1 Introduction

Exchange rate fluctuations impact a country’s trade competitiveness, inflation, and output and therefore have important consequences for its welfare and economic policy. It is common practice to estimate this impact by examining the pass-through of bilateral or trade-weighted exchange rates into export and import prices and volumes. This practice follows naturally from the classic Mundell-Fleming paradigm of sticky prices and producer currency pricing whereby exporting firms infrequently change prices denominated in their home currency. According to this paradigm, the price an importing country faces when expressed in the importing country’s currency fluctuates closely with the bilateral exchange rate. Accordingly, studies of exchange rate pass-through focus on trade-weighted or bilateral exchange rate changes (see Goldberg and Knetter (1997) and Burstein and Gopinath (2014) for surveys).

Recent evidence by Goldberg and Tille (2008) and Gopinath (2015) question the Mundell-Fleming assumption of producer currency pricing, as the vast majority of trade is invoiced in a small number of “dominant currencies”, with the U.S. dollar playing an outsize role. Accordingly Casas et al. (2016) develop a “dominant currency paradigm” where trade prices are sticky in dollars and demonstrate that the value of a country’s currency relative to the dollar is a primary driver of a country’s import prices and quantities regardless of where the good originates from, even at annual frequencies. They provide empirical evidence supporting these predictions using customs data for Colombia. However, no evidence exists on the the importance of dominant currencies for global trade. Such an analysis requires trade price and volume indices at the level of country-pairs, that excludes commodities, and covers many countries. This data is not readily available.

In this paper we contribute to the literature by first constructing annual trade price and volume data for country pairs broken down by commodities/noncommodities. Specifically, we construct harmonized annual bilateral import and export unit value and volume indices for 55 countries (yielding more than 2,500 dyads, i.e., trading pairs) from highly disaggregated UN Comtrade data starting as early as 1989, depending on the country, and covering through 2015. The countries in our sample comprise 91% of the world’s total goods exports and imports in 2015.

We then employ this data, merged with currency invoicing data from Gopinath (2015), to examine the implications of the dominant currency paradigm for a large number of countries. In order to estimate the cross-sectional heterogeneity of pass-through and its relation to dollar invoicing, we employ a novel Bayesian semiparametric hierarchical panel data model.

Our empirical results reject the Mundell-Fleming benchmark in favor of the dominant currency paradigm by establishing the following facts: 1) The dollar exchange rate quantitatively dominates the bilateral exchange rate in price pass-through and trade elasticity regressions. U.S. monetary policy induced dollar fluctuations have high pass-through into bilateral import prices. 2) Bilateral non-commodities terms of trade are essentially uncorrelated with bilateral exchange rates. 3) The strength of the U.S. dollar is a key predictor of rest-of-world aggregate trade volume and consumer/producer price inflation. Specifically, a 1% U.S. dollar appreciation against all other currencies in the world predicts a 0.6−0.8% decline within a year in the volume of total trade between countries in the rest of the world, holding constant various proxies for the global business cycle. 4) The importer’s share of imports invoiced in dollars explains about 15% of the variance of dollar pass-through/elasticity across country pairs. These results derive from fixed effects panel regressions as well as a novel Bayesian semiparametric hierarchical panel data model. We now elaborate further on each of these findings.

First, we introduce the U.S. dollar exchange rate as an additional explanatory variable in price pass-through and trade elasticity regressions. In standard price pass-through regressions, we document that when country j’s currency depreciates relative to country i by 10%, import prices in country j for goods imported from country i rise by 8%, suggestive of close to complete pass-through at the one year horizon. However, adding the dollar to the regression knocks the coefficient on the bilateral exchange rate from 0.80 down to 0.32 in the case when exporter-reported prices are used, and from 0.76 to 0.16 when importer-reported prices are used. The coefficient on the dollar exchange rate of 0.66 (0.78) dominates that of the bilateral exchange rate. This result cannot be attributed to the correlation of the dollar with the global business cycle, as we control for time fixed effects. Similarly, adding the dollar exchange rate to a bilateral volume forecasting regression knocks down the coefficient on the bilateral exchange rate by a substantial amount. The contemporaneous volume elasticity for the dollar exchange rate is around −0.20 across specifications and data sources, while the elasticity for the bilateral exchange rate is an order of magnitude smaller at −0.03. We also show that the dollar’s role as an invoicing currency is indeed special as it handily beats the explanatory power of the euro in price and volume regressions.

To evaluate the impact of U.S. monetary policy shocks on bilateral trade, we isolate daily dollar exchange rate changes on scheduled U.S. monetary policy announcement dates. As theory predicts, we document that a tightening of U.S. monetary policy is associated with a nominal appreciation of the dollar. Using only this subset of exchange rate changes in our regressions, we estimate a high pass-through of the dollar exchange rate into bilateral import prices. The impact of U.S. monetary shocks also significantly increases with the share of a country’s imports invoiced in dollars.

Second, the previous finding that prices respond to the dollar exchange rate but not the bilateral exchange rate would imply that the terms of trade should not be sensitive to the bilateral exchange rate, in sharp contrast to the Mundell-Fleming paradigm. More generally, if bilateral trade is invoiced in a dominant currency (for some countries this could be the euro), we should obtain low sensitivity to bilateral exchange rate changes. We indeed find this to be the case. In a regression of changes in the bilateral non-commodities terms of trade on changes in the bilateral exchange rate and the bilateral relative producer price index (PPI), the contemporaneous coefficient on the exchange rate is indistinguishable from zero, with the 95% confidence interval equal to [−0.013, 0.037] for data reported by exporters, and [−0.017, 0.035] for data reported by importers. For comparison, the coefficient should be 1 under producer currency pricing and − 1 under local currency pricing (i.e., pricing in the currency of the destination market), given fully rigid nominal prices.

Third, we demonstrate empirically that the strength of the U.S. dollar is a key predictor of rest-of-world trade volume and inflation, even controlling for measures of the global business cycle. We find that a 1% appreciation of the U.S. dollar relative to all other currencies is associated with a 0.6−0.8% contraction in rest-of-world (i.e., excluding the U.S. entirely) aggregate import volume within the year. Furthermore, countries with larger dollar import invoicing shares experience higher pass-through of the dollar exchange rate into consumer and producer price inflation. Whereas these findings are puzzling from the perspective of the traditional Mundell-Fleming model due to its emphasis on bilateral exchange rates, they emerge naturally in the dominant currency paradigm.

Fourth, we exploit our rich panel data set to show that the cross-dyad heterogeneity in pass-through coefficients is well explained by the propensity to invoice imports in dollars. We use the importer’s country-level dollar invoicing share from Gopinath (2015) as a proxy for the invoicing share of bilateral imports.¹ Standard panel regressions with interactions show that increasing the dollar invoicing share by 10 percentage points causes the contemporaneous dollar price pass-through to increase by 2.0−3.5 percentage points. Using a flexible hierarchical Bayesian framework to directly model pass-through heterogeneity, we estimate that the importer’s dollar invoicing share explains 15% of the overall cross-dyad variance in dollar exchange rate pass-through into prices. We also find that the importer’s dollar invoicing share affects the exchange rate elasticity of trade volumes. These findings confirm the quantitative importance of the global currency of invoicing, a key concept in the dominant currency paradigm.

The Bayesian estimation procedure we develop allows the data to speak flexibly about the extent and determinants of the cross-sectional heterogeneity of pass-through. We employ a random coefficients panel data model, where the distribution of the random coefficient on the dollar exchange rate is allowed to depend nonparametrically on the dollar invoicing share as well as unobserved determinants of pass-through heterogeneity. Unlike standard panel regressions that can be informative about the average pass-through and the statistical significance of the determinants of pass-through heterogeneity, the Bayesian approach allows us to quantify the overall cross-sectional heterogeneity of exchange rate pass-through/elasticities and the relation of this heterogeneity to dollar invoicing. The hierarchical aspect of the Bayesian inference procedure can be thought of as striking an optimal bias-variance balance between two extreme approaches: (1) constrained but precisely estimated panel regressions with cross-sectionally constant parameters and (2) unconstrained but noisily estimated dyad-by-dyad time series regressions. Despite the high dimensionality of our semiparametric model, posterior sampling is handled efficiently and robustly by the user-friendly open source software Stan (Stan Development Team, 2016).

Given the magnitudes of our results and the global coverage of our data, we conclude that the dominant currency paradigm is a more empirically relevant starting point than traditional modeling approaches when analyzing the international transmission of shocks and optimal monetary and exchange rate policy.

Literature review. Our exchange rate pass-through analysis appears to be among the first to exploit a globally representative data set on bilateral trade volumes in addition to values. This allows us to distinguish the effects of exchange rates on volumes and prices (more precisely, unit values) at the level of country pairs. We use the cross-sectional richness of our data set to investigate the determinants of differential pass-through, especially as it relates to currency of invoicing. Our analysis appears to be the first attempt in the literature at quantifying both the overall extent and the determinants of the heterogeneity of exchange rate pass-through and elasticity across trade dyads.

To our knowledge, the only other work that utilizes a similarly rich data set is Bussière et al. (2016), who analyze trade prices and quantities at the product level. The goal of that paper is to quantify the elasticity of prices and quantities to the bilateral exchange rate and check if Marshall-Lerner conditions hold. In contrast, our goal is to understand the prominence of dominant currencies in international trade by comparing the importance of dollar and bilateral exchange rates and by exploiting data on currency invoicing.

The remaining literature on exchange rate pass-through falls into two main camps.² First, many papers have used unilateral/country-level time series, which limits the ability to analyze cross-sectional heterogeneity and necessitates the use of trade-weighted rather than truly bilateral exchange rates (e.g., Leigh et al., 2015). Second, a recent literature has estimated pass-through of bilateral exchange rates into product-level prices (as opposed to unit values), but these micro data sets are only available for a small number of countries (see the review by Burstein and Gopinath, 2014).

Our paper confirms that the findings of Casas et al. (2016) are relevant for the majority of world trade, and we establish additional channels of U.S. dollar dominance. Casas et al. (2016) model the dominant currency paradigm and test its implications for the dollar’s role in pass-through regressions and for the terms of trade using product level data for Colombia. We depart from this paper by providing evidence for the dominant currency paradigm using newly constructed import and export price index data for 55 countries and over 2,500 country-pairs. Moreover, we relate the heterogeneity in pass-through estimates to the dollar invoicing shares of countries, estimate the impact of U.S. monetary policy shocks on bilateral import price pass-through, and demonstrate that the strength of the dollar is a key predictor of rest-of-world trade and consumer/producer price inflation.

The findings in this paper help shed light on the literature on producer (Obstfeld and Rogoff (1995), Corsetti and Pesenti (2001), Galí (2008)) versus local currency pricing (Betts and Devereux (2000), Devereux and Engel (2003), Engel (2011)). Corsetti et al. (2010) provide an in-depth analysis of both. By demonstrating that global trade data is less consistent with these frameworks than with dominant currency pricing, our results open up new avenues for research into the consequences and determinants of currency invoicing under sticky prices.

