How Do Exporters Adjust to Exchange-Rate Fluctuations?

Transcription

1 Working aer How Do Exorters Adjust to Exhange-Rate Flutuations? New Evidene from the East Afrian Community Alan Asrilla Niolas Berman Olivier Cadot Marguerite Duonhel Mélise Jaud Marh 2013 When iting this aer, lease use the title and the following referene number: F UGA-1

2 How Do Exorters Adjust to Exhange-Rate Flutuations? New Evidene from the East Afrian Community Alan Asrilla Niolas Berman Olivier Cadot Marguerite Duonhel Mélise Jaud Marh 2013 Abstrat We use a large samle of exort transations from ustoms files aross six develoing ountries and several years to exlore the extent of riing to market and volume resonses to exhange-rate variations in the East Afrian Community (EAC), a ustoms union, and a ontrol grou of exorters from develoing ountries outside the region. We find that, relative to the ontrol grou, EAC exorters seem to have a stronger ability to rie to market on the CU market, suggesting the existene of market ower. This market ower does not seem to relate to usual roxies for firm size, but is more marked for manufatured roduts. We also find that the suly (volume) resonse to exhange-rate variations is more subdued for EAC exorters than for the ontrol grou, suggesting the existene of suly onstraints. JEL lassifiation numbers: F41 Keywords: Exhange rate, riing to market, EAC 1

3 1 Introdution In a reent aer, Freund and Pierola (2012) identified 92 surges (seven-year growth aelerations) of aggregate manufatured exorts and unovered three striking emirial regularities: First, surges are more likely in oen eonomies or eonomies that are liberalizing. Seond, surges are reeded by a large dereiation of the real exhange rate and lower exhange rate volatility. In fat, in develoing ountries, the real dereiation is large enough so as to leave the exhange rate undervalued by 20% on average. And third, the extensive margin the disovery of new roduts and new markets is an imortant omonent of exort surges in develoing ountries, aounting for over 40% of total manufaturing exort growth during the surge. (. 389) The role of the exhange rate, whih ehoes revious work on maro growth aelerations (Hausmann, Prithett, and Rodrik, 2005; or Rodrik, 2008), has imortant oliy imliations for develoing-ountry governments. Many still try to engineer exort surges with exort-roessing zones and tax inentives, although exeriene suggests that tax inentives are ostly and rarely effetive. If a ometitive exhange rate was, as suggested by the reent literature, suh a entral driver of exort growth aross ountries, there would be a strong argument to shift oliy attention to exhange-rate management away from industrial oliy. However, miro evidene on the link between exhange rates on one hand and exort ries and volumes on the other is ambiguous, although it would be a key ste to show that the maro orrelations were not surious. For one thing, most of the literature on exhange-rate ass through (ERPT) has attemted to identify short-run orrelations rather than the long-run one unovered by Freund and Pierola, and this aer will be no exetion to that. But the ambiguity goes beyond short-run vs. long run. A large number of aers, inluding e.g. Goldberg and Knetter (1997), Cama and Goldberg (2005) and Goinath and Rigobon (2008), found low elastiities of imort ries to exhange-rate movements, a henomenon alled the exhange-rate disonnet. 1 Low rie ass-through an be related to a number of features of the destination market. First, there an be what Engel (2003) alled loal urreny riing, i.e. short-run nominal rigidities in the destination market. Seond, there an be riing to market (PTM), i.e. otimal marku determination by firms with market ower, leading to a damening of exhange-rate ass-through (we will formally exlore this hannel in Setion 2 below). Atkeson and Burstein (2008) and Berman, Martin and Mayer (2012), among others, rovide theoretial and emirial exlorations of the PTM hannel. Third, fixed loal distribution osts an ontribute to redue the reation of onsumer ries to exhange-rate flutuations (see Burstein, Neves and Rebelo 2003, or Goldberg and Cama 2010). A key argument in the PTM literature is that if large, more rodutive firms have more market ower, they will be the least susetible of assing through exhange-rate movements, reventing the transmission of exorter-urreny devaluations into added demand and growth. This an be shown in a model with CES referenes and additive distribution osts as in Berman, Martin and Mayer (2012) or in a model with quasi-linear referenes and variable markus as in Mayer and Ottaviano (2011). As large firms aount for a large fration of aggregate exorts (see Freund and Pierola 2012), aggregate exort ries and volumes will largely reflet the behavior of those firms. 1 The term goes bak to Obstfeld and Rogoff (2001). 2

