Journal of International Economics 88 (2012) 62–73

Contents lists available at SciVerse ScienceDirect

Journal of International Economics

j ourna l homepage: www.e lsev ie r .com/ locate / j i e

Border prices and retail prices☆

David Berger a, Jon Faust b, John H. Rogers c,⁎, Kai Steverson d

a Yale University, New Have, CT, USAb John Hopkins University, Baltimore, MD, USAc Federal Reserve Board, Washington, DC, USAd Princeton University, Princeton, NJ, USA

☆ We thank Ariel Burstein, Linda Goldberg and EmiCraig Brown, Rob McClelland, Daryl Slusher, Rozi Ultheir generous assistance. The views in this paper areauthors and should not be interpreted as reflecting the vof the Federal Reserve System or of any other person aserve System.⁎ Corresponding author.

E-mail address: rogersj3@frb.gov (J.H. Rogers).

1 Specifically, the distribution wedge is PCPI−PDOCK

PCPI

0022-1996/$ – see front matter. Published by Elsevier Bdoi:10.1016/j.jinteco.2012.02.011

a b s t r a c t

a r t i c l e i n f o

Article history:Received 17 April 2009Received in revised form 15 February 2012Accepted 16 February 2012Available online 24 February 2012

JEL Classification:F30

Keywords:PricesDistributionExchange rates

We analyze retail prices and at-the-dock (import) prices of specific items in the Bureau of Labor Statistics'(BLS) CPI and IPP databases, using both databases simultaneously to identify items that are identical in de-scription at the dock and when sold at retail. This identification allows us to measure the distributionwedge associated with bringing traded goods from the point of entry into the United States to their retail out-let. We find that overall U.S. distribution wedges are 50–70%, around 10 to 20 percentage points higher thanthat reported in the literature. We discuss the implications of this for measuring the size of the “pure” trade-ables sector, exchange rate pass-through, and real exchange rate determination. We find that distributionwedges are very stable over time but there is considerable variation across items. There is some variationacross the country of origin for the imported item, for our major trading partners, but not as much as thecross-item variation. We also investigate the determinants of distribution wedges, finding that wedges donot vary systematically with exchange rates, but are related to other features of the micro data.

Published by Elsevier B.V.

1. Introduction

An established but still-growing literature in international econom-ics has focused considerable attention on modeling and measuring thedistribution sector. On the theory side, it has been shown how distribu-tion can be crucial for generating models that display realistic real ex-change rate dynamics, accounting for exchange rate pass-through, andunderstanding several other classic questions including the internation-al transmission of real andmonetary shocks. Until now, estimates of thesize of the distribution sector have been constructed with aggregatedata. In this paper, we show that for the U.S., measures derived frommicro data are even larger than previous estimates. As detailed below,our primary statistic of interest is the CPI price relative to import price,a statistic that has confusingly been referred to as both the distributioncost or distribution (or profit) margin. We prefer to call this gap the dis-tributionwedgebecause it captures everything that encompasses the gapbetween the retail price and the dock price, including both profit mar-gins and local distribution costs.1 We think this term is conceptually

Nakamura for comments andics and Randal Verbrugge forsolely the responsibility of theiews of the Board of Governorsssociated with the Federal Re-

.V.

appealing becausewhile it is clear that at least one of these componentsis necessary to explain real exchange rate dynamics, it is an open ques-tion whether the failure of the law of one price for traded goods is pri-marily driven by variation in profit margins (Engel, 1999) or by localdistribution costs (Burstein et al. (2003)). 2

The distribution sector can be crucial for understanding and gen-erating models that display realistic real exchange rate dynamics.Engel (1999) and Burstein et al. (2005) examine the classic decompo-sition of the real exchange rate into changes in the relative price ofnon-tradeables and deviations from the law of one price for trade-ables, and present evidence on the importance of distribution servicesas a component of the prices of goods traditionally classified as “trad-ed.” Devereux et al. (2003) incorporate a distribution sector in theirwork on the welfare effects of moving to a single currency in theeuro area. In a series of papers, Corsetti, Dedola, and Leduc workedextensively on modeling the distribution sector [Corsetti and Dedola(2005), Corsetti et al. (2008a, 2008b)]. They revisit several classicquestions in international macroeconomics, including exchange ratepass-through, the lack of correlation between the real exchange rateand relative (home-foreign) consumption and the internationaltransmission of real and monetary shocks. Burstein et al. (2003),

2 The issue is further confused by the fact that Engel's paper focuses on the real ex-change rate of the U.S. and other large, developed economies whereas Burstein and co-authors focus on the real exchange rate between the U.S. and emerging economies.There is reason to believe that this difference underlies some their diverging results.For instance, Burstein et al. (2005) find that pass-through into dock prices for Argenti-na is almost 100% whereas Gopinath et al. (2010) find that the corresponding figure forthe U.S. is closer to 10%.

63D. Berger et al. / Journal of International Economics 88 (2012) 62–73

Goldberg and Campa (2010), and Choudri et al. (2005) show that in-corporating a distribution sector into an otherwise standard modelimproves the ability of the model to explain observed rates of ex-change rate pass-through. In models without a distribution sector,predicted rates of pass-through are counterfactually high.3

A large literature in international trade and finance argues that var-iable markups are essential for explaining real exchange rate dynamics,including Atkeson and Burstein (2008), Goldberg and Hellerstein(2008) and Nakamura and Zerom (2010). The latter two papers exam-ine specific industries (beer and coffee) to understand the sources of in-complete pass-through to retail prices. Consistent with the theoreticalwork of Atkeson and Burstein, Goldberg and Hellerstein find that 32%4

of the imperfect pass-through is a result of variable markups at thewholesale level. They also find that retail markup variation is muchless important and does not seem to vary systematically with exchangerates. This is consistent with Gopinath et al. (forthcoming), which findsthat border price differences are driven by differences in marginal costsnot by variable markups at the retail level. Despite the importance ofvariable markups in explaining incomplete pass-through, both industrystudies cited abovefind that local distribution costs explain themajorityof incomplete pass-through.5 Our distribution wedge captures the levelof these markups, not the changes, and while we cannot accuratelymeasure either the level or the change in markups, we show in thenext section that under plausible assumptions (about the relative mag-nitudes of the average markup and local costs and about the amount ofmarkup adjustment at the wholesale level in response to an exchangeshock), our estimates of the distribution wedge imply significantlyless exchange rate pass-through into retail prices than previous studieshave found. The findings of both Nakamura–Zerom and Goldberg–Hellerstein thus reinforce the results reported in the present paper, de-spite their focus on individual (and quite different) industries.

It is widely reported that distribution costs are large. In their au-thoritative survey, Anderson and van Wincoop (2004) emphasizethe importance of distribution costs as a crucial component of overalltrade costs. They note, “Trade costs, broadly defined, include all costsincurred in getting a good to a final user other than the marginal costof producing the good itself: transportation costs, policy barriers, in-formation costs, contract enforcement costs, costs associated withthe use of different currencies, legal and regulatory costs, and localdistribution costs (wholesale and retail).” They further estimate thecontribution of distribution to overall trade costs: “The 170% headlinenumber for overall trade costs [on an ad valorem tax equivalent basis]breaks down into 55% local distribution costs and 74% internationaltrade costs [1.7=(1.55∗1.74)−1]. ” Thus, according to the evidencein Anderson and vanWincoop, distribution costs for the United Statesare large and economically important.

Burstein et al. (2003) and Goldberg and Campa (2010) estimatethe size of distribution wedges at a fairly high level of aggregation,using national input–output tables. The Burstein et al. (2003) esti-mates are on average around 40% of the retail price for the UnitedStates and 60% for Argentina. Unlike us, they attribute 100% of thewedge to distribution costs which is why they refer to it as "costs"

3 In Devereux et al. (2003) the modeling was quite simple and designed solely tohave two different prices, one for domestic consumers and one for exports. In theirset-up, retailers did not use any resources such as labor. In the Corsetti, Dedola andLeduc frameworks, retailers use labor, so domestic factor costs matter for the consumerprice of imported goods. In addition the absence of substitutability between labor inretail and in the imported good makes the retail price of the imported good a linearbundle of the imported good and local labor. As a result the foreign exporter faces anon-constant elasticity of demand, leading to several interesting findings, such as lim-ited exchange rate pass-through even with flexible prices.

4 In their example complete pass-through would be 100% pass through. Local costsat the wholesale level lower pass-through to 50%. Variable markups at the wholesalelevel lower the pass-trough percent to 18%. (or a 32% markup variation) close towhat is observed in the data.

