Global Stock Exchanges Mechanisms to Reduce Home Bias: Evidence from Euronext

Transcription

1 Global Stock Exchanges Mechanisms to Reduce Home Bias: Evidence from Euronext by Grace Pownall Goizueta Business School, Emory University Maria Vulcheva Florida International University Xue Wang Goizueta Business School, Emory University November 2011 We appreciate generous research funding from the Goizueta Business School at Emory University and comments from workshop participants at Michigan State University and Goizueta Business School Archival Brown Bag Seminar. The MSCI data contained herein are the property of MSCI Inc. (MSCI, its affiliates and any other party involved in, or related to, making or compiling any MSCI data, make no warranties with respect to any such data. The MSCI data contained herein is used under license and may not be further used, distributed or disseminated without the express written consent of MSCI.) The ownership data used in these analyses were provided on a trial basis from the Ownership Module of Thomson One Banker.

2 Abstract This paper examines whether the integration of trading platforms at cross-border stock exchange mergers can reduce equity Home Bias in the context of the formation of Euronext. We test the transaction costs hypothesis as a potential explanation for Home Bias by examining changes in foreign relative to domestic ownership for Euronext listed firms using other publicly-traded EU firms as a control to isolate the effects of contemporaneous developments in the European financial and political regime. We also test the information costs hypothesis by comparing changes in domestic and foreign ownership between Euronext firms that voluntarily precommitted to enhanced reporting and disclosure quality by joining Euronext s NextPrime and NextEconomy segments and other non-segment Euronext firms. Overall, we find no diminution of Home Bias measured as foreign holdings relative to domestic holdings for the non-segment Euronext firms, but significant increases in all categories of foreign holdings relative to domestic holdings for the Euronext segment firms. Our results indicate that the segmentation mechanism of voluntary pre-commitment to enhanced financial reporting and corporate governance has the potential to reduce Home Bias for firms listed on the integrated global capital markets. 1

3 Global Stock Exchanges Mechanisms to Reduce Home Bias: Evidence from Euronext 1. Introduction Global financial markets are consolidating across national borders, with mergers involving U.S. and European exchanges such as the merger between NYSE and Euronext, the London Stock Exchange and the Milan Stock Exchange, NASDAQ and Sweden's OMX, and the ongoing merger negotiation between NYSE Euronext and Deutsche Börse. One of the goals for global stock exchange mergers is to create an integrated trading platform that makes listed firms more available to an expanded investor clientele, thus providing larger pools of liquidity. 1 However, there is no direct empirical evidence about the effects of the cross-border integration of trading platforms on foreign investment decisions. In this paper, we study the impact of the cross-border integration of trading platforms on Home Bias. Home Bias refers to the phenomenon that investors everywhere tend to over-invest in domestic securities and under-invest in foreign securities relative to optimal global portfolio diversification, and has been observed in many contexts and time periods (French and Poterba 1991; Kang and Stulz 1997; Ahearne et al. 2004; Bradshaw, Bushee, and Miller 2004; Covrig, DeFond, and Hung 2007; Yu 2009; Leuz, Lins, and Warnock 2009). Firms benefit from reducing Home Bias and having greater access to foreign investors. For example, prior research suggests that firms with greater access to foreign capital face a lower cost of capital because of improved investor diversification (Doidge, Karolyi, and Stulz 2004; Chan, Covrig, and Ng 2006). Given 1 Cross-border stock exchange mergers may or may not result in an integrated trading platform. It is the integration of trading platforms that makes securities more available to investors from the domiciles of all of the merged exchanges. The integration of trading platforms is desirable to maximize liquidity, and was achieved shortly after the formation of Euronext but has not been achieved for some cross-border mergers, such as that between the NYSE and Euronext. 2

4 that the integration of trading platforms has the potential to reduce cross-border investment barriers, it is important to understand whether and how the integration of trading platforms reduces Home Bias. The empirical setting in which we investigate these issues is the formation in 2000 and 2002 of the Euronext stock market from the merger of the Amsterdam, Brussels, Paris, and Lisbon stock exchanges. The merger of the four stock exchanges and the subsequent integration of the trading platform across the four markets made firms from each country readily available to investors from all four countries through a single marketplace. Euronext also made a first step toward the integration of de jure reporting, disclosure, and trading regulation by issuing one Rulebook covering many aspects of reporting and trading across the four exchanges and separately codifying other reporting and trading regulations that were specific to each exchange in four secondary Rulebooks. However, the common and jurisdiction-specific regulations were enforced by each country individually, creating potential unevenness in the de facto regulation. Therefore, to achieve even more consistent regulation of at least some firms from the markets in each country, Euronext established two voluntary sections of the integrated stock market on which firms could choose to list by pre-committing themselves to enhanced financial reporting quality and corporate governance. This mechanism of voluntary pre-commitment to transparency, if credible to investors, is a solution to the problem of regulating integrated global capital markets (Grundfest 1990; Leuz 2010). The two named segments were NextEconomy (for firms from high-tech industries) and NextPrime (for firms from more traditional sectors). While there are many explanations of Home Bias, two of the most likely are the costs of accessing trading opportunities and unfamiliarity with the mechanics of transacting in foreign firms (transaction costs hypothesis); and lack of familiarity with and the costs of accessing and 3

