Provided by the author() and Univerity College Dublin Library in accordance with publiher policie., Pleae cite the publihed verion when available. Title The Effect of Foreign Aid in Sub-Saharan Africa Author() Gillander, Robert Publication date -8 Serie UCD Centre for Economic Reearch Working Paper Serie; WP/6 Publiher Univerity College Dublin. School of Economic Link to online verion http://www.ucd.ie/tcm/wp_6.pdf Item record/more information http://hdl.handle.net/97/67 Downloaded 9--7T7:9:8Z The UCD community ha made thi article openly available. Pleae hare how thi acce benefit you. Your tory matter! (@ucd_oa) Some right reerved. For more information, pleae ee the item record link above.
UCD CENTRE FOR ECONOMIC RESEARCH WORKING PAPER SERIES The Effect of Foreign Aid in Sub-Saharan Africa Robert Gillander, Univerity College Dublin WP/6 Augut UCD SCHOOL OF ECONOMICS UNIVERSITY COLLEGE DUBLIN BELFIELD DUBLIN
The Effect of Foreign Aid in Sub-Saharan Africa Robert Gillander Univerity College Dublin School of Economic Augut Abtract Thi paper contribute to the aid effectivene debate by applying a vector autoregreion model to a panel of Sub-Saharan African countrie. Thi method avoid the need for intrumental variable and allow one to analye the impact of foreign aid on human development and on economic development imultaneouly. The full ample reult indicate a mall increae in economic growth following a fairly ubtantial aid hock. The ize of the effect put the reult omewhere between the argument of aid optimit and thoe of aid peimit. Economic growth i found to repond more to aid hock in group defined by better economic policie, poor intitution and high aid dependence. Human development, for which I ue the growth rate of life expectancy a a proxy, repond poitively to aid hock in democracie and in good intitutional environment. UCD School of Economic, Newman Building (Room G), Univerity College Dublin, Belfield, Dublin, Ireland. rgillander@gmail.com Funded by the Irih Reearch Council for the Humanitie and Social Science. I am indebted to Karl Whelan for detailed comment and uggetion and to Inea Love who kindly made her PVAR Stata program available to me. I alo wih to thank the participant at the European Doctoral Group in Economic Jamboree, the Irih Society for New Economit Conference, the Indian Statitical Intitute 6 th Annual Conference on Economic Growth and Development and the Irih Economic Aociation Annual Conference (Anna Gumpert and Marku Eberhardt in particular) for helpful comment and uggetion.
Introduction For better or for wore, Official Development Aitance (ODA) i the main tool employed by the rich world in it attempt to promote properity in the developing world. Given the importance of thi tool, it hould not be a urprie that many cholar of development conider it impact worthy of tudy, particularly it impact on economic growth. Unfortunately, from a policy maker perpective at leat, the approach taken by mot of thee work ha yielded finding acro the full pectrum of potential reult. The objective here i to employ a different method. A method that doe not require a many (often controverial) aumption. Thi paper etimate the impact of aid in a panel vector autoregreion (PVAR) framework. For the mot part, the exiting literature etimate tandard growth regreion augmented with aid term on panel data and attempt to overcome the endogeneity problem between growth and aid with tandard intrumental variable technique. The PVAR approach avoid the need for intrumentation trategie a all variable in a PVAR are recognied to be endogenou. Each variable i regreed on it own lag and thoe of the other variable. Only a minimal et of aumption i required to interpret the impact of hock in each variable on the ytem. Another merit of thi notoriouly atheoretical approach i that, a both Eaterly [] and Eaterly, Levine, and Roodman [] point out, much of thi literature uffer from a lack of a formal theory linking aid and growth to guide applied reearcher in ettling on an econometric pecification. A will be hown below, nothing approaching a conenu ha emerged from the literature and many key finding have been found to be le than robut to change in ample or pecification. Vector autoregreion have been ued by reearcher to tudy the impact of foreign aid in a particular country. For example, Oei, Morriey, and Lloyd [] examine the fical effect of aid in Ghana while Morriey, M Amanja, and Lloyd [6] examine the impact of ODA on growth in Kenya. While Hanen and Headey [] employ a PVAR model to examine the hort-run effect of aid on net import and pending, they do not examine growth. Thi paper main contribution then i to expand on the work of Morriey and hi co-author by widening the cope via method imilar to thoe ued by Hanen and Headey []. The econdary contribution of thi paper i that the PVAR approach allow one to examine the other ide of development imultaneouly. Taken a a whole, the foreign aid effectivene literature i vat. While mot of thi literature look at aid effect on economic development, there i a maller, yet growing, body of work that look at aid impact on human development a meaured by variable uch a the Human Development Index (I) and the infant mortality rate. However, ingle equation technique can only focu on one iue at a time. It make ene to make ue of the multi-equation nature of a PVAR to ex-