5 For example, Nakamura and Zerom (2010) conclude that local costs reduce pass-through after six quarters by 59% relative to a CES benchmark.

rather than the distribution "wedge".6 Goldberg and Campa (2010)document the size of the distribution sector for the Unites States and20 other OECD countries. Their primary data source is also input–outputtables so they are assuming that the entire wedge is due to distributioncosts. Across countries, distributionwedges on household consumptiongoods are between 30% and 50% of purchasers prices; the estimate forthe United States is 43%. For the eight countries for which Goldbergand Campa have time series data, it is found that distribution wedgesare sensitive to exchange rate movements. Bradford and Lawrence(2003) also use input–output sources to measure distribution costs inover 100 consumer categories for the United States and eight other in-dustrialized countries. For the United States, Bradford and Lawrence re-portwedges as a fraction of producer prices of 68% on average, or 40% asa fraction of purchaser prices. There is considerable variation across cat-egories of items and across countries, with Japan and the United Stateson the high end.

In this paper, we analyze retail prices and import prices of specificitems in the Bureau of Labor Statistics' (BLS) CPI and IPP databases tomeasure the distribution wedge associated with bringing tradedgoods from the point of entry into the United States to their retail out-let. Previous work has exploited these data separately, using eitherthe CPI (Bils and Klenow (2004) and Nakamura and Steinsson(2008)) or the IPP (Gopinath and Rigobon, 2008). We use both data-bases simultaneously. A “matching procedure,” described in detail inthe Appendix A, verifies that the items being compared are identicalin description. To our knowledge, no other study of distributionwedges uses as detailed a data set as ours. This allows for a cleanercalculation of the distribution wedge than was possible before andit confers the further advantage to investigate the determinants ofthe wedge. Of particular interest, given the focus on this question inthe existing literature, is whether wedges vary systematically withexchange rates. The total wedge that we measure is 10–20% largerthan previous estimates of the distribution wedge, a finding that inand of itself implies significantly less exchange rate pass-through toretail prices than previous estimates in the literature. After docu-menting the size of distribution wedges along several cuts of thedata, we explore the determinants of these wedges, including the re-lationship with exchange rate changes. We also relate wedges to var-ious features of the micro data, such as the frequency of price changesfor an item. This is something papers using input–output data are ofcourse unable to do.

We find that overall distribution wedges are around 50–70% forU.S. data between January 1994 and July 2007. This number is about10% to 20% age points higher than that reported by other researchers.Distribution wedges are quite stable over time but vary considerablyacross items. Wedges are typically lower for sale price CPI items, asexpected, but do not differ significantly across c.i.f. versus f.o.b. im-port price basis considerations. Surprisingly, intra-company transferpricing considerations do not have much of an effect on the size ofdistribution wedges. There is some variation across the country of or-igin for the imported item, for our major trading partners, but not asmuch as the cross-item variation. We do not find that wedges varysystematically with exchange rates, nor is there is a strong relation-ship between the response of the distribution wedge to exchangerate changes and that of the import price. Wedges are, however, sig-nificantly explained by other characteristics of the micro data. Wetake this lack of correlation with the exchange rate as evidence thatthe majority of our distribution wedge is capturing distributioncosts, not profit margins. If distribution wedges were largely

6 This result follows naturally from the fact that they assume in their theoreticalmodel that the distribution sector is perfectly competitive so these firms earn zeroprofits in equilibrium. Their data work implicitly makes the same assumption becausetheir primary source of data is national input–output tables and these tables are de-rived under the assumption that all production units have constant returns to scaletechnologies. Hence, in the absence of other distortions, these production units earnzero profits in equilibrium.

64 D. Berger et al. / Journal of International Economics 88 (2012) 62–73

composed of variable wholesale markups and if pricing to market isimportant, then changes in wedges would covary strongly negativelywith nominal exchange rate changes.

2. Distribution margin or distribution cost

As mentioned in Introduction, we measure the aggregate wedgebetween the retail price and the price at the dock—a wedge that in-cludes both retail and distributor markups and local distributionand marketing costs. Unfortunately, despite our intensive work withthese rich data sets, we are unable to disentangle these two compo-nents in a nice, nonparametric way. One way to proceed is to followthe previous literature (Burstein et al., 2003) and assume that the dis-tribution wedge is equal to the distribution cost. Given that we mea-sure the distribution wedge to be between 10% and 20% larger thanprevious estimates, if we performed a similar exercise to the onedone by Burstein et al. (2005), then one would find that our measuredwedge implies significantly less pass-through into retail prices thanwhat Burstein, Eichabaum and Rebelo found. This is shown explicitlyin the next section of the paper.

We think, however, that the assumption that the distributionwedge is equal to the distribution cost is unappealing because it con-tradicts a long empirical IO literature arguing that many firms are im-perfectly competitive. Furthermore, if there is no markup, there is norole for markup adjustment, which would contradict recent empiricalwork by Goldberg and Hellerstein (2008) and Nakamura and Zerom(2010), both of which find that markup adjustment at the wholesalelevel is important for explaining incomplete pass-through. Anotherapproach is to make a rough approximation of the relative magni-tudes of the markup components and the distribution costs compo-nents so that we can consider both margins in the exercise weperform in the next section. To fix ideas, consider a simple decompo-sition of the retail price for a single good:

PRt ¼ PNT

t þ μt þ PDt

where PR, PNT, PD and μ are the retail price, the distribution cost, theprice at the dock and the markup. Concretely, one can imagine thecase where the foreign manufacturer owns the wholesaler in theU.S. as is the case in the Beer industry. (Goldberg and Hellerstein,2008) PD is the foreign manufacturer's unit marginal cost, μ is themarkup the wholesaler charges the retail firm to purchase the item,and PNT is the total distribution costs required to bring the productto market. 7 The distribution wedge is defined as

dt ¼ 100� PRt −PD

t

PRt

!≈60%

Consistent with the upper estimates from the empirical IO litera-ture, assume that unit margins are equal to 25% μ

PR¼ :25

� �:8 This im-

plies that the fraction of the retail price spent on local distribution andmarketing costs is 35%. Consistent with the empirical literaturehighlighting the importance of pricing to market we also assumethat the markup varies negatively with the exchange rate. Specifical-ly, in response to a 1% unexpected depreciation of the dollar, we as-sume that the wholesale markup falls by 0.317%.9 Interestingly, nomatter which set of assumptions one makes, the implied pass-through into retail prices is much less than previous studies found.

7 Alternatively, one could consider the case where a wholesaler purchases the itemfrom an overseas manufacturer at price PD, it requires PNT total distribution costs tobring the item to retail and μ is the total markup of the wholesaler and the retailer.

8 Note that we are measuring the markup relative to the retail price, where tradi-tionally the markup is measured relative to marginal cost. Thus our assumption thatthe markup relative to the retail price is a (significant) lower bound on the convention-ally measured markup.

9 See Goldberg and Hellerstein (2008) Table 13 column 2.

3. Measuring distribution wedges

We measure distribution wedges in two ways. First, using the de-tailed information on product characteristics in the CPI and IPP data-bases, we match items at the dock to those sold at retail that areidentical in description. Our matching procedure is done on a catego-ry by category basis depending on available information, as describedin detail in the Appendix A. Under this procedure we are highly con-fident that we are comparing at-the-dock prices and retail prices ofitems that are identical in description. Unfortunately, this procedurealso necessitates that we discard a lot of data, either because exactmatches did not exist or because there was insufficient evidence todetermine the quality of a match.

In light of this last consideration, we check robustness using a sec-ondmeasure of distributionwedges. Under this procedurewe constructweighted-average price levels for fairly disaggregated item categories inthe CPI and import price data bases. The level of aggregation is by entrylevel item (ELI) in the CPI, or approximately 10-digit SIC code for im-ports.We use prices of only those CPI items thatwe could reliably deter-mine to have been imported rather thanmade in the United States. Thisalternative procedure allows us to measure the distribution wedge foritem categories such as (imported) “beer”, “televisions”, and “bananas”.Under this procedure we utilize the prices of manymore of the items inthe sample but use less of the item-specific information that is con-tained in the database.

Under both strategies, the distribution wedge for item category i,di, is calculated as

di ¼ CPIi−IPPið Þ=CPi

where CPIi is the retail price of the item (or its weighted average pricelevel) and IPPi is the import price.10 We use monthly data from Janu-ary 1994 to July 2007.

The calculation of di could be affected by several important “pricebasis” considerations.11 The first is whether the CPI item's price is asale price or a regular price. Second, is whether the imported item ispriced on a c.i.f. or f.o.b. basis. Finally, we must distinguish betweenimports that are intra-company transfers and those that are arm'slength transactions that more accurately reflect market prices. Eachof these could have non-trivial effects on the distribution wedge. Inlight of this, we report results in a few different ways reflecting com-binations of these price bases considerations.

In Table 1 we report the median distribution wedge for all itemsunder the first of our measurement procedures. Results using the“matching procedure” described in the Appendix A are contained inpart A of the table, while those of the alternative procedure usingweighted average price levels are in part B. For the former we reportwedges in four ways: when the CPI price is regular and the importprice basis is cif, CPI price is regular and import price is fob, and theanalogies for cases in which the CPI price is a sale price. In theupper panels intra-company transfer prices are excluded. In thelower panels we report the same calculations using only the intra-company transfer prices.

