate of rejection for the GRS F-statistic.

(13)

Literature review

They conclude with the notion that local slightly more so than global factors are important for global investors and researchers. This is especially true for the cash flow-to-price ratio and momentum. This paper represents one of the more relevant sources for our study, as we adopt from it several of the firm-specific variables as factors and the similar analytical approach. In the main regression model factor mimicking portfolios are used based on dividend and earnings yield in addition to the book-to-market value and cash-flow-to-price ratios.

2.4 Recent reversal in integration

The last literature subsection presents some of the most recent papers that have already shown more support for the idea of the decrease in levels of market integration after the crisis.

Akbari, Carrieri, and Malkhozov (2017) show recent reversals in global market integration with the help of funding liquidity shocks. They look at international CAPM models with funding constrains to compare different betting-against-beta (BAB) portfolios as a measure of cross-country funding barriers. They found it is lower for developed markets, but exhibits a bigger downward trend for emerging markets. In periods of low capital flows among countries, the measure is significantly higher, which is related to equity market segmentation. This result points to slight reversals in global market integration during the most recent financial crisis. They also find and increase in home bias, tendency to invest more in domestic stocks, during funding distress periods, even though there was no increase in investment barriers.

Akbari, Ng and Solnik (2018) examine the development of financial versus economic integration and the factors that drive them for the period 1989-2015 for 41 countries. Stock returns are decomposed into expectations of cash flows and discount rates and the fully integrated US is used as a benchmark. A higher degree of economic integration is implied by higher correlations of cash flow news among countries. Likewise, similar reactions of discount rates to global shocks imply stronger financial integration. Even in absence of financial integration, most of the markets have become more economically integrated. There is evidence of increasing integration in developed as well as emerging markets, but to a lower degree up until the financial crisis of 2008. After that, it has slightly declined due to financial regulations and protectionist policies. This is an important conclusion, as it offers further support for the rationale behind our study.

(14)

Literature review

Bekaert, Harvey, Kiguel and Wang (2016) similarly decompose integration into financial and economic globalization in connection to international asset returns, including bonds and exchange rates. A distinction is also made between regulatory and realized integration free of regulation changes. Besides looking at correlations of returns, they find evidence that the globalization of the world markets has not been steadily increasing, but has slowed down or even reversed in some countries after the crisis. They justify this by the collapse of trade flows or the late start of the sample that missed out one the period of the highest growing globalization of the late 20th century. Additionally, the acknowledgement of the effect of more national and protectionist policies like the “buy local” program in the States, for example, is critical for the assumption made in this thesis.

Most of the literature focuses on the rising market integration at the end of the 20th century and in the early 2000s. There are only a few recent papers (mentioned in the previous paragraph) that study the change in market integration in the post–crisis era and they use different approaches than the global versus local factor analysis. Additionally, the paper by Hao, Karolyi and Kho (2011) whose methodology we adopt, does not analyze the change of importance of global versus local factors over time or in individual sub-periods. They use a different sample of countries, including emerging markets, in the two decades prior and do not use the exact factors in any of the models or a regional version of any model. Consequently, this thesis could contribute to the existing literature by shedding light on the on the period of the last decade and how this somewhat less globalized and more isolationist economic and political climate impacted stock market integration trends among developed nations.

(15)

Literature review

2.5 Hypotheses

As mentioned, the objective of this paper is to examine how the integration among developed equity markets has changed in the wake of recent developments in the Western and other developed economies. More specifically, we would like to examine how the relative importance of local compared to global and regional factors, as determinists of international equity returns, has varied for different countries through time. For this, two main hypotheses, that are concerned with how degrees market integration have changed within and among different groups of developed countries and regions, are formulated.

H1: Stock market integration for countries within the European, North American and Asia Pacific regions has slowed down or even reversed after the financial crisis of 2007-2009. This means that local models have performed relatively better as compared to regional models for those countries in recent years.

H2: The trend in stock market integration among the European, North American and Asia Pacific regions has slowed down in years after the crisis. Pricing models based on local and regional factors have performed relatively better than models containing global factors in the years after the Great Recession.

These hypotheses will be tested by combining the approaches used by Hao, Karolyi and Kho (2011) (HKK from now on) and Bekaert, Hodrick, and Zhang (2009). The former paper first investigates the performance of three different factor models, namely the CAPM, the Fama and French three factor model and the Hao, Karolyi and Kho model. For each of those models, three separate versions are used depending on which factors are included in the regressions. The local version contains country-specific factors, the global version contains global factors and the international version contains both local and foreign factors. The results are aggregated for developed and emerging markets separately in addition to all countries together.

