Vehicle Currency Pricing, Trade, and Exchange Rate Pass-Through

Transcription

1 Vehicle Currency Pricing, Trade, and Exchange Rate Pass-Through Natalie Chen Wanyu Chung Dennis Novy y Preliminary and incomplete please do not quote or cite January 16, 2017 Abstract Using highly disaggregated rm-level data for UK imports and exports with non-eu countries between 2010 and 2013, this paper studies the relationship between invoicing currency choices and the response of prices and quantities of traded goods to exchange rate changes. On average, we nd that exchange rate pass-through into import prices is low at 25.8 percent, but varies substantially across invoicing currency choices. In addition, for the transactions invoiced in a vehicle currency, which is a widely used strategy among UK rms, we nd that the pass-through of the vehicle currency exchange rate is 49 percentage points larger compared to the pass-through of the bilateral exchange rate between the two trading partners. Our results therefore contribute to understanding the exchange rate disconnect puzzle whereby world import prices do not respond much to exchange rate movements. They also have implications for the setting of monetary policy and for understanding the dynamics of the trade balance. JEL Classi cation: F1, F31, F41. Keywords: Invoicing Currency; Vehicle Currency Pricing; Exchange Rate Pass-Through; Expenditure Switching; Trade Balance. Disclaimer: This study uses statistical data from Her Majesty s Revenue and Customs (HMRC) which is Crown Copyright. The research data sets used may not exactly reproduce HMRC aggregates. The use of HMRC statistical data in this study does not imply the endorsement of HMRC in relation to the interpretation or analysis of the information. y Natalie Chen, Department of Economics, University of Warwick, Coventry, CV4 7AL, UK and CEPR. N.A.Chen@warwick.ac.uk (corresponding author); Wanyu Chung, Department of Economics, University of She eld, She eld, S1 4DT, UK. Wanyu.Chung@sheffield.ac.uk; and Dennis Novy, Department of Economics, University of Warwick, Coventry, CV4 7AL, UK and CEPR. D.Novy@warwick.ac.uk.

2 1 Introduction One fact that is well established in the open economy macroeconomics literature is that the prices of internationally traded goods only react modestly to changes in exchange rates. In other words, the pass-through of exchange rate changes into import and domestic prices is incomplete. Factors that lead to incomplete pass-through include the decision of exporters to price in a foreign currency in combination with price stickiness (e.g., Gopinath, Itskhoki, and Rigobon, 2010), the pricing-to-market behavior of exporters when they di erentially adjust their markups across destinations in response to exchange rate movements (e.g., Knetter, 1989, 1993; Krugman, 1987), and the presence of local distribution costs in the destination market (Burstein, Neves, and Rebelo, 2003; Corsetti and Dedola, 2005). 1 One implication of partial pass-through is that the expenditure switching e ect of exchange rate uctuations is limited. That is, changes in exchange rates do not generate much substitution between domestic and foreign goods because their relative prices vary only very little (e.g., Burstein and Gopinath, 2014; Engel, 2002). In a large class of models with nominal rigidities, the currency in which traded goods are priced determines mechanically the degree of exchange rate pass-through and the extent of expenditure switching. 2 On the one hand, if prices are set in the exporter s currency ( Producer Currency Pricing, or PCP), the pass-through into import prices is complete. Exchange rate movements therefore allow relative prices to adjust, and this in turn induces a substitution between domestic and foreign goods. On the other hand, if prices are set in the importer s currency ( Local Currency Pricing, or LCP), exporters price-to-market and pass-through is zero. As a result, there is no expenditure switching. Ultimately, when prices adjust in the long run, the currency of invoicing no longer matters for the response of prices and quantities to changes in exchange rates (e.g., Gopinath, 2016; Gopinath et al., 2010). Using highly disaggregated rm-level data for UK imports and exports with non-eu countries, this paper examines the relationship between the currency of invoicing and the response of traded goods prices and quantities to changes in exchange rates. Our contribution is threefold. First, we estimate the sensitivity of prices and quantities to exchange rate uctuations by currency of invoicing. In contrast to papers that only consider pricing in producer or local currencies, we also allow for invoicing in a third, or vehicle, currency ( Vehicle Currency Pricing, or VCP). 3 The relation between vehicle currency pricing and pass-through has been rarely studied in the literature because the share of trade priced in a third currency is negligible for countries such as the US (e.g., Gopinath et al., 2010), and the unavailability of bilateral invoicing currency data precludes many papers from distinguishing between pricing in a vehicle or in the partner s currency (e.g., Gopinath, 2016). 4 5 Second, while 1 Exchange rate pass-through is the percentage change in local currency import prices resulting from a one percent change in the exchange rate between the exporting and importing countries. Additional factors behind incomplete passthrough include simultaneous importing and exporting at the rm level, see further below. For reviews of the literature, see Burstein and Gopinath (2014) and Goldberg and Knetter (1997). 2 For instance, Obstfeld and Rogo (1995) assume producer currency pricing, Betts and Devereux (2000) consider local currency pricing, while Devereux and Engel (2003) allow for both types of invoicing choices. 3 As documented by Casas, Díez, Gopinath, and Gourinchas (2016), the vast majority of world trade is invoiced in a small number of currencies, rejecting the idea that prices in international markets are set in local or producer currencies. 4 Some papers (e.g., Goldberg and Tille, 2008, 2016) study vehicle currency pricing, but not in relation to pass-through. 5 Exceptions are Casas et al. (2016) and Fabling and Sanderson (2015). 1

