Price Competition and the Fallacy of Composition in Developing Country Exports of Manufactures: Estimates of Short-Run Growth Effects

Transcription

1 Unversty of Massachusetts Amherst Amherst Economcs Department Workng Paper Seres Economcs 2005 Prce Competton and the Fallacy of Composton n Developng Country Exports of Manufactures: Estmates of Short-Run Growth Effects Robert A. Blecker Arslan Razm Unversty of Massachusetts - Amherst Follow ths and addtonal works at: Part of the Economcs Commons Recommended Ctaton Blecker, Robert A. and Razm, Arslan, "Prce Competton and the Fallacy of Composton n Developng Country Exports of Manufactures: Estmates of Short-Run Growth Effects" (2005). Economcs Department Workng Paper Seres. 54. Retreved from Ths Artcle s brought to you for free and open access by the Economcs at ScholarWorks@UMass Amherst. It has been accepted for ncluson n Economcs Department Workng Paper Seres by an authorzed admnstrator of ScholarWorks@UMass Amherst. For more nformaton, please contact scholarworks@lbrary.umass.edu.

2 DEPARTMET OF ECOOMICS Workng Paper Prce Competton and the Fallacy of Composton n Developng Country Exports of Manufactures: Estmates of Short-Run Growth Effects by Robert A. Blecker and Arslan Razm Workng Paper Revsed verson, September 2006 UIVERSITY OF MASSACHUSETTS AMHERST

3 Prce Competton and the Fallacy of Composton n Developng Country Exports of Manufactures: Estmates of Short-Run Growth Effects by Robert A. Blecker Amercan Unversty, Washngton, DC, USA and Arslan Razm Unversty of Massachusetts, Amherst, MA, USA Revsed August 2006 Abstract Ths paper studes whether ntra-developng country prce competton has sgnfcant effects on the short-run growth rates of developng countres that are specalzed n manufactured exports. Regresson estmates usng the generalzed method of moments (GMM) appled to annual panel data for 17 developng countres n show that these countres exhbt a fallacy of composton, n the sense that a real deprecaton relatve to competng developng country exporters ncreases the home country s growth rate n the short run. The results also suggest that real deprecatons for these developng countres relatve to the ndustralzed countres are contractonary. JEL Codes: F43, O19, O14, F32 Keywords: Export-led growth, fallacy of composton, terms of trade, manufactured exports, contractonary devaluatons, compettve devaluatons. Contact Author: Robert A. Blecker, Department of Economcs, Amercan Unversty, 4400 Massachusetts Ave.,.W., Washngton, DC , USA. Phone (202) , fax (202) , e-mal blecker@amercan.edu. Acknowledgements: The authors would lke to acknowledge helpful comments from Stephane Seguno, Peter Skott, two anonymous referees, and partcpants n semnars at the Schwartz Center for Economc Polcy Analyss (SCEPA) at The ew School and the Graduate Faculty of Economcs at the Unversdad aconal Autónoma de Méxco (UAM). The authors would also lke to thank the Economc Polcy Insttute and SCEPA for fnancal support.

4 Prce Competton and the Fallacy of Composton n Developng Country Exports of Manufactures: Estmates of Short-Run Growth Effects 1. Introducton In the past several decades, many developng countres have re-focused ther growth strateges on promotng manufactured exports, rather than manufactures for the domestc market ( mport substtuton ) or exports of prmary commodtes. Snce ths strategy has been adopted by some of the most successful developng countres, ncludng the four Tgers (South Korea, Tawan, Hong Kong and Sngapore) n the 1970s and 1980s and Chna more recently, economsts have devoted a great deal of attenton to analyzng the polces that are responsble for ths success and whether (or how) they could be emulated n other developng natons. One branch of ths lterature has focused on whether openness to trade has postve effects on longrun average growth rates, and to what extent openness as conventonally measured reflects free trade polces as opposed to other polces or nsttutons. 1 A related lterature has studed the degree to whch the most successful cases of manufacturng export-led growth can be attrbuted to free market forces versus nterventonst ndustral polces. 2 A less studed dmenson of ths topc s whether the efforts of so many developng countres to export ther way out of underdevelopment by specalzng n smlar types of manufactured products mostly targeted on the same ndustralzed country markets are leadng to a fallacy of composton. That s, wll the success of some countres necessarly lead to the falure of others, or even to dsappontng results for all countres attemptng to follow the same strategy. Ths concern has receved a new mpetus from the recent emergence of Chna as 1

5 an ndustral export powerhouse a phenomenon that s sometmes perceved as threatenng not only establshed manufacturng producers n the ndustralzed countres but also newly emergng producers of manufactures n other developng natons. It must be emphaszed that the ssue of a potental fallacy of composton arses ndependently of the polcy regme or nsttutonal settng that generates export-orented ndustralzaton although t s possble that dfferent polces or nsttutons may permt better or worse responses to the problem, f the problem exsts. Concern over a fallacy of composton has taken several forms. One verson s smply that the ndustralzed country markets for developng countres exports of manufactures are lmted quanttatvely, and hence the success of some of the latter countres nevtably dsplaces other developng countres from those markets. A second verson s the belef that oversupply of smlar types of manufactured exports by large numbers of developng countres creates a problem of fallng terms of trade. A thrd verson focuses on the exstence of ntra-developng country prce competton, through polces such as compettve devaluatons or wage represson. Most of the emprcal lterature on the fallacy of composton to date has studed the behavor of export quanttes and/or prces n global markets. These studes have found evdence for quanttatve dsplacement n export markets and prce competton between developng country exporters, although evdence on declnng terms of trade for developng country exports of manufactures s mxed. However, no studes have yet tested for effects of ntra-developng country prce competton on the growth of output n these countres. 3 Ths paper s the frst to test for whether changes n relatve prces or real exchange rates between dfferent developng country exporters have sgnfcant effects on the short-run growth rates of the exportng natons. Specfcally, we test the followng fallacy of composton (FOC) hypothess: that a reducton n the relatve prce of one developng country s exports (.e., a real deprecaton) wth respect to 2