4 Under that argument, the lak of reation of exort ries and volumes to exhange-rate flutuations is attributable to market ower. However, ass-through in a heterogeneous-firms model deends on the interlay of firm and destination-market harateristis whose net effet is an emirial question. Moreover, in develoing ountries, large firms are not neessarily those with more market ower as they may be ommodity exorters; whereas smaller firms serving regional markets roteted by tariffs, NTBs or high transortation osts may have more market ower. Freund and Pierola s third emirial regularity the relative imortane of the extensive margin in exort surges is also surrising, as reent deomositions of exort growth at the rodut level all suggested that it was dominated instead by the intensive margin (see e.g. Evenett and Keller 2002, Besedes and Prusa 2007, Brenton and Newfarmer 2007, or Amurgo-Paheo and Pierola 2007). However, miro evidene an hel reonile the onfliting evidene. Chaterjee, Dix-Carneiro and Vihyanond (2012) showed in a rodut-ladder model that an exhange-rate rise (a devaluation of the exorter s urreny) leads firms to broaden their rodut ortfolio, as infra-marginal roduts beome rofitable at the new exhange-rate. Thus, the missing miro linkage here may be the relationshi between exhange-rates and rodut-ortfolio deisions at the firm level. This aer exlores the onjetures above (riing-to-market and extensive-margin behavior) on the basis of a large, multi-ountry firm-level dataset with several million transations obtained from ustoms administrations in six develoing ountries. The ability to ool together firm-level data from several ountries is a first and lends itself to a systemati exloration of the drivers of riing to market. The artiular nature of the dataset, whih inludes four ountries from the same ustoms union the East Afrian Community makes it ossible to exlore PTM in a relatively large, integrated Southern regional market. Results for this samle of ountries suggest that there is less riing to market than observed on industrial-ountry datasets, and in artiular the Frenh dataset used by Berman, Martin and Mayer. Over the whole samle, the PTM oeffiient (the elastiity of exorter-urreny exorter ries to the exhange rate) is insignifiant. However, the PTM oeffiient is high (around , against around 0.1 on industrial-ountry samles) and highly signifiant on bilateral EAC trade, suggesting the existene of substantial market ower. The volume elastiity is very high between 2.3 and 3.6 (imlying an elastiity of substitution between 4 and 7 for lausible values of the arameters, a value itself within the onventional range). Thus, urreny flutuations seem to have strong effets on both exort ries and volumes in the EAC market. We also find that rodut-ladder effets, although quantitatively small, are resent; that is, there is more riing to market for ore roduts, in whih firms may have more market ower. The aer is organized as follows. Setion 2 gives the analytis behind the riing-to-market oeffiient. Setion 3 rovides some bakground desritive statistis on EAC exorters. Setion 4 disusses estimation issues and Setion 5 the results. Finally, Setion 6 onludes. 2. Bakground In order to highlight the assumtions that need to be embodied in a standard heterogeneous-firms model to generate inomlete exhange-rate ass-through and generate some reditive omarative-statis roerties we resent here a stried-down riing equation to whih we add 3

5 ingredients one by one, highlighting how the hanges introdued affet the relationshi between exhange rates, ries and volumes. The funtional form for demand assumed in the literature is tyially one of two alternatives: Dixit- Stiglitz, whih generates a onstant-elastiity demand, or linear-quadrati (see e.g. Ottaviano, Tabuhi and Thisse 2002; Ottaviano and Melitz 2008, or Mayer and Ottaviano 2011), whih generates a linear demand with variable elastiity. Consider the first ase, whih dominates the literature. Assume a ontinuum of differentiated varieties, eah rodued by a firm with rodutivity, with an elastiity of substitution. Letting x be the demand for variety in a generi destination, referenes are CES with 1/ U x d (1) X where 1 1/. As is well known, this funtional form generates a onstant-elastiity demand funtion of the form 1 (2) x Y P where Y is the destination ountry s inome, P its aggregate rie index and the onsumer rie of variety. This formulation imlies a onstant marku over marginal ost whih itself imlies mill riing a formula whereby the firm alies the same roduer rie to all of its exort markets. In that ase, there is full exhange-rate ass through (ERPT). To see this, let e be the exorter s exhange rate, 2 e the roduer rie in the exorter s urreny, and observe that marginal revenue is d x 1 1 dx where dln / dln is the elastiity of the onsumer rie to the roduer rie in the imorter urreny. Assume that osts are the sum of a fixed, destination-seifi ost f, e x / : At the firm s otimum, (3) and a variable ost x x,, e f, e (4) 1 1 where the term in arentheses is the firm s marku over marginal ost. Pries (5) 2 Throughout, we will define the exhange rate in the onventional manner as the number of units of the exorter s urreny er unit of the imorter s one, so a raise in e will be a dereiation of the exorter s urreny. 4

6 The degree of ERPT on ries deends on, i.e. on the assumtions made about transortation and distribution osts. With a simle ieberg transortation ost τ, so 1and (5) redues to so (6) 1, (7) 1 dln 0 dln e (8) and mill riing alies, whih means that there is full ERPT. Thus, under Dixit-Stiglitz referenes, heterogeneity of firms is not enough in itself to destroy the onstant-marku/millriing roerty of monoolisti-ometition models. Suose, however, that there is an additive distribution ost in the imorting ountry, as in Berman, Martin and Mayer (2012) or Chatterjee, Dix-Carneiro and Vihyanond (2012), so (9) where τ is an ieberg transortation ost and η is a distribution ost inurred in the imorting ountry and exressed in that ountry s urreny. 3 Then, after some maniulation,, (10) e and the exhange-rate elastiity of the roduer rie, in the exorter s urreny, is (11) dln e. (12) dln e e In that ase, mill riing ( 0) does not aly anymore, as β is stritly between zero and one. Instead, the degree of riing to market (how lose β is to one, as β =1 imlies full riing to market) deends on the arameters, with four testable omarative-statis roerties: 3 There is a surrising disonnet between reality and ommon ratie in trade theory in this formulation. First, freight rates harged by shiing omanies are not ad-valorem but by weight. For instane, two quote requests from Maersk for a 20 ontainer to be shied from Singaore to Rotterdam, one with fertilizers (low unit value) as the delared merhandise and the other with olletors antiques (high unit value) returned an exatly idential offer of $ , of whih base freight was $ Seond, wholesalers and retailers tyially harge roortional (ad-valorem) ommissions, as the authors verified from business intermediaries. Thus, in (9) the additive omonent η should be the transortation ost and the multiliative omonent τ should be the distribution ost. We will leave this issue for further researh and stik to the onventional formulation here. 5

7 1. 0: More rodutive firms rie more to market, 2. 0: More riing to market in destinations with higher distribution osts, 3. 0 : Less riing to market in faraway destinations, 4. 0 : Less riing to market in destinations where ometition is tougher. In addition, multi-rodut firms with indeendent rodut lines and varying levels of effiieny aross roduts will aly different riing rules aross their ortfolio of roduts. If firms are more effiient in their ore roduts (those having a high share in their exort sales) they will behave, for those roduts, like large firms, i.e. rie more to market (Chatterjee et al. 2012). This gives a fifth testable roerty, namely that if a firm rodues (and exorts) n roduts ranked from the largest to the smallest, n whene n. That is, there is more riing to market for ore roduts. Volumes The effet of exhange-rate variations on volumes an be alulated in a similar fashion. Logdifferentiating (2) with reset to e, q dln x ln ln dln e ln e ln e 1 onst. e onst. (13) As is stritly between zero and one, ries, whereas with full riing to market, q is stritly ositive. Seifially, with full ERPT on lim q 0 lim 0 (14) q. (15) 1 In the baseline ase with no distribution osts (η = 0), mill riing imlies a volume elastiity of q exorts to the exhange rate just equal to the elastiity of substitution ( ). With distribution 6

