The Dynamic Implications of Foreign Aid and Its Variability

Transcription

1 WP/05/119 he Dynamic Implicaions of Foreign Aid and Is Variabiliy Crisina Arellano, Aleš Bulíř, imohy Lane, and Leslie Lipschiz

2 2005 Inernaional Moneary Fund WP/05/119 IMF Working Paper IMF Insiue and Policy Developmen and Review Deparmen he Dynamic Implicaions of Foreign Aid and Is Variabiliy Prepared by Crisina Arellano, Aleš Bulíř, imohy Lane, and Leslie Lipschiz 1 June 2005 Absrac his Working Paper should no be repored as represening he views of he IMF. he views expressed in his Working Paper are hose of he auhor(s) and do no necessarily represen hose of he IMF or IMF policy. Working Papers describe research in progress by he auhor(s) and are published o elici commens and o furher debae. he paper examines he effecs of aid and is volailiy on consumpion, invesmen, and he srucure of producion in he conex of an ineremporal wo-secor general equilibrium model. A permanen flow of aid finances mainly consumpion, a resul consisen wih he hisorical failure of aid inflows o ranslae ino susained growh. Shocks o aid are refleced mainly in invesmen flucuaions, as a resul of consumpion smoohing. Aid shocks resul in subsanial welfare losses, suggesing ha aid variabiliy should be aken ino accoun in designing aid archiecure. hese resuls are consisen wih he evidence from cross-counry regressions of manufacured expors. JEL Classificaion umbers: C68, F35, F41 Keywords: Real business cycle, general equilibrium, aid, ransfer problem Auhor(s) Address: arellano@econ.umn.edu; abulir@imf.org; lane@imf.org; llipschiz@imf.org 1 his paper was mainly prepared while he firs auhor was visiing he Policy Developmen and Review Deparmen; she is an Assisan Professor a he Universiy of Minnesoa. We are indebed o Rober Flood, Caherine Paillo, and Mark Plan for commens, as well as o paricipans in a seminar a he Inernaional Moneary Fund.

3 - 2 - Conens Page I. Inroducion...3 II. he heoreical Model...5 III. Simulaions and Resuls...10 A. Calibraion for he Benchmark Model...10 B. Seady Sae Analysis...11 C. Sensiiviy Analysis...15 D. Sochasic Model...17 IV. Some Empirical Resuls...21 A. Explaining Expors...21 B. Regression Resuls...24 V. Issues for Furher Research...30 VI. Conclusions...31 References...32 ables 1. Large Recipiens of Aid, Seady Sae Simulaions Sensiiviy Analysis of he Seady Sae Resuls Properies of Model Economies Definiion of Variables and heir Sample Momens Marix of Correlaion Coefficiens Aid, Endowmen, ransacion Cos, and Manufacured Expors Sensiiviy Analysis: How Imporan Are he Individual Conribuing Facors?...29 Figures 1. Capial ransformaion Curve Seady Sae Variables as a Funcion of Aid Deerminans of Expors of Manufacured Goods, Appendix ables A1. he Parameers Used in he Baseline Model...37 A2. Seady Sae Simulaions..38 A3. Lis of Counries Used in Regressions...40 Annex Long-Run Equilibrium Soluion

4 - 3 - I. IRODUCIO Foreign aid remains an imporan fac of life for developing counries: many counries, especially he poores, rely on aid as a major exernal source of income. I is paricularly imporan o consider he macroeconomic implicaions of hese inflows o he exen ha hey may eiher suppor or impede he effors of low-income counries o foser heir own sources of economic growh. hese implicaions are expeced o become all he more imporan as donors increase aid flows o suppor hese counries progress oward he Millennium Developmen Goals. A saring poin for considering he impac of aid inflows is he ransfer problem: how is an inernaional ransfer effeced hrough an adjusmen of he exernal rade balance? his is an issue of real resource ransfers ha is relaed o, bu quie disinc from, he availabiliy of financing. Economic analysis of he ransfer problem yields sraighforward conclusions: effecing he ransfer ha is, achieving he required rade defici generally enails a combinaion of higher impors and lower expors associaed wih a real exchange rae appreciaion. 2 Anoher applicaion of he ransfer problem is so-called Duch disease he adverse effec of naural resource revenues on he manufacuring secor, associaed wih a real appreciaion, such as was experienced following he discovery of naural gas in he eherlands (Corden, 1984 or Gelb, 1988). A number of observers have poined ou ha he logic of Duch disease applies equally o he effec of large aid inflows in low-income counries. Foreign aid increases he supply of radable goods and, ceeris paribus, lowers heir price, while hrough he income effec of he ransfer increasing he demand for and price of nonradable goods. As a resul, facors of producion are redireced oward he secors producing nonradable goods. 3 his heoreical observaion has given rise o a lieraure examining he empirical significance of Duch disease, i.e., examining he exen o which aid is effeced via a decrease in expors raher han via an increase in impors (see a review in Bulíř and Lane, 2004). A second quesion is how he basic logic of he ransfer problem, usually developed in a saic wo-good model, carries over o he dynamic conex in which ransfers can also affec savings and capial accumulaion. his quesion is relaed, in urn, o he empirical quesion of how aid affecs invesmen: i has radiionally been argued ha aid may boos growh 2 he ransfer problem can be raced back o a discussion of Germany s war reparaions by Keynes (1929) and Ohlin (1929). Samuelson (1952) who was concerned wih he implicaions of he Marshall plan was he firs o frame he problem in a neoclassical apparaus. For a recen review, see Yano and ugen (1999). 3 See, for example, Michaely (1981) and Laplagne, readgold, and Baldry (2001) for simple saic models of his process. While he impac on invesmen is relaively rivial in hese models, he impac on labor depends on labor mobiliy and migraion, see Harris and odaro (1970) and Corden and Findlay (1975).