6 competng developng natons exports has a postve effect on that country s growth rate but a negatve effect on the growth rate of ts compettors (n the short run). Gven ths focus on relatve prces or real exchange rates, our analyss s necessarly lmted to the short run, and does not address long-run average growth rates as are typcally consdered n more general studes of the relatonshp between openness and growth. A related but dstnct lterature has argued that currency devaluatons 4 n developng countres are often contractonary n the short run. Devaluatons can be contractonary f the stmulus to exports s relatvely weak (e.g., because of low prce elastctes), and/or f the stmulus to exports s offset by other, negatve effects of a real devaluaton. These contractonary effects nclude regressve dstrbutonal effects (.e., lower real wages) that reduce domestc demand, reduced purchasng power over mports, and ncreased costs of servcng foregncurrency denomnated debt. The contractonary devaluaton dea mght appear to contradct our FOC hypothess, because the former mples that a cheapenng of a country s exports reduces the country s own growth. However, we thnk these two hypotheses are not mutually exclusve because the FOC apples to real deprecatons relatve to competng developng countres, whle contractonary effects of devaluatons are more lkely when there s a real deprecaton relatve to the ndustralzed countres (snce the latter countres are the prmary sources of developng countres mports and loans). Therefore, we also test the contractonary devaluaton (COD) hypothess: that a lower relatve prce of exports (real deprecaton) of a developng country wth respect to the ndustralzed countres has a negatve effect on the deprecatng country s own short-run growth rate, n spte of any possble stmulus to exports. To test these twn hypotheses (FOC and COD) emprcally, we utlze data for a sample of 18 developng countres and 10 ndustralzed countres coverng the years Ths 3

7 data set ncludes all major developng countres for whch manufactures consttuted more than 70% of total exports as of 2000, whch s approxmately the average for all developng countres. Ths cut-off was used for two reasons: frst, so that the behavor of manufactured exports can plausbly be hypotheszed to have a sgnfcant mpact on aggregate short-run growth; and second, so that aggregate export prce ndces (unt values) can plausbly be assumed to represent prmarly the prces of manufactured goods (dsaggregated export prce ndces for manufactures were not avalable for most of the countres n our sample). Instead of relyng on conventonal real exchange rate ndces based on blateral or multlateral trade shares wth all tradng partners, we construct separate trade-weghted real exchange rate ndces for the developng countres relatve to the mportng ndustralzed countres and relatve to each other. Dfferent sets of relatve prces or real exchange rates are used as a senstvty test. In addton to estmates usng the whole panel, we also estmate smaller panels consstng of more homogeneous groups of countres along four structural dmensons: country sze (and openness), ratos of manufactured exports to gross domestc product (GDP), the technologcal composton of exports, and the external debt-gdp rato. As one would expect, usng these more homogeneous groupngs allows for more precse estmates and reveals nterestng dfferences n the estmated coeffcents based on dfferent structural characterstcs. The qualtatve results of ths estmaton are not senstve to the measure of relatve prces (real exchange rates), but are senstve to the econometrc procedure used. Usng the generalzed method of moments (GMM), the results for most of the panels generally support the FOC and COD hypotheses, although there are some exceptons that reflect structural dfferences among dfferent panels of countres. Usng ordnary least squares (OLS) wth fxed effects, however, there s relatvely strong support for COD but not for FOC. Gven that GMM yelds more precse estmates n the presence of 4

8 endogenety problems and dynamc effects, the GMM results are preferred, but the OLS results suggest cauton n nterpretng the fndngs and the need for further research on ths topc. 2. Lterature revew and testable hypotheses A large lterature has found robust evdence for a postve relatonshp between openness to trade and long-run growth performance across countres. 5 However, as noted n the comprehensve survey by Wnters (2004, p. F4), Cross-country studes face problems n defnng openness, n dentfyng causaton and n solatng the effects of trade lberalsaton. Yankkaya (2003) reports evdence that more explct measures of trade lberalzaton do not have sgnfcant, postve effects on economc growth, and some measures of trade barrers have a postve assocaton wth growth especally n developng countres. Rodrk et al. (2004) fnd that when proxes for nsttutonal dfferences among countres are ncluded n an equaton for per capta ncome, the effects of openness become nsgnfcant or have the wrong (negatve) sgn, suggestng that deeper nsttutonal factors (such as rule of law) account for successful performance n regard to both trade and growth. Recently, some contrbutors to ths lterature (e.g., Prtchett, 2000; Rodrk, 2005) have begun to express radcal skeptcsm that any meanngful results can be gleaned from cross-sectonal or panel studes of long-run growth performance about ssues of causalty, especally wth regard to the effects of polces that are lkely to be endogenous to the growth process. Prtchett (2000, p. 222) argues n partcular that The use of hgh-frequency [.e., annual] panel data, partcularly wth fxed effects, to nvestgate long-run growth correlates s almost certanly pontless. However, the twn ssues of a fallacy of composton (FOC) and 5

