On the Determinants of International Equity Investment

Transcription

1 On the Determinants of International Equity Investment Sara B. Holland Terry College of Business, University of Georgia Sergei Sarkissian Faculty of Management, McGill University Michael Schill Darden Business School, University of Virginia Francis E. Warnock* Darden Business School, University of Virginia National Bureau of Economic Research Abstract Previous work suggests that any study of international equity portfolio allocations must have two features: the dependent variable must be float adjusted and free of a country-size bias. A third requisite feature comes from U.S.-centric studies: analysis of U.S. international equity investment must properly account for cross-listing. We first summarize evidence supporting the importance of these three features, focusing on evidence amassed using data from one source country, the United States. While a presentation of the existing evidence is useful in its own right, our main innovation is to apply these features to the global matrix of cross-border equity positions. Preliminary evidence from this global analysis suggests cross-listing is an important omitted variable in previous studies that include high standards countries. * This work is extremely preliminary, so please do not cite or post without the authors permission. The authors thank for helpful comments participants in seminars at University of Auckland, ECB, Federal Reserve Board, Notre Dame and Vanderbilt.

2 1. Introduction In both the finance and economics literatures, the past decade has witnessed a plethora of empirical studies on the determinants of international equity investment. In the economics literature, most such studies use a gravity-type model with some sort of barriers or frictions; the use of gravity models in studies of international portfolio allocation is described and critiqued (from a theoretical perspective) in Okawa and van Wincoop (2012). In the finance literature, some empirical studies on international equity investment are also based on gravity models, but many others are derived at least loosely from an international capital asset pricing model (CAPM) with barriers to international investment, such as that in Cooper and Kaplanis (1986). Findings both within and across these literatures have varied greatly. Andrade and Chhaochharia (2010) show that foreign direct investment (FDI), viewed as a proxy for marginal differences in the endowments of information about different countries, contributes to differences in U.S. investors international equity portfolios, consistent with the van Nieuwerburgh and Veldkamp (2009) equilibrium model of learning and portfolio choice under an information processing constraint. U.S. international equity investment has also been shown to be largely in countries with relatively favorable tax policies on dividend income earned by foreign investors (Amiram and Frank 2010) or that adopt IFRS (Yu 2010), and to tend to be reallocated toward countries that are tax-favored in U.S. laws (Desai and Dharmapala 2011). In papers that are less U.S. centric, international equity investment has been shown to be greater in countries with more bilateral trade, with a 10% increase in bilateral trade raising bilateral asset holdings by 6% to 7% (Aviat and Coeurdacier 2007); between countries with lower real exchange rate (RER) volatility, with a country pair with zero RER volatility having 20 percentage points lower equity home bias than a country pair with average RER volatility 2

3 (Fidora et al 2007); with markets that are large and have lower trading costs, a common language, and that are close geographically (Chan, Covrig, and Ng 2005); in country pairs with more bilateral trade, a common language, and common legal origins (Lane and Milesi-Ferretti 2008); in country pairs that have lower stock market return correlation (Coeurdacier and Guibaud 2011); and in countries with greater financial market development, capital market openness, information (measured by internet users), and familiarity (proxied by geographical closeness) (Bekaert, Siegel, and Wang 2012). Without meaning to suggest that any particular above-listed paper is flawed, it is somewhat difficult to discern what the determinants of international equity investment are because most papers fail in their empirical design along at least one of three dimensions. In the U.S. setting, compelling evidence and arguments point to three design features that must be present in any empirical study of U.S. international equity investment: the dependent variable must be float adjusted, the dependent variable should be free of a country-size bias, and the analysis must properly account for cross-listing. The first and second of these extend readily to the global n source countries by m destination countries case. The third is clearly important for U.S. international equity investment Ammer et al (2012) show that cross-listing is the single most important determinant of the amount of U.S. investment a foreign firm receives but it is less apparent that it should be vital when analyzing a global matrix of cross-border holdings. It is conceivable, for example, that cross-listing is special only in the U.S. case, but can be ignored for other cross-listing destinations. In this paper we lay out the case that these three factors a dependent variable that is both float-adjusted and free of a size bias, and cross-listing entering the empirical design in an appropriate way must be included in any study of the determinants of international equity 3

4 investment. To some extent existing work has emphasized these points separately. For example, that the dependent variable must be float adjusted is well established in Dahlquist, Pinkowitz, Stulz, and Williamson (2003) and Kho, Stulz, and Warnock (2009), in part because shares held by insiders are not available to dispersed portfolio shareholders and so should be omitted from portfolio analysis. Bekaert, Siegel and Wang (2012) argues convincingly that the dependent variable should be free of a country-size bias. Ammer et al (2012) show that cross-listing must be included in any U.S.-based study. But no one has yet established whether and how cross-listing should be included in a global n x m study of international portfolio allocation. A primary contribution of this paper is to do just that. Our main finding is that cross-listing should indeed be included in global n by m studies on international equity investment, but in a way that is subtle for two reasons. First, the determinants of cross-border investment differ at the tails of investment levels. Country pairs for which there is little or no bilateral investment and those that have so much bilateral investment that they are overweight global benchmark weights are different in ways we detail later. Second, for the sample of country pairs with meaningful (but not extreme) investment levels, our results suggest that cross-listing on a foreign exchange that has low standards (defined in various ways) is in an empirical sense irrelevant, consistent with the evidence in Sarkissian and Schill (2004) that cross-listing reflects rather than causes investment. But, consistent with Sarkissian and Schill (2009), a cross-listing in a high standards country is very different: Instrumented cross-listing is a significant determinant of the extent of holdings in high standards countries, and omitting it could change some inferences of past studies. The paper proceeds as follows. The next section summarizes reasons the dependent variable should be float adjusted and free of any size bias. Section 3 discusses existing evidence 4

