International Currencies and Capital Allocation

The paper asks what determines cross-border investment, and answers that the currency of denomination of assets – “an often neglected aspect” – drives portfolios at both macro and micro levels; the novelty is a security-level dataset of $32 trillion in global investment positions as of 2017. The authors motivate the question by noting that “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,” yet “surprisingly little is known about the determinants of cross-border investment,” in part for lack of detailed data (p. 1). The data are Morningstar’s complete position-level records from mutual funds and ETFs domiciled in over 50 countries, containing millions of individual positions – roughly 5 million unique positions held by about 9,000 US funds and about 6 million held by about 52,000 non-US funds in December 2017 (Section 2, p. 5). Positions carry a 9-digit CUSIP, which the authors use to attach currency, maturity, coupon, issuer location and industry.

The authors argue funds are both a large share of global portfolio investment and broadly representative of it, while conceding several specific gaps. Their US coverage reaches 93 percent of the ICI-reported value by 2017 and their non-US coverage captures between half and three-quarters of fixed-income assets under management over the last decade (Section 2.1, pp. 5-6). By OECD data, the share of total bond investment intermediated by investment funds in 2017 is 43 percent in the euro area and 23 percent in the US, averaging 36 percent across the ten countries analysed but ranging from 9 percent in Norway to 82 percent in Denmark (Section 2.2, p. 6). The conceded gaps are specific: US inward positions “do not align well” with TIC data, likely because large foreign entities such as Chinese and Japanese government institutions and large European insurers invest directly; the ECB reports a 37 percent dollar share of euro-area foreign bond holdings against 59 percent in this data, which the authors attribute to Luxembourg and Ireland being disproportionately important in the fund sector; no Japanese-domiciled funds are included and UK coverage is less complete than US; and the analysis excludes bank lending entirely, which matters because US non-financial corporations rely on bonds for 77 percent of debt financing against 17 percent for European firms (pp. 6-8).

Domestic bond investment is almost always in the domestic currency, while foreign investment in the same country’s corporate bonds is overwhelmingly not in that country’s currency. For 2017, the share of each country’s investors’ lending to that same country’s corporate issuers that is local-currency-denominated is above 0.8 for every country and close to 1 for most – about 95 percent for Canadian investors buying Canadian firms’ bonds. The mirror statistic for foreigners buying the same country’s corporate bonds is systematically and much lower: about 5 percent of Canadian corporate bonds bought by non-Canadians are in Canadian dollars (Figure 2, Section 3.1, pp. 9-10). Instead of local currency, foreigners hold bonds denominated either in their own currency or in dollars; excluding investment into the US, “the vast majority of all foreign investment is either denominated in the investing country’s currency or in US dollars” (Figure 3, p. 10). The same pattern holds for sovereign bonds but “is more muted,” which the authors attribute to most developed sovereigns issuing very little foreign-currency debt, and this is why the paper concentrates on corporate bonds (p. 10).

By exploiting variation in currency of denomination across multiple bonds issued by the same parent firm, so that nationality, industry, trade exposure and default risk are held fixed by a parent-firm fixed effect, with maturity and coupon controlled directly. The estimating equation regresses investor country j’s share of the total holdings of corporate bond c issued by parent p on a parent fixed effect and an indicator for the bond being denominated in j’s currency (equation 1, p. 11). The logic is stated plainly: “If Canadians, for instance, are much more likely to hold a given UK firm’s long-term Canadian dollar debt than that firm’s long-term British pound debt, this would support the conclusion that currency is the true underlying factor driving that investment decision” (p. 11). The estimates in Table 2 are all positive, statistically significant and large – the Canadian coefficient implies Canadian funds hold a share of a Canadian-dollar security about 90 percentage points larger than of a same-issuer security in another currency (p. 12). Table 3 reports eleven alternative samples and specifications – restricting to multi-currency issuers, to foreign issuers, to foreign issuers excluding home-market issuance, to financial and non-financial corporates separately, adding local government and supranational borrowing, adding sovereigns, and adding dummies for the security’s country of residence and for the investor country’s governing law – and the currency coefficient “remains economically large, stable, and precisely estimated” throughout (pp. 12-13).