5 - 4 - because i supplemens he limied supply of domesic saving available for invesmen; bu alhough some empirical evidence suggess ha invesmen increases wih aid (for example, Hansen and arp, 2001 or Clemens, Radele, and Bhavnani, 2004), his evidence is far from conclusive (Easerly, 1999). A hird quesion is why or indeed, wheher Duch disease maers: he ransfer problem is generally analyzed in he conex of models in which he resuling decline in producion of radables is an opimal adjusmen o he ransfer, so ha Duch disease is no really a disease a all. However, discussions of his phenomenon in he conex of low-income counries and developmen usually refer o he imporance of he expor indusries for growh (Michaely, 1981). Indeed, a number of heoreical sudies have elaboraed he idea ha rade can be he engine of growh for developing counries hrough echnological diffusion and learning by doing (Grossman and Helpman, 1991; Barro and Sala-i-Marin, 1997; Connolly, 1999; and Bigsen e al., 2002). Empirical papers have also found a posiive relaion beween rade and growh (Levine and Renel, 1993 or Sala-i-Marin, 1997). Bu such analyses wih he excepion of Adam and Bevan (2003) have no generally been inegraed ino he analysis of he decline in he radables secor in response o susained aid inflows. Fourh, a salien aspec of aid, which has recenly been receiving increasing aenion, is is variabiliy. Evidence indicaes ha aid flows are volaile, reflecing he vagaries of donors budge allocaions, donor condiionaliy, and oher facors (Pallage and Robe, 2001; Bulíř and Hamann, 2003; and Bulíř and Lane, 2004). his evidence makes i imporan o examine how aid volailiy affecs boh he level and variabiliy of key macroeconomic variables. Lile is known, however, abou long-erm oupu and welfare effecs of aid and is volailiy (Pallage and Robe, 2003, and urnovsky and Chaopadhyay, 2003). his paper examines he dynamic implicaions of aid and is variabiliy in he conex of a simple ineremporal wo-secor opimizing model, akin o hose used in he lieraure on real business cycles (RBC). he simplifying assumpions of he model leave aside hree imporan issues: firs, he composiion of aid and is fungibiliy, as well as any associaed condiionaliy; second, any aid-driven learning-by-doing effecs, and, hird, any domesic poliical economy effecs associaed wih ren-seeking behavior of aid recipiens. he model is calibraed based on plausible parameer values, comparable o hose used in previous lieraure. he resuls sugges, firs of all, ha he Duch disease effecs indeed carry over o a dynamic seing, wih some differences. Second, he model characerizes he response of aid on consumpion and invesmen: a consan, predicable sream of aid is refleced primarily in consumpion, while he effecs of shocks o aid are disribued beween consumpion and invesmen in proporions ha depend on he shape of he underlying uiliy and producion funcions. hird, aid variabiliy of he magniude found in previous lieraure may have subsanial derimenal welfare effecs, albei no large enough o wipe ou he welfare benefis of he aid iself.

6 - 5 - he paper also presens empirical resuls on he relaionship beween foreign aid and he producion of goods for expor. Cross-counry regressions for 73 aid-dependen counries indicae ha manufacured good expors, as prediced by he heoreical model, are negaively relaed o he level of aid. he resuls are significan afer conrolling for iniial endowmens, ransacion coss, he level of developmen, and a hos oher variables. he paper is organized as follows: Secion II develops he model and presens resuls for he numerical simulaions, Secion III presens regression resuls on he aid-o-expors nexus, Secion IV summarizes issues for furher research, and Secion V concludes. II. HE HEOREICAL MODEL he model used in his paper is a neoclassical dynamic general equilibrium model of a small economy ha has wo producive secors: radable and nonradable goods. he model displays ransiional growh driven by he accumulaion of capial, bu i is simplified in several imporan respecs. Firs, aid is assumed o be provided uncondiionally in he form of a gran. 4 Second, he recipien counry is assumed o have no access o inernaional capial markes, so ha invesmen in capial goods is he only available sore of value. hird, he model assumes simplified consumpion behavior, wih a represenaive household maximizing an ineremporal uiliy funcion characerized by consan elasiciy of subsiuion. Fourh, i does no model produciviy spillovers associaed wih he expansion of he radable goods secor. Fifh, he model also ignores he poliical economy effecs of aid (Bauer, 1979 or ornell and Lane, 1999), absorpion-capaciy consrains, or aid-relaed coss of invesmen (Asilis and Ghosh, 2002). his simplified model noneheless yields ineresing analyical resuls, which may provide a benchmark for oher analyses ha would move he model beyond hese limis. Households in he economy maximize heir expeced lifeime uiliy and have preferences over radable and nonradable goods: U = E 0 = 0 1 σ C β, 1 σ where σ is he coefficien of risk aversion and he consumpion funcion is of consan elasiciy of subsiuion (CES) form 1 µ µ [ ω( C ) + (1 ω) ( ) ] µ C, ( 1 ) C = ( 2 ) where ω is he weigh households place on radable consumpion (C ) and (1- ω) is he weigh on nonradable consumpion (C ). he elasiciy of subsiuion of consumpion beween radables and nonradables is 1 /(1 + µ ). I could be argued, however, ha 4 his means ignoring a large lieraure on he implicaions of aid condiionaliy (Killick, 1997, Svensson, 2000, and Mayer and Mourmouras, 2002).