amine both apect of development (very broadly defined) imultaneouly, a while GDP per capita i a good meaure of the overall tate of a ociety (due to it high correlation with thing we actually do care about), it i far from perfect. Economic development i urely a large part of the concept of development but there are other element that matter. Interet lie pecifically with the impule repone function (IRF) obtained from the model. Thee will how the repone path of economic growth and human development to a one time hock in foreign aid holding all other hock at zero. To reduce concern about biae that could arie from etimating the model without allowing for fixed effect, the ample i retricted to countrie in Sub-Saharan Africa. Thee IRF erve a the baeline reult and it i hown that thee reult are in line with GMM etimate which allow for country pecific fixed effect. The ample i then plit into group defined by economic policy, intitutional quality, democracy and aid dependence. Thi allow u to ee if aid i more effective in certain environment. Several important and policy relevant reult emerge from the analyi. Firtly, aid hock do eem to induce economic growth in the firt few year following the hock. However, thi initial repone i le than % extra growth per annum and i tempered by ome later negative repone. Secondly, aid appear to have a mall poitive impact on human development, though the error band do not allow u to rule out that the effect i negative or non-exitant. In countrie with good economic policy, the early impact of aid on economic growth jump to about %, though thi i once again tempered by ome negative repone later on. In democracie, aid ha a much larger impact on human development and a le pronounced effect on economic growth than in either the full ample or in autocracie. Relative to countrie with good intitution, thoe with poor intitutional environment make more ue of aid in the economic phere and le ue of aid in term of human development. Finally, countrie which are more dependent on aid ee a better repone in term of both economic and human development. One of the main implication of thee finding i that while aid doe lead to economic growth, the impact i not tranformative and to maintain even an extra % of growth would require frequent large injection of aid. Another i that while the repone of GDP per capita growth i higher in ome environment, it i not o much higher that donor could jutify focuing all aid monie on countrie with thee characteritic. Thi i further upported by the fact that often it i the cae that when aid fail to have an impact on one of the dimenion of development in one of the ub-ample, it impact on the other. Other ocial cience are at pain to make thi point to economit, but I would argue that we already know thi. Certainly, the mall but growing empirical literature on aid and human development (which i outlined below), not to mention the work of Amartya Sen, how that economit are aware of the multidimenional nature of development.
The remainder of the paper proceed a follow. Section outline the exiting literature. Section outline the data and dicue iue of meaurement. Section decribe the econometric approach in detail. Section and 6 preent the baeline and grouped reult repectively and Section 7 conclude. The Exiting Literature There i an extenive literature that eek to examine the impact of foreign aid on economic growth and a maller one that look at aid impact on human development. Some tudie have found poitive effect, ome conditional effect and other no effect. Given the wide range of finding and the debate that they have prompted, it i worth examining ome of the relevant recent paper in brief.. Aid and Economic Development There are many tudie that have examined the effect of aid on growth. I limit myelf here to the more recent work, both for brevity and becaue it highlight the iue that will be examined in Section 6. Recent work ha chiefly focued on conditional aid effectivene. Burnide and Dollar [] reignited the aid effectivene debate when they found that while aid ha no effect on growth on average, aid work in a good policy environment. They include an aid*policy interaction term and find that it i tatitically ignificant and robut to a number of pecification. Thi paper launched the debate on conditional aid effectivene. Eaterly, Levine, and Roodman [] recreate the Burnide-Dollar dataet and expand on it ignificantly. By following the approach of Burnide and Dollar to the letter, they find that not only i the crucial aid*policy coefficient inignificant but it ha the oppoite ign. Eaterly [] re-examine the iue in a different way. By employing Official Development Aitance (ODA) a the meaure of aid a oppoed to the meaure ued by Burnide and Dollar, Effective Development Aitance (EDA), Eaterly find that the aid*policy interaction term i no longer ignificant. He alo varie the pecification of good policy and again find that the aid*policy term i inignificant. The crucial interaction term i alo found to be inignificant by varying the definition of growth (Burnide and Dollar defined growth a real GDP growth over four year) to conider eight, twelve and twenty four year. Roodman [7] provide an excellent overview of the entire literature and Deaton [] dicue the potential problem with the IV trategie commonly employed in thi area. A many author note, the two meaure are highly correlated. The two are fully defined in the data ection below.
