The Contractionary Short-Run Effects of Nominal Devaluation in Developing Countries: Some Neglected Nuances

Transcription

1 The Contractonary Short-Run Effects of Nomnal Devaluaton n Developng Countres: Some Neglected Nuances by Arslan Razm Unversty of Massachusetts, Amherst Revsed, June 2006 Abstract Ths paper extends the model developed by Krugman and Taylor (1978) to take nto account nterestng features of the evolvng structure of global trade. The growng presence of transnatonal producton chans and dfferental prcng behavour of exports destned for ndustral and developng countres are accommodated. Indvdual country and panel data pass-through estmates derved from several econometrc approaches are provded to justfy the latter extenson. The lkelhood of contractonary short-run effects of devaluatons s shown to be postvely related to: (1) the proporton of a country's exports destned for other developng countres, and (2) the presence of TNCs n ether the export or home goods-producng sector. Unlke the Krugman-Taylor case, devaluaton wll generally have a contractonary mpact even f: (1) trade s ntally balanced, (2) consumpton behavour does not dffer between wage and proft earners, and (3) the government sector has a hgh margnal propensty to consume n the shortrun. The resultng polcy mplcatons underlne the need to take nto account these ncreasngly mportant nuances of nternatonal trade whle desgnng exchange rate polces for developng countres. JEL Codes: F14, F41, F23, O24 Keywords: Dfferental pass-through elastctes, contractonary devaluatons, transnatonal corporatons, error correcton models, autoregressve dstrbuted lag models, structuralst models. Correspondence Address: Arslan Razm, Department of Economcs, Unversty of Massachusetts, Amherst, MA 01003, Tel: (413) , Fax: (413) , E-mal: arazm@econs.umass.edu.

2 1. Introducton The relatonshp between nomnal exchange rate changes and output n developng countres has been debated at length amongst academc and polcy-makng crcles. Krugman & Taylor (1978) developed a coherent case for possble contractonary consequences of nomnal devaluatons. The frst part of ths paper extends ther orgnal model to accommodate some of the developments that have charactersed recent decades whle mantanng the concson of the orgnal framework. In partcular, the extent of exchange rate pass-through s hypothessed to be greater for exports to developng countres, and the tendency to consume domestc goods out of profts s hypothessed to be lower for transnatonal corporatons (TNCs) compared to domestc proft earners. The former hypothess requres more careful justfcaton n the form of emprcal support. The second part of ths paper, therefore, econometrcally tests for the exstence of sgnfcant dfferences n pass-through behavour dependng on whether the exported goods are destned for developng or ndustralsed countres. Based on these assumptons about TNC behavour and pass-through dfferences between exports to developng and ndustral countres, we show that the lkelhood of contractonary short-run consequences of nomnal devaluaton vares drectly wth: (1) the proporton of a country's exports destned for other developng countres, and (2) the presence of TNCs n ether the export or home goods-producng sector. Moreover, devaluaton s lkely to have a contractonary mpact even f: (1) trade s ntally balanced, (2) consumpton behavour does not dffer between proft and wage earners, and (3) the government sector has a hgh margnal propensty to consume n the shortrun. We fnd emprcal support for the exstence of systematc dfferences, on average, n pass-through behavour between developng and ndustralsed countres. Ths suggestve

3 fndng, whch s robust to several alternatve emprcal specfcatons and measures, underlnes the need for polcy-makers to consder related features of nternatonal trade whle determnng ther exchange rate polces. We make several contrbutons to exstng lterature. As mentoned earler, the Krugman-Taylor framework s extended to take nto account some nterestng features of the global economy. The degree of exchange rate pass-through s estmated for a large sample of (24 developng and 19 ndustralsed) countres for the perod Dfferences n pass-through behavour between developng and ndustral countres are nvestgated. Whle a substantal body of lterature has emerged over the years that estmates pass-through behavour n ndustral countres, developng countres have receved much less attenton. Even less attenton has been pad to the exploraton of systematc dfferences n pass-through behavour between developng and ndustral countres, and to pass-through nto mport prces rather than nto consumer prces. Both ndvdual country and panel data estmates are provded. Fnally, we focus on outputrelated aspects of exchange rate pass-through as opposed to many recent studes whch tend to emphasse monetary polcy- and nflaton-related ssues. 2. Background and Model Set-Up A substantal body of lterature has emerged on the possble contractonary effects of nomnal devaluatons n developng countres snce the poneerng work of rschman (1949) and Daz-Alejandro (1963). v Krugman and Taylor (1978) provded an elegant yet smple framework to analyze the potental short-run effects of nomnal devaluatons. In ther framework, a devaluaton can lead to short-run contracton through three channels: () the greater valuaton effects of a devaluaton on mports n the presence of a trade

4 defct and when measured n terms of the domestc currency, v () a redstrbuton of ncome towards proft earners who have a hgher propensty to save than wage earners, and () a redstrbuton of revenues from the prvate sector to the government sector whch, gven a fxed level of government spendng, reduces demand for the home good. Ths paper assumes no substtuton effects n consumpton or producton, whch may be justfed by the presence of J-curve effects and the short-run nature of the model. Moreover, the paper does not dstngush between developng country exports to ndustralsed countres and those to other developng countres. Fnally, the paper does not take nto account behavoural dfferences between domestc proft earners and TNCs. Our paper ntroduces some of these features to the Krugman-Taylor model n the lght of recent developments on the global scene, whle mantanng the concse elegance of the orgnal framework. The perod snce the debt crss of the early 1980s has seen a marked transformaton n the export structure of developng economes. Most strkngly, manufactured products now consttute more than 70 percent of total exports from these countres. A large proporton of these exports conssts of relatvely low value-added, labour-ntensve goods for whch nternatonal elastctes of substtuton are lkely to be relatvely hgh, at least n the medum to long run. Furthermore, some developng countres have seen the share of ther exports gong to other developng countres rse over ths perod. Wth ther transformaton nto exporters of manufactures, many of these developng countres are lkely to have acqured some ablty to prce dscrmnate, allowng for varyng degrees of pass-through across destnaton markets. Another major development over the last few decades has been the rsng presence of TNCs n both the tradable and non-tradable, but especally the exportable goods sectors of developng economes, a case n pont beng the accelerated establshment of export