9 contractonary devaluatons (COD) addressed n ths paper are not concerned wth the determnants of long-run, average growth rates (or average, per capta ncome levels). The COD hypothess s planly concerned wth short-run negatve effects of currency deprecatons on output, rather than effects on long-run growth. The FOC hypothess can be nterpreted n varous ways, some of whch may mply the exstence of an nternatonal addng-up constrant on the long-run growth of ndvdual countres. However, the tests for the FOC n ths paper focus on ntra-developng country prce competton, based on a short-run macroeconomc model n whch changes n a country s balance of payments can clearly mpact upon ts growth n the short run. 6 Ths paper s also dstngushed from the more conventonal trade-and-growth studes because we focus on a more lmted, relatvely homogeneous sample of countres and a more narrow research queston. Rather than use a large sample of countres wth dverse economc structures and ncome levels (many of the conventonal studes use over 100 countres), we have delberately selected a smaller subset of countres that have a common structural characterstc: a very hgh proporton of manufactures n ther exports. The 18 countres n our data set are ones that have already acheved a hgh degree of external orentaton and a common specalzaton n manufactures; the man queston we are concerned wth here s whether that common structural characterstc exposes them to sgnfcant prce competton wth each other that affects ther growth n the short run. Although we do not go to the level of ndvdual country case studes or analyze specfc epsodes of faster or slower growth as recommended by Prtchett (2000), we lmt our analyss to countres wth sgnfcant structural smlartes and a tme perod ( ) when all of them successfully promoted exports of manufactures. The exstng lterature on the FOC can be dvded nto three (overlappng) areas of focus: quanttatve crowdng-out n ndustralzed country markets; the commodtzaton of labor- 6

10 ntensve manufactures and the hypothess of a declne n ther terms of trade; and ntradevelopng country prce competton n global export markets. 7 Startng wth quanttatve crowdng out, Clne (1982) argued that the rates of export growth acheved by the east Asan four tgers n the 1970s could not plausbly be generalzed to a larger number of developng countres wthout provokng a protectonst response n the ndustralzed countres. Crtcs (e.g., Balassa, 1989) responded that competton among developng countres s lmted by the graduaton of the more advanced developng countres nto the producton of relatvely more captal-ntensve exports, leavng room for other countres to enter the market for relatvely less sophstcated exports. Such responses supported the dea of a flyng geese formaton, n whch successve waves of developng countres move from less technologcally advanced, laborntensve goods to more technologcally advanced, captal-ntensve goods, and the advance of some countres allows new entrants to succeed n the markets for the less advanced products. 8 Some studes have focused more narrowly on trends n US mports from Japan and the newly ndustralzng developng countres. Blecker (2002, 2003) found that new exporters (manly Chna and Mexco) substantally dsplaced Japan and the four tgers market shares n the US market. Palley s (2003) econometrc estmates showed that mports from Chna sgnfcantly crowded out mports from the four tgers durng the whole sample perod , whle Mexcan products dsplaced Japanese mports n the latter half of the perod. Other analyses have focused on the entry of Chna nto the global economy. Walmsley and Hertel (2000) used a dynamc general equlbrum model to analyze the effects of Chna s accesson to the WTO. Accordng to ther smulatons, although the world as a whole benefts from Chna s accesson, ts compettors n the labor-ntensve apparel ndustry experence sgnfcant losses. An econometrc study by Echengreen et al. (2004) found that Chna s entry nto the WTO would 7

11 have favorable effects on newly ndustralzng countres that export captal goods to Chna whle hurtng other developng countres that compete wth Chna as producers of consumpton goods. The dea of commodtzaton s found mostly n studes of partcular countres and markets, such as the work of Kaplnsky (1993) on export processng zones n the Domncan Republc and neghborng Carbbean natons. Accordng to Kaplnsky, commodtzaton has occurred because so many countres have specalzed n smlar types of low-technology goods (e.g., apparel) n whch they all have a statc comparatve advantage at the same tme. 9 He fnds that these countres have used exchange rate devaluatons to hold down wage costs n foregn currency (US dollars), and thereby to mantan external compettveness. The result s a fall n the terms of trade, whle real wages fal to rse (or even fall) leadng to mmserzng employment growth and very lmted gans from the expanson of exports. The hypothess of a declnng trend n the terms of trade for prmary commodty exports of developng countres n the 20 th century s now wdely accepted (see Sapsford and Snger, 1998; Sapsford and Chen, 1998), but the dea of a smlar trend for manufactured exports s more controversal (see, for example, Sarkar and Snger, 1991; Bleaney, 1993). Mazels, Palaskas, and Crowe (1998) and Mazels (2000) have found evdence for decreasng trends n the net barter terms of trade (BTT) for manufacturng goods exported by developng countres usng data for prces of mports of those goods nto the EU and the US. However, some of these studes also found that the declnes n the BTT were more than offset by volume gans, so that the ncome terms of trade rose substantally. There are also mportant qualfcatons to these studes n terms of the generalty, tmng, and magntude of the worsenng BTT, and other studes have found more mxed results. 10 Kaplnsky (1999, 2005) fnds that the trends n the terms of trade have vared among dfferent countres and for dfferent types of manufactures, makng any 8