3 most papers study price behavior only, we explore the response of both prices and quantities to exchange rate movements. Finally, by simultaneously analyzing exports and imports, we recognize that exporters are also importers (Amiti, Itskhoki, and Konings, 2014; Casas, Díez, Gopinath, and Gourinchas, 2016). To the best of our knowledge, our paper is the rst to simultaneously investigate the relation between three di erent invoicing choices and the exchange rate response of highly disaggregated prices and quantities of imports and exports. To this aim, we focus our analysis on the UK economy. There are several reasons for this choice. First, the sterling nominal exchange rate is freely oating with respect to other major currencies, and it has experienced signi cant uctuations over time. Second, we were granted access to a highly disaggregated data set from Her Majesty s Revenue and Customs (HMRC), which provides the universe of the Free on Board (FOB) export and the Cost, Insurance, and Freight (CIF) import transactions of the UK economy. For each transaction, we observe a unique trader identi er, the country of origin or destination, the date of the transaction, the 5-digit SITC Revision 3 and the 4-digit HS Revision 2007 classi cations, the 10-digit comcode product identi er (the rst eight digits correspond to the Combined Nomenclature), the value (in sterling), and the mass (in kilograms). Most importantly, for each transaction the currency of invoicing is reported from 2010 to 2013 for imports, and from 2011 to 2013 for exports, but for non-eu transactions only. As we do not directly observe prices, we compute FOB export and CIF import unit values at a quarterly frequency as a proxy for export and import prices at the trader-product-currency-destination/origin levels, using data on the value and the volume exported or imported. Finally, and crucially for our purposes, we observe large shares of import and export transactions invoiced in a vehicle currency. For imports, VCP accounts for 56 percent of non-eu transactions, whereas PCP and LCP represent 20 and 24 percent, respectively. For exports, the share of VCP is lower, but still accounts for 34 percent of non-eu transactions, whereas PCP and LCP amount to 50 and 16 percent, respectively. As we only observe import prices into a single country, we are unable to identify whether the response of prices to changes in exchange rates is due to changes in the markups or to changes in the marginal costs of the exporters (Gopinath et al., 2010). As a result, our estimates only measure exchange rate pass-through into import prices. In contrast, for exports we can identify the pricingto-market behavior of exporters by controlling for changes in marginal costs through the inclusion of trader-product-time xed e ects (e.g., Knetter, 1989, 1993). Our main results can be summarized as follows. For imports we nd that, on average, exchange rate pass-through is incomplete and low at 25.8 percent. We demonstrate, however, that pass-through varies substantially across invoicing choices. Pass-through is large at 75.8 percent for imports in producer currencies and insigni cantly di erent from zero for local currency (sterling). Pass-through is low at 21.6 percent for vehicle currencies when we use the standard bilateral exchange rate between trading partners. However, once we allow import prices invoiced in vehicle currencies to depend on the exchange rate with the vehicle currency rather than the bilateral exchange rate, we nd that pass-through is large at 74.3 percent, and thus in the same ballpark as for producer currency pricing. 6 This nding indicates that using the bilateral rather than the vehicle currency exchange rate underestimates pass-through for goods priced in vehicle currency by over 50 percentage points. 6 Pass-through is low at 18.9 percent for the exchange rate between the vehicle currency and the producer s currency. 2

4 Moreover, it is consistent with the premise that prices are sticky in their currency of invoicing. [For exports: TBC.] Given the strong presence of vehicle currency pricing in international trade, our results therefore contribute to understanding the exchange rate disconnect puzzle whereby world import prices do not respond much to exchange rate movements. Our research shows that bilateral exchange rates are inappropriate to evaluate exchange rate pass-through for traded goods invoiced in vehicle currencies. For policy purposes, this indicates that ignoring the currency of invoicing can lead to misleading predictions regarding the e ects of bilateral exchange rate changes on import prices and, in turn, on domestic in ation. Our results have thus implications for the setting of monetary policy. To forecast how currency uctuations a ect future in ation, central banks typically rely on rules of thumb based on estimates of exchange pass-through into import prices. For instance, the Bank of England considers that the pass-through into UK import prices is between 60 and 90 percent. As the import component of the consumer price index is about 30 percent, the pass-through to consumer prices is between 20 and 30 percent. In other words, a ten percent depreciation of sterling should increase UK consumer prices by two to three percent (Forbes, 2015; Forbes, Hjortsoe, and Nenova, 2015). Taking into account the currency of invoicing, our estimates instead suggest that a ten percent depreciation of sterling translates into seven to eight percent higher import prices invoiced in producer or vehicle currencies, and has no e ect on import prices invoiced in local currencies [TBC]. We therefore argue that policymakers should update their rules of thumb (Forbes et al., 2015). They should regularly revise their estimates by considering, among other factors, that imports are priced in di erent currencies, and that the shares of each currency are likely to change over time. It also matters against which other currencies sterling depreciates since some are more frequently used as vehicle currencies than others. For the UK, understanding the e ects of currency uctuations is of foremost importance as in June 2016 the country voted to leave the EU. At the time of writing (January 2017), in a period of just six months, sterling has depreciated by 17 percent against the US dollar, and by 15 percent against the euro. It is a key question for policymakers to understand to what extent this depreciation will translate into higher domestic in ation, lower imports and stronger exports. Also, the type of trade deals that the UK will strike with other countries after leaving the EU and whether the UK will stop participating in the Single Market could potentially lead to a shift in the invoicing currencies of traded goods. According to our results, a change in the composition of invoicing currencies is likely to a ect the sensitivity of prices and quantities to exchange rate movements, both on the import and export sides. [TBC] In addition, our results have implications for understanding the dynamics of the trade balance in response to exchange rate movements. Our baseline estimates indicate that, when not taking the currency of invoicing into account, exports rise by X percent and imports fall by X percent in response to a ten percent depreciation. Once we consider the role of currencies of invoicing, these numbers become X and X percent. [TBC] This paper builds on, and contributes to, three strands of the literature. The rst one is the literature on incomplete pass-through and pricing-to-market. These papers usually nd a low degree 3