5 processng zones (EPZs) to attract FDI and TNCs. Indeed, accordng to Chang (1998), TNCs now manage about 75 per cent of world trade n manufactured goods. Moreover, accordng to UNCTAD (2002), the shares of foregn afflates n total exports have grown to more than 50 percent n Chna and 80 percent n ungary. Consderng that the consumpton patterns of the owners of ths nternatonal captal are not expected to drectly nvolve domestc (host country) goods n any substantal sense, ths sgnfcant ncrease n the presence of TNCs n nternatonal producton chans calls for a more nuanced analyss of the short-run effects of nomnal devaluatons. In order to focus on these new developments whle keepng the analyss smple, we retan the followng assumptons from the orgnal Krugman-Taylor model: 1. There are two dstnct sectors; one produces the (non-tradable) home good for domestc markets whle the other produces the export good for nternatonal markets. The producton of the former requres labour and mported ntermedate nputs whle the latter requres labour only. Both sectors use fxed coeffcent technologes. 2. The prce of the home good s determned by a mark-up over drect nput costs, whle that of the mported nput s fxed n terms of the nternatonal currency. 3. The nomnal wage rate s constant n terms of the domestc currency. 4. Interest rates are kept constant by acton of the monetary authortes. We extend the Krugman-Taylor model partly by assumng dfferental pass-through behavour between exports to developng and ndustral countres. There are several plausble reasons for expectng prcng behavour to dffer (as assumed n our theoretcal model below) dependng on the destnaton of a developng country's exports. Passthrough s lkely to be postvely correlated wth the degree of substtutablty between a country's exports and the mport-substtutes produced by the tradng partner. v To the extent that developng country exports are closer substtutes for other developng country

6 products, v pass-through nto exports to other developng countres s lkely, therefore, to be hgher than that to ndustral countres. Pass-through may be hgher f the exporters are numerous relatve to the domestc compettors. v As Campa & Goldberg (2005) pont out, one mplcaton s that pass-through elastctes are nversely related to country real GDP. Ths tes n, to some extent, to the market share' hypothess of frm behavour under prcng to market. If developng country exporters are more nterested n mantanng market share n ndustral countres than n other developng countres, then they wll be less lkely to pass through the effects of currency changes to ndustral country consumers than to developng country consumers. The much larger economc sze of ndustral country markets could motvate such a dstncton. x Taylor (2000) explores the lnk between pass-through to aggregate prces and nflaton wthn a framework of staggered prcng and monopolstc competton, argung that regmes wth hgher nflaton tend to have more persstent cost changes, and thus, hgher exchange rate pass-through. x Devereux & Engel (2001) argue, on a related note, that countres that have a hghly volatle monetary polcy x wll fnd ther mport prces beng set n foregn currency, and wll therefore, experence a hgh rate of pass-through nto mported prces. To the extent that developng countres have hgher nflaton and less stable monetary polces, on average, ther mport prces would, therefore, be expected to experence hgher pass-through. x Fnally, many developng countres manly export textles and related products, whch untl recently were (and to some extent stll are) subject to mport quotas and non-tarff barrers n ndustralsed countres under the mult-fbre agreement. In the presence of such restrants, a small deprecaton s lkely to be absorbed by exporters n ther proft margns rather than beng passed through nto mport prces. x Indeed, an assumed hgher pass-through nto mport prces s one of the factors mplctly underlyng the frequently heard argument that the small open economy model

7 s more applcable to developng countres. In order to accommodate ths dfference n pass-through behavour, we assume that whle the nternatonal currency prce of exports to ndustralsed country markets s mantaned followng nomnal exchange rate changes, that of exports to other developng country markets s not because domestc exporters pass the effects of such changes on to the buyers. In other words, whle exporters to ndustral countres engage n local currency prcng (LCP), those to developng countres engage n producer currency prcng (PCP) when faced wth exchange rate changes. Note that such a set-up mples, ceters parbus, that whle a devaluaton vs-à-vs other developng countres has a negatve mpact on the terms of trade, that vs-à-vs ndustralsed countres does not. Mathematcally: P = ( 1 t ) P (1) D P ) I = e(1 t P (2) where P j, j = I, D, denote the prces (measured n domestc currency) receved by domestc producers of exports to ndustral and developng countres, respectvely, e denotes the nomnal exchange rate (unts of domestc currency per unt of nternatonal currency), t s the ad-valorem tax rate on exports, and P and P are the domestc and nternatonal prces of the export good, respectvely. xv The prce of the home good, P, can be expressed as: P = 1+ τ )( wa + ep a ) (3) ( M M where τ s the mark-up factor, w s the nomnal wage, a k, k =, M, are the fxed ntermedate nput coeffcents for labour and the mported ntermedate nputs, respectvely, and P M s the nternatonal prce of the mported ntermedate nput. The nomnal ncomes of wage and proft earners, denoted by W and R respectvely, can be