5 on where and why firms cross-list and cross-listing s impact on U.S. international equity investment. Section 4 assesses the proper role of cross-listing in studies of a global matrix of crossborder international equity holdings. Section 5 concludes. 2. The Dependent Variable 2.1 What Theory Suggests and Empirics Allow Gravity Gravity in empirical international portfolio allocation has its roots and theoretical underpinnings in the economics literature on international trade in goods, in which bilateral trade in goods is a function of country sizes, bilateral trade barriers, and multilateral resistance variables; see in particular Anderson and van Wincoop (2003) for the theoretical justification of using gravity in empirical trade models. 1 Okawa and van Wincoop (2012) point out three ways to derive a theoretical gravity equation for international asset holdings. One, suggested by Lane and Milesi-Ferretti (2008), is a multi-country extension of the model in Obstfeld and Rogoff (2000) that relates barriers in goods trade to portfolio home bias. As Okawa and van Wincoop note, this approach to bringing international asset positions into a gravity model is theoretically possible but has at least one important drawback: the real exchange rate hedge channel, through which barriers in goods trade affect asset trade in Obstfeld and Rogoff (2000), does not appear to be operative in practice. Using data on equity returns and real exchange rates, van Wincoop and Warnock (2010) show that hedging real exchange rate risk cannot account for portfolio home bias. Consistent with the van Wincoop and Warnock findings, Coeurdacier (2009) develops an extension of Obstfeld and Rogoff 1 Multilateral resistance variables in the trade literature (see Anderson and van Wincoop 2003) are based not only on the bilateral trade barrier between i and j but also i's and j s barriers with all other countries. For example, if j s barriers with other countries are higher than its barriers with i, all else equal it will import more from i. 5

6 (2000) to show that for realistic model parameters trade barriers cannot generate a portfolio home bias. This method for bringing gravity to international trade in assets seems, at least currently, to be a dead end. A second way Okawa and van Wincoop (2012) highlight to derive a theoretical gravity equation for international asset holdings is the Martin and Rey (2004) derivation of a gravity equation for financial holdings when countries trade claims on Arrow Debreu (AD) securities. Coeurdacier and Martin (2009) extend Martin and Rey to show that this can lead to a gravity equation that is similar to that for goods trade, with bilateral holdings depending both on bilateral frictions and multilateral resistance indices of source and destination countries. The key insight here is that under CES preferences demand for AD securities takes a similar form as the demand for goods, and empirical international finance studies can proceed along the well-worn paths of the empirical trade literature. Instead of differentiating goods by type, as in the trade literature, differentiating assets by states in which they have a payoff enables a standard constant relative risk-aversion expected utility to be written as a function of AD asset holdings in a way that is analogous to CES utility as a function of consumption of differentiated goods. As Okawa and van Wincoop note, however, the main limitation of this approach is that while any security can be written as a combination of AD securities, such a basket would have securities from many countries. In contrast, empirical work on international equity positions (or bond or bank holdings) is based on a clear separation between the countries of the investor and the security issuer. Okawa and van Wincoop (2012) focus on a third way to use gravity to derive empirical models of international portfolio investment, but show that this way is not robust to some features observed in the real world such as direct impediments (e.g., taxes) on international investments. Overall their verdict is clear: presently there is no justification for many of the existing 6

7 empirical gravity specifications (of international asset holdings) But they also soften this conclusion, noting that they cannot prove that many specifications of gravity in international finance have no theoretical foundation, just that currently there is no theory justifying such specifications. Needless to say (but they do), it is best for empirical work to be consistent with existing theory International CAPM An international CAPM-based model of international portfolio allocation is presented in Cooper and Kaplanis (1986), among others. The model includes country-specific proportional investment costs, representing both explicit and implicit costs of investing abroad. 2 Under standard international CAPM assumptions, the i th investor s optimization problem is to choose xi, the allocation of her wealth among risky securities in n countries, to maximize expected returns net of costs, or: max (x i R x i c i ) subject to x i Vx i = v and x i I = 1 where xi is a column vector, the nth element of which, xin, is the proportion of individual i s wealth invested in securities in country n R ci is a column vector of pre-cost expected returns is a column vector, the nth element of which, cin, is the cost to investor i of holding securities in country n v is a constant 2 The seminal papers modeling the effects of barriers to international investment are Black (1974) and Stulz (1981). 7

