Download Insights into Domestic Firms' Capital Accounts: Foreign Currency Issuance & Borrowing and more Summaries Literature in PDF only on Docsity!
International Currencies and Capital Allocation∗
Matteo Maggiori †^ Brent Neiman ‡^ Jesse Schreger §
August 2018
Abstract We establish that global portfolios are driven by an often neglected aspect: the currency of denomination of assets. Using a dataset of $27 trillion in security-level investment positions, we demonstrate that investor holdings are biased toward their own currencies to such an extent that each country holds the bulk of all debt securities denominated in their own currency, even those is- sued by foreign borrowers in developed countries. Surprisingly, currency is such a strong predictor of the nationality of a security’s holder that the nationality of the issuer – to date, the most powerful predictor in a voluminous literature on cross-border portfolios – adds very little explanatory power. While large firms issue bonds in foreign currency and borrow from foreigners, the vast majority of firms issue only in local currency and do not directly access foreign capital. These patterns hold broadly across countries with the exception of countries, like the United States, that issue an international currency. The global willingness to hold the US dollar means that even smaller US firms that borrow exclusively in dollars have little difficulty borrowing from abroad. Global portfolios shifted sharply away from the euro and toward the dollar starting with the 2008 financial crisis, further cementing the dollar’s international role and potentially amplifying the benefit that its status brings to the US.
JEL Codes: E42, E44, F3, F55, G11, G15, G23, G28. Keywords: Capital Flows, Exorbitant Privilege, Home Bias, Reserve Currencies.
∗This paper was previously circulated under the title “Unpacking Global Capital Flows.” We are grateful to Laura Alfaro, Luigi Bocola, Alberto Cavallo, Riccardo Colacito, Massimiliano Croce, Wenxin Du, Emmanuel Farhi, Gita Gopinath, Tarek Hassan, Arvind Krishnamurthy, Hanno Lustig, Gian Maria Milesi-Ferretti, Toby Moskowitz, Emi Nakamura, Jonathan Ostry, Monika Piazzesi, Diego Perez, Robert Ready, Kenneth Rogoff, Stephanie Schmitt-Grohe, Martin Schneider, Jeremy Stein, Jón Steinsson, Andrew Tilton, Harald Uhlig, Martin Uribe, Adrien Verdelhan, Frank Warnock, and Eric Van Wincoop for their comments, and we offer particular thanks to Steve Kaplan for his generous help with the project. Bob Freeman, Clark Hyde, Sara Lux, Christine Rivera, Ravi Wadhwani, and Matt Weiss offered outstanding technical assistance at various stages of the project. We thank Andrew Lilley, Antonio Coppola, Hillary Stein, Brian Wheaton, George Vojta, and Sanjay Misra for excellent research assistance. Our analysis makes use of data that are proprietary to Morningstar and/or its content providers. Neither Morningstar nor its content providers are responsible for any of the views expressed in this article. We thank the Becker-Friedman Institute, the NSF (1653917), the Sloan Foundation, and the Weatherhead Center for financial support. †Maggiori: Harvard University, Department of Economics, NBER, and CEPR. Email: maggiori@fas.harvard.edu. ‡Neiman: University of Chicago, Booth School of Business, and NBER. Email: brent.neiman@chicagobooth.edu. §Schreger: Columbia Business School and NBER. Email: jesse.schreger@columbia.edu.
1 Introduction
Capital crosses international borders far more today than only a few decades ago. In the late 1970s, almost none of the total outstanding value of US corporate debt was held by foreigners. Today, more than one-quarter is held abroad. In part due to a lack of detailed data, however, surpris- ingly little is known about the determinants of cross-border investment. We introduce a novel security-level dataset with, as of 2015, $27 trillion in global investment positions to demonstrate that portfolios at both the macro and micro levels are driven by an often neglected aspect: the currency of denomination of assets. We emphasize four findings. First, investors’ bond portfolios exhibit strong home-currency bias as they disproportionately invest in bonds denominated in their own country’s currency. Using micro data, we identify this effect by measuring the extent to which investors disproportionately hold bonds in their own currency relative to debt in other currencies issued by the same firm. This within-firm analysis allows us to disentangle the importance of the currency of denomination of a bond from possible confounding factors such as maturity, legal jurisdiction, and an issuer’s credit risk and sector of operation. This home-currency bias holds to such an extent that each country owns the vast majority of bonds issued in its currency, even when the issuer is foreign and resides in a developed country. In fact, given the currency of denomination of a bond, knowledge of the issuer’s nationality – the focus of a large and influential literature on home bias – adds very little information for predicting the investor’s nationality. If one considers only the global supply of bonds denominated in a country’s currency, that country’s investment portfolio exhibits little if any bias toward securities issued by domestic firms. Similarly, if one considers only bonds that are not denominated in a country’s currency, there is also little or no bias towards securities issued by domestic firms. Second, this home-currency bias is associated with a stark pattern of capital allocation across firms. In each country, a small number of large firms issue debt denominated in foreign currency and borrow from foreigners. By contrast, a large number of medium or smaller sized firms issue bonds only in the local currency (LC) and do not borrow substantially from foreigners. To demon- strate that this pattern does not simply reflect an unobservable characteristic of local currency borrowers that makes them unappealing to foreign investors, we show that these same local cur- rency borrowers do receive equity investments from abroad. These facts suggest that the currency of issuance itself is a key factor associated with the differential receipt of foreign capital. Third, the United States is the exception to the above patterns, with global investors uniquely willing to hold US dollars. In addition to their own currencies, foreigners invest a substantial portion of their portfolio in dollar-denominated securities, what we dub an international-currency bias, when they invest in all destination countries. This implies that when foreigners buy US
exhibit a bias toward local currency, only the largest firms would access foreign capital, much like the selection into exporting in the Melitz (2003) model of trade. As in the trade literature, quantitatively estimating the real economic impact of this selection will likely require a heavy structural apparatus that we leave for future work.