8 expressed as follows: W wa wa + wa I D = + (4) R = wa ep a P wa ep wa ) D I τ ( + M M ) + ( ) + ( (5) where k, k = I, D, denotes the volumes of exports to ndustral and developng countres, respectvely, s the output of the home good, whle a s the fxed labour coeffcent n the export sector. Snce all mports are assumed to be nputs nto home good producton, the latter beng the only good whch s consumed domestcally, xv the rest of the set-up follows Krugman & Taylor (1978). Thus, the zero excess demand condtons for the home and mport goods can be expressed respectvely as: W R ED = CW CR G P + P + (6) EDM = a M M (7) where G denotes nomnal government spendng. To avod clutter, we denote C ( W / P ) and C ( R / P ) W / R / by γ W and γ R, respectvely. 3. Income Effects of Devaluaton Reconsdered Followng Krugman & Taylor (1978), we smplfy by ndvdually consderng varous channels through whch the ncome effects of a devaluaton are propagated Devaluaton from an Intal Trade Balance Consder the specal case where γ W = γ R = γ,.e., the consumpton patterns of proft and wage earners are dentcal functons of real ncome. Further, let us abstract away for now from fscal effects by assumng that t = t M = G = 0. Consderable manpulaton

9 leads to the followng expresson for the elastcty of home goods output wth respect to the exchange rate: d de e = γ [(1 Λ)( ep P I ep M ) ΛP M D ] (8) where Λ <1 denotes the share of mported ntermedate costs n total nput costs, and P +τ wa τ = 1 γ W γ R 1 > 0. Notce that equaton (8) s dentcal to the expresson derved by Krugman & Taylor (1978) except for the addtonal term on the rght hand sde that results from dstngushng between exports to ndustral and developng countres. The lessons that emerge regardng the consequences of a nomnal devaluaton are, however, qualtatvely dfferent. Whle the effect transmtted from the ndustral country markets s expansonary, that orgnatng from developng country markets s contractonary. The overall effect of a devaluaton s contractonary barrng the unlkely scenaro where exports to ndustralsed countres are much greater than total mports and ntermedate nput costs as a proporton of total costs are low. Thus, what devaluaton takes wth one hand by rasng mport prces, t does not entrely gve back wth the other, even f trade s ntally balanced. xv Consderng that the typcal developng country sells a sgnfcant (and rsng) proporton of ts exports to other developng countres, the overall effect of a devaluaton s very lkely to be negatve. Ths result s made even more probable f we note that usually countres devalue when experencng trade defcts. Next, suppose that TNCs completely domnate the export sector, so that profts derved from exports are not spent on consumng the home good. Then: d de e γw = Λ < 0 P (9) A devaluaton s unambguously contractonary n ths case. The ncrease n mport costs s not offset by the ncrease n nomnal profts rrespectve of the ntal trade

10 balance, although nomnal ncome s redstrbuted from wages to profts. Snce the same analyss holds n all three cases when t s the export sector that the TNCs domnate, we do not analyze the dstrbutonal and fscal effects of a devaluaton separately n sectons 3.2 and 3.3. If, on the other hand, TNCs completely domnate the home goods sector then, perhaps paradoxcally, the effect of a devaluaton becomes ambguous, and depends on: () the proporton of exports destned for each group of countres, () the mark-up rate, and () ntermedate nput costs. If exports to ndustral countres are the major source of overall demand, ntermedate nputs are a relatvely small proporton of total costs, and the country starts wth a trade surplus, then a devaluaton could be expansonary. Mathematcally: d de e = γ [(1 Λ) ep P I (1 Λ τ ) ep M M ΛP D ] (10a) If, however, trade s ntally balanced, xv a devaluaton unambguously leads to a fall n output. Mathematcally, equaton (10a) reduces to: d de e γ = [ τep P M M + P D ] (10b) A comparson of equatons (8) and (10a) llustrates the underlyng ntuton. Wth only TNCs present n the home goods-producng sector, the ncrease n nomnal profts due to a constant mark-up over costs leaks abroad nstead of stmulatng domestc demand Dstrbutonal Effects Now suppose the devalung country starts wth balanced trade but that there s a greater tendency to save out of profts than wages. Ignorng fscal effects, the followng

11 expresson can be derved for the exchange rate elastcty of home goods output: d de e WΛ( γ R γ W ) γ RP = P D < 0 (11) As n Krugman & Taylor (1978), by shftng purchasng power from wages to profts, a devaluaton serves to lower overall demand for the home good. owever, the contracton follows even f consumpton behavour does not dffer between the two groups. The contractonary effect s even more pronounced f TNCs are present n ether or both of the two sectors. In the case where TNCs completely domnate the home goods sector, and agan assumng balanced trade as the pont of departure: d de e = Λ D [( γ R γ W ) W γ R ( τepm M + P )] (12) P The extent to whch the contractonary effect s magnfed by the presence of TNCs n the home goods sector vares drectly wth the (mported) nput ntensty of home goods' producton Fscal Effects Assumng that the government sets ts expendtures and tax rates at the begnnng of the perod, and further assumng a balanced budget, yelds the followng expresson: d de e γ I D = [ t xep + P ] < 0 P (13) where we assume balanced trade and γ W = γ R = γ. A devaluaton shfts ncome from the prvate sector to the government, whch has a margnal propensty to consume of zero. The effect on output complements the contractonary effect emanatng from lower profts from sales to developng countres. The presence of TNCs n the home goods sector does not affect the analyss snce that sector s not subject to export taxes.