8 osts, the exhange-rate elastiity of volumes an be smaller or larger than one, deending on the relative magnitudes of σ and η. Comarative-statis roerties have the same sign as ERPT (but oosite to β ): The more assthrough on ries, the larger the indued volume hange. 4 q 0: More rodutive firms have lower volume elastiity, q 0: Lower volume elastiity in destinations with higher distribution osts, q 0 : Higher volume elastiity for more faraway destinations, q 0 : Higher volume elastiity in destinations where ometition is tougher. Although omarative statis signs are oosite for ries and volumes, the o-movement between the two is omlex. As an illustration, onsider a stried-down examle with a develoing ountry exorting to two sharly (and somewhat artifiially) differentiated destinations: a Northern one haraterized by long distane, tough ometition, and high distribution osts, and a Southern one with short distane, low ometition, and low distribution osts. Seifially, assume that in the Northern market, σ N = 5, τ N = 1.6, and η N = 1; that is, transortation osts add 60% to FOB value and distribution osts add 100% to CIF value (at the baseline rie, sine they are not ad-valorem). With φ = 1, = 1, e = 1, β N = 0.11 and β qn = Assume now that in the neighboring Southern market, σ S = 2, τ S = 1.3, and η N = 0.2; that is, transortation ost is 30% of FOB rie and distribution ost is 20%. Then β S = and q qs = In this ase, riing to market is stronger on the Northern market (less ERPT), but so is the reation of volumes to a hange in the exhange rate; the ombination of arameter values indues ositive o-movement in rie and destination elastiities aross destinations. We now turn to an emirial exloration of these effets in a multi-ountry dataset with exorters from develoing ountries. 2 Data We use a new multi-ountry, firm-level dataset obtained from the Customs administrations of eight ountries : Bangladesh, Kenya, Moroo, Rwanda, Tanzania, and Uganda. The data were forwarded to us in the form of ASYCUDA sreadsheets and inlude all exort transations over a number of years (see Table 1) with, for eah transation, a firm identifier, a date (indiating month and year), the transation s destination ountry, the rodut s HS ode (tyially at ountry-seifi HS8- equivalent levels), the transation value in loal urreny, and a host of other variables of lesser interest for this aer. 4 This is under the assumtion of onstant marginal osts (although not equal aross firms), where all volume hanges are along the demand urve. 7

9 Table 1 gives basi samle-size and samle-eriod information by origin ountry. The overall samle is dominated by its three large origins, Bangladesh, Kenya and Moroo, in terms of transations (both total and yearly) and number of firms. All origin ountries have diversified destination ortfolios, and the total number of HS6 roduts exorted in one year or another ranges between (Rwanda) and (Kenya), out of a notional total of about HS6 lines. Number of years Table 1 Cross-ountry data summary Number of transations Transations er year Number of firms Number of destinations Number of roduts a/ Bangladesh 7 ( ) 412'000 58'857 13' '784 Kenya 7 ( ) 255'314 36'473 9' '660 Moroo 9 ( ) 463'386 51'487 17' '391 Tanzania 7 ( ) 44'408 6'344 4' '267 Uganda 8 ( ) 36'919 4'615 2' '940 Rwanda 7 ( ) 8'186 1'169 1' '415 Notes a/ Produts have been aggregated to the ommon HS6 lassifiation. Exhange rates vis-à-vis the U.S. dollar are from the IMF s International Finanial Statistis (IFS) and are deflated by onsumer rie indies to obtain real exhange rates (RER). They are all exressed in loal urreny units (LCU) er dollar in the IFS. Let eo and e d be resetively the origin and destination ountries exhange rates in LCU er dollar, and o and d their onsumer rie indies. Our bilateral exhange-rate variable, in logs, is thus eo / o e o d lneod ln ln ln. (16) ed / d ed o There was substantial volatility in exhange rates for EAC ountries in our samle eriod. Figure 1 shows bilateral nominal exhange rates vis-à-vis the dollar, in loal urreny units er dollar. Both volatility and lak of o-movement over the samle eriod are aarent. 8

10 (a) Uganda Figure 1 Bilateral dollar exhange rates, EAC ountries (b) Kenya Year Year () Rwanda (d) Tanzania Year Year In some of the regressions below, we will deomose RER effets into nominal exhange-rate effets and rie effets in order to exlore asymmetries in the rie resonse of exorters to nominal vs. real hanges. Table 2 shows desritive statistis for the variables used in the regressions. 9

11 Table 2 Desritive statistis Variable Obs Mean Std.Dev. Min Max Volume (Weight in kg) 1'168' '449 4'062' E+09 ln(volume) 1'168' Unit Value 1'168'627 11'669 2'299' E+09 ln(unit Value) 1'168' Real Exhange Rate (RER) 1'128' '982 ln(rer) 1'128' Nominal Exhange Rate (RER) 1'160' '126 CPIratio 1'129' Distane to Caital 1'168'627 4'530 3' '884 ln(dist. To Ca.) 1'168' GDP onstant (2000) 1'156' E e E E+13 ln(gdp) 1'156' GDP PPP (Const 2005) 1'156'005 23'179 16' '987 ln (GDP PPP) 1'156' Number of Produts (HS6) 1'168' ln(number of Produts (HS6) 1'168' Stylized fats As our dataset is the first firm-level trade dataset to over four EAC ountries, we summarize in this setion a few stylized fats of interest about East Afrian exorters as they emerge from our data. 3.1 East Afria s exorters : What do we know? It has been observed that the distribution of exort sales aross firms is heavily skewed (see e.g. Freund and Pierola 2012) with the bulk of exort values being aounted for by a few large firms. Skewness is aarent in our samle as the mean value of annual firm-level exort turnover is thirty to sixty times the median. There is muh less skewness in terms of number of destinations and roduts (Table 3), suggesting that large firms differ more from small ones at the intensive margin than at the extensive one. Table 3 Exorter harateristis 10