From a methodological perspective, our paper’s contribution is to introduce a semiparametric Bayesian panel data model with cross-sectionally heterogeneous slope coefficients. Our specification of the cross-sectional distribution of slope coefficients relies on the Mixture of Gaussian Linear Regression nonparametric conditional density prior in Pati et al. (2013), who derive high-level posterior concentration results. The MGLR prior extends the much-used Dirichlet Process Mixture density prior to conditional density estimation.³ Liu (2017) uses the MGLR prior specification but places it on the unit-specific panel intercepts (rather than slopes), and she focuses on forecasting rather than characterizing cross-sectional heterogeneity. Although our linear-in-parameters specification is more restrictive than the frequentist non-parametric approaches of Evdokimov (2010) and Chernozhukov et al. (2013), our Bayesian framework facilitates visualization of the conditional pass-through distribution, uncertainty assessment, and model selection.

Outline. Section 2 discusses a simple conceptual framework that motivates our empirical analysis. Section 3 describes our data set of bilateral trade unit values and quantities, exchange rates, dollar invoicing shares, and U.S. monetary shocks. Section 4 presents panel regression evidence on the average pass-through from bilateral, dollar, and euro exchange rates (or monetary shocks) into prices, quantities, and terms of trade. In Section 5, we employ the Bayesian model to characterize the cross-sectional heterogeneity of dollar pass-through and its relation to invoicing shares. Section 6 shows that the strength of the U.S. dollar is a key predictor of rest-of-world trade volume and inflation. Section 7 concludes. Appendix A details the data set and Bayesian approach. Appendix B contains supplementary material on data, empirics, and numerical procedures.

2 Conceptual framework

In this section we provide a simple conceptual framework along the lines of Proposition 2 in Casas et al. (2016) to motivate the empirical analysis that follows. Define p_ij to be the log price of goods exported from country i to country j measured in currency j, e_ij to be the log bilateral exchange rate between country i and country j expressed as the price of currency i in terms of currency j, and e_$j to be the log price of a U.S. dollar in currency j. Suppose a fraction θ_i of these exports are invoiced in the producer’s (country i) currency, a fraction θ_j in the local (destination country j) currency and a fraction θ_u in the dominant currency (dollar) with Σ_k∈{i,j,u} θ_k = 1. Import price inflation for country j for goods originating from country i is then

where $p_{ij,t}^k$ stands for the price of goods imported by country j from i that are invoiced in the currency k. Under Calvo pricing we have, $\mathrm{\Delta }{p}_{ij,t}^k=(1-{\delta }_p)(p_{ij,t}^{-k}-p_{ij,t-1}^k)$ where $p_{ij,t}^{-k}$ is the reset-price for (ij) in currency k. Through substitution we can express import price inflation as

where $\mathrm{\Delta }{p}_{ij,t}^{-k}\equiv p_{ij,t}^{-k}-p_{ij,t-1}^k$. In the very short run when δ_p → 1, we have the following benchmarks for the changes in import prices and the terms of trade (TOT).

• In the case of producer currency pricing (PCP), θ_i = 1 and θ_j = θ_u = 0,

• In the case of local currency pricing (LCP), θ_j = 1 and θ_i = θ_u = 0,

• In the case of dominant currency pricing (DCP), θ_u = 1 and θ_i = θ_j = 0,

In summary, the predicted coefficients from a regression of (log) changes in import prices on (log) changes in bilateral and dollar exchange rates differ qualitatively across the PCP, LCP, and DCP benchmarks, as do the implications for the terms of trade.

A primitive assumption of Keynesian environments is that prices are sticky in the currency of invoicing. This assumption is supported by direct measures of price stickiness for U.S. prices-at-the dock in Gopinath and Rigobon (2008) and for Irish prices-at-the dock in Fitzgerald and Haller (2013). Cravino (2017) provides indirect evidence of price stickiness using differential sensitivity of Chilean export prices invoiced in different currencies to exchange rate shocks, similar to the evidence in Gopinath et al. (2010).

At longer horizons, some prices do adjust and the predictions depend on the response of reset prices $p_{ij,t}^{-k}$ to changes in e_ij,t. With endogenous currency choice and Calvo timing, Gopinath et al. (2010) demonstrate that firms choose to price in currencies in which its reset price is most stable, i.e., its desired pass-through conditional on a price change is low. Consequently, even at longer (annual) horizons, we expect the divergent predictions across PCP, LCP, and DCP to hold.

3 Data

The core of our data set consists of panel data on bilateral trade values and volumes from Com-trade. To this global data set we append macroeconomic country aggregates from the World Bank’s World Development Indicators, financial variables from the Federal Reserve Bank of St. Louis’s FRED database, and currency invoicing shares from Gopinath (2015). Finally, we construct an annual series of U.S. monetary policy induced exchange rate changes.

Comtrade. UN Comtrade provides detailed customs data for a large set of countries at HS 6–digit product level with information about the USD value, quantity, and weight of imports and exports. This dataset makes it possible to compute volume changes over time for each product, and use the value data to infer unit values. Once unit values are calculated, we compute Laspeyres, Paasche and Fisher indices, both in their fixed base and chained forms to aggregate up from the product level. We do various aggregations, the main one being across products we classify as commodities and one for noncommodities. Given the inherent difficulty in drawing a line between commodities and noncommodities, we define commodities fairly broadly as HS chapters 1–27 and 72–83 which comprise animal, vegetable, food, mineral, and metal products. We use chained Fisher indices in the analysis that follows, except where noted.⁴

We conduct this exercise at annual frequency, and for exporter and importer reported data separately. Coverage of Comtrade at annual frequency over time and across countries is good. The longest time span of the data is 1989–2015, although the coverage varies by dyad. Appendix A.1 breaks down the coverage by country. In 2015, the 55 countries in our sample are responsible for 91.2% and 91.5% of the value of world goods imports and exports, respectively, as recorded in Comtrade. For each reporting country, we construct indices for both imports and exports. This gives us two sets of numbers for each bilateral trade flow, one as reported by the importer and one as reported by the exporter, and allows us to do robustness checks by comparing our empirical results across the two sets.

The biggest challenge for constructing price and volume indices using customs data is the so-called unit value bias, as argued by Silver (2007). Unit values, calculated simply by dividing observed values by quantities, are not actual prices. Even at the narrowly defined product categories at 6–digit product level, there is likely to be a wide range of products whose prices may not be moving proportionately. The implication is that if there are shifts in quantities traded within the narrowly defined product categories, unit values would be influenced even when there may not be any price movement. This creates a bias that the employed methodology takes a stab at correcting for by eliminating products whose unit values have a variance higher than a threshold and hence are more likely to suffer from the unit value bias.

Another challenge that arises from using Comtrade data is related to the use of different HS vintages over time. HS classification is updated about every five years to ensure that the available codings accurately reflect the variety of products being traded. This involves introducing codes for new products, eliminating the old ones, and often regrouping existing products. While concordances are readily available to facilitate the matching of HS codes across different HS vintages, this process inevitably leads to a loss of information, especially in the case of data on quantities, because the mapping of products across vintages is rarely one-to-one. To get around this problem, for the years in which there is a transition to a new HS vintage, we compute the indices twice, once under the old vintage (using concordances) and once under the new one. This way, only these transition years would be effected by the loss of information due to matching across vintages. After that year, we switch to working with the new vintage. This method not only minimizes the loss of information but also allows us to include new products in the construction of the indices. Boz and Cerutti (2017) provide further details of this method, including the strategy for dealing with outliers and missing values, and a comparison with a similar dataset constructed by Gaulier et al. (2008).

In the final stage, we compare our unit value indices to those provided by the BLS for the U.S., the only country, to our knowledge, that collects import price indices based on price surveys by origin. As shown in Appendix B.1, this comparison for the U.S. suggests that working with unit values is acceptable, as the growth rates of the two series are broadly aligned for most trading partners. Additionally, Boz and Cerutti (2017) find favorable results when comparing country-level indices with those from the WTO and IMF World Economic Outlook.

Currency invoicing share. For currency invoicing shares we use the data set constructed by Gopinath (2015). The export and import invoicing share for a country is a simple average of the respective shares for the years in which it is reported over the period 1999–2014. For all countries the invoicing shares tend to be fairly stable over time. Appendix A.1 lists the USD and euro import invoicing share for the 39 countries in our sample with available invoicing data.

U.S. monetary shocks. We compute currency-specific annual U.S. monetary policy induced exchange rate changes by summing across daily exchange rate changes on U.S. Federal Open Market Committee (FOMC) announcement dates. We use the scheduled FOMC announcement dates for 1990–2004 from Gürkaynak et al. (2005), supplemented with dates for 2005–2013 from Nakamura and Steinsson (2016), as well as dates for 2014–2015 from the FOMC meeting calendar on the website of the Board of Governors. There are a total of 219 scheduled announcement dates from 1990–2015. On each of these dates, we compute the daily log change (relative to the previous weekday) in the exchange rate vis-à-vis the U.S. dollar for each country in the Comtrade sample. Daily exchange rate series were obtained from FactSet (close mid price).⁵ To obtain an annual series for each country, we compute the unweighted sum of the log exchange rate changes on FOMC dates in a given year.

Country-level and global macro data. We use the World Bank’s World Development Indicators (WDI) database as the source for annual average exchange rates and macroeconomic data for the world and for the countries in our sample. We obtain the WTI oil price, 1–year Treasury bill rate, and VIX from the St. Louis Fed’s FRED database. See Appendix A.1 for details.

4 Bilateral pass-through, terms of trade, and the dollar

In this section we show that our global trade dataset is consistent with the dominant currency paradigm: The U.S. dollar plays an outsize role in driving international trade prices and quantities. We run fixed effects panel regressions at the dyad (country pair) level with exchange rates as the independent variable, and either prices, terms of trade, or volumes as the dependent variables. In all cases we find that bilateral (importer vs. exporter) exchange rates matter substantially less than the exchange rate vis-à-vis the U.S. dollar. The effects of the dollar are stronger when the importing country has a higher fraction of trade invoiced in dollars. The pass-through of the subset of dollar exchange rate fluctuations that are due to U.S. monetary policy actions is high. Finally, we show that the dollar’s quantitative dominance holds up in specifications that control also for the euro exchange rate.

All specifications below rely on Comtrade series that use Fisher price indices, data for non-commodities goods only, and the preferred imputation and outlier truncation technique of Boz and Cerutti (2017). In Appendix B.2 we show that our results are robust to alternative choices.

4.1 Exchange rate pass-through into prices

We first examine the pass-through of bilateral and dollar exchange rates to bilateral trade price indices. Define p_ij to be the log price of goods exported from country i to country j measured in currency j, e_ij to be the log bilateral exchange rate between country i and country j expressed as the price of currency i in terms of currency j, and e_$j to be the log price of a U.S. dollar in currency j. We estimate the following specifications,

$\begin{array}{cc}\text{Δ}{p}_{ij,t}={\lambda }_{ij}+{\delta }_t+\underset{k=0}{\overset{2}{\mathrm{\Sigma }}}{\alpha }_k\mathrm{\Delta }{e}_{ij,t-k}+\theta \prime X_{i,t}+{\varepsilon }_{ij,t}& \left(1\right)\end{array}$

$\begin{array}{cc}\text{Δ}{p}_{ij,t}={\lambda }_{ij}+{\delta }_t+\underset{k=0}{\overset{2}{\mathrm{\Sigma }}}{\alpha }_k\mathrm{\Delta }{e}_{ij,t-k}+\underset{k=0}{\overset{2}{\mathrm{\Sigma }}}{\beta }_k\mathrm{\Delta }{e}_{\$j,t-k}+\theta \prime X_{i,t}+{\varepsilon }_{ij,t}& \left(2\right)\end{array}$

$\begin{array}{lll}\text{Δ}{p}_{ij,t}=& {\lambda }_{ij}+{\delta }_t+\underset{k=0}{\overset{2}{\mathrm{\Sigma }}}{\alpha }_k\mathrm{\Delta }{e}_{ij,t-k}+\underset{k=0}{\overset{2}{\mathrm{\Sigma }}}{\eta }_k\mathrm{\Delta }{e}_{ij,t-k}\times S_j& \left(3\right)\\ & +\underset{k=0}{\overset{2}{\mathrm{\Sigma }}}{\beta }_k\mathrm{\Delta }{e}_{\$j,t-k}+\underset{k=0}{\overset{2}{\mathrm{\Sigma }}}{\psi }_k\mathrm{\Delta }{e}_{\$j,t-k}\times S_j+\theta \prime X_{i,t}+{\varepsilon }_{ij,t}& \end{array}$

where λ_ij and δ_t are dyadic and time fixed effects, respectively. X_i,t are other controls, namely the change in the log producer price index of the exporting country i measured in currency i (and two lags).