Related Literature. Our work relates to a large empirical literature linking net foreign asset dynamics to the differential composition of gross assets and gross liabilities, including important contributions by Lane and Milesi-Ferretti (2007), Gourinchas and Rey (2007), and Curcuru et al. (2008).^1 Our finding that foreigners’ portfolios are underweight local-currency debt to such an extent that the external debt liabilities of countries are in large part denominated in foreign currency complements the work by Lane and Shambaugh (2010) and Bénétrix et al. (2015). Our finding that home-country bias is largely attenuated within the set of local currency bonds expands upon the message in Burger et al. (2017), who first found using TIC data that US foreign investment across destination countries does not appear home-country biased in the subset of debt that is dollar denominated and suggested it might apply more generally across countries and debt markets. Boermans and Vermeulen (2016) find that a common currency is an important explanatory variable in a gravity portfolio setting for EMU-based investors. Koijen et al. (2018) examine the differential rebalancing behavior of domestic and foreign investors in response to ECB bond purchases. Our results on which firms select into foreign currency borrowing and the heterogeneity across countries in such selection have analogies both with the international corporate finance literature, including Gozzi et al. (2010, 2015) and Larrain and Stumpner (2017), and the trade literature following Melitz (2003). The model of Salomao and Varela (2016) features an endogenous funding choice by heterogeneous firms that must pay a fixed cost to borrow in foreign currency. They apply their framework to data on Hungarian firms and study the link between their borrowing and investment decisions. Liao (2016) shows that variation in the currency-hedged cost of debt across different currencies predicts firms issuance: firms issue the most in those currencies in which borrowing is cheaper (including the cost of currency hedging). Bruno and Shin (2015b,a) study how movements in the dollar affect capital allocation and corporate investment via a balance sheet (^1) Other recent work includes Alfaro et al. (2008), Bertaut et al. (2014), Du and Schreger (2017), and Lane and Milesi- Ferretti (2018). These papers make use of the IMF’s International Investment Position (IIP) and Coordinated Portfolio Investment Survey (CPIS), the US Treasury’s International Capital Flow (TIC) data, and the BIS’s Debt Security Statistics and Locational Banking Statistics. A related literature studies international mutual fund data, but typically concentrates on equity flows or includes only a small subset of countries (See, for example, Chan et al. (2005), Hau and Rey (2004, 2008b,a), Forbes et al. (2016), Jotikasthira et al. (2012), Raddatz and Schmukler (2012), and Didier et al. (2013)). Hau and Lai (2016) focus on European money market funds to study monetary policy. Hale and Obstfeld (2016) examine the effect of the euro on the geography of cross-border debt investment. Kalemli-Ozcan et al. (2017) uses loan-level data to examine how global shocks drive capital flows to Turkey. Koijen and Yogo (2017) demonstrate how to estimate a demand system for equity investments using a dataset of holdings at the institutional investor level. Our work suggests currency would be an important factor in such estimates for bond investments. Choi and Kronlund (2017) study Morningstar data on US corporate bond mutual funds.
