Redrawing the Map of Global Capital Flows: The Role of Cross-Border Financing and Tax Havens

Because the amount foreigners invest in securities issued by firms resident in tax havens “vastly exceeds the scale of these economies, making it implausible that the residency-based treatment reflects true economic activity.” The paper’s headline illustration is that CPIS records $3.9 trillion of foreign portfolio investment in the Cayman Islands in 2017 while Cayman GDP is $5 billion – “a thousand-fold difference” (Section 1.1). The authors do not dismiss residency: it “offers administrative ease and avoids conceptual ambiguity,” and when firms issue through operating affiliates outside havens “the residency concept may in some cases best represent the location where capital is deployed”; they also note that if capital raised never passes through the parent, tracing its use might be best done with residency positions plus intercompany lending data (fn. 8). Their claim is narrower and therefore stronger: for haven-resident issuers the residency relationship “most clearly lacks economic content and is not relevant for most economic analyses,” with the explicit exception of analysing taxes paid or collected (fn. 9). The concrete example carried through the paper is Petrobras International Finance Company, a Cayman-based subsidiary of Brazil’s largest energy company, whose bonds are recorded as euro-area investment in the Cayman Islands financial sector rather than in Brazil’s energy sector.

By combining seven commercial data sources to map each issuer of the 26 million stocks and bonds in the CUSIP Global Services master file to a single ultimate parent, with majority and priority rules resolving disagreements. The sources are the CGS Associated Issuer database, Refinitiv SDC Platinum New Issues, S&P Capital IQ, the Dealogic Debt Capital Markets feed, Bureau van Dijk’s Orbis, Factset Data Management Solutions, and Morningstar holdings data; the core procedure is CUSIP-based but also aggregates securities that have an ISIN but no CUSIP (Section 1.3, fn. 11). Coverage of haven issuance is high but uneven: the algorithm reallocates more than 90 percent of corporate bonds and equities issued by firms resident in each of Bermuda, Curacao, the Cayman Islands, the Channel Islands, Luxembourg, Macau, Panama and the British Virgin Islands, while Hong Kong, Ireland, the Netherlands and Singapore are “four exceptions with lower reallocation rates that range from 33 percent to 72 percent since these countries are destinations for offshore issuance but also have significant domestic issuance by companies actually operating there” (Section 1.3). The tax-haven list follows Hines (2010), itself an update of Hines and Rice (1994); the authors add the Netherlands (as in Torslov et al. 2018) and Curacao because both are large securities-issuance centres, and remove Switzerland “because our focus is on security issuance rather than taxation or illicit bank accounts, and Switzerland is not a large center for offshore security issuance” (fn. 10).

A matrix of shares, specific to investor country, asset class and year, that converts a residency-based position vector into a nationality-based one; the key assumption is that matrices estimated from Morningstar fund holdings are representative of all security investment, including by non-fund investors and by funds outside Morningstar. The construction is explicit: the entry for row i and column k is the fraction of investor j’s Morningstar holdings of country-i securities on a residency basis that map to country k on a nationality basis, each row sums to one, and nationality positions follow from pre-multiplying the residency vector by the matrix transpose (equations 1-3, Section 1.4). The worked illustration is that 20.1 percent of US corporate bond investments in the Cayman Islands reallocate to Brazil, 33 percent to China, and 13.3 percent to the United States itself. The assumption is needed because “TIC and CPIS cover the universe of security positions held by each country’s investors, a superset of those in the Morningstar data” (Section 1, p. 2). Matrices are recomputed each year, which recovers historical variation in haven use: the Brazil cell of the US corporate-bond matrix for the Netherlands rises from 0 percent in 2011 to over 10 percent by 2014 after Petrobras established Petrobras Global Finance BV there in 2012. The nine investor countries are the United States, the euro area, the United Kingdom, Canada, Switzerland, Australia, Sweden, Denmark and Norway; New Zealand is excluded relative to the earlier Maggiori-Neiman-Schreger work because key bilateral holdings are redacted in CPIS (fn. 12).

Against two entirely different investor types – US insurance companies and Norway’s sovereign wealth fund – and it holds closely. The two alternative datasets are chosen partly because they have very different coverage profiles: at end-2017 US insurers account for 20 percent of total US foreign bond investments and 3 percent of foreign equity investments in TIC while funds account for 31 and 47 percent; Norway’s sovereign wealth fund accounts for 71 percent of Norwegian foreign bond and 88 percent of foreign equity investment in CPIS while funds account for 4 and 6 percent (Section 3.2). Building alternative reallocation matrices from these holdings and comparing the implied restatements destination by destination, the best-fit lines through the origin have slope 0.98 with R-squared 0.95 for US corporate bonds, 0.99 and 0.97 for US equities, 0.91 and 0.95 for Norwegian bonds, and 1.00 and 0.98 for Norwegian equities (Figure 9). The authors’ conclusion is appropriately bounded: “Changing to nationality using our fund data or using the insurance and SWF data generate what are substantially the same restatements of the two countries’ bilateral investment positions.” A methodological aside from this exercise: in the course of the work the authors discovered and reported to Statistics Norway an error in Norway’s CPIS reporting of the sovereign fund’s bilateral composition, and use an internally amended version of Norway’s CPIS tables throughout (fn. 37).