(16)

Methodology

3 Methodology

In this paper, we compare the relative performance of three versions of the same factor model, namely the local, regional and global version, that each includes only one kind of factors correspondingly. Since the investigation concerns developed markets exclusively, we aggregate results based on the three regions: Europe, North America and the Asia Pacific. Unlike in HKK, which performs the examination of factor importance of the whole sample period, this paper divides the sample into four 5 year periods from 1998 to 2017 in order to observe the change in model performance through time.

The main factor model includes factors based on the following variables: market risk premium, value/growth factor (book-to-market ratio), dividend yield, cash flow-to-price ratio and the earnings yield (EPS/P). Along with the market risk premium, these firm-specific characteristics have proven to be significant determinants of stock returns in well-established asset pricing models like the Fama and French three factor model and the HKK model from Hao, Karolyi and Kho (2011). Besides this, the main reason for the inclusion of these four specific factors is data restriction, which is further elaborated on in the Data section.

In the regression, we use factor mimicking portfolios (FMPs) of returns, which have been pre-constructed based on the firm characteristic aggregated at country, regional or global level, as independent variables. The stocks are divided into two groups based on whether a particular factor is in the top (High) or the bottom (Low) 30%, while the middle 40% are not included. These groups are value-weight returns for each country, region and the world, calculated at the end of each year. For every factor, we construct a portfolio that goes long in the group stocks with above average performance and short into worst performing stocks. This means that the factors are then constructed by subtracting Low return quintile from the High return quintile and halving the result.

As HKK point out the BM ratio, CP ratio, dividend yield (DY) and earning yield (EY) are all significantly positive when it comes to explaining stock returns. Consequently, we form FMPs for all variables by subtracting the lower quantile from the higher quantile. Unlike the other four factors, the market return is simply the value weighted average of returns of all the stocks that have non-missing values of other factors.

(17)

Methodology

Local, regional and global versions of our five factor model with exactly specified factors:

𝑅

_𝑗,𝑡

= 𝛼

_𝑗

+ 𝑚

_𝑗𝐿

_𝑀𝐾𝑇

𝑡𝐿

+ ℎ

𝑗𝐿

𝐻𝑀𝐿

𝐿𝑡

+ 𝑐

𝑗𝐿

𝐶𝑃

𝑡𝐿

+ 𝑑

𝑗𝐿

𝐷𝑌

𝑡𝐿

+ 𝑒

𝑗𝐿

𝐸𝑌

𝑡𝐿

+



𝑗,𝑡

𝑅

_𝑗,𝑡

= 𝛼

_𝑗

+ 𝑚

_𝑗𝑅

𝑀𝐾𝑇

_𝑡𝑅

+ ℎ

_𝑗𝑅

𝐻𝑀𝐿

𝑅_𝑡

+ 𝑐

_𝑗𝑅

𝐶𝑃

_𝑡𝑅

+ 𝑑

_𝑗𝑅

𝐷𝑌

_𝑡𝑅

+ 𝑒

_𝑗𝑅

𝐸𝑌

_𝑡𝑅

+



_𝑗,𝑡

𝑅

_𝑗,𝑡

= 𝛼

_𝑗

+ 𝑚

_𝑗𝐿

𝑀𝐾𝑇

_𝑡𝐿

_{+ ℎ}

𝑗𝐿

𝐻𝑀𝐿

𝐿𝑡

+ 𝑐

𝑗𝐿

𝐶𝑃

𝑡𝐿

+ 𝑑

𝑗𝐿

𝐷𝑌

𝑡𝐿

+ 𝑒

𝑗𝐿

𝐸𝑌

𝑡𝐿

+



𝑗,𝑡

where 𝑅𝑗,𝑡 is the excess return of a particular portfolio of returns j we would like to test (dependent variable). Lower case letters (i.e. 𝑚_𝑗𝐿_{) represent regression coefficients and upper case} acronyms (i.e. 𝑀𝐾𝑇_𝑡𝐿) are the returns of the factor mimicking portfolios (independent variables). α_j is a constant and _j,t is the error term. Subscript L indicates local factor, R indicates a regional factor and G indicates a global factor for portfolio.

For each version of the model (i.e. local) and for every country, High group portfolios based on the 4 different factors are regressed the factor mimicking portfolios from the corresponding countries or regions. In the local 4 factor models, country returns are regressed on the corresponding country-specific factor mimicking portfolios. In the regional versions they are regressed on the three distinct regional FMPs and in the global models, all country returns are regressed against the same globally aggregated FMPs. This is done so local, regional and global model performance can be compared.

In order to assess the performance of the distinct versions of the model in each period, we look at three separate statistics: the adjusted R-squared, the absolute pricing error  or intercept and the absolute t-statistic for the intercept. They are averaged across countries for each of the four factors. Based on the t-statistic we also report the number of countries for which the null hypothesis, that the intercept is equal to zero, was rejected at the 10% confidence level. The number of rejections can then be compared among models and 5 year sub-periods along with the three aforementioned statistics. Higher R-squared is an indicator of superb model performance, while higher pricing error and their t values along with more rejections of the null are all signs of poor model performance.