5 interpreting information about foreign firms (information costs hypothesis). Based on the two explanations, we expect that the integration of the trading platforms across the four stock exchanges reduces Home Bias. The Euronext empirical setting also allows us to test the relative strength of the transaction costs hypothesis and the information costs hypothesis in explaining the Home Bias puzzle. If transaction costs explain Home Bias, we predict that the reduction in transaction costs through the integrated trading platform will be associated with reduced Home Bias for firms traded on Euronext after the merger. The attenuation of Home Bias should be particularly reflected in increased foreign relative to domestic ownership for companies traded on Euronext by investors from the other countries with stock exchanges involved in the merger (the Netherlands, Belgium, Portugal, and France). Alternatively, if information costs explain Home Bias, we should observe that the reduction in Home Bias for firms traded on Euronext after the merger is a function of the information quality of these firms. Pownall, Vulcheva, and Wang (2011) present evidence that firms choosing to be listed on the named segments increased their financial reporting quality and disclosure relative to their practices prior to joining the named segments and relative to firms that did not choose to join the segments. Segment firms were more likely than non-segment firms to have fully functioning websites, to use IFRS and global auditors, to use English in their primary or secondary financial statements, and to report quarterly. Pownall et al. (2011) also find that segment firms had increased trading liquidity relative to their experience prior to joining the segments and also relative to non-segment Euronext-listed firms. These empirical regularities suggest reduced costs of accessing, processing, and interpreting information about the segment firms' financial position and performance. Therefore, if information costs are a credible 4

6 explanation for Home Bias, we will observe reduced Home Bias for the segment firms, driven by increases in foreign ownership by investors from all over the world as opposed to only investors from the four countries with national exchanges involved in the Euronext merger. Further, we predict Home Bias will be reduced for the segment firms relative to the non-segment firms even after controlling for the firm-specific adoption of IFRS. Using a database that reports firm-level holdings of mutual funds from around the world, we conduct empirical tests on a sample of 21,760 firm-year observations (2,310 unique firms) in 16 European countries during the period We compare changes in foreign and domestic mutual fund ownership of Euronext-traded firms with concurrent changes in foreign and domestic ownership of other European companies to isolate the effects of the integrated trading platform from the effects of other developments in the European financial and political regime taking place concurrently. For these comparisons we use a sample of publicly-traded EU firms with data available from Worldscope and Datastream as a control group. Briefly, we find that the integration of trading platforms is associated with a reduction in Home Bias for firms listed on the named segments of the Euronext exchange, but not for the non-segment Euronext firms. We interpret this as suggesting that the reduction in trading costs from the integration of trading platforms did not make the non-segment Euronext firms more attractive to the specific investors for whom the transaction costs were reduced. While there is no diminution of Home Bias measured as the holding of other Euronext countries relative to domestic holdings for non-segment firms, we document significant increases in all categories of foreign holdings relative to domestic holdings for Euronext segment firms. We interpret this as suggesting that the decrease in information costs due to the pre-commitment to enhanced transparency made the segment firms more attractive to all categories of foreign investors, 5

7 consistent with the information costs hypothesis. Our empirical results on segment firms are robust after using propensity score matching to select a sample of non-euronext firms and a sample of non-segment Euronext firms using firm characteristics that the literature suggests are associated with attractiveness to foreign investors. Our paper makes the following contributions. First, to the best of our knowledge, our study is the first to examine the real investment consequences of the cross-border integration of trading platforms. These results are important in light of the ongoing consolidation of global stock exchanges. Integrating the trading platforms of stock exchanges from different countries has the potential to create ever larger pools of liquidity, especially in the presence of a mechanism such as the Euronext named segments that allows firms to commit to higher reporting standards than those effectively imposed by their home countries. The enhanced financial reporting and corporate governance supported by the segmentation mechanism also has the potential to reduce Home Bias and allow investors to garner more of the benefits of international diversification. Second, our paper sheds more light on the Home Bias puzzle by employing the unique empirical setting of the integration of trading platforms in Euronext, which allows us to isolate the transaction cost hypothesis from the information cost hypothesis. Collectively, our empirical evidence suggests that the reduction in Home Bias for Euronext listed firms after the integration is more consistent with the information cost hypothesis. The rest of this paper is organized as follows. Section 2 describes the merger of the four predecessor exchanges into Euronext and the formation of the two named segments, reviews the literature on the equity Home Bias puzzle, and develops hypotheses. Section 3 describes the data, 6