10 Various authors report wedges in different ways. With the formula above thewedge is bounded by zero (when CPI price is greater than IPP) and unity. It has the in-tuitive interpretation as the fraction of the retail price, which consumers do observe,that is accounted for by transportation costs, overhead, retailer profit, etc.11 This is relevant for the matching procedure but not the alternative procedurewhere we calculate weighted-average price levels for ELI categories. Under the latterwe do not utilize such information as we are trying to use prices of as many items aspossible, irrespective of whether the database contains specific information on theitem.

Table 1Distribution wedges, all items.

A. Matching procedure

Regular Sale

Intra-company transfer prices excludedcif 0.57 0.50fob 0.68 0.60

Intra-company transfer prices onlyregular sale

cif 0.58 0.57fob 0.62 0.49*Cells report the median of the distribution wedge, =(P{CPI}−P{IPP}) /P{CPI},across all items in the sample. Regular (sale) denotes that the CPI price of theitem was a regular (sale) price. Cif (fob) denote the price basis of the importprice in the IPP database. The calculations in the upper panel exclude all importswhose prices are reported as being intra-company transfer prices. The calculations in the lower panel include only imports whose prices are reported as beingintra-company transfer prices.

B. Alternative procedurePrice levels

Mean Weighted-average

μ 0.70 0.64*The distribution wedge is calculated using weighted-average price levels for several disaggregated item categories in the CPI and import price data bases. Thelevel of aggregation is by entry level item (ELI) in the CPI, or approximately 10-digit SIC code for imports. The item categories are listed in Table 2. The cellsabove report the simple mean and the expenditure share-weighted average distribution wedge across those item categories.

65D. Berger et al. / Journal of International Economics 88 (2012) 62–73

3.1. Distribution wedges: all items

According to the upper panel of Table 1A, when transfer prices areexcluded from the sample the median distribution wedge across allregular-priced CPI items is 0.57 (0.68) for imports priced on a cif(fob) basis. For sale-price CPI items the respective distributionwedges are 0.50 (0.60). The analogous numbers for transfer pricesare, contrary to our prior expectations, generally quite similar: 0.58(0.62) and 0.57 (0.49), as seen in the lower panel of Table 1A.

The distribution wedges reported in Table 1 are distinctly higherthan the estimates reported for U.S. consumption goods by other re-searchers. Burstein et al. (2003) estimate U.S. distribution wedges 12

to be 42% in 1992 and 43% in 1997, using the national input–outputtables. The wedge is about the same when the authors use datafrom the 1992 U.S. Census of Wholesale and Retail Trade. Goldbergand Campa's (2010) cross-country evidence confirms the 43% esti-mate of the distribution wedges for all U.S. final household consump-tion in 1997 (also using national input–output data), estimating thatmost of this is due to distribution wedges in the wholesale-retail sec-tor rather than transportation. Bradford and Lawrence (2003) reportan overall distribution wedge for the United States in 1992 of 40%as a percentage of purchaser price.

13 We were able to construct reliable estimates of weighted-average price levels for

3.1.1. Alternative procedureTable 1B reports distribution wedges computed under the alterna-

tive procedure where we construct weighted-average price levels forfairly disaggregated item categories. These wedges are slightly higherthan those obtained from the matching procedure: 0.70 or 0.64depending on how we weight item categories. This indicates a gener-al robustness to using prices of considerably more items than waspossible under the matching procedure.

12 Remember, they make assumptions so that the distribution margin is equal to thedistribution cost.

3.1.2. Stability over timeThe wedges are quite stable over time. Lumping all items together

without distinguishing between cif and fob, sale price or not, etc., ourmatching procedure gives us wedges of 0.62, 0.67, 0.63, 0.57, 0.59,0.60, 0.58, 0.61, 0.59, 0.57, 0.58, 0.60, 0.60 and 0.61 in the years1994 through 2007 respectively. As noted above, a relatively stableoverall distribution wedge is also found by Burstein et al. (2003).This stability of wedges against the backdrop of considerable fluctua-tions in the dollar foreshadows our finding below concerning the lackof a systematic relationship between distribution wedges and ex-change rates.

3.2. Results by item

In our data sample, there is considerable variation across items,with wedges ranging from around 20% to 80%. These results arepresented in Table 2A for the 21 item categories from which wewere able to uncover a sufficient number of high-grade matches(see the Appendix A tables for the number of observations ineach category). The lowest wedges are for televisions, video cam-eras, VCRs, cameras, telephones and microwave ovens. The highestwedges are found for drugs, our two apparel categories (men's andwomen's pants), watches, film, and our two fresh foods categories(bananas and tomatoes). As expected, wedges are typically lowerfor sale price items, and in some cases the difference is nearly 20percentage points. Wedges do not differ systematically betweenthe cif and fob price basis, though on average fob wedges are higheras expected.

Table 2B reports results by item category when we compute distri-bution wedges using the alternative procedure.13 Consistent with theresults under the matching procedure, the largest wedges are ob-served for watches, olive oil, and bananas, with wedges for televisionsets (and alcoholic beverages here) being at the low end. Below werelate the cross-section of distribution wedges to features of themicro data underlying our sample.

3.3. Composition effects and results by brand

The results so far could be masking important composition ef-fects, in principle across brand, time, and country of origin. Theitem categories above, while certainly disaggregated, still containproduct heterogeneity. There are, for example, the wedges associat-ed with small-screen television sets (13-inch diameter) and thoseassociated with large, high-end televisions. These wedges are aver-aged together in the results above. If there are important composi-tion effects, it may be misleading to compare our estimates to thoseof the existing literature, or to compare results across various slicesof our own data set.

In fact, however, composition effects are likely to be unimportant.We compute distribution wedges by brand for cases in which we haveat least ten observations.14 For Alcoholic Beverages, the standard devia-tion across the 25 brands is 0.08 (for the case in which transfer pricesare excluded), compared to a mean wedge of around 0.50–0.55(Table 2). For beer (18 brands) and television sets (six brands), thecross-brand standard deviations are 0.13 and .07, respectively.

only about half of the item categories used in the matching procedure. This was dueto data limitations. Particularly constraining was getting information on whether a par-ticular CPI item was produced in the United States or abroad.14 See the working paper version at www.federalreserve.gov/pubs/ifdp/2009/972/ifdp972.pdf

Table 2Distribution wedges by item categories.

A. Matching procedure

Regular price (CPI) Sale price (CPI)

cif fob cif fob

Intra-company transfer prices excludedAlcoholic beverages 0.55 0.58 0.51 0.44Audio players 0.58 0.55 0.52 0.47Bananas – 0.72 – 0.59Beer 0.53 0.66 0.42 0.62Calculators – 0.72 – 0.70Cameras – 0.47 – 0.40Computer accessories 0.29 0.36 – 0.31Drugs 0.67 0.84 – –

Film 0.86 0.74 0.82 –

Men's pants – 0.75 – 0.70Microwave ovens – 0.46 – 0.36Kitchen equip. (misc.) – 0.66 – 0.62Olive oil – 0.72 – 0.62Telephones 0.35 0.42 – 0.27Stoves 0.56 0.78 0.55 0.61Tomatoes 0.83 0.78 0.76 0.70Televisions 0.28 0.21 0.24 0.35VCRs 0.44 0.40 0.41 0.34Video cameras 0.32 0.29 – 0.23Watches – 0.78 – 0.79Women's pants – 0.64 – 0.66

Intra-company transfer prices onlyAlcoholic beverages 0.65 0.58 0.57 –

Audio players 0.52 0.48 0.51 0.42Bananas – 0.73 – 0.61Beer – 0.65 – 0.57Calculators – 0.54 – –

Cameras – 0.51 – 0.39Computer accessories – 0.43 – 0.47Drugs – 0.85 – –

Film – 0.71 – –

Men's pants 0.61 0.58 – 0.49Microwave ovens – 0.55 – 0.34Kitchen equip. (misc.) – 0.60 – –

Olive oil – 0.81 – 0.82Telephones – 0.39 – 0.35Stoves 0.57 0.53 0.56 –

Tomatoes 0.31 0.84 – –

Televisions 0.47 0.35 0.53 0.36VCRs – 0.36 – 0.32Video cameras – 0.33 – 0.29Watches – 0.86 – –

Women's pants 0.70 0.82 0.66 –

B. Alternative procedure

Category Wedge

Alcoholic beverages 0.41Bananas 0.72Beer 0.69Computer accessories 0.69Refrigerator 0.58Men's pants 0.62Olive oil 0.74Televisions 0.50Watches 0.90Women's pants 0.59

Table 3Distribution wedges by country of origin.

cif fob

Euro Area 0.48 0.60Canada 0.55 0.59China 0.54 0.50Japan n.a. 0.36Mexico 0.75 0.55United Kingdom 0.56 0.59

*Cells contain the median distribution wedge across all items imported from the listedcountry, under the matching procedure. Sale price and regular price CPI items are in-cluded together, and intra-company transfer prices are included along with “arm'slength transaction” prices. (n.a.) Insufficient number of item categories.

