Dominant Currency Paradigm

They fail because they assume prices are sticky in either the producer’s or the destination’s currency, whereas trade is overwhelmingly priced in a few third currencies; DCP replaces them with dominant-currency pricing plus variable markups plus imported inputs. The first-generation New Keynesian literature – Mundell and Fleming, Svensson and van Wijnbergen, Obstfeld and Rogoff (1995) – assumes producer currency pricing, under which “the law of one price holds and a nominal depreciation raises the price of imports relative to exports (the terms-of-trade) thus improving competitiveness.” Because “there is, however, pervasive evidence that the law of one price fails to hold,” Betts and Devereux and Devereux and Engel developed local currency pricing, under which a depreciation instead worsens the terms of trade. The authors argue recent invoicing evidence “questions the validity of both approaches”: “the vast majority of trade is invoiced in a small number of ‘dominant currencies,’ with the U.S. dollar playing an outsized role,” exporters “price in markets characterized by strategic complementarities in pricing that give rise to variations in desired mark-ups,” and “most exporting firms employ imported inputs in production, reducing the value added content of exports,” whereas workhorse models “assume constant demand elasticity and/or abstract from intermediate inputs” (Section 1). Under DCP, “firms set export prices in a dominant currency (most often the dollar) and change them infrequently. They face strategic complementarities in pricing, and there is roundabout production using domestic and foreign inputs.”

With prices fully rigid in their currency of invoicing, the three paradigms give three different, non-overlapping predictions for the terms of trade: +1 under PCP, −1 under LCP, and exactly 0 under DCP. Proposition 1 writes bilateral import price and quantity pass-through as functions of the invoicing shares and the bilateral and dollar exchange rates, and then evaluates them at each paradigm’s corner. Under PCP the bilateral terms of trade move one-for-one with the exchange rate; under LCP they move one-for-one against it; under DCP “∆tot = 0” and quantities respond to the dollar exchange rate rather than the bilateral one (Section 2.4). The authors note the prediction is shock-agnostic in the very short run: “the predictions for prices, when prices are yet to change, do not depend on what drives the exchange rate variation, that is, whether it arises from monetary policy shocks, financial shocks or other shocks.” The economics they emphasize is that DCP’s stable terms of trade is not a story of muted relative price movements: “the terms-of-trade stability under DCP is associated with volatile movements of the relative price of imported to domestic goods for non-dominant (currency) countries,” volatility “driven by fluctuations in the value of the country’s currency relative to the dominant currency, regardless of the country of origin of the imported goods. Consequently, demand for imports depends on the value of a country’s currency relative to the dominant currency” (Section 1).

That a uniform depreciation against the dollar reduces trade between non-US countries, because every periphery country’s imports become more expensive while none of them gains export volume. The derivation aggregates bilateral import volumes over the set of country pairs in which the US is neither origin nor destination, weighting by each pair’s share of non-commodity import value, and yields a strictly negative response of rest-of-world trade to a uniform dollar appreciation (Section 2.4, equation 19). The comparison is the point: “Under either PCP or LCP, the growth of the rest-of-the-world trade is instead zero, either because bilateral non-dollar exchange rates are unchanged (under PCP) or because there is no bilateral pass-through (LCP).” The intuition given earlier: “Under DCP, a strengthening of dominant currencies relative to non-dominant ones is associated with a decline in imports across the periphery without a significant increase in exports to dominant currency markets, thus negatively impacting global trade. In contrast, in the case of PCP, the rise in competitiveness for the periphery generates an increase in exports. … In the case of LCP, both the import and export response is muted so the impact on global trade is weak.”

Chained Fisher bilateral price and volume indices built from UN Comtrade HS 6-digit annual customs data, restricted to non-commodities, covering more than 2,500 dyads and 91 percent of world trade, merged with World Bank, FRED and Gopinath (2015) invoicing shares. The construction starts from “detailed annual customs data for a large set of countries at the HS 6-digit product level with information about the destination country, dollar value, quantity, and weight,” from which volume changes and unit values are computed and then “chained Fisher price indices to aggregate up from the product level to the bilateral country level” (Section 3.1). Commodities are excluded and defined deliberately broadly: “Given the inherent difficulty in drawing a line between commodities and non-commodities, we define commodities fairly broadly as HS chapters 1-27 and 72-83.” Both unweighted and trade-weighted specifications are reported throughout, with weights given by each dyad’s average share of world non-commodity trade value over 1992-2015. Standard errors are clustered by dyad. The authors note the limitation the second dataset is meant to address: “the global evidence has the virtue of covering 91% of world trade, [but] it lacks granularity and the indices are at the annual frequency.”