One of the bet example of upport for the Burnide-Dollar reult come from Collier and Dollar []. They expand the Burnide-Dollar dataet to include 9 aid-growth-policy epiode a oppoed to the original 7. The other modification they make i to employ the World Bank Country Policy and Intitutional Aement (CPIA) a their policy variable. Their finding agree with the Burnide-Dollar reult. However, Dalgaard, Hanen, and Tarp [] quetion the uitability of the CPIA for growth regreion. They include a climate*aid interaction term and find that it i ignificant. The aid*policy term loe it ignificance once climate*aid enter the pecification. They ugget that the climate variable may be a proxy for deep determinant uch a intitution. Thi i a ignificantly different conditional effectivene reult and one which will be examined in Section 6. Given the ketch of the literature above, it will come a no urprie that the role good policy ha to play will alo be invetigated. Svenon [999] examine whether civil and political libertie play any role in aid effectivene. He find that an aid*democracy interaction term i highly ignificant and that an aid*policy term à la Burnide and Dollar i inignificant. It i important to note that he conider, and reject, the poibility that democracy i imply a proxy for good policie. Thi ugget we hould add democracy to our lit of conditional reult to be examined. Thi ample of the aid and economic growth literature i chiefly included to motivate the diviion of the ample into group defined by economic policy, intitution and democracy in Section 6. However, it alo how that the traditional approach of running either tandard cro-ectional or panel growth regreion augmented with aid and aid interaction term lead to fragile reult. Indeed, the title of Roodman [7], The Anarchy of Number: Aid, Development, and Cro-Country Empiric, um up the literature well. Each of the paper i a fine econometric work, and it may be that thee factor do indeed matter, but the fragility of the reult i undeniable. The propect of a freh perpective i a major motivating factor in employing the PVAR methodology.. Aid and Human Development While nowhere near a extenive a that which concern itelf with economic development, there i a mall but growing literature that eek to empirically ae the impact of foreign aid on human development. Roodman examine even leading paper in the aid-growth literature, (including Burnide and Dollar [], Collier and Dollar [] and Dalgaard, Hanen, and Tarp []) and find that each of them i uceptible to change in the ample and in pecification.
Koack [] find that aid ha a poitive effect on I growth but only in democratic countrie. Hi etimate alo ugget that aid will have a negative effect on I growth in autocracie. Interetingly, he find that democracy alone ha a negative effect on I growth. He interpret thee finding a implying that more-democratic poor countrie have, on their own, lower growth in the quality of life, but that aid to thee countrie may revere thi negative tendency (p6). McGillivray and Noorbakhh [7] examine the impact of aid on the level of the I and allow conflict to enter the analyi. They find that aid alone ha a negative impact on I core but diagree with Koack [] in that they do not find either a negative effect of democracy on the I or a poitive aid*democracy interaction term. Thee two tudie give u a econd reaon to divide the ample along line of democracy. Uing quantile regreion, Gomanee, Girma, and Morriey [a] examine aid effect on human development a meaured by both the I and the infant mortality rate. They argue that while aid might not have a direct impact on welfare, it may have an indirect one via pro-poor expenditure (PPE). By contructing a PPE index they find that aid ha a poitive impact on welfare through public expenditure and that the effect i greater in countrie with lower welfare. They alo argue that good economic policy i not required for aid to be effective in promoting human development. A related paper, Gomanee, Morriey, Moley, and Verchoor [b], find to the contrary that there i a direct impact of aid on human development and little evidence of an indirect effect via PPE. It i clear that there i nearly a much diagreement and tendency for conflicting reult in the aid-human development literature a there i in the aid-growth literature. There are of coure more work that concern themelve with the effect of aid on economic and human development. However, the paper above provide a ufficient overview of the evidence and ugget that it may be illuminating to examine whether aid ha different effect in group of countrie defined by economic policy, intitutional quality and democracy. The final diviion of the ample i inpired by Hanen and Headey []. They utilie a PVAR model to examine the hort-run macro effect of aid. They plit the countrie in their ample into thoe that are highly dependent on ODA and thoe that are not. In the context of the current work, it i plauible that there i ome critical level of aid dependence under which aid cannot be effective. It i equally plauible that being overly dependent on aid reult in macro ditortion. Rather they find that aid and conflict have negative effect on human development and that aid i no more or le effective in conflict ituation. 6