12 Total exort turnover a/ Destinations Produt Mean Median Mean Median Mean Median Uganda 2'405'939 37' Kenya 1'042'485 34' Tanzania 2'667'656 46' Rwanda 584'311 17' Relatedly, the diversifiation of EAC exorters in terms of roduts follows a non-monotone ath. Table 4 shows anel regression results for the firm-level Herfindahl index of rodut onentration on the log of exort turnover in U.S. dollars. Both between and within firms, the relationshi is non-monotone, with firms first diversifying and then re-onentrating. 5 The turning oint, alulated as x ex / where 1 and 2 are the regression oeffiients on the linear and quadrati terms resetively, is between $ and $ deending on the estimation method ($ for manufatured roduts), leaving, on average, about 16% of exorter-year observations to the right; that is, about 16% of all exorter-year observations in our ross-ountry samle are in the re-onentration hase. A lausible exlanation is a roess of exerimentation whereby firms try a number of roduts on exort markets and then onentrate on the most rofitable, 6 imlying that rodutivity imrovements aearing as learning by exorting might in fat be within-firm, between-roduts omosition effets. 5 This attern was observed at the ountry level in terms of rodution (Imbs and Waziarg 2002) and in terms of exorts (Cadot, Carrère and Strauss-Kahn 2011). To our knowledge, it has not been observed on firms. Whereas Table 4 refers to EAC exorters, the same attern holds for the whole samle, inluding Moroo and Bangladesh. 6 Whether firms learn from exorting has been a long-standing debate in the literature. At the firm level, the early ( re-melitz ) emirial literature on the rodutivity-exort linkage was rediated on the idea that firms learn by exorting (see e.g. Haddad 1993, Aw and Hwang 1995, or Tybout and Westbrook 1995). However, Clerides, Lah and Tybout (1998) argued theoretially that the rodutivity differential between exorting and non-exorting firms was a seletion effet, not a learning one, and found suort for this interretation using lant-level data in Columbia, Mexio and Moroo. Subsequent studies (Bernard and Jensen 1999; Eaton et al. 2004, 2011; Helman et al. 2004) onfirmed the imortane of seletion effets at the firm level. More reently, however, aers fousing on miro-level data have found some evidene of learning-by-exorting (e.g., Girma, Greenaway and Kneller 2004, van Biesebroek 2005; De Loeker 2007; Aw, Roberts and Winston 2007 or Cresi et al. 2008). 11

13 Table 4 Produt onentration at the firm level, EAC exorters Estimator OLS FE FE De. Var. Herfindahl onent. index, by rodut (1) (2) (3) Log exort turnover (55.68)*** (36.11)*** (29.61)*** Log exort turnover, squared (46.20)*** (29.36)*** (23.66)*** Constant (84.42)*** (55.43)*** (44.06)*** Firm fixed effets No Yes Yes Time effets Yes Yes Yes Turning oint (USD) 652' ' '413 Observations 36'349 36'349 24'416 R-squared Number of firms 15'142 10'812 Note : Robust t statistis in arentheses, * signifiant at 10%; ** signifiant at 5%; *** signifiant at 1%; turnover in U.S. dollars for all firms. There is strong satial seialization, as the majority (slightly more than 60%) of the samle s EAC exorters realize over 95% of their exort sales on EAC or regional (EAC lus DRC and South Sudan) markets (anels (a) and (b) of Figure 2 resetively). Figure 2 Share of regional sales in total exort turnover, by firm (a) East Afrian Community (EAC) (b) EAC + DRC & South Soudan EAC share in firm exorts Region's share in firm exorts 12

14 However, the share of exort sales going to EAC markets is muh smaller, as EAC exorters tend to be smaller in terms of overall exort turnover than non-eac exorters. This is shown in Figure 3 whih lots a smoother (non-arametri) regression of the ratio of regional over total exorts at the firm-year level against entiles of the distribution of exort turnover, ordered from smallest to largest. The relationshi is strongly dereasing, with the largest entile (whih aounts for most of the ountry-level exort value) exorting less than 40% to regional markets. Figure 3 Share of regional sales in exort aross exort-turnover distribution Centiles of ln exort turnover bandwidth =.8 The heterogeneity of the oulation of exorters in terms of destination markets highlights the need to identify effets within firms in order to eliminate onfounding influenes due to seletion into destination markets. 4 Estimation The database is very large, but relatively oor in ovariates as it ontains no firm harateristis (ountries in the samle either do not arry out firm surveys are regular dates or, like Moroo, do not ommuniate them to researhers). 7 Thus, identifiation of the key omarative-statis roerty on firm rodutivity/size (φ) will have to rely on roxies. In the literature, rodut soe is the firmlevel observable that orrelates most losely, aross firms, with rodutivity. However, within firms, both the theoretial literature (Bernard et al. 2011, Ekel and Neary 2010) and the emirial one (Chatterjee, Dix-Carneiro and Vihyanon 2012) suggest that rodut soe is endogenous to the firm s environment. For instane, Bernard et al. (2011) Ekel and Neary (2010) show that firms otimally redue rodut soe (fous on their ore ometenies) after a trade liberalization as a result of ro-ometitive effets. The same ro-ometitive effets an be exeted from an exhange-rate dereiation (an areiation of the exorter s urreny), as shown by Chatterjee, Dix-Carneiro and Vihyanon (2012). This reates an obvious identifiation roblem. For now, this will be treated in two ways: First, using rodut soe at the firm level rather than at the firmdestination level, as endogeneity to the bilateral exhange rate would be at the firm-origin- 7 Uganda has a firm survey, but the ombined data is not suffiiently balaned to be usable for this aer. 13