Bilateral terms of trade are essentially uncorrelated with bilateral exchange rates – a contemporaneous coefficient of 0.037 unweighted – which rules out both PCP and LCP for the average country. Table 2 reports 0.0369 (standard error 0.00863) in the unweighted specification and 0.0813 (0.0235) trade-weighted, with “the 95% confidence interval equal[ling] [0.02, 0.05] in the unweighted regression, and [0.04, 0.13] in the weighted regression”; lag coefficients “are also small in magnitude,” and controlling for relative producer prices shrinks the point estimates further toward zero while “confidence intervals remain narrow” (Section 3.2). The authors head off two alternative readings. First, “although the lack of correlation could in principle be consistent with a world of 50% PCP and 50% LCP, the next subsections refute that possibility,” and “while the lack of correlation is consistent with any currency being a dominant currency, we provide evidence next that the major dominant currency is indeed the dollar.” Second, a flexible-price model with strong strategic complementarities (Atkeson and Burstein; Itskhoki and Mukhin) would also deliver stable terms of trade, but it is rejected “because, as we show next, the import pass-through into destination country prices at-the-dock is high, contrary to the presence of strong complementarities in pricing.” They also note attenuation bias is not a concern since exchange rates are precisely measured.

It changes it enormously: the bilateral coefficient falls from 0.76 to 0.16 unweighted (0.77 to 0.34 weighted) while the dollar coefficient picks up nearly all of the effect. The benchmark bilateral-only specification implies that “when country j’s currency depreciates relative to country i by 10%, import prices in country j rise by 8%, suggestive of close to complete pass-through at the one year horizon,” which with year fixed effects “should be interpreted as fluctuations in excess of world annual fluctuations” (Section 3.3). Adding the dollar exchange rate “sharply reduces the relevance of the bilateral exchange rate,” and the introduction reports the dollar coefficient at 0.78, which “largely dominates that of the bilateral exchange rate.” The authors stress that time fixed effects rule out an obvious confound: “the apparent dominance of the dollar cannot be an artifact of special conditions that may apply in times when the dollar appreciates or depreciates against all other currencies, for example due to global recessions or flight to safety in asset markets.” They characterize this as revealing “a potential misspecification in the standard pass-through regressions that ignore the role of the dollar” (Section 1).

The dollar effect is systematically larger where more imports are dollar-invoiced – 3.5 to 7.6 percentage points more contemporaneous pass-through per 10 percentage points of invoicing share – which ties the finding to the mechanism rather than to something else about dollar movements. Interacting both exchange rates with the importer’s country-level dollar invoicing share gives a role that is “economically and statistically significant,” and “depending on whether we use trade weights or not, the regression results indicate that increasing the dollar invoicing share by 10 percentage points causes the contemporaneous dollar pass-through to increase by 3.5-7.6 percentage points,” while “the R2 values of the panel regressions are substantially improved by adding the invoicing share interaction terms” (Section 3.3). The illustration uses three real cases: Switzerland with a dollar share of 0.13, Turkey at 0.59, and Argentina at 0.88, with dollar pass-through “highest for Argentina with the largest dollar invoicing share and the least for Switzerland with its low dollar share.” An important measurement caveat is stated: “we do not have data on the fraction of bilateral trade invoiced in dollars, so we use the importer’s country-level share as a proxy.” The authors also report that the pattern is not confined to emerging markets: “although flows between emerging markets exhibit stronger dollar dominance, our results are not limited to flows involving emerging markets.”

The dollar elasticity is an order of magnitude larger than the bilateral one, but the authors are explicit that these are predictive relationships, not structural trade elasticities. The contemporaneous dollar elasticity is “about -0.19 to -0.13 across specifications, while the elasticity for the bilateral exchange rate is an order of magnitude smaller,” and unlike the price regressions “the interactions of exchange rate changes with the importer’s dollar invoicing share are mostly imprecisely estimated here” (Section 3.4). The interpretive caveat is unusually direct: “These regressions do not capture structural demand elasticity parameters, since we do not attempt to control for all relevant relative prices, and the importer’s GDP growth is an imperfect proxy for the level of import demand. In particular, we cannot simply add importer x year fixed effects since these would absorb the dollar exchange rate. Hence our results will invariably conflate expenditure switching and shifts in aggregate import demand. The correct interpretation is to view these regressions as predictive relationships that may inform potential structural estimation exercises.” One finding is also sample-dependent and flagged as such: in the full sample the effect on the level of volume “is essentially neutral at horizons of 1-2 years,” which the authors attribute to lagged adjustment of domestic prices, but “this particular finding is driven by the early years in our sample, as results on the 2002-2015 subsample point toward a large and persistent negative effect of dollar appreciations on the volume of bilateral trade.”