8 empirical design, and results. Section 4 presents diagnostics and extensions, and section 5 concludes with a summary and discussion of implications. 2. Euronext Background, Literature Review, and Hypothesis Development 2.1. Euronext Background 2 The Euronext equity market, Europe's second largest, was formed between 2000 and 2002 by the merger, demutualization, and initial public offering of the exchanges in Amsterdam, Brussels, Paris, and Lisbon. All Euronext equities listed in Amsterdam, Brussels, Paris, and Lisbon are traded through Nouveau System Cotation (NSC) and cleared by Clearnet, which allows all Euronext members, regardless of their location, to access all securities listed on Euronext. The centralized, order-driven trading system, along with the central clearing system, settles transactions on a net basis to guarantee performance (Poser 2001). Euronext and regulators in the individual jurisdictions have harmonized their rules and regulations over the years to produce two Euronext Rulebooks: Rulebook I contains harmonized rules that are contractual agreements among the market participants of Euronext, and the four versions of Rulebook II contain the remaining rules of the individual markets that have not been harmonized. The coverage of Rulebook I has gradually increased from harmonized membership, trading, and enforcement rules in the early periods to a common set of listing qualifications and disclosure requirements governing listed companies in later periods. Today, a Euronext-listed company complies with the unified listing requirements as specified in Rulebook I, and after admission, with the ongoing financial reporting requirements of its home member state. 2 Much of the descriptive material in Sections 2.1 and 2.2 was gathered from the Euronext NV Annual Reports of 2001 through For additional description of the Euronext market and its two segments, see Euronext Product Information on the NYSE Euronext website at EN.html?selected Mep=2& idinstrument=15427&isin Code=NL See also Pownall et al. (2011). 7

9 2.2. The NextEconomy and NextPrime Segments Euronext created two market segments NextEconomy and NextPrime for two reasons: (1) to meet the needs of investors seeking greater transparency and liquidity; and (2) to allow companies to increase their visibility and appeal by pre-committing to enhanced governance, disclosure, and financial reporting quality. NextEconomy and NextPrime did not replace any existing regulated market but were in addition to being listed on one of the four national Euronext exchanges. Euronext-listed companies joined the segments voluntarily by signing Commitment Agreements to comply with a series of enhanced and standardized financial reporting, investor relations, and corporate governance practices. 3 Once firms signed the Commitment Agreements, Euronext featured the segment firms on a specific section of the exchange's website including detailed corporate and financial information; initiated advertising campaigns and roadshows; and included coverage of the segments firms in investor guidebooks. At 12/31/01, NextPrime listed 107 companies and NextEconomy listed 93 companies, with market capitalization of billion (Euronext NV Annual Report 2001). Over the years the segments existed (2002 through 2007), Euronext dropped firms from the segments for cause, and added more firms to the segments at their petition. 4 Given Euronext's goal of capturing liquidity in European firms' shares, it is not surprising that few firms were actually dropped for cause. Pownall et al. (2011) compare segment firms' reporting and disclosure attributes with those of the non-segment Euronext-listed firms. They find that although compliance with the 3 The Commitment Agreements' main requirements were to: (1) publish quarterly financial reports beginning in 2004; (2) adopt international accounting standards beginning in 2004; (3) publish financial documents in English beginning in 2002; (4) schedule meetings for analysts; (5) describe corporate governance policy in the annual report; (6) schedule regular publications and meetings; (7) publish key financial data on their websites. 4 See archived weekly notices of removals and additions on the NYSE Euronext website. 8

10 segment commitments was less than complete, on average segment firms had materially higher incidence of global auditors, English language reporting, IFRS adoption, fully functioning websites, and quarterly reporting than did the non-segment firms. On October 23, 2007, Euronext announced the discontinuation of the NextEconomy and NextPrime segments, 5 because the EU Transparency Directive's requirements for enhanced transparency and disclosure for all publicly-traded firms in Europe duplicated most of the requirements of the Commitment Agreements, 6 which eliminated firms ability to voluntarily distinguish themselves by pre-committing to the higher standards Home Bias Literature Review and Hypothesis Development According to the precepts of finance theory, the optimal proportion of domestic assets in an average investor s portfolio should be equal to the share of the domestic market in global market capitalization. For example, in 2009 a US investor should have invested 31.6% of her money in US stocks, because the US capital market during that year represented 31.6% of the world stock market capitalization. 7 In reality, domestic assets make up a disproportionately large share of the average investor s asset holdings. This lack of international portfolio diversification is commonly referred to as equity Home Bias, and despite extensive research over the past few decades, it continues to puzzle economists and finance researchers (Amadi 2004). One of the main and earliest hypotheses attempting to explain Home Bias is the transaction costs hypothesis. This hypothesis attributes Home Bias to the existence of market frictions, which makes it costly for investors to acquire foreign assets. Several papers document 5 See Suppression des segments Nextprime et Nexteconomy" published by NYSE Euronext on 10/23/ See Christensen, Hail, and Leuz (2010) for a description of the Transparency Directive and the cross-country variation in its effects. 7 World Federation of Exchanges