Very much larger: US corporate bond holdings in the BRICS rise from $19 billion to $126 billion and euro-area holdings from $152 billion to $389 billion, with the reallocations traced to a handful of haven jurisdictions and a small number of very large firms. The residency-based starting point is strikingly small – US corporate bond investment of $8 billion in Brazil, $3 billion in China, $6 billion in India and close to zero in Russia and South Africa, against $390 billion in Canada, $308 billion in the United Kingdom and $144 billion in Australia, so that the BRICS account for just 1 percent of all US foreign corporate debt investment in 2017 and 2 percent of euro-area foreign bond holdings (Section 2.1). After reallocation the BRICS rise to 16 percent of the nine investors’ tax-haven bond positions in 2017, up from 5 percent in 2007, when more than half was Russian and virtually none Indian or South African (Figure 4). The conduits are specific: $12 billion from the British Virgin Islands and $27 billion from the Cayman Islands become US investment in China; $16 billion from the Cayman Islands and $22 billion from the Netherlands become US investment in Brazil; Russian firms use Luxembourg and Irish affiliates, and Indian and South African haven issuance occurs “almost exclusively in the Netherlands” (Figure 2). The exposures differ sharply by investor: US investors have nearly $12 billion of exposure to Russian firms through Luxembourg and Irish affiliates, while the restatement uncovers nearly $90 billion of European exposure to Russian firms (Figure 3). Concentration is high – five affiliates including Petrobras Global Finance BV (whose $28 billion of bonds reallocate to Brazil), Petrobras International Finance Company, Vale Overseas Limited and Odebrecht Finance Limited account for $61 billion, or 80 percent of the corporate debt reallocated to Brazil, with the corresponding top-five shares 59 percent for China, 75 percent for India, 68 percent for Russia and 70 percent for South Africa (Table 6). The authors draw a policy implication from this concentration: “Policymakers and analysts should pay attention to these large firms as even their idiosyncratic behavior can drive sudden stops or rapid changes in total portfolio investment at the country level.”

It shifts the picture toward portfolio rather than direct investment, toward corporate rather than government bonds, and toward foreign rather than local currency – with an important caveat about total liabilities. On the portfolio-versus-FDI margin, the mechanism is that when a haven affiliate issues a bond and transfers the proceeds to its parent, “this latter transfer would typically appear in the emerging market’s external accounts as an intercompany loan, a type of FDI,” whereas the nationality restatement records the bond itself as a portfolio liability (Section 2.1.2). This matters because “economists and policymakers view portfolio flows as more volatile than FDI investments” and countries often regulate the two differently on that presumption (fn. 5). On corporates versus sovereigns, reallocations are minimal for government bonds because “governments, after all, almost always issue under their own name and not via affiliates,” and even sovereign issuance in international markets keeps the two bases aligned; the result is that corporate bonds account for 25 percent of US holdings of Brazilian bonds under residency but 66 percent under nationality. On currency, the local-currency share of foreign-held bonds in the nine investors’ portfolios falls from 63 to 33 percent for Brazil, 66 to 52 percent for India and 70 to 40 percent for Russia, with declines of 5 and 11 percentage points for China and South Africa. The authors twice flag the accounting caveat that limits how far these results can be pushed: “Our restatements need not have any implication for a country’s total liabilities because the increase in portfolio investment may implicitly come from a decrease in other investment categories,” and to the extent an offshore affiliate passes identical funds in the same currency to its parent, the restatement “may raise the foreign currency share of portfolio liabilities but need not change the currency composition of overall external liabilities” (fn. 22).