5 of pass-through into import prices (e.g., Campa and Goldberg, 2005; Gopinath and Rigobon, 2008), or consumer prices (e.g., Campa and Goldberg, 2010). A number of recent papers relate pass-through to rm- or product-level characteristics. Amiti et al. (2014) document that Belgian exporters with high import shares and high export market shares have a lower pass-through to destination prices. Berman, Martin, and Mayer (2012) show that highly productive French exporters price-to-market more, leading to lower pass-through in the destination market. 7 Within this literature, papers most closely related to our work are those that investigate the relationship between the invoicing currency and exchange rate pass-through. Due to the di culty of obtaining invoicing currency data, especially at a disaggregated level, the available evidence is, however, limited. 8 Gopinath et al. (2010) provide evidence that pass-through is low for disaggregated US imports priced in US dollars, and large when priced in non-dollars. Gopinath (2016) shows that this pattern also holds in other countries as pass-through rises with the share of imports invoiced in a foreign currency. Casas et al. (2016) use rm-level price and quantity data for Colombian exports and imports but observe the currency of invoicing for exports only. Interestingly, more than 98 percent of exports are priced in US dollars. 9 They show that, for non-dollarized economies, controlling for the peso to US dollar exchange rate knocks down the e ect of the bilateral exchange rate in explaining both the prices and the quantities of exports and imports. To the best of our knowledge, Fabling and Sanderson (2015) are the only authors to allow for vehicle currency pricing (based on export data for New Zealand), and they nd that pass-through is high for rm-level exports invoiced in domestic currency, and low when priced in local and vehicle currencies. Cravino (2014) shows that a nominal appreciation increases the prices, and reduces the quantities, of rm-level exports invoiced in the exporter s currency, but has no e ect when priced in the destination s currency. 10 We di er from these papers in that we study the behavior of export and import prices simultaneously. Import prices allow us to assess the sensitivity of domestic in ation to exchange rate shocks, while export prices shed light on the pricing strategies of exporters. We also emphasize the role of vehicle currency pricing in impacting pass-through, and in that context scrutinize the response of both prices and quantities to exchange rate shocks. 11 Second, our paper is related to studies that investigate the factors in uencing invoicing currency choices. 12 At the country level, Goldberg and Tille (2008) provide evidence that country size matters, whereas hedging considerations and transaction costs play a minor role. At a disaggregated level, 7 Other recent papers that study pass-through using disaggregated data include Auer and Schoenle (2016), Bernini and Tomasi (2015), Chatterjee, Dix-Carneiro, and Vichyanond (2013), Chen and Juvenal (2016), Fitzgerald and Haller (2014), Gopinath and Itskhoki (2010), Gopinath and Rigobon (2008), and Nakamura and Zerom (2010), among others. 8 A large body of the literature on currency of invoicing and pass-through is, indeed, theoretical. Engel (2006) and Gopinath et al. (2010) develop models where a rm s desired pass-through determines the currency of invoicing. In Bacchetta and van Wincoop (2003), currency choice explains why the pass-through into consumer prices is lower than into import prices. Also see Choudhri and Hakura (2012), and Devereux, Engel, and Storgaard (2004). 9 These authors de ne the Dominant Currency Paradigm as the case where rms price exports in a dominant currency, in general the US dollar. Dominant currency pricing therefore combines vehicle and local currency invoicing. 10 See, also, Devereux, Tomlin, and Dong (2015) who show that the market shares of both exporting and importing rms impact exchange rate pass-through and the currency of invoicing. 11 Fabling and Sanderson (2015) allow for vehicle currency pricing, but only study prices on the export side. Cravino (2014) looks at quantities in addition to prices, but only considers exports and ignores vehicle currency pricing. 12 For models that investigate the determinants of currency choice, see Bacchetta and van Wincoop (2005), Devereux et al. (2015), Engel (2006), Friberg (1998), Goldberg and Tille (2008), and Gopinath et al. (2010), among others. 4