channel, and Bruno and Shin (2017) provide evidence that the recent increase in dollar borrowing by emerging market non-financial corporates is driven by these firms running a carry trade. Our results on the special role of the dollar and its use in denominating internationally held bond contracts complements a growing body of research. The existing literature including Ca- ballero et al. (2008), Mendoza et al. (2009), Gourinchas et al. (2011), He et al. (2018), Maggiori (2017), and Farhi and Maggiori (2018) has mostly focused on the safe-haven properties of the US dollar and the lower risk-free rate it affords to US government bonds, whereas we focus on the allocation of capital among corporate borrowers and offer evidence that the US “exorbitant privilege” includes the unique ability of US corporates that only borrow in dollars to raise capital from foreigners. Our finding that most cross-border bond positions are denominated in dollars, even when neither the investor nor the issuer are based in the US, has a mirror in the dominance of the dollar in invoicing traded goods, discussed in Goldberg and Tille (2008), Goldberg (2010), Gopinath (2016), and Gopinath and Stein (2018). It also relates to the international use of the dollar as a unit of account and means of payment modeled by Matsuyama et al. (1993), Doepke and Schneider (2017), and Chahrour and Valchev (2017). Finally, the empirical patterns that we document offer a challenge as well as new guidance for international macro models. Benchmark models cannot match our facts because they generate no bond trading, as in Lucas (1982), or because they predict that foreign investors, conditional on investing in a country, tend to take on direct exposure to the borrower’s local currency, as in Alvarez et al. (2009), Bacchetta and Van Wincoop (2010), Pavlova and Rigobon (2012), and Lustig and Verdelhan (2016).^2 A few models do generate home-currency bias either as the optimal solution of a frictionless portfolio choice (Solnik (1974), Adler and Dumas (1983), Engel and Matsumoto (2009), Coeurdacier and Gourinchas (2016)) or exogenously by postulating that households invest abroad in bonds denominated in their own domestic currency (Gabaix and Maggiori (2015)). Even these few models, however, would struggle to match the skewed foreign capital allocation – where foreign currency issuers receive the bulk of foreign investment – that we show is a critical feature of the data. We conclude in Section 6 by elaborating on these points and suggesting how future work might generate models in which currency is critical for both debt investors and issuers and in which the US dollar plays a special global role.
2 Mutual Fund Investment Data
Morningstar, Inc., one of the world’s largest providers of investment research to the asset man- agement industry, provided us with their complete position-level data collected from mutual funds (^2) Also see Corsetti et al. (2008), Tille and Van Wincoop (2010), Devereux and Sutherland (2011), Dou and Verdelhan (2015), Colacito and Croce (2011), Colacito et al. (2017), Hassan (2013), and Hassan et al. (2016).
US. Figures 2c and 2d further show that our data on funds domiciled in the European Monetary Union (EMU) and the UK closely track over time the equivalent aggregates provided by ICI.^5 To ensure that analyses are not influenced by domiciles for which Morningstar data are unrepresenta- tive, our analysis is performed on a subsample of the data that includes those developed economies for which Morningstar’s coverage of fixed-income funds is at least one-quarter of what ICI reports for that market at the end of 2015. These criteria select a final sample of 21 countries, about half of which are subsumed into the EMU. Table 1 lists the remaining 10 effective countries, ranked by the order of their AUM in 2015 in our data. While the US and EMU clearly account for the bulk of global AUM, we observe about $1 trillion in AUM for the UK and Canada.
2.2 Representativeness of Mutual Fund Investments
Mutual fund data is valuable for studying global capital allocation both because mutual funds directly constitute a sizable share of all global portfolio investments and because mutual fund investments are in many ways representative of aggregate cross-border portfolio investment. While mutual funds are differentially important across countries, they always constitute one of the main holders of securities. The left panel of Table 2 uses OECD data to show that the share of total bond investment in 2015 that is intermediated by mutual funds is 40 percent in the EMU, 21 percent in the US, and averages about one-third across the 10 countries included in our analysis. Comparisons with publicly available datasets suggest that, in the characteristics that we em- phasize, our data appear largely representative of the broader set of portfolio investments. In the appendix, we include figures demonstrating that the country and currency shares of US outward investment in our data closely match their equivalents in TIC data. Since TIC covers all portfolio investment, including positions by pensions and hedge funds, for example, this suggests that US mutual fund positions are broadly representative of US portfolio positions. We also report similar statistics for inward investments, which do not align well with our data. This likely owes to large foreign entities directly investing in US securities, such as government institutions in China and Japan or large European insurance companies. To examine the representativeness of non-US mutual funds, we compare our data with reported positions from the CPIS, a survey of cross-border portfolio holdings conducted by the IMF that includes information on the currency of foreign debt holdings for a few countries in recent years. In the appendix, we include tables demonstrating that the currency composition of Canadian, Danish, Swiss, and US portfolios in 2015 are similar in our data and in CPIS, as is also the case for a (^5) The ICI data for non-US domiciled funds are available quarterly on their web page when they release their “World- wide Public Tables”. We were able to obtain these tables for most quarters since the first quarter of 2005 using the Internet Archive (https://web.archive.org/). We log-linearly interpolate between the ICI values in the first quarter of 2005 and their values in the second quarter of 2002, which we obtained from Khorana et al. (2005).