7 - 6 - consumpion in poor counries may be inflexible a he subsisence level and ha consan relaive risk aversion (CRRA) preference used in his paper may underesimae he welfare losses. his is one of he issues lef for furher research. Households receive labor income, ren capial o firms, and make invesmen decisions. In addiion, he economy receives a sochasic ransfer of radable goods, X, ha is, aid. he budge consrain in radable good erms is as follows: C + p C = r K + w L i + X, ( 3 ) where p is he relaive price of nonradables in erms of radables (i.e., he real exchange rae), K is he economy s capial sock, i is invesmen (which, for he sake of simpliciy, we assume is in he form of radable goods only), w is he real wage rae paid o labor inpu, L is he labor endowmen se o 1, r is he real domesic ineres paid on capial, and x X = X exp( ε ) is a sochasic amoun of aid inflows given o he economy. Here, an imporan underlying assumpion is ha he counry has no access o inernaional capial markes, so ha he only mechanism for saving is invesing in he domesic capial sock. Capial follows he usual accumulaion process: and depreciaes a rae δ. K = 1 i + (1 + δ ) K ( 4 ) Firms in boh secors are compeiive, choose labor and capial o maximize profis, and produce oupu wih a Cobb-Douglas, consan reurns o scale echnology: 1 exp( ) α ε α, Y = A K L ( 5 ) Y = A K L ( 6 ) 1 exp( ) η ε η. Boh secors are subjec o produciviy shocks, ε, ε. Firms and households have he same informaion se: hey know he disribuion of boh he produciviy and aid shocks. However, households canno insure perfecly agains negaive shocks because asse markes are incomplee and he only asse available for ineremporal smoohing is domesic capial. Labor is perfecly homogenous and mobile across secors: L = L + L = 1. ( 7 ) Capial, however, is assumed o be somewha secor-specific, in he sense ha capial becomes less effecive as more of he exising capial sock is allocaed o one secor. his assumpion is capured by he facor ransformaion curve (Mendoza and Uribe, 2000): K = κ( K, K ), ( 8 )

8 - 7 - where ( ) subsiuion beween κ is assumed o be a CES funcion, K = [ K + K ] ν K and capial is he special case where = 1 K + ν, wih he elasiciy of ξ = 1/ 1+ ν wih ν 1. Perfecly homogenous ν. Figure 1 shows he cos κ ( ) for a given level of K being ( ) K (which for simpliciy is se o 1) as a funcion of he relaive allocaion of capial o he nonradable goods secor. he producion possibiliy fronier is concave, owing o differences in facor inensiies in he wo secors as well as o he curvaure of he aggregae capial sock as given by κ ( ). In equilibrium, he slope of he producion possibiliy fronier is equal o he relaive price of nonradables, which in urn is equal o he marginal rae of subsiuion beween radables and nonradables. ν 1 1 Figure 1. Capial ransformaion Curve 1/ Capial ransformaion Cos Kapa Kn/(K+Kn) 1/ he cos of reallocaing capial from one secor o anoher, κ(.), is shown as a funcion of he relaive allocaion of capial o he nonradable secor. he relaive price of nonradables is deermined by relaive echnologies and he relaive secoral capial socks. Firms in he wo secors hire labor and ren capial from he households so ha in equilibrium he wage rae equals he marginal produciviy of labor and he rae of reurn equals he marginal produciviy of capial. Since capial is secor-specific, he effecive rae of reurn in each secor incorporaes he degree of facor subsiuabiliy beween he wo secors given by he derivaive of κ wih respec o he secoral capial. In equilibrium, marginal produciviies across secors are equalized: w A exp( )(1 )( K L ) α η = ε α = p A exp( ε )(1 η)( K L ), ( 9 )

9 - 8 - A exp( ε ) α( K L ) A exp( ε ) η( K L ) r = =. ( 10 ) α 1 η 1 p κ1( K, K ) κ2( K, K ) In his seing, aid affecs he wage rae and he rae of reurn on capial by affecing he relaive levels of labor and capial used in he wo secors. he effecs depend on he relaive facor inensiies of he wo secors, as refleced in he parameers α and η (he Solper- Samuelson effec). Here, we focus on he case in which radables are relaively laborinensive (α < η): increased aid raises he demand for capial-inensive nonradables, raising he economy-wide reurn o capial, ha is, he equilibrium real rae of ineres, r. he marke clearing condiions for he wo secors are: and C = Y ( 11 ) C + i = Y + X. ( 12 ) he firs order condiions of he maximizaion problems of households and firms can be combined in he following se of equilibrium condiions: and (1+ η ) 1 ω C p = ω ( 13 ) C where σ σ α 1 C C 1 A exp( ε 1) α( K 1 L 1) = β E 1, C C + δ p p+ 1 κ1( K+ 1, K+ 1) C p is he CES price index for aggregae consumpion: ( 14 ) 1+ η η 1 1 η C 1+ η 1+ η p = ω + (1 ω) ( p ) 1+ η (15 ) Equaion (13) equaes he marginal rae of subsiuion beween he consumpion of radable and nonradable goods o he relaive price, p, and equaion (14) corresponds o he Euler equaion ha equalizes he marginal cos of sacrificing a uni of curren consumpion wih he marginal benefi of allocaing he resuling exra savings ino he aggregae capial.

10 - 9 - he compeiive equilibrium of his model is defined as he sae coningen sequences of allocaions and prices { C }, C, K, K, K + 1, L, L, p, r, w such ha (i) households = 0 maximize uiliy subjec o heir budge and ime consrains aking prices as given, (ii) firms maximize profis subjec o heir echnology aking inpu prices as given, and (iii) markes clear. he sequence of decisions in he economy is as follows. A ime households and firms realize he produciviy and aid shocks, make invesmen decisions, and choose consumpion and capial and labor in each secor. Prices are deermined a his ime. he invesmen decisions are based on he expeced rae of reurn ha households foresee for he nex period. A +1 he uncerainy is resolved and he rae of reurn on he previous period s invesmen is realized. I is worh noing ha alhough he acual rae of reurn is equalized across secors, he level a which i is equalized was uncerain when invesmen was allocaed. he equilibrium real exchange rae (16) is a funcion of he relaive echnologies of he wo secors, as well as of he secoral allocaions of capial. 5 he model incorporaes he Balassa- Samuelson hypohesis ha variaions in he real exchange rae come from labor produciviy differenials, modified for he possibiliy of secor-specific capial: 1 α α K A 1 α α α K ln p = ( α η)ln ln ln. + ( 16 ) L A 1 η η ξ K In addiion o equaion (16), which is essenially a supply side condiion coming from equalizaion of marginal produciviies in he wo secors, he real exchange rae is subjec o a demand side condiion, equaion (13), coming from equalizaion of he marginal rae of subsiuion beween consumpion of he wo goods. Aid affecs he real exchange rae hrough he level of radable consumpion as well as hrough he raio of he capial sock of he wo secors. 6 An increase in aid increases radables consumpion, lowering he relaive price of radable goods, however, he resuling appreciaion of he real exchange rae is aenuaed by he increase in he relaive share of capial used in he nonradable secor. he soluion of he model predics wha porions of aid are consumed and invesed, and how he oucome is affeced by aid volailiy. he resuls depend, in par, on he elasiciy of subsiuion of capial in he wo secors. 5 We combine equaions (9) and (10) and ake logs. 6 Permanenly higher aid would increases consumpion permanenly wih consan aid he agens would need no addiional (or reduced) invesmen. Shocks o aid affec invesmen, bu only o he exen ha he resuling diminishing marginal rae of ransformaion of capial ineracs wih he increasing marginal rae of subsiuion of consumpion. hus, he magniude of he effec depends on he specificaions of he model, such as aversion o ineremporal variaions of consumpion or he scope for fungibiliy of capial across secors.

