• No results found

Eastorwest,homeisbest:Thepersistenceofthehome-biaseffectinequityinvesting M U G

N/A
N/A
Protected

Academic year: 2021

Share "Eastorwest,homeisbest:Thepersistenceofthehome-biaseffectinequityinvesting M U G"

Copied!
56
0
0

Bezig met laden.... (Bekijk nu de volledige tekst)

Hele tekst

(1)

U

NIVERSITY OF

G

RONINGEN

M

ASTER THESIS FINANCE

East or west, home is best: The persistence of

the home-bias effect in equity investing

Author

Jarmo

VAN

B

EURDEN

Supervisor

dr. Steffen E

RIKSEN

(s3105687)

Abstract

The home-bias effect is a phenomenon in behavioral finance where investors hold a too-large portion of their portfolio in domestic equity assets. Using Mundlak models based on unbalanced panel data for 32 OECD countries for the period 2001–2018, this paper observed the persistence of this effect and found that internet use decreases it. When investors do invest abroad, they prefer to invest in countries that have a similar language. Evidence regarding the effect of geo-graphical distance is inconclusive.These findings are interpreted as supportive of the conjecture that the home-bias effect is a result of asymmetric information between investors.

(2)

1

Introduction

THE HOME-BIAS EFFECTin equity investing is an observed and well-documented

phe-nomenon in behavioral finance. This bias can be described as the tendency of investors to overrepresent domestic equity in their portfolios even though the benefits of in-ternational equity diversification are well established in the modern portfolio theory literature (Levy and Sarnat, 1970; Grubel, 1968; Grauer and Hakansson, 1987). This systematic suboptimal portfolio construction results in suboptimal portfolio perfor-mance, and investors miss out on complementary portfolio improvements. By show-ing the presence of this phenomenon and examinshow-ing potential explanations, the lit-erature could make investors aware of this phenomenon and help to mitigate it. This could have positive effects on portfolio performance of, for example, pension or hedge funds.

A considerable body of literature has tried to find an explanation for this nonopti-mal investment behavior, resulting in multiple convincing explanations. The main ex-planation is that it is driven by information asymmetry between investors. Domestic investors appear to have—or believe they have—more knowledge of domestic stocks compared to foreign stocks. It is for this reason that they are more familiar with and optimistic about domestic stocks and therefore prefer to hold these stocks in their port-folios over foreign stocks.

The aim of this paper is to contribute to the current literature by investigating the persistence of the home-bias effect in equity investing and the influence of internet use, linguistic proximity, and geographical distance on home bias using unbalanced yearly panel data for 32 Organisation for Economic Co-operation and Development (OECD) countries over the period 2001–2018. It is hypothesized that internet use increases the accessibility of information and thus reduces the information asymmetry between do-mestic and foreign investors. Furthermore, internet use gives access to financial tech-nologies (fintechs) that could help investors with portfolio decisions, make it easier for investors to manage their portfolios, and remind them of the benefits of international portfolio diversification (Agarwal and Chua, 2020; Goldfarb and Tucker, 2019; Hueb-ner, Vuckovac, Fleisch, and Ilic, 2019; D’Acunto, Prabhala, and Rossi, 2019). Moreover, investors might possess more information about foreign firms located in countries that are geographically closer to and use similar languages to that of the investor (Coval and Moskowitz, 1999; Hau, 2001; Massa and Simonov, 2006; Grinblatt and Keloharju, 2001).

(3)

distinction between investments in OECD and non-OECD countries. Then, Mundlak models were used to find the overall effect of internet use on the portfolio internation-alization of OECD investors in these same groups of countries. Thereafter, this paper focused on OECD country investments in other OECD countries, again using a Mund-lak model, to assess the effects of internet use, linguistic proximity between countries, and the geographical distance between countries on portfolio internationalization in these OECD countries.

The main findings in this paper are that the home-bias effect is a persistent phe-nomenon for OECD country investors and that this bias is decreasing more rapidly regarding foreign investments in other OECD countries than regarding foreign invest-ments in non-OECD countries. This implies that OECD investors are beginning to reap the benefits of international diversification for portfolio performance but are still reluctant to invest in non-OECD countries. Internet use increases portfolio interna-tionalization and therefore reduces the home-bias effect. However, this effect is not present considering OECD investor investments in non-OECD countries in the 2010– 2018 period. Moreover, OECD country investors, when investing abroad, prefer to invest in countries that are linguistically similar. Weak or no evidence is found that investors prefer to invest in countries that are geographically closer. The robustness of these results based on Mundlak models was tested extensively by running the Mund-lak models excluding each country in the data once, running random- and fixed-effects models, and testing for structural breaks in the data.

This paper contributes to the current literature by adding a piece of the puzzle re-garding the information asymmetry explanation of the home-bias effect. It does so by simultaneously analyzing variables that the literature suggests can affect the home-bias effect and using a significant number of control variables. The findings regarding the effects of internet use and the observation of a difference in the magnitude and de-velopment of the home-bias effect of OECD investors regarding investments in OECD and non-OECD countries in the 2001–2018 period especially is complementary to the current literature.

(4)

2

Background

2.1

Modern portfolio theory

In 1952, Harry Markowitz published an article titled “Portfolio Selection” (Markowitz, 1952). This work formulated a basic framework to maximize the expected return for a given level of risk and became the basis of modern portfolio theory. Extending Markowitz’s work, Sharpe (1964) also argues that rational investors should maximize profit for a given level of risk, where risk is the variance of portfolio return. Investors can do so with the use of mean-variance analysis. The main idea is not to put all money in one or a few stocks but to create a diversified portfolio. Through diversification, some of the risks inherent in an asset can be avoided. Consequently, diversification can reduce the riskiness of a portfolio for a given level of expected return. Lintner (1965) added to the modern portfolio theory literature by introducing the separation theorem, a two-step method to construct a portfolio where it is assumed that investors can choose between risky and risk-free assets. The combined contributions of Sharpe (1964) and Lintner (1965) enable investors to work with the main ideas of Markowitz (1952) using the capital asset pricing model.

(5)

2.2

Home-bias effect

Obstfeld and Rogoff (2000) listed six major observed, irrational phenomena in inter-national macroeconomics for which proper explanations are yet to be found. A large number of papers have been written to test possible explanations for these six phe-nomena or to contribute to possible solutions. The home-bias effect in equity invest-ing, which is described as the too-strong preference of investors for domestic equity, is one of these six puzzles. Even though the benefits of diversification through inter-nationalization are well established in modern portfolio theory, research appears to suggest that investors make limited use of it and hold a too-large portion of domes-tic stocks in their portfolios. The notion of the portion of domesdomes-tic equity being too large is based on modern portfolio theory, which suggests, based on the international version of the capital asset pricing model as developed by Sharpe (1964) and Lintner (1965), that investors hold investments in specific countries based on the market cap-italization size of that specific country compared to total world market capcap-italization (Bekaert and Wang, 2009; Bekaert, Hoyem, Hu, and Ravina, 2017; Lewis, 1999). The portion of domestic stocks in an investor’s portfolio should therefore be close to the relative market capitalization of the investor’s country of residence.

