Jesse Schreger | Macro Paper Warehouse

CIP deviations, the dollar, and frictions in international capital markets

This is a survey chapter, written for Volume V of the Handbook of International Economics, rather than a paper presenting new results; its claims are drawn from the literature it reviews and its own descriptive statistics. (The full text used here is the freely available NBER working-paper version of May 2021.) The organising fact is the dollar’s outsized role in global finance: the United States is about 15 percent of world trade and 25 percent of global GDP, but the dollar accounts for roughly 50 percent of cross-border loans and international debt securities, 90 percent of FX transactions, 60 percent of official reserve holdings and 50 percent of trade invoicing. Because “the global market for dollar funding is highly fragmented” and many participants who need dollars cannot reach dollar-rich lenders directly, large global banks have to intermediate – and since the Global Financial Crisis their balance sheet constraints have tightened, partly through regulatory reform. The most visible symptom is the failure of covered interest rate parity, measured by the cross-currency basis: the difference between the cash-market dollar rate and the synthetic dollar rate implied by borrowing in foreign currency and swapping into dollars. The chapter documents a sharp pre- and post-crisis dichotomy – CIP “held remarkably well prior to the GFC,” with only fleeting deviations of 30 seconds to 40 minutes – and shows the post-crisis deviations survive replacing Libor with OIS or with government-collateralised repo rates, so they are not simply a credit spread. Three stylised facts follow. The basis is generally negative, with the Australian and New Zealand dollars the G10 exceptions; it correlates 90 percent in the cross-section with the level of nominal interest rates since 2008, which means the hedged CIP arbitrage runs opposite to the unhedged carry trade; and it has a strong factor structure, with the first principal component of quarterly changes in the five-year G10 bases explaining 51 percent of variation over 2008Q1-2020Q3 and correlating 96 percent with the average basis, so the basis widens in bad times alongside a strong broad dollar, high VIX, wide BBB-Treasury spreads and negative intermediary capital shocks. The explanatory framework is a supply-and-demand diagram for swapped dollars. Pre-crisis supply was perfectly elastic at a zero basis; post-crisis the leverage ratio requirement, which “mandate[s] banks to maintain capital against all assets, regardless of their risk characteristics,” makes even a riskless matched-book trade costly, tilting the supply curve upward so that demand shifts now move the equilibrium basis. On the demand side the chapter identifies three client types willing to pay the basis as an intermediation fee: non-top-tier non-US banks with local-currency insured deposits but dollar assets, non-US institutional investors with local-currency liabilities and dollar portfolios, and multi-currency corporate issuers exploiting currency-segmented bond markets. Central bank swap lines are the crisis backstop, priced at a fixed spread over OIS that fell from 100 basis points in the Global Financial Crisis to 50 in November 2011 and 25 in March 2020, with peak outstanding of about $580 billion in 2008-09, $110 billion in the European debt crisis and $450 billion during COVID. The chapter then separates government bond CIP deviations, which need not be arbitrage at all, since they can reflect sovereign default risk, capital controls and market segmentation, or cross-country differences in convenience yields. A final section surveys two views of what CIP deviations mean for exchange rates – one treating them as a signal of intermediaries’ risk-bearing capacity, the other as a determinant working through bond convenience yields – and the chapter closes with open questions about whether post-crisis regulation is calibrated correctly, about the growing role of non-banks, and about the macroeconomic consequences.