number of EMU member countries. We cannot directly compare the data for the EMU as a whole since the CPIS does not report a consolidated EMU figure that removes intra-EMU investment. Our data align less well with aggregates reported by the European Central Bank (ECB). For example, the ECB reports the dollar share of EMU foreign bond holdings in 2015 to be 37 percent, below the 57 percent in our data. The discrepancy likely reflects the fact that Luxembourg and Ireland, countries that are disproportionately important in the mutual fund sector, have a higher share of their foreign holdings in dollars than the EMU average. Finally, it is important to highlight that our analysis focuses on bond finance and therefore excludes information on bank lending. The right panel of Table 2 compares the shares of bonds and loans (so they sum to 100 percent) in non-financial corporate liabilities in 2015. As is well known, US firms rely more heavily on bond financing (77 percent of total debt financing) than do European firms (17 percent).^6 The share of bonds is between one-third and one-half in countries like Australia, Canada, and the UK. Despite this heterogeneity, we note that the key patterns we highlight hold similarly among all non-US countries.
2.3 Mapping Positions to Firms, Industries, and Countries
Morningstar reports the domicile country of each mutual fund but does not have information on the nationality of individuals who invest in each fund. In general, tax optimization and regulatory restrictions make it unlikely that investors buy mutual funds domiciled in other countries.^7 Based on this principle, we assume that the domicile of a mutual fund is also the country of residency of its investors and we use the two concepts interchangeably in the rest of the paper. Notable exceptions are funds domiciled in Ireland and Luxembourg, which include a large number of Un- dertakings for Collective Investment in Transferable Securities (UCITS) funds that are designed to be sold throughout the European Union under a harmonized regulatory regime. Given our focus on currency, we pool all data for countries within the EMU, including Luxembourg and Ireland, and treat the EMU itself as a single consolidated country in our benchmark analyses.^8 We demonstrate in the appendix the robustness of our main analyses to the removal of Luxembourg, Ireland, and the EMU from our dataset. Turning from investors to issuers, one benefit of working with security-level data is that we can trace issuers to their ultimate parent company, which allows us to associate security issuance (^6) See De Fiore and Uhlig (2011) for an analysis of the sources of the differential reliance on bond and loan finance in the US and Europe. 7 In the appendix, we provide support for this assumption using TIC data that shows that US outward investment is only rarely directed to foreign funds and that foreign investment into the US is only rarely directed to US funds. 8 This leaves open the possibility of some cross-border holdings of mutual fund shares for countries that are in the EU but not in the EMU (such as Sweden or the United Kingdom), as well as the possibility of investors outside the EU buying some UCITS in Luxembourg and Ireland.
holder that the nationality of the issuer – to date, the most powerful predictor in a voluminous literature on portfolio determination – has little additional explanatory power. We also introduce the notion of international-currency bias, the tendency in our data of investors to disproportionally hold securities denominated in an international currency such as the US dollar.
3.1 Country Level Results
We find that domestic bond investments are almost always denominated in the domestic currency. For example, when Canadian investors buy bonds issued by Canadian issuers, the bonds are almost always denominated in Canadian dollars. However, foreigners invest differently. When Australians buy bonds issued by Canadian issuers, the bonds are rarely denominated in Canadian dollars. Figure 3a plots the shares of investment that are in the issuer’s currency for the bond portfolios in our data as of December 2015. The shaded red bars on the left illustrate for each country the share of all lending by that country’s investors to that same country’s issuers that is denominated in the local currency. For example, the second red shaded bar from the top shows that about 95 percent of lending by Canadian investors to Canadian issuers is denominated in Canadian dollars, as per the example above. The red shaded bars are all above 0.75 and most are quite close to
- Unsurprisingly, and consistent with conventional modeling assumptions in the literature, all countries invest overwhelmingly in local currency when buying the bonds of domestic issuers. More surprising, however, is our finding that foreigners invest differently. The hollow blue bars on the right of Figure 3a show the same statistic but for foreign investment portfolios, i.e. the share of foreign investment in each country’s bonds that is denominated in the issuer’s currency.^9 For example, the second blue hollow bar from the top shows that less than 20 percent of bonds purchased by non-Canadian investors and issued by Canadian entities are denominated in Canadian dollars. If foreign and domestic investors held similar portfolios in each market, then the length of red and blue bars would be identical in each row. On the contrary, Figure 3a shows that the blue bars are systematically (much) smaller than the red bars for each row. Domestic investment is almost always in the local currency. Excluding (for now) investment in the United States, foreign investment is rarely in the local currency. Figure 3b performs the analysis separately for sovereign bonds, where this pattern still holds but is more muted. Most developed countries’ sovereigns issue a very limited amount of foreign currency bonds (the US government, for example, does not issue in foreign currency). While we (^9) The hollow blue bars on the right are calculated by simply adding up positions over multiple foreign investors that purchase from each issuer country. The relative weight of these foreign investors therefore implicitly relates to its scale of AUM in our data and therefore may differ from equivalent values reported by national statistical agencies. We have disaggregated the hollow blue bars into the portfolios from individual investor countries and verified that these patterns hold robustly across bilateral pairs. See the appendix for details.