(18)

Methodology

3.1 Detailed factor description

In this section, we offer a more detailed description of the dependent and independent market return in addition to the factors of which the factor mimicking portfolios are based on.

Excess returns: Returns on country portfolios are calculated from the stock prices for each

month. This is done by dividing the difference of indices in period t and t-1 by the price in period t-1: (𝑃𝑡− 𝑃𝑡−1) 𝑃⁄ 𝑡−1. In our case, the following equation is used: 𝑙𝑜𝑔 (𝑃𝑡⁄𝑃𝑡−1), which is a good approximation of the former for small returns. To get to the excess returns the risk-free rate (𝑟𝑓,𝑡) in each period is subtracted from the total portfolio returns (𝑟𝑗,𝑡) accordingly. For percentile returns, as denominated in this paper, they are multiplied by 100%.

Market return: This excess market return is calculated by subtracting the risk-free rate 𝑟_𝑓from the market return 𝑟_𝑚 aggregated at national, regional or global level.

Book-to-market ratio: Value/growth factor measured by book-to-market equity ratio is the

fraction of the book and market (B/M) value of the firm. As Fama and French (1993) point out that firms with higher B/M ratio (value stocks) earn relatively higher returns as compared to firms with low B/M ratio (growth stocks). The ratio is obtained by dividing common shareholders’ equity by the market capitalization of a firm.

Cash flow-to price ratio: To obtain the cash flow to price ratio, the cash flow per share is

divided by the closing price of a firm’s share. The cash flows denote the earnings in cash of a firm before taxes, depreciation or amortization. This factor is the inverse of the price-to-cash flow ratio and measures how much cash a company makes as compared to its value.

The earning yield is obtained by dividing the earnings per share at the end of the year or quarter

by the closing price of a share. It is the inverse of the P/E ratio, which is a popular valuation measure also known as a price multiple. It basically shows how much an investor earns per dollar invested.

Dividend yield factor is the fraction of the annual dividend payout per share divided by the price

of one share of a company. For the owners or shareholders, it is an indication of how much they earn each year in payouts per dollar invested.

(19)

Methodology

3.2 Correlation

The main methodology described at the beginning of this section is relevant for analyzing the integration of country stock returns within a region and among regions. Despite the fact that correlation is not a good measure of integration it is still insightful to see how it has developed through the years, as it usually follows the level of integration. Higher correlations should not be directly interpreted as higher degrees of market integration. Though, it is worth noting that highly integrated countries usually also exhibit high correlations of returns. Therefore, finding that correlation patterns match the patterns of model significance, would offer support for the validity of the main results. At the beginning of the Results section, before the main analysis of the five-factor model, it will be investigated to what extent individual country indices or market–wide returns co-move with their respective regional indices an the globally aggregated market return in each of the four 5-year sub-periods. For each country, two pairwise correlation coefficients will be reported, with the regional and global index respectively, in each sub-period. The straightforward formula for the correlation coefficient is explained in the next paragraph. Besides the individual countries’ correlations, it is also interesting to investigate how correlations between the different regions have changed over time, as posed by H2. The following simple correlation measure will be used

Corr

Yi

(𝑅

𝐺𝑃𝑟

, 𝑅

𝐺𝑃𝑞

) =

𝑐𝑜𝑣𝑌𝑖(𝑅𝐺𝑃𝑟,𝑡𝑅𝐺𝑃𝑞,𝑡)

√𝑣𝑎𝑟𝑌𝑖(𝑅𝐺𝑃𝑟,𝑡)√𝑣𝑎𝑟𝑌𝑖(𝑅𝐺𝑃𝑞,𝑡)

,

where 𝑐𝑜𝑣𝑄𝑖(𝑅𝐺𝑃𝑟,𝑡𝑅𝐺𝑃𝑞,𝑡) is the covariance in sub-period Yi between regional indices or the value-weighted returns of regions r and q and 𝑣𝑎𝑟_𝑄𝑖 is the variance of that region’s returns in that sub-period. We compare the correlation matrices in different sub-periods for the three regions studied including the whole developed world. Based on previous studies, especially those published in recent few years, we would expect correlations of returns among countries and regions to have been rising up until the height of the crisis and then slightly decreasing or at least increasing to a lesser degree. For example, Akbari, Carrieri, and Malkhozov (2017) show the development of market correlation in Figure 1 as a comparison to their BAB correlation. Global market correlation has been steadily increasing form the early 2000s until late 2008, stayed mostly constant for two to three years and then started decreasing slightly after 2011.

(20)

Data and descriptive statistics

4 Data and descriptive statistics

4.1 Data

The study is conducted from the perspective of the US investor, so all the values are denominated in US dollars. The interest rate on a three-month U.S. Treasury bill from Interest Rate section of Federal Reserve Bank on WRDS can be used as a proxy for the risk-free rate.