Because almost all US trade is dollar-invoiced, pass-through into US export prices in the destination currency is complete while US import prices barely move, and US import volumes are essentially unresponsive to exchange rates. “Consistent with the very high fraction of U.S. exports and imports being invoiced in dollars (97% and 93%, respectively), bilateral exchange rate pass-through into prices is 100% on impact for U.S. exports and close to zero for U.S. imports” (Section 3.5). On volumes, the paper interacts everything with an indicator for the US being the importer: “when the importing country is not the U.S., the within-year bilateral trade volume response is estimated at -0.12% (unweighted) following a 1% depreciation of the importer currency,” whereas “we find U.S. imports to be completely insensitive to the bilateral exchange rate on impact, with an implied contemporaneous import volume response of 0.003% following a 1% depreciation of the dollar,” and “the difference between the contemporaneous import elasticity for the U.S. vs. that for the rest of the world is highly significant.” The policy-relevant reading: “the data indicates that U.S. trade balance adjustment following exchange rate movements occurs primarily through exports rather than imports, a consequence of the predominance of dollar invoicing in U.S. trade.”

By aggregating the richest bilateral panel specification – with bilateral, dollar and euro exchange rates all interacted with the relevant invoicing shares – up to a weighted average across all non-US trading pairs, with global business and financial cycle proxies replacing time fixed effects. Because the object of interest is a dollar move against every other currency, “we do not control for time fixed effects. Instead, we control for several proxies for the global business cycle” plus lags of importer real GDP growth (Section 3.6, equation 22). The aggregation assumes the trade elasticity is “heterogeneous across importers but homogeneous across exporters,” so the rest-of-world response can be read off the estimated equation at the import-weighted average dollar invoicing share, which “fluctuates little around a mean of 0.40 in the 2002-2015 sample.” The result: “A 1% ceteris paribus dollar appreciation leads to a 0.6% contraction in rest-of-world trade volume within the year (regardless of whether we use unweighted or trade-weighted regressions), and this contractionary effect persists out to at least two years.” The authors immediately add the qualifier: “While our regression specification cannot be interpreted structurally, the magnitude of the predictive effect underscores the importance of the dollar’s role in world trade.” A companion country-level exercise finds “the average pass-through of the dollar into CPI (resp., PPI) to be 11% (resp., 28%) within the year, and is higher for countries with a higher dollar invoicing share of imports.”

Robust to dropping the global financial crisis and to controlling for the euro exchange rate, with the rest-of-world effect stronger before the crisis. “The estimated average exchange rate pass-through and trade elasticity computed on the 1992-2007 sample are almost identical to our baseline,” and for the uniform dollar appreciation exercise “we find even stronger effects during the pre-crisis period 2002-2007.” On the euro: “the euro exchange rate is much less quantitatively important than the dollar exchange rate in price and volume regressions” – which matters because it distinguishes a dollar-specific invoicing mechanism from a general “big-currency” effect (Section 3.7). The authors also report robustness to adding importer PPI and GDP growth as controls, and note they avoid a vector error correction approach because prior work finds “VECM results to be highly unstable across specifications,” a problem “likely to be compounded by measurement error in our bilateral data.”

Granularity at quarterly frequency and at the firm-product-country level, ruling out composition effects, plus a setting where the invoicing shares are directly observed and the model can be estimated. Colombia is chosen as “a small open economy that is representative of emerging markets in its heavy reliance on dollar invoicing with 98% of exports invoiced in dollars,” and prices and quantities are defined “at the firm-10-digit product-country (origin or destination)-quarter (or year) level for manufactured goods (excluding petrochemical and basic metal industries)” (Sections 1 and 4). The terms-of-trade evidence is instructive on the commodity caveat: the peso is a commodity currency, and for the overall terms of trade the correlation with the peso-dollar rate is 0.62 with a regression coefficient of 1.15 and an R-squared of 0.38 – which looks like PCP – whereas “if we focus instead on the non-commodity terms-of-trade … the terms-of-trade is far more stable with a regression coefficient of 0.33,” consistent with DCP (Section 4). Firm-level pass-through matches the paradigm’s dynamic signature: “all pass-throughs start out close to one and decline slowly over time,” with export prices to dollar destinations at 0.84 contemporaneously falling to 0.56 after two years, import prices from dollar origins around 1 declining to 0.8, and import prices from non-dollar origins starting near 0.87 and falling to 0.49 after two years. The annual firm-level regressions “re-confirm the findings in Section 3 of the important role of the dollar in pass-through regressions,” and on quantities “the relevant exchange rate is the peso/dollar exchange rates as opposed to the bilateral exchange rate” for both imports and exports from non-dollarized partners.