11 III. SIMULAIOS AD RESULS A. Calibraion for he Benchmark Model he numerical analysis sars from a baseline scenario in which he model s parameers are calibraed o mimic some of he characerisics of poor, aid-dependen counries, or aken from oher developing-counry business cycle sudies. he model is solved numerically by value funcion ieraion. he long-run soluion is described in he Annex, and able A1 summarizes he parameers used in he baseline simulaions. For he preference parameers, he elasiciy of subsiuion beween radable and nonradable consumpion, 1/(1 + µ ), is aken from Osry and Reinhar (1992), where heir esimae for µ is equal o he risk aversion coefficien σ is se o 5 (Reinhar and Vegh, 1995), and he ime-preference parameer is se o 0.95 which is a sandard value in RBC sudies in developing counries. he weigh of radable consumpion ω is aken from he weighs of radable goods in he CPI indexes, where we found he share of radable goods o be roughly equal o 0.5. he nonradable secor is assumed o be more capial inensive ha he radable secor: he capial shares in he radable and nonradable secors, α and η, are equal o 0.3 and 0.4, respecively. wo sylized facs suppor our choice of calibraion parameers: firs, exporers in developing counries specialize ypically in labor-inensive, low-skill echnologies (foowear, apparel, and so on) and, second, mos nonradable infrasrucure projecs in hose counries are highly capial inensive (elecriciy, elecommunicaions, and so on), see, for example, Brock and urnovsky (1994) and Goldsein and Lardy (2005) for his argumen. he elasiciy of subsiuion beween capial used in he radable and nonradable secors, 1/(1 + ν ) is se o 10. he rae of depreciaion of capial δ is se o 0.09 so ha in seady sae he invesmen rae is equivalen o 18 percen of GDP, an average of long-erm invesmen raes aken from he World Economic Oulook. he shocks are se o replicae broadly he sochasic srucure of produciviy and aid in aid-dependen counries. Firs, in our baseline model we assume ha produciviy shocks affec he radables and nonradables secors equally, ε = ε, wih a sandard deviaion of 10 percen and a firs order auocorrelaion coefficien of 0.4. his seup implies ha 60 percen of any deviaion of produciviy from is mean level dissipaes in each period. For comparison, his calibraion of he produciviy process produces a sandard deviaion of GDP of abou 10 percen and of radable oupu of 15 percen. Second, he properies of aid flows in some of he heavily aid-dependen counries are summarized in able 1. Our simulaions were no designed o mach economic performance of specific counries, insead, our calibraions are based on he sample of aid-dependen counries lised above. Accordingly, he mean of aid inflows is se equal o 25 percen of seady sae GDP; he volailiy is se a 15 percen of ha consan mean level. Aid is ypically mildly procyclical, i.e., i ends o decline in hose years when GDP is lower, and vice versa. hus he correlaion coefficien beween he produciviy and aid shocks is se o equal o 0.4.

12 Counry able 1. Large Recipiens of Aid, Aid-o-GI raio (In percen) Commimens-odisbursemens raio Relaive volailiy of aid1/ Cyclicaliy of aid 2/ Bhuan Comoros Gambia, he Guinea-Bissau Guyana Malawi Mozambique icaragua Rwanda anzania Zambia Average Median Source: World Developmen Indicaors; World Economic Oulook; Bulíř and Hamann (2005). 1/ Raio of variances of derended aid and fiscal revenue, Bulíř and Hamann (2005). 2/ Correlaion coefficien of aid and fiscal revenue, Bulíř and Hamann (2005). B. Seady Sae Analysis In his par of he analysis we focus on he seady sae, in which he capial sock has converged o a consan long-run level while he counry receives a consan level of aid in he absence of produciviy shocks. his seady sae generalizes o an ineremporal seing he implicaions of aid ha arise in a simple saic model. I also provides a reference poin for he analysis of he implicaions of shocks o produciviy and aid, which will be developed in he nex secion. We find ha wih more aid he marginal rae of saving remains very low, invesmen declines in relaion o oal income, he real exchange rae rises, and radable oupu shrinks. he seady sae allocaions shown in able 2 are calculaed for four alernaive levels of aid: 0 percen, 10 percen, 25 percen and 50 percen of GDP, where GDP is defined as a sum of radable and nonradable goods, expressed in raded good erms, GDP = Y + p Y. he impac of aid changes he composiion of oupu, labor shares, invesmen raes, and consumpion, as well as he relaive price of nonradable goods.