15 Goldberg and Campa (2010) use national data, at a fairly high level of aggregation,over the period 1995–2001. Their estimates indicate that a 1% real depreciation resultsin a 0.47% decline in the distribution wedge. Burstein et al. (2005) also use fairly aggre-gated data. Their results rely on there being significant pass-through to import pricesand little or no pass-through to retail prices.

66 D. Berger et al. / Journal of International Economics 88 (2012) 62–73

3.4. Distribution wedges by country-of-origin

We also calculate distribution wedges based on a different cut ofthe data. Here we lump together all items that were imported froma particular trading partner and calculate distribution wedges basedon the matching procedure. Table 3 presents the results for ourmajor trading partners. wedges range from a low of 0.36 for Japan(for imports priced on an fob basis) to 0.75 for Mexico (cif basis).

However, most of the wedges fall within the range of 50% to 60%reported in the all-items tables above.

4. Do Distribution Wedges Move Systematically with ExchangeRates?

Few issues in international macro have been as pervasive as thetransmission of exchange rate movements into domestic consumerprices (e.g., Bernanke (2007), Engel (1999), Burstein et al. (2005)and Goldberg and Campa (2010)). Because a potentially crucialdampening channel is the distribution sector, a lot of effort has goneinto estimating the impulse from exchange rates into border pricesand distribution wedges. Burstein et al. (2005) find a significant rela-tionship between exchange rates and wedges, especially for emergingcountries like Argentina. Goldberg and Campa (2010) report thathome currency depreciations are associated with statistically signifi-cantly lower distribution wedges in a panel regression containingthe United States and 9 European countries. 15 In our data, however,recall that there is prima facie evidence against finding a relationshipbetween distribution wedges and exchange rates: our average annualdistribution wedge across all items fluctuates between 0.57 and 0.67during the period 1994–2007, with all of the estimates after 1996lying between 0.57 and 0.61. In these years the dollar moved by aconsiderable amount: against the currencies of our major tradingpartners, the dollar first appreciated by more than 20% and subse-quently depreciated by more than 30%.

Composition effects associated with such aggregated results could,of course, be masking significant relationships between wedges andexchange rates at a more detailed level. In addition, exchange ratescould be exerting only a small effect on distribution wedges becauseneither the import price nor consumer price responds to exchangerate changes, or alternatively, when the import price changes in re-sponse to the exchange rate, this is also passed through to the con-sumer price. So we ask: if there is pass-through to import prices,even if atypically, is there still low pass-through to consumer pricesand hence significant pass-through to wedges? Or is that importprice response passed through to consumer prices? Our data set onborder prices and matched retail price for the same item allows us,uniquely, to investigate this issue directly.

In Tables 4A, 4B and 5 we present the results of standard pass-through regressions of the form:

Δy tð Þ ¼ cþ b0Δs tð Þ þ b1Δs t−1ð Þ þ…þ bnΔs t−nð Þ þ a1ΔCPI t−1ð Þþ a2ΔCPI t−2ð Þ þ d1ΔCPI � t−1ð Þ þ d2ΔCPI � t−2ð Þ;n

¼ 2;12

Table 4AExchange rate pass-through to Wedges, IPP, and CPI; Results by Item.

Category (no. observations) Wedge IPP CPI

b0 F-stat R2 b0 F-stat R2 b0 F-stat R2

All observationsVideo cameras (316) 1.45 1.07 .02 −0.06 5.24** .02 0.46 0.35 .03Telephones (398) −0.18 1.42 .01 0.06 5.20** .01 −0.81* 0.65 .00Watches (165) −0.15 0.25 .00 0.44† 3.27* .00 −0.27 0.00 .00Computer accessories (137) −1.10 0.36 .03 0.52 4.70** .10 −0.21 2.21† .04Alcoholic beverages (3837) −0.15 0.01 .00 −0.10 4.08** .00 −0.20* 7.60** .00Televisions (1386) −0.67 0.80 .00 −0.13 0.15 .01 −0.32* 2.57† .00Women's pants (207) 0.63 0.56 .00 0.77† 2.22† .00 −0.25 0.17 .00Olive oil (217) 0.48 8.78** .05 0.57 2.98* .01 −0.18 1.51 .00Beer (4048) −0.11 0.15 .00 0.03 4.85** .00 0.03 3.17* .00Bananas (7603) −0.15 7.11** .00 0.96** 283** .04 0.47* 11.6** .00Audio players (484) 0.15 0.14 .00 0.28 2.37† .00 −0.26 0.03 .00Cameras (337) 0.18 0.13 .00 −0.37 2.13† .02 −0.30 0.67 .00Drugs (113) 0.04 2.88* .02 0.61 0.47 .04 0.02 1.79 .00Film (515) −0.98** 7.33** .01 0.45 4.00** .00 0.23 0.16 .00Men's pants (477) 0.34 0.51 .01 −0.37† 1.25 .00 0.20 4.32** .00Kitchen equip. (misc.) (243) −0.87 0.52 .01 −0.05 0.04 .00 −2.35* 2.88* .00Microwave ovens (137) −0.65 2.04† .00 0.79† 1.53 .00 −2.89* 1.17 .08Stoves (392) −1.16* 1.71 .00 0.00 0.06 .00 −2.13* 7.13** .02Tomatoes (1805) 0.57* 5.96* .01 −0.89* 11.8** .02 0.58 1.58 .01VCRs (713) −0.14 0.00 .00 0.01 1.00 .01 −0.11 1.40 .00Δy(t)=c+b0Δs(t)+b1Δs(t−1)+b2Δs(t−2)+a1ΔCPI(t−1)+a2ΔCPI(t−2)+d1ΔCPI*(t−1)+d2ΔCPI*(t−2). s is the trade-weighted nominal exchange rate; CPI (CPI*)is the U.S. (foreign aggregate) consumer price index. F-stat: b1, b2 jointly zero. **,*,† significant at 1%, 5%, 10%.

Conditional on a contemporaneous price change in that itemVideo cameras (213) 0.38 0.48 .00 −0.03 0.38 .00 0.60 1.50 .00Telephones (223) −3.09† 6.60** .00 0.63 1.08 .02 0.70 0.68 .00Watches (86) 0.06 0.52 .00 0.12 0.45 .00 0.64 0.01 .00Computer accessories (123) −1.30 1.13 .00 1.15† 8.32** .08 0.48 4.78* .01Alcoholic beverages (2438) −0.18 0.07 .00 0.10 1.13 .00 −0.22* 7.25** .00Televisions (736) −0.41 0.04 .00 0.11 0.10 .00 −0.29 1.79 .00Women's pants (139) 2.35 0.26 .00 0.31 0.37 .00 0.16 1.26 .00Olive oil (218) −0.32 0.30 .02 −0.67 10.8** .04 −0.21 0.11 .02Beer (2524) 0.24* 5.37** .01 0.01 2.28† .00 0.11 0.38 .00Bananas (6585) -0.31* 10.8** .00 0.89** 235* .04 0.16 6.85** .00Audio players (239) −0.63 2.57† .04 −0.04 0.08 .01 −0.58 2.18 .01Cameras (163) 0.03 0.87 .00 −0.66† 2.47† .02 −0.60 3.58* .02Drugs (71) 0.20 2.57† .00 0.004 0.33 .08 0.72 2.07 .00Film (331) −0.69 4.47** .02 0.05 4.97* .05 −0.04 0.40 .01Men's pants (251) 0.76 0.52 .00 −0.28 0.003 .00 0.91 9.91** .03Kitchen equip. (misc.) (133) −1.86 1.40 .00 0.08 0.82 .00 −0.28 0.07 .00Microwave ovens (90) −3.24* 8.81** .07 0.29 0.78 .00 −0.53 1.69 .07Stoves (116) −0.65 0.20 .06 −0.11 0.005 .00 −1.31 6.40** .06Tomatoes (1189) 0.90** 0.17 .01 −0.68 2.69* .00 0.58 2.40† .01VCRs (418) 0.79 7.82** .01 −0.17 0.02 .00 0.28 2.56† .01The regression and tests statistics are the same as for the table above. Now the regression is run only for periods in which the IPP price changes (for the Wedges and IPPregressions) or in which the CPI price changes (for CPI regression in the final columns).

Table 4BExchange rate pass-through to Wedges, IPP, and CPI; results by country of origin.