No. The bias is pervasive at the fund level. Taking the 300 funds with the largest external corporate bond holdings, the large majority hold either all or none of their foreign investment in their own currency, and a lowess fit of the home-currency share on the size rank of funds’ foreign investment “is effectively flat” (Figure 4a, Section 3.3, p. 13). Repeating the plot for the share denominated in either the home currency or the dollar, the points are “nearly universally clustered near one,” across funds of different type, mandate and domicile (Figure 4b, p. 13). In fund-level regressions reported in the appendix, funds specialising in foreign investment – those with larger foreign shares of total assets – show less home-currency bias, and there is less robust evidence that the bias declines very mildly with total fund size (fn. 20, pp. 13-14).

It claims that, for corporate bonds, inference of home-country bias is confounded by home-currency bias – while being explicit that distinguishing the two definitively would require exogenous variation the authors do not have. Running the home-country indicator alone reproduces the literature: coefficients are positive for every country, ranging from 10 percent to 71 percent, and country information alone explains roughly one-third of the variation in securities’ holdings (Panel A of Table 4, pp. 14-15). Replacing it with a home-currency indicator gives point estimates and R-squared values “both approximately twice as large” (Panel B). Including both, the currency coefficient is “little changed” from its univariate value and the R-squared only slightly larger, while the country coefficient is “dramatically reduced” (Panel C, p. 15). The authors state the caveat directly: “Ultimately, distinguishing 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 relative explanatory power of country and currency” (p. 14). They credit Burger, Warnock and Warnock (2017) with first suggesting this possibility from US TIC data.

In most countries the firms that borrow substantially from foreigners are essentially only those that issue in foreign currency, and the relationship is close to one-for-one; the evidence against a pure firm-quality story is that the same local-currency-only firms do attract foreign equity. Plotting each Canadian firm’s foreign-currency share of debt against its foreign-held share of debt, a large mass of smaller firms sits at or just above the origin, and as firms borrow more in foreign currency they borrow more from foreigners, with points “clustered along the 45 degree line”; the euro area and the UK look similar (Figure 5, Section 4.1, p. 16). For local-currency-only firms, domestic portfolio shares lie almost uniformly above foreign portfolio shares in Canada, the euro area and the UK (Figures 7-9a, pp. 18-19). The equity test is the paper’s answer to the confound: “if something 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,” and in fact the domestic-foreign gap in those same firms’ equity portfolio shares “is far more muted,” with only small positive differences for Europe, Sweden and Norway and a negative gap for Denmark, New Zealand and Australia (p. 20). The authors flag the key caveat themselves: bank loans are unobserved, so local-currency firms might access international capital indirectly via domestic banks borrowing abroad – though they argue loan finance is generally more expensive than bond finance, so even then those firms would plausibly be worse off (p. 16).

Firm size predicts foreign-currency issuance robustly; the fixed-cost interpretation is offered as the natural reading of that size-dependence rather than as a separately tested mechanism. A probit for whether a firm has any foreign-currency debt, run country by country with two-digit SIC fixed effects and four alternative size proxies (bond principal outstanding, EBIT, total assets, revenues), yields average marginal effects that are “all positive and statistically significant” (Table 5, p. 17). The authors then note that “this type of size-dependence is a hallmark of selection in the presence of fixed costs,” and enumerate the costs: enriched accounting infrastructure, arranging and paying for currency hedges, a more sophisticated corporate treasurer’s department, a relationship with an international investment bank, foreign roadshows and investor meetings (pp. 17-18). They address the obvious alternative – that size proxies for foreign sales and hence a hedging motive – by conditioning on the foreign share of sales from Worldscope segment tables, and report that a higher foreign sales share matters significantly for the UK and US but is insignificant or negative for Canada and the euro area, while firm size “remains strongly and positively correlated” with foreign-currency issuance “across the vast majority of specifications” (p. 18). They also note the direction is not uniform in theory: large exporters may want foreign-currency debt to match receipts, but large importers have the opposite incentive (fn. 26, p. 18).

US firms that borrow exclusively in dollars place their bonds into foreign and domestic portfolios with comparable ease, which is not true of local-currency-only firms in any other country in the data. In the US version of the firm-level plot, “there is a significant mass of medium-sized firms that issues only in US dollars but receives substantial financing from foreigners” (Figure 5d, p. 16). In the LC-only firm plots, the US is “the one exception,” with domestic shares running through the middle of foreign shares rather than above them (Figure 8, p. 19), and in the aggregated bar chart the domestic and foreign bars are of similar height only for the US (Figure 9a, p. 19). LC-only firms also matter more there: they account for nearly 60 percent of total US corporate bonds in the data, against roughly 15 to 25 percent for Canada, the euro area and the UK (fn. 28, p. 19). The authors’ interpretation is that this constitutes a previously unremarked component of the “exorbitant privilege”: “international currencies effectively open up the capital account for firms that only borrow in domestic currency” (p. 2). They are careful to add that “measuring the benefits of selling bonds to foreigners or quantifying the ‘privilege’ from issuing in a global currency like the US dollar is beyond the scope of this paper” (p. 3).