12 In summary, complete (or hgher) exchange rate pass-through nto prces of exports destned for other developng countres ncreases the lkelhood of output contracton followng a nomnal devaluaton. The presence of TNCs n ether sector generally further enhances such a lkelhood. 4. Nomnal Devaluaton and Prce Adjustment A number of studes have explored the relatonshp between exchange rate changes and the prce of tradables. Few studes, however, have focused on developng countres. Even fewer have systematcally nvestgated dfferences n pass-through behavour between ndustral and developng countres. As dscussed n secton 2, there are several plausble reasons for expectng prcng behavour to dffer dependng on the destnaton of a developng country's exports. Furthermore, assumng relatvely greater pass-through nto export prces for goods destned for other developng countres, t follows from the theoretcal model developed n sectons 2 and 3 that the contractonary short-run effect of a devaluaton vares drectly wth the share of such exports n a country's total exports. The next two sectons emprcally explore pass-through behavour for exports to ndustralsed and developng countres usng several dfferent ndvdual country and panel data approaches. We fnd support for the hypothess that exchange rate passthrough s markedly hgher, on average, for exports to developng countres. 5. Emprcal Models and Sample Our emprcal strategy conssts of two stages. In the frst stage, we seek to nvestgate the possble exstence of systematc dfferences between the pass-through behavour of exports destned for ndvdual developng and ndustral countres. In the second stage,

13 we pool the data nto two groups; one for developng countres and the other for ndustral countres. The fndngs are consstent wth those orgnatng from the ndvdual country studes. Ideally, we would lke to have mport prces for transactons among pars of countres n order to see the correspondence between changes n these and the blateral nomnal exchange rates. owever, such data are not readly avalable for most developng countres. An alternatve s to study correlatons between nomnal exchange rate and export prce changes for each country. owever, ths approach would not yeld any nformaton regardng dfferences between pass-through nto developng versus ndustralsed country mport prces. Yet another alternatve s to estmate pass-through nto mport prces. Although ths approach does not dstngush between mports comng from developng and ndustral countres, t helps address the ams of our study more drectly. We therefore, use the latter approach to nvestgate systematc dfferences n pass-through behavour for mports nto countres at dfferent levels of ncome. Our sample conssts of 24 developng and 19 ndustral countres. xv All the countres for whch mport prce or unt mport value data were avalable from the Internatonal Monetary Fund's Internatonal Fnancal Statstcs (IFS) are ncluded. xx The sample spans the perod The choce of the start year s desgned to exclude the debt crss of We explored the tme seres propertes of our varables, wth the help of the Augmented Dckey-Fuller (ADF) test. The results, whch are reported n Table 1, ndcate that an overwhelmng number of ndvdual country seres are non-statonary, and ntegrated of order one,.e., I(1), at the tradtonal levels of statstcal sgnfcance. xx In carryng out our analyss, we therefore, pursued three approaches: (1) an autoregressve dstrbuted lag (ADL) model, (2) an error correcton approach wth

14 unknown long-run (contegratng) coeffcents, and (3) an error correcton approach that assumes complete pass-through n the long run. Whle we only report the results from the frst and second approaches, those from the thrd approach were vrtually dentcal, and are avalable on request. The ADL model can be expressed as follows: P n n n n M, t = 0 + β, p E, t p + β 2, p P, t p + β q P 3, M, t q + ε 1, t p p= 0 p= 0 q= 1 p= 0 β (14) where, for the th country, P M s the mport prce (n the mporter's currency), E s the nomnal exchange rate (mporter's currency per US dollar), whle P denotes the exportng country's unt producton costs (n terms of US dollars). All varables are n logs and the seres used have an annual frequency. The specfcaton n frst dfferences reflects our focus on short-run pass-through behavour and ts consequences for the shortrun relatonshp between devaluatons and output changes. xx The reason for the ncluson of the nomnal exchange rate on the rght hand sde of equaton (14) s obvous. A bref explanaton for the other explanatory varables s n order. An ncrease n a country's mport prce transmtted from relatve prce changes abroad may be due ether to changes n the blateral nomnal exchange rate or to changes n the cost structure abroad. xx The ncluson of the foregn prce level (world CPI as a proxy for world producton costs) controls for the latter, allowng us to solate the effects of nomnal exchange rate changes. Fnally, although we use varables n ther frst dfferences, the possblty of persstence or nertal effects of external shocks suggests the addton of lagged dfferenced values of the dependent varable. As a robustness check, we re-estmate exchange rate passthrough elastctes wth the domestc prce level added as an explanatory varable. xx As a further robustness check, we re-estmate usng the Internatonal Monetary Fund's Specal Drawng Rghts' (SDR) as the unt of value nstead of the dollar. The results, whch are reported n the next secton, are found to be qualtatvely robust.