12 generalzatons dffcult. Usng more dsaggregated data than most prevous studes, Kaplnsky and Santos-Paulno (2006) fnd that prces of EU mports of manufactures have fallen the most for mports from lower-ncome countres and of less technologcally advanced products. Several econometrc studes have confrmed that developng country exports of manufactures are senstve to relatve prces (real exchange rates) vs-à-vs competng exports from other developng countres. Fan et al. (1992) estmated export demand functons for 23 developng countres and found that, for most of those countres, competton wth other developng country exporters was more sgnfcant than wth ndustralzed country exporters. Muscatell et al. (1994) found econometrc evdence that ntra-developng country competton was more mportant among a sample of fve east and southeast Asan newly ndustralzng economes than between them and the ndustralzed countres. Razm and Blecker (2006) have obtaned broadly smlar results usng data for the same 18 developng countres consdered here, ncludng more recent data (for 1983 to 2001) than the prevous studes, although wth some exceptons (notably Korea, Tawan and Mexco). Razm and Blecker s panel estmates suggest that ntra-developng country competton s sgnfcant only among countres exportng manly low-technology products, whle countres that export more hgh-technology products compete more wth ndustralzed country producers and also have hgher ncome (expendture) elastctes for ther exports. However, none of the econometrc studes of ntra-developng country prce competton has yet studed the effects of ths competton on the growth of output rather than exports. Of course, changes n a country s export prces (whether due to changes n currency values or domestc producton costs) wll affect the country s real exchange rate wth the ndustralzed countres as well as n relaton to competng developng countres. Ths rases the 9

13 classc queston of whether real deprecatons are expansonary or contractonary, an ssue that has generally been consdered (for the developng countres) wth regard to ther exchange rates wth the major currences of the ndustralzed natons. Standard open economy macro models mply that currency deprecatons are expansonary on the demand sde, provded that prce elastctes satsfy the Marshall-Lerner condton (so that the trade balance mproves) and there are unemployed resources n the devalung country (so that output can ncrease). A real deprecaton can also yeld benefts on the supply sde by ncreasng the prce of traded goods relatve to nontraded goods, whch creates ncentves to shft domestc producton towards tradables and demand towards non-tradables, thereby freeng up a greater surplus for export. A substantal body of lterature has challenged ths tradtonal vew for developng countres, and has argued for potentally contractonary effects of real currency deprecaton on several grounds. 11 The real balance and wealth effects of a devaluaton are lkely to be negatve, f the devaluaton leads to hgher domestc prces. Especally, countres wth sgnfcant foregncurrency denomnated debts wll face ncreased debt servcng burdens as a result of a devaluaton. A devaluaton may also shft the dstrbuton of ncome n favor of captal and aganst labor, by enablng frms to ncrease prce-cost margns. If captal owners have a hgher propensty to save than workers, then overall aggregate demand may fall n spte of ncreased exports (although ths result can be reversed f nvestment demand s hghly responsve to proftablty). 12 On the supply sde, an ncrease n the cost of mported nputs may ncrease domestc costs of producton, and f wages are ndexed, labor costs may also rse. The lterature on contractonary deprecatons has focused manly on nomnal exchange rate changes or trade-weghted aggregate real exchange rates relatve to the major currences of the ndustralzed natons. Ths focus s probably justfed, because some of the causes of 10

14 contractonary devaluatons especally ncreased real debt burdens and reduced terms of trade (lower purchasng power over mports) depend manly on exchange rates relatve to the ndustralzed countres, whch are the man sources of loans and mports for developng countres. On the other hand, the exstence of ntra-developng country competton means that a real deprecaton could provde a boost to a developng country s export and output growth. For these reasons, we hypothesze that a real deprecaton relatve to the ndustralzed country currences s more lkely to be contractonary, whle a deprecaton relatve to competng developng countres s more lkely to be expansonary. Thus, the specfc hypotheses that wll be tested n ths paper consst n the followng: FOC: A real deprecaton (lower relatve prce of exports) wth respect to competng developng countres n ndustralzed country markets for manufactures wll boost short-run growth n the devalung country but reduce t n the competng developng countres; and COD: A real deprecaton (lower relatve prce of exports) wth respect to ndustralzed countres wll reduce short-run growth n a developng country f export gans are offset by contractonary effects such as reduced purchasng power over mports, ncreased foregn debt burdens, and regressve ncome redstrbuton. It should be noted that these hypotheszed effects of a deprecaton of a gven country s currency are lkely to vary dependng on whether t s matched by compettve deprecatons of other developng countres and hence whether there s a compettve gan vs-à-vs the latter countres or not, as well as on structural characterstcs of the countres such as ther external debt burdens. 3. Data set and modelng approach 11

15 3.1 Country selecton and panel defntons In selectng the countres to nclude n ths study, we restrcted ourselves to those developng countres for whch exports of manufactures could plausbly have a sgnfcant nfluence on ther aggregate growth performance. Therefore, we ncluded only the 18 major developng countres for whch manufactured products consttuted at least 70% of ther exports n These countres are: Bangladesh, Chna, the Domncan Republc, Hong Kong, Inda, Jamaca, South Korea, Malaysa, Maurtus, Mexco, Pakstan, the Phlppnes, Sngapore, Sr Lanka, Tawan, Thaland, Tunsa, and Turkey. 14 The ndustral countres n the sample are the 10 largest mporters of manufactured products from developng countres as of 1990: Belgum, Canada, France, Germany, Italy, Japan, etherlands, Swtzerland, the Unted Kngdom, and the Unted States. Usng data for these ndvdual countres allows for the constructon of the country-by-country trade weghts used n calculatng the prce ndces descrbed above. 15 An alternatve crteron that mght have been used to select the developng countres n the sample s manufactured exports measured as a percentage of GDP, whch s arguably more relevant for the growth mpact of those exports. 16 The frst two columns n Table 1 show both of these ndcators,.e., manufactured exports as percentages of total exports and GDP, respectvely. The countres dffer much more by the latter crteron, rangng from a low of 7.1% n Inda to a hgh of 129.0% n the cty-state of Sngapore. However, there are very few major developng countres (excludng former Sovet-block transton economes) that have a hgher rato of manufactured exports to GDP than Inda, whch do not also meet the 70% of total exports threshold. The few countres that are n ths category (e.g. Costa Rca, Indonesa and South Afrca) have percentages of manufactures n total exports that are suffcently low that ther export prce (unt value) ndces would not be relable ndcators of the prces of ther 12