8 V I is the variance/covariance matrix of the gross (pre-cost) returns of the securities is a unity column vector For simplicity, assume that the covariance matrix, V, is diagonal with all variances equal to s 2. Impose the world capital market clearing condition, W i x i = M, where Wi is the proportion of world wealth owned by country i and M is a column vector, the ith element of which, Mi, is the proportion of world market capitalization in country i s market. Then the solution to this problem simplifies to 3 hs 2 (x in M n ) = c in + b n + a i d (1) where a i = z c i (weighted average marginal cost for investor i) b n = M j c jn (world weighted average cost in country n) d = z M i c i (world weighted average cost) and z = V 1 I/(I V 1 I) (global minimum variance portfolio) In the case with no costs to investing, ci is a zero vector and the right-hand side of (1) is zero. Hence xin=mn; that is, investor i allocates her wealth across countries according to market capitalizations. In the more general case with non-zero and non-uniform costs, a logic similar to that of multilateral resistance in the gravity theory holds: if the actual cost to investor i of investing in country n (cin) is high relative to investor i s average cost to investing (ai) or relative to all investors costs to investing in country n (bn), then the right-hand side of (1) is likely negative and investor i will underweight country n in her portfolio. The higher are costs in a particular foreign market, the more severely underweighted that country will be in the investor s portfolios. 3 Here h is the Lagrange multiplier on the constraint x i Vx i = v. 8

9 Moreover, since investors do not face such costs in their home market, equation (1) predicts an overweighting of domestic stocks the equity home bias when costs exist in other countries. Thus, the international CAPM yields a dependent variable that might be obtainable to the empiricist: the proportion of the investor s financial wealth allocated to country i s assets. But in practice, some hand-waving is required to get from theory to a readily constructible dependent variable. Country i s assets becomes streamlined to just one asset class (for example, country i s equities, bonds, or bank assets). And measures of financial wealth are not as easily constructed as one might think, so the focus is almost always on allocations within the investor s portfolio in a single asset class (e.g., within the foreign equity portfolio), rather than how wealth is allocated across and within a broader set of asset classes A Measure Without a Size Bias The above discussion suggests that gravity, used by many to motivate international equity analysis, is not readily applicable to such use. The international CAPM is more appropriate, although it still requires some hand-waving to get to a readily constructible dependent variable. The dependent variable that arises from the international CAPM is essentially a portfolio allocation: the share of the investors portfolio that is allocated to asset i. When controlling for Mi (i s share in world market capitalization), coefficients on other variables should indicate whether various costs are associated with deviations from the international CAPM allocations. However, empirically there is a problem with a dependent variable based solely on portfolio shares, a problem explored in detail in Bekaert, Siegel and Wang (2012) and also briefly discussed in Ammer et al (2012). Portfolio share is biased by size in a way that can bias inference on the explanatory variables of interest. That portfolio share is strongly related to country size is readily apparent in the graph in Panel A of Figure 1: the larger the country, the greater is U.S. 9

10 investment in its equities. Bekaert et al show that including a size variable to control for this association between size and investment in no way solves the problem as inference on other variables of interest typically the fundamental point of any one study is muddied in ways that are not easy to predict. Some studies have attempted to remedy this size bias problem by analyzing, consistent with Bekaert et al (2012), deviations from the international CAPM benchmark rather than portfolio shares. Bekaert, Siegel and Wang (2012) suggest a measure they label as a foreign bias of country j vis-à-vis country i (FBj,i) that is similar but not identical to the following: FB FB MC /MC - H /H i world j, i j, world j, i, if H j, i/h j, world MC i/mc world MC i/mc world H /H MC /MC j, i j, world i world j, i, if H j, i/h j, world MC i/mc world 1- MC i/mc world (2) where MC is market capitalization and Hj,i is j s holdings of i s equities, both in U.S. dollars. The FB measure is a version of a home bias measure that, when the actual portfolio weight is less than the benchmark weight, varies monotonically with the ratio of actual to benchmark weights. In an overweight situation in which the actual weight exceeds the benchmark weight, the FB measure is a muted linear function of the ratio. Our preferred measure which, like FB, is both suggested by theory (international CAPM) and free of any country-size bias is observationally equivalent to FB measure for underweight positions but does not smooth overweight measures. We define relative portfolio weight, or RW, as follows: H /H j, i j, world RW j, i (3) MC i/mc world 10

11 The relative weight of country j in country i s portfolio is just the ratio of its weight in country i s portfolio to its weight in the world portfolio. When i holds no j equities, RWj,i is zero. If the weight of j in i s portfolio is identical to j s weight in the world portfolio, RWj,i equals one. When RWj,i is greater than one, j is overweight in i s portfolio; that is, i s holdings of j equities exceed the global benchmark weight. The below chart is drawn for a benchmark weight of 0.2, so that there is no home bias (i.e., RW=1 and FB=0) at an actual portfolio weight of 0.2. For RW less than or equal to one (i.e., for all observations that are underweight or at the benchmark weight), RW and FB are perfectly negatively correlated. The only difference between the two measures is for overweight observations (actual weight greater than 0.2 in the chart), where RW does not smooth and FB does. In practice, the choice of FB or RW makes little difference empirically. As quick aside on size bias for difference measures, the reader might readily accept that holdings expressed in dollars (or log dollars) or as a share of the U.S. portfolio has a country size 11