Data The data on yearly economic growth come from the World Bank World Development Indicator (WDI). 6 While economic growth i a concrete concept, the other two variable that will be ued in thi analyi are omewhat nebulou. Thu, it i good practice to devote a little time to defining exactly what the aid and human development variable ued in the analyi actually are.. Meauring Aid Thi paper ue the notation aid it to repreent total net ODA per capita. Thi include flow from all donor (a meaured by the OECD) to recipient i at time t. aid it i meaured in contant 7 US dollar. 7 The data come from the OECD Development Aitance Committee (DAC). 8 In chooing how to meaure aid, one mut make two important deciion. Firtly, what count a foreign aid? Thi effectively boil down to the choice between ODA and EDA. ODA i defined on the OECD webite a: Grant or loan to countrie and territorie on the DAC Lit of ODA Recipient (developing countrie) and to multilateral agencie which are: (a) undertaken by the official ector; (b) with promotion of economic development and welfare a the main objective; (c) at conceional financial term (if a loan, having a grant element of at leat per cent). In addition to financial flow, technical cooperation i included in aid. Grant, loan and credit for military purpoe are excluded. Tranfer payment to private individual (e.g. penion, reparation or inurance payout) are in general not counted. 9 EDA i an adjutment to ODA that replace the official loan component with the grant equivalent of official loan and diregard grant that are tied to technical aitance. Although thi may eem like an important modification, the two are hugely correlated. I chooe to ue ODA a the aim of the paper i to ae the impact of the Wet effort and ODA i the tool they employ. 6 The WDI can be acceed at http://data.worldbank.org/indicator 7 Given that ODA contain conceional loan, thee net flow can be negative if the repayment of loan are greater than new loan and grant. 8 The data can be acceed at www.oecd.org/dac/tat/data 9 See www.oecd.org/dac/gloary Chang, Fernandez-Aria, and Serven [999]. Dalgaard and Hanen [] how that the correlation between the two (a a fraction of GDP) in nominal term i.98 and between real EDA and nominal ODA it i.89. They alo ugget that the difference between the two meaure eem to be a imple tranformation (p6). 7
The other iue one mut decide on i how to normalie the aid variable. Many tudie expre aid a a fraction of GDP. In my cae thi would lead to a mechanical relationhip between two of my main variable of interet - aid and economic growth. To avoid thi, I divide yearly aid data by yearly population data from the Penn World Table. Thu, aid it i fully defined a total net ODA per capita.. Meauring Human Development One alo face two iue when attempting to quantify human development ( it ). The firt i finding an ideologically neutral meaure. While ome of u would conider low level of poverty or inequality to be good indicator of human development, other would diagree. Claiming either of thee variable a good proxie for human development require one to take an implicit ideological tance. Uing democracy or meaure of government accountability i imilarly unatifactory. The econd problem i the lack of ufficiently long time erie data relating to human development. The UN Human Development Index (I) i equal part economic properity, education and life expectancy and i largely free of ideology. However, the erie i not of ufficient length to be uable in an analyi uch a thi. I conider life expectancy to be an ideologically clean proxy for human development. While there are other apect to human development, a long lifepan i eential to purue many of them. The data erie, which again come from the World Bank WDI, i of ufficient length and cover a ufficient number of countrie. The pecific variable, denoted a it, i the growth rate of total life expectancy at birth (in year). The data allow for a balanced panel of countrie over the period 97-. The main reaon for the omiion of 7 countrie i GDP growth erie that tart too late. Some of thee data poor countrie have very mall population and other had particularly turbulent tart to their independence. The lit of countrie that are ued in the analyi, along with ummary tatitic, can be een in Appendix A. Of coure, if one et out to examine the impact of ODA on inequality or poverty then thi iue of an ideological bia doe not arie. Though it i omewhat arbitrary in that it give equal weight to each component. For more detail ee United Nation Development Programme [variou year]. I ue the growth rate to minimize concern about tationarity. 8
Econometric Approach Thi paper i concerned with the IRF aociated with the following PVAR model: p Z it = α + β q Z it q + ǫ it () q= where Z it i the vector (aid it, growth it, it ), α i a vector containing the contant term, β q i the matrix of coefficient for lag q and ǫ it i ditributed a (,σi ) with E(ǫ itǫ i ) = for all t. Thi model can be viewed a the mot retrictive poible in that it impoe common lope and common intercept and can be etimated by Pooled Ordinary Leat Square (POLS). It i well known that POLS yield biaed and inconitent etimate if the true data generating proce contain a fixed effect. To reduce uch concern, only countrie in Sub-Saharan Africa are conidered. All countrie are of coure different, but thee countrie hould form a ufficiently homogeneou group to allow POLS etimate of the relationhip between ODA, economic growth and human development to be meaningful. Such concern hould be further reduced by the diviion of the ample into the ubgroup outlined above. A le retrictive model i a dynamic panel data model with country fixed effect. Such a model can be etimated via GMM technique a i done by Love and Zicchino [6]. Thi approach will be ued a a check on the main reult. A it i not the reduced form diturbance of Equation () that we are intereted in, we need a method to recover the pure tructural hock. The mot common way to obtain orthogonalied hock i to employ a Choleki decompoition which order the hock in a enible way. The channel of influence I impoe i the following: aid it growth it it Aid hock can impact on growth and human development contemporaneouly; growth can impact on human development contemporaneouly but only influence aid flow with a lag; human development only operate on the other two variable with a lag. The choice of An even le retrictive variant of Equation () allow for both country pecific intercept and lope. Pearan and Smith [99] how that thee ort of model can be etimated by applying the Mean Group Etimator (MGE). Thi i the method adopted by Hanen and Headey []. However, Rebucci [] how, uing Monte Carlo imulation, that the diperion of the lope parameter around their mean mut be high in abolute term for the heterogeneity bia of pooled etimator to be ubtantial (p6). In addition, the imulation indicate that for MGE to be a ueful alternative, the time dimenion of the panel mut be longer than that which i to be found in mot macroeconomic dataet. 9