16 manufactured exports, and would be lkely to ncorporate sgnfcant measurement error due to fluctuatons n prmary commodty prces. Hence, these countres have been omtted from the sample. evertheless, the results of our analyss could vary between countres wth dfferent ratos of manufactured exports to GDP, and therefore we wll dvde the panel nto two subpanels of countres that are above and below a 25% threshold for ths ndcator, referred to as HIMFRGDP and LOMFRGDP, respectvely. (A complete lst of the countres ncluded n each sub-panel s gven n Table 2.) [Insert Table 1 about here] The thrd and fourth columns of Table 1 are used to classfy countres nto two alternatve sub-panels, consstng of small open economes and large countres. Openness s measured n the standard way by total merchandse trade (exports plus mports) as a percentage of GDP (whch s, as one would expect, postvely correlated wth the share of manufactured exports n GDP). For country sze, we used an admttedly arbtrary cut-off of a GDP of US$100 bllon n We then classfed any country wth a trade share n GDP of greater than 50% and a GDP of less than US$100 bllon n the SMALLOPE panel and put all other countres n the LARGE panel. We expect FOC effects to be stronger n the SMALLOPE and HIMFRGDP panels. [Insert Table 2 about here] Table 1 also reports the percentage of hgh technology goods n each country s manufactured exports. We use ths ndcator based on the expectaton that the FOC hypothess may be more lkely to apply to countres that are specalzed n less technologcally sophstcated, more commodtzed exports, such as textles and apparel. To obtan sub-panels of roughly smlar szes, we desgnated countres wth 30% hgh technology exports or above n 2000 as hgh-technology (HITECH) exporters, and those wth less than 10% as low- 13

17 technology (LOTECH) exporters; 17 Chna and Mexco were ncluded n both panels because of ther ntermedate status (these two countres had approxmately 20% hgh technology exports n 2000) and on the assumpton that they compete wth countres n both groups. Fnally, Table 1 lsts the ratos of external debt to GDP for each country, whch allow us to dstngush countres wth relatvely hgh and low debt burdens (HIDEBT and LODEBT, respectvely) usng a onethrd (33%) cut-off. Ths wll permt us to test the suggeston n the COD lterature that currency devaluatons are more lkely to be contractonary n developng countres wth large external debts. 18 Because debt ratos vary consderably over sample perod, we used the average of 1990 and 2000 for ths ndcator. Ths gves us a total of eght sub-panels consstng of more structurally homogeneous countres, whch wll allow us to test for dfferences among these varous country groupngs n the results of our model estmates. 19 As may be seen n Table 2, there are some smlartes between some of the panels; especally, the LOMFRGDP and LOTECH panels are dentcal except for Maurtus, whch s ncluded only n the latter. evertheless, there are some subtle dfferences n the econometrc results for these dfferent panels, as wll be seen below. 3.2 Emprcal specfcaton and prce measures The emprcal specfcaton adopted n here s based on a theoretcal model of exports and growth for a group of developng countres, all of whch export smlar manufactured goods (.e., mperfect substtutes) to a common set of ndustralzed country markets. The mathematcs for ths model are gven n an appendx; ths secton presents a bref ntutve summary of (and justfcaton for) the model followed by a dscusson of the econometrc approach and data set. Based on the model of balance-of-payments constraned growth (see McCombe and 14

18 Thrlwall, 1997, 2004), we assume that the growth rates of developng countres that are open to trade are constraned by ther need for mports n the growth process (for example, mports of captal goods, ntermedate goods, and consumer goods not produced at home), and the fact that these mports (whch requre foregn exchange) can only be fnanced through ether export earnngs or net captal nflows. Hence, the parameters of a standard model of export and mport demand (essentally, prce and ncome elastctes) along wth the volume of captal nflows place constrants on the expanson of the economy n the short run. These constrants can be releved ether by an ncrease n net fnancal nflows or, more to the pont of ths paper, by mprovng the compettveness of a country s tradable goods relatve to rval producers. Also, gven that the ndustralzed countres consttute the prncpal market for the exports of the developng countres, the growth rate of that market (and ts openness to mports) s another constrant on the growth of the developng natons that export manufactures. Ths vew of a balance-of-payments constrant may appear to be a revval of the oldfashoned gap approach to growth, whch n some crcles s consdered obsolete (for an especally entertanng attack on gap models, see Easterly, 2001). Several ponts are n order to justfy our use of a gap-type analytcal framework n ths paper. Frst, we are concerned only wth a foregn exchange gap, not a savng-nvestment gap; thus, crtques of savng-nvestment gaps are not relevant to ths study. Second, the dea of a foregn exchange gap s relevant to the countres n our data set, because all of them have hgh mport coeffcents for captal and ntermedate goods that are essental for both export and domestc producton. Whether or not more savng (domestc or foregn) would ncrease nvestment n these countres may be doubtful, 20 but more foregn exchange can certanly allow them to purchase more necessary mports and may also relax credt constrants on prvate nvestment and government spendng. 15

