Banking, Trade, and the Making of a Dominant Currency

Dominance is a constellation of five facts spanning trade, banking, corporate borrowing, reserves and returns – and the paper’s goal is a single model that produces all of them together. On invoicing, “an overwhelming fraction of international trade is invoiced and settled in dollars,” and crucially “the dollar’s share in invoicing is far out of proportion to the U.S. economy’s role as an exporter or importer of traded goods”: 60 percent of Turkey’s imports are dollar-invoiced while only 6 percent of its imports come from the US, and in a 43-country sample “the dollar’s share as an invoicing currency for imported goods is approximately 4.7 times the share of U.S. goods in imports,” against a corresponding multiple of only 1.2 for the euro. On bank funding, “the dollar liabilities of non-U.S. banks, which are on the order of $10 trillion, are roughly comparable in magnitude to those of U.S. banks,” and BIS locational statistics put 62 percent of banks’ foreign-currency local liabilities in dollars. On corporate borrowing, 60 percent of banks’ foreign-currency local claims are dollar-denominated, and much of that borrowing “is in many cases done by firms that do not have corresponding dollar revenues, so that these firms end up with a currency mismatch, and can be harmed by dollar appreciation.” On reserves, the dollar accounts for 64 percent of world official foreign exchange reserves, the euro 20 percent and the yen 4 percent. And on returns, “U.S. dollar risk-free assets generally pay lower expected returns (net of exchange-rate movements) than the risk-free assets of most other currencies,” a UIP violation sometimes called the dollar’s “exorbitant privilege” (Section 1).

Because with sticky invoice-currency prices, safety in real terms means holding the currency your future purchases are priced in. The authors’ fundamental observation is that “in a multi-currency world, one cannot think about the structure of safe asset demands without taking into account invoicing patterns. Simply put, a financial claim is only meaningfully ‘safe’ if it can be used to buy a known quantity of some specific goods at a future date, and this necessarily forces one to ask about how the goods will be priced.” Walking through an emerging-market household holding a buffer stock of deposits: “If most of its imports are priced in dollars – and crucially, if these dollar prices are sticky – the household will tend to prefer deposits denominated in dollars, as these are effectively the safest claim in real terms from its perspective. In other words, while deposits in any currency may be free of default risk, in a world in which exchange rates are variable, only a dollar deposit held today can be used to purchase a certain quantity of dollar-invoiced goods tomorrow” (Section 1). The authors connect this to Calvo’s “Price Theory of Money,” in which sticky prices in a unit of account give money an “output anchor,” and to Keynes’s remark that fixed contracts and sticky money wages play a large part in money’s liquidity premium (fn. 8).

Importers get utility directly from a Cobb-Douglas aggregate of safe dollar and safe local-currency deposits, with weights meant to stand for invoice shares; the authors call this “arguably an ad-hoc way” to do it and solve a consumption-based version in the appendix to show the results survive. The utility function is linear in date-0 consumption and date-1 wealth plus a term theta·log(M), following the money-in-the-utility-function approach of Krishnamurthy and Vissing-Jorgensen, Stein, Sunderam, Greenwood et al. and Nagel, with M a Cobb-Douglas aggregate of home-currency and dollar deposits that has constant returns to scale (Section 2.1, equation 1). The authors state the limitation directly: the formulation “captures these features in what is arguably an ad-hoc way, because we take the shortcut of assuming that importers derive utility directly from their portfolio mix across safe dollar and safe local-currency claims, without relating this mix to their ultimate time-1 consumption.” They then report that “to assure the reader that our main results follow even from a more conventional model where the utility function of importers is defined only over current and future consumption, in the Appendix we solve a variant of the model” with a concave, risk-averse utility over the date-1 consumption bundle – at the cost of having to solve it numerically, and at the cost of an additional segmentation assumption restricting importer households to safe assets (fn. 9). Two further first-order-condition results are noted: the risky asset trades at the lowest price, because it “does not provide any monetary services,” and the two safe claim prices “need not be equalized,” which is exactly a UIP violation whose sign cannot yet be determined.