One of the first papers to examine the home-bias effect is French and Poterba (1991), who recognized the benefits of international diversification and pointed out that most investors hold nearly all their wealth in domestic assets. Tesar and Werner (1995) also reported strong evidence of home bias in the national portfolios of Canada, Germany, Japan, the United Kingdom, and the United States during the period 1975–1990. An-other paper that established the existence of a home bias in the United States was Bekaert, Hoyem, Hu, and Ravina (2017), which examined the 401(k) retirement plans of 296 firms in the period 2005–2011 to determine the equity allocations of 3.8 million individuals and found a clear home bias in the portfolios.

(6)

2.3

Explanations for the home-bias effect

There is a sizeable body of literature seeking to explain the home-bias effect. The most important factors that the literature suggests affect home bias are discussed below, beginning with a variable that, counterintuitively, does not influence the degree of portfolio internationalization: transaction costs. Glassman and Riddick (2001), follow-ing Tesar and Werner (1995), examined the effects of transaction costs on international investment patterns and argued that transaction costs cannot explain the home-bias effect. Cooper and Kaplanis (1994) determined that international risk hedging or di-rect observable costs of international investments like transaction costs cannot explain the home-bias effect.

Other studies have found factors that affect the degree of portfolio internation-alization. First, focusing on capital controls and corporate governance are Dahlquist, Pinkowitz, Stulz, and Williamson (2003) and Kho, Stulz, and Warnock (2009). Dahlquist, Pinkowitz, Stulz, and Williamson (2003) concluded that foreign investor protection significantly increases the portion of equity invested abroad. The intuition behind this result is that higher investor protection reduces the risk of investing abroad, which in turn reduces the home-bias effect. Kho, Stulz, and Warnock (2009) quantified the logic that one must be able to buy foreign stocks to invest abroad and found that the stocks of firms that have less insider ownership are more easily attainable by foreign investors. Reducing the degree of insider ownership therefore reduces the home-bias effect.

The preference for domestic equity could also be driven by cultural factors or na-tionalism. Morse and Shive (2011) argue that a higher degree of patriotism results in less foreign equity investments. This argument is in line with the findings of Prad-khan (2016). Moreover, Benos and Jochec (2013) illustrated that investors are subcon-sciously driven toward stocks with patriotic names especially in times when patriotic sentiments are high, for example during a war.

Another explanation that has an empirically significantly effect on international investments is the effect of exchange rate volatility. Fidora, Fratzscher, and Thimann (2007) argue that exchange rate volatility induces a bias toward domestic financial assets because it increases the riskiness of foreign equity. This exchange rate volatility is not present if two countries use the same currency, which therefore positively affects the degree of portfolio internationalization.

(7)

diver-sification. Furthermore, Graham, Harvey, and Huang (2009) found that individuals with higher education hold more internationally diversified portfolios, which could be because these individuals better understand the benefits of international diversi-fication. Bekaert, Hoyem, Hu, and Ravina (2017) determined that individuals with access to financial advice also invest more internationally. If investors do not have the knowledge themselves to construct internationally diversified portfolios, they could benefit from consulting a financial expert.

The possible explanation that is most widely covered and supported by the litera-ture is that the home-bias effect is driven by information asymmetries between domes-tic and foreign investors, where domesdomes-tic investors have an information advantage over foreign investors regarding domestic equity (Dvoˇrák, 2005). French and Poterba (1991) also found that domestic investors have more information about domestic eq-uity than foreign investors. They argue that investors are more familiar and optimistic about stocks they have more information about. This familiarity and opportunism could be the main explanation of the home-bias effect. This view is supported by Gra-ham, Harvey, and Huang (2009), who also found that domestic investors are more optimistic about domestic stocks in comparison with foreign stocks. These findings are in line with Strong and Xu (2003).

2.4

Information asymmetry

The literature suggests that investors are more familiar or optimistic about domestic equity because they possess more information about these stocks compared to for-eign stocks. This contributes to the home-bias effect because investors prefer to hold domestic equity over foreign equity, which results in investors overrepresenting do-mestic equity in their portfolios. If this is in fact the case, factors influencing infor-mation availability should influence the degree of portfolio internationalization. The literature describes several variables that affect the degree of information asymmetry between domestic and foreign investors.

(8)

signif-icant positive information advantage for investors located closer to the headquarters of the traded stock. By examining Swedish and Finnish investors, Massa and Simonov (2006) also determined that investors invest more in stocks that are geographically or professionally close to them because they have more information. Grinblatt and Kelo-harju (2001) also observed a geographical preference. Using a dataset of Finnish and Swedish investors, they found that these investors preferred stocks of Finnish firms that are located close to the investor. All these papers, after finding a preference for investors to invest in firms that are geographically close by, argue that the home-bias effect can be explained by familiarity, where investors are more familiar with stocks they have more information about.

Second, the ease of communication trough language between an investor and the firm to be invested in could affect information availability. Besides the effect of the ge-ographical proximity between the investor and the headquarters of the firm invested in, Hau (2001) also found that German-speaking investors investing in German firms have an informational advantage over non-German speaking investors due to lan-guage similarity. Grinblatt and Keloharju (2001) also determined that besides a prefer-ence for firms that are closer by, Finnish investors prefer the stocks of Finnish firms that communicate in the same language as the investors and have chief executives with the same cultural background as they have because it makes the transfer of information easier. These findings are in line with Melitz and Toubal (2014).

Today, one of the most important factors regarding the accessibility of information is the internet. Over the last few decades, the World Wide Web has allowed easier access for many people to a great deal of information. Amadi (2004) found that in-ternet access significantly positively affects portfolio internationalization. One reason for this effect is that internet use decreases the cost of information and thus decreases the information asymmetry between domestic and foreign investors. According to Et-tredge, Richardson, and Scholz (2002), the internet offers firms the opportunity to en-hance their communication quality and reduces the cost of information distribution, further reducing information asymmetry. For example, since January 2007, the French Financial Authority has required firms listed on the Euronext-Paris stock exchange to disclose all mandatory information via the internet. Gajewski and Li (2015) eval-uated the effect of online disclosures on information asymmetry and found that the obligation for firms listed on the Euronext-Paris stock exchange to disclose their finan-cials online lowered information asymmetry in the French financial market. Hodge, Kennedy, and Maines (2004) also found that the internet increases the transparency of firms’ financial statements and therefore reduces information asymmetry.

(9)

accessibil-ity. Internet access also allows investors to use fintechs, which, according to Agar-wal and Chua (2020), improves portfolio diversification because these fintechs pro-vide financial advice. Goldfarb and Tucker (2019) found that internet use lowered the cost of financial advice and therefore increased its accessibility for household in-vestors. Moreover, fintechs can make it easier for households to manage their port-folios (Huebner, Vuckovac, Fleisch, and Ilic, 2019). These findings are inline with D’Acunto, Prabhala, and Rossi (2019) which analyzed the effect of fintechs and es-pecially robo-advicing on investor behaviour.

3

Data

1

3.1

Data description

The analyses in this paper focus on OECD country investors’ international investment behavior in the period 2001–2018, using yearly financial as well as nonfinancial data. This paper analyzes the home-bias effect of 32 of these OECD countries, using the data from the year they became an OECD member to 2018. Appendix 1 lists all the coun-tries that are in the OECD, whether the respective country is included in the analyses, and the year they entered the OECD. Hence, the analyses do not include all OECD countries. For example, in 2018, the OECD had 36 countries, 32 of which are examined in the analyses (data regarding Luxembourg, South Korea, Lithuania, and Latvia is lacking). For Luxembourg and South Korea, respectively no educational and Linguis-tic data could be obtained, and for Lithuania and Latvia, there are too few data points because they only entered the OECD in 2016 and 2018 respectively. Because not every country became a member in the same year, the timespan of the data differs for each country. Unbalanced panel data is therefore used.