11 empirical results in support of this hypothesis. Martin and Rey (2004) develop a model in which the demand for foreign assets changes with transaction costs in a non-linear fashion, and conclude that it is possible for a small increase in transaction costs to cause a larger increase in Home Bias. Studying implicit and explicit trading costs, Domowitz et al. (2000) find that in the period both types of costs are substantial, especially for emerging markets. Therefore, these costs have the potential to affect the returns of a globally diversified portfolio. Fidora et al. (2006) document that real exchange rate volatility is an important determinant of bond and equity Home Bias for 40 investor and 120 destination developed and developing markets they show that reducing volatility from its sample mean to zero would result in a decrease in bond Home Bias of up to 60 percentage points and equity Home Bias of up to 20 percentage points. In addition, Balli et al. (2010) find that following the introduction of the Euro in 1999, investors in the Eurozone replaced Home Bias with Eurobias or overweighting of Eurodenominated assets partly due to the elimination of exchange rate risk. Even within the Eurozone itself, investors tend to invest in markets with lower transaction costs. On the other hand, a number of papers raise issues with regard to the validity of transaction costs in explaining Home Bias. Lewis (1999) suggests that an implicit assumption of this hypothesis is that investment costs are larger than the possible diversification benefits, and estimates that in the case of international diversification, investors stand to gain between 20 and 100 percent of their life time consumption depending on their level of risk aversion. Consequently, the costs needed to divert investors from international diversification seem prohibitively high. Ahearne et al. (2004) also suggest that even when trading costs are a determinant of Home Bias, they are only of secondary importance. Further, some authors suggest that some of the market frictions cited by the transaction costs hypothesis have been reduced or 10

12 have disappeared over time (French and Poterba 1991; Cooper and Kaplanis 1994). For instance, the importance of such barriers to foreign investment as restriction of foreign exchange transactions, withholding taxes, and controls on capital repatriation, has been reduced by international tax treaties, the removal of foreign exchange controls, and the reduction of capital controls in many countries. Given the mixed evidence in the literature about the extent to which the transaction costs hypothesis can explain Home Bias, we conduct empirical analyses in the context of the Euronext merger. We hypothesize that if the transaction costs hypothesis explains Home Bias prior to the Euronext merger, there will be a reduction in Home Bias following the creation of an integrated trading platform. We also hypothesize that such attenuation of Home Bias will be particularly reflected in increased foreign ownership for companies traded on Euronext by investors from the other countries with stock exchanges involved in the merger. Therefore, we formulate the following hypotheses: H 01 : Foreign ownership did not increase relative to domestic ownership for Euronext-listed companies at the time of the merger that created Euronext, compared to other companies in the European Union. H A1: Foreign ownership increased relative to domestic ownership for Euronext-listed companies at the time of the merger that created Euronext, compared to other companies in the European Union, and this increase was more pronounced for foreign investors from the other countries with stock exchanges involved in the merger relative to other foreign investors. 11

13 The information costs hypothesis provides a main alternative explanation of Home Bias. According to this hypothesis, investors refrain from investing in foreign stocks because they are at an information disadvantage compared to domestic investors. Coval and Moskowitz (1999) show that even within a country, investors prefer geographically proximate firms because they can more easily communicate with their employees, managers, and suppliers, and may obtain information on these firms from local media. Similarly, Kang and Stulz (1997) find that foreign investors in Japan tend to opt for larger companies, with which they are more familiar, even at the cost of 0.60% increase in the volatility of their monthly returns compared to their holding the market portfolio. 8 While information accessibility is a problem in the international environment, additional issues arise for foreign investors from the need to understand and analyze such factors as differences in accounting standards, disclosure requirements, and regulatory environments across countries. For example, Ahearne et al. (2004) find that countries in which a larger percentage of firms cross-listed on US stock exchanges and complied with the stricter disclosure and reporting requirements are more heavily weighted in US investors portfolios than countries in which fewer firms did so. Bradshaw et al. (2004) find that U.S. institutional investors invest more and therefore are less biased against firms whose accounting choices are closer to US GAAP, and that the increase in investment follows the increase in conformity. There is also a growing research stream on foreign investment decisions following voluntary and mandatory adoptions of IFRS. Covrig et al. (2007) examine mutual fund 8 In addition, Nieuwerburgh and Veldkamp (2009) suggest that small initial information asymmetry is sufficient for investors to prefer acquiring information on local firms and therefore, a larger share of local assets. This, in turn, makes it more cost effective to acquire further information on local assets and less beneficial to acquire information on foreign assets. The result is a further exacerbation of information asymmetry, which the authors term information immobility. 12