12 bias, but might expect a difference bias measure (essentially the numerator from equation 2) to be free of such bias. 4 However, the difference measure also has a substantial country size bias (Figure 1 Panel B), whereas a ratio relative weight measure (equation 3) for U.S. investment does not (Figure 1 Panel C). Extending to a fuller global matrix, the portfolio share and difference measures also exhibit size bias (Figure 2 Panels A and B), but there is no relationship between the relative weight measure and size (Figure 2 Panel C). 2.2 Scaled by float or market capitalization? In the above discussion, for expository ease it was assumed that the entire world market capitalization is available to dispersed portfolio investors. But Dahlquist et al (2003) argue that shares held by insiders are not available to dispersed portfolio shareholders and so should be omitted from portfolio analysis. Kho et al (2009) further develop this point in their optimal corporate ownership theory of the home bias in which foreign portfolio investors exhibit a large home bias against countries with poor governance because their investment is limited by high optimal ownership by insiders (the direct effect of poor governance) and domestic monitoring shareholders (the indirect effect ) in response to poor governance. 5 That the dependent variable in empirical studies of international equity investment must be float adjusted is well established in Dahlquist, Pinkowitz, Stulz, and Williamson (2003) and Kho, Stulz, and Warnock (2009). Accounting for this issue changes Eq. (3) to a float-based relative weight measure: 4 Note that the Bekaert et al measures are also free of a size bias. Also, for cross-sectional studies (i.e., studies limited to one source country and one time period) note that at a point in time the top line in equation (2) is a reasonable measure that is essentially equivalent to j s holdings as a percent of the destination country s market cap (times a measure that is constant across countries at a point in time, the share of the j in the world portfolio). 5 Kho et al s theory has a second component: Foreign direct investors from good governance countries have a comparative advantage as insider monitors in poor governance countries, so that the relative importance of foreign direct investment is negatively related to the quality of governance. 12

13 H /H j, i j, world RW j, i (4) F i/fworld where Fi is the float (market capitalization less insider holdings) in country i, Fworld is world float, and Hj,world is calculated using float-adjusted holdings. While there is no exact measure of insider ownership that is available both across a range of countries and through time, a country-level measure built up from the firm-level closely-held field in Worldscope is a reasonable proxy. Note that we are not claiming that the Worldscope closely-held field is perfect since it is a measure based on numerical cutoffs that cannot truly discern who has controlling interest and who does not. Moreover, reporting requirements and their enforcement varies across the world, and the coverage of Worldscope has changed over time. 6 That said, the Worldscope closely held variable is available over time and for a large number of countries, and potentially includes insiders who are not controlling shareholders but might be part of the controlling coalition The Importance of Cross-listing Often (but not always) independent of the international portfolio allocation literature, the literature on cross-listing has burgeoned over the past decade. For U.S. cross-listings that is, non-u.s. firms that list on a U.S. exchange Karolyi (1998) and Miller (1999) can be considered seminal papers that detail ADR programs, the instrument through which most foreign firms cross-list in the U.S. See also Stulz (1999). For discussions of cross-listings in many markets, see Sarkissian and Schill (2009) and cites therein. 6 To mitigate the effect of outliers, we use a smoothed measure of closely held shares. In particular, if the change in the market value weighted aggregate country-level closely held measure is greater than 1.5 standard deviations, then country-level closely held is set to the average of the previous year, the current year, and the next year. 7 Some studies painstakingly gather information on the holdings by controlling shareholders (e.g., Claessens, Djankov, and Lang 2000, Faccio and Lang 2002, and Lins 2003). Theirs should be truer measures of insider holdings, but such datasets are typically as of a point in time. 13

14 3.1 Where Do Firms Cross-List? As Sarkissian and Schill (2004) note, in an integrated and frictionless world capital market a firm should have no preference for the markets on which it lists its shares, since this choice has no effect on its pool of investors. But we do not live in that world. Our world has frictions, sometimes substantial, and not all markets are fully integrated; in such a world firms might want to choose where to cross-list. In one sense the firm s listing decision collapses to one question: Can the overseas listing help to overcome portfolio investors home bias? If yes, the firm might choose a foreign listing location that would help overcome the home bias of an important set of foreign investors (Merton, 1987, Baker et al 2002, Lang et al, 2003, Ahearne et al 2004, Ammer et al 2012). If no that is, if overseas listings do not help overcome home bias then the choice of overseas listing venues might reflect the same bias as that of investor holdings (Sarkissian and Schill 2004). In that case, firm managers would be aware that foreign investors have a home bias perhaps because they are just less willing to invest in equities of lesser-known companies (Kang and Stulz 1997) that the cross-listing cannot overcome, so firms might list their stocks only in the markets where they already have a substantial investor base (servicing the investor base, as it were). In another sense the decision is more complicated than what we have just described, because there are scores of potential overseas listing venues. If it is costless (in every way) to list, the firm might just list everywhere. But some markets, such as U.S. exchanges, impose substantial costs on foreign firms who list there. These costs can be monetary, regulatory, and legal. When there is a cost to cross-listing (which itself varies across countries and markets), a firm choosing to cross-list must weigh the costs and benefits of listing in each market. 8 One benefit could simply be to increase the firm s visibility for foreign investors. For example, consistent with Merton's 8 The different possible reasons a firm would cross-list could have different implications for valuations, a topic covered in the literature anchored by Doidge, Karolyi, and Stulz (2004), Gozzi, Levine, and Schmukler (2008) and Sarkissian and Schill (2009, 2012a). 14