For reference, we first consider the specification without the dollar exchange rate as an explanatory variable. Regression Eq. (1) is a standard pass-through regression where bilateral import prices are regressed on bilateral exchange rates. The estimates from such a regression are reported in columns (1) and (4) of Table 1. Column (1) uses import prices of country j originating from country i as reported by country i, and column (4) uses prices reported by country j. Under perfect construction, these two should be the same, but that is not the case in practice. However, we obtain similar estimates for pass-through. According to these estimates, 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.⁶ The second and third lags are economically less important.

Table 1:

The first (resp., last) three columns use Comtrade data reported by exporting (resp., importing) countries. S.e. clustered by dyad. *** p<0.01, ** p<0.05, * p<0.1.

Exchange rate pass-through into prices

Table 1:

The first (resp., last) three columns use Comtrade data reported by exporting (resp., importing) countries. S.e. clustered by dyad. *** p<0.01, ** p<0.05, * p<0.1.

Exchange rate pass-through into prices

VARIABLES	(1) export Δpij,t	(2) export Δpij,t	(3) export Δpij,t	(4) import Δpij,t	(5) import Δpij,t	(6) import Δpij,t
Δeij,t	0.801***	0.315***	0.318***	0.757***	0.164***	0.209***
	(0.0113)	(0.0127)	(0.0177)	(0.0132)	(0.0126)	(0.0169)
Δeij,t−1	0.0762***	0.115***	0.126***	0.0674***	0.0521***	0.0270*
	(0.00675)	(0.0133)	(0.0206)	(0.00818)	(0.0123)	(0.0154)
Δeij,t−2	0.0113**	−0.0656***	−0.0863***	0.0306***	−0.0727***	−0.0721***
	(0.00489)	(0.0114)	(0.0147)	(0.00608)	(0.0127)	(0.0167)
Δeij,t × Sj			−0.0152			−0.0841***
			(0.0230)			(0.0240)
Δeij,t−1 × Sj			−0.0156			0.0319
			(0.0260)			(0.0232)
Δeij,t−2 × Sj			0.0205			−0.000922
			(0.0156)			(0.0164)
Δe$j,t		0.656***	0.522***		0.781***	0.565***
		(0.0142)	(0.0274)		(0.0143)	(0.0283)
Δe$j,t−1		−0.128***	−0.118***		−0.0737***	0.0844***
		(0.0146)	(0.0290)		(0.0157)	(0.0276)
Δe$j,t−2		0.0857***	0.143***		0.104***	0.117***
		(0.0124)	(0.0227)		(0.0146)	(0.0259)
Δe$j,t × Sj			0.196***			0.348***
			(0.0320)			(0.0326)
Δe$j,t−1 × Sj			−0.0145			−0.185***
			(0.0353)			(0.0358)
Δe$j,t−2 × Sj			−0.0716***			−0.0495*
			(0.0261)			(0.0290)
Exp. PPI	yes	yes	yes	yes	yes	yes
Time FE	yes	yes	yes	yes	yes	yes
R-squared	0.411	0.447	0.507	0.356	0.398	0.515
Observations	45,945	45,945	33,291	46,820	46,820	34,513
Dyads	2,611	2,611	1,867	2,647	2,647	1,900

View Table

Table 1:

The first (resp., last) three columns use Comtrade data reported by exporting (resp., importing) countries. S.e. clustered by dyad. *** p<0.01, ** p<0.05, * p<0.1.

Exchange rate pass-through into prices

VARIABLES	(1) export Δpij,t	(2) export Δpij,t	(3) export Δpij,t	(4) import Δpij,t	(5) import Δpij,t	(6) import Δpij,t
Δeij,t	0.801***	0.315***	0.318***	0.757***	0.164***	0.209***
	(0.0113)	(0.0127)	(0.0177)	(0.0132)	(0.0126)	(0.0169)
Δeij,t−1	0.0762***	0.115***	0.126***	0.0674***	0.0521***	0.0270*
	(0.00675)	(0.0133)	(0.0206)	(0.00818)	(0.0123)	(0.0154)
Δeij,t−2	0.0113**	−0.0656***	−0.0863***	0.0306***	−0.0727***	−0.0721***
	(0.00489)	(0.0114)	(0.0147)	(0.00608)	(0.0127)	(0.0167)
Δeij,t × Sj			−0.0152			−0.0841***
			(0.0230)			(0.0240)
Δeij,t−1 × Sj			−0.0156			0.0319
			(0.0260)			(0.0232)
Δeij,t−2 × Sj			0.0205			−0.000922
			(0.0156)			(0.0164)
Δe$j,t		0.656***	0.522***		0.781***	0.565***
		(0.0142)	(0.0274)		(0.0143)	(0.0283)
Δe$j,t−1		−0.128***	−0.118***		−0.0737***	0.0844***
		(0.0146)	(0.0290)		(0.0157)	(0.0276)
Δe$j,t−2		0.0857***	0.143***		0.104***	0.117***
		(0.0124)	(0.0227)		(0.0146)	(0.0259)
Δe$j,t × Sj			0.196***			0.348***
			(0.0320)			(0.0326)
Δe$j,t−1 × Sj			−0.0145			−0.185***
			(0.0353)			(0.0358)
Δe$j,t−2 × Sj			−0.0716***			−0.0495*
			(0.0261)			(0.0290)
Exp. PPI	yes	yes	yes	yes	yes	yes
Time FE	yes	yes	yes	yes	yes	yes
R-squared	0.411	0.447	0.507	0.356	0.398	0.515
Observations	45,945	45,945	33,291	46,820	46,820	34,513
Dyads	2,611	2,611	1,867	2,647	2,647	1,900

Our main result in this section establishes the dominance of the dollar over bilateral exchange rates in predicting trade price movements. Columns (2) and (5) report estimates from regression Eq. (2). As is evident, including the dollar exchange rate sharply reduces the relevance of the bilateral exchange rate. It knocks the coefficient on the bilateral exchange rate from 0.80 down to 0.32 in the case when exporter-reported prices are used, and from 0.76 to 0.16 when importer-reported prices are used. Instead, almost all of the effect is absorbed by the dollar exchange rate.⁷ Notice that, due to our inclusion of time fixed effects, 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. Appendix B.2 shows that our results are robust to the choice of time sample.

The cross-dyad heterogeneity in pass-through coefficients is related to the propensity to invoice imports in dollars. Columns (3) and (6) interact the dollar and bilateral exchange rates with the share of invoicing in dollars at the importer country level, as in regression Eq. (3). Notice that 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. As expected, the import invoicing share plays an economically and statistically significant role for the dollar pass-through. Depending on whether we use prices reported by exporters or importers, the regression results indicate that increasing the dollar invoicing share by 10 percentage points causes the contemporaneous dollar pass-through to increase by 2.0−3.5 percentage points. In Section 5 below we further quantify the importance of the dollar invoicing share for explaining the cross-sectional variation in pass-through.

Figs. 1 and 2 depict the regression results visually in the form of impulse response functions. Fig. 1 shows the impulse responses of the bilateral import price level. The top row uses exporter reported data, the bottom row importer reported data. The left column shows the bilateral pass-through in the specifications without the dollar exchange rate, while the right column compares the bilateral and dollar pass-throughs in specifications with both exchange rates. Fig. 2 illustrates the pass-through heterogeneity as a function of the invoicing share S_j, as implied by the regression specifications with interactions. The figure focuses on three representative countries: Switzerland (S_j = 0.13), Turkey (S_j = 0.59), and Argentina (S_j = 0.88). Although Fig. 2 suggests that dollar pass-through is high even for low dollar invoicing shares, this particular finding appears to be an artifact of using country-level dollar invoicing shares rather than bilateral shares. If we weight the pass-through regression by trade value shares, giving higher weight to dyads that are responsible for a larger share of the importer’s total imports, the bilateral pass-through is estimated to exceed the dollar pass-through when the dollar invoicing share is low (e.g., S_j = 0.13).

Impulse responses of bilateral price level to bilateral e_ij,t and USD e_$j,t exchange rates. Based on the regressions in Table 1 without interactions. Top row: exporter reported data, bottom row: importer reported. Left column: specifications (1) and (4), right column: specifications (2) and (5). Error bars: 95% confidence intervals, clustering by dyad.

Citation: IMF Working Papers 2017, 239; 10.5089/9781484327975.001.A001

• Download Figure
• Download figure as PowerPoint slide

View Full Size

Impulse responses of bilateral price level to bilateral e_ij,t and USD e_$j,t exchange rates. Based on the regressions in Table 1 without interactions. Top row: exporter reported data, bottom row: importer reported. Left column: specifications (1) and (4), right column: specifications (2) and (5). Error bars: 95% confidence intervals, clustering by dyad.

Citation: IMF Working Papers 2017, 239; 10.5089/9781484327975.001.A001

• Download Figure
• Download figure as PowerPoint slide

Impulse responses of bilateral price level to bilateral e_ij,t and USD e_$j,t exchange rates. Based on the regressions in Table 1 without interactions. Top row: exporter reported data, bottom row: importer reported. Left column: specifications (1) and (4), right column: specifications (2) and (5). Error bars: 95% confidence intervals, clustering by dyad.

Citation: IMF Working Papers 2017, 239; 10.5089/9781484327975.001.A001

• Download Figure
• Download figure as PowerPoint slide

Impulse responses of bilateral price level to bilateral e_ij,t and USD e_$j,t exchange rates, as a function of importer’s dollar invoicing share S_j. Based on regression specifications (3) and (6) in Table 1 with interactions. Top row: exporter reported data, bottom row: importer reported. Error bars: 95% confidence intervals, clustering by dyad.

Citation: IMF Working Papers 2017, 239; 10.5089/9781484327975.001.A001

• Download Figure
• Download figure as PowerPoint slide

View Full Size

Impulse responses of bilateral price level to bilateral e_ij,t and USD e_$j,t exchange rates, as a function of importer’s dollar invoicing share S_j. Based on regression specifications (3) and (6) in Table 1 with interactions. Top row: exporter reported data, bottom row: importer reported. Error bars: 95% confidence intervals, clustering by dyad.

Citation: IMF Working Papers 2017, 239; 10.5089/9781484327975.001.A001

• Download Figure
• Download figure as PowerPoint slide

Impulse responses of bilateral price level to bilateral e_ij,t and USD e_$j,t exchange rates, as a function of importer’s dollar invoicing share S_j. Based on regression specifications (3) and (6) in Table 1 with interactions. Top row: exporter reported data, bottom row: importer reported. Error bars: 95% confidence intervals, clustering by dyad.