ordering i a crucial factor in any VAR exercie. 6 The rationale behind the choen ordering i a follow: Donor countrie (epecially their bureaucrat) need time to oberve and react to change in the recipient country. Thu aid flow will only repond to change in economic and human development with a lag. It take time for increaing human development to tranlate into economic development. It i much more plauible that economic growth can have a contemporaneou effect on human development. While thi i only one of ix poible ordering, it eem the mot plauible and the enitivity of the reult to an alternative Choleki decompoition i examined in Appendix B. I follow Love and Zicchino [6] and Hanen and Headey [] and contruct 9% confidence interval uing Monte Carlo Integration method. 7 Finally, to determine the apropriate number of lag to include, that i the p in Equation (), I employ tandard information criteria. Full Sample Reult Both the Akaike information criterion and the Schwarz-Bayeian information criterion indicate that the model hould include eight lag.. POLS The IRF obtained from the full ample of countrie can be een in Figure. For completene I preent all the IRF from the ytem though interet only really lie with the repone of the growth rate of GDP per capita and life expectancy to a hock in aid. Each IRF how the repone path of the variable in quetion to a one time poitive one tandard error increae in the etimate of the hock variable holding all other hock at zero. The value of the hock can be read off the Y-axi of the diagonal element of Figure at time zero. The aid hock i approximately $ per peron. It i, however, reauring that the none of the other IRF diplay inexplicable behaviour. Aid tend to decline following an economic growth hock and tend to increae following a hock to the growth of human development. It i plauible that when donor ee an increae in economic growth they cale back aid but when they ee ome improvement in 6 The importance of the ordering increae with the correlation between the reduced form innovation. See Ender [] page 76. 7 I ue RATS MONTEVAR procedure which make draw from an invere Wihart ditribution and ue the Jeffery prior.
Figure : Full Sample IRF: POLS Impule repone 7 6 - - - - -. 8.7 Repone of..8.6... -. -. -.6 7 6 -... -. -. -.7.8..8.6..6. -.... -.. -. -.. -. -. -. -.6 -. -. Note: Each IRF depict the repone of the row variable to a one tandard error hock in the etimate of the column variable. The upper and lower line are 9% error band generated uing Monte Carlo imulation with draw. human development they attempt to conolidate thee gain. Economic growth i lightly lower for a few period following a hock to human development growth perhap due to people witching from economic to leiure or other non-market activitie though the effect quickly die out. Finally, a hock to economic growth increae the growth rate of human development until about even period after the hock, at which point the growth of human development i lower than it would have been. Thi i perhap due to increaed pollution or other retarding influence that can arie from increaed economic activity. Turning to our main concern, we can ee that an aid hock at time zero ha an immediate, but mall, poitive impact on GDP per capita growth. Thi poitive impact increae to about.8% extra growth one period after the hock and the poitive impact till exit in period two. Thee hort run increae in growth can be explained a increae in government invetment and perhap conumption. The next three period ee growth that i lower than it would have been in the abence of the initial aid hock. A plauible explanation for thi i
that recipient government treat the aid hock at time zero a a permanent increae and are taken by urprie when it prove to be tranitory. 8 After thi we ee a return to a poitive repone. Thi would fit with government invetment on infratructure made with the initial aid hock paying off. After the th tep the initial hock ha no more influence. The error band are generally tight enough for u to have confidence in the etimated repone path. So i aid effective in term of generating economic growth? Given the reult of Figure, one would have to ay ye. The poitive repone to aid eem to outweigh the period where growth i lower than it would have been. However, there are two important iue one hould bear in mind:. The extra injection of $ per peron i not cheap. Even at thi level, the reult don t fit the decription of tranformative.. Given that the effect die out by the th period, can we ay that the induced economic growth i utainable? Thee reult thu it omewhere between the argument of aid optimit and aid peimit. Ye aid eem to be effective in that it doe induce ome economic development, but to make it a promiing tool for eliminating global poverty it would need frequent large injection. So are the reult in term of human development more upportive of aid advocate? Not really. The etimated repone path i poitive for the mot part but very mall in magnitude. That aid, the human life pan i meaured on a (adly) mall cale o an increae in it yearly growth rate of even a fraction of a percent may be quite good, epecially given that the repone perit until the th period following the initial hock. However, the error band allow for the poibility that the impact may be negative. The reult in thi repect are thu ambiguou.. Robutne: GMM It i important to check if the baeline reult are robut to allowing for country pecific effect i.e. if the POLS approach give different reult to a model that allow for fixed effect. To do thi, I employ the PVAR Stata program written by Inea Love. For pecific detail of the procedure, I direct the reader to Love and Zicchino [6]. Briefly, the etimator allow for country fixed effect which are removed by forward mean differencing. The model i then etimated by GMM. 8 They may not be entirely wrong in thi a Figure how that the repone of aid to it own hock perit beyond fifteen period, though it doe drop rapidly to roughly a third of the value of the hock.