show that foreigners are disproportionately likely to buy those few foreign-currency denominated sovereign bonds, we also show that they buy substantial amounts of local-currency bonds.^10 The picture for corporate bonds is extremely stark. As seen in Figure 4, which restricts the analysis to corporate bonds, foreigners are very unlikely to hold local-currency corporate debt. For example, whereas roughly 20 percent of foreign investment in all Canadian bonds was in Canadian dollars, less than 10 percent of foreign investment in Canadian corporate bonds is in Canadian dollars. Unlike sovereigns, many corporations issue a substantial fraction of their debt in multiple foreign currencies, thus offering investors the possibility to invest in the same issuer but in the currency of their choice. Since our focus is precisely on this currency choice, both from the investor and the issuer perspective, we focus our analysis in the rest of the paper on the corporate bond market.^11 Rather than holding local-currency bonds, foreigners tend to hold bonds denominated either in their own domestic currency or in an international currency, such as the US dollar.^12 Figure 5a shows the currency composition of each country’s external bond investments. We exclude investment in the United States to focus purely on the international role of the dollar. The vast majority of all foreign investment is either denominated in the investing country’s currency or in US dollars.^13 Our results imply a strong sorting of foreign investment away from local currency bonds, de- spite the fact that these bonds constitute the bulk of the corporate bond market in each country. This sorting underlies the importance of studying portfolio holdings and not just the stock of securities outstanding to understand the external positions of countries. For example, a naive assumption that foreign and domestic investors buy securities in each country according to market-value weights would imply that developed countries have external liabilities denominated in their own currency and external assets denominated in foreign currency to a greater extent than is in fact the case.^14 (^10) For an analysis of determinants of the currency composition of sovereign debt, see Ottonello and Perez (2016), Engel and Park (2018), and Du et al. (2016). (^11) We rule out that the stark currency selection in corporate bonds is purely an artifact of rules preventing mutual funds from investing in foreign currency. In fact, we have shown that the same class of investors, open-end mutual funds, buys sovereign bonds predominantly in foreign currency. (^12) Foreigners’ holdings of dollar-denominated securities do not fully explain their low holdings of local-currency se- curities. To see this, Figure 5b simply replicates the results in Figure 4 after dropping all dollar-denominated holdings and excluding the United States as an issuer. (^13) The outsized role of the US dollar in cross-border portfolios of corporate debt that do not involve the US as either the investor or the borrower provides a possible channel for the outsized role of US monetary policy in global economic activity, as discussed in Bruno and Shin (2015b) and Rey (2015). See also Wiriadinata (2018), Zhang (2018), and Mukhin (2018). (^14) A large literature on “Original Sin” such as Eichengreen and Hausmann (1999) and Eichengreen and Hausmann (2005) has emphasized the similar fact that emerging economies borrow from foreigners in “hard” currencies like the US dollar, presumably due to their inflation risk, weaker institutions, or less developed internal capital markets. We show, however, that even rich and developed economies that do not suffer from these problems borrow in foreign currency from foreigners to a surprising extent via their corporate sector. Adams and Barrett (2018) and Fanelli (2017)
tionately holds securities denominated in its home currency. If country j had no home-currency bias then β (^) j would be zero.^16 Our benchmark estimates are run using data for 2015, are weighted by the total holdings in our data of each security, and control for maturity and coupon payment.^17 Table 4 reports our estimates of equation (1). Looking across the top row, the β (^) j coefficients are all positive, statistically significant, and large in magnitude. For example, the top row of column 1 shows that if a security is denominated in Canadian dollars, Canadian mutual funds hold a share of the total holdings of this security that is 93 percentage points larger than what they hold of securities that are not denominated in Canadian dollars but issued by the same issuer. This implies that Canadian investors hold the vast majority of Canadian dollar securities that are issued around the world. A similar effect holds for all other countries. Even among bonds issued by the same issuer, investors disproportionately hold those bonds that are denominated in their home currency. Table 5 demonstrates the robustness of our results by reporting the same β (^) j coefficients from various alternative samples of our data.^18 The first specification estimates equation (1) when we drop firms that only issue in local currency and restrict the sample to only those firms that issue in multiple currencies (MC), since variation within these firms is what identifies the currency bias. To be included in this specification as an MC issuer, a firm must issue in the local currency of the investor country and at least one other currency. The second specification only includes foreign issuers and the third specification additionally excludes any issuance by these firms that is done in the issuer’s domestic market. The fourth and fifth specifications restrict the sample to financial and non-financial corporates, respectively. The sixth and seventh specifications also examine financial and non-financial corporates separately, but additionally restrict the sample to only include for- eign firms. The eighth specification includes borrowing by local governments and municipalities, sovranationals such as the World Bank, and various structured fixed income products. The ninth specification includes all bonds in our dataset (including sovereigns). Finally, our tenth specifica- tion distinguishes securities not only by issuer and currency, but also by residence (i.e. the country where the security is issued). In particular, we add to the currency dummy in equation (1) a dummy for the security being issued in the investors’ country ( j). This specification allows us to ensure that our results are driven by currency and not by investors exhibiting a preference for bonds issued (^16) Our approach differs from that more commonly used in the home-bias literature in two ways. First, we use in our benchmark regressions of equation (1) a country’s share of total holdings rather than measure the ratio of the share that a security accounts for in a country’s portfolio relative to the share that security accounts for in total holdings. These two measures are linear transformations of each other within countries, so regressions that use either measure as the dependent variable contain the same information. Second, whereas the literature often uses worldwide market capitalization to measure total holdings, we measure total holdings internal to our mutual fund data. 17 We control for maturity with dummies corresponding to the categories: less than 2 years, between 2 and 5 years, between 5 and 10 years, and greater than 10 years. We treat coupon similarly by using seven equally spaced buckets from below 1 percent to greater than 6 percent. 18 We denote statistical significance at 1 percent using asterisks, but to improve the presentation, we do not report standard errors. Standard errors are clustered at the level of the fixed effects.