Category (no.observations)

Wedge IPP CPI

b0 F-stat R2 b0 F-stat R2 b0 F-stat R2

All observationsCanada (1185) −0.44 0.60 .00 −0.04 1.37 .00 −0.08 0.51 .01Mexico (5162) 0.05 1.35 .00 −0.12 2.09 .00 0.06 1.25 .00U.K. (1027) 0.06 0.31 .00 0.45* 6.63** .01 0.17 3.02* .00China (1377) 3.22 0.40 .00 0.41 0.03 .01 6.49* 11.9** .01Japan (901) 0.11 0.02 .00 −0.02 0.28 .00 0.09 0.16 .00Euro Area (2344) 0.05 4.19* .00 0.07 0.33 .01 0.11 0.41 .00

Conditional on a contemporaneous price changeCanada (847) −0.86* 3.20* .01 0.08 0.00 .00 −0.19† 2.52† .00Mexico (2820) 0.05 1.26 .00 −0.18† 2.71† .00 0.05 1.88 .00U.K. (710) 0.06 0.16 .00 0.34 1.02 .00 0.43** 13.6** .01China (703) 1.52 0.46 .00 0.92 0.67 .00 3.16* 1.91 .00Japan (593) 0.04 2.05 .00 −0.07 2.45† .00 −0.05 0.70 .01Euro Area (1504) 0.04 4.98* .00 0.00 7.93** .02 0.10 0.03 .00

Δy(t)=c+b0Δs(t)+b1Δs(t−1)+b2Δs(t−2)+a1ΔCPI(t−1)+a2ΔCPI(t−2)+d1ΔCPI*(t−1)+d2ΔCPI*(t−2). s is the bilateral nominal exchange rate; CPI (CPI*) is the U.S.(foreign world aggregate) consumer price index. F-stat: b1, b2 jointly zero. **,*,† significant at 1%, 5%, 10%.

67D. Berger et al. / Journal of International Economics 88 (2012) 62–73

19 These average pass-through estimates thus closely resemble those from the exist-ing literature that has used aggregated price data. Hummels et al. (2010) provide a the-oretical explanation for why the large cross-item dispersion of pass-through estimatesthat we find here would arise. Depending on the shock and market structure, pass-through estimates outside the 0 to 1 range are easily rationalized.20 The absence of a significant relationship between distribution wedges and ex-change rate changes would seem to contradict Goldberg and Campa (2010). However,

68 D. Berger et al. / Journal of International Economics 88 (2012) 62–73

The dependent variable is, alternately, the change in the (1) distri-bution wedge, (2) import price and (3) consumer price. These areregressed on the contemporaneous change in the exchange rate,two or twelve lagged changes in the exchange rate, and two laggedchanges in the foreign CPI.16 The data are monthly from January1994 to July 2007. In these regressions we use the most comprehen-sive sample of prices, grouping together, e.g., regular and sale prices,market and transfer prices, etc., but have examined robustness forseveral different cuts of our data along these lines. We run the regres-sions by item (Tables 4A and 5A) and by country of origin of the im-port item (Tables 4B and 5B). We report in the lower half of each tableresults conditioning on there being a contemporaneous change in theitem price, i.e., running the pass-through regression only for thosemonths in which an actual price change occurred. 17

Consider the first column of Table 4A, the pass-through regres-sions for distribution wedges by item, with two lagged exchangerate terms. The coefficient estimates on the contemporaneous ex-change rate changes, b0, have a weighted-average value of −0.21(std. error 0.60). The coefficients are insignificantly different fromzero for all but 3 of the 21 items: Film, Stoves, and Tomatoes; the F-statistic from a test of the null hypothesis that the exchange ratechanges are jointly zero also rejects in two of these cases, again sug-gesting some pass-through. Note, however, that the regression R2

?

values are miniscule, and that only for Tomatoes is the coefficienton the contemporaneous exchange rate change positive as expectedfrom theory and earlier studies. Thus, there is only scant evidence ofsignificant pass-through to distribution wedges.

Thenext two columns of Table 4Apresent results for the correspond-ing import and consumer price changes. The short-run pass-throughelasticity for IPP prices appears to be significant for Bananas, Men'sPants, and Tomatoes. For the latter two categories, there is no significantpass-through to the consumer price. For Stoves, significant negativepass-through is for the consumer price is matched with zero pass-through to the import price, resulting in significant negative pass-through to the wedge. Only in the case of Tomatoes is pass-through tothe import price large enough to give rise to significant positive pass-through to the distribution wedge, as in Burstein et al. (2005).18

Although necessarily tedious to digest, these results exemplify an im-portant feature of our data: there is no consistently significant relation-ship between exchange rates and distribution wedges, import prices, orconsumer prices. This conclusion is unaffected by how we slice thedata, e.g., by excluding sale prices and/or intra-company transfer prices;by conditioning on there being a contemporaneous change in the itemprice (see the lower half of Table 4A); or by running the regressions ona country-by-country basis lumping all items together (Table 4B). Ourpass-through regressions produce short-run elasticity estimates thatare generally zero, and in all cases the regression R2

? values are tiny.In the Table 5A and B regressions with 12 lags of the exchange rate

we examine longer-run considerations. The weighted-average coeffi-cient on b0 in the upper half of Table 5A is −0.23 (std. error 0.65) forwedges, −0.27 (0.52) for IPP, and 0.16 (0.30) for CPI. The general lackof significance of b0 in Tables 4A, 4B holds on an item-by-item basishere too, as seen in the individual columns. Turning to the sum of theb(n) coefficients, the “long-run elasticities” in the parlance of Campa–Goldberg, we observe more evidence of significant pass-through, espe-cially for import prices and when the regression is run by country of

16 This is the prototype regression found in the literature on exchange rate pass-through. Goldberg and Campa (2010) provide details. We follow them in reportingthe estimated b0, akin to their “short-run pass-through elasticity”, and the sum of b0through bn, the equivalent of their “long-run pass-through elasticity.”17 In Tables 4A and 5A the exchange rate is the trade-weighted value of the dollar,while in Tables 4B and 5B the exchange rate is the bilateral rate of the dollar againstthe currency of the exporting country.18 Larger (smaller) estimated pass-through coefficients for the IPP regressions do cor-respond with larger (smaller) estimates for the corresponding CPI item, but the corre-lation is small: 0.23.

origin rather than by item (Table 5B). For wedges, estimates in theupper half of the Table 5A range from 7.52 for olive oil to −7.26 forComputer Accessories, and produce a cross-item average of 0.40.When the regression is run only for periods in which there is a pricechange, estimates range from 9.26 to −5.55 (mean=0.08), as seen inthe lower half of Table 5A. Large dispersion across items is also foundfor the IPP and CPI item regressions, as seen in the columns further tothe right, producing cross-item averages of −0.32 and −0.02 for IPPand CPI, respectively, in the upper half of Table 5A, and 0.40 and−0.06 in the lower half.19 The strongest evidence of significant long-run pass-through is for import prices when we pool items by countryof origin. As seen in Table 5B, pass-through to IPP is negative and signif-icant for items coming from Canada, Mexico and the Euro area.

In summary, our matched data set of border prices and retail pricesreveals that: (1) there is no strong correlation between (changes in) ex-change rates and distribution wedges in the aggregate; (2) pass-through regressions reveal a significant relationship for relatively fewitem categories; (3) although this could be because there is significantpass-through to border prices and offsetting pass-through to thematched retail item price, that is rarely found; (4) instead, pass-through to border prices is insignificant for most items, consistentwith Gopinath and Rigobon's (2008) “sticky borders” finding, and sois pass-through to retail prices; and (5) we find only one case of aBurstein et al. (2005)-style result where pass-through to the borderprice is significantly negative, pass-through to the retail price is not,and so pass-through to the wedge is significantly positive.20

We think of our investigation as shedding light on issues initiallyraised by Engel (1999). Our work is the logical progression of the em-pirical evidence provided by Burstein and co-authors, Goldberg andCampa. Engel (1999) examined whether distribution costs could ex-plain his basic finding on the predominance of failures of the law ofone price for tradeables in accounting for real exchange rate variabil-ity. He showed that if the distribution cost story is correct the wedgeshould be highly correlated with the real exchange rate.21 Subsequentstudies, including ours, have used more direct and more micro-basedevidence on distribution wedges and their relationship with ex-change rate changes. Our finding that distribution wedges are largesuggests low pass-through to retail prices. The (direct) estimates oflow pass-through to distribution wedges and their components inthis section, furthermore, provides confirming evidence for Engel's(1999) hypothesis on the sources of real exchange rate variability,at least for the United States.