19 Thrd, we are concerned here only wth short-run growth effects, and n that context t s wdely accepted that foregn exchange constrants can be bndng. 21 For example, ths pont s recognzed n the lterature on sudden stops n captal nflows durng fnancal crses: The channels through whch large reversals of captal nflows lead to collapses n output and employment are well known. A net wthdrawal of fnancal captal curtals the avalablty of credt for purchasng the mported materals and ntermedate goods that are necessary to produce output n the short run and for fnancng nvestment needed to sustan output over the medum run... Furthermore, n a Keynesan world, declnes n output, ncomes, and nvestment are exacerbated through multpler effects. (Isard, 2005, p. 165) Logcally, these short-run effects of captal nflow reversals should be reversed and output should be stmulated when captal nflows turn postve or ncrease at least, t s an emprcal queston whether the stmulus effects of relaxng foregn exchange constrants are found n our sample of countres, n normal years as well as durng perods of crss and recovery. In our econometrc analyss we wll estmate equatons of the followng general form (gnorng lags and country fxed effects at ths pont for smplcty): Y ˆ = α + β ˆ ˆ ˆ Fˆ + u t L 1Zt + β2rt + β3rt + β4 t t (1) where Y t s real GDP n country at tme t, α s a constant term, Z t s real total expendtures on mports of manufactured goods by the ndustralzed countres, R t s the real exchange rate wth respect to the ndustralzed countres (.e., the relatve prce of the ndustralzed countres goods to developng country s exports), L R t s the real cross-exchange rate relatve to other developng country exporters (.e., the relatve prce of competng developng countres goods to F t developng country s exports), s net captal (fnancal) nflows (measured as a percentage of GDP) nto country at tme t, and u t s a random error term. All varables are expressed n annual growth rates, ndcated by a ^ and measured by dfferences n natural logarthms. Ths use of (logarthmcally) dfferenced varables n ths model s approprate for three 16

20 reasons. Frst, t s consstent wth the theoretcal model outlned n the appendx, n whch all data are expressed as rates of change. Second, all of the varables are statonary n log dfferences, whle some of them have unt roots n log levels. 22 Thrd, dfferencng the data confnes the econometrc analyss to short-run effects (ncludng dynamcs of up to a few years, once lags are added). The twn hypotheses of fallacy of composton (FOC) and contractonary devaluaton (COD) outlned earler have clear nterpretatons n terms of the sgns of the coeffcents β 2 and β 3 n equaton (1). Because R and R L are defned as relatve prces of foregn goods, an ncrease n the growth rate of ether varable ndcates a faster real deprecaton. Therefore, β 3 > 0 suggests support for FOC, whle β 2 < 0 provdes support for COD. More rgorously, the real exchange rate varables n (1) are defned as R = P p and t t t L R = P p, where p t s the export prce of developng country, and and are ndces L t L t t P t P t of the prces of competng goods produced n the ndustralzed and developng countres, respectvely, approprately trade weghted for each country. Although nomnal exchange rates do not appear n these defntons, all of the prces and ndces are expressed n US dollars so that L nomnal exchange rate conversons have already taken place. To calculate and, we use a dual weghtng scheme prevously utlzed n Razm and Blecker (2006). 23 Indexng the developng countres by and the ndustral countres by j, the prce ndex for the ndustral bloc correspondng to each developng country s defned as: P t P t P t π π 1 2 jt jt = j 1 2 π jtπ jt j P jt (2) 1 2 where s the share of ndustral country j n developng country s exports at tme t, s π jt the share of ndustral country j n total ndustral country mports from developng countres at π jt 17

21 tme t, and P jt s the prce ndex for mport-competng domestc products n ndustral country j at tme t. The dual weghts are motvated by the dea that the actual share of a gven ndustralzed country n the exports of a partcular developng country may not be an adequate measure of the latter country s potental nterest n (or ablty to) export to the former market f the latter country can make ts exports more compettve, and hence ths share s nteracted wth the 2 average share of that ndustralzed country n the exports of all developng countres. 24 The prce ndex for the competng developng country exporters for each developng country s constructed as follows: 1 π jt π jt L 1 3 P t = π jt π kjt p j k kt (3) where 3 π kjt s the share of ndustral country j s mports orgnatng n ( s compettor) developng country k (where k ) at tme t, and p kt s the export prce ndex of developng country k (k ) at tme t (n US dollars). 25 In ths case, the dual weghts take nto account both the share of each developng country s exports gong to a certan ndustral country and the shares of other 3 developng countres competng exports destned for that same ndustral country. 26 The real expendture ndex (scale varable) for each developng country s defned as: 27 1 π jt π kjt Z t = j π 1 jt t P X jt (4) where X jt s the total value (n nomnal US dollars) of mports of manufactured products by ndustral country j from all the developng countres n the sample at tme t. Ths varable s weghted by the share of developng country s exports sold n ndustralzed country j ( construct an ndex that s relevant for each developng country s potental export volume. 28 Real 1 π jt ) to 18