15 Whle the ADL approach s useful for our purposes gven the length of our tme seres, t does not explctly take nto account the possble exstence of long-run contegratng relatonshps between the levels of the regressors. We therefore, derved another set of estmates after embeddng a hypothessed long-run relatonshp between the levels of the varables n an error correcton framework n order to nvestgate short-run reverson to equlbrum. The specfcaton can be expressed n ts general form as follows: t t t t p t p n p p t p n p t M P E P E P, 1, 5 1, 4 1, 3,, 2 0,, 1 0 0, ε ε θ θ θ θ θ θ = = = (15) where captures the hypothessed long-run relatonshp between the prce of mports, the nomnal exchange rate, and nternatonal producton costs, whle = 1, 1, 1, 1, t t t M t P E P ε 5 θ s the error correcton coeffcent. xxv Fnally, we apply another verson of the error correcton approach whch assumes homogenety n exchange rates and prces, or n other words, unt long-run contegratng coeffcents: t t p t p n p p t p n p t M P E P ε ε φ φ φ φ = = = 1, 3,, 2 0,, 1 0 0, (16) where 3 φ s the error correcton coeffcent. Ths approach mplctly assumes long-run reverson towards purchasng power party, or alternatvely, towards a fundamental real exchange rate. The results, although not reported here, are avalable from the author on request. << INSERT TABLE 1 ABOUT ERE >> 6. Econometrc Results Ths secton dscusses the results of our econometrc analyss of ndvdual country and panel data for 43 countres.

16 6.1. Indvdual Country Estmates The ADL Approach Table 2 presents estmates for 24 developng countres. Consderng the relatvely short length of our tme seres, the ndvdual country model s specfed n contemporaneous frst dfferences throughout. The pass-through estmates have the expected (postve) sgn and magntude (between 0 and 1) for most countres. owever, the magntudes of the pass-through coeffcent for Chna, ong Kong, South Korea, and Sr Lanka are much greater than one. Moreover, the pass-through coeffcent for Sngapore s negatve. Fnally, Inda and Pakstan report negatve adjusted coeffcents of determnaton. These countres are, therefore, excluded whle reportng averages. Of the remanng 17 countres, 11 report pass-through estmates of greater than 0.5, ndcatng that exporters to these countres pass through most of the mpact of nomnal devaluatons wthn a year. The coeffcent (for estmates that are sgnfcant at the 5 percent level) ranges from for South Afrca to for Argentna. The coeffcent on the world prce level varable s generally much lower n magntude suggestng that most of the ncrease n mport prces orgnates n exchange rate changes rather than n changes n global producton costs. The coeffcent of the lagged mport prce term s generally not sgnfcant at the conventonal levels, ndcatng weak nertal effects. xxv The adjusted coeffcents of determnaton are generally hgh enough to nspre confdence n the results. Table 3 presents the results for 18 ndustralsed countres. Indvdual country estmates could not be derved for the US for obvous reasons. xxv The pass-through estmates have the expected sgn and magntude for almost all the countres. owever, the coeffcent for New Zealand s much greater than one. The estmated pass-through

17 elastcty for Norway s negatve. These two countres are, therefore, excluded whle reportng averages. Of the remanng 16 ndustral countres, only three report passthrough estmates of greater than 0.5, ndcatng that exporters to these countres mostly accommodate nomnal exchange rate changes n ther margns, at least n the short run. The coeffcents (for estmates that are sgnfcant at the 5 percent level) range from for Denmark to for Japan. Agan, the coeffcent of the world prce level varable s generally much lower n magntude suggestng that most of the ncrease n mport prces orgnates n nomnal exchange rate changes rather than n changes n global producton costs. For most countres, the coeffcent of the lagged mport prce term s not statstcally sgnfcant at the conventonal levels, ndcatng weak nertal effects. As noted above, a much greater proporton of developng countres report a passthrough elastcty of greater than 0.5 n magntude. Further comparson of the two sets of estmates for ndustral and developng countres ndcates that the mean value of the nomnal exchange rate coeffcent s markedly greater for the group of developng countres. Excludng the seven developng countres whch ether report negatve nomnal exchange rate pass-through elastctes, yeld coeffcents that are greater than 1.25 n absolute value, or report negatve adjusted coeffcents of determnaton, yelds averages of for the exchange rate coeffcent and for the world prce coeffcent for the 17 remanng developng countres. xxv Excludng ndvdual ndustral countres on the same grounds yelds averages of for the exchange rate coeffcent and for the world prce coeffcent for the 16 remanng ndustral countres. xxv Fnally, further excludng countres that reported exchange pass-through coeffcents that are not sgnfcant at the 5 percent level yelds averages of and for developng countres and and for ndustral countres, respectvely. xxx Thus, exchange rate pass-through elastctes are much hgher, on average, for developng

18 countres. The correlaton coeffcent between the estmated pass-through elastctes and the GDP per capta (averaged over the sample perod) of ndvdual countres s -0.25, provdng further support to our hypothess. << INSERT TABLES 2 and 3 ABOUT ERE >> Dagnostc tests (not reported here but avalable on request) ncludng the Jarque-Bera statstcs, the Breusch-Godfrey LM tests for seral correlaton n the resduals, and LM tests for Autoregressve Condtonal eteroscedastcty (ARC) n the resduals, xxx do not ndcate major problems at the 5 per cent level of sgnfcance. Only the resduals from the New Zealand and Venezuela equatons ndcate seral correlaton. Furthermore, estmates for Inda and Poland ndcate autoregressve condtonal heteroscedastcty. The resduals for ungary, Poland, and Fnland ndcate volaton of the normalty assumpton. Reassurngly, CUSUM test plots do not suggest parameter nstablty wth the excepton of Inda. xxx Controllng for domestc competton: Some prevous studes have controlled for the effects of competton from domestc producers of mport-substtutes n the mportng countres by ncludng a domestc producer prce ndex varable as a regressor. owever, such data are not avalable for many developng countres n our sample. We therefore, carred out a robustness check by addng the domestc consumer prce ndex for each mportng country as a regressor nstead. The addton of the addtonal control varable does not qualtatvely change the results. owever, consderng that we use annual data, the domestc CPI s lkely to be hghly correlated wth the nomnal exchange rate. We therefore, only report results for regressons run wthout ths control varable. xxx Usng SDRs nstead of US dollars: One potental drawback of usng the dollar as the benchmark' exchange rate for both nomnal exchange rates and the world prce level s