14 ownership for 25,000 mutual funds in 29 countries, and find that their investment in voluntary IFRS adopters is 45% higher than that in non-adopting firms and that it increases by 35% in the year following IFRS adoptions. The authors attribute this attenuation of Home Bias to more disclosure by adopting firms. In addition, several papers document that cross-border investment increased significantly following mandatory IFRS adoptions in the European Union (see Yu 2009 and DeFond et al for mutual fund investment, and Florou and Pope 2009 for institutional investment). DeFond et al. (2011) attribute the findings to increased financial statement comparability as a result of using a uniform set of accounting standards, while Yu (2009) focuses on the effect of accounting distance (i.e. the difference between the local standards of the investor and the local standards of the investee), and suggests that accounting distance mitigates information asymmetry and explains the increase in foreign investment. Despite the overall evidence in support of the information costs hypothesis, the extent to which it can explain Home Bias is not completely clear. Beneish et al. (2009) study mandatory IFRS adoptions and find no evidence of reduction in Home Bias. Beneish and Yohn (2008) suggest that the lack of effects on Home Bias from IFRS adoptions might be due to other factors, such as behavioral biases, being more important to explain the phenomenon. To test the information costs hypothesis in the Euronext context, we compare changes in domestic and foreign ownership of Euronext-listed segment companies with similar ownership changes of other Euronext-listed firms (that is, non-segment firms). We formulate the following hypotheses: 13

15 H 02 : Foreign ownership did not increase relative to domestic ownership for segment-listed companies at the time of the merger that created Euronext, compared to other Euronextlisted companies and to other European Union companies. H A2 : Foreign ownership increased relative to domestic ownership for segment-listed companies at the time of the merger that created Euronext, compared to other Euronextlisted companies and to other European Union companies. 3. Data, Empirical Design, and Results 3.1. Data Our ownership data come from the Ownership Module in Thomson One Banker, which covers 53,000 stocks in more than 70 countries. The data include ownership levels and percentages by year according to investor types and sub-types as well as individual owners. It is reported to Thomson One Banker by investors who own shares over the minimum regulatory threshold of their domestic countries. A discussion with Thomson One representatives indicates two possible shortcomings of the database. First, the information is self-reported by investors, thus certain inaccuracies exist because investors misreport. Second, it is collected at different points in time during the year. Shares that change hands might be double-counted if reported first by the seller and later on by the buyer. We conduct our regression analyses using only mutual fund ownership because mutual fund ownership data are less likely subject to these shortcomings than the overall data provided by various investors. The focus on the mutual fund ownership data 14

16 is also consistent with prior literature (see, DeFond et al and Yu 2009, for example). The Ownership module covers more than 34,000 global mutual funds. 9 Panel A of table 1 describes our sample selection. We start with an initial sample of 6,081 firms (46,141 firm-years) from the European Union with non-missing mutual fund ownership data for the period Of these firms, 1,012 are from the four Euronext countries (and the majority of those are from France), and 5,069 are from the rest of the European Union. We require that total ownership not exceed 100%, resulting in the loss of 18 firms and 1,939 firm-years. 10 We exclude 406 firms (5,037 firm-years) due to zero domestic mutual fund ownership, 283 firms (2,051 firm-years) due to missing Worldscope data, and 451 firms (5,541 firm-years) for missing control variables. Further, we lose 61 firms (1,141 firmyears) when we exclude the top and bottom half percent of each continuous control variable distribution. Finally, we require a balanced sample with firms appearing both prior and subsequent to the formation of Euronext, which excludes 2,552 firms (8,672 firm-years), leaving a sample of 2,310 firms with 21,760 firm-years. [Insert Table 1 About Here] Panel B of table 1 disaggregates the primary sample into Euronext country firms and other EU firms. Approximately one quarter of the sample are firms from the four Euronext countries, including 169 (376) segment (non-segment) firms. The majority of both segment and non-segment firms are from France (98 segment firms and 276 non-segment firms). In contrast, 9 _factsheet.pdf. 10 Despite the shortcomings of the Thomson One ownership data, we use them as a proxy for true ownership with the assumption that there is no bias between the segment firms and the non-segment firms and between Euronext listed firms and non-euronext listed firms that would shift the percentage of foreign relative to domestic ownership. Further, assuming that errors and omissions in ownership data occur randomly between foreign owners and domestic owners or are stable over time, we control for changes in domestic ownership when evaluating changes in foreign ownership. 15

17 Portugal contributes only eight segment and 26 non-segment firms. The other EU firms come from 12 countries, with the majority (787 firms) coming from the London Stock Exchange. The second largest concentration of other EU firms is from Germany (400 firms), and only two (one) firms come from Denmark (Luxembourg). Table 2 presents summary statistics for the mutual fund ownership data, broken down in panel A by Euronext vs. other EU firms and by pre- vs. post-merger time periods. There is a statistically significant increase in mutual fund holdings for both Euronext and other EU firms. In addition, all mutual fund ownership categories increase by a statistically significant amount from the pre- to the post-merger period for the Euronext firms. For the other EU firms, although the undeflated values of all ownership categories increase significantly at the mean and median, when these values are deflated by domestic ownership only mutual fund ownership from Euronext countries increases significantly at the mean and median, and all foreign and European mutual fund ownership increase significantly at the median. These observations, especially for the foreign ownership categories standardized by domestic ownership to control for secular trends in mutual fund percentage ownership, are strongly consistent with diminishing Home Bias for Euronext firms from before to after the merger. Panel B presents the same statistics broken down by segment vs. non-segment firms and pre- vs. post-merger time periods. Total mutual fund ownership increases from pre- to post-merger for the segment and the non-segment firms. Domestic mutual fund ownership increases for the non-segment firms, but it exhibits little change for the segment firms. While all categories of foreign ownership increase significantly (at the mean and median) for the segment firms, only total foreign and EU ownership increase significantly at the mean and median for non-segment firms when foreign ownership is deflated by domestic ownership. 16