A bank pools risky local-currency projects and can promise safe claims only up to the worst-case value of its collateral; because that collateral is local-currency revenue, it is a worse technology for producing dollar safe claims than local-currency safe claims. The bank has N projects paying a risky return in domestic currency, funded by safe local-currency claims, safe dollar claims or risky local-currency bonds, and faces the collateral constraint E-bar·B-dollar + B-home <= gamma_L·N, where gamma_L is the worst productivity realization and E-bar > 1 the most depreciated local currency value (Section 2.2). The intuition the authors extract: “the bank has a comparative advantage in manufacturing local-currency safe claims relative to dollar-denominated safe claims. This is because the bank’s underlying collateral is a collection of projects that pay off in local currency. Given the risk of currency depreciation, an amount of local collateral that is sufficient to back one unit of safe local-currency claims is only enough to back 1/E-bar units of safe dollar claims.” Remark 2 is careful about who “the bank” is: it is “an agglomeration of banks and the local non-financial firms that the banks lend to,” since no individual firm with an idiosyncratic chance of zero payoff can issue safe claims while a pool of N uncorrelated projects can – the pooling-and-tranching role identified by Gorton and Pennacchi. The authors add that whether banks or non-financial firms ultimately bear the currency risk does not matter for the model’s logic, but that “as a matter of empirical reality, the existing evidence suggests that a significant amount of the exchange-rate risk is borne by the non-financial corporate sector in emerging markets,” so their mismatch propositions “are best taken as statements that refer at least in part to mismatch among non-financial firms.”

In an interior equilibrium where the bank issues all three liabilities, the dollar safe claim has the highest price and lowest expected return, with the wedge pinned at exactly E-bar; the point is that this UIP violation is derived from the invoice share rather than assumed. Proposition 1 states Q-dollar > Q-home > Q-risky and (Q-dollar minus beta)/(Q-home minus beta) = E-bar, so “UIP is violated, and dollar deposits benefit from an ’exorbitant privilege’ relative to local-currency deposits: they have a higher price and a lower expected return.” It is “a direct consequence of the bank’s comparative disadvantage in creating dollar safe claims out of local-currency-denominated collateral. Because of this disadvantage, the bank will only be willing to fund these local projects with dollar borrowing if doing so is cheaper than funding with domestic deposits.” The introduction spells out why the direction matters: “if one takes the UIP violation as exogenous, it seems obvious why some firms might be willing to court exchange-rate risk by borrowing in dollars – it can be worth it to do so simply because dollar borrowing is on average cheaper. But this leaves open the question of where the UIP violation comes from in the first place. … Thus the primitive in our story is the share of internationally-traded goods invoiced in dollars, which in turn drives the demand for safe dollar claims; the UIP violation then emerges endogenously as the equilibrium ‘price’ required to bring supply into line with demand.”

Only when the dollar invoice share is large enough relative to the externally supplied stock of safe dollar claims; and the gap must jump to a strictly positive level first, because even the first unit of currency conversion is costly in proportion to (E-bar minus 1). Proposition 2 establishes the threshold condition: mismatch requires the dollar share of the money aggregator to exceed E-bar·X-dollar/(gamma_L·N + E-bar·X-dollar), which the authors gloss as “if the dollar invoice share is large enough relative to the supply X-dollar of safe dollar claims available from abroad, the bank will necessarily get drawn into the business of manufacturing dollar deposits backed by local-currency projects, which in turn requires the rate of return on these dollar deposits to be lower than that on own-currency deposits” (Section 2.3). On the discreteness: the UIP deviation “has to become significantly positive – in particular, it has to reach a value of theta(E-bar minus 1)/(gamma_L·N + E-bar·X-dollar) – before the banks start using local-currency collateral to back dollar claims. This is because the cost of doing even the first unit of this kind of currency conversion is discretely positive, and is proportional to (E-bar minus 1), which is effectively a proxy for the variability of the exchange rate.” The economics the authors emphasize is the supply curve: “as worldwide demand for safe dollar claims expands, we exhaust the supply that can be provided by low-cost producers (the U.S. Treasury, and firms that naturally have dollar-denominated revenues) and therefore must turn to less efficient, higher cost producers, namely firms that have to take on currency risk.”