6 Goldberg and Tille (2016) analyze import transactions by size and conclude that the invoicing choice results from a bargaining process between trading partners. Lyonnet, Martin, and Méjean (2016) document that exporters using nancial instruments to hedge against exchange rate risk are more likely to price in a foreign currency. Using UK data, Chung (2016) shows that exporters relying more on foreign currency-denominated imported inputs are less likely to invoice in their home currency. 13 In this paper, we do not explain the currency of invoicing choice. Instead, we investigate how invoicing strategies and the patterns of pass-through and of expenditure switching interact with each other. Finally, as partial pass-through leads to deviations from the Law of One Price, our paper is also part of a wider literature on product-level real exchange rates. 14 Recent studies that rely on disaggregated product-level data include Cavallo, Neiman, and Rigobon (2014) who nd that the Law of One Price holds very well within currency unions, and Burstein and Jaimovich (2012) who show that product-level real exchange rates across countries are more volatile than nominal exchange rates. The paper is organized as follows. Section 2 describes rm-level imports and exports, and the balance sheet and macroeconomic data we use. It also provides descriptive statistics. Section 3 presents our main empirical results while Section 4 investigates pass-through to consumer prices. Section 5 provides robustness checks while Section 6 analyzes the determinants of exchange rate pass-through mismeasurement. Section 7 concludes. 2 Data and Descriptive Statistics Our data set combines information from three di erent sources: transaction-level customs data, rmlevel balance sheet information, and macroeconomic data. 2.1 Customs Data Transaction-level export and import data for the UK are obtained from Her Majesty s Revenue and Customs (HMRC), a non-ministerial Department of the UK government responsible for the collection of taxes, the payment of state support, and the collection of trade in goods statistics. Access to the data is only granted to approved projects, and all empirical output is subject to HMRC s code of statistical disclosure. For each FOB export and CIF import transaction, the data set provides us with a unique trader identi er, the country of destination or origin, the transaction date, the 5-digit SITC Revision 3 and the 4-digit HS Revision 2007 classi cations, the 10-digit comcode product code (the rst eight digits correspond to the Combined Nomenclature), the value (in sterling), the mass (in kilograms) and, most importantly, the currency of invoicing, but for non-eu transactions only. 15 While the trade data are available from 1996 to 2013, we concentrate our analysis on the period because reporting the currency of invoice has only become an obligation since 2010 for non-eu imports, 2011 for non-eu 13 At a disaggregated level, see also Donnenfeld and Haug (2003), Friberg and Wilander (2008), or Ito, Koibuchi, Sato, and Shimizu (2010, 2013, 2016). On the euro as an invoice currency, see Kamps (2006) and Ligthart and Werner (2012). 14 Producer currency pricing implies that the Law of One Price holds, while it fails under local and vehicle currency pricing (e.g., Devereux and Engel, 2003). 15 Data collection for trade with EU countries is not compulsory and therefore not available at the level of transactions. In general, the currency of invoice and the currency of settlement are the same (Friberg and Wilander, 2008). 5

7 exports, and for transaction values exceeding 100,000 only. 16 In our data set, non-eu imports and exports represent 50 and 54 percent of total UK imports and exports, respectively. Given that actual export and import prices are not available, we compute the unit values of exports and of imports as the ratio between the value of the transaction in sterling and the corresponding volume in kilograms. As we rely on unit values, we are unable to observe when rms adjust their prices. We clean up the data in several ways. First, we drop the few transactions for which the currency of invoicing is missing. Second, as we are interested in the behavior of rms which have some pricing power, we exclude homogeneous commodities from the sample whose prices are determined by world supply and demand (Gopinath, 2016, TBC). Third, we drop the observations where the value of exports or imports is indicated as positive but the corresponding volume is zero. Finally, we aggregate the data at a quarterly frequency. 2.2 Balance Sheet Data [TBC] 2.3 Macroeconomic Data The consumer price indices (CPI) and nominal exchange rates are from the International Financial Statistics (IFS) of the International Monetary Fund (IMF). As shown in Figure 1, between January 2010 and December 2013 the UK experienced nominal and real exchange rate uctuations between sterling and the US dollar exceeding a range of ten percent. The real exchange rate is de ned as the ratio of consumer price indices times the monthly nominal exchange rate, where an increase indicates a real depreciation of sterling. On a monthly basis, sterling depreciated the most relative to the US dollar in February 2010 (by 6.01 and 5.57 percent in nominal and real terms), while the largest appreciation occurred in September 2013 (of 4.09 in nominal and 4.28 percent in real terms). 2.4 Descriptive Statistics As shown in Table 1, our import sample between 2010 and 2013 includes 82,131 rms, 9,812 products (at the 10-digit comcode level), and 134 origin countries with a total of observations. These rms import an average of 4 5 di erent products from 1 81 origin countries (at the 5 and 95 percentiles, the products per importer are 1 and 14, and the origin countries per importer are 1 and 5). 17 The mean import transaction is valued at pound sterling, or 834 pound sterling per kilogram. The mean change in import unit values is equal to percent per quarter (at the 5 and 95 percentiles, the mean changes are and percent). For exports between 2011 and 2013, we have fewer rms ( exporters) and products (5 842) but more destination countries (181) with a total of observations. On average, these rms 16 However, our data set contains transactions below the 100,000 threshold because in practice, most rms report all transactions even if they fall below the threshold. Overall, for around ve percent of imports and ten percent of exports, the currency of invoicing is missing. 17 Due to con dentiality reasons, we are unable to report the maximum and minimum values. 6