19 that the varaton n pass-through estmates could be a functon of the share of an ndvdual country's mports from the US. For example, whle Mexco purchased 64 percent of ts mports from the US n 2003, the correspondng statstc for Poland was less than 2 percent. xxx owever, to the extent that a large proporton of nternatonal trade that does not nvolve the US s nevertheless denomnated n US dollars, these statstcs radcally understate the use of the dollar n nternatonal trade. That the dollar remans the vehcle currency of choce for nternatonal transactons (especally for developng country traders), provdes justfcaton for ts use as the benchmark. owever, to further test the robustness of our concluson that nomnal exchange rate pass-through s hgher on average for developng countres, we re-estmated ndvdual country pass-through equatons usng a weghted nomnal exchange rate ndex, namely the Internatonal Monetary Fund's reserve asset known as the SDR. xxxv The SDR s based on a basket consstng of the fve key nternatonal currences, the US dollar, the pound sterlng, the yen, the Deutsche mark, and the French franc. xxxv The average estmated exchange rate pass-through coeffcents were somewhat close to the earler dollar estmates. Furthermore, exchange rate pass-through elastctes were hgher, on average, for developng countres, although the dfference between the two groups of countres was somewhat smaller than n the other sets of estmates. The correlaton coeffcent between the estmated pass-through elastctes and the GDP per capta (averaged over the sample perod) of ndvdual countres was xxxv The Error Correcton Approach Tables 4 and 5 present estmates for 24 developng and 18 ndustral countres, respectvely. xxxv The short-run pass-through estmates have the expected sgn and magntude for most countres. The coeffcent of the lagged error correcton term s

20 correctly sgned n all cases, ndcatng reverson to equlbrum over tme, although t s not statstcally sgnfcant n many nstances. xxxv The latter observaton s not surprsng gven the relatvely short length of our tme seres. Agan, a much greater proporton of developng countres reports a pass-through coeffcent of greater than 0.5 n magntude. Excludng countres whch ether report negatve or nsgnfcant (at the 5 percent level) nomnal exchange rate pass-through elastctes, yeld coeffcents that are greater than 1.25 n absolute value, or report negatve adjusted coeffcents of determnaton, yelds averages of and for developng and ndustral countres, respectvely. Thus, the results are qute smlar to those yelded by the ADL approach. In partcular, exchange rate pass-through coeffcents are much hgher, on average, for developng countres. The correlaton coeffcent between the estmated pass-through coeffcents and the GDP per capta of ndvdual countres (averaged over the sample perod) was Usng SDRs nstead of US dollars: As another robustness test, we pursued the error correcton approach to derve regresson results wth varables defned n terms of the SDR. Agan, estmated exchange rate pass-through coeffcents were much hgher, on average, for developng countres. Also, the error correcton terms were correctly sgned, although n many cases, statstcally nsgnfcant. The correlaton coeffcent between the estmated pass-through coeffcents and the GDP per capta (averaged over the sample perod) of ndvdual countres was xxxx << INSERT TABLES 4 and 5 ABOUT ERE >> 6.2. Panel Data Estmates In order to ncrease the number of degrees of freedom avalable, and to nvestgate

21 possble lagged effects, we next pool the data nto panels, one each for developng and ndustralsed countres. In addton to the above-mentoned benefts, poolng generally also reduces collnearty among the regressors, thus ncreasng the effcency of the estmates (sao, 2003). An obvous drawback of poolng s the mposton of the rather strngent condton that the coeffcents be the same across members. We used a Chow test of the jont restrctons ( 0 : β 1 = β1 β 2 = β 2 β 3 = β3 ) n equaton (14) n order to nvestgate whether the data for the developng and ndustralsed countres n our sample can be pooled together n separate panels. The null hypothess of poolablty was rejected n both cases at the conventonal levels of statstcal sgnfcance. owever, although pooled estmaton may ntroduce some bas n ths case, t s stll lkely to be more effcent. xl Moreover, snce we are manly nterested n dfferences n behavour, and there s no obvous reason why pooled estmates for one set of countres would be more based than those for the other, poolng offers an mportant test of the robustness of our overall results. Ths secton therefore, presents the panel data estmates. Panel data unt root tests ndcated that all the seres are I(1), wth the excepton of the nomnal exchange rate seres for developng countres whch was found to be I(2). xl Table 6 reports the results derved from the ADL approach. We specfy the emprcal model as an autoregressve dstrbuted lag model of order one, utlsng a least squares dummy varable (fxed effects) model to ncorporate country-specfc unobserved heterogenety. As seen from columns (2) and (3) of Table 6, whch report dollar regressons for the perod , the panel data estmates support our earler fndng of sgnfcantly hgher exchange rate pass-through elastctes for mports nto developng countres. xl The estmated coeffcents for the contemporaneous nomnal exchange rate are and for developng and ndustral countres, respectvely. xl Moreover, takng nto