Citation: IMF Working Papers 2017, 239; 10.5089/9781484327975.001.A001

• Download Figure
• Download figure as PowerPoint slide

4.2 Terms of trade and exchange rates

The previously established fact that prices respond to the dollar exchange rate but not the bilateral exchange rate implies that the terms of trade should not respond to the bilateral exchange rate. We now test this hypothesis directly in our data by relating bilateral terms of trade to bilateral exchange rates. In this subsection, a cross-sectional unit is defined to be an unordered country pair, so that all trade flows between two countries i and j are associated with the cross-sectional unit {i, j}. Define the bilateral log terms of trade tot_ij = p_ij − p_ji − e_ij, a unitless quantity (i.e., export and import price indices are measured in the same currency). Moreover, let ppi_ij denote the log ratio of PPI in country i divided by PPI in country j, with indices expressed in the same currency.

We consider the following regressions:

$\begin{array}{cc}\mathrm{\Delta }to{t}_{ij,t}=\begin{array}{c}{\lambda }_{ij}+{\delta }_t+\underset{k=0}{\overset{2}{\mathrm{\Sigma }}}{\alpha }_k\mathrm{\Delta }{e}_{ij,t-k}+{\varepsilon }_{ij,t}\end{array},& \left(4\right)\end{array}$

$\begin{array}{cc}\mathrm{\Delta }to{t}_{ij,t}=\begin{array}{c}{\lambda }_{ij}+{\delta }_t+\underset{k=0}{\overset{2}{\mathrm{\Sigma }}}{\alpha }_k\mathrm{\Delta }{e}_{ij,t-k}+\underset{k=0}{\overset{2}{\mathrm{\Sigma }}}{\beta }_k\mathrm{\Delta }{ppi}_{ij,t-k}+{\varepsilon }_{ij,t}\end{array},& \left(5\right)\end{array}$

where λ_ij and δ_t are dyadic and time fixed effects. Regression Eq. (4) relates the growth rate of the bilateral terms of trade to the growth rate of the bilateral nominal exchange rate (and two lags). As discussed in Section 2, if exporting firms set prices in their local currencies (producer currency pricing, PCP) and prices are sticky, the contemporaneous exchange rate coefficient α₀ should equal 1. On the other hand, if exporting firms set prices in the destination currency (local currency pricing, LCP) and prices are sticky, the contemporaneous exchange rate coefficient should be − 1. However, if most prices are invoiced in U.S. dollars and are sticky in nominal terms, the coefficients α_k should be close to zero. Regression Eq. (5) controls for lags 0−2 of the growth rate of the ratio of PPI in the two countries, since firms’ optimal reset prices should fluctuate with domestic cost conditions.

In line with the dominant currency paradigm, we find that bilateral exchange rates are virtually uncorrelated with bilateral terms of trade. The results of the panel regressions are shown in Table 2. If we do not control for relative PPI, the regression results indicate that the contemporaneous effect of the exchange rate on the terms of trade is negative, in direct contradiction of PCP. While the negative sign is consistent with LCP, the magnitude is not, as the 95% confidence interval equals [−0.11, −0.07] for data reported by exporters, and [−0.05, −0.02] for data reported by importers.⁸ The coefficients on the first and second lags have opposite sign of the contemporaneous coefficient but are very small in magnitude. When controlling for relative PPI, the point estimates of the coefficients on the bilateral exchange rate shrink further toward zero, and confidence intervals on these coefficients remain narrow. Hence, the results lend strong support to the prediction of the dominant currency paradigm: Terms of trade are unresponsive to bilateral exchange rates.

Table 2:

The first (resp., last) two columns use Comtrade data reported by exporting (resp., importing) countries. S.e. clustered by dyad. The number of dyads is about half that in Table 1 since here the two ordered country tuples (i, j) and (j, i) are collapsed into one cross-sectional unit {i, j} *** p<0.01, ** p<0.05, *p<0.1.

Terms of trade and exchange rates

Table 2:

The first (resp., last) two columns use Comtrade data reported by exporting (resp., importing) countries. S.e. clustered by dyad. The number of dyads is about half that in Table 1 since here the two ordered country tuples (i, j) and (j, i) are collapsed into one cross-sectional unit {i, j} *** p<0.01, ** p<0.05, *p<0.1.

Terms of trade and exchange rates

VARIABLES	(1) export Δtotij,t	(2) export Δtotij,t	(3) import Δtotij,t	(4) import Δtotij,t
Δeij,t	−0.0881***	0.0121	−0.0369***	0.00938
	(0.00941)	(0.0127)	(0.00863)	(0.0130)
Δeij,t−1	0.0157	−0.0126	0.0447***	−0.0167
	(0.0102)	(0.0169)	(0.0104)	(0.0157)
Δeij,t−2	0.0269***	−0.00807	0.00174	0.00710
	(0.00875)	(0.0105)	(0.00788)	(0.00877)
Δppiij,t		0.239***		0.0340
		(0.0246)		(0.0260)
Δppiij,t−1		0.0605**		−0.131***
		(0.0257)		(0.0263)
Δppiij,t−2		−0.0687***		−0.0511**
		(0.0195)		(0.0212)
Time FE	yes	yes	yes	yes
R-squared	0.007	0.015	0.008	0.011
Observations	22,928	18,757	24,270	19,847
Dyads	1,322	1,172	1,347	1,200

View Table

Table 2:

The first (resp., last) two columns use Comtrade data reported by exporting (resp., importing) countries. S.e. clustered by dyad. The number of dyads is about half that in Table 1 since here the two ordered country tuples (i, j) and (j, i) are collapsed into one cross-sectional unit {i, j} *** p<0.01, ** p<0.05, *p<0.1.

Terms of trade and exchange rates

VARIABLES	(1) export Δtotij,t	(2) export Δtotij,t	(3) import Δtotij,t	(4) import Δtotij,t
Δeij,t	−0.0881***	0.0121	−0.0369***	0.00938
	(0.00941)	(0.0127)	(0.00863)	(0.0130)
Δeij,t−1	0.0157	−0.0126	0.0447***	−0.0167
	(0.0102)	(0.0169)	(0.0104)	(0.0157)
Δeij,t−2	0.0269***	−0.00807	0.00174	0.00710
	(0.00875)	(0.0105)	(0.00788)	(0.00877)
Δppiij,t		0.239***		0.0340
		(0.0246)		(0.0260)
Δppiij,t−1		0.0605**		−0.131***
		(0.0257)		(0.0263)
Δppiij,t−2		−0.0687***		−0.0511**
		(0.0195)		(0.0212)
Time FE	yes	yes	yes	yes
R-squared	0.007	0.015	0.008	0.011
Observations	22,928	18,757	24,270	19,847
Dyads	1,322	1,172	1,347	1,200

4.3 Trade volume elasticity

Having demonstrated the outsize role of the U.S. dollar in determining international prices, we now show that the dollar also dominates the bilateral exchange rate when predicting bilateral trade volumes.

Table 3 shows the results from panel regressions of trade volume on bilateral and dollar exchange rates. Let y_ij denote the log volume of goods exported from country i to country j. Our volume regressions take the same form as in the price pass-through regressions Eqs. (1) to (3), except that the dependent variable is now the log growth rate Δy_ij,t of bilateral trade volumes, and the extra controls X_jt (here indexed by j rather than i) consist of the log growth rate of real GDP (and two lags) for the importing country j. 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, our results will invariably conflate expenditure switching and shifts in aggregate import demand (for example due to valuation effects on net foreign assets). We view these regressions as predictive relationships that may inform future structural estimation exercises. Nevertheless, we will refer to the coefficients on exchange rates as “trade elasticities” for simplicity.

Table 3:

The first (resp., last) three columns use Comtrade data reported by exporting (resp., importing) countries. S.e. clustered by dyad. *** p<0.01, ** p<0.05, * p<0.1.

Trade elasticity with respect to exchange rate

Table 3:

The first (resp., last) three columns use Comtrade data reported by exporting (resp., importing) countries. S.e. clustered by dyad. *** p<0.01, ** p<0.05, * p<0.1.

Trade elasticity with respect to exchange rate

VARIABLES	(1) export Δyij,t	(2) export Δyij,t	(3) export Δyij,t	(4) import Δyij,t	(5) import Δyij,t	(6) import Δyij,t
Δeij,t	−0.148***	−0.0384**	0.0171	−0.119***	−0.0310*	−0.0765*
	(0.0148)	(0.0180)	(0.0403)	(0.0139)	(0.0160)	(0.0403)
Δeij,t−1	0.0755***	0.00728	−0.0682	0.0757***	−0.00245	0.00742
	(0.0148)	(0.0198)	(0.0485)	(0.0126)	(0.0165)	(0.0354)
Δeij,t−2	0.0416***	0.0128	0.0495	0.0393***	0.0235**	0.00410
	(0.0102)	(0.0141)	(0.0311)	(0.00919)	(0.0115)	(0.0241)
Δeij,t × Sj			−0.0606			0.118*
			(0.0655)			(0.0684)
Δeij,t−1 × Sj			0.0976			−0.0433
			(0.0746)			(0.0640)
Δeij,t−2 × Sj			−0.0441			0.0419
			(0.0494)			(0.0428)
Δe$j,t		−0.237***	−0.188***		−0.186***	−0.140**
		(0.0294)	(0.0587)		(0.0250)	(0.0600)
Δe$j,t−1		0.148***	0.359***		0.168***	0.221***
		(0.0278)	(0.0666)		(0.0248)	(0.0635)
Δe$j,t−2		0.0542***	−0.114**		0.0365*	0.111**
		(0.0201)	(0.0580)		(0.0198)	(0.0525)
Δe$j,t × Sj			−0.0104			−0.0903
			(0.0906)			(0.0871)
Δe$j,t−1 × Sj			−0.244**			−0.0465
			(0.100)			(0.0922)
Δe$j,t−2 × Sj			0.192**			−0.0952
			(0.0797)			(0.0706)
Imp. GDP	yes	yes	yes	yes	yes	yes
Time FE	yes	yes	yes	yes	yes	yes
R-squared	0.073	0.075	0.084	0.069	0.071	0.074
Observations	50,761	50,761	36,757	52,272	52,272	38,582
Dyads	2,773	2,773	1,982	2,807	2,807	2,014

View Table

Table 3:

The first (resp., last) three columns use Comtrade data reported by exporting (resp., importing) countries. S.e. clustered by dyad. *** p<0.01, ** p<0.05, * p<0.1.