15 (1987) investor recognition hypothesis, Baker, Nofsinger, and Weaver (2002) find a cross-listing on the NYSE is associated with increased analyst coverage (and decreased cost of capital). 9 But, as argued in Stulz (1999), because the quality or reliability of information varies around the world, a cross-listing that increases visibility might not matter if firms from countries with lower disclosure requirements cannot credibly communicate their prospects to investors; to raise external funds, such firms might cross-list on an exchange with higher disclosure standards, thereby sending a positive signal to investors. A reduction in information asymmetries is another possible benefit; Lins, Strickland, and Zenner (2005) argue that firms from developing countries that list on U.S. exchanges gain enhanced access to capital markets due to a reduction in information asymmetries that reflects greater disclosure requirements, shareholder rights protection, liquidity, and analyst following. Another benefit is bonding to stricter investor protections, which can reduce the agency costs of controlling shareholders (Coffee 1999, 2002; Stulz 1999; Reese and Weisbach 2002). 10 This discussion has important implications for our empirical analysis in Section 4. Some listing destinations might reflect investors existing preferences firms are just cross-listing where their existing clients are whereas other destinations could potentially open the market to the firm. In the former case, a cross-listing variable would have no effect on home bias, as the cross-listing is purely a function of other variables investors consider when making international investment decisions. In the latter case, cross-listing is special and has explanatory power above and beyond other home bias variables. And, of course, whether cross-listing matters or not could vary by destination market. 9 Baker et al also find that these benefits are greater than for foreign firms that list on the London Stock Exchange, partially offsetting the higher costs associated with a NYSE listing. 10 In general, the frictions that make the selection of trading venue relevant include cross-border barriers to investments (Black, 1974; Solnik, 1974; Stulz, 1981; Errunza and Losq, 1985) and information flow (Merton, 1987; Foerster and Karolyi, 1999), as well as market differences in liquidity (Tinic and West, 1974; Foerster and Karolyi, 1998; Domowitz et al., 2001; Werner and Kleidon, 1996), disclosure requirements (Biddle and Suadagaran, 1992; Fuerst, 1998; Huddart et al., 1999), and minority shareholder protection (Coffee 1999, 2002; Reese and Weisbach, 2002; Doidge et al., 2003; Lee, 2003; Benos and Weisbach, 2004). 15

16 3.2 Cross-listing and U.S. international investment This section summarizes the recent Ammer et al (2012) findings and can be skipped by the reader familiar with those results. In the U.S. setting, much work has noted that U.S. investment in foreign stocks tends to be higher in stocks that cross-list on a U.S. exchange, suggesting that there is something special about cross-listing (Ahearne, Griever, and Warnock 2004, Bradshaw, Bushee, and Miller 2004, Edison and Warnock 2004, Aggarwal, Klapper, and Wysocki 2005, Kho, Stulz, and Warnock 2009). But Ammer et al (2012) note that this stylized fact cannot be interpreted without reference to the underlying causal links between cross-listing and U.S. investment. Three features of crosslisting are important to consider: (i) it is a voluntary decision that (ii) is made primarily by firms that are large, well-established, and highly liquid the types that might attract substantial U.S. ownership even without the cross-listing and (iii) these firms might choose to cross-list exactly because the listing might attract large U.S. investment. To determine whether there is an actual cross-listing effect, Ammer et al estimate the unobservable component of what U.S. holdings in cross-listed firms would have been had they not cross-listed. Thus, the cross-listing effect is an estimate of the treatment effect E(H L i X L 1) E(H X 0), i where X is an indicator variable set to one when a firm has cross-listed on a U.S. exchange, L E(H i X 1) is the expected level of U.S. holdings in cross-listed firm i conditional on it being cross-listed, and E(H L i X 0) is the expected level of holdings in cross-listed firm i if it had not cross-listed. Ammer et al apply three different methods a Heckman-based two-stage 16