13 able 2. Seady Sae Simulaions 1/ (Seady saes resuls for various aid levels) Aid Inflows (in percen of GDP) 0 percen 10 percen 25 percen 50 percen radable oupu-o-gdp 2/ onradable oupu-o-gdp 2/ Relaive price of nonradable goods radable consumpion-o-gdp oal consumpion-o-oal income 3/ Invesmen-o-GDP 2/ Invesmen-o-oal income 3/ Capial-o-GDP 2/ Raio of capial sock in he radable o nonradable secor Raio of labor sock in he radable o nonradable secor Source: Auhors calculaions. 1/ For hese simulaions, produciviy and aid shocks are se o nil. 2/ Aggregae oupu is defined as GDP = Y + p Y, using radables as numéraire. hus he raio of nonradables oupu o GDP, p Y / GDP, depends boh on physical producion of nonradables and on nonradable price, p. 3/ oal income equals o GDP and aid.

14 Figure 2. Seady Sae Variables As a Funcion of Aid 1/ onradable price 1.1 radables secor capial percen 10 percen 25 percen 50 percen percen 10 percen 25 percen 50 percen radables secor labor Invesmen-o-oal income percen 10 percen 25 percen 50 percen percen 10 percen 25 percen 50 percen radable consumpion radable oupu percen 10 percen 25 percen 50 percen percen 10 percen 25 percen 50 percen Source: Auhors' calculaions. 1/ he verical axis indicaes he seady saes values of he variables. he horizonal axis shows he differen seady sae aid values (as percenage of GDP).

15 One sriking resul of he model is ha a permanenly higher level of aid is associaed wih a permanenly higher level of consumpion, wih lile incremenal effec on saving and invesmen. While he invesmen-o-gdp raio increases from 21 percen o 23 percen wih aid rising from 0 percen of GDP o 50 percen of GDP, invesmen declines sharply as a raio o oal income, ha is, GDP plus aid (Figure 2). Indeed, he marginal rae of saving is close o zero. Moreover, as we show in he sensiiviy analysis secion, his resul is robus o changes in he key parameers of he model. his finding reflecs he logic of opimal consumpion and invesmen behavior where capial goods are he only sore of value and he arge level of capial is independen of he level of aid. In he seady sae, consumpion increases slighly in relaion o oal income from 82 percen o 88 percen wih aid rising from 0 percen of GDP o 50 percen of GDP. Addiional aid does no affec he consumpion profile, bu simply shifs up he consumpion level, while keeping ha higher level of consumpion consan over ime. he arge capial sock is deermined by he ime preference parameer and he depreciaion rae and hus i remains unchanged and all he aid is consumed. Saving and invesmen decisions are affeced by aid only o he exen ha aid leads o ineremporal variaions in consumpion. A higher consan level of aid does no give rise o any such variaion, so here is lile difference in invesmen as a percenage of GDP for differen aid levels. he reason why invesmen as a percenage of GDP is higher is ha producion of nonradables is relaively capial-inensive; he rae of reurn o capial is higher han in he absence of aid, providing a greaer incenive for invesmen. 7 A second resul, in line wih he classical ransfer paradox, is ha he relaive price of nonradables is higher in he case where aid inflows are higher. In he simulaion resuls, aid equal o 50 percen of GDP is associaed wih a relaive price of nonradables higher (i.e., real appreciaion) by one hird han in he absence of aid. his change can be viewed on boh he demand and he cos side: aid increases he availabiliy of radables relaive o nonradables, raising he equilibrium price of he laer; a he same ime, i pushes up he reurns o capial, he facor assumed o be used inensively in nonradable producion, hus increasing he relaive cos of producing nonradables. A hird resul, semming from he firs wo, is ha he relaive size of he radable secor is subsanially smaller wih higher levels of aid. Wihou aid he radable secor comprises more han 60 percen of GDP in he seady sae; is share shrinks o abou 36 percen of GDP when aid is equivalen o 50 percen of GDP. he relaive decline of he radable secor is almos linear: a 1 percenage poin increase in aid is associaed wih abou a ½ percenage poin decline in he share of he radable secor in GDP. Alhough overall GDP (measured in erms of radables) is larger wih aid, a par of he increase in GDP is due o higher nonradable prices, p, raher han addiional physical producion of nonradables. he 7 his resul is an inversion of he Rybczynski heorem: an increase in a counry s endowmen of a facor will cause an increase in oupu of he good which uses ha facor inensively.

16 relaively higher oupu of he nonradable secor is also associaed wih a higher share of labor: he raio of labor in he radables o he nonradables secor decreases from 1.90 wihou aid o 0.66 when aid is equivalen o 50 percen of GDP. he allocaion of capial in he wo secors changes much less owing o he assumpion ha capial is secor-specific: he raio of capial in he radables o he nonradables secor declines from 1.02 o 0.92 beween he wo poles of he aid coninuum. 8 Wih aid, households also change he consumpion bundle oward radable goods consumpion, reflecing he increase in he relaive price of nonradables: radables consumpion increases from 40 percen of GDP o more han 60 percen of GDP. While hese resuls are in line wih he saic model findings, we observe some new feaures. Firs, aid is associaed wih a decline in he invesmen rae as households rely on aid inflows as opposed o invesmen for consumpion smoohing and we do no observe any significan increase in aid-induced invesmen. Second, he resuls are sensiive o he subsiuabiliy of secoral capial. We will explore hese findings furher in he subsequen secions. C. Sensiiviy Analysis In his secion we perform a series of sensiiviy ess on he deerminisic model o undersand he mechanisms by which aid changes he composiion of oupu. In hese experimens we vary he level of aid, he elasiciy of subsiuion of secoral capial, and he capial inensiy of he wo secors. All oher parameers are held consan. In able 3 we show four ses of resuls: he baseline model of he previous secion; a model wih idenical capial shares in boh secors and perfecly subsiuable capial (Model A); a model wih idenical capial shares in boh secors bu imperfecly subsiuable capial (Model B); and a model wih differen capial shares in boh secors bu perfecly subsiuable capial (Model C). he full se of numerical resuls is shown in Appendix, able A2. We find ha one of he key resuls a decline in radables oupu remains inac in all experimens, while he oher real appreciaion remain inac as long as capial is secor specific. Firs, radables oupu diminishes even in he absence of real appreciaion as aid subsiues for domesic radables (all models). his follows direcly from he fac ha aid is a gif of radable goods: even if higher supply of radables does no affec he price of nonradables, because capial is seamlessly reallocaed oward he producion of nonradables, i sill implies ha demand for radables was saisfied from aid proceeds. 8 Given he assumpion ha nonradables are relaively capial inensive, an increase in he relaive size of ha secor requires a lower capial inensiy in boh secors, associaed wih a rise in he reurns o capial.