The focus on OECD countries was chosen because these countries provide exten-sive and trustworthy data and are located all around the globe. The OECD is an orga-nization of countries that are committed to an open, transparent, and free market and together try to solve market inefficiencies. Moreover, these countries must be demo-cratic and protect human rights. The countries are generally high-income countries that rate high on the Human Development Index. The 2001–2018 time period was chosen because it is recent and therefore not extensively covered in the literature. The outcome variables used to capture the home-bias effect are, equal to Amadi (2004)

1This section motivates and explains the variables used to analyze the effect of internet use,

(10)

and Mishra (2015), based on the degree of portfolio internationalization. This paper used two variables for portfolio internationalization to assess the effects on invest-ments both in groups of countries and in specific individual countries.

The first variable is referred to as total portfolio internationalization and was mea-sured for three groups. This is defined as the portion of total equity investments invested by a specific OECD country in all foreign OECD countries, all non-OECD countries, and all foreign countries in the world (the sum of OECD and non-OECD countries). For example, for each of the 32 OECD countries in the dataset, the por-tion of total equity investments in 2018 invested in the 35 other OECD countries, all non-OECD countries, and all foreign countries in the world, which is the sum of the 35 foreign OECD countries and the non-OECD countries was determined.

The second variable is referred to as specific portfolio internationalization. This is defined as the portion of total equity investments invested by an OECD country in one other OECD country. Using the same example, for 2018, this variable measures the portion of equity invested by each of the 32 OECD countries in the dataset indi-vidually in each of the 35 individual foreign OECD countries. The combination of two specific OECD countries is referred to as a country pair. The sum of the specific portfolio internationalizations of one country in a given year is equal to the total port-folio internationalization of that country in OECD countries in that specific year. Both portfolio internationalization variables were calculated by establishing the portion of foreign investments of total investments made by all players in a country using data on international equity assets, domestic market capitalization, and foreign equity lia-bilities. The exact equational forms can be found in Appendix 2.

One of the independent variables of interest is internet use. It is reasonable to believe that the internet is a main determinant of how information travels across the world, and for this reason, internet use could reduce the information asymmetry between investors (Amadi, 2004). Internet use also gives investors access to fintechs that can help investors manage their portfolios (Agarwal and Chua, 2020). To capture internet use, this paper followed Amadi (2004) and used the percentage of the population of a certain country using the internet in a given year.

(11)

capitals of each country pair in kilometers was used. Regarding language, the litera-ture suggests that investors have more information about firms in countries that have a similar language as their home country, because they can understand reports, state-ments, and foreign news regarding these firms (Hau, 2001; Grinblatt and Keloharju, 2001; and Melitz and Toubal, 2014). The ease of communication trough language be-tween players in each country pair was measured using a common language index, which ranges from zero to one. The higher the number, the easier it is for inhabitants of these two specific countries to communicate through spoken and written language. This common language index was designed by the Centre d’Études Prospectives et d’Informations Internationales and is based on Fearon and Laitin (2000) and Fearon (2003). The index is calculated using country data regarding countries’ official lan-guages, national lanlan-guages, common spoken lanlan-guages, and etymological factors. To achieve reliable and meaningful outcomes regarding the effect of the independent variables of interest on portfolio internationalization, control variables are needed to control for external factors. First, investors’ information availability about firms in a specific foreign country might be influenced by the degree of trade between their domestic country and this specific foreign country. As a measure of bilateral trade in-tensity between countries, the imports and exports of a country with another specific country are taken as the portion of that country’s total imports and exports. Aviat and Coeurdacier (2007) found evidence that entrepreneurs located in different coun-tries can learn about each other through trade, which has a positive effect on the trade of financial assets. The intensity of trade could thus positively influence information availability and therefore increase the degree of specific portfolio internationalization. On the other hand, if two countries are heavily connected through trade, their mar-kets have higher correlation, which reduces the efficiency of diversification (Calderon, Chong, and Stein, 2007; Chen and Zhang, 1997). Therefore, it might also be logical from a modern portfolio theory perspective that higher bilateral trade intensity results in lower equity investments in that specific country.

(12)

Third, the analysis should include domestic market capitalization as a percentage of total world market capitalization, relative market capitalization. Modern portfolio theory suggests investors hold domestic and foreign equity in proportions equal to the relative market capitalization of these countries (Bekaert and Wang, 2009; Bekaert, Hoyem, Hu, and Ravina, 2017; Lewis, 1999). If the relative domestic market capital-ization increases, for example, it is logical from a modern portfolio theory perspective for domestic investors to hold less foreign equity and more domestic equity. The de-viation of investor behavior from this logic is the essence of the home-bias effect.

Fourth, portfolio diversification is a complex task that necessitates financial knowl-edge, which might be obtained through education. With this financial knowlknowl-edge, or financial literacy, investors might become aware of the benefits of diversification and able to exploit these benefits. Hence, education could, as proven by Giofré (2017) and Graham, Harvey, and Huang (2009), positively affect the degree of portfolio in-ternationalization. As a variable for education, this paper used the percentage of the population aged between 25 and 64 years that has an educational degree on the ter-tiary level in a specific year for each specific country. Terter-tiary education is defined as academic education that is normally taught in universities.

Fifth, economic factors could play a role in the degree of portfolio internationaliza-tion. If the home country’s economy is growing, it might make sense to invest more in domestic equity, and conversely, if it is shrinking, in foreign equity. On the other hand, Riff and Yagil (2016) found that the home-bias effect increases in times of finan-cial crises because investors perceive the riskiness of foreign stocks to be higher than domestic stocks. These finding are complementary to the conclusions made in Gian-netti and Laeven (2012) which found that lenders rebalance their lending portfolios to domestic borrowers during financial crises. To capture a country’s overall economic trend, a measure based on the composite leading indicator (CLI) of the specific OECD countries was used. The CLI forms an index that provides signals on turning points in business cycles of countries. In other words, the index provides signals on fluctua-tions in economic activities of a specific country around its long-term potential, where the long-term potential is indexed at 100. The CLI can accurately predict these fluctua-tions six to nine months before the actual economic activity. The variable used consists of the deviation of the CLI of a specific country from the long-term potential in a given year. Using this method a negative (positive) sign indicates expected economic growth to be lower (higher) than the long-term potential.

(13)

treaties regarding equity investments between member states. Moreover, both using the euro takes away some investment risk because there is no exchange rate volatility, which has been found to have an effect on home bias (Fidora, Fratzscher, and Thi-mann, 2007).

There are other variables that reasonably influence portfolio internalization that are not discussed in this paper. A seemingly obvious variable affecting the degree of port-folio internationalization is transaction costs. However, Glassman and Riddick (2001), Tesar and Werner (1995), and Cooper and Kaplanis (1994) have proven that transac-tion costs cannot explain the home-bias effect. It is for this reason that this paper did not explicitly analyze the effect of transaction costs on portfolio internationalization.

By combining different datasources this paper suffers from missing data. The an-alyzes conducted in this paper were executed using rows with data for all variables. Please consult Appendices 2 and 3 for a more thorough explanation of variables and data sources.

3.2

Descriptive statistics

Table 1 lists the descriptive statistics for the above mentioned variables, and Figure 1 consists of graphs regarding six of these variables.