Monthly frequency of stock returns between 1998 and 2017 is needed in order to attain a sample size of 60 observations for each five year sub-period. More specifically, the whole study period is divided into 2 sub-periods before the crisis (1998-2002 and 2003-2007), one during the crisis (2008-2012) and another sub-period after (2013-2017). The portfolios are calculated from firms that have data on all four variables and gathered for the countries, for which there is data for the whole sample period. These include 14 European (Austria, Belgium, Denmark, Finland, France, Germany, Ireland, Italy, Netherlands, Norway, Spain, Sweden, Switzerland and the United Kingdom), 5 Asia Pacific (Australia, Hong Kong, Japan, New Zealand and Singapore) and the two North American countries.

The data for the four distinct types of factor mimicking portfolios used in the regression is obtained from the Kenneth R. French’s website in premade form. The portfolios of returns are sorted based on the book-to-market ratio, the earnings yield, the cash-flow-to-price ratio and the dividend yield. The country, regional and global index portfolios formed on the basis of those factors will be downloaded from the International Research Returns Data section of the website.1 These specific factors have been chosen not only because of their significance in stock return predictability but also due to data restriction issues. Since we do not construct the FMPs of returns ourselves, we have to use the limited information provided. Since the website is mostly focused on the US data, it only provides monthly observations on the required factors for other developed nations. This restriction on the frequency of observation conveniently avoids the problem of non-synchronous trading around the world due to time differences and distinct public holidays or other potential non-trading day conventions. For all the countries we use the value-weighted returns denominated in US Dollars. The number of firms changes every year as the

1

(21)

country portfolios are rebalanced on a yearly basis to include only the firms with all factors. The returns have been pre-filtered to exclude missing values and account for dividend payouts.

The data is selectively copied for our specific time period and merged into a panel containing 6000 rows (240 months for 25 countries/regions) and 14 columns (country and region indices, year, month, market return, 4 High and 4 Low FMPs and the risk-free rate). Additionally, from the same files, we also extract the yearly observations (the only available) of the four firm characteristics, namely book-to-market ratio, earnings yield, cash-flow-to-price ratio and dividend yield for the purpose of reporting summary statistics. These are also merged into a separate panel of 420 rows (20 years for 21 countries) by 8 columns (country and region indices, year, number of firms included at each given year and the 4 yields and ratios). The data on the American factor mimicking portfolios is gathered separately from several files of the Kenneth French Website and added to the two panels, as return portfolios for the US are not included in the same file. When it comes to the yearly observations on US factors themselves all are provided with portfolios of returns except for the dividend yield. This was instead obtained from the historical values of the S&P 500’s historical dividend yield on their official website.2

Figure 1. MSCI ACWI Index investment distribution. The figure shows a pie chart representing individual

country or region weights of the All Country World Index. This index includes 24 emerging and 23 developed markets that together represent around 85% of the global equity investment opportunities.

(22)

Additionally, the North American (NA) return index had to be constructed separately, as it is not provided like the other two regional indices for the Europe and the Asia Pacific regions. Returns for the NA region are calculated by weighing Canadian and the US returns by 5.42% and 94.58% respectively. These weights are obtained from the MSCI North America Index provided by Morgan Stanley Capital International website3, as is done on the Kenneth French’s website. Furthermore, the Global index of returns also has to be constructed from the three regional indices used in this analysis. The weights are obtained from the same website but from the MSCI All Country World Index (ACWI), which is value-weighted from stocks of over 2400 companies from all around the globe. The weights that this index assigns to each region of the world are presented in Figure 1.

Excluding the emerging markets (12%) from the MSCI World Index and neglecting the minimal influence of Portugal and Israel, which are not present in our sample, the final weights for our three regional indices are: 23.86% (21/88) for Europe, 62.50% (55/88) for North America and 13.64% (12/88) for the Asia Pacific region. From these percentages, the global factor mimicking portfolio of returns, which are the same for every country in the global model, can be calculated.

4.2 Descriptive statistics

From all 21 developed countries, there are almost exactly 7600 firms represented in the sample on average each year. We report the mean of the number of firms because the sample of firms in each country changes from year to year. Table 1 below provides the summary statistics for the four factors of interest across countries and through time. Out of all the developed markets included Japan and US companies represent considerably more than half of the sample with 18.6% and 38.7% respectively. When it comes to the regional distribution of firms in the sample, all three regions seem to have a somewhat more even representation. Europe has the smallest share of 26.7%, North America the largest percentage of 42.5% and the Asia Pacific somewhere in the middle with 30.8%. In Europe, the country with the highest percentage of average firms (25.0%) in the sample is as expected the UK, but it does not completely dwarf other countries in the region like Japan and the US.