18 [Insert Table 2 About Here] 3.2. Empirical Design Our basic empirical design is to regress foreign relative to domestic mutual fund ownership measures 11 for EU firms on indicator variables for (1) the pre- vs. post-merger periods (to determine whether the integration of the trading platforms was associated with an overall change in Home Bias); (2) the segment firms vs. the non-segment firms (to determine whether choosing to become listed on the named segments was associated with a different level of Home Bias), and (3) interactions between the post-merger period and Euronext non-segment firms (to determine whether the change in Home Bias from before to after the integration was significant for the non-segment firms) and between the post-merger period and the segment firms (to determine whether the change in Home Bias from before to after the integration was significant for the segment firms). Our basic regression model is: Log(1 + MF_for/dom ijt ) = + 1Post + 2Segment + 3NonSegment + 4Post*Segment + 5Post* NonSegment + kcontrols) + it (1) where: MF_for/dom j = relative foreign to domestic mutual fund ownership, where foreign mutual fund ownership is measured as total foreign ownership when j=1, non-euronext EU ownership when j=2, Belgian, Dutch, French, and Portuguese ownership but excluding the firm's home country owners when j=3, and UK ownership when j=4; and domestic ownership is measured as total domestic mutual fund ownership; Post = an indicator variable equal to 1 for 2002 (for the other EU firms) and the year in which the firm s exchange joined Euronext (for the Euronext firms), and all years thereafter; Including, for the Euronext-listed firms, Belgian, Dutch, French, and Portuguese investors relative to domestic investors. 12 This year is 2002 for Amsterdam, Brussels, and Paris, and 2004 for Lisbon. 17

19 Segment = an indicator variable equal to 1 for companies included in the NextPrime or NextEconomy segments during any year of the sample period; NonSegment = an indicator variable equal to 1 for companies listed on Euronext but not included in the NextPrime or NextEconomy segments during any year of the sample period; and firm-specific variables are included to control for each firm's appeal to mutual fund investors, along with industry, exchange, and year fixed effects. The variables included in the regressions (including the control variables) are summarized in the appendix. As the appendix shows, we include leverage, profitability, the number of foreign exchanges on which the firm is traded, size, Big-5 auditor, conformity with US GAAP or IFRS, cross-listing in the US, stock return volatility, liquidity (measured as the number of trading days with zero returns, following Ashbaugh et al. 2005), being included in the Morgan Stanley Capital International (MSCI) Index in 2007, 13 dividend yield, earnings-price ratio, analyst following, number of stock indices in which the company is included, and ownership concentration. 14 Table 3, panel A gives summary statistics for the control variables for the full sample. The table shows that sample firms are quite heterogeneous on most dimensions, thus our empirical design contains a number of controls. By splitting the time period into pre- and postmerger periods, we are using each firm as its own control. By comparing the segment firms and 13 We use the MSCI World Index as a proxy for the company being included in a major market index. It is a freefloat-adjusted market cap weighted index that combines 24 developed and 21 developing country indices, and therefore measures the performance of a broad range of developed and developing economies ( Since the information that MSCI Inc. provided was for the composition of the index in the period , and therefore failed to cover the whole sample period, we do not use the year-specific constituents of the index, but instead rely on the 2007 index composition only, given 2007 is the middle point of the data we have available (see also DeFond et al for a similar approach). 14 These controls are taken from Florou and Pope (2009); Bradshaw et al. (2004), Covrig et al. (2007), Hamberg, Mavruk, Sjogren (2009), and Yu (2009), among others. In untabulated diagnostic analyses, we also included free float (measured as the percentage of closely held shares) and being traded as an ADR either OTC or on a national exchange in the US. Including these variables used in the literature did not change our inferences, and severely reduced our sample size due to lack of free float data. 18