Trade elasticity with respect to exchange rate

VARIABLES	(1) export Δyij,t	(2) export Δyij,t	(3) export Δyij,t	(4) import Δyij,t	(5) import Δyij,t	(6) import Δyij,t
Δeij,t	−0.148***	−0.0384**	0.0171	−0.119***	−0.0310*	−0.0765*
	(0.0148)	(0.0180)	(0.0403)	(0.0139)	(0.0160)	(0.0403)
Δeij,t−1	0.0755***	0.00728	−0.0682	0.0757***	−0.00245	0.00742
	(0.0148)	(0.0198)	(0.0485)	(0.0126)	(0.0165)	(0.0354)
Δeij,t−2	0.0416***	0.0128	0.0495	0.0393***	0.0235**	0.00410
	(0.0102)	(0.0141)	(0.0311)	(0.00919)	(0.0115)	(0.0241)
Δeij,t × Sj			−0.0606			0.118*
			(0.0655)			(0.0684)
Δeij,t−1 × Sj			0.0976			−0.0433
			(0.0746)			(0.0640)
Δeij,t−2 × Sj			−0.0441			0.0419
			(0.0494)			(0.0428)
Δe$j,t		−0.237***	−0.188***		−0.186***	−0.140**
		(0.0294)	(0.0587)		(0.0250)	(0.0600)
Δe$j,t−1		0.148***	0.359***		0.168***	0.221***
		(0.0278)	(0.0666)		(0.0248)	(0.0635)
Δe$j,t−2		0.0542***	−0.114**		0.0365*	0.111**
		(0.0201)	(0.0580)		(0.0198)	(0.0525)
Δe$j,t × Sj			−0.0104			−0.0903
			(0.0906)			(0.0871)
Δe$j,t−1 × Sj			−0.244**			−0.0465
			(0.100)			(0.0922)
Δe$j,t−2 × Sj			0.192**			−0.0952
			(0.0797)			(0.0706)
Imp. GDP	yes	yes	yes	yes	yes	yes
Time FE	yes	yes	yes	yes	yes	yes
R-squared	0.073	0.075	0.084	0.069	0.071	0.074
Observations	50,761	50,761	36,757	52,272	52,272	38,582
Dyads	2,773	2,773	1,982	2,807	2,807	2,014

The volume regressions underline the dominant role played by the U.S. dollar. As in the case of the price pass-through regressions, adding the dollar exchange rate to the volume regressions knocks down the coefficient on the bilateral exchange rate by a substantial amount. The contemporaneous elasticity for the dollar exchange rate is around −0.2 across specifications and data sources, while the elasticity for the bilateral exchange rate is an order of magnitude smaller. The data point to an interesting reversal in the years following the contemporaneous effect, whereby import volumes tend to increase one year after a ceteris paribus depreciation of the importer currency rate vis-à-vis the dollar. We discuss these lag dynamics below. Unlike the price pass-through regressions, the interactions of exchange rate changes with the importer’s dollar invoicing share are mostly imprecisely estimated here.

Fig. 3 visually depicts the regression results in the form of impulse responses. The figure shows the response of the level of bilateral trade volume to exchange rate shocks. The right column shows results from regression specifications with both bilateral and dollar exchange rates. It is apparent from the figure that the dollar exchange rate has a much more negative impact effect than the bilateral exchange rate. Yet, the figure also shows that the effect of either exchange rate on the level is essentially neutral at horizons of 1−2 years, due to the negative lagged terms in Table 3. One potential explanation is that the ratio of import prices and domestic prices adjust with a lag to exchange rate changes, implying that a year after the initial shock, relative prices faced by consumers are mostly unchanged compared to the period before the shock. However, we show in Appendix B.2 that this particular finding is driven by the early years in our sample, as results on the 2002–2015 subsample point toward a large, permanent negative effect of dollar appreciations on the volume of bilateral trade.

Impulse responses of bilateral volume to bilateral e_ij,t and USD e_$j,t exchange rates. Based on regressions in Table 3 without interactions. Top row: exporter reported data, bottom row: importer reported. Left column: specifications (1) and (4), right column: specifications (2) and (5). Error bars: 95% confidence intervals, clustering by dyad.

Citation: IMF Working Papers 2017, 239; 10.5089/9781484327975.001.A001

• Download Figure
• Download figure as PowerPoint slide

View Full Size

Impulse responses of bilateral volume to bilateral e_ij,t and USD e_$j,t exchange rates. Based on regressions in Table 3 without interactions. Top row: exporter reported data, bottom row: importer reported. Left column: specifications (1) and (4), right column: specifications (2) and (5). Error bars: 95% confidence intervals, clustering by dyad.

Citation: IMF Working Papers 2017, 239; 10.5089/9781484327975.001.A001

• Download Figure
• Download figure as PowerPoint slide

Impulse responses of bilateral volume to bilateral e_ij,t and USD e_$j,t exchange rates. Based on regressions in Table 3 without interactions. Top row: exporter reported data, bottom row: importer reported. Left column: specifications (1) and (4), right column: specifications (2) and (5). Error bars: 95% confidence intervals, clustering by dyad.

Citation: IMF Working Papers 2017, 239; 10.5089/9781484327975.001.A001

• Download Figure
• Download figure as PowerPoint slide

Fig. 4 illustrates the trade elasticity heterogeneity as a function of the dollar invoicing share, similar to Fig. 2. The point estimates of the impulse responses point toward an association between higher dollar invoicing shares and more negative trade elasticities, but the confidence intervals are wider than in our earlier price pass-through regressions. Section 5 provides further quantitative evidence on trade elasticity heterogeneity and its determinants.

Impulse responses of bilateral volume to bilateral e_ij,t and USD e_$j,t exchange rates, as a function of importer’s dollar invoicing share Sj. Based on regression specifications (3) and (6) in Table 1 with interactions. Top row: exporter reported data, bottom row: importer reported. Error bars: 95% confidence intervals, clustering by dyad.

Citation: IMF Working Papers 2017, 239; 10.5089/9781484327975.001.A001

• Download Figure
• Download figure as PowerPoint slide

View Full Size

Impulse responses of bilateral volume to bilateral e_ij,t and USD e_$j,t exchange rates, as a function of importer’s dollar invoicing share Sj. Based on regression specifications (3) and (6) in Table 1 with interactions. Top row: exporter reported data, bottom row: importer reported. Error bars: 95% confidence intervals, clustering by dyad.

Citation: IMF Working Papers 2017, 239; 10.5089/9781484327975.001.A001

• Download Figure
• Download figure as PowerPoint slide

Impulse responses of bilateral volume to bilateral e_ij,t and USD e_$j,t exchange rates, as a function of importer’s dollar invoicing share Sj. Based on regression specifications (3) and (6) in Table 1 with interactions. Top row: exporter reported data, bottom row: importer reported. Error bars: 95% confidence intervals, clustering by dyad.

Citation: IMF Working Papers 2017, 239; 10.5089/9781484327975.001.A001

• Download Figure
• Download figure as PowerPoint slide

4.4 Effects of U.S. monetary policy shocks

We now directly estimate the effects of U.S. monetary policy induced exchange rate changes on bilateral trade. The dominant role played by the dollar in our earlier price and volume regressions suggests that U.S. monetary policy actions that cause the dollar to fluctuate could have large spillover effects on international trade prices and volumes. In this section we directly measure the impulse responses with respect to U.S. monetary shocks.

As described in Section 3, for each country we construct an annual measure of U.S. monetary policy induced exchange rate fluctuations by summing daily log exchange rate changes on dates with scheduled FOMC announcements. Let z_j,t be the sum of these daily log changes in year t for country j, where the exchange rate is measured in units of country j currency per U.S. dollar. The use of daily exchange rate changes on FOMC announcement dates allows for a specific interpretation of the shocks z_j,t, whereas the overall exchange rate change Δe_$j,t arguably is an amalgam of the effects of several distinct shocks. However, we do not claim that z_j,t is solely driven by U.S. monetary policy actions, or that all fluctuations in z_j,t are necessarily “surprises”. Because we are interested in a large number of currencies for a sample going back to 1990, we cannot use high-frequency intra-day data as in Andersen et al. (2003), Gürkaynak et al. (2005), and Nakamura and Steinsson (2016), for example.

Daily and annual data are consistent with our interpretation of the U.S. monetary shock series. U.S. monetary tightenings are associated with appreciations of the dollar against the currencies in our sample: The daily correlation of log exchange rate changes (across all currencies) and changes in the U.S. 1–year Treasury bill rate is 0.064 on FOMC dates in the period 1990–2007 (i.e., excluding the zero lower bound period). The standard deviation of daily log exchange rate changes across all currencies is 0.72% on FOMC dates; it is 0.92% across all days in the 1990–2015 sample. The standard deviation of the annual (i.e., summed) monetary shock series z_j,t is 2.6%, compared to an annual standard deviation of 31.6% for overall log exchange rate changes Δe_$j,t on the same 1990–2015 sample. The annual correlation of z_j,t and Δe_$j,t is 0.17. We conclude that our monetary shock series captures a small but nontrivial fraction of shocks to exchange rates vis-à-vis the dollar.

We measure the effect of U.S. monetary policy actions on bilateral trade prices using panel fixed effects regressions as in Eqs. (1) to (3) in Section 4.1, except here we replace raw exchange rate changes with our U.S. monetary shock series. Specifically, we replace the log change in the bilateral exchange rate Δe_ij,t with the U.S. monetary policy induced change in the bilateral exchange rate, z_j,t − z_i,t, and we replace the log change in the dollar exchange rate Δe_$j,t with the dollar exchange rate shock z_j,t. We include lags 0−6 of the exchange rate shocks on the right-hand side of our regression specifications, as our results below suggest that the pass-through of U.S. monetary shocks is more gradual than the pass-through of raw exchange rate changes. We continue to control for dyad and time fixed effects, as well as lags 0−2 of exporter PPI log growth.

Fig. 5 shows that the average pass-through of U.S. monetary policy induced exchange rate changes into bilateral import prices is high in the long run. The figure depicts the impulse response functions of the bilateral price level to shocks.⁹ The left column is based on specifications with only shocks z_j,t — z_i,t to the bilateral exchange rate, while the right column is based on specifications that add shocks z_j,t to the dollar exchange rate. We split the results by exporter (top row) and importer (bottom row) reported data. The impact effect of shocks to both bilateral and dollar exchange rates is about 0.3. However, in specifications with both bilateral and dollar exchange rate shocks, the dollar pass-through rises to about 0.6−1.0 in years 4−6, depending on specification, whereas the bilateral pass-through is negligible in the long run.

Impulse responses of bilateral price level to bilateral z_j,t − z_i,t and USD z_j,t U.S. monetary policy induced exchange rate shocks. Based on regression specifications without interactions. Top row: exporter reported data, bottom row: importer reported. Left column: specifications with only bilateral shock, right column: specifications with bilateral and USD shocks. Error bars: 95% confidence intervals, clustering by dyad.

Citation: IMF Working Papers 2017, 239; 10.5089/9781484327975.001.A001

• Download Figure
• Download figure as PowerPoint slide

View Full Size

Impulse responses of bilateral price level to bilateral z_j,t − z_i,t and USD z_j,t U.S. monetary policy induced exchange rate shocks. Based on regression specifications without interactions. Top row: exporter reported data, bottom row: importer reported. Left column: specifications with only bilateral shock, right column: specifications with bilateral and USD shocks. Error bars: 95% confidence intervals, clustering by dyad.

Citation: IMF Working Papers 2017, 239; 10.5089/9781484327975.001.A001

• Download Figure
• Download figure as PowerPoint slide

Impulse responses of bilateral price level to bilateral z_j,t − z_i,t and USD z_j,t U.S. monetary policy induced exchange rate shocks. Based on regression specifications without interactions. Top row: exporter reported data, bottom row: importer reported. Left column: specifications with only bilateral shock, right column: specifications with bilateral and USD shocks. Error bars: 95% confidence intervals, clustering by dyad.