Purely for financing reasons: sticky dollar prices make its future dollar revenues more predictable, which is better collateral for cheap dollar borrowing – and the first-order condition makes the dollar-invoiced share proportional to the UIP gap. Exporters choose the fraction eta of convertible projects to invoice in dollars at a quadratic cost interpretable as “a proxy for the risk aversion of the ultimate owners of the EM’s exporter firms,” who consume a home-currency basket (Section 3). The invoicing first-order condition is eta = (gamma_L/(beta·phi))·(Q-dollar minus Q-home), which means “as soon as the UIP deviation (Q-dollar minus Q-home) > 0, it must be that eta > 0, i.e., there is some amount of dollar invoicing by EM exporters in equilibrium,” because “the marginal cost to an exporter of doing the first unit of dollar invoicing is zero.” The authors are explicit that this is a deliberately narrow channel, complementary to existing accounts: the literature “sets aside financing considerations and instead focuses on factors that influence the optimal degree of exchange rate pass-through into prices,” or on the vehicle-currency transaction-cost story of Rey and Devereux and Shi, whereas “in our model the only reason exporters choose to invoice in dollars is because by doing so they are able to more cheaply finance their projects.” Equilibrium eta is the intersection of an upward-sloping “Invoicing Choice” line and a downward-sloping “Dollar Premium” curve, and Proposition 3 divides the parameter space into three regions: no dollar funding, dollar claims backed only by dollar-invoiced projects, and dollar claims backed additionally by local-currency projects, that is, mismatch.

The authors take the objection seriously and answer it with collateral frictions: hedging requires posting collateral, whereas invoicing in the buyer’s currency sources the hedge from a counterparty who cannot be defaulted on. Remark 3 concedes the point in principle – “invoicing in dollars bundles together a goods-pricing decision with a risk-management decision, and in principle these two decisions could be unbundled” – and then appeals to Rampini and Viswanathan and Rampini et al., who argue “that, due to financial contracting frictions, hedging of this sort by both operating firms and financial intermediaries tends to be quite constrained,” since a forward contract requires collateral that “draws resources away from real investment activities.” The worked example is a Mexican exporter selling machines in Brazil: hedging via a derivatives dealer requires assuring the dealer that the sale will happen and that revenues will not be diverted, which for a liquidity-constrained firm “will in turn compromise its ability to do real investment.” By contrast, invoicing in dollars means “it sources its hedge from somebody (the Brazilian importer) who is already fully protected from default on the part of the exporter, because the importer does not have to turn over any cash until it receives its machines, and is not promised anything other than the machines in any state of the world.”

By letting the invoice share facing each country’s importers depend on other countries’ invoicing choices: a strong enough feedback coefficient makes the complementarity self-sustaining, so both a dollar-intensive and a dollar-light equilibrium can exist. With one large economy and a unit continuum of small open economies, the dollar invoice share facing importers in country i is a plus b times the average dollar invoicing choice of all other emerging markets, with a, b > 0 and a + b < 1 (Section 4). The authors give both parameters an interpretation: “the parameter a corresponds to the share of U.S. goods in country-i expenditures, and the parameter b corresponds to the share of goods from all other EM countries,” so “a acts as an exogenous anchor on import-invoice shares, while b serves as a feedback coefficient.” The mechanism is stated symmetrically in both directions: high dollar invoicing elsewhere raises country i’s dollar invoice share, raising demand for dollar deposits, pushing dollar rates down, which “in turn validate the original decision on the part of country-j exporters to price in dollars”; and “conversely, a less dollar-intensive equilibrium can also be self-sustaining.” Proposition 4 gives three regions in a, with multiplicity in the middle one when b exceeds an explicit bound, and the authors draw two messages: first, “as the share of EM imports from the U.S. … gradually increases from zero, we eventually must get a discrete jump in the global role of the dollar”; second, “because of these strategic complementarities, there can be some indeterminacy in the outcome when imports from the U.S. are in a middle range. This indeterminacy may leave the door open for historical factors to pin down what actually happens.”