in their own legal jurisdiction. While for some countries, the residence of bond issuance does enter statistically significantly, it only very slightly attenuates the coefficient on currency. In all these analyses, despite the extensive differences in the included sample of issuers and the variation used to estimate fixed effects, the coefficient on home currency bias remains economically large, stable, and precisely estimated.
3.3 Home-Country Bias and Home-Currency Bias
A voluminous prior literature has documented the strength and pervasive presence of home-country bias, more commonly referred to as simply “home bias.” The influential work of French and Poterba (1991) found that investors disproportionally hold equity securities issued by domestic firms. The subsequent literature demonstrated that the same is true, to an even greater extent, for bonds. Furthermore, while equity home-country bias has seen a marked decline over the recent years, bond home bias has declined much less, as shown in Coeurdacier and Rey (2013). Home- country bias is to date the singularly effective force for empirically characterizing global portfolios and is essential for the quantitative performance of models in international macroeconomics and finance.^19 Our results, however, offer the intriguing possibility that home-country bias largely reflects home-currency bias, since the propensity to issue in local currency is greater for local borrowers. Indeed, Burger et al. (2017) first suggested this possibility by demonstrating with US TIC data that home bias measures greatly attenuate when excluding non-dollar securities. Ultimately, dis- tinguishing a bias for home-currency from a bias for home-country requires exogenous variation in either country or currency. While we do not have such exogenous variation, we compare the rela- tive explanatory power of country and currency by estimating equation (1), adding a home-country indicator ( (^1) {Countryp= j}, equal to one when parent issuer p is located in country j) and dropping the firm fixed effects (since the country and firm indicators are collinear). We run three related regressions:
s (^) j,p,c = α (^) j, 0 + γ (^) j, 01 {Countryp= j} + ε (^) j,p,c, (2) s (^) j,p,c = α (^) j, 1 + β (^) j, 01 {Currencyc=Currency (^) j} + ε (^) j,p,c, (3) s (^) j,p,c = α (^) j, 2 + γ (^) j, 11 {Countryp= j} + β (^) j, 11 {Currencyc=Currency (^) j} + ε (^) j,p,c. (4)
Equation (2) is a classic home-country bias regression that measures the extent to which a coun- try is overweight securities issued by domestic firms. Panel A of Table 6 reports the estimates of (^19) Additionally, see Fidora et al. (2007), De Moor and Vanpée (2013a,b), and Adams and Barrett (2018) for studies of home-country bias in bond portfolios and Lewis (1999), Sercu and Vanpée (2007), and Bekaert and Wang (2009) for surveys of the literature.
borrowing firms. We show that in each country a small number of foreign-currency borrowers are typically the only firms that borrow substantially from foreigners. In each country, most firms borrow only in local currency and their debt is mostly held by domestic investors. We also show that, consistent with the country level results in Figure 4, the United States is an exception to this rule: US firms that only borrow in dollars place their debt into foreign and domestic portfolios with comparable ease.