15 destination rather than the firm level; seond (additionally), using lagged values of rodut soe as a regressor. We will reort results from both aroahes. In a robustness setion, we reort results using the (lagged) number of destinations and total exort value, both at the firm level, as alternative roxies for firm size. Transortation osts will be aroximated by bilateral distane, the degree of ometition by the size of the destination eonomy (GDP), and distribution osts by its inome level (GDP er aita). Let us define the following indies: o is origin ountry, d is destination ountry, f is firm, is rodut, and t is year. Let eodt be the real exhange rate between the origin and destination ountries in year t, as in (16) and fdt the roduer rie of rodut exorted by firm f to destination d at t, in ountry o s urreny (what was alled in the algebra of setion 2). 8 k 1 Let also. x odt odt (bilateral 2 distane), x dt Ydt, ydt,eacdt (destination GDP, aroximating σ, destination GDP/aita, aroximating η, and whether the destination belongs to the EAC ustoms union, to identify tradeoliy effets); x 3 4 ft lnn ft (firm f s number of roduts, aroximating φ), and x m (a dummy variable marking manufatured roduts). Finally, let δ od and δ fd be origin-destination and firm-rodut-destination fixed effets resetively. The baseline estimation equation is x, k = 1,,4, be four vetors of exlanatory variables. Seifially, ln ln e 4 x x x fdt 0 odt 1 odt 3 dt 4 ft 14 k 5 x k ln e k odt x. (17) u ot fd fdt Equation (17) is estimated by FE-OLS with robust standard errors. 4 Results This setion resents baseline results on the riing-to-market (PTM) oeffiient in terms of unit values and volumes, for both the whole samle and EAC ountries, as well as reliminary results on rodut-ladder effets and the effet of exhange-rate volatility on exort volumes. As results tables are bulky, they are relegated to the end of the aer. 4.1 Baseline In order to benhmark results for EAC exorters, Table 5 resents baseline regression results for the whole samle inluding non-eac exorters. The deendent variable is the log of exort unit values, and all regressions are estimated by OLS with fixed effets by firm-rodut-destination and by origin-year. The riing-to-market oeffiient the equivalent of β in (12) is the oeffiient on the log of the real bilateral exhange rate. As highlighted in Setion 2, we exet it to be between zero and one and to be higher (loser to one) for larger firms and markets with larger distribution osts (resumably higher-inome ountries), while being lower for more faraway 8 Origin subsrits an be omitted in the resene of firm subsrits given that firms in the samle are treated as if all ountry-level subsidiaries were indeendent entities.

16 destinations and destinations haraterized by tougher ometition. As high-inome destinations are likely to be haraterized by both tough ometition and high distribution osts, the effet of destination inome is, a riori, ambiguous. These roerties are exlored through interation terms between the bilateral RER and various roxies inluding distane, destination GDP er aita, and the number of roduts exorted by the firm (a roxy for its rodutivity). We also exlore asymmetries between areiations and dereiations of the exorter s urreny. Table 5 Baseline rie regression results, all samle By and large, the results reorted in Table 5 suggest that, omared to the literature, we observe little riing to market on our samle of develoing-ountry firms, whih aords with the intuition that those firms may have less market ower than firms loated in industrial ountries. The oeffiient on the log of the bilateral RER, whih is the equivalent of β in setion 2, is signifiant and lose to the usual estimate in industrial ountries (0.108) when inluded alone in the regression, but one other regressors are inluded, it beomes unstable and insignifiant. In other words, when the exorter s urreny dereiates, the tyial exorter in our samle kees his exort rie onstant and lets the buyer ature the rent. Unit values orrelate ositively with the destination ountry s inome level, in aordane with the lassi findings of Shott (2004). They orrelate negatively with the destination ountry s GDP, suggesting tougher ometition on larger markets. None of the interation terms of these exlanatory variables turns out signifiant, so there is at this stage no identifiation of our omarative-statis roerties. However, a key interation term, between the log of the lagged number of roduts and the real exhange rate, is ositive and highly signifiant, imlying more riing to market for firms with a broader soe. If broader soe is taken as a roxy for firm rodutivity (a higher φ), the sign of this interation term is in aordane with the basi redition of the CES model with additive distribution osts or with that of the QL model. However, quantitatively, this effet is very small. Table 6 Baseline volume regression results, all samle Table 6 reorts the estimated elastiity of volumes to the exhange rate, the equivalent of β q in setion 2. Here the results are stable and suggest a strong elastiity. Note that unlike β, β q is not onstrained by the model s logi to lie between zero and one. Assuming β = 0 (as the estimates in Table 5 are not signifiant), if transortation osts added on average 20% to the FOB rie of goods and distribution osts added 100% to the CIF rie, the estimates in Table 6 would imly, by inversion of (14), values of σ, the elastiity of substitution, between 4.26 and 6.6, a lausible range. Table 7 resents estimates restrited to the samle of EAC exorters. As for the whole samle, the PTM oeffiient is not signifiant. However, for seifiation (9) where the number of roduts (as a roxy for firm size) is not lagged, the PTM oeffiient interated with a dummy for manufatured roduts beomes signifiant, suggesting there is less ass-through for manufatured roduts. More imortantly, the PTM oeffiient is very large and signifiant for EAC bilateral trade. This means that the limited ometitiveness of EAC markets rovides market ower not only to foreign firms selling there, but also to EAC firms. PTM oeffiients between (olumn 10) and (olumn 12) means ass-through of at most a quarter of exhange-rate variations to FOB exort 15