The euro was a genuine international currency until 2008 and then lost that status to the dollar, and the shift survives removing the US and euro area, holding exchange rates fixed, controlling for sample composition, and dropping banks. In 2005 the dollar denominated 41 percent of global cross-border corporate debt holdings in the data and the euro 38 percent; after the global financial and eurozone crises the euro’s share fell to 22 percent by late 2017 while the dollar’s rose to 63 percent (p. 2; Figure 10, Section 5, pp. 21-22). Four robustness moves are reported. Excluding the US and the euro area as either investor or issuer leaves the pattern strong, which the authors read as showing the shift is “not simply attributable to changes in the relative size of the US and EMU markets nor is it directly driven by the unconventional monetary policy (quantitative easing) of the Fed or the ECB” (Figure 11b, p. 21). Re-generating the series with exchange rates fixed at 2005 levels shows the dollar’s appreciation “can only directly explain a small portion” of the trend (p. 22). Regressing currency shares on time and country-pair fixed effects rules out compositional change in the sample (Figure 11c, p. 22). Restricting to non-financial corporate borrowers shifts levels but leaves the reallocation intact (Figure 11d, p. 22). Across most specifications in Table 6 the dollar share rises by about 10 to 20 percentage points and the euro share falls by about the same. The authors add the cross-sectional counterpart: the dollar share of foreign investment into US corporate bonds was about 40 percent in 2005 against about 75 percent in 2017, and they describe these results as “suggestive that the roles of the dollar and euro in shaping cross-border capital allocation have changed during this period,” with identifying the driver left to future work (pp. 22-23).

That models must generate home-currency bias, not merely home-country bias, and must do so with differential strength across currencies – and that frictionless portfolio models cannot deliver the specific pattern observed. Benchmark models fail either because they generate no bond trading, as in Lucas (1982), or because they predict foreign investors take direct exposure to the borrower’s local currency (p. 4). Even the models that do produce home-currency bias frictionlessly, such as Solnik (1974) and Adler and Dumas (1983), “do not replicate our finding that foreign investors almost entirely avoid debt exposure to firms that issue only in local currency even when they buy the equity of those same firms,” because with perfect markets investors would not distort their allocation across firms but would adjust unwanted currency exposure with long-short positions in short-term risk-free bonds (Section 6, p. 23). The authors’ own reading is that home-currency bias “reflects a combination of financial frictions, like hedging costs, and behavioral biases that effectively segment the investor pool for firm debt by currency,” pointing toward models with currency segmentation as in Gabaix and Maggiori (2015) combined with size-based selection as in Melitz (2003) – and they say explicitly “we suspect many of our facts would emerge in such an environment, but leave it to future work to formalize the logic” (pp. 23-24). They also argue against a regulatory explanation: the bias appears across countries with different regulatory regimes, some hedging activity is observed in the data, US outward allocation shares match TIC data that includes entities not regulated like mutual funds, and it may simply be natural for issuers rather than funds to hedge, since a firm hedges once at issuance and holds the position to maturity while funds change exposures frequently (p. 24).

That the common presumption of a positive wealth effect from domestic depreciation may be overstated, because foreign investors hold far less local-currency debt than market-weight benchmarks would imply. The authors note that “a naive assumption that foreign and domestic investors buy securities in each country in proportion to their 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,” so “a domestic currency depreciation might not have as much of a positive wealth effect as is commonly conjectured” (p. 10). They immediately qualify this: the wealth effect also depends on the extent of hedging and on the residency of hedging counterparties, and “the lack of systematic data on derivatives use precludes us from drawing too strong a conclusion” (fn. 14, p. 11). They connect the finding to the “original sin” literature of Eichengreen and Hausmann, with the twist 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” (fn. 13, pp. 10-11), and note it complements Lane and Shambaugh (2010) and Benetrix, Lane and Shambaugh (2015) by linking the currency composition of external liabilities to the country composition of foreign investors in micro data.