20 non-segment firms, we are controlling for other financial and political factors in the four countries that may have had an impact on mutual fund ownership patterns. Finally, by including a large sample of other European Union firms in our regressions, we are controlling for financial, political, and cultural shifts in Europe which may have had an impact on ownership patterns. [Insert Table 3 About Here] Panel B of table 3 compares the Euronext and other EU firm subsamples. Before the merger, Euronext firms were significantly more leveraged, more profitable, larger, more liquid, with more volatile returns, more analyst following, and more concentrated ownership (all differences significant at the mean and median), relative to other EU firms. After the merger, Euronext firms are still significantly more leveraged, more profitable, and larger, with more analyst following and more concentrated ownership (all differences significant at the mean and median), but the significance of the differences in return volatility and liquidity is attenuated. Panel C compares the segment and non-segment firms. Before the merger, firms that eventually signed Commitment Agreements and became listed on the segments were less leveraged, smaller, and more liquid (all differences significant at the mean and median) than the non-segment firms. After the merger, segment firms were still less leveraged, smaller, and more liquid, and also had less concentrated ownership than did non-segment firms Regression Results Table 4 shows the results of estimations of eq. (1) using four categories of foreign mutual fund ownership (standardized by domestic mutual fund ownership to control for secular trends in the data) for the full sample of European Union firms. Column 1 uses the log of (1 + total foreign mutual fund ownership divided by domestic mutual fund ownership) as the dependent 19

21 variable, column 2 uses other EU mutual fund ownership as the measure of foreign ownership, column 3 uses only foreign mutual fund owners domiciled in Euronext countries other than the sample firm s domicile as the measure of foreign ownership, and column 4 uses UK mutual fund ownership as the measure of foreign ownership to capture the idea that mutual fund managers in a major European financial center may have different patterns of buying and selling foreign securities than mutual fund managers not so centrally located. The coefficient on Post is positive and significant when foreign ownership is measured as total foreign ownership, suggesting an overall decrease in Home Bias from the pre- to the postperiod, but not because European mutual funds increased their investment in EU firms more than domestic mutual funds did. The coefficients on both Segment and Non-segment are positive and significant except when foreign ownership comes from the other Euronext jurisdictions, consistent with less Home Bias on average for the Euronext firms outside the Euronext countries than for the other EU firms. The coefficient on the interaction between Post and Segment is always positive and significant, indicating that the decrease in Home Bias for the segment firms relative to other EU firms is more pronounced no matter how foreign ownership is measured. Finally, the coefficient on the interaction between Post and Non-segment is not significant in any of the four specifications, suggesting that the integration of the trading platforms after the merger did not have significant effect on Home Bias for the non-segment firms. Tests of hypothesis 1 (that transaction costs fall more significantly for investors from the Euronext countries at the merger, which should be associated with a diminution of Home Bias if Home Bias is driven by transaction costs) in their most straightforward form focus on the column 3 results (other Euronext countries' ownership only), specifically the interactions between Post and Segment or Non-segment. These results support the alternate hypothesis for the segment firms, but fail to 20

22 reject the null for the non-segment firms. We conclude that there is only limited evidence consistent with the transaction costs hypothesis. [Insert Table 4 About Here] Tests of hypothesis 2 (that the enhancements in accounting, disclosure, transparency, and corporate governance pledged by the segment firms reduced perceived information costs to foreign investors, and were associated with reductions in Home Bias) focus on the coefficients in all four columns on the interaction between Post and Segment. The coefficient is positive and significant in all columns, offering strong support to the information costs hypothesis. Among the control variables, leverage and being included in more stock indices are positively associated with Home Bias; and size, return volatility, and analyst following are negatively associated with Home Bias; but none of the other control variables are consistently significant. The R 2 s of the regressions range from 20% to 34%. Because our hypotheses are framed as tests on changes in Home Bias for Euronext firms at the time of the merger, including other EU firms as controls for European events and forces may be redundant to including all Euronext firms. Therefore, we present results for only the Euronext firms in table 5. We are unable to estimate the coefficient for NonSegment and for the interaction between Post and NonSegment when we include only Euronext firms in the sample. The coefficient on Post is insignificant, indicating no general reduction in Home Bias for Euronext firms from pre- to post-merger. The coefficient on Segment is negative and significant when foreign ownership is either all foreign ownership or other EU ownership, meaning that in general other European and all foreign mutual funds were less likely to hold segment firms equities. However, the interaction between Post and Segment is positive and significant using all four proxies for foreign mutual fund ownership, indicating a bigger reduction in home bias for 21

23 the segment firms from pre to post-merger. This result suggests that the ability to join the segment and credibly pre-commit to enhanced disclosure and transparency decreased information costs for foreign investors (including European investors) and was associated with increases in their ownership of segment firms but not of non-segment firms. This result is consistent with the information costs hypothesis but not with the transaction costs hypothesis. [Insert Table 5 About Here] 4. Diagnostics and Extensions We perform several replications of our primary analyses to assess the sensitivity of our results to including various combinations of control variables. Specifically, we rerun the regressions in equation (1) including only the test variables and no control variables, and including the Closelyheldsharespct variable, which results in the loss of almost 4,000 observations. The results of these re-estimations support the same inferences as those reported in table 4 and table 5. We also estimate equation (1) comparing results for the segment firms to a propensity score matched sample of non-segment firms, and separately to a propensity score matched sample of other EU firms. Our descriptive statistics in table 3 show that segment firms differ significantly from non-segment firms and other EU firms on a number of firm characteristics both before and after the formation of Euronext. Among these characteristics are leverage, profitability, number of foreign exchange listings, size, Big-5 auditor, IFRS or US GAAP, US cross-listing, return volatility, and trading liquidity. The accounting literature has demonstrated that these characteristics affect not only the reporting decisions of firms but also their decisions to list or cross-list on a highly-regulated exchange (see Lang et al for an example). Thus, 22