Figure : Full Sample IRF: GMM Impule repone for 6 lag VAR of aid growth hd 8.87 (p ) aid aid (p 9) aid.989 (p ) growth growth (p 9) growth 8.6 (p ) hd hd (p 9) hd. 6 repone of aid to aid hock (p ) aid aid (p 9) aid.786.99 6 repone of aid to growth hock (p ) growth growth (p 9) growth 7.6.9 6 repone of aid to hd hock (p ) hd hd (p 9) hd.97.968 6 repone of growth to aid hock (p ) aid aid (p 9) aid.9. 6 repone of growth to growth hock (p ) growth growth (p 9) growth.879.6 6 repone of growth to hd hock (p ) hd hd (p 9) hd.78.6.. 6 6 6 repone of hd to aid hock repone of hd to growth hock repone of hd to hd hock Error are % on each ide generated by Monte Carlo with rep Note: Each panel how the repone of the indicated repone variable to a one tandard deviation hock in the hock variable. Reult obtained uing the PVAR program kindly made available to me by Inea Love. Inea Love PVAR program will not allow me to obtain error band for model with more than ix lag. The reult are preented in Figure while the reult of a model with eight lag (with no error band) can be found in Appendix C. The IRF only trace out the repone to ix period after the initial hock and ue a lightly different value for the hock, but we can immediately ee that the pattern of repone to an aid hock are very imilar to thoe obtained uing POLS. The aid hock i maller yet it generate a larger, though till rather mall, repone in the growth rate of human development and the economic growth repone i roughly % of the POLS outcome. However, the reult of the eight lag GMM model how a imilar level of repone to that of the POLS model. The error band are alo wider in the cae of the economic growth repone, though the bulk of the error band lie above zero. Overall we can be confident that POLS i a atifactory method and that the reult preented in Figure are valid.
6 Conditional Reult To invetigate whether aid work in certain environment, that i conditionally, previou author in thi literature have included interaction term in their pecification. While one could do likewie in the VAR framework, the interpretation of hock in thee multiplicative term would be difficult and the extra term would have to be made to fit into a Choleki decompoition. I take a impler alternative approach and merely run the model over group defined by certain characteritic. 6. Good Veru Bad Economic Policy Given how influential the Burnide-Dollar reult ha been in donor circle, it make ene to ue the model to ee if we can find evidence of aid working better in a good policy environment. One leon to be taken from the literature i that often the definition of good policy i omewhat arbitrary in what i included and once it i changed the ignificant reult diappear. With thi in mind, I have choen the mot objective meaure of economic policy that I have come acro. The World Bank Doing Buine project collect information acro a wide variety of apect of the buine environment and rank countrie on the overall eae of doing buine. The full methodology can be found on the project webite. 9 Here, it i ufficient to note that the urvey are a objective a poible and cover mot apect of a country buine environment. Thu, the ranking hould provide a good proxy for overall economic policy. Thi i quite a different meaure of policy from the weighted indice of macroeconomic indicator that are commonly ued in thi literature. I ue the mot recent ranking a my meaure. While thi i open to criticim, the data only goe back to and additional important variable have been added over time. I believe that the data can afely be regarded a a good meaure of a country tock of economic policie. I conider a country with a ranking in the top to have good economic policy. Countrie below thi, admittedly arbitrary, cutoff form the bad policy ample. A benefit of dividing the ample thi way i that even if we would ee countrie move around omewhat in the ranking over time had we the data, uch broad range make it more likely that they remain within the bound I have et for good and bad policy. In any event, the cutoff would have to be much more forgiving for many more countrie to make into the good ample. Figure and preent the IRF obtained from running the model on each of the policy group. It will come a no urprie that only five countrie made the cut for good policy. I continue to include eight lag in the model over all ample. 9 http://www.doingbuine.org Appendix A how which countrie are in thi and each ubequent ub-ample.