(23)

Table 1: Summary statistics of factors by country/ region and by sub-period

The table displays the descriptive statistics for the yearly observations of the four factors of interest, namely the book-to-market ratio, the earnings yield, the cash-flow-to-price ratio and the dividend yield. The results shown are for the 21 countries, three regions and the whole developed world, for the period between 1998 and 2017.

Panel A contains the average number of firms rounded to the nearest whole number during the whole sample period. Additionally, the panel shows the means and the standard deviations for each of the four factors denominated in percentages (%).

In panel B the Summary statistics are presented for developed markets across the three regions and separately for the four 5-year sub-periods: two before the start of the crisis, one during and one after the crisis has mostly ended. For each region and sub-period, it reports the time-series mean, standard deviation, the minimum and the maximum value reached by globally aggregated factors that are weighed by the number of firms. In the second panel all the values represent percentages.

Panel A: Across countries/regions

Country or region Average number of firms Average book-to- market Std. dev. book-to- market Average earnings yield Std. dev. earnings yield Average cash-flow- to-price Std. dev. cash-flow- to-price Average dividend yield Std. dev. dividend yield Austria 60 77.962 29.643 8.201 3.531 18.217 5.667 2.559 1.058 Belgium 65 53.943 20.070 7.233 2.737 11.784 3.195 2.941 1.153 Denmark 65 40.183 13.526 5.554 2.310 10.253 4.042 1.847 0.765 Finland 63 41.829 16.774 6.239 2.553 10.214 3.835 3.406 1.743 France 306 52.699 20.528 6.473 2.595 12.435 3.919 2.994 0.983 Germany 307 51.510 14.652 6.465 2.199 14.467 3.589 2.724 0.705 Ireland 24 54.295 34.078 7.856 6.367 11.732 7.804 2.476 2.338 Italy 120 62.506 31.030 7.153 2.741 15.546 4.668 3.714 1.649 Netherlands 66 42.976 18.089 6.950 3.651 11.882 4.744 3.163 1.299 Norway 77 59.857 15.783 9.210 3.023 17.447 4.490 3.737 1.754 Spain 81 51.082 17.208 7.689 4.656 14.272 6.167 3.056 1.387 Sweden 158 43.124 12.666 6.794 2.494 10.193 3.057 2.997 1.419 Switzerland 134 32.655 10.148 5.310 1.636 7.520 1.957 2.186 0.965 UK 508 45.194 12.236 6.843 2.218 10.626 2.798 3.386 0.892 Canada 285 45.568 8.059 6.185 1.542 11.060 2.047 2.396 0.930 US 2944 58.420 11.342 6.368 1.084 9.021 1.448 1.839 0.442 Australia 325 42.799 8.541 5.871 1.500 8.864 1.506 3.796 1.097 Hong Kong 301 72.435 19.189 8.993 4.327 10.887 4.529 3.266 0.868 Japan 1414 66.674 19.667 5.315 2.030 12.673 3.613 1.501 0.670 New Zealand 49 57.396 11.437 6.406 1.471 11.402 1.772 4.676 0.822 Singapore 249 63.190 17.284 7.927 3.034 10.959 3.234 3.102 1.375 Europe 2032 49.109 14.834 6.775 2.310 12.221 3.196 3.032 0.973 North America 3229 57.286 10.663 6.352 1.063 9.201 1.406 1.888 0.455 Asia Pacific 2339 63.527 16.141 6.166 2.123 11.703 3.073 2.284 0.772 World 7600 57.020 11.795 6.408 1.589 10.779 2.203 2.316 0.631

(24)

Table 1 continued Panel B: Through time

Europe

Sub-period: 1998 – 2002 2003 – 2007

Factor Average Standard deviation Minimum Maximum Average Standard deviation Minimum Maximum

Book-to-market 33.27 11.26 9.61 70.22 43.94 10.81 23.17 79.37

Earning yield 4.71 1.34 1.43 11.59 6.38 1.93 3.82 24.21

CP ratio 9.46 3.23 2.56 22.04 12.28 3.84 5.76 36.01

Dividend yield 2.04 0.59 0.58 3.31 2.59 0.71 1.10 4.58

Sub-period: 2008 – 2012 2013 – 2017

Book-to-market 62.23 22.36 30.16 159.31 57.03 14.44 32.42 111.73 Earning yield 9.23 3.39 3.90 32.52 6.78 1.45 3.80 11.05 CP ratio 15.30 5.24 6.78 41.39 11.85 3.25 6.35 21.39 Dividend yield 3.97 1.41 1.52 11.50 3.52 0.76 1.89 6.06 North America Sub-period: 1998 – 2002 2003 – 2007