Citation: IMF Working Papers 2017, 239; 10.5089/9781484327975.001.A001

• Download Figure
• Download figure as PowerPoint slide

Fig. 6 shows that the pass-through of shocks to the dollar exchange rate is strongly positively associated with the importer’s dollar invoicing share S_j. As in Section 4.1, we report impulse response functions for low, medium, and high levels of dollar invoicing, where the point estimates and confidence intervals are computed from regressions with interactions. The association between pass-through and dollar invoicing is even stronger for the U.S. monetary policy induced shocks than for the overall exchange rate changes analyzed in Section 4.1. While the estimated long-run pass-through of shocks to the dollar exchange rate is close to zero for a country like Switzerland with S_j = 0.13, it is about 2.0−3.5 for a country like Argentina with S_j = 0.88, depending on specification. These magnitudes are not directly comparable with the pass-through magnitudes in Section 4.1, since a 1% monetary shock z_j,t represents a much rarer occurrence than a 1% appreciation of the dollar exchange rate e_$j,t, as discussed above.

Impulse responses of bilateral price level to bilateral z_j,t − z_i,t and USD z_j,t U.S. monetary policy induced exchange rate shocks, as a function of importer’s dollar invoicing share S_j. Based on regression specifications with interactions. Top row: exporter reported data, bottom row: importer reported. Error bars: 95% confidence intervals, clustering by dyad.

Citation: IMF Working Papers 2017, 239; 10.5089/9781484327975.001.A001

• Download Figure
• Download figure as PowerPoint slide

View Full Size

Impulse responses of bilateral price level to bilateral z_j,t − z_i,t and USD z_j,t U.S. monetary policy induced exchange rate shocks, as a function of importer’s dollar invoicing share S_j. Based on regression specifications with interactions. Top row: exporter reported data, bottom row: importer reported. Error bars: 95% confidence intervals, clustering by dyad.

Citation: IMF Working Papers 2017, 239; 10.5089/9781484327975.001.A001

• Download Figure
• Download figure as PowerPoint slide

Impulse responses of bilateral price level to bilateral z_j,t − z_i,t and USD z_j,t U.S. monetary policy induced exchange rate shocks, as a function of importer’s dollar invoicing share S_j. Based on regression specifications with interactions. Top row: exporter reported data, bottom row: importer reported. Error bars: 95% confidence intervals, clustering by dyad.

Citation: IMF Working Papers 2017, 239; 10.5089/9781484327975.001.A001

• Download Figure
• Download figure as PowerPoint slide

Unlike for price pass-through, regressions of trade volumes on U.S. monetary shocks yield very noisy results. This is a consequence of relatively low R² values for volume regressions, which we also found in Section 4.3, coupled with the limited variation in measured monetary shocks relative to the variation in overall exchange rates. We therefore refrain from reporting volume regression results here. It is an interesting topic for future research to develop more comprehensive (and thus more volatile) measures of U.S. monetary policy shocks that can be related directly to trade volumes.

4.5 The dollar versus the euro

We conclude this section by comparing the explanatory power of the dollar exchange rate with that of the euro. We show that the dollar dominates both the bilateral exchange rate and the euro in regression specifications that include all three exchange rates.

The preceding panel regressions do not directly imply that the U.S. dollar is a uniquely important vehicle currency. In our regression specifications without interactions, we would have obtained exactly the same coefficient estimates if we had used the euro exchange rate, say, in place of the dollar exchange rate, since we control for time fixed effects. Nevertheless, our specifications with interactions indicated that the dollar may play a special role. Now we directly compare the explanatory power of the dollar against that of the euro in panel regressions that do not control for time fixed effects but instead control for observed global real and financial variables.

To measure bilateral price pass-through from the dollar and the euro, we run panel regressions of the form

$\begin{array}{cc}\mathrm{\Delta }{p}_{ij,t}=\begin{array}{c}{\lambda }_{ij}+\underset{k=0}{\overset{2}{\mathrm{\Sigma }}}{\alpha }_k\mathrm{\Delta }{e}_{ij,t-k}+\underset{k=0}{\overset{2}{\mathrm{\Sigma }}}{\beta }_k\mathrm{\Delta }{e}_{\$j,t-k}+\underset{k=0}{\overset{2}{\mathrm{\Sigma }}}{\xi }_k\mathrm{\Delta }{e}_{€j,t-k}+\theta \prime X_{ij,t}+{\varepsilon }_{ij,t}\end{array},& \left(6\right)\end{array}$

where e_€j,t denotes the log euro exchange rate in units of currency j per euro. Notice that we omit time fixed effects, as is necessary to identify β_k and ξ_k separately. In addition to lags 0−2 of exporter PPI log growth, the controls X_ij,t consist of the contemporaneous values of global real GDP growth, global GDP deflator inflation, global export volume growth, growth in the WTI oil price deflated by the global GDP deflator, and the log VIX. The time sample for regressions in this subsection is 2002–2015 due to the introduction of the euro in 1999 and our use of lagged exchange rate changes.

Fig. 7 shows that the euro pass-through into prices is negligible on average, while the dollar pass-through remains high when we control for the euro. The figure displays the regression results in the form of impulse responses of the bilateral price level; corresponding regression tables are available in Appendix B.2. We split by exporter (top) and importer (bottom) reported data. The left column shows results for specifications that do not include the dollar exchange rate, i.e., restricting β_k = 0 in Eq. (6). The right column displays estimates of bilateral, dollar, and euro pass-through from regressions with all three exchange rates. In the latter specifications, the dollar pass-through is quantitatively much larger than the euro pass-through; indeed, the results are close to those of Section 4.1 where we did not consider the euro at all. The euro pass-through is quantitatively small at all horizons, and it is in fact estimated to be slightly negative. The difference between the dollar and euro pass-through is statistically significant at conventional significance levels. In Appendix B.2 we show through regressions with interactions that the importer’s dollar and euro invoicing shares help explain the heterogeneity in pass-through.

Impulse responses of bilateral price level to bilateral e_ij,t, USD e_$j,t, and euro e_€j,t exchange rates. Based on regressions in Table 6, Appendix B.2. Top row: exporter reported data, bottom row: importer reported. Left column: specifications with only bilateral and euro ER, right column: specifications adding USD. Error bars: 95% confidence intervals, clustering by dyad.

Citation: IMF Working Papers 2017, 239; 10.5089/9781484327975.001.A001

• Download Figure
• Download figure as PowerPoint slide

View Full Size

Impulse responses of bilateral price level to bilateral e_ij,t, USD e_$j,t, and euro e_€j,t exchange rates. Based on regressions in Table 6, Appendix B.2. Top row: exporter reported data, bottom row: importer reported. Left column: specifications with only bilateral and euro ER, right column: specifications adding USD. Error bars: 95% confidence intervals, clustering by dyad.

Citation: IMF Working Papers 2017, 239; 10.5089/9781484327975.001.A001

• Download Figure
• Download figure as PowerPoint slide

Impulse responses of bilateral price level to bilateral e_ij,t, USD e_$j,t, and euro e_€j,t exchange rates. Based on regressions in Table 6, Appendix B.2. Top row: exporter reported data, bottom row: importer reported. Left column: specifications with only bilateral and euro ER, right column: specifications adding USD. Error bars: 95% confidence intervals, clustering by dyad.

Citation: IMF Working Papers 2017, 239; 10.5089/9781484327975.001.A001

• Download Figure
• Download figure as PowerPoint slide

Table 4:

Summary statistics for countries in the merged Comtrade/WDI sample. #dyads: number of non-missing dyads that the country appears in. avg T: average number of years per dyad that the country appears in; a dyad-year observation is counted if at least one UVI or volume observation is reported by the exporter or importer, and exchange rate data exists for both countries. InvS: share of imports invoiced in USD/euro.

Country summary statistics

Table 4:

Summary statistics for countries in the merged Comtrade/WDI sample. #dyads: number of non-missing dyads that the country appears in. avg T: average number of years per dyad that the country appears in; a dyad-year observation is counted if at least one UVI or volume observation is reported by the exporter or importer, and exchange rate data exists for both countries. InvS: share of imports invoiced in USD/euro.

Country summary statistics

	As exporter		As importer
Country	#dyads	avg T.	#dyads	avg T	Inv S$	Inv S€
Africa
Algeria	20	14.2	49	21.1		0.49
Egypt	54	20.4	52	21.6
South Africa	54	15.0	53	14.7
Americas
Argentina	54	21.8	52	21.8	0.88	0.08
Brazil	54	24.2	53	23.1	0.84	0.11
Canada	54	24.4	54	24.1	0.75	0.05
Chile	52	20.9	50	22.0
Colombia	53	19.5	52	21.4	0.99	0.00
Mexico	54	23.4	52	23.3
United States	54	24.0	54	23.5	0.93	0.02
Venezuela	21	10.9	48	19.9
Asia
China	54	23.7	54	22.6
Hong Kong	54	23.2	52	22.5
India	54	25.3	53	24.4	0.86	0.10
Indonesia	54	23.9	52	23.0	0.81	0.04
Israel	53	21.5	51	21.3	0.73	0.21
Japan	54	25.6	52	25.5	0.71	0.03
Kazakhstan	39	14.6	52	18.2
Malaysia	54	24.1	53	23.4
Philippines	54	22.1	50	21.5
Saudi Arabia	50	20.1	53	21.2
Singapore	54	24.7	51	24.0
South Korea	54	25.0	52	24.6	0.81	0.05
Thailand	54	24.5	53	24.5	0.79	0.04
Turkey	54	24.4	54	23.9	0.59	0.31
Vietnam	54	19.3	49	19.0
Europe
Austria	54	23.1	52	23.0	0.06	0.70
Belgium	54	15.9	53	15.9	0.14	0.82
Czech Republic	54	20.6	53	21.3	0.19	0.68
Denmark	54	22.3	52	24.4	0.25	0.32
Estonia	47	17.9	52	19.3	0.34	0.53
Finland	54	25.6	52	25.0	0.42	0.38
France	54	23.1	54	22.7	0.21	0.75
Germany	54	23.3	54	23.0	0.23	0.75
Greece	54	23.0	51	23.6	0.40	0.58
Hungary	54	23.6	52	22.6	0.27	0.57
Ireland	54	23.4	53	22.5	0.23	0.47
Italy	54	23.1	54	22.5	0.29	0.67
Lithuania	53	17.3	50	18.9	0.51	0.39
Luxembourg	54	15.8	51	14.0	0.16	0.78
Netherlands	54	23.7	54	23.2	0.37	0.46
Norway	54	23.1	52	23.0	0.21	0.29
Poland	54	22.9	52	22.3	0.30	0.58
Portugal	54	24.9	53	24.8	0.22	0.76
Romania	54	22.6	52	21.4	0.31	0.67
Russia	54	21.4	52	21.0
Slovak Republic	54	20.7	51	20.4	0.12	0.79
Slovenia	54	21.1	52	20.7	0.20	0.75
Spain	54	24.8	54	24.9	0.35	0.58
Sweden	54	23.7	54	23.1	0.25	0.36
Switzerland	54	25.6	54	25.1	0.13	0.53
Ukraine	53	19.3	52	19.8	0.75	0.16
United Kingdom	54	23.4	54	23.3	0.47	0.15
Oceania
Australia	54	25.1	52	25.2	0.53	0.08
New Zealand	54	22.7	50	24.0

View Table

Table 4:

Summary statistics for countries in the merged Comtrade/WDI sample. #dyads: number of non-missing dyads that the country appears in. avg T: average number of years per dyad that the country appears in; a dyad-year observation is counted if at least one UVI or volume observation is reported by the exporter or importer, and exchange rate data exists for both countries. InvS: share of imports invoiced in USD/euro.