Yes – asymmetric dominant-currency equilibria arise naturally, and for an intermediate range of the anchor parameter they are the only possible outcome. The symmetric setup is deliberate: equal external supplies of dollar and euro safe assets, equal maximal depreciation of the local currency against both (so that using local collateral is equally costly for either), and an invoicing cost that allows a diversification gain from mixing the two, governed by a parameter the authors read as “a proxy for the covariance of the dollar and euro exchange rates versus the local EM currency” (Section 5.1). Three equilibrium types can arise: no global-currency invoicing, symmetric positive invoicing in both, and “an asymmetric dominant-currency equilibrium, where exporters exclusively use only one of the two currencies … to price their exports.” The virtuous circle is stated plainly: “once one currency – say the dollar – gets a bit of an edge in invoice share, this tends to feed on itself: as more global trade is invoiced in that currency, there is more demand for it as a safe store of value. This in turn makes it a cheaper currency to borrow in, which leads exporters in search of lower borrowing costs to invoice their sales in that currency.” The authors note there is no “sharp analytical characterization of how the cut-offs line up,” and that while they display one intuitively natural ordering, “our experimentation suggests that other orderings are also possible” – with the noteworthy feature of the displayed ordering being “an intermediate range of values of a … where the only possible equilibrium is one with a single dominant currency.”

That it reproduces the specific pattern seen in the data: heavy dollar invoicing far in excess of the US trade share, positive dollar premium, dollar mismatch on bank balance sheets, and none of these for the euro. In the single-dominant-currency case, “there is positive invoicing in one of the two global currencies,” while “the euro on the other hand is not used in trade invoicing, and EM banks do not create any safe euro claims”; the dollar’s weight in importer preferences jumps sharply while the euro’s falls, and the dollar UIP gap “is always positive and exceeds” the euro’s (Section 5.2). A subtle feature is that the euro gap “is negative for lower values of a and then turns positive, but even at this point there is still no incentive to invoice in euros as long as” it stays below the diversification-adjusted threshold. The mismatch and invoicing patterns line up with the facts: “in this equilibrium the bank-exporter coalition bears a dollar currency mismatch – in the sense that dollar deposits exceed dollar-denominated collateral – while there is no euro mismatch. In addition, the dollar’s use in trade invoicing greatly exceeds the U.S. share in world trade, while that same ratio equals one for the euro. This is very much in line with the empirical evidence on trade invoicing.” The authors also report a stability finding that matters for which equilibria they emphasize: for their parameters they verified numerically that symmetric dual-currency equilibria without mismatch are unstable, whereas mismatch equilibria are always stable.

By assigning a role to history: they propose selecting, among multiple equilibria, the one that was uniquely pinned down at the most recent earlier date, and treating it as focal until parameters make it unviable. The authors are explicit about the limitation: the model “is unable to speak to which currency that will be, given that it treats the U.S. and Europe as being identical on all fundamental dimensions. Taken literally, the model says that the outcome is indeterminate” (Section 5.3). Their selection device: “any time we are faced with multiple possible equilibrium outcomes at some date t, a plausible selection mechanism would be to go back in time to the first date prior to t when one of those equilibria is uniquely pinned down by the model, and posit that it then remains as the focal equilibrium until the parameters change to the point where it is no longer viable.” Applied to the euro, the pre-1999 situation – when eurozone members had separate currencies and “the largest individual member, Germany, had a GDP only about a fifth that of the U.S.” – would have made dollar dominance the unique outcome, which then persists. The implication: “even if the European economy grows to the point where it catches up with – or even somewhat surpasses – the U.S., this may not be enough to dislodge the now-entrenched dollar,” which “might require the European economy to get substantially bigger than the U.S.” The same argument is applied to the renminbi, which “is likely to be handicapped by history.”