Book-to-market 58.55 10.30 31.57 73.58 52.00 7.80 34.51 66.96

Earning yield 6.39 0.85 3.32 7.22 6.03 0.48 5.03 6.77

CP ratio 9.49 1.10 7.18 12.70 8.88 1.08 8.06 13.47

Dividend yield 1.39 0.23 1.17 1.84 1.71 0.12 1.35 1.87

Sub-period: 2008 – 2012 2013 – 2017

Book-to-market 66.31 13.32 35.82 88.60 52.28 6.35 45.41 63.88 Earning yield 7.21 1.55 4.92 10.77 5.78 0.62 5.22 7.56 CP ratio 10.29 1.92 8.64 16.80 8.13 1.19 7.11 12.52 Dividend yield 2.37 0.57 1.83 4.30 2.08 0.36 1.84 3.54 Asia Pacific Sub-period: 1998 – 2002 2003 – 2007

Book-to-market 48.13 12.37 28.34 84.48 57.03 13.58 30.77 102.67

Earning yield 4.18 2.79 1.60 18.31 5.33 0.83 4.51 8.13

CP ratio 9.33 2.86 4.56 19.50 11.10 2.06 7.08 13.66

Dividend yield 1.48 1.06 0.41 5.24 1.83 0.95 1.04 4.56

Sub-period: 2008 – 2012 2013 – 2017

Book-to-market 76.97 22.93 33.00 116.84 72.00 12.30 47.34 92.78

Earning yield 8.05 3.27 5.45 19.92 7.11 1.64 4.91 12.01

CP ratio 14.74 4.70 7.79 23.55 11.65 1.81 8.29 15.53

(25)

Panel A of Table 1 displays the time-series averages and standard deviations of the four factors of our model besides the aforementioned average number of firms across countries and regions. The time-series means are equally weighed, as we assume that an equally representative sample of firms was picked each year. The data on the factors for the three separate regions was not available along with returns as for the individual countries. Consequently, the factors had to be aggregated for countries belonging to their distinct regions. Since nor the individual companies’ sizes nor the averages across countries are provided, the regional averages of the factors are weighed based on the number of firms from each country. Doing so, we assume, the average firm sizes are roughly equal across all countries. This means that the averages of the factors from the North American region are almost identical to those of the US, as the country is assigned a weight of nearly 95%. This, in turn, makes the values of the factors for North America substantially biased towards the US.

The average book-to-market ratios, denoted in percentages, vary fairly substantially from country to country and can differ by more than a factor of two, as is the case for Austria and Switzerland. The former has the highest book-to-market ratio mean while the later appear to have the lowest, which is especially surprising for two countries that are that close, geographically and economically. Investment, trade and cross-border employment between the two countries have been rising steadily in the past decades as indicated by Switzerland’s federal Department of Foreign Affairs.4 Ireland’s book-to-market ratio is changing the most while Canada’s is the most stable throughout the years. Earnings yields are similar among the developed markets, especially when comparing the regional weighed averages. Ireland again stands out when it comes to standard deviation. Switzerland is again the country with the lowest average cash-flow-to-price ratio, while North America exhibits a relatively low ratio. Besides the usual culprit, Spain also went through high levels of variation of the cash-flow-to-price ratio. The dividend yield is comparably even across countries. Japan stands out with the lowest time-series mean of the yield, bringing down the average of the whole Asia Pacific region. Even this factor is no exception when it comes to Ireland’s fluctuation. The main takeaway from panel A is that factors are more volatile for the countries that have been hit be the financial crisis most extensively, specifically Ireland. The averages of factors by country are very similar to those of HKK (2011)

(26)

Panel B of Table 1 displays even more detailed statistics adding the minimum and maximum percentages, besides the means and standard deviations. The data is aggregated at the regional level and weighed by the number of firms across the four 5 year sub-sample periods. These include two before the crisis, on sub-period during the main development of the recession and one after when the worst was over for most of the countries studied in this paper. The second panel might seem slightly oversaturated and redundant, however, we think it is essential, as the main results are reported in a similar fashion.

All values in panel B are denominated in percentages for easier interpretation and to avoid unreadable clusters of numbers. Looking at the panel as a whole, the first major observation is that for almost all factors in the three regions there is an apparent pattern developing. Most of the averages of factors are increasing between the years 1998 and 2012 but they drop considerably in the last sub-period. Standard deviations follow a similar pattern; however, they peak even more in the third sub-period, which was the height of the Great Recession. Furthermore, the differences between the maximum and minimum values, also known as ranges, change through time in the same manner. Literally, all the maximum values of the four factors of all three regions are observed in the third sub-period between 2008 and 2012.

On one hand, we can rarely be completely certain about the true cause behind the above described pattern. On the other hand, the pattern is distinctive of the entire sample of countries, which could only be caused by a globally significant macroeconomic event. The global financial crisis of 2008 fits the criteria. It is fairly intuitive that the book-to-market ratio is higher during the crisis. The prices drop due to investors’ negative outlook along with the risk aversion, increasing the ratio. The same goes for the other three firm characteristics, as they all have the market price in the denominator. However, the pattern is the least considerable for the earnings yield, as profits also fall during crises.