8 export 2 95 di erent products to 3 01 destination countries (at the 5 and 95 percentiles, the products per exporter are 1 and 8, while the destinations per exporter are 1 and 10). Exporters charge on average pound sterling per transaction, or pound sterling per kilogram. The mean change in export unit values is equal to percent per quarter (at the 5 and 95 percentiles, the mean changes are and percent). The two samples cover a large range of origin and destination countries that di er in terms of economic development, including OECD countries such as Canada, Switzerland and the US, but also emerging markets such as China, India, Nigeria and Vietnam, as well as developed Asian countries such as Hong Kong and Japan. 18 As shown in Table 2, China and the US are the top markets for imports and exports, respectively. We remind the reader that our sample does not include EU countries, and those therefore do not appear in Table 2. Table 3 reports descriptive statistics for imports and exports by invoicing currency. For imports, vehicle currency pricing represents the largest share of the sample (in terms of number of observations, number of importers, number of products, number of origin countries, and the value share of imports in the sample). In particular, the value share of imports invoiced in a vehicle currency amounts to percent, whereas the shares in producer or local currencies are only and percent. In total, 76 di erent vehicle currencies are used, but percent of the value of these transactions is in US dollars, and 9 04 percent in euros. 19 In terms of transaction counts, these correspond to shares of 88 and 10 8 percent, respectively. On average, unit values are the highest for goods priced in producer currency at a value of pound sterling per kilogram. For exports, the picture is somewhat di erent. The largest share of exports is invoiced in producer currency (sterling) at percent, followed by percent in vehicle currency, and percent in local currency. A total of X di erent vehicle currencies are used. Similarly to imports, percent of the value of these transactions is in US dollars, and percent in euros. In terms of transaction counts, these correspond to shares of 76 and 22 3 percent. Unit values are the highest for vehicle currency priced goods at 3, pound sterling per kilogram. It is interesting to note that rms use a mix of di erent pricing strategies when trading with di erent countries. [TBC] The upper left panel of Table 4 reports the average import shares by invoicing currency and industry (at the 1-digit SITC level). Vehicle currency pricing is the dominant strategy for most sectors, and its share is the largest for Mineral fuels (90 67 percent) and Animal and vegetable oils (85 08 percent), both of which tend to produce homogeneous goods (Gopinath et al., 2010; Goldberg and Tille, 2008). The vehicle currency share is also large at percent for Manufactured goods. Instead, local currency pricing is the most widely adopted strategy for Beverages and tobacco at percent, while producer currency pricing is the least used among most sectors (with the exception of Beverages and tobacco, Animal and vegetable oils, and Not classi ed ). The lower left panel of the table splits the data by region of origin. With the exception of the US, vehicle currency pricing is the dominant strategy for all regions, and varies from percent for Asia to percent for other 18 [Which countries are included for exports but not for imports? There are 47 of them.] 19 The other main vehicle currencies include the Japanese yen and the Canadian dollar [TBC]. 7

9 regions (which include Oceania, the Middle East, and Africa). Given that the US mostly exports in US dollars (Goldberg and Tille, 2008), UK imports from the US are mainly invoiced in the producer s currency (the producer share is percent). For exports, with the exception of Manufactured goods which are largely invoiced in vehicle currencies, producer currency pricing (sterling) is instead the dominant strategy for all industries, consistent with Table 3. Its share varies from percent for Chemicals to percent for Animal and vegetable oils. It is also the most widely used strategy for exports to the US, China, and Europe (excluding the EU), whereas vehicle currency pricing dominates for Asia, other Americas, and other regions. [Add description of intensive versus extensive margin. Add descriptive stats using the FAME data.] 3 Empirical Analysis This section describes the various speci cations we estimate to evaluate the degree of exchange rate pass-through into import prices by currency of invoicing. We then turn to the case of export prices. 3.1 Import Prices To compare exchange rate pass-through in our sample with the estimates reported in the literature, we estimate a standard pass-through regression (e.g., Gopinath et al., 2010): P ln = P ln (1) =0 where is the import unit value of product (de ned at the comcode level) imported by rm from country in quarter, expressed in sterling per kilogram. It is our proxy for import prices. The bilateral exchange rate between sterling and the currency of country in quarter is denoted by (an increase in indicates a bilateral depreciation of sterling), and is the quarterly foreign in ation rate calculated using the consumer price index. We include up to eight lags for the nominal exchange rate and the foreign in ation rate, where is the number of lags. Given our quarterly data this corresponds to lags of up to two years. is the rst di erence operator, and is an error term. We include rm-quarter,, as well as product-destination xed e ects,. Shortrun pass-through is given by the coe cient 0 on the contemporaneous change in the exchange rate, whereas the cumulative estimate ( ) P =0 evaluates long-run pass-through. Given the level of disaggregation of the data, changes in bilateral exchange rates are assumed to be exogenous to the import prices faced by individual rms. As a benchmark, we rst estimate equation (1) on the full sample of imports. Next, to investigate whether invoicing currency choices are associated with di erent pass-through rates, we regress equation (1) separately on three subsamples of import transactions invoiced in producer, local, and vehicle currencies. =0 8

10 For the transactions invoiced in producer or local currencies, it is intuitive to regress the sterling import price on the bilateral exchange rate between sterling and the origin country s currency, as we do in equation (1). For the transactions priced in vehicle currencies, we would instead expect that it is the exchange rate with the vehicle currency that matters. One way to explore this possibility is to decompose the bilateral exchange rate in equation (1) as follows (Fabling and Sanderson, 2015): ln ln = ln + ln (2) where in this case is the bilateral exchange rate between sterling, or the destination country s currency ( ) and the currency of the origin country ( ). Its change can be decomposed into the change of the sterling to vehicle currency exchange rate, and the change of the vehicle to origin country s currency exchange rate. For the transactions priced in vehicle currencies, we can then estimate: ln P = P ln + P ln =0 =0 (3) where we allow for separate coe cients and for the two exchange rates. If prices are sticky in the currency in which they are invoiced, we would expect pass-through into import prices to be larger when sterling uctuates against the vehicle currency. As an alternative, we run the following speci cation on the full sample of import transactions: P P ln = ln (4) =0 where is a dummy variable for the three invoicing currency choices. When estimating equation (4), we rst include the bilateral exchange rate for all transactions. Then, for the vehicle currency transactions, we decompose the bilateral exchange rate according to equation (2). Short-Run Pass-Through We start by analyzing short-run exchange rate pass-through. We estimate equations (1), (3), and (4), but only report and discuss the contemporaneous exchange rate elasticities. Column (1) of Table 5 reports the results of estimating equation (1) on the full sample of imports, as is typically done in the literature. The coe cient on the contemporaneous change in the exchange rate is equal to 0.258, and is signi cant at the one percent level. In response to a ten percent bilateral depreciation, import prices (in sterling) rise by 2.58 percent, therefore pass-through for import prices is low at 25.8 percent. This nding is consistent with other papers that nd a low degree of exchange rate pass-through into import prices (e.g., Campa and Goldberg, 2005; Gopinath and Rigobon, 2008). To investigate whether invoicing currency choices are associated with di erent pass-through rates, we then regress equation (1) separately on three subsamples of import transactions invoiced in producer, local, and vehicle currencies. 20 The results are reported in columns (2) to (4), respectively, and 20 Focusing on the UK economy, Forbes et al. (2015) show that pass-through varies substantially depending on the nature of the shocks that move the exchange rate in the rst place. Allowing for endogenous exchange rate changes can =0 =0 9