Figure : Good Policy IRF Impule repone 6 8 - - 6-6 -8 - - - 7.. Repone of - -... -..... -. -. -.......7...... -... -. -. -. -.7 -. -. -. -. Note: Each IRF depict the repone of the row variable to a one tandard error hock in the etimate of the column variable. The upper and lower line are 9% error band generated uing Monte Carlo imulation with draw. The ample i compoed of thoe countrie which rank in the top in the World Bank eae of doing buine ranking. The reult are not a precie a thoe of Section but are triking nonethele. It i worth noting that the magnitude of the aid hock in both cae are very imilar. In the good policy group, economic growth i actually lower than it would have been at t= abent the aid hock. However, by t= growth i roughly % higher than it would have been. We ee the ame pattern of negative repone that we aw in the full ample after t= with ome mall potitive repone after t=. While the error band do not allow one to dicount the poibility that thee long lating effect are negative, they were not preent in the full ample reult. The bad policy ample repone are markedly different and (not urpriingly given that thee countrie form the vat majority of the full ample) are imilar to the full ample reult. The level of repone i maller though and we ee ome long lating negative repone that were abent in the full ample.
Figure : Bad Policy IRF Impule repone 7.. -. - -. - - -.. 8.8 Repone of.7... -. -. -.7 7 6 -.6.. -. -. -. -.6 -.8........... -. -. -... -. -. -.6. -. -.8 -. Note: Each IRF depict the repone of the row variable to a one tandard error hock in the etimate of the column variable. The upper and lower line are 9% error band generated uing Monte Carlo imulation with draw. The ample i compoed of thoe countrie which rank outide of the top in the World Bank eae of doing buine ranking. In term of human development, the error band are too broad to draw much inference. In the bad policy group the etimated effect i inditinguihable from zero for the mot part. In the good policy group the error band are alo too wide but we can ee initial negative repone. After that we cannot ay with any certainty what ign the effect take. Thee reult ugget that, in term of increaing GDP per capita growth, aid doe work better in a good policy environment. However, I would argue that it till doe not reach the level of tranformative power that aid advocate claim. A % boot to economic growth i certainly impreive but once again it i not cheap and i tempered by ome negative repone. On the other hand, the reult certainly do not conform to the argument of the mot depairing of aid peimit in that we ee ome beneficial impact of aid, even in countrie with bad policy. 6
Figure : Democratic IRF Impule repone. 7.... 7.... 7..... -. -. -. -... Repone of..7... -. -. -.7 -. -... -. -. -....... -... -... -. -. -. -. -. -. -. -. -. -. -. -. -. Note: Each IRF depict the repone of the row variable to a one tandard error hock in the etimate of the column variable. The upper and lower line are 9% error band generated uing Monte Carlo imulation with draw. The ample compoed i of thoe countrie with an average value of or higher in the Polity IV index of political regime characteritic. 6. Democracy V. Autocracy We aw in Section that there i evidence that being a democracy matter for both dimenion of aid effectivene. To examine thi iue, I utilie the meaure of democracy created by the Polity IV project. Thi variable combine information on democratic and autocratic feature of a country and place them on a cale from (hardcore autocracy) to + (hardcore democracy). Thi ugget a imple diviion of the ample. Countrie with an average Polity core (over the ample period) le than are conigned to the autocratic group. Thoe with a core of or higher form the democratic group. Figure and 6 how the IRF for both ub-ample. A wa the cae with the bad policy group, the autocracy group form the vat majority of the full ample and it i therefore unurpriing that the IRF are very imilar to the baeline reult. The big difference i 7
Figure 6: Autocratic IRF Impule repone 6 - - - - -. 8. Repone of... -. -. 7 6 -.8.6.. -. -. -. -.6 -.8.6..8...6... -.. -.. -.. -.6 -. -. -.8 -. -. Note: Each IRF depict the repone of the row variable to a one tandard error hock in the etimate of the column variable. The upper and lower line are 9% error band generated uing Monte Carlo imulation with draw. The ample i compoed of thoe countrie with a negative average value in the Polity IV index of political regime characteritic. that we ee no appreciable poitive impact of aid on human development. Turning to the democratic ample, we ee a relatively large repone of human development to an initial aid hock. Not only i it relatively large, it i poitive and perit to the th tep. A the hock are once again of very imilar magnitude, thi reult partially upport the finding of Koack [] aid eem to work in term of promoting human development only in democratic countrie but we cannot ee any clear evidence of it retarding human development in autocracie. In term of aid impact on economic growth, a noted above, the autocratic ample repone path i very imilar to that of the full ample. The major difference i the late negative repone that were abent in the full ample. The repone path in the democratic ample ha error band that are too wide to draw any firm inference for the mot part, though the initial repone eem to be poitive. While thee initial poitive repone are maller than 8