(14)

effect.

Figure 1 graph 1 also indicates that the largest portion of equity that is invested abroad by OECD country investors is invested in OECD countries, with only a small portion invested in non-OECD countries. This is also illustrated in Table 1, where the unweighted average total portfolio internationalization in OECD countries over time is 32.54% but only 5.35% in non-OECD countries, indicating that OECD investors prefer to invest in other OECD countries when investing abroad. Moreover, there is a clear difference in the development of the degree of total internationalization in OECD countries compared to non-OECD countries. While there was an increase in the per-centage of total equity invested by OECD country investors in all foreign countries in the world, the percentage invested in non-OECD countries stayed approximately the same over the period. This indicates that the decrease in home-bias effect of the OECD countries is almost entirely driven by investments in OECD countries. This ob-servation cannot be explained by the increase in the number of countries that became a members of the OECD over the years, because Figure 1 graph 4 reveals that the com-bined market share of the OECD countries is decreasing over time, which, according to Bekaert and Wang (2009), Bekaert, Hoyem, Hu, and Ravina (2017), and Lewis (1999), should result in an increase in equity invested in non-OECD countries. These obser-vations indicate that the home-bias effect is a persistent and decreasing phenomenon, and the decrease is mostly driven by investments in OECD countries.

Figure 1 graph 2 illustrates the unweighted average rate of internet use for the OECD countries included in the data. As expected, the use of the internet has dramatically increased over the years, from an average of about 38% in 2001 to an average of about 82% in 2018.

Table 1 reveals that the average value of openness is 0.6978. This means that on av-erage, the total value of imports and exports is 69.78% of the GDP of a specific country in a specific year. Figure 1 graph 3 illustrates the average unweighted openness of the OECD countries analyzed over time. The graph clearly indicates an increase in openness over the years. This increase could be a sign of increasing globalization. The drop around 2007–2009 implies a reduction in international trade, possibly due to the 2007–2008 financial crisis.

(15)

capital-Table 1: Descriptive statistics

Observations Mean Median Standard Min Max Deviation

Total Portfolio Internationalization World 520 0.3789 0.3888 0.2181 0.0003 0.9408 Total Portfolio Internationalization OECD 520 0.3254 0.3424 0.2046 0.0003 0.7979 Total Portfolio Internationalization Non-OECD 520 0.0535 0.0377 0.0532 0.0000 0.4128 Specific Portfolio Internationalization 13410 0,0127 0.0015 0.0343 0.0000 0.4649

Internet Use 539 0.6689 0.7240 0.2198 0.0519 0.9901

Linguistic Proximity 15405 0.2299 0.1561 0.1831 0.0237 0.9230

Geographical Distance 15936 5262 2373 5409 60 19586

Bilateral Trade Intensity 15926 0.0247 0.0079 0.0479 0.0000 0.7415

Openness 549 0.6978 0.5821 0.3669 0.1733 1.8259

Relative Market Capitalization 530 0.0273 0.0049 0.0744 0.0000 0.5215

Education 511 0.2999 0.3074 0.1076 0.0844 0.5789

Composite Leading Indicator 549 0.004 0.0016 0.0192 -0.1068 0.0807

European Union 15861 0.3693 0.0000 0.4826 0.0000 1.0000

Eurozone 15861 0.1614 0.0000 0.3679 0.0000 1.0000

Note: This table displays the unweighted descriptive statistics of the variables using data regarding the OECD countries in the dataset over 2001–2018. With unweighted it is meant that all countries have the same weight in contributing to the average in a given year. The portfolio internationalization variables, internet use, openness, relative market capitalization, bilateral trade intensity, education, and Composite Leading Indicator are percentages expressed in decimal form. The variable geographical distance covers the great circle distance between the capitals of a country pair. Linguistic proximity is expressed using an index ranging from 0 to 1 where the higher the number the easier individuals in a country can communicate trough language with individuals in another specific country. European Union and Eurozone are dummy variables taking value one if both countries of a country-pair are in the respective association and zero otherwise. Please consult Appendices 2 and 3 of a thorough explanation of the variables

ization has decreased over time. A possible cause could be the increasing importance of emerging countries like China in the financial market.

Figure 1 graph 5 indicates that the unweighted average percentage of people who have a tertiary education degree in an OECD country is steadily increasing. The high-est degree of tertiary education is 57.89% (Canada in 2017), and the lowhigh-est is 8.44% (Turkey in 2001).

In Table 1, the CLI ranges from -10.68% below the long-term potential to 8.07% above the long-term potential. Figure 1 graph 6 reveals that there are significant fluc-tuations in the unweighted average index deviation around zero. The large downward spike around 2007 indicates that GDP growth was expected to be lower than the long-term potential, which again is a sign of the financial crisis.

The bilateral trade intensity variable ranges from 0% to 74.15%. On average, 2.47% of total trade of an OECD country is with another OECD country.

(16)

0.0237 to 0.9230, where the higher the index, the better investors of the country pair can communicate.

Finally, the table reveals that for 36.93% of all country pairs both countries are in the EU, and for 16.41% of all country pairs both countries are in the Eurozone.

0 10 20 30 40 50

Average Portfolio Internationalization in %

2000 2005 2010 2015 2020 Year World OECD Non−OECD (1) Portfolio Internationalization 20 40 60 80

Average internet use in %

2000 2005 2010 2015 2020 Year (2) Internet Use 60 65 70 75 Average openness in % 2000 2005 2010 2015 2020 Year (3) Openness 75 80 85 90 95

Cumulative market capitalization in %

2000 2005 2010 2015 2020 Year (4) Market Capitalization 20 25 30 40 35

Average tertiary education in %

2000 2005 2010 2015 2020 Year (5) Tertiary Education −4 −2 0 2

Average deviation from the CLI in percentage points

2000 2005 2010 2015 2020

Year

(6) Composite Leading Indicator

Figure 1: Graphs of various variables in the data over time

(17)

4

Methodology

4.1

Regressions and variable transformations

To examine the effects of internet use, linguistic proximity, and geographical distance on the home-bias effect, linear regression models were used. First, the effect of internet use on total portfolio investments of OECD country investors in all countries in the world (OECD and non-OECD countries combined), OECD countries, and non-OECD countries was analyzed. This way, it can establish whether the effect of internet use is equal or different when considering investments in different groups of countries. This paper then focused on OECD countries and analyzed the effects of internet use, linguistic proximity, and geographical distance of OECD countries on investments in other OECD countries.

To see the effect of internet use on the degree of portfolio internationalization in the different groups of countries, linear regression specifications (1.1–1.3) were used, which are formulated as:

TPIworldit =α+β1IUit+β2Oit+β3MCit+β4Eit+β5CLIit+εit (1.1)

TPIoecdit =α+β1IUit+β2Oit+β3MCit+β4Eit+β5CLIit+εit (1.2)

TPInonoecdit =α+β1IUit+β2Oit+β3MCit+β4Eit+β5CLIit+εit (1.3) Where:

TPIworldit =TPIoecdit+TPInonoecdit (2)

(18)

the percentage of the population of country i that used the internet in year t. To capture the effect of the degree of openness of country i in year t on total portfolio internation-alization, Oit was added. Because modern portfolio theory suggests that investors hold portfolios in which the weights of the domestic and foreign equity investments are similar to their relative market capitalization, the variable MCit was added to in-clude the relative market capitalization of country i in year t. Because portfolio diver-sification is a difficult subject for which knowledge is needed, the educational variable Eit was added. To capture the effect of economic factors, the percentage point differ-ence of the CLI from its long-term potential was used as CLIit. Finally, the regression specifications have a constant, α, and an error term, εit. The equation for TPIworldit, is equal to Amadi (2004) and Mishra (2015). The equations for TPIoecdit, TPInonoecdit are a natural deviation from this equation. The equational forms of Oit, and MCit are equal to Amadi (2004) and Campa and Fernandes (2006). All the equations and vari-albe descriptions can be found in Apendices 2 and 3.