5. Endogenous exits, law of one price deviations and sticky prices

The proximate determinants of the size of distribution wedges re-main to be uncovered. In this section, we relate distribution wedgesto various features of the BLS micro data.22 Two of these features

the data sets used in the two papers are quite different, with ours being based on microdata for the United States alone and the Goldberg-Campa data set being considerablymore aggregated (using item categories) but for several countries. No matter howwe sliced our data set, there was no consistently significant relationship between dis-tribution wedges and exchange rates, IPP prices and exchange rates, or CPI prices andexchange rates on an item-by-item basis. Thus differences in the two papers are due tothe level of aggregation of item prices, as evidenced by the greater significance of pass-through on a country-of-origin basis, as in Table 5B.21 See in particular Sections 5 and 6.22 The evidence presented in this section is intended to be suggestive and serve tomotivate future work. A deeper investigation of the determinants of the size of distri-bution wedges is beyond the scope of this paper.

Table 5Exchange rate pass-through: longer-run considerations.

A

Category (no. observations) Wedge IPP CPI

b0 Σbi R2 b0 Σbi R2 b0 Σbi R2

All observationsVideo cameras (320) 1.29 4.58 .00 0.20 0.25 .03 0.54 1.06 .01Telephones (380) −1.89 −2.47 .02 −0.68 0.70 .01 −0.56 1.82† .04Watches (163) −0.21 −0.07 .00 0.54† 1.02 .00 0.66 −0.58 .00Computer accessories (136) 0.74 −7.26 .00 0.97 6.88** .14 0.87 2.06 .00Alcoholic beverages (3832) −0.30 0.33 .00 −0.05 −1.17** .00 −0.12 −0.41† .00Televisions (1372) 0.70 2.08 .01 −0.03 0.33 .01 −0.34* 0.37 .01Women's pants (204) 0.19 4.79 .00 0.33 1.12 .00 −0.52 0.11 .01Olive oil (213) −0.09 7.52** .06 0.23 −15.3** .06 −1.54 −7.42* .09Beer (4065) −0.05 −0.30 .00 0.13* 0.68** .01 0.01 0.57** .00Bananas (7598) −0.38* 0.06 .01 0.90** 1.20** .05 0.35† 1.09† .00Audio players (501) −0.33 −0.85 .00 −0.02 0.88† .00 −0.28 0.16 .00Cameras (349) −0.48 0.56 .00 0.09 −0.03 .05 −0.66* 0.71 .01Drugs (112) −0.07 0.40 .10 0.56 −3.07** .00 0.53 −1.48 .07Film (519) −0.46 −0.39 .05 0.50 4.56** .13 −0.39 −1.58 .00Men's pants (493) 0.57 −1.72 .01 −0.48* 0.86 .00 0.03 −0.39 .01Kitchen equip. (misc.) (252) −1.21* −0.57 .01 0.02 0.17 .00 −2.23* −1.38 .00Microwave ovens (128) −1.00 3.11 .01 1.00* 0.83 .00 −0.10 −0.24 .02Stoves (376) −1.59** −1.08 .04 0.15* −0.28 .01 −2.06** −1.00 .06Tomatoes (1800) 0.28 0.93 .01 −0.70† −0.67 .02 −0.10 0.30 .03VCRs (712) 0.15 0.01 .00 −0.30† −0.12 .01 0.35 0.65 .00Regression: Δy(t)=c+b0Δs(t)+b1Δs(t−1)+…+b12Δs(t−12)+a1ΔCPI(t−1)+a2ΔCPI(t−2)+d1ΔCPI*(t−1)+d2ΔCPI*(t−2). We report estimated b0, the sum of b0through b12, and regression R2. **,*,† significant at 1%, 5%, 10%.

Conditional on a contemporaneous price change in that itemVideo cameras (214) 0.41 −0.13 .04 −0.08 −0.09 .11 0.45 1.59 .05Telephones (213) −1.93 −1.17 .00 1.08 5.38* .06 −0.71 −1.06 .00Watches (88) 0.29 −0.25 .00 0.12 0.60 .04 0.79 1.68 .02Computer accessories (123) −2.41 −5.55 .00 0.74 4.52* .00 0.15 −0.64 .12Alcoholic beverages (2462) −0.30 0.11 .00 −0.17 −1.38** .00 −0.29* −0.63* .01Televisions (719) 1.17 5.87** .03 −0.22 0.03 .00 −0.33 1.78** .03Women's pants (141) 0.21 9.26† .00 0.58 0.45 .00 −3.22* 0.24 .03Olive oil (220) −0.43 1.24 .00 0.14 −11.2** .08 −0.35 −0.05 .03Beer (2531) 0.07 −0.17 .00 0.09 0.59** .01 0.11 0.27 .01Bananas (6587) −0.37* −0.22 .01 1.09** 1.11** .06 0.19 1.37* .00Audio players (239) −1.11* −1.36 .00 −0.18 1.61* .04 −0.35 −1.10 .05Cameras (166) −0.42 −1.57 .00 0.69† 2.09* .21 −0.64 0.51 .05Drugs (77) 0.14 1.54** .36 0.10 −3.51* .06 0.71 −3.05 .10Film (307) −1.31* −0.87† .02 0.82 3.71† .07 −0.84† −1.41 .01Men's pants (249) −0.61 −2.53 .00 −0.02 1.80* .01 0.37 2.39 .00Kitchen equip. (misc.) (126) −0.83 1.11 .00 0.16† 0.68** .20 0.74 2.02 .00Microwave ovens (83) −3.09* −1.37 .00 0.33 1.85 .00 0.50 2.87 .02Stoves (112) −0.12 0.99 .09 0.06 −0.82† .02 −2.02† 1.19 .00Tomatoes (1198) 0.73† 0.47 .01 −0.70 −0.22 .05 −0.45 −1.44 .01VCRs (419) 1.12 2.29 .01 −0.20 0.60 .00 1.11* 2.09* .01The regression and test statistics are the same as for the table above. Now the regression is run only for periods inwhich the IPP price changes (for theWedges and IPP regressions) orin which the CPI price changes (for CPI regression in the final columns).

BAll observationsCanada (1029) −0.59 −0.34 .00 0.01 −1.04** .00 0.01 0.001 .00Mexico (5008) 0.16 −0.53 .00 −0.16* −0.92** .00 0.04 −1.08** .00U.K. (932) −0.34* 0.92† .00 0.12 2.95** .00 0.19 2.90** .03China (1342) 2.27 1.28 .00 0.13 1.17 .00 4.70* 4.51 .00Japan (861) 0.50* 0.50 .00 0.01 −0.18 .02 0.14 0.38 .00Euro Area (2122) −0.06 0.57** .01 0.04 −0.39** .02 −0.07 to −0.30 .01

Conditional on a contemporaneous price changeCanada (716) −0.63† −0.56 .00 0.04 −1.16** .00 −0.30* 0.16 .00Mexico (2749) 0.15 −0.41 .00 −0.20† −2.40** .01 0.05 0.24 .00U.K. (648) −0.10 1.37† .00 −0.05 1.49 .00 0.34* 2.56** .04China (678) −0.35 −0.001 .00 0.57 3.34 .00 3.59† 4.09 .01Japan (557) 0.08 0.67 .01 0.06 −0.15 .00 0.29* 0.52 .03Euro Area (1334) 0.10 0.73* .02 −0.21* −0.76** .04 −0.05 −0.34 .01Regression: Δy(t)=c+b0Δs(t)+b1Δs(t−1)+…+b12Δs(t−12)+a1ΔCPI(t−1)+a2ΔCPI(t−2)+d1ΔCPI*(t−1)+d2ΔCPI*(t−2). We report estimated b0, the sum of b0through b12, and regression R2. **,*,† significant at 1%, 5%, 10%.

69D. Berger et al. / Journal of International Economics 88 (2012) 62–73

are measures of law of one price deviations and price stickiness. Theyare well known, and we simply follow the existing literature in calcu-lating them from the BLS micro data. The third feature is our own con-struct, whose explanation we turn to next.

5.1. Endogenous exits

One striking feature of the micro data in the IPP database is thatparticular items imported from particular countries are relatively

Distribution Margins vs IPP StickinessCorrelation = -.31

Tomatoes

BananasOliveoil

0

0.1

0.2

0.3

0.4

0.5

0.6

0.7

0.8

0.9

Probability of a Price Change

Dis

trib

uti

on

Mar

gin

Distribution Margins vs Absolute LOP DeviationsCorrelation = .68

0

0.1

0.2

0.3

0.4

0.5

0.6

0.7

0.8

0.9

LOP Deviations

Dis

trib

uti

on

Mar

gin

s

Distribution Margins vs Endogenous ExitsCorrelation = -.37

0

0.1

0.2

0.3

0.4

0.5

0.6

0.7

0.8

0.9

Endogenous Exits

Dis

trib

uti

on

Mar

gin

s

Distribution Margins vs CPI StickinessCorrelation = -.42

Tomatoes

0

0.1

0.2

0.3

0.4

0.5

0.6

0.7

0.8

0.9

0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8

0 0.1 0.2 0.3 0.4 0.50 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9

0 0.1 0.2 0.3 0.4 0.5 0.6

Probability of a Price Change

Dis

trib

uti

on

Mar

gin

Fig. 1. Proximate determinants of distribution wedges.