22 expendtures for each country are then calculated by deflatng the weghted nomnal expendtures by the prce ndex for ndustral countres P t, as defned n equaton (2) above The trade data used n computng the varous market shares (,, and ) and the π jt π jt π kjt value of manufactured mports of the ndustralzed countres from the developng countres X jt were obtaned from the Unted atons, Commodty Trade Statstcs (COMTRADE) database, wth the excepton of Tawan for whch data were obtaned from the Organsaton for Economc Co-operaton and Development (OECD), SourceOECD database. Manufactures were defned as all products n standard nternatonal trade classfcaton (SITC) categores 5-8 excludng category 68, whch conssts manly n mnerals. Real GDP was taken from the World Bank, World Development Indcators, on-lne database. Perod average exchange rates and prce ndces (wth a few exceptons as noted below) were obtaned from the Internatonal Monetary Fund, Internatonal Fnancal Statstcs (IFS). As a senstvty test, we used two dfferent sets of prce ndces (plus addtonal ones not reported here for reasons of space). 29 The frst set conssts of export unt values for the developng countres and producer prce ndces (PPIs) for the ndustralzed countres. The use of a PPI rather than an export prce ndex for the ndustralzed countres prces (P jt n equaton 2) s based on the assumpton that developng country exports compete manly wth domestcally produced goods n the ndustralzed countres, rather than wth the specalzed exports of the latter. 30 For the developng countres, the export unt values reported n the IFS were used for the prces p kt n equaton (3), wth a few exceptons. 31 Snce the sample was restrcted to developng countres n whch manufactures accounted for at least 70% of total exports, these ndces should predomnantly reflect trends n manufactured export prces. 32 et captal nflows F were measured by net fnancal nflows n US dollars n the IFS, and were expressed as a share of 19

23 nomnal GDP (n US dollars) to normalze for country sze. The use of prce ndces based on export unt values may be consdered problematc for three reasons. Frst, export prces may be endogenous f they are affected by a country s own domestc growth (tests and solutons for ths problem are dscussed n secton 4 below). Second, there are well-known problems wth export unt values due to the changng composton of exports over tme (fxed weght export prce ndces would be preferable, but are not avalable for most developng countres). Thrd, export prces may be correlated across countres f goods are prced to market or f small-country exporters are prce-takers n global markets. For example, apparel exports may be sold at dollar prces that are determned n world apparel markets, rather than by domestc costs and exchange rates n the exportng country. In response to ths last problem, we would lke to have ndces of underlyng costs of producton, such as unt labor costs. However, these are not generally avalable for the developng countres n our sample. Instead, we use an admttedly mperfect proxy for domestc costs, whch s the consumer prce ndex (CPI) of each country converted to US dollars by the perod average exchange rate (producer prce ndces were not avalable for most developng countres). For consstency, we also use CPIs for ndustralzed country prces n ths second set of estmates. Usng CPIs means that the relatve prces R and R L n equaton (1) are measured as CPI-adjusted real exchange rates. It s hoped that these real exchange rates may reflect the ndvdual countres underlyng compettveness better than export prce ndces whch may be more correlated across countres, n spte of the mperfect correlaton between consumer prces and producton costs n export ndustres. At least, ths second set of prce measures gves us a senstvty test for whether our results depend on the use of partcular prce ndces. Table 3 shows coeffcents of varaton for the four relatve prces for each country (two 20

24 dfferent prce measures relatve to both the ndustralzed countres and other development countres). These statstcs show consderable proportonal varaton n these seres for each country, and that the degree of varaton dffers across countres. The fact that the average coeffcents of varaton for all countres wth respect to other developng countres are lower than those wth respect to the ndustralzed countres for both prce measures s consstent wth a vew that exports are prced to market and/or that the developng countres engage n compettve deprecatons to some extent. [Insert Table 3 about here] 4. Econometrc analyss and results All equatons are estmated usng annual panel data for We use two alternatve regresson procedures, OLS wth fxed effects and the Arellano-Bond two-step GMM method for dynamc panel data estmaton. The former s used as a baselne manly because of ts wdespread use n econometrc studes. Usng country fxed effects s helpful because (as emphaszed by Prtchett, 2000) they are lkely to capture other causes of nternatonal dfferences n long-run, average growth rates,.e., the nsttutonal, geographc or polcy factors usually consdered n long-run growth studes. However, the OLS results are subject to smultanety bas f some of the rght-hand sde varables are endogenous, whch s confrmed n the present case by Hausman weak exogenety tests. Moreover, OLS estmates of a fxed-effects model yeld based estmates of the coeffcents n the presence of a lagged dependent varable (snce the latter s correlated wth the error term by constructon), and therefore we cannot nclude the lagged dependent varable n the equatons estmated by OLS. However, a dynamc specfcaton 21

25 wth a lagged dependent varable s lkely to consderably mprove our estmates n the presence of persstence effects and omtted supply-sde factors (such as nsttutonal varables). Usng GMM corrects for both of these problems, because t allows us to nclude the lagged dependent varable and also controls for endogenety through the use of nstrumental varables, as descrbed n more detal below. 34 The lagged dependent varable s hghly sgnfcant for almost all of the panels estmated by GMM, ndcatng the presence of dynamc adjustment over tme. The general form of all the estmated equatons ncludes both the current and one-year lagged value of each ndependent varable (plus a lagged dependent varable for the GMM estmates). Each general model was then tested and reduced to a more specfc form followng the general-to-specfc (GTS) modelng strategy (see Cuthbertson et al., 1992; Charemza and Deadman, 1997). Restrctons were based on the sgnfcance levels of the reported estmates, and varables (current or lagged) were elmnated based on Wald and redundant varables tests for both the omsson of ndvdual varables and the jont omsson of several varables. A sgnfcance level of 10% was used for all these tests. 35 As a result of the GTS method, some ndvdual lags of each varable (.e., current or oneyear lag) may be omtted from each fnal specfcaton ( specfc model ). In general, more varables are omtted usng OLS than usng GMM, due to the greater precson n the latter estmates. For reasons of space, we wll focus our presentaton of the OLS results on the summed coeffcents for all ncluded lags of each varable (detaled results ncludng the ndvdual coeffcents for the current and one-year lags are contaned n the statstcal appendx, whch s avalable on request). Smlarly, for the GMM results, the coeffcents reported are the longrun (n a tme-seres sense) coeffcents,.e., the sums of the coeffcents of the ncluded lags dvded by one mnus the coeffcent on the lagged dependent varable