4.1 Foreign Currency Issuers Borrow from Foreigners
In most countries, only firms that issue in foreign currency place substantial shares of their bond debt in foreign portfolios. For example, Figure 6a plots for each Canadian firm with debt in our data in 2015 the share of the total firm debt that is denominated in foreign currency, i.e. currencies other than the Canadian dollar, against the share of the total firm debt that is held by foreigners. The scale of each firm’s bubble captures the market value of its total bond borrowing. We have aggregated the data across all debt securities issued by each firm, including those issued by subsidiaries or other associated issuers. This plot exemplifies two common features of the data. First, a large mass of smaller (by debt) firms are at the origin or slightly above it. These are smaller Canadian firms that borrow only in Canadian dollars and almost entirely borrow from Canadian investors. Second, as firms borrow more and more in foreign currency, they borrow more and more from foreigners. The relationship is nearly one for one, with the data points clustered along the 45 degree line. Figures 6b and 6c show similar patterns for the European Monetary Union and the United Kingdom. An important caveat is that we do not observe firm loan financing by banks. Hence, our data do not rule out the possibility that local-currency firms access the international market indirectly by receiving loans from domestic banks that themselves borrow from abroad in foreign currency. Even in this case, however, local-currency firms might be adversely affected since the loans are likely to come at a premium over direct bond financing from the foreigners. An extensive corporate finance literature has indeed shown that loan financing is in general more expensive than bond financing, including Diamond (1991), Rajan (1992), Bolton and Scharfstein (1996), and De Fiore and Uhlig (2011, 2015). The relationship between foreign currency issuance and foreign borrowing is markedly differ- ent for firms in the United States, as shown in Figure 6d. While it is still true that foreign currency borrowers tend to borrow more from foreigners, there is a significant mass of medium sized firms that issues only in US dollars but receives substantial financing from foreigners. One way to in- terpret these data is that the global taste for holding dollar debt securities effectively opens up the capital account for local currency borrowers in the US, whereas local currency borrowers in other countries are relegated to borrowing almost exclusively from domestic investors.
The fact that the bubbles located away from the origin in Figure 6 are generally larger shows that bigger firms are more likely to borrow in foreign currencies. For example, for the case of Canada, Figure 7a ranks firms along the x-axis in terms of their total borrowing, from the the largest borrower on the left to the smallest borrower on the right. The y-axis plots the number of currencies in which the debt of each firm is denominated. Toward the right end of the plot, nearly all firms only issue bonds denominated in a single currency (which, in this case, is typically Canadian dollars). Moving to the left, as firms’ borrowing increases, firms issue in an increasing number of currencies. The largest Canadian borrower in our data issues bonds denominated in 7 different currencies. Figures 7b, 7c, and 7d show a similar pattern in the EMU, the UK, and the US. Together with Figure 6, this implies that large borrowers issue in foreign currency and borrow from abroad, whereas small and medium borrowers issue in domestic currency and borrow from domestic investors. We can more formally analyze selection into foreign currency borrowing by estimating on data for 2015 the following probit model: Pr
(^1) {MCp} = 1
α (^) j + β (^) jSizep + γ (^) j,pIndustryp
where (^1) {MCp} is an indicator for a firm p having debt in foreign currency, Sizep is a measure of firm size, and Industryp are a set of fixed effects capturing the firm’s two-digit SIC. Unlike our prior analyses, we estimate equation (5) using operating and balance sheet data from Compustat (North America and Global) and Worldscope and using issuance data from the SDC New Issues database.^22 We proxy for firm size using four alternative measures: total bond principal outstand- ing, profits (EBIT), total assets, and revenues. We include industry fixed effects to account for differences in capital intensity, the collateral value of the firm, and propensity to be involved in ex- port/import activity since these might in turn affect the capital structure decision by the firm. This regression is run separately for each country in our sample, and so the intercept α (^) j, the industry fixed effects γ (^) j,p, and coefficients on the different proxies for size β (^) j are allowed to vary across countries. Table 7 presents the average marginal effects for the country atop each column from estimates of equation (5) using each of our four size proxies. All estimates are positive and statistically significant: Bigger firms, all else equal, are more likely to issue in foreign currency. All the different measures of firm size point in the same direction. This type of size-dependence is a hallmark of selection in the presence of fixed costs. In- deed, issuing in foreign currency often involves substantial set-up costs. Firms need to build an (^22) In the regressions, we use data from SDC instead of our data from Morningstar since the SDC data captures all issued bonds, not just those held by mutual funds. The results are robust, however, to instead using Morningstar data. We merge the SDC database with firm-level balance sheet data using the CUSIP6 of the Ultimate Parent as reported in SDC. Figures 6 and 7 use only the Morningstar data.