International Currencies and Capital Allocation

Using a new security-level dataset covering $32 trillion in global investment positions, this paper establishes that the currency a bond is denominated in – not the nationality of its issuer – is the dominant predictor of who holds it, and that this home-currency bias leaves most firms borrowing only at home while a small number of large foreign-currency issuers capture nearly all foreign bond capital. The data are Morningstar’s complete position-level holdings of open-end mutual funds and exchange-traded funds domiciled in over 50 countries, filtered to the 23 countries (14 of them inside the euro area, leaving 10 effective country units) where Morningstar’s coverage of fixed-income assets under management is at least a quarter of what the Investment Company Institute reports. Four facts follow. First, home-currency bias is strong and is identified within firm: comparing an investor country’s share of two bonds issued by the same parent but denominated differently, and controlling for maturity and coupon, Canadian funds hold a share of a Canadian-dollar bond about 90 percentage points larger than of a non-Canadian-dollar bond from the same issuer, with similarly large and precisely estimated coefficients for every other country (Table 2, p. 12). Running home-country and home-currency indicators side by side, the currency coefficient and R-squared are roughly twice those on country alone, and adding currency collapses the country coefficient while barely moving the currency one – so, at least for corporate bonds, the classic home-country bias documented since French and Poterba (1991) is largely confounded by home-currency bias (Table 4, pp. 14-15). Second, home-currency bias travels with a stark allocation of capital across firms: in each country a small number of large firms issue in foreign currency and borrow from foreigners, while most firms issue only in local currency and are held almost entirely by domestic investors. Probit estimates using Compustat, Worldscope and SDC data show that bigger firms are significantly more likely to issue in foreign currency on all four size proxies used (Table 5, p. 17). That this is not simply about which firms foreigners find unappealing is shown by the fact that the same local-currency-only firms do receive foreign equity investment (Figure 9b, p. 20). Third, the United States is the exception: a significant mass of medium-sized US firms issues only in dollars yet receives substantial foreign financing, which the authors read as the global taste for dollar debt effectively opening the capital account for local-currency US borrowers – a pattern found for no other country in the data (Section 4, pp. 16, 19-20). Fourth, in the time series the dollar’s role is recent rather than permanent: the dollar denominated 41 percent of global cross-border corporate debt holdings in the data in 2005 and the euro 38 percent, shares that were largely stable until 2008, after which the euro’s fell to 22 percent and the dollar’s rose to 63 percent (Introduction, p. 2; Section 5, pp. 21-22). The paper is explicit about its scope: the dataset contains quantities but not prices, so it cannot assess borrowing costs or quantify the value of the dollar’s privilege; it covers bond finance only and excludes bank lending; the analysis is of corporate rather than sovereign bonds; and the authors deliberately establish the four facts without identifying the mechanisms behind them, offering hedging costs, market segmentation by currency and fixed issuance costs as candidate explanations for future work to formalize.

Local Currency Sovereign Risk

Sovereigns that borrow in their own currency can always print the money to pay, so a common modeling assumption is that such debt is free of default risk; this paper builds a measure to test that and finds it false. The authors define the local currency credit spread as the yield on a local-currency (LC) government bond minus a synthetic LC risk-free rate assembled from the U.S. Treasury yield plus the long-dated forward premium implied by cross-currency swaps – equivalently, the promised dollar spread a global investor locks in by holding the LC bond with a matched swap, and equivalently the size of the failure of long-term covered interest parity between emerging-market and U.S. government bond yields. Using a new hand-built dataset of daily zero-coupon LC and foreign-currency (FC) yield curves and swap curves for 10 emerging markets from January 2005 to December 2011, at a 5-year benchmark tenor, they find the mean spread of LC nominal yields over U.S. Treasuries is 5 percentage points, of which their decomposition attributes 3.72 points to currency risk and 1.28 points to credit risk. The LC credit spread averages 128 basis points, is positive and statistically significant for every one of the ten countries, and stays significantly positive even after deducting half the bid-ask spread on the swaps to allow for transaction costs (that half-spread averages 19 basis points). It is nonetheless generally lower than the same sovereign’s FC credit spread, which averages 195 basis points – a gap of 67 basis points, widening to 86 basis points once swap transaction costs are netted out – and significantly negative in every country except Brazil, where a financial-transactions tax on foreign fixed-income investment drove the two apart. The two spreads also differ in structure: the first principal component explains only about 53 percent of LC credit spread variation across countries versus over 81 percent for FC spreads, and FC spreads are far more tightly tied to global risk factors (a 93 percent correlation between the first principal component of FC spreads and the VIX, against 76 percent for LC spreads). The ex-ante pattern is mirrored ex post: once currency risk is swapped away, LC bond excess returns carry no significant loading on global equity returns while FC excess returns do, so hedged LC debt is safer than FC debt on this measure despite the reputation of emerging-market local debt. Turning to why the spreads differ, the paper distinguishes differential cash-flow risk, differential liquidity, and differential pass-through of global risk aversion, and shows in panel regressions with country fixed effects that the VIX and bid-ask liquidity measures alone explain 46.7 percent of the within-country variation in the LC-minus-FC spread differential – the large majority of what is explained even after a full set of local and global macroeconomic fundamentals is added. The scope conditions matter: ten countries, a single seven-year window spanning the global financial crisis, a 5-year tenor, and a measure that is model-free about default but silent on which of taxes, convertibility restrictions, selective default, or risk premia drives the level of the spread in any individual country.