Because Chinese firms in restricted industries list offshore through Variable Interest Entities, overwhelmingly in the Cayman Islands, and residency-based statistics record their shareholders as investing in the Cayman Islands. Under residency the United States holds $547 billion of Cayman common equities – similar to its holdings in Canada and larger than those in Germany or France – plus $195 billion in Bermuda-resident companies, more than its position in Indian companies (Section 2.2). Reallocating these, US equity exposure to China rises from about $150 billion to almost $700 billion and the euro area’s from under $100 billion to over $300 billion. Of the $542 billion increase in the US position, $477 billion comes from the Cayman Islands and the next largest amount, $48 billion, from Hong Kong; of the $227 billion euro-area increase, $187 billion from the Cayman Islands, $30 billion from Hong Kong and about $10 billion from everywhere else (Section 2.2.1). The structural evidence that this is about ownership restrictions rather than taxes is compositional: of the 25 firms that are Chinese by nationality and receive the most equity investment, only nine are resident in China, four in Hong Kong and twelve are Cayman-resident VIEs, and the prominent VIEs cluster in the internet and telecommunications sectors where Chinese restrictions bite – Alibaba, Baidu, JD.com and Tencent among them. “The stark differences in the industrial composition of the VIEs compared to the companies resident in China that directly raise financing from foreign investors corroborates that circumventing ownership restrictions is a key driver of China’s use of tax havens to raise equity financing.” The paper also draws out an investor-protection point: VIEs carry legal risk – Chinese regulators “might change the tax treatment of VIEs or even recognize them as illegal,” a risk Alibaba’s own IPO prospectus discloses – and while that risk was known, “our work demonstrates that the scale of exposure to these risks has been underappreciated due to residency-based reporting and represents a concern for financial stability.”

The argument is that foreign claims on VIEs enter China’s liabilities as intercompany or FDI positions whose recorded value does not track the listed companies’ market value, so a decade of large valuation gains to foreign shareholders is missing from China’s accounts. Two pieces of evidence support the premise. The market value of all VIEs rose from a few billion dollars in 2005 to almost $2 trillion by mid-2019, gaining more than $1 trillion in the six quarters from 2016Q4 to 2018Q1 – yet China’s reported inward FDI stock from Hong Kong, the British Virgin Islands and the Cayman Islands, which should be a superset of VIE-related investment, “displays none of the recent surge in the VIEs’ market value and toward the end of our sample even lies below the market value of VIEs” (Figure 7a). Nor can the positions be hiding in another liability category: while VIEs gained $1.1 trillion in market value between 2016Q4 and 2018Q1, China’s total recorded external liabilities excluding official reserves and trade credits rose only $390 billion, most of it a $180 billion increase in portfolio debt, “highly unlikely to include the VIE equity investments” (Figure 7b). The adjustment then assumes each foreign-held VIE position is recorded at the cumulative value of that VIE’s equity offerings, with Chinese residents’ own VIE holdings estimated from Bloomberg (Section 2.2.2). Against an official net credit position of $2.1 trillion in 2018 – 15 percent of Chinese GDP, comparable to Germany’s and against Japan’s $3.1 trillion – the estimated overstatement is close to zero in 2008 and reaches $1.1 trillion by end-2018. The sensitivity analysis moves the number modestly: assuming the 16 percent of VIE ownership that Bloomberg cannot identify is all Chinese reduces the overstatement to $0.9 trillion, while assuming China’s external assets are not overstated at all raises it to $1.4 trillion. Crucially, the authors do not allege misreporting: “We have corresponded with China’s statisticians and have no reason to believe their treatment of these FDI positions is inconsistent with official guidelines,” noting that BPM6 permits fallback valuation methods such as cumulated flows and that Chinese law does not recognise the listed shares as equity claims on the Operating Company (fn. 27-28).

That much of the feared external adjustment for China has already occurred unnoticed, and that the “world banker” reading of imbalances is reinforced. The authors’ framing is that “China’s net credit position is closer to that of Norway or Switzerland than it is to Japan’s,” and that although China has run large current account surpluses since the early 2000s it “is a much smaller net creditor today than statisticians, economists, and policymakers believe because its NFA does not reflect massive valuation changes” (Section 2.2.2). The policy inference is explicit: “Our estimates suggest that much of this external adjustment has already happened during 2008-2018 but went unnoticed as it was obscured in the statistics due to offshore issuance. Since foreigners realized very large capital gains on Chinese equities during this period, they retain substantial claims on China. Therefore, significantly less external adjustment will be required in the future than was previously thought.” They contrast the attention given to the $1.1 trillion of US Treasuries held by China with the near-absence of attention to the $700 billion of US holdings in Chinese equities, and note that developed-country investment in Chinese VIEs combined with Chinese investment in Treasuries “reinforce the world banker view of global imbalances.” They also flag that their adjustment covers only offshore VIEs and that other parts of China’s external accounts, such as Chinese holdings of US real estate documented by Li et al. (2020), could add further mismeasurement (fn. 33).