the solid red dots roughly split through the center of the hollow blue diamonds, indicating that LC-only firms in the US are almost equally likely to represent a given share of domestic or for- eign portfolios. US firms that borrow only in dollars, unlike LC-only firms in the other countries, borrow substantially from foreigners.^23 Aggregating across firms, we sum the solid red dots and hollow blue diamonds in each of the sub-plots in Figure 9 and plot in Figure 10a for these four countries (and the other 6) the aggregate shares of LC-only issuers’ debt in domestic portfolios as red bars and the aggregate shares of LC- only issuers’ debt in foreign portfolios as blue bars. The red bars are almost always dramatically taller than the blue bars, confirming that LC-only firms account for a far larger share of domestic than of foreign investment portfolios. The one exception is the United States, where the red and blue bars are of similar height. US firms that issue only dollar-denominated debt account for similar shares of domestic and foreign investment portfolios. Figure 10b shows that LC-only firms account for vastly different percentages of the overall corporate debt outstanding across countries. LC-only firms in the US account for nearly 60 percent of the country’s total borrowing, as shown in the first bar on the left. The equivalent value for Canada, the European Monetary Union, and the United Kingdom ranges from 15 to 25 percent. Taken together, the above results are consistent with the view that selection into foreign cur- rency borrowing leads to different outcomes across countries. In this view, US firms face ample demand for their bonds, both by domestic and by foreign investors, even when just borrowing in dollars. These firms, consequently, mostly borrow in dollars and only issue in foreign currency when their borrowing needs grow extremely large. Firms in countries with a smaller local-currency debt market, like Sweden, quickly outgrow the demand for their local currency debt and in order to borrow more (without pushing interest rates too high) switch to foreign currency borrowing. In these countries, even relatively small firms borrow in multiple currencies and MC-firms account for most of the countries’ overall borrowing. One might worry that the above patterns, at least for countries other than the US, reflect dif- ferences between the local-currency and multi-currency firms that are distinct from, though cor- related with, the currency of the debt security. Perhaps local-currency firms are in industries for which foreign investors naturally lack expertise or interest. Alternatively, multi-currency firms might be those that export a lot to foreign destinations and are therefore well known to foreign investors. To evaluate this possibility, we proxy a firm’s appeal to foreign investors using the firm’s equity portfolio shares. After all, though debt and equity do not offer identical payoffs, if some- thing about a firm caused it to be a fundamentally unappealing investment for foreigners, foreign investors should avoid both the firm’s equity and its debt. If equity markets are unaffected by (^23) In the appendix, we repeat this analysis separating issuers into financial and non-financial corporations. The docu- mented patterns hold across both subsamples.
currency-related frictions (for example, because equities are real assets not affected by the cur- rency of denomination), then the equity portfolio shares provide a helpful model-free benchmark for what optimal debt portfolio shares might look like in the absence of home-currency bias. Fig- ure 11 considers the same LC-only firms as in Figure 10a, but plots the share of their equities in domestic and foreign equity portfolios for that market. It is clear that the difference in LC-only firms’ shares of foreign and domestic equity portfolios, if any, is far more muted than is the case for their debt securities, even for countries other than the United States. For example, there is only a small positive difference for Europe, Sweden, and Norway, and the gap is actually negative for New Zealand. To investigate this further, Figure 12 explores the joint holdings of equity and debt of the same firm by foreign and domestic investors. We define a measure of how overweight foreigners are in the debt or equity of a firm p by taking the log of the ratio of the foreign portfolio share of firm p to the domestic portfolio share of firm p in that asset class. The higher this ratio, the more overweight the foreign investors are for that firm. A log ratio value of zero means a firm represents the same portfolio weight in domestic and foreign portfolios. We include all firms with both an equity and a bond security in our sample and plot the foreign overweight ratio for debt on the vertical axis and for equity on the horizontal axis.^24 LC firms are depicted with red circles and MC firms with blue ones, with the size of each circle proportional to the total market value of the total debt of the firm. For the MC firms in countries other than the US, there is a strong positive correlation of the debt and equity foreign overweight ratios, as seen in the upward sloping blue best-fit lines.^25 If MC firms attract a lot of foreign equity investment, they also attract a lot of foreign debt investment. By contrast, the red best-fit lines for the LC firms are flat. Unlike the case for MC firms, even when LC firms receive a lot of foreign equity investment, this is not also associated with large foreign debt investments. Finally, we again see that the US constitutes an exception, with the foreign overweight ratios for debt and equity behaving more similarly. Whereas the two best-fit lines are flatter than with the MC firms for the other countries, the US is the one case in which the dollar-only issuers and foreign-currency issuers exhibit a similar relationship between debt and equity foreign overweight ratios. In sum, investor home-currency bias and the firm-size dependency for foreign-currency is- suance together imply that most firms issue only local-currency debt and do not borrow much from abroad. The United States, however, issues an international currency and represents an exception to these patterns. Even smaller US firms place their dollar-denominated bonds into foreign portfo- (^24) We drop firms for which either the domestic or the foreign portfolio share is zero since in these cases the log ratio is not defined. We winsorize the log ratio for both debt and equity at the 1% level. In unreported results, we confirmed the robustness of our analysis to introducing these data points by setting the corresponding log ratio to a very large or very small constant. 25 The best fit lines are weighted by the amount of debt issued by each firm owned by mutual funds in the dataset.