Country summary statistics

	As exporter		As importer
Country	#dyads	avg T.	#dyads	avg T	Inv S$	Inv S€
Africa
Algeria	20	14.2	49	21.1		0.49
Egypt	54	20.4	52	21.6
South Africa	54	15.0	53	14.7
Americas
Argentina	54	21.8	52	21.8	0.88	0.08
Brazil	54	24.2	53	23.1	0.84	0.11
Canada	54	24.4	54	24.1	0.75	0.05
Chile	52	20.9	50	22.0
Colombia	53	19.5	52	21.4	0.99	0.00
Mexico	54	23.4	52	23.3
United States	54	24.0	54	23.5	0.93	0.02
Venezuela	21	10.9	48	19.9
Asia
China	54	23.7	54	22.6
Hong Kong	54	23.2	52	22.5
India	54	25.3	53	24.4	0.86	0.10
Indonesia	54	23.9	52	23.0	0.81	0.04
Israel	53	21.5	51	21.3	0.73	0.21
Japan	54	25.6	52	25.5	0.71	0.03
Kazakhstan	39	14.6	52	18.2
Malaysia	54	24.1	53	23.4
Philippines	54	22.1	50	21.5
Saudi Arabia	50	20.1	53	21.2
Singapore	54	24.7	51	24.0
South Korea	54	25.0	52	24.6	0.81	0.05
Thailand	54	24.5	53	24.5	0.79	0.04
Turkey	54	24.4	54	23.9	0.59	0.31
Vietnam	54	19.3	49	19.0
Europe
Austria	54	23.1	52	23.0	0.06	0.70
Belgium	54	15.9	53	15.9	0.14	0.82
Czech Republic	54	20.6	53	21.3	0.19	0.68
Denmark	54	22.3	52	24.4	0.25	0.32
Estonia	47	17.9	52	19.3	0.34	0.53
Finland	54	25.6	52	25.0	0.42	0.38
France	54	23.1	54	22.7	0.21	0.75
Germany	54	23.3	54	23.0	0.23	0.75
Greece	54	23.0	51	23.6	0.40	0.58
Hungary	54	23.6	52	22.6	0.27	0.57
Ireland	54	23.4	53	22.5	0.23	0.47
Italy	54	23.1	54	22.5	0.29	0.67
Lithuania	53	17.3	50	18.9	0.51	0.39
Luxembourg	54	15.8	51	14.0	0.16	0.78
Netherlands	54	23.7	54	23.2	0.37	0.46
Norway	54	23.1	52	23.0	0.21	0.29
Poland	54	22.9	52	22.3	0.30	0.58
Portugal	54	24.9	53	24.8	0.22	0.76
Romania	54	22.6	52	21.4	0.31	0.67
Russia	54	21.4	52	21.0
Slovak Republic	54	20.7	51	20.4	0.12	0.79
Slovenia	54	21.1	52	20.7	0.20	0.75
Spain	54	24.8	54	24.9	0.35	0.58
Sweden	54	23.7	54	23.1	0.25	0.36
Switzerland	54	25.6	54	25.1	0.13	0.53
Ukraine	53	19.3	52	19.8	0.75	0.16
United Kingdom	54	23.4	54	23.3	0.47	0.15
Oceania
Australia	54	25.1	52	25.2	0.53	0.08
New Zealand	54	22.7	50	24.0

Table 5:

Definition of BLS country groups in Fig. 17. Countries marked with an asterisk (*) are not available in the Comtrade sample.

BLS country groups

Table 5:

Definition of BLS country groups in Fig. 17. Countries marked with an asterisk (*) are not available in the Comtrade sample.

BLS country groups

BLS group	Country ISO codes
ASEAN	BRN* IDN KHM* LAO* MMR* MYS PHL SGP THA VNM*
Asia Near East	ARE* BHR* IRN* IRQ* ISR JOR* KWT* LBN* OMN* QAT* SAU SYR* YEM*
European Union	AUT BEL BGR* CYP* CZE DEU DNK ESP EST FIN FRA GBR GRC HRV* HUN IRL ITA LTU LUX LVA* MLT* NLD POL PRT ROU SVK SVN SWE
Latin America	ARG BRA CHL COL MEX VEN (plus other unspecified Central American, South American, and Caribbean countries*)
Asian New. Ind.	HKG KOR SGP TWN
Pacific Rim	AUS BRN* CHN HKG IDN JPN KOR MAC* MYS NZL PHL PNG* SGP TWN

View Table

Table 5:

Definition of BLS country groups in Fig. 17. Countries marked with an asterisk (*) are not available in the Comtrade sample.

BLS country groups

BLS group	Country ISO codes
ASEAN	BRN* IDN KHM* LAO* MMR* MYS PHL SGP THA VNM*
Asia Near East	ARE* BHR* IRN* IRQ* ISR JOR* KWT* LBN* OMN* QAT* SAU SYR* YEM*
European Union	AUT BEL BGR* CYP* CZE DEU DNK ESP EST FIN FRA GBR GRC HRV* HUN IRL ITA LTU LUX LVA* MLT* NLD POL PRT ROU SVK SVN SWE
Latin America	ARG BRA CHL COL MEX VEN (plus other unspecified Central American, South American, and Caribbean countries*)
Asian New. Ind.	HKG KOR SGP TWN
Pacific Rim	AUS BRN* CHN HKG IDN JPN KOR MAC* MYS NZL PHL PNG* SGP TWN

Table 6:

The first (resp., last) three columns use Comtrade data reported by exporting (resp., importing) countries. S.e. clustered by dyad. *** p<0.01, ** p<0.05, * p<0.1.

Euro vs. dollar exchange rate pass-through into prices

Table 6:

The first (resp., last) three columns use Comtrade data reported by exporting (resp., importing) countries. S.e. clustered by dyad. *** p<0.01, ** p<0.05, * p<0.1.

Euro vs. dollar exchange rate pass-through into prices

VARIABLES	(1) Δpij,t	(2) Δpij,t	(3) Δpij,t	(4) Δpij,t	(5) Δpij,t	(6) Δpij,t
Δeij,t	0.438***	0.316***	0.286***	0.305***	0.181***	0.207***
	(0.0154)	(0.0169)	(0.0702)	(0.0159)	(0.0174)	(0.0695)
Δeij,t−1	0.146***	0.119***	0.276***	0.162***	0.134***	0.0277
	(0.0151)	(0.0157)	(0.0821)	(0.0133)	(0.0140)	(0.0621)
Δeij,t−2	−0.101***	−0.0533***	−0.0898	−0.0793***	−0.0389***	−0.0723
	(0.0148)	(0.0151)	(0.0744)	(0.0123)	(0.0129)	(0.0666)
$\mathrm{\Delta }{e}_{ij,t}\times (S_j+S_j^{€})$			0.0449			−0.0357
			(0.0834)			(0.0784)
$\mathrm{\Delta }{e}_{ij,t-1}\times (S_j+S_j^{€})$			−0.178*			0.0814
			(0.0919)			(0.0720)
$\mathrm{\Delta }{e}_{ij,t-2}\times (S_j+S_j^{€})$			0.0311			0.0640
			(0.0827)			(0.0745)
Δe$j,t		0.666***	0.512***		0.754***	0.614***
		(0.0358)	(0.0448)		(0.0373)	(0.0405)
Δe$j,t−1		−0.0459***	−0.0121		−0.0646***	0.155***
		(0.0161)	(0.0311)		(0.0161)	(0.0252)
Δe$j,t−2		0.00180	0.0456		0.0576***	0.0154
		(0.0158)	(0.0299)		(0.0148)	(0.0228)
Δe$j,t × Sj			0.216***			0.510***
			(0.0470)			(0.0439)
Δe$j,t−1 × Sj			−0.0841			−0.447***
			(0.0613)			(0.0555)
Δe$j,t−2 × Sj			−0.0817			0.101**
			(0.0499)			(0.0465)
Δe€j,t	0.396***	−0.0688**	−0.108**	0.467***	−0.0800**	−0.347***
	(0.0179)	(0.0330)	(0.0434)	(0.0175)	(0.0332)	(0.0430)
Δe€j,t−1	−0.174***	−0.109***	−0.0923***	−0.187***	−0.103***	0.0450
	(0.0168)	(0.0201)	(0.0346)	(0.0166)	(0.0209)	(0.0398)
Δe€j,t−2	0.119***	0.0552***	0.0867***	0.132***	0.0294	−0.0982***
	(0.0168)	(0.0204)	(0.0295)	(0.0156)	(0.0188)	(0.0284)
$\mathrm{\Delta }{e}_{€j,t}\times S_j^{€}$			0.285***			0.694***
			(0.0860)			(0.0821)
$\mathrm{\Delta }{e}_{€j,t-1}\times S_j^{€}$			0.000			−0.203**
			(0.0893)			(0.0982)
$\mathrm{\Delta }{e}_{€j,t-2}\times S_j^{€}$			−0.0615			0.295***
			(0.0819)			(0.0851)
Exp. PPI	yes	yes	yes	yes	yes	yes
Agg. controls	yes	yes	yes	yes	yes	yes
Time FE	no	no	no	no	no	no
R-squared	0.185	0.197	0.211	0.131	0.143	0.210
Observations	32,825	32,825	23,668	33,802	33,802	24,463
Dyads	2,611	2,611	1,867	2,647	2,647	1,900

View Table

Table 6:

The first (resp., last) three columns use Comtrade data reported by exporting (resp., importing) countries. S.e. clustered by dyad. *** p<0.01, ** p<0.05, * p<0.1.

Euro vs. dollar exchange rate pass-through into prices

VARIABLES	(1) Δpij,t	(2) Δpij,t	(3) Δpij,t	(4) Δpij,t	(5) Δpij,t	(6) Δpij,t
Δeij,t	0.438***	0.316***	0.286***	0.305***	0.181***	0.207***
	(0.0154)	(0.0169)	(0.0702)	(0.0159)	(0.0174)	(0.0695)
Δeij,t−1	0.146***	0.119***	0.276***	0.162***	0.134***	0.0277
	(0.0151)	(0.0157)	(0.0821)	(0.0133)	(0.0140)	(0.0621)
Δeij,t−2	−0.101***	−0.0533***	−0.0898	−0.0793***	−0.0389***	−0.0723
	(0.0148)	(0.0151)	(0.0744)	(0.0123)	(0.0129)	(0.0666)
$\mathrm{\Delta }{e}_{ij,t}\times (S_j+S_j^{€})$			0.0449			−0.0357
			(0.0834)			(0.0784)
$\mathrm{\Delta }{e}_{ij,t-1}\times (S_j+S_j^{€})$			−0.178*			0.0814
			(0.0919)			(0.0720)
$\mathrm{\Delta }{e}_{ij,t-2}\times (S_j+S_j^{€})$			0.0311			0.0640
			(0.0827)			(0.0745)
Δe$j,t		0.666***	0.512***		0.754***	0.614***
		(0.0358)	(0.0448)		(0.0373)	(0.0405)
Δe$j,t−1		−0.0459***	−0.0121		−0.0646***	0.155***
		(0.0161)	(0.0311)		(0.0161)	(0.0252)
Δe$j,t−2		0.00180	0.0456		0.0576***	0.0154
		(0.0158)	(0.0299)		(0.0148)	(0.0228)
Δe$j,t × Sj			0.216***			0.510***
			(0.0470)			(0.0439)
Δe$j,t−1 × Sj			−0.0841			−0.447***
			(0.0613)			(0.0555)
Δe$j,t−2 × Sj			−0.0817			0.101**
			(0.0499)			(0.0465)
Δe€j,t	0.396***	−0.0688**	−0.108**	0.467***	−0.0800**	−0.347***
	(0.0179)	(0.0330)	(0.0434)	(0.0175)	(0.0332)	(0.0430)
Δe€j,t−1	−0.174***	−0.109***	−0.0923***	−0.187***	−0.103***	0.0450
	(0.0168)	(0.0201)	(0.0346)	(0.0166)	(0.0209)	(0.0398)
Δe€j,t−2	0.119***	0.0552***	0.0867***	0.132***	0.0294	−0.0982***
	(0.0168)	(0.0204)	(0.0295)	(0.0156)	(0.0188)	(0.0284)
$\mathrm{\Delta }{e}_{€j,t}\times S_j^{€}$			0.285***			0.694***
			(0.0860)			(0.0821)
$\mathrm{\Delta }{e}_{€j,t-1}\times S_j^{€}$			0.000			−0.203**
			(0.0893)			(0.0982)
$\mathrm{\Delta }{e}_{€j,t-2}\times S_j^{€}$			−0.0615			0.295***
			(0.0819)			(0.0851)
Exp. PPI	yes	yes	yes	yes	yes	yes
Agg. controls	yes	yes	yes	yes	yes	yes
Time FE	no	no	no	no	no	no
R-squared	0.185	0.197	0.211	0.131	0.143	0.210
Observations	32,825	32,825	23,668	33,802	33,802	24,463
Dyads	2,611	2,611	1,867	2,647	2,647	1,900