22 account lagged effects does not notceably affect the dvergence between the two nomnal exchange rate pass-through elastctes. xlv The coeffcent for the developng country panel s hgher than that found for most ndvdual developng countres. The contemporaneous coeffcent on the world prce level s much greater for the ndustral country group. The adjusted coeffcent of determnaton s much hgher for the developng country panel ndcatng that the specfed varables do a much better job of explanng varaton n mport prces for the developng country group. Retracng the robustness checks employed for the ndvdual country estmates, we next re-estmated the panel specfcatons for ndustral and developng countres usng SDR unts nstead of US dollars. Columns (4) and (5) of Table 6 present the results. The countres whose currences are ncluded n the calculaton of the SDR (Germany, Japan, US, and UK) were excluded to mtgate possble smultanety concerns. The estmated exchange rate coeffcents are qute smlar to the prevous (dollar-based) ones for both groups of countres, although the estmated coeffcent for ndustralsed countres s somewhat hgher. The coeffcents of the contemporary nomnal exchange rate term are and for developng and ndustral countres, respectvely, whle those for contemporary world prce are and respectvely. Agan, takng nto account lagged effects does not notceably affect the dvergence between the exchange rate passthrough elastctes. Notce that the lagged pass-through coeffcents are generally not sgnfcant at the 5 percent level, rrespectve of the currency unt used. As a senstvty test, and n order to explore any changes n behavour over tme, we splt the sample nto two perods, and Columns (7)-(15) present the estmates. The contemporaneous pass-through estmates are remarkably stable across tme perods and currency unts, although the gap between the two sets of pass-through estmates ncreased between the two perods. Recently, Taylor (2000) has suggested that

23 pass-through elastctes have declned n response to a less nflatonary envronment world-wde, leavng frms more wllng to accommodate small prce changes n ther margns. We fnd supportve evdence only for ndustral countres. Indeed, n the case of developng countres, pass-through appears to have ncreased. xlv More conclusve results, however, must awat a more exhaustve treatment that takes nto account the evoluton of nflatonary expectatons. xlv Table 7 reports the results derved from the error correcton approach. As seen from columns (2) and (3), whch report regressons for the perod , the panel data estmates support our fndng of hgher exchange rate pass-through elastctes for mports nto developng countres. Usng the US dollar as the unt of measure, the estmated coeffcents of the contemporary nomnal exchange rate are and for developng and ndustral countres, respectvely. Columns (4) and (5) of Table 7 present the SDR-based estmates. The estmated short-run exchange rate coeffcents are vrtually dentcal to the dollar estmates. Columns (7)-(15) present the estmates for the sub-perods. The short-run pass-through estmates are even more stable across tme perods and currency unts than n the dollar-based estmates. Agan, short-run passthrough nto ndustral country mport prces appears to have declned n the second perod. The error correcton coeffcents are correctly sgned n all nstances and statstcally sgnfcant at the 5 percent level n almost half of the cases. The range of estmated values of these coeffcents ndcate reasonably rapd reverson, suggestng that about percent of the effect of a dsequlbratng shock s removed wthn a year. Fnally, as another robustness check, we estmated panel equatons whle takng nto account tme-specfc effects n addton to cross-secton fxed effects. The estmates, whch are avalable from the author on request, reman essentally unchanged. In bref, all sets of panel data estmates xlv ndcate hgher nomnal exchange rate pass-

24 through nto developng country mport prces. Ths n lght of our theoretcal model renforces the theoretcal possblty that devaluatons are more lkely to be contractonary n the short-run than suggested orgnally by Krugman & Taylor (1978). << INSERT TABLES 6 and 7 ABOUT ERE >> 7. Concludng Remarks The objectve of the frst part of ths paper was to extend the framework developed by Krugman & Taylor (1978) for analyzng the short-run effects of a nomnal devaluaton on the level of output of a sem-ndustralsed developng economy. In dong so, we have attempted to ncorporate some recent developments n the nternatonal economy whle mantanng the smplcty of the orgnal framework. We assume that exchange rate pass-through s hgher for developng country exports sold to other developng countres. Moreover, owners of TNCs are lkely to consume a much smaller proporton of ther earnngs n the host country than the owners of host country frms. Some of the nterestng fndngs emergng from the extended theoretcal framework can be summarsed as follows: The lkelhood of contractonary short-run consequences of a devaluaton vares drectly wth: (1) the proporton of a country's exports destned for other developng countres, and (2) the presence of TNCs n ether the export or home goods-producng sector. xlv Unlke the orgnal Krugman-Taylor framework, a devaluaton wll generally have a contractonary mpact even f: (1) trade s ntally balanced, (2) consumpton behavour does not dffer between wage and proft earners, and (3) the government sector has a hgh margnal propensty to consume n the short-run.

25 As long as the government sector's margnal propensty to consume n the short-run s lower than that of the domestc proft earners, the lkelhood of a negatve fscal effect followng a devaluaton vares nversely wth the presence of TNCs n the export sector. Interestngly enough, the presence of TNCs n the export sector s more lkely to result n a contracton of demand (and thus, of output) followng a devaluaton than ther presence n the home goods-producng sector. Ths s because ncreased profts for domestc exportng frms from sales to ndustralsed countres may make up for the fall n demand orgnatng from other sources. Thus, our extended model ndcates that some evolvng features of global trade and producton make the short-run effects of a devaluaton much more lkely to be contractonary than suggested orgnally by Krugman & Taylor (1978). The postulated dfference between exchange rate pass-through nto prces of goods exported to developng and ndustralsed countres s emprcally explored n the second part wth the help of ndvdual country and panel data regressons. The man fndng emergng from the estmates s that developng country mports exhbt hgher pass-through elastctes. Moreover, ths concluson s robust to dfferent methodologes, specfcatons and unts of account. Thus, a major polcy lesson that emerges from our study s that the short-run consequences of a nomnal devaluaton for output may vary dependng on whether t s relatve to other developng or ndustral countres, and on the composton of a country's export markets. Developng countres that have a major presence of TNCs, and that sell a major share of ther exports to other developng countres, face an ncreased lkelhood of short-run contractonary effects of nomnal devaluatons. Our emprcal study should be seen as a prelmnary exploraton of an ntrgung