That the estimated model is essentially DCP and matches the observed dynamics of pass-through, and that all three DCP ingredients are needed quantitatively – removing complementarities and imported inputs halves four-quarter export pass-through. The estimated model is driven by three shocks – productivity, an oil endowment shock standing in for commodity prices, and shocks to the dollar-rest-of-world exchange rate – with a reduced-form link between the two real exchange rates so that their effects can be separated (Section 4.2). “The estimated model is very close to DCP. The export invoicing shares for Colombia are measured in the data directly and is 100% for exports to $ and 93% for exports to R. The estimated import invoicing shares are 100% for imports from $ and 93% for those from R.” The identification comes from the shape of the impulse responses: “PCP implies low initial pass-through into export prices, which then gradually increases over time, as prices are sticky in the exporting currency. LCP implies low pass-through into import prices, which then increases over time,” and neither matches. The model also reproduces the regression pattern – bilateral rates matter alone but “drop significantly as a predictor of prices once the dollar exchange rate is also included.” On the necessity of the other two ingredients: “Under our benchmark DCP specification we find, in line with the data, the export pass-through at four quarters to both dollar and non-dollar destinations to be 65%. Instead, when we shut down strategic complementarities and imported input use, the predicted pass-through declines by half to 30%” (Section 1), and “strategic complementarities in pricing and imported input use in production are important factors controlling the (slow) dynamic of price pass-throughs.”

Quarterly, with the elasticity of substitution set to 2 and the markup elasticity to 1 from the micro pass-through literature, a home-bias share of 0.7, and standard rigidity parameters. The demand system follows Klenow and Willis, with a super-elasticity parameter that generates strategic complementarities when positive and nests the constant-elasticity case at zero. On the elasticity of substitution, the authors calibrate to an average of available estimates: “Broda and Weinstein (2006) obtain a median elasticity estimate of 2.9 for substitution across imported varieties, while Feenstra et al. (2010) estimate a value close to 1 for the elasticity of substitution across domestic and foreign varieties. Thus, we set sigma = 2.” On complementarities, “we rely on estimates from the micro pass-through literature that converges on very similar values” and set the markup elasticity to 1. “The home bias share is set to 0.7. This implies steady-state spending on imported goods in the consumption bundle and intermediate input bundle equal to thirty percent,” the wage rigidity parameter is set to 0.85 “corresponding roughly to a year and a half average duration of wages,” and the intermediate input share is two-thirds (Section 2.5.2, Table 1). Stationarity in the small open economy is induced by the standard Schmitt-Grohé and Uribe debt-elastic interest rate device, which the authors note “delivers almost identical results” for the one-time shocks they study.

The inflation-output trade-off worsens: a rate cut buys less output for more inflation than under PCP, because the depreciation passes into import prices without generating an export expansion. Following a 25 basis point cut, the exchange rate depreciates by around 0.8 percent, and “the increase in inflation in the case of DCP and PCP far exceeds that of LCP since exchange rate movements have a smaller impact on the domestic prices of imported goods when import prices are sticky in local currency” (Section 2.5.3). The terms of trade barely move under DCP while depreciating almost one-for-one under PCP and appreciating almost one-for-one under LCP. On quantities, DCP “generates a significant decline in trade-weighted imports (0.43%), despite the expansionary effect of monetary policy, and only a modest increase in trade-weighted exports (0.1%).” Output expansion “is muted under DCP relative to PCP, with the lowest impact under LCP,” because “under DCP, there is an expenditure switching effect from imports towards domestic output that is absent under LCP, while DCP misses out on the expansionary impact on exports under PCP.” The paper reports the ratio of the inflation response to the output response as 0.4 under DCP, against “0.2/0.8 = 0.25” under PCP and “lowest for LCP at 0.07.” Markups rise under DCP – though by less than under LCP, “because of the increase in marginal costs arising from the higher price of imported inputs” – and fall under PCP.