Similarly, we find that the dollar exchange rate has the largest predictive power for trade volumes. We run panel regressions similar to Eq. (6), except with volume growth Δy_ij,t on the let-hand side, and we replace exporter PPI with lags 0−2 of importer real GDP growth in the list of controls X_ij,t. Fig. 8 shows impulse responses of the level of bilateral trade volume to the bilateral, dollar, and euro exchange rates. The dollar exchange rate is the only one of the three that has a quantitatively large negative association with trade volumes.¹⁰ Curiously, the trade elasticity of the euro exchange rate is estimated to be positive both in the short and long run, although the confidence intervals are consistent with economically small magnitudes.¹¹ The difference between the dollar and euro trade elasticities is significant at conventional significance levels.

Impulse responses of bilateral volume to bilateral e_ij,t, USD e_$j,t, and euro e_€j,t exchange rates. Based on regressions in Table 7, Appendix B.2. Top row: exporter reported data, bottom row: importer reported. Left column: specifications with bilateral and euro ER, right column: specifications adding USD. Error bars: 95% confidence intervals, clustering by dyad.

Citation: IMF Working Papers 2017, 239; 10.5089/9781484327975.001.A001

• Download Figure
• Download figure as PowerPoint slide

View Full Size

Impulse responses of bilateral volume to bilateral e_ij,t, USD e_$j,t, and euro e_€j,t exchange rates. Based on regressions in Table 7, Appendix B.2. Top row: exporter reported data, bottom row: importer reported. Left column: specifications with bilateral and euro ER, right column: specifications adding USD. Error bars: 95% confidence intervals, clustering by dyad.

Citation: IMF Working Papers 2017, 239; 10.5089/9781484327975.001.A001

• Download Figure
• Download figure as PowerPoint slide

Impulse responses of bilateral volume to bilateral e_ij,t, USD e_$j,t, and euro e_€j,t exchange rates. Based on regressions in Table 7, Appendix B.2. Top row: exporter reported data, bottom row: importer reported. Left column: specifications with bilateral and euro ER, right column: specifications adding USD. Error bars: 95% confidence intervals, clustering by dyad.

Citation: IMF Working Papers 2017, 239; 10.5089/9781484327975.001.A001

• Download Figure
• Download figure as PowerPoint slide

Table 7:

The first (resp., last) three columns use Comtrade data reported by exporting (resp., importing) countries. S.e. clustered by dyad. *** p<0.01, ** p<0.05, * p<0.1.

Euro vs. dollar trade elasticity

Table 7:

The first (resp., last) three columns use Comtrade data reported by exporting (resp., importing) countries. S.e. clustered by dyad. *** p<0.01, ** p<0.05, * p<0.1.

Euro vs. dollar trade elasticity

VARIABLES	(1) Δyij,t	(2) Δyij,t	(3) Δyij,t	(4) Δyij,t	(5) Δyij,t	(6) Δyij,t
Δeij,t	−0.110***	−0.0234	0.581***	−0.0631*	0.0229	0.619***
	(0.0418)	(0.0432)	(0.224)	(0.0371)	(0.0386)	(0.229)
Δeij,t−1	−0.0349	−0.00530	−0.177	−0.0827**	−0.0694*	−0.366*
	(0.0339)	(0.0354)	(0.184)	(0.0343)	(0.0367)	(0.188)
Δeij,t−2	−0.0151	−0.0393	0.167	−0.0419	−0.0679**	0.121
	(0.0307)	(0.0318)	(0.190)	(0.0270)	(0.0277)	(0.165)
$\mathrm{\Delta }{e}_{ij,t}\times (S_j+S_j^{€})$			−0.629**			−0.668***
			(0.260)			(0.253)
$\mathrm{\Delta }{e}_{ij,t-1}\times (S_j+S_j^{€})$			0.160			0.348*
			(0.218)			(0.207)
$\mathrm{\Delta }{e}_{ij,t-2}\times (S_j+S_j^{€})$			−0.207			−0.203
			(0.221)			(0.196)
Δe$j,t		−0.703***	−0.663***		−0.695***	−0.316***
		(0.0894)	(0.115)		(0.0806)	(0.103)
Δe$j,t−1		−0.0615	0.116		−0.00291	−0.120*
		(0.0441)	(0.0822)		(0.0410)	(0.0693)
Δe$j,t−2		0.00568	−0.257***		0.00554	0.0595
		(0.0441)	(0.0797)		(0.0400)	(0.0613)
Δe$j,t × Sj			−0.243*			−0.760***
			(0.131)			(0.121)
Δe$j,t−1 × Sj			−0.221			0.357**
			(0.159)			(0.145)
Δe$j,t−2 × Sj			0.446***			0.00131
			(0.131)			(0.108)
Δe€j,t	−0.244***	0.264***	0.302***	−0.179***	0.320***	0.427***
	(0.0525)	(0.0870)	(0.110)	(0.0413)	(0.0759)	(0.0960)
Δe€j,t−1	0.117***	0.119***	0.218***	0.170***	0.145***	0.00391
	(0.0416)	(0.0454)	(0.0741)	(0.0374)	(0.0406)	(0.0779)
Δe€j,t−2	−0.0147	0.0161	−0.116*	−0.0295	0.00521	0.0250
	(0.0377)	(0.0460)	(0.0661)	(0.0333)	(0.0406)	(0.0639)
$\mathrm{\Delta }{e}_{€j,t}\times S_j^{€}$			0.0101			−0.313*
			(0.214)			(0.185)
$\mathrm{\Delta }{e}_{€j,t-1}\times S_j^{€}$			−0.115			0.194
			(0.229)			(0.197)
$\mathrm{\Delta }{e}_{€j,t-2}\times S_j^{€}$			0.560***			0.0756
			(0.200)			(0.201)
Imp. GDP	yes	yes	yes	yes	yes	yes
Agg. controls	yes	yes	yes	yes	yes	yes
Time FE	no	no	no	no	no	no
R-squared	0.072	0.075	0.085	0.068	0.071	0.078
Observations	36,269	36,269	26,129	37,437	37,437	27,109
Dyads	2,773	2,773	1,982	2,807	2,807	2,014

View Table

Table 7:

The first (resp., last) three columns use Comtrade data reported by exporting (resp., importing) countries. S.e. clustered by dyad. *** p<0.01, ** p<0.05, * p<0.1.

Euro vs. dollar trade elasticity

VARIABLES	(1) Δyij,t	(2) Δyij,t	(3) Δyij,t	(4) Δyij,t	(5) Δyij,t	(6) Δyij,t
Δeij,t	−0.110***	−0.0234	0.581***	−0.0631*	0.0229	0.619***
	(0.0418)	(0.0432)	(0.224)	(0.0371)	(0.0386)	(0.229)
Δeij,t−1	−0.0349	−0.00530	−0.177	−0.0827**	−0.0694*	−0.366*
	(0.0339)	(0.0354)	(0.184)	(0.0343)	(0.0367)	(0.188)
Δeij,t−2	−0.0151	−0.0393	0.167	−0.0419	−0.0679**	0.121
	(0.0307)	(0.0318)	(0.190)	(0.0270)	(0.0277)	(0.165)
$\mathrm{\Delta }{e}_{ij,t}\times (S_j+S_j^{€})$			−0.629**			−0.668***
			(0.260)			(0.253)
$\mathrm{\Delta }{e}_{ij,t-1}\times (S_j+S_j^{€})$			0.160			0.348*
			(0.218)			(0.207)
$\mathrm{\Delta }{e}_{ij,t-2}\times (S_j+S_j^{€})$			−0.207			−0.203
			(0.221)			(0.196)
Δe$j,t		−0.703***	−0.663***		−0.695***	−0.316***
		(0.0894)	(0.115)		(0.0806)	(0.103)
Δe$j,t−1		−0.0615	0.116		−0.00291	−0.120*
		(0.0441)	(0.0822)		(0.0410)	(0.0693)
Δe$j,t−2		0.00568	−0.257***		0.00554	0.0595
		(0.0441)	(0.0797)		(0.0400)	(0.0613)
Δe$j,t × Sj			−0.243*			−0.760***
			(0.131)			(0.121)
Δe$j,t−1 × Sj			−0.221			0.357**
			(0.159)			(0.145)
Δe$j,t−2 × Sj			0.446***			0.00131
			(0.131)			(0.108)
Δe€j,t	−0.244***	0.264***	0.302***	−0.179***	0.320***	0.427***
	(0.0525)	(0.0870)	(0.110)	(0.0413)	(0.0759)	(0.0960)
Δe€j,t−1	0.117***	0.119***	0.218***	0.170***	0.145***	0.00391
	(0.0416)	(0.0454)	(0.0741)	(0.0374)	(0.0406)	(0.0779)
Δe€j,t−2	−0.0147	0.0161	−0.116*	−0.0295	0.00521	0.0250
	(0.0377)	(0.0460)	(0.0661)	(0.0333)	(0.0406)	(0.0639)
$\mathrm{\Delta }{e}_{€j,t}\times S_j^{€}$			0.0101			−0.313*
			(0.214)			(0.185)
$\mathrm{\Delta }{e}_{€j,t-1}\times S_j^{€}$			−0.115			0.194
			(0.229)			(0.197)
$\mathrm{\Delta }{e}_{€j,t-2}\times S_j^{€}$			0.560***			0.0756
			(0.200)			(0.201)
Imp. GDP	yes	yes	yes	yes	yes	yes
Agg. controls	yes	yes	yes	yes	yes	yes
Time FE	no	no	no	no	no	no
R-squared	0.072	0.075	0.085	0.068	0.071	0.078
Observations	36,269	36,269	26,129	37,437	37,437	27,109
Dyads	2,773	2,773	1,982	2,807	2,807	2,014