17 procedure, 11 propensity-matching, and difference-in-differences to estimate the cross-listing effect robustly. The main Ammer et al firm-level results are reproduced in Table 1. In the 1997 crosssection (Panel A), focusing on the market-capitalization based results (just for ease), L L E(H i X 1) is observed to be 16.5 percent (row 1). Estimates of E(H i X 0) are 6.4 or 7.0 percent (row 2 or 4), larger than average holdings of non-cross-listed firms (2.9 percent); crosslisted stocks have attractive features even without the cross-listing. But correcting for selection still suggests a sizeable, statistically significant cross-listing effect: U.S. holdings in a typical cross-listed stock are 10.2 (or 9.5) percentage points of market capitalization higher than they would be without the U.S. listing. Across time (Panel B), stocks that were not cross-listed in March 1994 but cross-listed by December 1997 saw increased U.S. holdings of 8.5 percent of market capitalization, while those who remained not-cross-listed increased only 0.6 percent (row 8). This 7.9 percent average cross-listing effect decreases only a bit to 6.8 percent (row 10) when instruments such as firm-level accounting quality, MSCI membership, and firm size are included. In sum, using security-level data Ammer et al find that selection adjustments matter. Based on three techniques, the average cross-listing effect that ranges from 7 to 10 percent of market capitalization (or about 15 percent of float). Yes, foreign firms with characteristics that attract ample U.S. investment even without the cross-listing (such as large firms that are financially transparent and liquid) are more likely to elect to cross-list in the United States. But a dramatic cross-listing effect remains once selection bias is controlled for: average U.S. holdings in foreign firms that cross-list on a U.S. exchange are two to three times higher than they would 11 This method is analogous to Lee s (1978) study of unionization and wages, which extends the Heckman (1979) selection-bias correction to a simultaneous system. 17

18 have been without cross-listing. 4. Does Cross-Listing Belong in Global Studies of International Portfolio Allocation? Cross-listing should not be omitted from studies of U.S. international investment, but is crosslisting important for global studies? Not every stock market in the world is the NYSE. Not all countries have regulators like the SEC and regulations like U.S. securities laws. At one extreme it could be that there is just something special about the U.S. as a cross-listing destination, but that cross-listing in other destinations do not influence international investment patterns. At the other extreme, perhaps every destination country is special. Or perhaps some are and some are not Description of the Global Holdings and Cross-Listing Dataset The most substantial obstacle to addressing the appropriate role of cross-listing in a global study is that it requires a global panel dataset of cross-listing. For this we updated the Sarkissian and Schill (2012a) annual dataset of 2,838 listings on foreign stock exchanges (i.e., not OTC listings). Data is available from 1985 (the start date of Worldscope market and accounting data) to 2012, based on the authors surveys of world stock exchanges as of the end of 1998, 2003, 2006, and Surveys were completed for all country exchanges indicated as having foreign listings by the World Federation of Exchanges, except for corporate tax havens (such as the Cayman Islands, Bermuda, and Jersey) and exchanges outside main boards of country stock exchanges. Exchange research departments were asked for a summary of all foreign companies, excluding investment funds and trusts, listed on their exchange. In all but the initial (1998) survey, the history of all foreign companies that were once listed but had since delisted their shares was also requested. 12 Listings of foreign shares were received for all exchanges. 12 For some, but not many, exchanges, foreign listing and delisting data are posted on the exchange website. 18

19 Delistings data are less complete; in some cases delisted share histories were only partial (e.g, going back 10 years) or unavailable. 13 See Sarkissian and Schill (2012a) for more details. Table 2, reproduced from Sarkissian and Schill (2012a), shows the distribution of foreign listing between all pairs of home and host markets. The table also reports the total number of listings from each home country and in each host market. The ten largest suppliers of listings are Canada (483 listings), the U.S. (288), the U.K. (239), Australia (163), India (162), Japan (142), Israel (137), Netherlands (120), France (97), and Germany (93). Almost 90% of Canadian listings are placed on U.S. exchanges and about 75% of Indian listings are placed in Luxembourg. The U.S. and U.K. are the most active host markets, with 1198 and 315 listings, respectively. They are followed by Luxembourg (251 listings), Germany (183), France (104), New Zealand (91), Canada (89), Switzerland (87), and Netherlands (71). The global cross-listing dataset presents the first limits to our working sample: The foreign listings data is for 53 source countries and 32 destination (or host) countries. The availability of global international investment data is the next constraint. We obtain dollar-denominated foreign equity investment data from the IMF Coordinated Portfolio Investment Survey (CPIS) for 1997 and 2001 to In the CPIS, there are 27 source (i.e. investor) countries in the 1997 CPIS and more than 60 starting in 2001; we add source countries to our sample as they become available in the CPIS dataset. Our main analysis focuses on the more complete data beginning in The CPIS data, used in Bekaert et al (2012) and Lane and Milesi-Ferretti (2012), among many others, are not pristine. For example, almost any international investment dataset, whether this is highlighted or not, will have a severe financial center bias that at the very least renders data vis-à-vis financial centers meaningless for the purpose of analyzing international investment. Thus, as is usually done in the literature, we omit Ireland and Luxembourg. The primary benefit of the CPIS dataset is that it is readily available for a range of countries across a range of dates For the U.S., incomplete delisted history was complemented with ADR delist codes from CRSP, following the procedure of Chaplinsky and Ramchand (2008). 14 CPIS data are also subject to a geographical bias because, especially in the first decade of the CPIS, most countries did not follow best practices and conduct security-level surveys to report data to the CPIS. Without 19