24 the investment decisions of foreign mutual funds might not be independent of firms' decisions to list on a named segment and our results might be influenced by selection bias. To control for such bias, we match the segment and non-segment firms (other EU firms) on these firm characteristics to obtain a sample of non-segment firms (other EU firms) with similar likelihood of listing on a named segment. Given the large number of independent variables which need to be matched, we expect that a simple matching procedure will result in a large loss of observations. Instead, to minimize this loss, we choose to conduct a propensity score matching (PSM) analysis. The PSM analysis allows us to calculate a propensity score for each observation in our sample, as a single balanced representation of all firm characteristics of interest (Guo and Fraser, 2010). We conduct the PSM analysis separately pre- and post-merger. We first present the results from the PSM sample of segment firms and the matched other EU firms in Table 6. Panel A reports two logistic regressions used to calculate the propensity scores for the sample of segment firms and the sample of other EU firms. The results of these regressions indicate that both before and after the formation of Euronext, the probability of joining a named segment is positively and significantly related to leverage, profitability, trading liquidity, analyst following, and ownership concentration. The probability of joining a named segment is also negatively and significantly related to the number of stock exchanges and the number of stock indices. Using the calculated propensity scores from these logistic regressions, we perform a nearest neighbor within caliper matching of segment and non-segment firms. We keep in our PSM sample only pairs of segment and other EU firm-years for which the absolute value of the difference in their propensity scores is no larger than In the process of matching we lose segment and other 23

25 EU firm observations that are not in the common support region of the calculated propensity scores, leaving 1,141 firms and 3,564 firm-years (see panel A of table 1). [Insert Table 6 About Here] To evaluate whether our PSM procedure was successful in eliminating the differences in firm characteristics between the segment and other EU firms, we recalculate the descriptive statistics by period, the difference between means and medians, and the statistical significance of these differences. The comparison in panel B, table 6 indicates that the PSM procedure successfully eliminated the difference in means and medians between the segment and non-eu firms for most firm characteristics. The firms in our PSM sample pre-merger differ significantly only on mean and median ownership concentration and closely held shares percentage and on median analyst coverage, and post-merger on mean and median closely held shares percentage and on median analyst following and ownership concentration. 15 Panel C presents results of the PSM Home Bias analyses, and shows inferences very similar to those on table 4. The coefficient on the interaction term for segment firms after the formation of Euronext is positive and significant for all three columns, indicating that the diminution of Home Bias for the segment firms was higher than that for other EU firms using any of the proxies for Home Bias. Similarly, we form another PSM sample of segment firms and non-segment Euronext firms following the same method. Panel A, table 7 reports two logistic regressions used to calculate the propensity scores. The results of these regressions indicate that both before and 15 We did not include the closely held shares variable in the logit regression to calculate the PSM scores as shown in panel a, table 6 (table 7), because the requirement of non-missing closely held shares severely reduces the sample size. So there are significant differences in closely held shares between segment firms and other EU firms (nonsegment firms) in the PSM sample. To increase our confidence in the inferences drawn from the PSM sample, we repeat the regressions by including control variables including closely held shares, and we obtain similar results as those in panel c, table 6 (table 7). 24

26 after the formation of Euronext, the probability of a Euronext firm joining a named segment is positively and significantly related to being audited by a Big-5 auditor and trading liquidity. The probability of a Euronext firm joining a named segment is also negatively and significantly related to the number of stock exchanges and firm size. We again keep in our PSM sample only pairs of segment and non-segment firm-years for which the absolute value of the difference in their propensity scores is no larger than After this requirement, we are left with 461 firms and 2,882 firm-years (see panel A of table 1). [Insert Table 7 About Here] Panel B, table 7 reports the descriptive statistics of the PSM sample by period, the difference between means and medians, and the statistical significance of these differences. The comparison indicates that the PSM procedure successfully eliminated the difference in means and medians between the segment and non-segment firms for most firm characteristics. Finally, panel C presents results of the PSM Home Bias analyses, and shows inferences very similar to those on table 5. The coefficient on the interaction term for segment firms after the formation of Euronext is positive and significant for columns (1), (2), and (4), indicating that the diminution of Home Bias for the segment firms was higher than that for non-segment firms using most proxies for Home Bias. 5. Summary and Conclusions In this paper, we examined whether and how the integration of trading platforms at crossborder stock exchange mergers can reduce equity Home Bias the phenomenon that investors everywhere tend to over-invest in domestic securities and under-invest in foreign securities relative to optimal global portfolio diversification. Our research setting was the formation in 25