11 show that exchange rate pass-through varies substantially across invoicing choices. Pass-through is large at 75.8 percent for producer pricing (column 2), low at 21.6 percent for vehicle pricing (column 4), and is insigni cantly di erent from zero for local pricing (column 3). These results highlight that estimating a single pass-through coe cient across all transaction types, as in column (1), hides a signi cant amount of heterogeneity in the pass-through elasticities across invoicing choices. They are in line with the ndings of Fabling and Sanderson (2015) who show that, for exports, pass-through is high for rm-level exports invoiced in domestic currency, and low when priced in local and vehicle currencies. They are also consistent with other studies showing that local currency priced goods typically display a low response to exchange rate movements (references). Next, in column (5), for the subsample of vehicle currency transactions, we regress equation (3) and decompose the bilateral exchange rate according to equation (2). This exercise has a dramatic e ect on exchange rate pass-through. Pass-through is large at 74.3 percent for the sterling to vehicle currency exchange rate, which is similar in magnitude to the pass-through for producer priced transactions in column (2) at 75.8 percent. Instead, it is low at 18.9 percent for the vehicle to origin country s currency exchange rate. 21 These ndings are consistent with prices being sticky in the currency in which they are invoiced (Fabling and Sanderson, 2015). Finally, Table 6 reports the results of estimating equation (4). Consistent with Table 5, column (1) shows that, relative to the bilateral exchange rate, pass-through is relatively large (at 58.6 percent) for producer currency pricing, low (at 30.3 percent) for vehicle currency pricing (we can reject at the ve percent level that the two elasticities are equal), and zero for local currency pricing. Once we decompose the bilateral exchange rate for vehicle currency priced goods, column (2) shows that passthrough is large, and of similar magnitude, for the producer priced and for the vehicle currency priced transactions relative to the sterling to vehicle currency exchange rate (the estimated coe cients are equal to and 0.797, respectively, and are not signi cantly di erent from each other). In contrast, the response of import prices invoiced in third currencies to changes in the vehicle to origin country s currency exchange rate is low at 16.7 percent, and zero for local currency priced goods when the bilateral exchange rate uctuates. Our contribution is twofold. First, we show that exchange rate pass-through varies substantially across invoicing currency choices. This nding contrasts with the low degree of pass-through that is typically estimated in the literature. This means that for policy purposes, ignoring the currency of invoicing can lead to misguided predictions regarding the e ects of bilateral exchange rates on import prices and, in turn, on domestic in ation. Second, by comparing columns (1) and (2) of Table 6, we show that using the bilateral rather than the sterling to vehicle currency exchange rate underestimates pass-through for goods priced in third currencies by 49.4 percentage points ( = 49 4 percent). Given the strong presence of vehicle currency pricing in international trade, our results therefore contribute to understanding the exchange rate disconnect puzzle whereby world import prices do not respond much to exchange rate movements. therefore, they claim, explain why pass-through rates change over time. In our paper, we do not allow for pass-through to be shock-dependent. We argue, however, that this should not a ect our analysis as the pass-through rates we estimate by currency of invoicing are all impacted by the same mix of exchange rate shocks over the period. 21 We can reject at the ve percent level that the two estimated coe cients are equal. 10

12 Long-Run Pass-Through Due to the inclusion of eight lags of the exchange rate change in our regressions, we depict long-run pass-through graphically. [TBC] 3.2 Export Prices To analyze exchange rate pass-through for export prices, we estimate the following speci cation: P ln = P ln + P (5) =0 =0 where is the export unit value of product exported by rm to country in quarter, expressed in sterling per kilogram. In contrast to equation (1), now denotes the destination country for exports, rather than the origin country of imports. In addition to controlling for the foreign in ation rate, we also control for the growth of GDP in the destination country,, included contemporaneously and with one lag (e.g., Gopinath et al., 2010). To identify the pricing-to-market behavior of exporters, and therefore the response of markups to changes in exchange rates, we need to control for changes in marginal costs. As marginal costs are product and exporter-speci c, we control for exporter-product-quarter xed e ects,. In contrast to imports, recall that the share of exports priced in a vehicle currency is percent only. 3.3 Import Volumes [TBC] 3.4 Export Volumes [TBC] 3.5 Firm-Level Characteristics [TBC] =0 4 Pass-Through to Consumer Prices It is a common notion that changes in exchange rates a ect import prices fairly quickly and within a year. However, the e ect of changes in import prices on the domestic price level, and therefore on in ation, is instead believed to be quite slow and to take between three and ve years (Forbes, 2015). As consumer prices contain a higher non-traded component compared to import prices, due for instance to local distribution costs, and also include the prices of traded goods that are sold only domestically, exchange rate pass-through into consumer prices is generally lower than into import prices (Burstein and Gopinath, 2014; Gopinath, 2016). [TBC] 11