20 Panel A of Table 3 shows summary statistics and data construction for the relative investment weight, proportion of market capitalization listed abroad, and investment determinants. Previous research has focused on investments determinants that include familiarity and information, economic and financial development, accounting and legal standards, as well as other theoretically motivated variables. We follow the literature in our choice of determinants: log of distance between capital cities as a measure of geographic proximity, a dummy variable that equals 1 when two countries share a similar official language, bilateral trade to proxy for familiarity, and stock return correlations as a measure of diversification opportunities (Chan et al (2005), Coeurdacier and Guibaud (2011), Lane and Milesi-Feretti (2012), and Bekaert et al (2012)). We include as measures of economic and financial development unilateral trade, the ratio of market capitalization to GDP, and capital controls. 15 To gauge the importance of (a particularly simple form of) information, we include the number of internet users in a country as in Bekaert et al (2012). Because we are investigating bilateral investment between many countries, we follow Bekaert et al (2012) and include both destination and investor country unilateral variables. Panel B of Table 3 shows the unilateral and bilateral variables we use to instrument for the proportion of destination market capitalization listed in the investor country. Three unilateral instruments are log of destination GDP per capita, used in Doidge et al (2004) and Ammer et al (2012), the Djankov et al (2008) anti self-dealing index used in Sarkissian and Schill (2012b), and trading volume scaled by GDP, used in Ammer et al (2012). The one bilateral variable is a disclosure measure that takes the value of 1 if the investor country has higher disclosure standards than the destination country; following Sarkissian and Schill (2012a), we use the disclosure measure from Bae et al (2008) that indicates when a country s accounting requirements differ from or do not conform to IAS rules. knowing the exact security the investor holds, in many cases it can be very difficult to assign it to a particular country. 15 We use the measure of capital openness from Chinn and Ito (2008), which is time varying and updated through our sample period. 20

21 Our final working sample is the 32 countries for which investment determinants and cross-listing instruments are available Results Full sample In this section, we report results from a regression of (bilateral) relative weight on bilateral and unilateral variables. In particular, we test the hypothesis using the following regression: RW ijt = βcrosslistmcap ijt + γ 1X ijt + γ 2X it + γ 3X jt + θ t + ε ijt (5) where RW ijt is the share of country i's float-adjusted wealth invested in country j scaled by country j s float-adjusted share of world market capitalization, crosslistmcap ijt is the fraction of country j s market capitalization listed in country i, X ijt is a matrix of bilateral investment determinants, X it is a matrix of investor country i characteristics, X jt is a matrix of destination country j characteristics, θ t are year fixed effects, and ε ijt is an error term. We calculate standard errors clustered at the destination-country pair following Bekaert et al (2012). The first specification in Table 4 shows that full sample results from a least squares regression of the relative weight on investment determinants are generally consistent with previous findings. Across the full sample of countries, international investment increases when bilateral trade is higher, the distance between countries is smaller, and the home and destination country share an official language. Controlling for other variables, diversification opportunities do not appear to drive investor behavior, contrary to Coeurdacier and Guibaud (2011). Higher unilateral trade in both the holder and destination countries increases relative investment. While more internet users in the holder country lead to more outward foreign investment, capital account openness does not have a statistically significant effect on foreign investment. 16 We added a few countries to the Sarkissian and Schill (2012a) cross-listing dataset that host no foreign firms. 21

22 In the second specification, we regress foreign investment on the proportion of market capitalization listed on a foreign exchange, without including other variables. The coefficient on the cross-listed market capitalization proportion is positive and statistically significant, suggesting that in a full sample, bivariate setting cross-listing is associated with increased foreign investment. In the third specification both cross-listing and other explanatory variables are included. Controlling for other variables, the cross-listed proportion of market capitalization is still positive and statistically significant but with a smaller magnitude. As discussed in Section 3, however, estimating the effect of cross-listing on investment with OLS is incorrect if cross-listing is an endogenous choice by firms. Thus, the fourth specification of Table 4 uses an IV regression. The first stage regression is reported in the fourth column. Countries with a higher proportion of market capitalization listed abroad are less likely to conform to international accounting standards (relative to the market hosting the foreign listing) and have a higher volume of stock trading relative the GDP (consistent with idea that firms may be more likely to list abroad when they have outgrown the home market). 17 Results from the second stage of the IV regression of relative weight on the instrumented crosslist fraction (last column) indicate that cross-listing increases international equity investment. The coefficient on the proportion of destination market capitalization listed in the holder country is positive and statistically significant. Moreover, the p-value for the test of exogeneity of the instruments is 0.457, indicating that we cannot reject the null hypothesis that the instruments are exogenous. We note, however, that including cross-listing as an endogenous regressor does not alter inferences about the effect of other international investment determinants. For example, the statistical significance and magnitudes of coefficients on familiarity and economic development variables are not very sensitive to the inclusion of an endogenous cross-listing regressor Distribution of relative investment weight 17 Two of the four proposed instruments are statistically significant in the full sample first-stage regressions, but if the cross-listing decision is only endogenous for firms from a subset of countries, these instruments effects on the proportion of market capitalization listed abroad might differ across countries. We return to this point below. 22