17 ries, a very high degree of riing to market. Interestingly, the ability to rie to market does not seem to relate to firm size in the EAC, onfirming our onjeture that in oor ountries, firm size is not a good roxy for market ower (as larger firms tend to be ommodity traders with little or no market ower). Table 7 Baseline rie regression results, EAC exorters Table 8 reorts estimates for the volume oeffiient (β q ) for EAC exorters. Here the effet is either not signifiant or barely so (and with the wrong sign in olumn 12), a very notable differene with the result for the whole samle where it was signifiant in olumns with very high oint estimates (Table 6). Interated with a dummy variable for manufatured roduts, it beomes signifiant with oint estimates between in olumn 10 and in olumn 12. Thus, there is a suly resonse to exhange-rate variations for EAC exorters selling manufatured roduts, however, the striking differene with Table 6 is that the oeffiient is now substantially (three times) smaller, suggesting a muh weaker suly resonse. This is suggestive of onstraints and rigidities ossibly of lak of exess aaity as well in EAC ountries omared to the ontrol grou of Bangladeshi and Moroan exorters. Lak of exess aaity is onsistent with the ability to rie to market. Table 8 Baseline volume regression results, EAC exorters In sum, the omarison between the adjustment attern of EAC exorters to exhange-rate variations with that of a ontrol grou of out-of-region exorters suggests substantially more market ower and less ability to resond in terms of quantities to rie signals. 5.2 Extensions We noted in Setion 2 that if firms are more rodutive in their ore roduts, the riing-to-market oeffiient would be higher. We test this onjeture, due to Chatterjee et al. (2012), by inluding in our baseline regression additional interation terms at the firm-rodut-destination level between the log of the real exhange rate and the rank of the rodut in the firm s exort sales. We use several roxies for rodut s rank in firm f s sales in destination d at time t: Bottom fdt is a dummy variable equal to one when rodut is the smallest in firm f s ortfolio; not ore fdt is equal to one when rodut is not firm f s to rodut; seond fdt is equal to one when rodut is firm f s seond-largest; finally, ln(ranking) fdt is simly rodut s rank. Results are shown in Table 9. All rodut-ladder variables exet seond fdt have negative and signifiant oeffiient; however, the magnitudes are very small. To see this, note that, as before, the riing-to-market oeffiient is insignifiant for the whole samle, but large and highly signifiant for EAC bilateral trade. When rodut is firm f s bottom rodut, its PTM oeffiient is redued from to 0.446, an effet whih is very small even if it is identified with reision. Table 9 16

18 Regression results, rodut ladder In Baldwin and Gu (2009) and Noke and Yeale (2006), trade liberalization auses multi-rodut firms to onentrate on their most rofitable roduts, whih raises firm-level rodutivity through a seletion effet at the rodut level. In Mayer et al. (2010), firms onentrate on their highesterforming roduts when they exort to markets where ometition is intense, whih, again, translates into higher rodutivity at the firm level. In Table 10, we take a first ass at the effet of short-run exhange-rate volatility on exort volumes. We onstrut a new regressor equal to the absolute value of the jum in the exhange rate in the revious eriod. We find that exhange-rate volatility has a ositive effet on trade volumes for the whole samle, but not for EAC exorters. These results, however, are very reliminary. Table 10 Effet of exhange-rate volatility on exort volumes 6 Conluding remarks Our large ross-ountry dataset has made it ossible to exlore the effet of exhange-rate variations on the behavior of EAC exorters at the rie and volume margins in omarison to a treatment grou of out-of-region exorters. Our results are very reliminary and should be treated with the utmost aution, but they are, as they stand, suggestive of a number of distintive features of EAC markets and the fabri of exorters in member ountries. We find that the ability to rie to market, a roxy for market ower, aears substantial for EAC exorters when the sell in EAC markets. It also does not seem to related to firm size although the nature of our data obliges us to use rude roxies for firm size and rodutivity, the key arameters in a heterogenous-firms model suggesting that market ower is not about size on EAC markets. This aords with intuition, as medium-sized firms in manufatured-rodut setors roteted by high tariffs (the 25% band of EAC s CET) and non-tariff barriers are likely to have more market ower than larger firms selling ommodities on highly ometitive international markets. We also find that the volume resonse to exhange-rate variations, while signifiant and reasonably large (a unit elastiity) is substantially more subdued for EAC exorters than for the ontrol grou (for whih we find an elastiity around three), suggesting the existene of suly onstraints for EAC exorters. These onstraints may inlude aaity onstraints, exlaining the ability of EAC exorters to ature the rents of urreny devaluations (i.e. their ability to rie to market noted above). Finally, we find little evidene that short-term volatility er se enalizes EAC exorters in terms of volume exorted, although our results here require further researh. In terms of oliy imliations, our reliminary results suggest that exhange-rate oliy may not be enough, in itself, to trigger exort surges in the resene of other binding onstraints to exort growth. 17