The pound’s pre-1914 dominance and the dollar’s rapid takeover in 1914-1924, which they interpret as a small shift in the invoicing anchor triggering a large jump. Citing Eichengreen et al. (2017), “the pound was the dominant global currency prior to World War I, with over 60% of world trade invoiced, financed and settled in pounds. This was despite the fact that the U.S. economy had already overtaken Britain as the world’s largest economy (in the 1870s) and was almost as large as Britain in world trade.” The transition “was triggered by two events: by the Federal Reserve Act of 1913 that allowed U.S. banks to deal in instruments of trade credit (also known as ‘bankers acceptances’) and by World War I, which was relatively more disruptive to Britain.” The Fed’s role was active and quantified: “a policy decision was taken to make dollar trade credit available at concessionary rates, and as a result, between 1917 and 1930 the Federal Reserve held over half of all trade acceptances.” Mapped to the model, “such a policy can be thought of as an increase in a-dollar, the exogenous component in the dollar invoice share facing other countries,” and “a small change in this parameter can lead to a large and abrupt change in the global role of the currency” (Section 6.1). The authors also note the ranking reversed again in the 1930s, when the Depression hit the US harder, before the dollar took over for good in the 1950s (fn. 24). On the renminbi, “between 2010 and 2015 the renminbi’s share as a settlement currency in China’s trade has gone from 0% in 2010 to 25% in 2015,” and it “has now surpassed the euro as the second most widely used currency in global trade finance” – while remaining “far behind other major currencies in international financial transactions unrelated to trade.”

It tests the paper’s basic premise on a cross-section of ten countries and finds a strong association, but the authors themselves call the evidence preliminary and the sample is very small. The testable implication comes from the importers’ first-order conditions: the ratio of a country’s dollar to euro deposits equals the ratio of its dollar to euro invoice shares times the inverse ratio of the two safe-claim premia (equation 42), so “if importers in country i have a greater share of their imports invoiced in dollars – relative to euros – than importers in country j, they will hold a greater share of their deposits in dollars as well.” Operationalizing it requires two auxiliary assumptions: that country-i importers bank domestically, and that holdings of US Treasuries and the like are intermediated through the banking system rather than held directly (Section 6.2). Using import invoice shares from Gopinath (2015) and the BIS Locational Banking Statistics, “for the ten countries for which both of these data are available, there is indeed a strong positive relation between the two variables, with the regression having an R-squared of 0.72.” A cleaner liability measure – restricted to loans and deposits from non-bank counterparties, to strip out interbank and wholesale funding – “cuts our sample down from ten to eight countries, but leads to a very similar picture: the R-squared of the regression is now 0.82.” The pattern is intuitive in the cross-section: “it is those countries that are geographically closer to the Eurozone, and that trade heavily with Eurozone countries – namely, Denmark, Norway, Sweden, and Switzerland – that, unsurprisingly, have both lower dollar invoicing of their imports and fewer dollar-denominated banking sector liabilities.” Sample construction is disclosed: euro countries and the US are excluded deliberately “to demonstrate that this is not simply a phenomenon of countries in a currency union using their own currency in trade and in finance,” and Brazil and India are excluded for restricting foreign-currency deposits, with the authors noting that including them “only strengthens the result.” The introduction labels all of this “some preliminary evidence which is consistent with our basic premise,” and the relationship is a correlation across a handful of countries, not an identified causal estimate.

They are the third link in the chain, asserted here and modelled in a companion paper: a central bank that must be lender of last resort to a dollarized banking system wants dollar reserves. In this paper the argument is stated informally – “given that an important role for the central bank is to act as a lender of last resort to its commercial banking system, the fact that the commercial banks’ hard-currency deposits are primarily in dollars means that the central bank will want to have stockpile of dollars so as to be able to replace any sudden loss of bank funding” – with the authors noting explicitly that “we do not model this last link in the chain formally here” but do so in Gopinath and Stein (2018) (Section 1). Section 6.3 sketches the companion model: a rare banking-crisis state in which the local currency depreciates and failing banks’ deposits must be bailed out, financed either by distortionary ex post taxes or by pre-accumulated reserves. “The key proposition is that the larger is the share of dollar (as opposed to local-currency) deposits in bank liabilities, the more the government – i.e., the central bank – will choose to rely on dollar reserves to finance bailouts, as opposed to ex post taxation,” because “holding dollar reserves effectively hedges against the currency risk associated with being a lender of last resort to a dollarized banking system.” The supporting evidence is again a small-sample correlation: “in a sample of 15 countries for which the data is available, there is a strong correlation between the dollar’s share in import invoicing, and its share in central-bank foreign exchange reserves.”