23 Results in Table 4 were for the full sample, which like previous studies attempts to explain the determinants of overall international investment, but given the distribution of relative investment weights, the drivers of international investment may vary depending on the level and types of constraints that investors face. As a first attempt to explore potential variation in investment determinants, we break the sample into three subsamples based on the float-adjusted relative investment weight. Table 5 shows summary statistics for relative investment weight and the proportion of market capitalization cross-listed according to the distribution of relative investment weight. The first breakpoint, 0.1, is approximately the 50 th percentile. Relative investment in the lower half of the distribution (between 0 and 0.1) country pairs with zero or near zero bilateral investment has a mean of only Unsurprisingly, the proportion of cross-listed market capitalization is quite low, with a mean of only The second breakpoint is 1, where the country s portfolio and global market weights are identical. An RW of one, close to the 90 th percentile of the relative investment weight distribution, can be interpreted as the cutoff between underinvestment and overinvestment if the destination country share of world market capitalization is the appropriate benchmark from a model of international investment. 18 When the relative investment weight is between 0.1 and 1, mean relative investment weight is and the proportion of cross-listed market capitalization is markedly higher with a mean of Finally, when the relative investment weight is greater than 1 that is, the portfolio weight exceeds the benchmark global weight the mean of the proportion of cross-listed market capitalization is Table 6 shows that the determinants of international investment vary somewhat by relative investment weight. For the subsample of observations with relative weight less than 0.1 countries with zero or near zero bilateral investment the coefficient on cross-listed market capitalization is actually negative but statistically insignificant. For the subsample with relative weight between 0.1 and 1 country pairs for whom home bias exists but is not extreme the coefficient on the cross-listed market 18 To be clear, however, we recognize that this share may not be the appropriate benchmark as it depends critically on specifying the correct model and make no claims as such. 23

24 capitalization is positive and statistically significant. That is, for observations in the section of the distribution where investment is substantial but below world benchmark weights, cross-listing does indeed increase international equity investment. For the subsample with relative weights greater than 1 country pairs with more bilateral investment than a global benchmark would suggest the coefficient on cross-listed market capitalization is positive but not statistically significant. Turning to the remaining determinants of foreign investment, the results for distance are consistent across the distribution of relative weights greater distance decreases foreign investment. Many other determinants change sign, however, as the relative weight increases. For example, the coefficient on return correlation is positive and significant in the underweight samples (in columns 1 and 2), consistent with previous puzzling results on the lack of diversification motives in foreign investment. In contrast, when there are no barriers (perceived or otherwise) to cross-border investment (relative weight greater than 1), the coefficient on returns correlation is negative and significant, suggesting greater weights for countries that might provide more diversification benefits Splits based on different accounting and legal standards Results in Table 6 show that the determinants of investment vary across relative investment weights. These differences beg the question of what types of constraints matter for international investment. Moreover, the cross-listing literature discussed in Section 3 suggests that there is no theoretical reason a cross-listing in one country should have the same effect on international investment as a cross-listing in another country. For example, not all exchanges have the same disclosure requirements; if disclosure quality matters to dispersed portfolio investors (as argued in Stulz 1999) and a cross-listing on market i can force a change in a firm s disclosure, one would expect the cross-listing effect to differ across countries. In un-tabulated results, we find that in fact the effect of cross-listing in the home market does vary across countries. In OLS regressions at the country-level, cross-listing mitigates bias for many countries, but in many others, cross-listing does not appear to matter. Moreover, cross-listing might be endogenous in some countries but not in others. In other words, for some types of countries cross-listing is likely to be both endogenous and matter more as a driver of investment. 24

25 We hypothesize that cross-listing is likely to be both endogenous and a more important determinant of international investment in countries with high accounting and legal standards. We test this conjecture by splitting the sample of investor countries according to measures that reflect accounting rules, shareholder rights, and rule of law. In particular, we split the sample into home investor countries with low standards and high standards, with OLS results for low standards countries and IV results for high standards countries (under the assumption that cross-listing in low standards countries is not exogenous but cross-listing in high standards countries is endogenous). Given our previous results on how the effects of investment determinants vary with the distribution of relative weight, we show results separately for relative investment weight between 0 and 0.1, between 0.1 and 1, and greater than 1. To examine differences in the role of cross-listing on foreign investment across countries, we first start with a measure of accounting standards from the Center of Financial Analysis and Research (CIFAR) reported in Bushman et al (2008) that is the average number of accounting and non-accounting items disclosed from a broad set of 90 items in the annual reports by a sample of firms across countries. We label high accounting standards those countries with above median CIFAR score (based on countries in our sample); low accounting standards countries have a below-median CIFAR score. Our sample spans many years and accounting standards may change over time, so we adopt a dynamic measure of accounting standards. To capture changes in accounting standards, we reclassify low standards countries that mandate International Financial Research Standards (IFRS) adoption as high standards countries using IFRS adoption dates from International Accounting Standards Plus published by Deloitte Touche Tohmatsu. Table 7 shows results for countries with low standards and high standards, split also by relative weight. Cross-listing is positive and significant for low standards countries when the relative weight is less than 0.1 (Column 1) and greater than 1 (Column 5), and is positive and significant for high standards countries when the relative weight is between 0.1 and 1 (Column 4). For low accounting standards countries, the OLS coefficient on the proportion of market capitalization listed abroad is positive and significant for low standards countries when the relative weight 25