17 able 3. Sensiiviy Analysis of he Seady Sae Resuls 1/ Baseline Model 2/ Model A 3/ Model B 4/ Model C 5/ onradable prices unchanged unchanged radable oupu-o-gdp Raio of capial sock in he radable o nonradable secor oal capial unchanged Invesmen-o-GDP unchanged unchanged Invesmen-o-oal income GDP unchanged 1/ indicaes ha he variable increases as he economy receives more aid. 2/ A model economy wih all parameers as in able A1, ha is α = 0.3, η = 0.4, and ν = 10. 3/ A model economy wih idenical capial shares and perfecly subsiuable capial ( α = η = 0.3 and ν = 1). 4/ A model economy wih idenical capial shares and imperfecly subsiuable capial ( α = η = 0.3 and ν = 10 ). 5/ A model economy wih differeniaed capial shares and perfecly subsiuable capial ( α = 0.3, η = 0.4, and ν = 1). Second, he increase in he relaive price of nonradables depends on imperfec subsiuabiliy of capial: if he capial sock were freely inerchangeable beween secors, no real appreciaion would occur (Model A and C). his is inuiive: labor would be reallocaed oward nonradable goods and he greaer capial inensiy of non-radables would be refleced in a higher seady-sae capial sock. Furhermore, he lower he elasiciy of subsiuion beween capial employed in he radables and nonradables secors, he larger he relaive decline in radables oupu. he argumen is as follows: when capial is highly secor-specific, he aid-induced shif in he composiion of oupu oward nonradables akes place in a less efficien way. I is more cosly o reallocae capial, and more of he burden hus falls on a reallocaion of labor. Alhough he oupu of radables declines less when capial is secor-specific, he price of nonradables increases subsanially, boosing GDP (Baseline model and Model B). hird, he srucure of he economy also affecs he seady sae sock of capial and GDP. oal capial is unchanged if secoral capial is freely inerchangeable and capial inensiies are idenical (Model A), bu i increases in he oher versions of he model. In economies where he share of capial in oupu is idenical (Model A and B) he invesmen-o-gdp raio

18 does no change. Moreover, we find ha oupu increases less when he elasiciy of subsiuion beween radable and nonradable capial is high and indeed oupu is unchanged in Model A. D. Sochasic Model In his secion we develop he case in which he economy may be subjec o produciviy shocks and aid receips may also be uncerain and compare i wih he seady sae in he absence of any shocks (able 4, firs column). 9 Developing counries are subjec o srong produciviy flucuaions and in many of hem aid consiues a significan share of income and is poenially an imporan addiional source of volailiy. Developing counries limied financial inegraion resrics heir abiliy o insure agains hese shocks inernaionally, so he shocks may have paricularly imporan economic implicaions (Mendoza, 1997, Pallage and Robe, 2003, or urnovsky and Chaopadhyay, 2003). Sochasic produciviy and fixed aid I is useful o consider firs a case in which here are produciviy shocks and a fixed level of aid, assumed equal o 25 percen of seady sae GDP (second column of able 4). he produciviy shocks are assumed o be somewha persisen, i.e., a posiive shock o oal facor produciviy his period implies higher oal facor produciviy nex period and, o a lesser degree, in subsequen periods. In his seing, invesmen varies subsanially in response o produciviy shocks due o he absence of capial adjusmen coss: for insance, a negaive shock lowers invesmen boh for consumpion smoohing reasons and because of he negaive effec of he shock on he fuure marginal produciviy of capial. Bu consumpion also varies significanly in response o he shock: given ha capial is he only sore of value and is subjec o diminishing marginal produciviy, consumpion smoohing is no complee. 10 Produciviy shocks also give rise o variaions in producion, owing boh o heir direc effec and o he resuling flucuaions in invesmen. he seady-sae capial sock and invesmen are marginally higher han in he absence of shocks, reflecing he role of capial as a buffer sock. Moreover, volailiy in produciviy resuls in an increase in he average producion of radables relaive o nonradables: his reflecs he model assumpion ha he capial sock can be increased by radable goods only. 9 he model is solved by value funcion ieraion over a discreized sae space. In he case wih shocks o boh produciviy and aid, he sae space corresponds o ε x = {-0.15, 0.15} and ε = ε = {-0.1, 0.1}. he probabiliy disribuion marix is calculaed such ha i replicaes he persisence and correlaion among shocks in he daa. he endogenous sae variable is he aggregae capial sock wih a grid corresponding o 50 equally spaced poins. he poins are chosen so ha hey cover all he posiive probabiliy poins in he limiing disribuion. 10 he limied degree of consumpion smoohing is in par a reflecion of he model s specificaion, i.e., a combinaion of he CES uiliy and Cobb-Douglas producion funcions.