13 5 Robustness Import Prices To ensure the robustness of our ndings, this section discusses our results using alternative data samples. We estimate equation (4), and for simplicity only report short-run passthrough estimates. 22 Overall, the broad similarity of the results supports the paper s main conclusions. Pass-through is large for producer currency transactions, and zero for local currency priced goods. For vehicle currency transactions, the response of import prices is large when the sterling to vehicle currency exchange rate uctuates, but low when the vehicle to origin s currency exchange rate changes. As commodity prices are determined by demand and supply conditions in world markets, whereas we are essentially interested in the behavior of rms which have some pricing power (Gopinath, 2015), in columns (1) and (2) of Table 7 we focus on manufacturing industries only (SITC 6 8), and exclude raw materials from the sample (SITC 2 3), respectively. In column (3), we exclude the US from the sample as its exports and imports are mostly in US dollars. In column (4), we restrict the sample to the goods produced in the origin country (and exclude the ones produced in third countries). Using information on the end use of goods as provided by the BEC classi cation, columns (5), (6), and (7) restrict the sample to intermediate, nal, and capital goods, respectively. The results remain qualitatively similar, with the exception of capital goods which prices react to exchange rate movements only when invoiced in the producer s currency. Export Prices [TBC] 6 The Determinants of Pass-Through Mismeasurement In this section, we rst formalize the measurement error of pass-through estimates with a model of monopolistic competition under exible price setting, based on Engel (2006). We then test the determinants of pass-through mismeasurement in our data [TBC]. 6.1 Theory Unconditional Pass-Through Elasticity We consider a monopolistic rm that sells to the UK market. We begin with the case where the rm can choose the price for its product with full information about demand, costs, and so forth. The representative rm is a exible-price setter that chooses, the log of the foreign currency price of its export to maximize the twice-di erentiable concave pro t function ( ) is a vector of variables that a ect the rm s pro ts but are exogenous to the rm. Engel (2006) proves that the unconditional pass-through elasticity the coe cient on the projection of on (the log of the exchange rate expressed as the exporter s currency price of the home currency) when prices are set exibly equals: ^ 1 = 0 (~ ¹ ) ( ) (~ ¹ ) 2 [0 1] (6) ( ) where ~ is the optimal price expressed in the home currency indexed by and ¹ is the mean of x around which we linearize the pro t function. 22 The corresponding robustness checks for long-run pass-through are available upon request. 12

14 6.1.2 Vehicle Currency Pricing Next, we consider the same exible-price case conditional on the rm s consideration of the exchange rate vis-à-vis a vehicle currency, denoted as (the log of the exchange rate expressed as the exporter s currency price of the vehicle currency). The unconditional pass-through elasticity the coe cient on the projection of ~ on becomes: ^ 2 = 0 (~ ¹ ) (~ ¹ ) 2 [0 1] (7) Note that ^ 2 is the unconditional pass-through elasticity into the price in a vehicle currency. Since the exporter only responds to the exchange rate the pass-through elasticity into the price in the importer s currency is exactly ^ 2 (i.e., full pass-through from the vehicle currency price into the importer s price when uctuates) Measurement Error Note that we are interested in the measurement error of exchange rate pass-through for VCP, which arises due to the use of bilateral exchange rates rather than the actual rates that the exporters react to. We develop the following proposition. Proposition 1 The real pass-through elasticity ^ 2 is measured as ^ 1 with. A measurement error occurs as ª j(^ 1 ^ 2 ) ^ 1 j: ª n "1 + 2 # (8) ( ) ( ) Also, ª is higher with: ( ² Higher ) ( ) Note here that the two exchange rates are expressed from country 0 s perspective ( = ). The gap is higher when exporters costs are more linked to the vehicle currency, relative to sterling. ² Lower ( ) ( ) When the bilateral exchange rate is more volatile than ² Higher When the two exchange rates are highly correlated. [TBC] 7 Concluding Remarks One limitation of our analysis is that we do not observe the currency of invoicing for EU imports and exports, which in 2015 represented 53 and 44 percent of total UK imports and exports, respectively. 23 Using aggregate data on the currency of invoicing we can, however, discuss some related evidence on how the prices and the quantities of total UK imports and exports respond to changes in exchange 23 Total and EU exports and imports are from the Direction of Trade Statistics of the International Monetary Fund. 13

15 rates. According to Gopinath (2016), total UK imports are priced in euros (15 percent), US dollars (47 percent), and sterling (32 percent). While the LCP share is 32 percent, we cannot recover the import shares in producer or vehicle currencies as the data are not available by country of origin. However, given that a proxy for a country s in ation sensitivity to exchange rate movements is the fraction of its imports invoiced in a foreign currency (Gopinath, 2016), these data suggest that with 62 percent of imports invoiced in a foreign currency, UK in ation is not much insulated from exchange rate shocks. A depreciation should indeed make around two-thirds of UK imports more expensive, which helps to improve the trade balance. Besides, total UK exports are priced in euros (13 percent), US dollars (28 percent), and sterling (51 percent). Therefore, a depreciation should make about half of UK exports (in sterling) cheaper, while for the other half it mainly raises markups and hence pro ts. [TBC] 14