Because the data indicate that most foreign investments by OECD investors are in other OECD countries, formula (3) was used to focus on OECD country investments in other OECD countries. This formula allows to focus specifically on the effects of in-ternet use and the interaction variables linguistic proximity and geographical distance on the portion of total equity investors of one OECD country invest in another specific OECD country. This formula also includes the control variables and has the functional form:

SPIjt=α+β1IUit+β2LPj+β3GDj+β4BTIjt+β5Oit+β6MCit+β7Eit+β8CLIit+ β9EUjt+β10Cjt+εjt (3) In this formula, specific portfolio internationalization, SPIjt, is calculated as the per-centage of total equity assets country i invests in country x in year t. The country pair of i and x is denoted as j where i is one of the 32 OECD countries included in the analysis and x is any other OECD member country in year t (Appendix 1). As noted earlier, the countries in the analyses, i, do not cover all OECD countries x while this paper does analyze investments of these i countries in all other OECD countries x2 So, trough out the entire timespan considered, x exceeds i. This paper wanted to test the effect of internet use IUit on specific portfolio internationalization in other OECD countries. As explained in the data section, it is logical to assume that

in-2This paper lacks data for the following OECD countries: Latvia, Lithuania, Luxembourg, and South

(19)

creasing trade between countries results in higher availability of useful information for investor while potentially reducing the effectiveness of diversification through in-ternationalization. Therefore, a bilateral trade intensity variable BTIjtwas added for country pair j in year t. The geographical distance and linguistic proximity for country pair j are represented by the GDj and LPj variables, and these variables are constant over time. EUjtand Cjt are dummy variables indicating whether both countries in the country pair are member of the EU or the Eurozone, respectively. The formulas used to calculate SPIjt and BTIjt can be found in Appendix 2. The variables Oit, MCit, Eit, and CLIit have the same explanations and were added following the same reasoning as for (1.1–1.3).Equation (3) also contains a constant α and an error term εjt.

This paper used yearly data for multiple entities over different points in time, also called panel data or longitudinal data. Unbalanced panel data was used because the number of points in time per entity are not equal. For regression specifications (1.1– 1.3), the different panels are the 32 i OECD countries. For regression specification (3), the different panels are the j country pairs. To find the appropriate model to run re-gressions (1.1–1.3) and (3), some characteristics of the individual variables must first be established.

The first assumption in most time series models is that individual variables are stationary over time. A stationary series can be defined as one with a constant mean, constant variance, and constant autocovariance for each given lag (Brooks, 2019). This is a desirable feature, because nonstationary variables will indicate that previous val-ues of the error term have a nondeclining effect on the current value of the error term. Models using nonstationary data can therefore be unpredictable by falsely indicating significant relationships between variables.

(20)

For both regressions (1.1–1.3) and (3), the test suggests that there is a unit root present in the variables for openness, relative market capitalization, and education (Appendices 4.1 and 4.2). An additional Fisher augmented Dickey-Fuller test is used for these variables considering a trend in the data. In other words, it tests whether the data is stationary if a time trend is included. Based on this test, it can be concluded that the variables openness and relative market capitalization are not trend-stationary (Appendices 3.3 and 3.4) in both (1.1–1.3) and (3). To make these variables stationary, they are transformed by using the first difference of these variables. The variable for education is trend-stationary for both regressions (Appendices 3.3 and 3.4). This vari-able is made stationary by de-trending the varivari-able by conducting linear regression analyses, per panel, of education as the dependent variable with a time trend for the years as an independent variable. The residuals of this regression are the de-trended values of education, which are stationary.

The corrected regression specifications therefore have the following functional form where d. indicates that the first difference of the variable was used and s. that the vari-able is stationary using de-trending:

TPIworldit =α+β1IUit+β2d.Oit+β3d.MCit+β4s.Eit+β5CLIit+εit (4.1)

TPIoecdit =α+β1IUit+β2d.Oit+β3d.MCit+β4s.Eit+β5CLIit+εit (4.2) TPInonoecdit =α+β1IUit+β2d.Oit+β3d.MCit+β4s.Eit+β5CLIit+εit (4.3)

(21)

variables results in coefficients that are sensitive to small changes in the model and larger standard errors. Hence, multicollinearity reduces the precision of the estimated coefficients and reduces the statistical power of the model. Therefore the model might not show significant results even when the variable has a significant effect. In other words, the model might be too conservative, which is not wrong but is something to be considered.

4.2

Model specification

Classical models for unbalanced panel data are either fixed-effects or random-effects models. A fixed-effects model adds a time consistent variable per entity to account for entity specific unobserved variables (Brooks, 2019). Fixed-effects models therefore do not produce estimates based on variables that are constant over time per entity because these effects are captured by the added time consistent variables per entity (Wooldridge, 2012; Wooldridge, 2019). A limitation of the fixed-effects model is that it treats these unobserved variables as if they affect the observed variables, while this is not necessarily the case, because this effect could also be random. This can result in too-high standard errors.

A random-effects model also accounts for these entity specific unobserved vari-ables but assumes that their effects are random and uncorrelated with the independent variables in the model (Wooldridge, 2012). This is also a major assumption. A bene-fit of random-effects models over fixed-effects models is that random-effects models do produce estimates for explanatory variables that are constant over time per en-tity. Moreover, random-effect models are estimated using generalized least squares (GLS) to account for possible substantial positive serial correlation in the error term Wooldridge, 2012). This has the advantage that the estimated coefficients are not af-fected by potential heteroskedasticity in the model (Brooks, 2019).

(22)

observed variables. Moreover, the Mundlak model, just like the random-effects model, uses GLS and therefore solves for potential heteroskedasticity in the model (Brooks, 2019).

To test whether the Mundlak model or fixed-effects model better fits regressions (4.1–4.3) and (5), it was tested whether the ‘Mundlak estimates’ are significantly dif-ferent from zero using Wald tests. Based on the results of these tests (Appendix 6), it can be concluded that with a 10% level of significance, at least one Mundlak estimate per regression is significantly different from zero. Based on these results, it can be con-cluded that the Mundlak model is most appropriate for these regressions. It is for this reason that the Mundlak model was chosen as the main model for the analysis in this paper.

To correct for country or country pair specific effects, standard errors clustered for the panels in the regression were used. The combination of GLS and panel clustered standard errors will, according to Nichols and Schaffer (2007), result in unbiased and correctly sized standard errors. The Mundlak model, using clustered standard er-rors, stationary variables, and GLS, appears to be fairly resistant to possible biases. It is for this reason that no special corrections regarding serial autocorrelation were made. Moreover, the model was corrected for possible heteroskedasticity by using GLS Brooks, 2019). Endogeneity should also not be a problem since precise and reli-able data sources were used. Finally, the significant number of control varireli-ables should prevent omitted variable biases.