Tightening in the dominant-currency country transmits strongly abroad and shrinks both rest-of-world and world trade; tightening in a non-dominant country does neither. With three symmetric economies differing only in the dollar’s role in international pricing and bond markets, a 25 basis point tightening in the dominant country U depreciates both other currencies by 0.65 percent against the dollar. In U itself, output falls 0.6 percent and consumption 0.36 percent, but “the decline in inflation is, however, negligible (in contrast to PCP) because dollar pricing generates a low pass-through of the dollar appreciation into the price of imported goods” – inflation falls only 0.02 percent (Section 2.5.4). Abroad the pattern inverts: “The monetary tightening in U has a larger effect on inflation on impact in G/R (0.2%) than in U because the depreciation has high pass-through into import prices of the former countries. This in turn generates an endogenous increase in interest rates (0.15%) in G/R via the Taylor rule,” with output falling 0.03 percent and consumption 0.13 percent. Their exports to U fall 0.4 percent despite the depreciation, “because dollar prices to U change by little so there is no significant positive expenditure switching effect,” and trade between G and R falls 0.85 percent because their depreciation against the dollar “makes all imports more expensive.” Aggregate effects: rest-of-world trade falls 0.83 percent and global trade 0.73 percent. A 25 basis point tightening in non-dominant country G instead produces a 0.2 percent decline in G’s own inflation, high pass-through into G’s import prices but low pass-through into its export prices, so “there is only a small negative impact on exports from G,” and “the monetary tightening in G is associated with an expansion in global trade and almost no effect on rest-of-world trade.”

It improves on it by using bilateral rather than trade-weighted exchange rates, excluding commodities, and estimating pass-through coefficients rather than correlations. Obstfeld and Rogoff (2000) “conduct one of the earliest tests of the Mundell-Fleming paradigm against the Betts-Devereux-Engel paradigm,” examining “the correlation between country-level terms of trade and the trade-weighted exchange rate for 21 countries, using quarterly data for 1982-1998,” reporting “an average correlation of 0.26, which they interpret as a rejection of local currency pricing” – while themselves conjecturing that the low correlation “could be because of the construction of the trade-weighted exchange rates and/or because their terms of trade measures include commodity prices.” This paper takes up both conjectures: “we examine the bilateral terms of trade, excluding commodity prices and we estimate pass-through coefficients as opposed to correlations. Moreover, we test additional predictions of the different pricing paradigms” (Section 1). The Colombian evidence confirms the commodity conjecture concretely, since the overall terms-of-trade coefficient there is 1.15 while the non-commodity coefficient is 0.33.

Invoicing is exogenous in the model, but the authors argue their empirical predictions survive endogenizing it, because firms choose the currency in which their reset prices are most stable. The limitation is stated in the conclusion: “Our framework takes the invoicing currency choice as given. Yet we have been careful to point out that most of our results would hold even with endogenous currency invoicing.” Two reasons are given. First, “some ingredients from our model, namely imported input use in production and strategic complementarities in pricing, are precisely those that would give rise endogenously to dominant currency in pricing,” as shown by Gopinath, Itskhoki and Rigobon in partial equilibrium and Mukhin in general equilibrium. Second, “firms choose to price in currencies in which their reset prices are most stable, i.e., the desired medium-run pass-through into prices (expressed in the invoicing currency) is low. In other words, our empirical findings will continue to be relevant in an environment with endogenous currency choice” (Section 5 and fn. 8).

Because exchange rates in the data are driven largely by an unexplained residual, so the ideal test has too little power. They state the ideal test and the obstacle together: “The ideal test would be to examine the joint response of exchange rates, prices, and quantities to an exogenous shock such as a monetary policy shock. The problem is that in the data exchange rate fluctuations have little to do with monetary policy shocks or other identified policy shocks. Instead exchange rates appear to be driven by a ‘residual’ that the literature names ‘financial shocks.’ Practically this shows up as low power in testing the channel from identified exogenous shocks to exchange rates and to trade” (Section 2.4). This is why the very-short-run predictions, which are shock-agnostic when prices have not yet adjusted, carry so much of the paper’s identification weight, and why the longer-horizon evidence leans on the structural estimation in the Colombian section instead.

That the dominant-currency country is insulated from the inflationary consequences of its own exchange rate while exporting them, and that a small open economy under DCP cannot hit both its inflation and output targets. The conclusion states both sides: “Monetary policy shocks in the dominant currency country also have strong spillovers to the rest of the world, while the converse is not true: the dominant currency country is largely insulated from the inflationary consequences of fluctuations in its currency, which are absorbed instead into prices and trade in the rest of the world.” And on optimal policy: “under DCP, a small open economy’s optimal monetary policy is no longer able to attain both zero producer price inflation and zero output gap in circumstances where producer currency pricing would,” with the fuller treatment referred to Casas et al. (2016). The authors place the result in a wider pattern: “the dominance of the U.S. dollar is pervasive, from the structure of external balance sheets …, the currency composition of private portfolios …, the choice of anchor currency … and trade invoicing, with important and complex interactions which we are only starting to explore.”