19 Referenes Amurgo-Paheo, Alberto, and D. Pierola (2007), Patterns of Exort Diversifiation in Develoing Countries: Intensive and Extensive Margins. World Bank Poliy Researh Working Paer Atkeson, Andrew, and A. Burstein (2008), Trade Costs, Priing-to-Market, and International Relative Pries ; Amerian Eonomi Review 98, Aw, B.Y., and A. Hwang (1995), Produtivity and the Exort Market: A Firm-Level Analysis ; Journal of Develoment Eonomis 47, , M. Roberts and T. Winston (2007), Exort Market Partiiation, Investments in R&D and Worker Training, and the Evolution of Firm Produtivity ; The World Eonomy 30, Berman, Niolas; P. Martin and T. Mayer (2012), How do Different Firms Reat to Exhange-rate Changes? ; Quarterly Journal of Eonomis 127, Bernard, Andrew, and J.B. Jensen (1999) Exetional exorter erformane: ause, effet, or both? ; Journal of International Eonomis 47, 1-25., S. Redding and P. Shott (2011), Multi rodut Firms and Trade Liberalization, Quarterly Journal of Eonomis 126, Besedes, Tibor, and T. Prusa (2011), The role of extensive and intensive margins and exort growth ; Journal of Develoment Eonomis 96, Burstein, Ariel; J. Neves and S. Rebelo (2003), Distribution osts and real exhange rate dynamis during exhange-rate-based stabilizations ; Journal of Monetary Eonomis 50, Cadot, Olivier; C. Carrère and V. Strauss-Kahn (2011), Exort Diversifiation: What s Behind the Hum?, with Céline Carrère and Vanessa Strauss-Kahn, Review of Eonomis & Statistis 93, Cama, Jose Manuel, and L. Goldberg (2005), Exhange-Rate Pass Through into Imort Pries ; Review of Eonomis and Statistis 87, Chatterjee, Arita; R. Dix-Carneiro and J. Vihyanond (2012), Multi-Produt Firms and Exhange-Rate Flutuations ; Amerian Eonomi Journal, forthoming. Clerides, S.K., S. Lah and J. Tybout (1998) Is Learning By Exorting Imortant? Miro-Dynami Evidene From Colombia, Mexio, And Moroo ; Quarterly Journal of Eonomis 113, Cresi, G., C. Crisuolo and J. Haskel (2008), Produtivity, exorting, and the learning-byexorting hyothesis: diret evidene from UK firms ; Canadian Journal of Eonomis 41, De Loeker, J. (2007), Do Exorts Generate Higher Produtivity? Evidene from Slovenia ; Journal of International Eonomis 73, Eaton, Jonathan, S. Kortum, and F. Kramarz (2004), Disseting Trade: Firms, Industries, and Exort Destinations ; Amerian Eonomi Review 94, , and (2011), An Anatomy of International Trade: Evidene from Frenh Firms, Eonometria 79,

20 Ekel, Carsten, and J. Peter Neary (2010), Multi rodut Firms and Flexible Manufaturing in the Global Eonomy ; Review of Eonomi Studies 77, Engel, Charles (2003), Exenditure Swithing and Exhange-Rate Poliy ; NBER Maroeonomis Annual , Freund, Caroline, and D. Pierola (2012), Exort Surges ; Journal of Develoment Eonomis 97, and (2012), Exort Suerstars ; World Bank Poliy Researh Working Paer #6222; Washington, DC: The World Bank. Girma, Sourafel, D. Greenaway and R. Kneller (2004), Does Exorting Inrease Produtivity? A Miroeonometri Analysis of Mathed Firms ; Review of International Eonomis 12, Goldberg, Linda, and M. Knetter (1997), Goods Pries and Exhange Rates: What Have We Learned? ; Journal of Eonomi Literature 35, and J. M. Cama (2010), The Sensitivity of the CPI to Exhange Rates: Distribution Margins, Imorted Inuts, and Trade Exosure; Review of Eonomis and Statistis 92, Goinath, Gita, and R. Rigobon (2008), Stiky Borders ; Quarterly Journal of Eonomis 123, Haddad, M. (1993), How Trade Liberalization Affeted Produtivity in Moroo ; World Bank Poliy Researh Working Paer 1096; Washington, DC: The World Bank. Hausmann, Riardo; L. Prithett and D. Rodrik (2005), Growth Aelerations ; Journal of Eonomi Growth 10, Helman, Elhanan, M.J. Melitz and S.R. Yeale (2004), Exort versus FDI with heterogeneous firms, Amerian Eonomi Review 94, Imbs, Jean, and R. Waziarg (2003), Stages of Diversifiation, Amerian Eonomi Review 93, Mayer, Thierry, and J. Ottaviano (2011), Market Size, Cometition, and the Produt Mix of Exorters ; CEPR Disussion Paer 8349; London, UK: CEPR. Obstfeld, Martin, and K. Rogoff (2001), The Six Major Puzzles in International Maroeonomis: Is There a Common Cause? ; NBER Maroeonomis Annual , Ottaviano, Gianmaro; T. Tabuhi and J.-F. Thisse (2002), Agglomeration and Trade Revisited ; International Eonomi Review 43, and J. Melitz (2008), Market Size, Trade, and Produtivity ; Review of Eonomi Studies 75, Rodrik, Dani (2008), The Real Exhange Rate and Eonomi Growth ; Brookings Paers on Eonomi Ativity, Fall 2008, Shott, Peter (2004), Aross-rodut versus Within-rodut Seialization in International Trade, Quarterly Journal of Eonomis 119, Tybout, J. and D. Westbrook (1995), Trade Liberalization and the Dimensions of Effiieny Change in Mexian Manufaturing Industries, Journal of International Eonomis 39,

21 Van Biesebroek, J. (2005), Exorting raises rodutivity in sub-saharan Afrian manufaturing firms ; Journal of International Eonomis 67,

22 Deendent var.: ln (Unit Value) Estimator: OLS Table 5 Baseline regression results, unit values, all samle (1) (2) (3) (4) (5) (6) (7) (8) (9) (10) (11) (12) Log bilateral RER 0.108*** ** 1.622*** *** *** 0.137*** ** (0.0316) (0.0332) (0.369) (0.127) (0.212) (0.0692) (0.0317) (0.0303) (0.0309) (0.390) (0.370) (0.352) Interation terms ln (RER) deval. a/ ( ) ( ) ( ) ( ) ln (RER) ln (dist.) *** (0.0439) (0.0430) (0.0434) (0.0397) ln (RER) ln (dest. GDP/a) * (0.0128) (0.0252) (0.0238) (0.0237) ln (RER) ln (dest. GDP) * ( ) (0.0144) (0.0133) (0.0131) ln (RER) manuf. Prod *** 0.301*** ** * (0.0777) (0.0707) (0.0572) (0.0568) ln (RER) ln (1+number rod.) b/ *** *** ( ) ( ) ln (RER) ln (lag number rod.) b *** ** ** ( ) ( ) ( ) ln (RER) EAC bilateral trade / 0.692*** 0.341** 0.525*** (0.153) (0.164) (0.179) Devaluation (Real) *** ** ( ) ( ) ( ) ( ) ln (dest. GDP/a) *** 0.546*** 0.476*** 0.515*** (0.0480) (0.0999) (0.104) (0.103) ln (dest. GDP) *** *** *** *** (0.0476) (0.0897) (0.0921) (0.0912) ln (1+number rod.) ( ) ( ) ln (lag number rod.) ( ) ( ) ( ) Observations 568, , , , , , , , , , , ,556 R-squared Firm-rodut-destination FE Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Origin--year FE Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes 21