19 able 4. Properies of Model Economies (Means and Sandard Deviaions of Limiing Disribuion) Seady sae 1/ Fixed Aid, Sochasic Produciviy 2/ Sochasic Aid, o Produciviy Shocks 3/ Mean Sochasic Aid and Produciviy 4/ radable oupu-o-gdp 5/ Relaive price of nonradables radable consumpion-o-gdp Invesmen-o-GDP Invesmen-o-oal income Capial-o-GDP Raio of capial sock in he radable o nonradable secor Raio of labor sock in he radable o nonradable secor Sandard Deviaion (In percen) Aid Gross domesic produc Relaive price of nonradables radable oupu onradable oupu radable consumpion CES consumpion Invesmen Labor in radable secor Memorandum Iem: Aid o GDP raio (in percen) Source: Auhors calculaions. 1/ Idenical as in he hird column of able 1. 2/ Produciviy shocks are idenical for boh he radable and nonradable secors. heir sandard deviaion is se equal o 10 percen and he firs order auocorrelaion coefficien is se equal o / o produciviy shocks. he sandard deviaion of aid inflows is se equal o of 15 percen and he aid-o- GDP correlaion coefficien is se equal o / Produciviy shocks are idenical for boh he radable and nonradable secors. heir sandard deviaion is se equal o 10 percen and he firs order auocorrelaion coefficien is se equal o 0.4. he sandard deviaion of aid inflows is se equal o of 15 percen and he aid-o-gdp correlaion coefficien is se equal o / Aggregae oupu is defined as GDP = Y + p Y. Hence, all he GDP raios depend also on he change in prices of nonradables, p.

20 Sochasic aid and no shocks o produciviy ex, one may consider he case in which here are no produciviy shocks bu aid flucuaes around is mean of 25 percen of seady sae GDP (hird column of able 4). On he one hand, as in he previous case, aid shocks are primarily refleced in variaions in invesmen. On he oher hand, variaions in consumpion are much smaller han in he case of produciviy shocks. We see hree reasons for his difference. Firs, aid shocks do no direcly affec producion, second, hey are ransiory and, hird, comparaively small in absolue erms (25 percen of oupu). Give ha nonradables consumpion is supply-deermined, here is lile variabiliy in oupu. We also find ha he share of radables in oal oupu is similar o he baseline case as is he sock of capial. In summary, while he firs momens (means) in hese wo economies are fairly similar, he second momens (sandard deviaions) are quie differen, reflecing he separae channels hrough which shocks o produciviy and aid propagae. Sochasic aid and produciviy Finally, we consider he case where boh aid and produciviy are sochasic (fourh column of able 4). he resuls are in line wih our previous findings, namely ha a more volaile economy encourages invesmen in order o smooh consumpion; and variabiliy in produciviy and aid are largely refleced in variabiliy in invesmen. he average share of radables in oal oupu is somewha larger han in he oher cases, reflecing mainly he effec of produciviy shocks on oupu shares as discussed earlier. 11 Wih a larger capial sock, nonradables oupu is larger as well (alhough as a share of oal oupu i is lower) and nonradables prices are marginally lower. In summary, he level and he volailiy of aid have differen implicaions for he srucure of economic aciviy. Whereas a seady, predicable level of aid is refleced mainly in higher consumpion, shocks o aid are absorbed mainly by variaions in invesmen, wih smaller effecs on consumpion. Moreover, he sock of capial and radable oupu are higher under more volaile condiions. hese effecs reflec he ineracion of several facors: primarily he easy access o invesmen for consumpion smoohing purposes; bu also flucuaions in he real exchange rae; and he allocaion of labor and capial across secors, reflecing boh heir average levels and response o shocks. Welfare implicaions We have shown ha even volaile aid increases he average level of consumpion, however, aid volailiy increases he variabiliy of consumpion, wih negaive consequences for 11 Shocks o aid have very differen effecs on invesmen han a change in he seady-sae level of aid a reminder ha, even if i is esablished empirically ha shor-run changes in aid are correlaed wih changes in invesmen, i does no follow ha a higher seady level of aid provision would resul in a commensuraely higher average level of invesmen.

21 welfare. his makes i paricularly imporan o look a he overall welfare implicaions of aid and is variabiliy. We examine hree cases: one where volaile aid is compared o aid ha is sable wih no produciviy shocks; a second where produciviy is also sochasic; and a hird where aid flows are delivered so as o sabilize consumpion. he shock srucure in he hree scenarios is idenical o he one discussed earlier (able 3) and we use wo merics, welfare coss in he erms of CES consumpion and CES consumpion denominaed in radable goods. he laer meric removes he downward bias caused by nonradables price changes. o ranslae values from CES consumpion ino CES consumpion denominaed in radable goods we use he relaive price of aggregae consumpion (15). he esimaed welfare coss of aid volailiy are comparable o wha was found in previous research on business cycle coss in developing counries. 12 In he firs case no produciviy shocks welfare coss of aid volailiy in erms of CES consumpion and in erms of CES consumpion denominaed in erms of radables are 1 10 of 1 percen and 1 6 of 1 percen, respecively. his is a pure impac of volaile aid in an economy wih no oher shocks. A more realisic welfare calculaion would incorporae also produciviy shocks: in such a case welfare coss increase o 1 6 of 1 percen and ⅓ of 1 percen, respecively. If aid could be delivered in such a way as o insure households in he recipien counry agains he effecs of produciviy shocks, i.e., fully sabilizing aggregae consumpion, he welfare gains would be equivalen o 3 4 of 1 percen of CES consumpion and 1½ percen of CES consumpion denominaed in radable goods. We noe wo caveas of hese model resuls. Firs, our calculaions of welfare losses are on he very low bound of he range of plausible esimaes: had we inroduced capial adjusmen coss, we would have obained much higher welfare coss of volailiy. In such a case, invesmen would be a less effecive insrumen o insure agains consumpion flucuaions. Second, in a model wih a governmen, aid shorfalls would resul in disrupion in public services and invesmen wih associaed welfare implicaions. We can also ask wha level of aid would render households indifferen, in erms of uiliy, beween an environmen of volaile and procyclical aid and one where aid is consan a he mean he so-called compensaing variaion. his is equivalen o he reducion in consumpion ha households would be willing o give up in order o receive consan aid flows. We calculae he compensaing variaion as he welfare coss above, denominaed in 12 Our welfare calculaion is idenical o ha of Lucas (1987), who esimaed welfare cos as ½ imes he risk aversion coefficien (σ) imes he difference in he variance of consumpion. here are well-known measuremen problems wih his approach: aid volailiy no only changes he volailiy of consumpion, bu also changes consumpion as a resul of precauionary savings ha enlarge he sock of capial. On he one hand, shocks increase consumpion volailiy and decrease welfare; on he oher hand, consumpion increases in he long run due o a higher capial sock, increasing welfare. hus, we base our welfare calculaions solely on consumpion volailiy, ignoring he changes in levels.