A comparison among regions shows that for almost all factors in every sub-period the variation and the ranges are the largest for the Asia Pacific region, which also has the highest book-to-market ratio. The earning yield is not specifically distinguishable for any particular region. The cash-flow-to-price ratio moves very similarly for European and Asian countries but is a bit lower for North America. The low dividend yield of the US brings the North American average way down, making the regions yield the lowest though the whole sample period.

(27)

Table 2: Descriptive statistics of factor-mimicking portfolio and market by sub-period

The table presents descriptive statistics for the returns of the factor mimicking portfolios of returns and for the market-wide return. The FMPs are constructed by taking the highest 30% of returns minus the lowest 30% of returns based on the four factors. These include: book-to-market ratio (BM), earnings yield (EY), the cash-flow-to-price ratio (CP) and the dividend yield (DY). These FMPs along with the market return (Mkt) are all labeled with abbreviations in the parentheses, to conserve space and to differentiate them from the actual factors from Table 1. Panel A displays the average, standard deviation, minimum and maximum in percentages (%), as well as the t-statistic and the number of months for every 5-year sub-period. Panel B exhibits the pairwise correlations coefficients for the returns in the years before and after the crisis. Panel C tests for serial correlations at 1, 3, 6 and 12 month intervals for the same sub-periods as in the previous panel, by trying to reject the null that there is no auto-correlation present. The values of the LM(k) statistic are comparable to the critical value of the normal distribution.

Panel A: Summary statistics for FMPs of returns

Sub- period: 1998 – 2002 2003 – 2007

Factor Mean

Return

Standard

Deviation t-statistic Min. Max.

No. of Months

Mean Return

Standard

Mkt -0.23 5.88 -0.30 -30.22 31.64 60 1.18 3.25 2.81 -19.05 19.09 BM 0.07 5.77 0.10 -39.56 59.60 60 0.03 2.86 0.07 -19.17 42.30 EY -0.04 5.58 -0.06 -29.84 42.88 60 0.07 2.83 0.20 -15.95 21.27 CP 0.30 5.67 0.41 -28.63 44.87 60 -0.08 3.24 -0.19 -51.14 46.84 DY 0.12 6.02 0.16 -36.41 46.66 60 0.10 3.00 0.26 -23.73 29.43 Sub- period: 2008 – 2012 2013 – 2017 Factor Mean Return Standard

No. of Months

Mean Return

Standard

Mkt 0.23 6.33 0.29 -34.11 29.76 60 1.08 3.28 2.54 -14.14 13.83

BM 0.25 4.48 0.44 -43.85 41.36 60 -0.01 2.97 -0.02 -17.48 19.62

EY 0.17 3.75 0.35 -34.91 38.62 60 -0.07 2.47 -0.23 -15.38 22.94

CP 0.12 3.34 0.28 -35.14 36.55 60 0.02 2.35 0.06 -13.96 13.90

DY -0.04 4.80 -0.07 -49.54 53.67 60 0.07 2.39 0.24 -20.47 15.45

Panel B: Correlations among FMP returns

Sub- period: 1998 – 2007 2008 – 2017 Factor Mkt BM EY CP DY Mkt BM EY CP DY Mkt 1.000 1.000 BM -0.158 1.000 0.309 1.000 EY -0.192 0.545 1.000 0.292 0.610 1.000 CP -0.158 0.654 0.702 1.000 0.238 0.713 0.728 1.000 DY -0.259 0.511 0.531 0.592 1.000 0.131 0.504 0.582 0.581 1.000

Panel C: LM(k) test statistic for serial correlation of FMP returns

Sub- period: 1998 – 2007 2008 – 2017

Lag in

months Mkt BM EY CP DY Mkt BM EY CP DY

Lag 1 2.87 3.32 2.38 2.78 2.91 8.68 2.70 2.71 2.10 1.53

(28)

4.3 Factor mimicking portfolios

So far we have only presented the descriptive statistics for the factors themselves. However, the main question of this paper deals with the returns based on those four factors. More specifically, factor mimicking portfolios of returns are calculated from the firms’ stock returns based on which factor quantile they fall into. As in Fama and French (1993) the final portfolios of returns are basically the differences in returns of the top 30% and the bottom 30% of firm’s returns sorted on the basis of the individual factors. A more detailed description of the procedure, screening and weighing can be found in the above Methodology and Data sections. Finally, these return portfolios are then used in different types of regression models, with either local, regional or global factors, in different time periods, to ultimately answer the question about the changing degree of developed market integration.