23 Deendent var.: ln (Volume) Estimator: OLS Table 6 Baseline regression results, volumes, all samle (1) (2) (3) (4) (5) (6) (7) (8) (9) (10) (11) (12) Log bilateral RER 0.403*** 0.514*** *** 3.094*** *** 0.469*** 0.438*** 3.629*** 3.035*** 2.324*** (0.0655) (0.0710) (0.589) (0.276) (0.441) (0.123) (0.0658) (0.0749) (0.0666) (0.811) (0.866) (0.789) Interation terms ln (RER) deval. a/ ( ) ( ) ( ) ( ) ln (RER) ln (dist.) ** (0.0699) (0.0917) (0.102) (0.0840) ln (RER) ln (dest. GDP/a) *** (0.0274) (0.0530) (0.0550) (0.0549) ln (RER) ln (dest. GDP) *** *** *** *** (0.0163) (0.0316) (0.0327) (0.0320) ln (RER) manuf. Prod *** 0.682*** 0.674*** 0.652*** (0.133) (0.134) (0.142) (0.141) ln (RER) ln (1+number rod.) b/ ( ) ( ) ln (RER) ln (lag number rod.) b *** ( ) ( ) ( ) ln (RER) EAC bilateral trade / *** *** * (0.227) (0.291) (0.360) Devaluation (Real) *** *** *** *** (0.0106) (0.0107) (0.0108) (0.0108) ln (dest. GDP/a) 1.015*** *** ** *** (0.113) (0.230) (0.250) (0.250) ln (dest. GDP) 1.024*** 1.544*** 1.687*** 1.733*** (0.100) (0.199) (0.216) (0.215) ln (1+number rod.) 0.250*** 0.244*** (0.0129) (0.0128) ln (lag number rod.) *** *** *** (0.0122) (0.0122) (0.0122) Observations 568, , , , , , , , , , , ,558 R-squared Firm-rodut-destination FE Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Origin--year FE Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes 22

24 Deendent var.: ln (Unit Value) Estimator: OLS Table 9 Baseline regression results, unit values, EAC ountries (1) (2) (3) (4) (5) (6) (7) (8) (9) (10) (11) (12) Log bilateral RER *** *** *** (0.106) (0.110) (0.670) (0.334) (0.596) (0.142) (0.107) (0.0926) (0.115) (1.340) (1.087) (0.929) Interation terms ln (RER) deval. a/ * ( ) ( ) ( ) ( ) ln (RER) ln (dist.) *** ** (0.0851) (0.161) (0.140) (0.0910) ln (RER) ln (dest. GDP/a) (0.0358) (0.0815) (0.0676) (0.0676) ln (RER) ln (dest. GDP) (0.0227) (0.0505) (0.0420) (0.0401) ln (RER) manuf. Prod *** 0.645*** (0.178) (0.177) (0.162) (0.161) ln (RER) ln (1+number rod.) b/ ** ( ) ( ) ln (RER) ln (lag number rod.) b ( ) ( ) ( ) ln (RER) EAC bilateral trade / 0.862*** 0.725** 0.888*** (0.188) (0.327) (0.312) Devaluation (Real) (0.0127) (0.0127) (0.0121) (0.0121) ln (dest. GDP/a) *** 0.624* (0.166) (0.361) (0.290) (0.290) ln (dest. GDP) *** *** ** ** (0.148) (0.295) (0.248) (0.248) ln (1+number rod.) (0.0161) (0.0161) ln (lag number rod.) (0.0132) (0.0133) (0.0133) Observations 145, , , , , , , , , , , ,189 R-squared Firm-rodut-destination FE Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Origin--year FE Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes 23

25 Deendent var.: ln (Volume) Estimator: OLS Table 8 Baseline regression results, volumes, EAC ountries (1) (2) (3) (4) (5) (6) (7) (8) (9) (10) (11) (12) Log bilateral RER *** 3.214*** *** * (0.177) (0.183) (0.937) (0.564) (0.993) (0.225) (0.178) (0.177) (0.188) (1.949) (1.850) (1.594) Interation terms ln (RER) deval. a/ *** *** *** *** ( ) ( ) ( ) ( ) ln (RER) ln (dist.) *** 0.694*** (0.118) (0.250) (0.246) (0.157) ln (RER) ln (dest. GDP/a) *** ** (0.0601) (0.127) (0.120) (0.120) ln (RER) ln (dest. GDP) *** (0.0377) (0.0802) (0.0756) (0.0714) ln (RER) manuf. Prod *** 1.003*** 1.159*** 1.160*** (0.257) (0.280) (0.291) (0.290) ln (RER) ln (1+number rod.) b/ ( ) ( ) ln (RER) ln (lag number rod.) b ( ) ( ) ( ) ln (RER) EAC bilateral trade / (0.276) (0.540) (0.581) Devaluation (Real) *** *** *** *** (0.0225) (0.0226) (0.0226) (0.0226) ln (dest. GDP/a) 0.807*** (0.301) (0.655) (0.651) (0.650) ln (dest. GDP) 0.407* (0.243) (0.540) (0.538) (0.538) ln (1+number rod.) 0.168*** 0.169*** (0.0258) (0.0258) ln (lag number rod.) * (0.0216) (0.0219) (0.0219) Observations 145, , , , , , , , , , , ,189 R-squared Firm-rodut-destination FE Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Origin-year FE Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes 24