26 The OLS regresson results are shown n Table 4 for the estmates usng relatve export prces and producer prces and n Table 5 usng CPI-adjusted real exchange rates. The jont sgnfcance of the fxed effects s tested usng the redundant fxed effects test, whch shows that they are sgnfcant n all panels n both tables at the 1% level. 37 These estmates confrm two of the key assumptons of our theoretcal model: that short-run growth rates n our sample of developng countres depend postvely on the growth of total expendtures by the ndustralzed countres on manufactured mports from those developng countres ( Ẑ ) and on the growth rate of the rato of net foregn captal nflows to GDP ( Fˆ ). These two varables have postve and sgnfcant sums of coeffcents for almost all of the panels and usng both prce measures (n two panels, one usng each prce measure, the postve effects of GTS, but n no case are they negatve). [Insert Tables 4 and 5 about here] Ẑ are nsgnfcant or omtted by In addton, the OLS results generally support the COD hypothess: the effect of the relatve export prce or real exchange rate change wth respect to the ndustralzed countres ( Rˆ ) s negatve n almost every panel n both Tables 4 and 5 (the only exceptons are the HIDEBT panel n Table 4 and the LARGE panel n Table 5, for whch both the current and lagged coeffcents were elmnated by GTS), and sgnfcant n the vast majorty of these regressons. However, the FOC hypothess does not receve much support from the OLS regressons. The sum of coeffcents on the relatve export prce wth respect to competng L developng countres ( ) s postve and sgnfcant only for the HIMFRGDP panel n Table 4, Rˆ PX and the sum of coeffcents on the real exchange rate wth those same countres ( ) s postve and sgnfcant only for the SMALLOPE and HIDEBT panels n Table 5, and these summed coeffcents are negatve or nsgnfcant (or elmnated by GTS) n all other equatons. Some of L RˆCPI 23

27 the results of the OLS estmaton may appear anomalous, 38 but as wll be seen below these results change dramatcally when we shft to the GMM estmaton. Consderng the model n equaton (1), the three varables that are most lkely to be endogenous are the two relatve prces (because both Rˆ and L Rˆ have home country export prces or CPIs n the denomnator) and net captal nflows Fˆ. It s plausble, for example, that more rapd growth n any gven country would put upward pressure on ts own prces or attract more captal nflows. Furthermore, captal nflows may be correlated wth exchange rate changes. We therefore ran Hausman weak exogenety tests for the null hypothess that changes n the home country prce pˆ (as measured ether by an export unt value or the CPI) and net fnancal nflows Fˆ are jontly exogenous wth respect to output growth, Y ˆ. The p-values for ths test are and usng export prces (unt values) and real exchange rates (CPIbased), respectvely. 39 Ths means that we can reject the null hypothess of weak exogenety at the 5% level for each relatve prce measure and net captal nflows. Both a pror reasonng and the Hausman tests thus rase serous concerns about the OLS estmates dscussed above. Therefore, and also n order to be able to nclude the lagged dependent varable for reasons noted earler, we turn to the GMM approach whch addresses these concerns. We used second and thrd lags of the dependent varable and lagged nstances of the regressors as the nstruments. Perod SUR (seemngly unrelated regresson) weghted matrces were used to correct for both perod heteroskedastcty and general correlaton of observatons wthn cross-sectons. Orthogonal devatons were used to remove ndvdual specfc effects. The Sargan test was used to verfy the valdty of the overdentfyng assumptons. Under the null hypothess that the overdentfyng restrctons are satsfed, the test statstc s asymptotcally χ 2 wth degrees of freedom equal to the number of overdentfyng 24

28 restrctons, and the results ndcate that those restrctons are not rejected n any panel. The estmaton results usng GMM are gven n Tables 6 and 7, usng our two alternatve sets of prce measures. In the GMM estmates, the coeffcents on expendtures Ẑ and net captal nflows Fˆ are unformly postve and are sgnfcant n almost all panels. The COD effect (negatve coeffcent on Rˆ ) s sgnfcant n all but one panel n Table 6 and all panels n Table 7. Usng the CPI-adjusted real exchange rates (see Table 7), the HIDEBT panel has a notably greater (.e., more negatve) COD effect than the LODEBT panel, as we would expect snce a deprecaton ncreases the burden of servcng a foregn currency-denomnated external debt. 40 [Insert Tables 6 and 7 about here] Usng GMM, the FOC hypothess (postve coeffcent on L Rˆ ) s supported n a majorty of the regressons usng both prce measures. Exceptons nclude the ALL, LOMFRGDP, and HIDEBT panels usng relatve export prces and producer prces (Table 6), and the HIMFRGDP, HITECH, and LODEBT panels usng the real exchange rate (Table 7). Both the LARGE and SMALLOPE panels exhbt sgnfcant postve FOC effects usng ether set of prce measures, but (as we would expect) the coeffcents on ether measure of L Rˆ are greater for SMALLOPE. The FOC hypothess s supported usng ether set of prce measures for the LOTECH countres, whch s consstent wth our expectatons based on the dea that these countres export the most commodtzed manufactures. The unstable results for the panels grouped by the rato of manufactured exports to GDP, where the sgns for the prce varables reverse when we change prce measures, suggest that perhaps ths s not a useful way to dvde the countres. 41 As for the HITECH countres, they are found to experence postve FOC effects usng relatve export prces, but not usng the CPI-adjusted real exchange rate. The HIDEBT countres are found to have sgnfcant FOC effects usng the real exchange rate (Table 7) but not usng the relatve 25