16 References Amiti, M., Itskhoki, O., Konings, J., Importers, exporters, and exchange rate disconnect. American Economic Review 104 (7), Auer, R.A., Schoenle, R.S., Market structure and exchange rate pass-through. Journal of International Economics 98, Bacchetta, P., van Wincoop, E., Why do consumer prices react less than import prices to exchange rates? Journal of the European Economic Association 1 (2/3), Bacchetta, P., van Wincoop, E., A theory of the currency denomination of international trade. Journal of International Economics 67 (2), Berman, N., Martin, P., Mayer, T., How do di erent exporters react to exchange rate changes? Quarterly Journal of Economics 127 (1), Bernini, M., Tomasi, C., Exchange rate pass-through and product heterogeneity: Does quality matter on the import side? European Economic Review 77, Betts, C., Devereux, M.B., Exchange rate dynamics in a model of pricing-to-market. Journal of International Economics 50 (1), Burstein, A., Gopinath, G., International prices and exchange rates, in: Gopinath, G., Helpman, E., Rogo, K. (Eds), Handbook of International Economics, Vol. 4, Elsevier, Amsterdam, pp Burstein, A., Jaimovich, N., Understanding movements in aggregate and product-level real exchange rates. University of California, Los Angeles, mimeo. Burstein, A., Neves, J.C., Rebelo, S., Distribution costs and real exchange rate dynamics during exchange-rate-based stabilizations. Journal of Monetary Economics 50 (6), Campa, J.M., Goldberg, L.S., Exchange-rate pass-through into import prices. Review of Economics and Statistics 87 (4), Campa, J.M., Goldberg, L.S., The sensitivity of the CPI to exchange rates: Distribution margins, imported inputs, and trade exposure. Review of Economics and Statistics 92 (2), Casas, C., Díez, F.J., Gopinath, G., Gourinchas, P.-O., Dominant Currency Paradigm. NBER Working Paper Cavallo, A., Neiman, B., Rigobon, R., Currency unions, product introductions, and the real exchange rate. Quarterly Journal of Economics 129 (2), Chatterjee, A., Dix-Carneiro, R., Vichyanond, J., Multi-product rms and exchange rate uctuations. American Economic Journal: Economic Policy 5 (2), Chen, N., Juvenal, L., Quality, trade, and exchange rate pass-through. Journal of International Economics 100, Choudhri, E.U., Hakura, D.S., The exchange rate pass-through to import and export prices: The role of nominal rigidities and currency choice. IMF Working Paper WP/12/226. Chung, W., Imported inputs and invoicing currency choices: Theory and evidence using UK transaction data. Journal of International Economics 99, Corsetti, G., Dedola, L., A macroeconomic model of international price discrimination. Journal of International Economics 67 (1), Cravino, J., Exchange rates, aggregate productivity and the currency of invoicing of international trade. University of Michigan, mimeo. 15

17 Devereux, M.B., Engel, C., Monetary policy in the open economy revisited: Price setting and exchange-rate exibility. Review of Economic Studies 70 (4), Devereux, M.B., Engel, C., Storgaard, P.E., Endogenous exchange rate pass-through when nominal prices are set in advance. Journal of International Economics 63 (2), Devereux, M.B., Tomlin, B., Dong, W., Exchange rate pass-through, currency of invoicing and market share. NBER Working Paper Donnenfeld, S., Haug, A., Currency invoicing in international trade: An empirical investigation. Review of International Economics 11 (2), Engel, C., Expenditure switching and exchange rate policy. NBER Macroeconomics Annual 17, Engel, C., Equivalence results for optimal pass-through, optimal indexing to exchange rates, and optimal choice of currency for export pricing. Journal of the European Economic Association 4 (6), Fabling, R., Sanderson, L., Export performance, invoice currency and heterogeneous exchange rate pass-through. The World Economy 38 (2), Fitzgerald, D., Haller, S., Pricing-to-market: Evidence from plant-level prices. Review of Economic Studies 81 (2), Forbes, K., Much ado about something important: How do exchange rate movements a ect in ation? Speech given at the 47 Money, Macro and Finance Research Group Annual Conference, Cardi. Forbes, K., Hjortsoe, I., Nenova, T., The shocks matter: Improving our estimates of exchange rate pass-through. Bank of England Discussion Paper 43. Friberg, R., In which currency should exporters set their prices? Journal of International Economics 45 (1), Friberg, R., Wilander, F., The currency denomination of exports A questionnaire study. Journal of International Economics 75 (1), Goldberg, L.S., Tille, C., Vehicle currency use in international trade. Journal of International Economics 76 (2), Goldberg, L.S., Tille, C., Micro, macro, and strategic forces in international trade invoicing: Synthesis and novel patterns. Journal of International Economics 102, Goldberg, P.K., Knetter, M.M., Goods prices and exchange rates: What have we learned? Journal of Economic Literature 35 (3), Gopinath, G., The international price system. Jackson Hole Symposium Proceedings, forthcoming. Gopinath, G., Itskhoki, O., Frequency of price adjustment and pass-through. Quarterly Journal of Economics 125 (2), Gopinath, G., Itskhoki, O., Rigobon, R., Currency choice and exchange rate pass-through. American Economic Review 100 (1), Gopinath, G., Rigobon, R., Sticky borders. Quarterly Journal of Economics 123 (2), Ito, T., Koibuchi, S., Sato, K., Shimizu, J., Why has the yen failed to become a dominant invoicing currency in Asia? A rm-level analysis of Japanese exporters invoicing behavior. NBER Working Paper