Table 2 shows the descriptive statistics for the returns of the FMPs and the market-wide return in different time periods. To reiterate, these FMPs are the returns obtained by taking the High group of returns minus the Low group of returns based on the four factors. For panel A the summary statistics are calculated for the four 5-year periods as for the factors, but in panels B and C correlation statistics are only separately analyzed for the two decades before and after the start of the crisis

In the first panel A of Table 2, summary statistics of FMPs of returns are presented for all the 21 developed markets together, weighed based on the MSCI weights. The aggregated local means would very likely resemble those of the global FMPs; however, the standard deviations and the spreads would be smaller for the Global FMPs due to aggregation (smoothing). Apart from these statistics, which are denominated in percentages (%), we also report the t-statistic in order to determine how much returns depart from zero. It is calculated by dividing the mean by the standard deviation and the square root of the sample size, in this case, 60 months. Looking at the standard deviations and ranges, we see that returns were more volatile during the first and third sub-periods, which could be attributable to the Great Recession and the crisis in the early 2000s. The average market return is significantly positive in the calmer years of the second and fourth sub-period. FMPs’ averages on the other hand display more random behavior and are mostly close to zero and become even less significant over time. This could be due to the fact that we use only the top and bottom 30% of returns to construct them, which might not be enough of a

(29)

difference. Hao, Karolyi and Kho (2011) use deciles portfolios, so they take the top 10%. We, however, are limited b, the data that is available.

In panel B we can observe the correlation of FMP returns and the market return in two different periods. The returns of the FMPs are relatively strongly correlated with each other and even more so in the second period. Another important finding is that that the market return is negatively correlated with the FMPs’ returns before the crisis but positively after it. This could result from the fact that markets are more correlated during crises, known a contagion, and that the factors have lost their importance through time in determining stock returns, due to wider adoption and recognition of the finance and investing community. These results are quite different from previous literature on correlation, such as Goetzmann, Li and Rouwenhorst (2005), where correlations have been steadily increasing thorough time.

Finally, we test for autocorrelation of portfolio returns, since we are using them in a time sries regression. Panel C reports the bias-corrected Q(P) statistic for serial correlation. In Stata, the module for testing it is named “XTQPTEST” and was developed by Born and Breitung (2016) as a bias-corrected LM-based test for auto-correlation of panel data. The test essentially investigates, whether there is no serial correlation of order k in all separate time-series (H0) and if there is at least some serial correlation in the panel (H1). The LM(k) test values are comparable to the critical values of the normal distribution. So when the LM(k) is smaller than 1.96, we can except H0 that there is no serial auto-correlation in all of the individual time-series with 95% probability. The higher value of the LM statistic means there is higher chance that more time-series exhibit serial correlation. Looking directly at the numbers in Table 2, most of the FMPs exhibit weak or no auto-correlation in the first decade, except for dividend yield based FMP lagged one year that is negative. The market returns seem to be significantly positively auto-correlated for the half year period. The FMPs’ show similar results, mostly there is no serial correlation, except perhaps for the one month lag. The market returns exhibit surprising values, as there seems to be a strong positive correlation at one and three month periods and a strong negative auto-correlation every six months. This would imply that we would have to regression models robust to autocorrelation.

(30)

Results

5 Results

In the following section, we present the results of our main analysis that tries to answer the questions posed by the two hypotheses. That is, whether the process of regional and global equity market integration has slowed down or even reversed in the recent decade, after the Great Recession, among developed countries and regions. This is done by looking at the performance of the three different versions of our five-factor model, namely local regional and global versions, in four separate 5-year periods between 1998 and 2017. Specifically, we are interested in how the relative importance of local versus global factors, as determinants of stock returns, has evolved in the past 20 years. Usually, the higher the importance of global factors for a country’s equity returns, the more that country is integrated into the global market.

5.1 Correlations

We first look at how the correlation of market-wide returns of developed markets with respect to the regional and global returns has evolved through the years. Despite stressing several times that correlation is not an adequate measure of integration and has been criticized by many financial academics like Pukthuanthong and Roll (2009), it still insightful to look at its development. The evolution of correlation may offer support for the results of the regression models later, as returns of more integrated markets also co-move more with the global returns most of the time. Table 3 shows the correlations of individual countries with their respective regions and the globe.

Panel A contains the correlation coefficients of 21 developed countries with respect to their particular regions and the global developed market. The two correlation coefficients for each country are reported for each of the four 5-year sub-period of interest. The first thing to notice is that for almost all of the countries the correlation with the regional returns is larger than the one with the global market returns for the whole sample period. This finding makes sense as we would expect for a country’s financial market to be more closely tied with its neighbors in the region. Nonetheless, there are some exceptions to this rule especially in the decade before the start of the crisis. Canadian along with some Scandinavian and Asian market returns co-move more globally then regionally in that period, but form most regional correlation prevails in the decade after. This notion is in line with our expectation, as the financial markets were rapidly becoming more globalized in the early 2000s, but after the crisis regional integration became

Developed stock market integration : recent change in the importance of global factors for international stock returns

Master Thesis