70 D. Berger et al. / Journal of International Economics 88 (2012) 62–73

short-lived. We construct a variable that summarizes the short-livednature of such items, and see if this is systematically related to distri-bution wedges. We label this new variable the “endogenous exits”ratio. An endogenous exit is said to occur any time (1) the importingcompany has gone out of business; (2) the BLS industry analyst, inconsultation with the company, concludes that a product is “out ofscope”, indicating that there is no longer a meaningful market forthe product; or (3) highly significant changes in quality are made toan existing item. We then count, within each of the item categoriesused in the matching procedure, the number of items in that categoryexperiencing an endogenous exit during the sample period. The vari-able of interest, the endogenous exits ratio, is the ratio of this count tothe total number of items in that category.23

The unconditional relationship between endogenous exits anddistribution wedges is displayed in the scatter plot in the upper leftpanel of Fig. 1. Each dot represents a single item category, for examplebeer. The relationship is negative, with a simple correlation coeffi-cient of −0.37 and no clear outlier observations. Thus, in our data,item categories with small distribution wedges are those with

23 In any given year, about one-fifth to one-fourth of the items exit the sample “en-dogenously” in this way. This figure has remained fairly constant over time. Somewhatto our surprise, there is not a lot of cross-country variation when we count endogenousexits by country of origin of the imported item.

relatively many endogenous exits. We offer interpretations of thisfinding below.

5.2. Law of one price deviations

We next examine the relationship between distribution wedgesand deviations from the law of one price for the CPI items used inthe matching procedure. Conceptually, if the law of one price is closerto holding in the market for a particular item we may expect thatmarket to be more competitive and hence exhibit smaller distributionwedges. To be specific, we calculate absolute deviations from the lawof one price across cities for a particular type of, e.g., television set. Re-call that we have already determined particular items to be “identicalin description” through our matching procedure. These are the onlyitems whose (cross-city) price we are comparing in this exercise.For each category like televisions we calculate one number: the medi-an law of one price deviation across all of the individual city-pair ob-servations. We then relate this to the distribution wedge alreadycalculated for that item category.

The relationship is depicted in the upper right panel of the figure.There is clearly a positive relationship in our data: categories such astelevisions, VCRs, microwave ovens have very small deviations fromthe law of one price at the retail level; these are also the categorieswith the smallest distribution wedges.

71D. Berger et al. / Journal of International Economics 88 (2012) 62–73

The tri-variate relationship among distribution wedges, endoge-nous exits, and law of one price deviations at retail provides someeconomic insights. In item categories where the distribution wedgeis small there is a relatively large amount of product churning or turn-over, directly observed at the import stage, either because of signifi-cant quality changes or other market forces that render productsobsolete relatively quickly. These small-wedge (and high exits) itemcategories are also those in which market forces keep prices relativelyin line with the law of one price at the retail level.

5.3. Sticky prices

Finally we ask whether distribution wedges are related to mea-sures of price stickiness. On a priori grounds, we may expect thatthe sectors with low wedges and in which the law of one pricecomes close to holding, are also characterized by relatively flexibleprices. In the bottom row of the figure we depict scatter plots ofdistribution wedges against the probability that an item in thatcategory experienced a price change. We calculate these probabil-ities for both the CPI price (lower left panel) and the IPP price(lower right) of that item. We follow Nakamura and Steinsson(2008) in calculating the probabilities (we include both salesprices and regular prices in our CPI calculations). As seen in the fig-ure, the relationship is affected by a small number of outlier obser-vations. Excluding the outliers, the relationship is stronglynegative, −0.42 for CPI and −0.31 for IPP, so that lower wedgesare associated with more frequent price changes.24 On the CPIside, the one outlier category is tomatoes, for which prices arequite flexible while distribution wedges are high. This presumablyreflects a relatively unique combination of (1) supply-side compe-tition, product homogeneity and low demand elasticities inducingfrequent price changes and (2) costly transport and storageneeds that keep wedges high. On the IPP side, tomatoes are againan outlier, as are bananas and olive oil.

This simple, non-structural examination of the determinants ofdistribution wedges suggests an interesting relationship among dis-tribution wedges and three “micro features” of the BLS data: endoge-nous exits, law of one price deviations, and sticky prices. Therelationship points to the likely strong role of factors that we wouldexpect to see influencing distribution wedges—competition, productsubstitutability, transportation and storage costs. These relationshipsshould motivate future work that more precisely explains the chan-nels of these influences.

6. Conclusions

Using the detailed information on product characteristics in theCPI and IPP databases of the U.S. Bureau of Labor Statistics, wematch items imported into the United States to those sold at retailthat are identical in description. We compute the size of the resultingdistribution wedge of CPI price relative to import price and then in-vestigate the determinants of these wedges. We find the following,

1. Distribution wedges for the United States are large.Our calculation is in the range of 50–70% for U.S. data betweenJanuary 1994 and July 2007. Wedges are slightly higher underthe “alternative procedure” than baseline calculations obtainedfrom the detailed “matching procedure.” Back of the envelopecalculations using a simple modeling framework of Burstein etal. (2005) imply that the size of the “pure” tradeables sector inthe U.S. is thus in the range of 7–16%.

24 With the outliers, the correlation is essentially zero. Note that this negative rela-tionship is consistent with the theoretical and empirical work presented in Gopinathand Ishtoki (2010).

2. Wedges are larger than previously reported.Our headline number is about 10 to 20 percentage points higherthan a consensus estimate of 40–45% which was essentiallyobtained using NIPA data (Burstein–Neves–Rebelo, Goldberg–Campa, Bradford-Lawrence). This maps into a calculation ofthe “pure” tradeables sector that is 5 to 10 percentage pointslower than the 22% number reported by BER. Differences be-tween our results and those of the existing literature appear tobe driven by differences in the data sets used, rather than bycompositional effects. Since our calculations using the BLS dataare built up from the microeconomic level, we hope they pro-vide a cleaner calculation of distribution wedges than was pos-sible before.

3. Wedges are stable over time but vary considerably across items.Under the matching procedure, the average annual distributionwedge across all items is 0.62, 0.67, 0.63, 0.57, 0.59, 0.60, 0.58,0.61, 0.59, 0.57, 0.58, 0.60, 0.60 and 0.61 in the years 1994 through2007 respectively. The relative stability of wedges coincides withlarge fluctuations in the dollar over time. Across item categories,several exhibit low wedges: televisions, video cameras, VCRs, cam-eras, telephones, microwave ovens, while other categories havehigh wedges: drugs, apparel (men's, women's pants), watches,film, bananas, tomatoes.

4. Wedges do not vary dramatically with exchange rates or acrossmajor exporters.Our measures of the distribution wedge are relatively steady,during a period when dollar first appreciated by more than 20%and subsequently depreciated by more than 30%. Regression re-sults confirm the lack of a relationship between changes inwedges and exchange rates. Underlying this result is a lack of aconsistently significant relationship between exchange ratesand IPP or CPI prices. When we slice the data by the country ofexport, most of the wedges fall within the range of 50% to 60%.This sheds new light on issues first raised by Engel (1999),and further examined empirically by Burstein and co-authors,Goldberg and Campa.

5. Variation in wedges is explained by proxies for sectoral character-isticsBetween categories, distribution wedges vary negatively with en-dogenous exits and frequency of price changes, and positivelywith law of one price deviations in the retail market. Thus, in cat-egories where the wedge is small there is a relatively large amountof product churning or turnover, directly observed at the importstage. This turnover occurs (?) because significant quality changesare made to the product or because other market forces renderthat product obsolete relatively quickly. These small-wedge itemcategories are also those in which market forces lead to relativelyfrequent price changes and keep prices relatively in line with thelaw of one price at the retail level.

Appendix A. The Matching Procedure

As noted in the text, in calculating distribution wedges d wecompare the price of an item in the IPP database to that of amatched item in the CPI database. We match items that are identi-cal in description. This appendix provides details on the criteria weused to construct these matches. Naturally these criteria differedacross item categories. Each potential match was given a gradethat depended on how many of the criteria were met successfully.For example, as described below, there were 5 criteria that had tobe met in order for there to be an “A Grade” match for that item:product (e.g., vodka), proof (e.g., 80), size of the container (e.g.,1 l), brand, and country of origin. When a particular criteria wasnot met, it was usually because that piece of information was miss-ing. In those cases when there was an obvious mismatch on a

72 D. Berger et al. / Journal of International Economics 88 (2012) 62–73

criteria, e.g., brand of beer, an F grade was given. In our empiricalwork we used only A grade and B grade matches.