5

Results

In the data section, it was observed that most of the equity invested abroad by OECD countries is invested in other OECD countries. The degree of portfolio international-ization by these OECD countries in OECD countries is also growing more rapidly over time than in non-OECD countries. To assess the effect of internet use on the degree of total portfolio internationalization, the Mundlak models for the linear regression models (4.1–4.3) were ran to distinguish between investments in OECD countries and non-OECD countries. The results regarding the coefficients of these regressions are displayed in Table 2.

The levels of significance are equal for all different groups except for the constant. This has some implications.

(23)

increase or decrease in the degree of openness of country i from one year to another does not affect the home bias of investors in country i. This outcome is in line with Amadi (2004) and contrary to Lane and Milesi-Ferretti (2003). Second, the coefficient of the first difference of relative market capitalization has no significance. Therefore, the degree of total portfolio internationalization appears to be unaffected by changes in relative market capitalization from one year to another. This result indicates in-vestor irrationality, because the literature (Bekaert and Wang, 2009; Bekaert, Hoyem, Hu, and Ravina, 2017; Lewis, 1999) suggests the weights of domestic and foreign eq-uity assets in portfolios to be determined by relative market capitalization. Third, the stationary education variable is insignificant, indicating that the degree of a country’s population that has a tertiary educational degree, above or below the time trend, does not affect the degree of total portfolio internationalization. This result is in contrast to Giofré (2017); however, that paper used finance-specific education, or financial liter-acy, instead of education overall. Finally, the CLI variable is insignificant, indicating that the expected economic growth above or below the long-term potential does not affect total portfolio internationalization. Based on Riff and Yagil (2016) we expected this variable to be signficantly positive. These variables are insignificant when consid-ering investments in all countries in the world (column 1), investments only in OECD countries (column 2), and investments in non-OECD (column 3) countries.

(24)

Table 2: Mundlak analysis of total portfolio internationalization

(1) (2) (3)

World OECD Non-OECD

Internet Use 0.3906*** 0.3254*** 0.0654***

(0.0628) (0.0590) (0.0191)

d.Openness 0.0168 0.0154 0.0010

(0.0552) (0.0432) (0.0197) d.Relative Market Capitalization -0.1549 -0.3977 0.2419

(0.4209) (0.2851) (0.1752)

s.Education -0.1380 -0.3226 0.1827

(0.4634) (0.4940) (0.2089) Composite Leading Indicator 0.0002 -0.0005 0.0006

(0.0025) (0.0022) (0.0006)

Constant -0.4122*** -0.4044*** -0.0072

(0.1493) (0.1314) (0.0427)

Observations 451 451 451

Number of Countries 32 32 32

Note: This table displays the effect of various variables on portfolio internationalization by OECD coun-try investors considering internationalization in all foreign countries in the world (column 1), foreign OECD countries (column 2), and foreign non-OECD countries (column 3) using Mundlak models based on the respective regressions (4.1), (4.2), and (4.3). Please consult Appendices 2 and 3 for a detailed description of the variables. d. indicates that a variable transformation is used to make the variable stationarity by taking the first-difference, s. indicates that a variable transformation is used to make the variable stationary by de-trending the original variable as described in section 4.1. The standard errors are presented in parentheses where *** p<0.01, ** p<0.05, * p<0.1.

Internet use appears to affect the degree of portfolio internationalization of OECD country investors in other countries. Regression (5) is used to analyze the effect of the internet and other variables on specific portfolio internationalization by focusing on the portion of total equity invested by one OECD country in another OECD coun-try. The outcomes based on this regression can be found in Table 3. This regression specification produces outcomes that are more precise than the outcomes based on regressions (4.1–4.3), presented in Table 2, because it uses more variables including control variables for bilateral trade, both countries being in the EU, and both countries using the euro. Moreover, because regression (5) observes the relationship between in-dividual OECD countries, the results produced are based on more observations than the results of (4.1–4.3). It is for these reasons that differences between the outcomes based on (4.1–4.3) and (5) might be observed.

(25)

pos-itive coefficient, and Calderon, Chong, and Stein (2007) and Chen and Zhang (1997), who found a negative coefficient.

Contrary to the results as presented in Table 2, evidence was found that deviations in a country’s relative market capitalization from one year to another affect specific portfolio internationalization. This indicates that a one percentage point increase (de-crease) in relative market capitalization from one year to another results in a 0.0152 percentage point decrease (increase) in equity invested by OECD country investors in another specific OECD country in a specific year, significant at the 10% level. Hence, OECD investors do react to the relative market capitalization of their domestic coun-try.

Also contrary to Table 2, the CLI has a small and positive significant effect on spe-cific portfolio internationalization in this model, significant at the 5% level. This indi-cates that more favorable economic prospects in the home country of an OECD coun-try investor result in more investments by this investor in other OECD countries. A possible explanation for this result could be that investors prefer to reduce their de-pendency on the domestic economy by investing internationally. On the other hand, negative economic prospects result in a decrease in relative specific portfolio interna-tionalization and therefore an increase in the home-bias effect. These results are in line with Giannetti and Laeven (2012) and D’Acunto et al. (2019).

Whether both countries of a country pair are in the EU has no significant effect. On the other hand, both countries of a country pair being in the Eurozone has a pos-itive effect on the degree of specific portfolio internationalization significant at the 5% level. The Eurozone outcome is similar to the findings of Fidora, Fratzscher, and Thimann (2007), which suggest that both countries using the euro should positively affect investments between these countries because it removes the exchange rate risk of international investments.

Overall, these results regarding the control variables based on regression (5) are similar to those based on regressions (4.1–4.3) except for the relative market capital-ization and CLI variables.

(26)

internationaliza-tion by 0.0131 percentage points in a specific year, ceteris paribus, significant at the 1% level. This effect could again be explained by a reduction in information costs and the accessibility of fintechs due to internet use.

The coefficient for linguistic proximity is positive and significant at the 1% level. This indicates that higher (lower) linguistic proximity between countries results in higher (lower) portfolio investments between these countries. A potential explanation of this result is that information is more easily attained when an investor speaks a sim-ilar language as that in the country where a firm is located, which increases optimism and familiarity toward stocks of a specific country. In other words, ease of communica-tion trough language can affect informacommunica-tion asymmetry and therefore the home-bias effect. This outcome is in line with Hau (2001), Grinblatt and Keloharju (2001), and Melitz and Toubal (2014). However, since the common language index ranges from 0 to 1, the difference between countries that have very different (common language in-dex zero) or similar (common language inin-dex one) languages is only 2.14 percentage points, ceteris paribus.

In line with Coval and Moskowitz (1999), Hau (2001), Massa and Simonov (2006), and Grinblatt and Keloharju (2001), a significant and negative effect for geographi-cal distance was found. This result implies that the distance between countries does influence the proportion of a portfolio invested in another specific country by these in-vestors. An increase (decrease) in the distance between two specific countries of 1000 kilometers results in a decrease (increase) in specific portfolio investments of 0.0003 percentage points, ceteris paribus. However, with a 10% level of significance, this effect is only weakly significant. An explanation for this result, grounded in the literature, is that investors possess more information about firms that are located closer to them. Hence, the geographical distance between the country of an investor and the country where a firm is listed could influence information asymmetry between investors and therefore the home-bias effect.