26 possblty rased by our extenson of the Krugman-Taylor model. One lmtaton of our study s that, n order to mantan the concseness of the orgnal Krugman-Taylor framework, t assumes away substtuton effects of prce changes on trade volumes, thus focusng on the short run and almost certanly overstatng structural nerta n developng countres. Such substtuton effects are lkely to appear both n producton and consumpton. On the producton sde, substtuton effects are lkely to be much stronger for exports to ndustralsed countres gven that, due to greater pass-through, a devaluaton has a much lower effect on profts for exporters sellng to developng countres. On the other hand, and to the extent that developng country exports are closer substtutes for domestc producton n other developng countres, substtuton n consumpton s lkely to be greater for mports nto those countres. Thus, substtuton effects n consumpton and producton are lkely to counteract each other (to a lesser or greater degree), as far as the destnaton of exports s concerned. xlx Another lmtaton of our study s that t ether uses US dollars or IMF's SDR as unts of account. owever, to the extent that the dollar s the currency of choce for nternatonal trade, and that the currences ncluded n the SDR cover an overwhelmng proporton of nternatonal transactons, ths lmtaton s mtgated. Fnally, an extended study of dfferences n pass-through behavour between ndustralsed and developng countres wll take nto account addtonal factors such as scale economes, asymmetrc responses to apprecatons and deprecatons, varable mark-up behavour, and unt labour costs. Data Items and Sources E Blateral nomnal exchange rates n US dollar terms obtaned from the Internatonal

27 Monetary Fund's Internatonal Fnancal Statstcs (IFS) database, wth the excepton of Tawan for whch data were obtaned from the web ste of the Central Bank of Chna. SDR data obtaned from IFS. P M Import prces or unt mport values. Aggregated values were obtaned from the IFS wth the excepton of Tawan, Mexco, and Chna for whch the values were obtaned from the Tawanese Drectorate-General of Budget, Accountng, and Statstcs, the Bank of Mexco, and the World Bank, respectvely. P World consumer prce ndex obtaned from IFS. P Consumer prce ndces obtaned from IFS. For Tawan, these data were obtaned from the Tawanese Drectorate-General of Budget, Accountng, and Statstcs. GDPPC GDP per capta obtaned from the World Bank's World Development Indcators. For Tawan, the seres was obtaned from the Tawanese Drectorate-General of Budget, Accountng, and Statstcs.

28 Table 1: Augmented Dckey-Fuller (ADF) Test Statstcs for Indvdual Countres Varable P M E P* P M E P* Argentna Australa Belgum-Luxemburg Brazl Canada Chle Chna Colomba Denmark Fnland Germany Greece ong Kong ungary Iceland Inda Ireland Italy Japan Jordan Kenya Korea Maurtus Mexco Morocco Netherlands New Zealand Norway Pakstan Phlppnes N/A N/A Poland Portugal Sngapore South Afrca Span Sr Lanka Sweden Syra N/A N/A Tawan Thaland Turkey UK US N/A N/A Venezuela The reported results are based on the followng regresson: x t = µ + µ x + γ x + υ 1 2 n t 1 t 1 t = 1 t Lag lengths based on Schwarz Informaton Crteron. The 1, 5, and 10 percent crtcal values for most seres are -3.89, -3.05, and -2.67, respectvely. Results for frst-dfferences not presented where the seres s found to be I(0). N/A ndcates not avalable due to sngularty ssues.

29 Table 2: Pass-Through Estmates for Developng Countres Usng the Autoregressve Dstrbuted Lag (ADL) Approach (US dollars, ) Country Constant E t P t * P M,t-1 Adj. R 2 St. Error Argentna (-2.04) (102.4) (1.546) (-1.25) Brazl (-0.68) (16.30) (0.373) (1.062) Chna (0.373) (-0.13) (5.764) (-0.66) (0.569) Chle (2.365) (4.782) (0.394) (-3.13) Colomba (0.626) (2.488) (0.982) (0.913) ong Kong (-2.15) (1.641) (2.676) (1.231) ungary (0.419) (3.099) (1.080) (0.092) Inda (0.063) (0.545) (1.119) (0.130) (-0.24) Jordan (0.014) (0.175) (5.263) (-0.04) (0.642) Kenya (0.396) (0.382) (7.187) (1.850) (-2.95) Korea (0.040) (0.122) (1.024) (13.38) (-0.20) (-1.67) Sr Lanka (0.087) (0.000) (-1.07) (2.112) (1.249) (-0.00) Maurtus (0.183) (3.175) (1.080) (-0.47) Mexco (-0.17) (7.558) (0.616) (0.333) Morocco (-2.34) (1.366) (2.637) (-0.770) Pakstan (1.257) (1.270) (0.858) (-0.820) Poland (0.925) (2.950) (0.192) (0.042) Sngapore (0.718) (-0.60) (-1.021) (-0.502) South Afrca (0.758) (2.810) (-0.03) (2.497) Syra (-0.04) (1.421) (0.347) (-0.861) Tawan (-0.87) (1.699) (0.757) (0.051) Thaland (1.348) (5.191) (0.489) (-1.921) Turkey (-0.61) (8.509) (1.087) (1.099) Venezuela (0.928) (9.125) (-0.14) (1.415) Dependent varable: P M,t. t-statstcs n parentheses. All varables n logs.