Existing accounts tie the dollar’s low return to insurance properties – country size, crisis appreciation, fiscal fundamentals and market liquidity, or monopoly power – whereas this paper ties it to trade invoicing and, distinctively, requires the marginal supplier of dollar claims to be mismatched. The authors survey the alternatives as resting on “the superior insurance properties of U.S. bonds that arise either from country size (Hassan (2013)); from the tendency of the dollar to appreciate in a crisis (Gourinchas and Rey (2010), Maggiori (2017)); from better fiscal fundamentals and liquidity of debt markets (He et al. (2016)); or from the monopoly power of the U.S. as a safe asset provider (Farhi and Maggiori (2018)),” and state their own contribution as “a distinct explanation that is tied to the invoicing role of the dollar in international trade. In our model, it is this invoicing behavior that generates the demand for dollar safe assets and importantly, that implies that the marginal supplier of dollar claims must have a mismatch of its assets and liabilities in equilibrium” (Section 1). On the invoicing side they are equally explicit that they are adding a channel rather than replacing one, setting aside pass-through and vehicle-currency considerations to isolate the financing motive. The only contemporaneous work they identify as also joining invoicing and finance is Chahrour and Valchev (2017), “who focus on the medium of exchange role of currencies”; adding a medium-of-exchange role, they argue, “only strengthens our conclusions.”

Exchange rates are exogenous with zero expected appreciation, so the results are statements about average interest differentials, not about high-frequency UIP dynamics; and the UIP violations modelled here need not imply violations of covered interest parity. Remark 1 states that taking exchange rates as exogenous “is not important for our key conclusions,” because the first-order conditions “fundamentally pin down the net-of-exchange-rate expected returns,” and that an alternative with active monetary policy would preserve the same UIP violations while letting rates move – with the dollar then “expected to weaken going forward so as to maintain the same relative expected return.” Either way, “we would still be making the same statements about on-average interest-rate differentials – i.e. rate differentials when monetary policy in both countries was at its neutral level.” The authors add that “either version of our model is silent with respect to any higher frequency aspect of UIP violations such as the forward premium puzzle,” and that “instead, we are interested in on-average cross-country rate of return differentials, of which we take the ’exorbitant privilege’ to be a leading example” (fn. 11). On CIP: because what savers pay for is a certain payoff in a given currency and they are “indifferent between a dollar deposit that pays out one dollar for sure and a synthetic dollar deposit” built from a local deposit plus a forward, “this indifference on the part of depositors will tend to enforce covered interest parity” (fn. 10).

That dominance is resilient in the medium run precisely because it is self-reinforcing, but that when it does break the break may be abrupt. The conclusion states both halves: “the model suggests that the dollar’s dominance is likely to be quite resilient in the medium run, even in the face of rapid growth in global exports from other leading economies like those of Europe and China. However, in the longer run, if the gap between the U.S. and one of these other economies widens far enough, the dollar may potentially fall off the world stage to a very substantial extent, much as the British pound sterling did in the early part of the 20th century. In other words, change may be slow to come, but when it finally does, the forces in our model suggest that the change may well be quite dramatic in magnitude” (Section 7). The authors also note a third possible destination: with enough growth in both the US and Europe, both anchor shares could become large enough that “the only possible outcome is a symmetric one” with both currencies used for invoicing (Section 5.3). These are conditional statements about the model’s equilibrium structure, not forecasts with dates attached.