(27)

Table 3: Mundlak analysis of specific portfolio internationalization in OECD countries (1) Internet Use 0.0131*** (0.0029) Linguistic Proximity 0.0214*** (0.0063) Geographical Distance -0.0003* (0.0002)

Bilateral Trade Intensity -0.0442

(0.0568)

d.Openness -0.0014

(0.0015) d. Relative Market Capitalization -0.0152* (0.0082)

s.Education -0.0142

(0.0190)

Composite Leading Indicator 0.0001**

(0.0001)

Both in European Union -0.0024

(0.0019) Both in Eurozone 0.0099** (0.0045) Constant -0.0095** (0.0045) Observations 9,843

Number of Country pairs 944

Note: This table displays the results of various variables on portfolio internationalization considering internationalization by investors in one OECD country in another OECD country using a Mundlak model based on regression (5). The geographical distance variable is expressed per 1000 kilometers. Please consult Appendices 2 and 3 for a detailed description of the variables. d. indicates that a variable transformation is used to make the variable stationarity by taking the first-difference, s. indicates that a variable transformation is used to make the variable stationary by de-trending the original variable as described in section 4.1. The standard errors are presented in parentheses where *** p<0.01, ** p<0.05,

* p<0.1.

exceeds the long-term potential, this investor invests a larger portion of their port-folio in another OECD country. These results are somewhat ambiguous, because no significant effect was found for investments by OECD countries in other OECD coun-tries based on regressions (4.1–4.3) presented in Table 2. Moreover, both councoun-tries of a country pair using the euro positively affects specific portfolio internationalization.

6

Robustness

Four checks to test the robustness of the models were used to determine whether the conclusions made in the result section regarding the effect of internet use, linguistic proximity, and geographical distance hold when the regressions (4.1–4.3) were ran using different models.

(28)

were ran, excluding each country from the sample once. With this method, one can identify whether the outcomes depend on data for one specific country. The outcomes of this robustness check can be found in Appendices 7.1–7.4. The tables list the results based on the independent variable(s) of interest and the control variables. However, the tables do not display the results regarding the control variables. Based on this ro-bustness test, it can be concluded for regressions (4.1–4.3) that the results are similar in signs and levels of significance to those in Table 2. The results for regression (5), excluding each country once (Appendix 7.4), are also similar to Table 3 in signs and significance except for the geographical distance variable. This variable is insignifi-cant when specific countries are removed from the regression. Hence, the effect of geographical distance on specific portfolio internationalization is weakly present or not present. Overall, it can be concluded that the conclusions based on the Mund-lak models do not strongly depend on data regarding specific countries except for the geographical distance variable.

As a second check for the robustness of the models, the regressions were ran using effects models instead of Mundlak models. The results based on effects models can be found in Appendices 8.1 and 8.2. The outcomes of the random-effects model for regressions (4.1–4.3) are similar to the outcomes of the Mundlak model as presented in Table 2. Regarding regression (5), three deviations in the con-trol variables were found when comparing the results from the random-efilar in sign and significance levels as the outcomes using Mundlak models. This again proves the robustness of the cofects model with the results from the Mundlak model. First, the bilateral trade intensity variable is positive and significant at the 5% level, indicat-ing that more trade between countries results in relatively more equity investments between countries. Second, the coefficient of the variable for both countries in a coun-try pair being in the EU is significant at the 10% level and negative, a result that is somewhat counterintuitive. Third, the coefficient for relative market capitalization is insignificant. Even though these differences in control variables between the mod-els, the outcomes of the main independent variables are still similar to those from the Mundlak models. Hence, both models would result in the same conclusion. It is for this reason that the outcomes based on the Mundlak models can be regarded as robust. As a third check, the regressions were ran using fixed-effects models. These results can be found in Appendices 9.1 and 9.2. Appendix 9.2 does not show the coefficients and standard errors for linguistic proximity and geographical distance because these variables are constant over time between country pairs. The outcomes of these fixed-effects models are simnclusions based on the Mundlak models.

(29)

a test was ran to determine whether outcomes produced using Mundlak models based on equation (4.1–4.3) and regressions (5) apply to the whole period 2001–2018. The presence of a structural break would imply that there are significant differences in one or more coefficients if the regression were ran using a different period subset within the total period considered.

Appendix 10.1 displays the results of a Chow test for structural breaks. The results of the tests imply that there is a significant difference in coefficients if these coefficients are produced using data from 2001–2009 versus 2010–20183. This significant difference could potentially be the result of different behavior before and after the financial crisis of 2007–2008.

Hence, based on this Chow test, it is concluded that there is a structural break in the data. The data is therefore subdivided in two subsets, one for 2001–2009 and one for 2010–2018. The results regarding regressions (4.1–4.4) and (5) based on these two subsamples using Mundlak models are presented in Appendices 10.2 and 10.3. The results for regressions (4.1–4.4) presented in Appendix 10.2 deviate from the results presented in Table 2. The first difference variable for openness has a negative coef-ficient in the 2001–2009 period and a positive coefcoef-ficient in the 2010–2018 period for investments in all foreign countries, both significant at the 10% level. This indicates that a country becoming more internationally open from one year to another had a surprisingly negative effect on total portfolio internationalization in 2001–2009 and a positive effect, in line with Lane and Milesi-Ferretti (2003), in 2010–2018. If the ef-fect are examined more closely by focusing on investments in OECD countries versus those in non-OECD countries, the effect is negative for investments in OECD countries in 2001–2009 and insignificant in 2010–2018. For investments in non-OECD countries, it is insignificant in both periods. More interestingly, the coefficient for the effect of in-ternet use on total portfolio internationalization in non-OECD countries in the period 2010–2018 (column 6) is insignificant. This indicates that the positive effect of inter-net on total portfolio internationalization regarding investments of OECD investors in non-OECD countries as found in Table 2 only applies for the 2001–2009 period. Another interesting result is that the effect of internet on total portfolio internation-alization by OECD country investors in all other countries in the world and in other OECD countries appears to be larger in the 2010–2018 period than in the 2001–2009 period indicating that the effect of internet is nondeclining over time.

Regarding the results of regression (5) when split into two subsets, presented in Appendix 10.3, all outcomes are similar to Table 3 except for geographical distance,

3Multiple Chow tests where conducted examining structural breaks at different points in the

(30)

countries in a country pair using the euro and CLI. The geographical distance vari-able is only significant in 2010–2018 and is not significant in 2001–2009. This same effect is found for the variable regarding both countries using the euro. These results could potentially be explained by the difference in countries included in the analysis in both periods. This same reasoning might also explain the insignificance of the CLI coefficient in 2001–2018. Difference might also be explained by differences in investor behavior before and after the financial crisis. Other interesting results are that the coef-ficient of the first difference variable for relative market capitalization and the variable for internet use are larger in 2010–2018 than in 2001–2009. These results might explain the decrease of the home-bias effect in equity investing over the years. Moreover, it proves that the effect of internet use on portfolio internationalization is not decreasing over time.

Overall, regarding this robustness check, the important conclusion was made that internet use only affect total portfolio internationalization of OECD investors in non-OECD countries in 2001–2009. Moreover, CLI only appears to affect specific portfolio internationalization in 2001–2009. The effect of geographical distance is only found in 2010–2018. Furthermore, it was concluded that the degree of openness of a country might affect specific portfolio internationalization, but this effect is different between investments in different groups of countries and different periods.