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

Because global firms raise capital through subsidiaries incorporated in tax havens, official residency-based statistics attribute those securities to the haven rather than to the parent’s country; this paper matches the universe of traded securities to their ultimate parents and restates bilateral investment positions, finding developed-market financing of large emerging market firms to be dramatically larger than reported and China’s net creditor position to be roughly half its official size. The scale of the problem is set by two numbers: the corporate sector globally raises 7 percent of its equity and 9 percent of its bond financing through foreign subsidiaries located in tax havens, and CPIS records $3.9 trillion of foreign portfolio investment in the Cayman Islands in 2017 against a Cayman GDP of $5 billion. The method has three steps. First, combining seven commercial data sources, the authors map each issuer of the 26 million stocks and bonds in CUSIP Global Services’ master file to a single ultimate parent, reallocating more than 90 percent of the corporate bonds and equities issued in each of Bermuda, Curacao, the Cayman Islands, the Channel Islands, Luxembourg, Macau, Panama and the British Virgin Islands. Second, merging that mapping with Morningstar security-level holdings of 61,000 funds reporting over 11 million positions worth $32 trillion as of December 2017, they build “reallocation matrices” giving, for each investor country, asset class and year, the share of residency-based holdings in each country that belongs to each other country on a nationality basis. Third, they apply those matrices to two public residency-based datasets – the US Treasury’s TIC and the IMF’s CPIS – for nine developed investor economies with adequate fund coverage. Two patterns dominate the redrawn map. Bond positions in the BRICS are far larger: US corporate bond holdings in the BRICS rise from $19 billion to $126 billion, a 560 percent increase, and euro-area holdings from $152 billion to $389 billion, because emerging market corporates issue through haven affiliates partly to spare foreign bondholders withholding taxes that are 15 percent in Brazil and 20 percent in Russia but zero in the British Virgin Islands, the Cayman Islands, Luxembourg and the Netherlands. Equity exposure to China is far larger still: US holdings rise from about $150 billion to almost $700 billion, the euro area’s from under $100 billion to over $300 billion, overwhelmingly reflecting Variable Interest Entities listed in the Cayman Islands. Because foreign claims on VIEs enter China’s accounts as intercompany positions valued without reference to listed share prices, China’s reported net creditor position of $2.1 trillion at end-2018 is overstated by $1.1 trillion. The paper is careful about what it does and does not establish. Its central identifying assumption is that reallocation matrices built from fund holdings are representative of all security investment, which it tests against US insurance-company and Norwegian sovereign-wealth-fund holdings, obtaining best-fit slopes of 0.98 to 1.00 with R-squared of 0.95 to 0.98. On China’s accounts it states that it has “corresponded with China’s statisticians and have no reason to believe their treatment of these FDI positions is inconsistent with official guidelines” – the claim is one of mismeasurement relative to market value, not of misreporting. And it insists there is no single correct restatement: alongside the baseline it offers full-nationality, guarantor-based and sales-based alternatives, since “the most appropriate concept in accounting for these positions will depend on the question at hand.”

The Costs of Sovereign Default: Evidence from Argentina

The question behind the paper is the oldest one in sovereign debt – why governments repay creditors who have almost no recourse – and the obstacle is that “governments usually default in response to deteriorating economic conditions, which makes it hard to determine if the default itself caused further harm to the economy.” The authors’ solution is a natural experiment. After Argentina’s 2001 default, the hedge fund NML Capital bought defaulted bonds, refused the 2005 and 2010 restructurings, and sued in New York courts under the pari passu clause; the courts eventually blocked Argentina from paying its restructured bondholders unless the holdouts were paid too, and Argentina refused. Rulings for NML therefore raised the probability of a default on the restructured bonds and rulings for Argentina lowered it, while carrying no information about Argentine fundamentals – the identifying assumption being that the rulings move firms’ stock returns only through the sovereign’s risk-neutral default probability, which requires both that US judges have no private information about Argentina’s economy and that the rulings do not hit the firms directly. Using daily data from 3 January 2011 to 29 July 2014, fifteen isolated court rulings, two-day event windows, credit default swaps to measure the five-year cumulative risk-neutral default probability, and a heteroskedasticity-based estimator in the manner of Rigobon and Sack, the paper finds that a 10 percent increase in that default probability causes a 6.043 percent negative log return on a value-weighted index of Argentine American Depository Receipts, and a 1 percent depreciation in each of three measures of the unofficial exchange rate (statistically significant only for the black-market Dolar Blue). Because the five-year risk-neutral default probability rose from roughly 40 percent to 100 percent over the sample, linear extrapolation of the log return implies the episode cut the value of the indexed firms by about 30 percent – 39 percent for banks, 30 percent for non-financials, 43 percent for the oil company YPF – and a wholly unanticipated default (0 to 100 percent) would imply a 45 percent fall. Scaled against earnings, an unanticipated default would destroy market value worth “roughly eight years of annual earnings,” which is one to two orders of magnitude larger than the cash-flow news a standard calibrated model would generate; the authors offer very persistent or permanent output losses as one “(speculative) explanation” while stressing it is not the only one. Sorting firms by the characteristics the theoretical literature points to yields only “suggestive evidence” that export-intensive firms, foreign subsidiaries and large firms are hurt more than their betas would predict, with banks economically large but statistically insignificant; the authors call this “modest support” for several existing theories and note both directions in which external validity could fail – Argentina’s costs may be understated because it was already shut out of international markets, or overstated because it chose to default despite an ability to pay.