Three reasons, all specific to what the BIS International Debt Securities data contain. First, IDS reports only the total multilateral value of each issuing country’s securities outstanding under each basis, not the bilateral composition, and “many possible bilateral configurations are consistent with any given multilateral statistic” (Section 3.1). The authors test the obvious shortcut – scaling each bilateral residency position by the country’s nationality-to-residency ratio of outstanding bonds – and report in an appendix that it “has significant shortcomings” (fn. 34). Second, IDS covers no equities, “a major part of our results,” and includes only debt issued on international markets, where registration domain, listing place or governing law differs from the issuer’s residence; the authors give a vivid measure of the gap, noting the local-currency share of foreign-held Brazilian bonds is under 10 percent on both bases in IDS against roughly 70 percent and 34 percent in their own data, the difference largely owing to foreign holdings of local-currency Brazilian government securities that IDS excludes (fn. 35). Third, their approach is open-source and flexible, so users can vary the notion of nationality to suit the question. Where holdings data are unavailable they offer the issuance distribution matrix as a second best, restating CPIS for all reporting countries, but warn against treating it as equivalent because of home bias in tax havens – the baseline reallocates $50 billion more US corporate bond holdings back to the United States than the global matrix does (Sections 3.3-3.4).

Three, each matched to a different economic question: full nationality for corporate control, guarantor for credit risk transmission, and sales-based for demand exposure. Full nationality reallocates every affiliate’s issuance to its ultimate parent, haven or not – the paper’s recurring example being European holdings of Toyota Motors North America bonds, American under residency and the baseline but Japanese under full nationality. It strengthens rather than weakens the emerging-market result, “because emerging market companies also own subsidiaries in developed countries that issue bonds,” which matters particularly for India and South Korea, and for Brazil through US-operating subsidiaries such as JBS USA (Section 4.1). The guarantor treatment assigns a bond to the firm that guarantees it rather than the ultimate parent, on the reasoning that haven issuing vehicles often hold little or no assets so creditors demand explicit group guarantees; its illustration is the $1.2 billion of bonds the South African conglomerate Naspers issued via its Dutch subsidiary Prosus NV, which the baseline assigns to South Africa but the guarantor basis leaves associated with the Netherlands, because Prosus explicitly guarantees them with its own capital. The result is reassuring rather than disruptive: guarantor-based bond positions “show only muted differences relative to our baseline restatement, confirming that corporate control and financial backing typically coincide.” The sales-based treatment, using Factset GeoRev revenue-segment data, splits a single security across countries in proportion to where the issuer earns revenue, and is offered as potentially the most useful for “calibrating the geographic exposures of investors’ wealth to demand shocks in multi-country trade and macro models.” It raises China’s share of the US external equity portfolio further: 2 percent under residency, 10 percent under baseline nationality, 15 percent under sales, with 10 percent for the euro area; and the three measures imply three different trends over 2007-2017 – essentially no growth under residency, about 5 percentage points (almost a doubling) under nationality, and almost 10 percentage points under the sales basis (Section 4.2, Figure 11). Where sales data are missing for both issuer and ultimate parent the security’s residency association is left unchanged, and governments are treated as earning all revenue domestically (fn. 42).

As adding a new role for havens – conduits by which emerging market firms reach developed market capital – and as a global, open-source complement to official restatement efforts. On the first, the authors note that the tax-haven literature (Hines and Rice 1994, Desai et al. 2004, Gravelle 2009, Zucman 2013, Guvenen et al. 2018, Torslov et al. 2018) has largely concerned “the use of tax havens by wealthy households to shield assets from taxation and by developed market firms to minimize corporate tax exposures,” while “our results shed light on a different role of tax havens as conduits for emerging market firms to access developed market capital.” On the second, they acknowledge the shortcoming of residency statistics “has long been recognized,” cite initiatives at the BIS, the Federal Reserve and the IMF, and place their contribution as offering “a global analysis of portfolio investment for many countries and under different conceptual treatments,” with emphasis on open availability of code and data – their code runs even when supplied with only a subset of the commercial datasets (fn. 7). They note adjacent work: Bertaut et al. (2019) comparing US TIC under both bases, Damgaard et al. (2019) on FDI in the Coordinated Direct Investment Survey, Lane and Milesi-Ferretti (2018) and Avdjiev et al. (2016) on the growing role of financial centres. They also flag the breadth of downstream consequences: the restated positions have “clear relevance for any analyses using TIC or CPIS data,” including the gravity literature and recent demand-system estimation by Koijen and Yogo (2019) and Jiang et al. (2020). One further measurement point worth noting is that the restatement reveals substantial spurious foreign investment: 9 percent of all US holdings of foreign common equities and 11 percent of foreign bond holdings in official statistics “are better thought of as domestic investments,” largely reflecting Cayman-issued collateralized loan obligations backed by US assets and tax inversions into Ireland by US firms (fn. 14).