Using four techniques to test the robustness of the model, only small differences were found between the outcomes of these different techniques and the models as presented in Tables 2 and 3. No extensive attention was paid to the robustness of the control variables because the aim of the robustness section was to check the robustness of the conclusions derived from the Mundlak models presented in Table 2 and 3.

(31)

7

Conclusion & Discussion

The goal of this paper was to contribute to the literature on the home-bias effect in equity investing by adding another piece of the puzzle regarding the main explanation that it exists due to information asymmetries.

Using a dataset including 32 OECD countries in the period 2001–2018, this pa-per began by observing that the home-bias effect is a pa-persistent but decreasing phe-nomenon. However, the decrease in the home-bias effect, in other words the increase in the degree of portfolio internationalization, was stronger for investors from OECD countries in other OECD countries and not in non-OECD countries. Investors in OECD countries appear to hold a too-large portion of their portfolios in domestic eq-uity, and when they do invest abroad, they focus on investments in other OECD coun-tries and seem blind, unable, or unwilling to invest in non-OECD councoun-tries. These investors are improving portfolio performance by using diversification through inter-nationalization, but there is still room for improvement.

This paper tried to answer the research question on the effects of internet use, lin-guistic proximity, and geographical distance on the home-bias effect. The main hy-pothesis was that an increase in internet use would result in a decrease in the home-bias effect, because the use of the internet improves information availability, allows for easier portfolio management, and gives investors easy access to financial advice through fintechs. It was also hypothesized that language similarity, or ease of commu-nication trough language, would reduce information asymmetry and therefore reduce the home-bias effect and that investors prefer to invest in stocks that are geographi-cally closer because they have more information about firms that are located close by. The results indicate that internet use indeed has a positive effect on portfolio interna-tionalization for investors in OECD countries in equities in other OECD countries as well as in non-OECD countries. However, this last effect is only present in the 2001– 2009 period. The use of the internet results in OECD investors investing a larger por-tion of their portfolio abroad, reducing the home-bias effect and therefore improving the performance of these portfolios by exploiting the benefits of portfolio diversifi-cation through internationalization. Furthermore, significant evidence is found that investors prefer to invest in countries that have a similar language. These results are similar to the current body of literature. Inconclusive evidence regarding the effect of geographical distance was found.

(32)

model was tested using extensive robustness checks. However, the model possesses imperfections that should be mentioned.

First, for the linguistic proximity between countries, a language index was used. This index is based on country data from 2000 until 2008 and is not dynamic. That is, it does not change over the years. It is reasonable to assume that the degree of specific languages spoken in a country changes over time. For example, the Education First English Proficiency Index (EF2, 2016; EF2, 2017; EF2, 2018) suggests that the average level of English spoken in countries is increasing over time. Hence, the ease of com-munication between investors and companies located in different countries may be increasing through communications in English.

Second, this paper uses the percentage of individuals between 25 and 64 years of age who have a tertiary educational degree in a specific country as a variable for education. Although the level of tertiary education likely affects whether investors can comprehend modern portfolio theory and can therefore understand and exploit the benefits of diversification through internationalization, it may be better to specify this variable as financial education or financial literacy as done in Giofré (2017).

Finally, there are stocks that are traded on stock exchanges in two different coun-tries at the same time, so-called dual-listed stocks (e.g., Royal Dutch Shell, Carnival Corporation, Dexia). This paper treats these companies as domestic firms for both countries in which the stocks are listed, even though the company might be more ac-tive in one of these two countries. Moreover, there are stocks that are listed in a country other than their country of origin or where their activities take place (e.g., Manchester United). These stocks then count as foreign investments for investors living in the country of origin of this firm or where the activities of this firm take place. This paper does not account for the discrepancy between the country where a firm is listed and where the firm is actually active.

Future research is recommended to account for these factors. Moreover, it is suggested that future research on the home-bias effect focuses on the geographical distance vari-able. Although a significant number of papers have found that this variable has a significant effect on home bias (Coval and Moskowitz, 1999; Hau, 2001; Massa and Simonov, 2006; Grinblatt and Keloharju, 2001), this paper did not find convincing ev-idence. It might be that this effect is diminishing over time or that it is currently not even present. The increasing use of the internet and technological advances together with the increase in English proficiency might reduce the information benefits of in-vestors living closer to firms.

(33)
(34)

References

(2016). Education first english proficiency index. Technical report, Eduation First. (2017). Education first english proficiency index. Technical report, Eduation First. (2018). Education first english proficiency index. Technical report, Eduation First.

Agarwal, S. and Chua, Y. H. (2020). Fintech and household finance: a review of the empirical literature. China Finance Review International.

Amadi, A. A. (2004). Equity home bias: A disappearing phenomenon? Capital Markets: Market Efficiency.

Aviat, A. and Coeurdacier, N. (2007). The geography of trade in goods and asset holdings. Journal of International Economics, 71(1):22–51.

Bekaert, G., Hoyem, K., Hu, W.-Y., and Ravina, E. (2017). Who is internationally diversified? evidence from the 401 (k) plans of 296 firms. Journal of Financial Economics, 124(1):86–112. Bekaert, G. and Wang, X. S. (2009). Home bias revisited. SSRN Working Paper Series.

Benos, E. and Jochec, M. (2013). Patriotic name bias and stock returns. Journal of Financial Markets, 16(3):550–570.

Brooks, C. (2019). Introductory econometrics for finance. Cambridge University Press, fourth edition.

Calderon, C., Chong, A., and Stein, E. (2007). Trade intensity and business cycle synchroniza-tion: Are developing countries any different? Journal of international Economics, 71(1):2–21. Campa, J. M. and Fernandes, N. (2006). Sources of gains from international portfolio

diversifi-cation. Journal of Empirical Finance, 13(4-5):417–443.

Chen, N.-f. and Zhang, F. (1997). Correlations, trades and stock returns of the pacific-basin markets. Pacific-Basin Finance Journal, 5(5):559–577.

Coeurdacier, N. and Guibaud, S. (2011). International portfolio diversification is better than you think. Journal of international money and finance, 30(2):289–308.

Cooper, I. and Kaplanis, E. (1994). Home bias in equity portfolios, inflation hedging, and international capital market equilibrium. The Review of Financial Studies, 7(1):45–60.

Referenties

GERELATEERDE DOCUMENTEN

Bij deze locatie lijkt het niet langer te gaan om samenwerking over de grens heen, maar speelt de grens geen rol meer. Zo waren er klanten en werk- nemers van beide zijden van

The literature study includes a comprehensive literature review (from books and journals) regarding the purpose, usefulness, reliability and relevance of financial

economy, and a stable economy on its turn has a positive influence on the exchange rate in general and can lead to a decrease in inflation. Both developments might have

While scholars have shown that the companies’ corporate social responsibility decisions (Chin et al., 2013), tax avoidance strategies (Christensen et al., 2015), compensation

This thesis has investigated the effects of institutions on Chinese investments in Africa and found that Chinese firms tend to invest in countries with the same level of

These fluctuations influence FDI flows and literature tends to show that an increase in the level of exchange rate (appreciation) reduces the amount of FDI received

The Netherlands, Vietnam, ASEAN, emerging market, developing countries, foreign direct investments, locational factors, Dutch multinationals.. SUPERVISORS:

Among the basic emotional states (PA(D)), we confirm that the emotion of pleasure is strongly positively correlated to bid, amplifies the purchase intention, and leads to