Category: AlcoholGrades: size, proof, BRAND, product, country of origin (mostly A's)Category: AudioplayerGrade Scale

A=Brand, model number+other characterB=Model number+other characteristicsC=partial model number+other characteristics

Category: BananasMatch criteria: Brand, Country, Quality=A

B=a country (or brand) discrepancy from a mid sample switch inthe CPIC=country & brand tend to be off

Category: BeerGrades: Brand, type, bottles vs. cans, country of origin

A=got them allB=usually type and/or country of origin unknown in IPPC=unknown type and container (usually in IPP) and unknowncountry of origin (usually in CPI)

Cans vs. bottles given an FDiscrepancy in container size given an FCategory: CalculatorGrades: Model Number and Brand (C tends to be off on a TI 83

"Plus" vs. no Plus)Category: Cameras

If it matched on brand and model number gave it A;If model number was missing but brand matched gave it a C;If it matched model number but brand was missing gave it a B/Cdepending on how unique the model number seemed;If brands were different gave it an F.

Category: Computer Accessories

A match on brand, model number, screen size/resolutionB match on screen size and model number+other characteristicsBC match on screen size, other characteristics and partial serialnumber (not enough info in IPP to know for sure)

Category: DrugsGrade Scale

A=matches on brand, type and size of package and form of drugB=believed to be an exact match but a major product character-istic is not listed in one of the descriptionsC=one major characteristic is off matches on brand but ipp size ishalf of what we want (C2) ; or (brand x) vs. (brand x max strength)F=not a match

Category: FilmCriteria: Brand, Shutter Speed, ExposuresNumber=Ratio of #rolls in CPI to IPPCategory: Men's PantsQualities: style number, brand, item description

A=style number, brand, and basic similarity of item descriptionB=often a downgrading from an A when the style number chan-ged in the CPI sampleC=info just too spottyF=clear country of origin contract (cpi=made in USA)

Category: MicrowavesLetter Grade: Brand, cubic feet, watts

A=all 3B=brand info missing in IPP, cu ft and watt matchC=cu ft or watt info missing or off in IPP, brand info missing inIPPF=watt and cu ft off or missing, brand info missing

Category: Miscellaneous Kitchen Appliances

Brand+model number=A.Really long model number and other miscellaneouscharacteristics=A.Model number match=B.

Anything else C or below.Category: Olive oil

A Matches on Brand, Type, Size and Bottle Type (plastic vs glass vscan)B Matches on Brand, Type, SizeC Matches on Brand and SizeF Not a Match

Category: Phones

A: if it matched model/brand and serial OR serial was at least 7digits and it matched other characteristicsB: matched serial only and serial was at least 4 digitsC: Matched serial and serial was 3 digits or lesF: matched nothing OR there was definitive evidence the twowere different products.

Category: StovesGrades: mostly matched on serial numbers (Brand like George

Foreman grill, specs like Bun Warmer) (not much else to go onother than proximity of the serial numbers)

Category: TomatoesWe matched on brand, country of origin, type (cherry vs roma)

and how it was grown (vine vs green house). If it hit brand andtype and at least one of country and how it was grown (and no dis-crepancy in other) then it was an A. Otherwise if it type and at leastone of country and how it was grown, then it was a B.

Category: Televisions

A matches on brand, model # and size at least

B matches on model# and size at least (and may have contradicto-ry country of origin info)C partial model # match and sizeF not a match (wrong size etc.) or made in USA

Category: VCRQualities: Verbal description of item, model number, brand, coun-

try of origin

A=got everything essentially

B=model number and item description mostly, sometimes abrand match as well (still gave it B)

Category: VideocamerasGrades: mostly Bs=model number (good matches) and basic

item description. Country of origin typically not in CPI, brand typicallynot in IPP

Category: Watch

A full serial number, country of origin/brand and otherindentifiers

B serial number, country of origin+other indentifiers

C partial serial number, country of origin

F not a match

73D. Berger et al. / Journal of International Economics 88 (2012) 62–73

Category: Women's Pants

A identifiable brand, style number type of pants and country oforiginB usually no brand or country of originC incomplete style number, no brandF made in USA or not a match.

Appendix to Table 1a and 2a: number of observations (intra-com-pany transfer prices excluded)

Regular price (CPI) Sale price (CPI)

cif fob cif fob

All items 6054 14,090 1018 3316Alcoholic beverages 2959 1108 660 166Audio players 150 290 23 58Bananas – 3482 – 793Beer 1509 3155 148 463Calculators – 214 – 19Cameras – 199 – 39Computer accessories 20 110 – 8Drugs 7 89 – –

Film 581 114 42 –

Men's pants – 128 – 328Microwave ovens – 126 – 81Kitchen equip. (misc.) – 157 – 42Olive oil – 170 – 58Telephones 65 341 – 41Stoves 82 246 29 71Tomatoes 513 2231 62 544Televisions 105 820 22 308VCRs 56 633 29 153Video cameras 7 206 – 37Watches – 134 – 60Women's pants – 103 – 35

Appendix to Table 1a and 2a: number of observations (intra-com-pany transfer prices only).

Regular price (CPI) Sale price (CPI)

cif fob cif fob

All items 546 6210 173 1259Alcoholic beverages 230 7 92 –

Audio players 172 130 34 9Bananas – 3447 – 573Beer – 122 – 14Calculators – 36 – –

Cameras – 299 – 46Computer accessories – 30 – 14Drugs – 51 – –

Film – 21 – –

Men's pants 12 149 – 96Microwave ovens – 10 – 7Kitchen equip. (misc.) – 171 – –

Olive oil – 69 – 12Telephones – 226 – 18Stoves 76 14 21 –

Tomatoes 6 10 – –

Televisions 18 797 6 287VCRs – 283 – 72Video cameras – 266 – 93Watches – 49 – –

Women's pants 27 10 20 –

References

Anderson, J., van Wincoop, E., 2004. Trade costs. Journal of Economic Literature 42 (3),691–751.

Atkeson, A., Burstein, A., 2008. Pricing to market, trade costs, and international relativeprices. American Economic Review 98 (5), 1998–2031.

Bernanke, B., 2007. Globalization and monetary policy. Speech at the Fourth EconomicSummit. Stanford Institute for Economic Policy Research, Stanford, CA.

Bils, M., Klenow, P., 2004. Some evidence on the importance of sticky prices. Journal ofPolitical Economy 112 (5), 947–985.

Bradford, S., Lawrence, R., 2003. Paying the price: the cost of fragmented internationalmarkets. The Peterson Institute for International Economics.

Burstein, A., Neves, J., Rebelo, S., 2003. Distribution costs and real exchange rate dy-namics. Journal of Monetary Economics 52 (6), 1189–1214.

Burstein, A., Eichenbaum, M., Rebelo, S., 2005. Large devaluations and the real ex-change rate. Journal of Political Economy 113 (4), 742–784.

Choudri, E., Faruqee, H., Hakura, 2005. Explaining exchange rate pass-through in differ-ent prices. Journal of International Economics 65 (2), 349–374.

Corsetti, G., Dedola, L., 2005. A macroeconomic model of price discrimination. Journalof International Economics 67 (1), 129–156.

Corsetti, G., Dedola, L., Leduc, S., 2008a. International risk sharing and the transmissionof productivity shocks. Review of Economic Studies 75, 443–473.

Corsetti, G., Dedola, L., Leduc, S., 2008b. High exchange rate volatility and low pass-through. Journal of Monetary Economics 55 (6), 1113–1128.

Devereux, M., Engel, C., Tille, C., 2003. Exchange rate pass-through and the welfare ef-fects of the euro. International Economic Review 44 (1), 223–242.

Engel, C., 1999. Accounting for real exchange rate changes. Journal of Political Economy107, 507–538.

Goldberg, L., Campa, J., 2010. The sensitivity of the CPI to exchange rates: distributionmargins, imported inputs, and trade exposure. The Review of Economics and Sta-tistics 92 (2), 392–407.

Goldberg, P., Hellerstein, R., 2008. A structural approach to explaining incomplete ex-change rate pass-through and pricing to market. American Economic Review 98(2), 423–429.

Gopinath, G., Ishtoki, O., 2010. Frequency of price adjustment and pass-through. Quar-terly Journal of Economics 125 (2).

Gopinath, G., Rigobon, R., 2008. Sticky borders. Quarterly Journal of Economics 123 (2),531–575.

Gopinath, G., Ishtoki, O., Rigobon, R., 2010. Currency choice and exchange rate pass-through. American Economic Review 100 (1), 304–336.

Gopinath, G., Gourinchas, P.O., Hsieh, C., Li, N., forthcoming. Cross-border prices, costs,and mark-ups. American Economic Review.

Hummels, D., Kejriwal, M., Naknoi, K., 2010. Exchange rate pass-through and marketstructure in a multi-country world. Working paper, Purdue University.

Nakamura, E., Steinsson, J., 2008. Five facts about prices. Quarterly Journal of Econom-ics 123 (4), 1415–1464.

Nakamura, E., Zerom, D., 2010. Accounting for incomplete pass-through. Review ofEconomic Studies 77 (3), 961–1013.