22 erms of radable goods, as a percenage of mean aid inflows. Firs, if donors delivered aid in a sable manner and ignoring produciviy shocks, hey could mainain he same level of recipien well-being while reducing aid by ¾ of 1 percen. Second, if aid could insure agains produciviy shocks, donors could reduce aid inflows by 3 percen wihou reducing he presen level of welfare. hese esimaed welfare effecs are comparaively large: as one reference poin, hey far exceed esimaes in he lieraure for he welfare coss of business cycles in indusrial counries (beween 1 10 and 3 10 of 1 percen of annual U.S. consumpion for he pos-war period in he Unied Saes, see İmrohoroğlu, 1989). In summary, if donors were o allocae aid eiher in a sable fashion or, beer sill, o smooh ou produciviy shocks, he benefis o poor counries would be equivalen o hose of a subsanial increase in he overall level of aid. 13 IV. SOME EMPIRICAL RESULS In he heoreical secion we have shown ha aid lowers radable goods oupu subsanially, while aid volailiy has a comparaively small impac on radables. In his secion, we will link hese resuls o oupu of radable goods in aid-dependen counries. he cross-counry regressions sugges ha a high degree of aid dependence is indeed associaed wih lower manufacuring expors and ha aid volailiy appear o play a negaive role in expor deerminaion. However, he laer resuls are no robus. A. Explaining Expors We explore he relaive imporance of various facors ha may explain radable goods performance in cross-counry regressions. Unforunaely, empirical invesigaions of radable goods have been plagued wih measuremen problems. On he one hand, he disincion beween radable and nonradable goods has become increasingly ambiguous. 14 On he oher hand, even clearly radable goods are no homogeneous: some noionally radable commodiies are no raded exernally because of heir qualiy, price, or producer locaion. 15 Moreover, he valuaion of radables in domesic currency erms is ofen complicaed by he volailiy of he exchange rae in developing counries. aional accouns daa are generally of inferior qualiy o inernaional rade daa. 13 he possible use of counercyclical aid o insure low-income counries agains economic flucuaions is explored by Pallage, Robe, and Bérubé (2004). 14 Čihák and Holub (2003) demonsraed his argumen for ransiion counries. 15 Goldsein and Lardy (2005) use he example of car manufacuring in China ha is predominanly serving he domesic marke. Indeed, secors producing radable goods for he domesic marke and for expor end o be fairly disinc in mos newly indusrialized and developmen counries. While he former are capial inensive, he laer ofen specialize in labor inensive, low-skill echnologies ha have been ousourced from indusrial counries.

23 Empirical research on he srucure of producion in developing counries has hus focused on manufacured expors in relaion o oal expors as opposed o radable goods in relaion o GDP and we follow in his radiion (Sekka and Varoudakis, 2000). Firs, inernaional rade daa are compiled a reasonable qualiy and frequency. Second, he moivaion for he narrow focus on manufacured expors reflecs ha fac ha primary-secor expors (agriculure and mining in paricular) are likely o be affeced o only a limied degree by he facors examined in his paper; for example, mining oupu depends largely on discovered reserves which end o be price-inelasic in he shor o medium erm and he associaed rens serve as a cushion agains exchange rae flucuaions. hird, by expressing manufacured expors as a percenage of oal expors we sidesep he problem of assessing overall economic aciviy and he measuremen error caused by exchange rae volailiy. everheless, o conrol he robusness of our resuls, we esimaed he impac of aid on wo oher dependen variables: he value added in he merchandise secor as a percenage of GDP and he nonmining expors as a percenage of oal expors, finding ha he impac of aid and aid volailiy remained saisically significan. (hese resuls are available on reques.) In assessing manufacured expor performance, several consideraions are likely o be imporan (see Elbadawi, 1998 for a review). Firs, differences in counries relaive endowmens including human capial, naural resource endowmen, and locaion are arguably imporan deerminans of expors (Wood and Berge, 1997 and Rodrik, 1998). his consideraion would sugges, in paricular, ha sub-saharan Africa s poor human capial resources, vas supply of primary commodiies, and long disance o exernal markes are imporan reasons is manufacured good expors are, on average, significanly lower han hose in oher regions. Second, ransacion coss associaed, for insance, wih expor axes, poor governance, and dilapidaed infrasrucure may pose imporan barriers o expor performance for a given se of endowmens (Collier, 2000). I is imporan o conrol for hese facors in examining he effec of aid inflows on he producion of radables and on expors, as prediced in he model developed earlier in his paper. We are adding aid and aid volailiy o he lis of poenial explanaory variables. While he measuremen of he former variable is sraighforward, aid volailiy can be measured in a variey of ways and we show resuls for wo of hem. Firs, he volailiy of aid can be compared vis-à-vis expecaions of aid, ha is, he donor commimens of aid. A higher raio of commimens o disbursemens (denoed as Aid volailiy I in able 5) implies more aid volailiy. his raio is invariably larger han 1 in our sample: donors commi more aid han hey disburse. Second, he volailiy of aid can be compared o oher macroeconomic variables, such as domesic fiscal revenue. he raio of variances of aid and domesic revenue (denoed as Aid volailiy II in able 5) is an obvious benchmark as aid receips supplemen domesic fiscal revenue. 16 Hence, while he firs measure can be relaed o he noion of aid predicabiliy, he second measure relaes aid volailiy o he insabiliy of he overall 16 We express he annual series of aid and domesic revenue in percen of GDP, ake logs, and filer he series by he Hodrick-Presco filer. Variances of aid and revenue are hen compued from he derended series. See Bulíř and Hamann (2003) for addiional deails.