Figure 7: Good Intitution IRF Impule repone 8 6 - - - - - -.6. Repone of.. -. -. -. -.6 -.8 -.7... -. -. -.7...6.8 -...6.. -.... -. -... -.. -. -. -. -.6 -. -. Note: Each IRF depict the repone of the row variable to a one tandard error hock in the etimate of the column variable. The upper and lower line are 9% error band generated uing Monte Carlo imulation with draw. The ample i compoed of thoe countrie with an average core of or higher in the Polity IV contraint on executive power index. thoe in the autocratic ample, they lat longer. After that, it i generally not poible to make any inference about the ign of the repone. Thee reult, therefore, do not conform with Svenon [999] in that aid doe not eem to be more effective in promoting economic growth in democratic environment. 6. Good V. Bad Intitution The next diviion of the ample i inpired by Dalgaard, Hanen, and Tarp [], who, a we noted in Section, argue that deep determinant of a ociety may play a big role in making aid effective. They point pecifically to intitutional quality. Thi ha enormou intuitive appeal good intitution hould lead to a well ordered and table ociety which in turn hould lead to an effective ue of reource uch a foreign aid. 9
Figure 8: Bad Intitution IRF Impule repone. 8 6.. -. - -. -.. Repone of..... -. -. -. -. 8 6 -... -. -. -......7....7.. -. -. -.. -. -. -.7 -. -.. -. -. -. -. Note: Each IRF depict the repone of the row variable to a one tandard error hock in the etimate of the column variable. The upper and lower line are 9% error band generated uing Monte Carlo imulation with draw. The ample i compoed of thoe countrie with an average core of le than in the Polity IV contraint on executive power index. To invetigate, I divide the ample uing a tandard meaure of intitutional quality. The Polity IV dataet contain a variable which quantifie the contraint on the power of the executive along a even point cale. I define countrie with an average value over the ample period of three or higher a poeing good intitution. Thi i inecapably and unavoidably arbitrary to ome degree, but the definition of point three on the cale ugget it i a valid cutoff point. Slight to Moderate Limitation on Executive Authority: There are ome real but limited retraint on the executive. Marhall and Jagger [8].
Figure 7 and 8 preent the reult which ugget that good intitution do not lead to more effective aid in term of economic growth. In fact, the IRF how very little repone of economic growth to aid. Bad intitution on the other hand, eem to yield better economic growth repone to aid in the early period following the aid hock. Compared to the baeline reult, the initial repone of economic growth i nearly twice a big. A in the baeline cae, thi i tempered by ome negative repone after period two: negative repone which are likewie larger than in the baeline cae. We can alo ee that there are ome marginal negative repone after t= that perit until t=. Overall, the repone of economic growth to aid, both poitive and negative, are larger in the bad intitution ub-ample. The tory i quite different when we turn to human development. In good intitutional environment, we ee a higher level of repone than in the full ample and the error band are more upportive of a poitive repone. In the bad intitution ample, the repone i imilar to the baeline reult, though we ee early negative repone. Why might it be the cae that bad intitutional environment get more economic bang for their aid buck and le in term of human development? A noted above, common ene would ugget that aid work bet along both dimenion in good intitutional environment. One poibility i that a they are not a contrained by legilature repreenting the popular will and thu can concentrate the aid monie in a few large cale national project rather than having many mall cale local one. Thi i mere peculation and highlight the bigget drawback of the PVAR approach while we can ee outcome quite well, the mechanim are completely obcured. 6. High V. Low Aid Dependence The final diviion i along line of aid dependency. There are no firm theoretical reaon to do o, but there are ome plauible mechanim that could relate aid effectivene to the level of aid dependency. It could be that a certain level of aid relative to the ize of the economy i needed for it effect to be ubtantial. On the other hand, it could be that high aid dependence foter corruption, retard reform and induce other uch harmful behaviour. There are alo iue of aborption that may induce macroeconomic problem which could limit aid effectivene. Thee poitive and negative potential conequence of the level of aid dependence are not necearily mutually excluive. One effect may temper the other or they could operate on different time cale. Thi reult i upported if one ue the World Bank Rule of Law (RL) variable a the meaure of intitutional quality. The Polity variable i preferred a RL i only available for relatively recent year. The reult in term of human development are quite different however, with neither group howing much repone. Reult available on requet.
Figure 9: High Aid Dependence IRF Impule repone. 8. 6. -. - -. -... Repone of.... -. -. 7.... -.... -. -. -.........8...6. -. -.... -.. -. -. -. -. -. -. -. -.6 Note: Each IRF depict the repone of the row variable to a one tandard error hock in the etimate of the column variable. The upper and lower line are 9% error band generated uing Monte Carlo imulation with draw. The ample i compoed of thoe countrie with an average nominal aid to nominal GDP ratio of % or higher. While a PVAR model cannot get at thee mechanim, it would be intereting if we could oberve outcome conitent with thee argument. There i no firm definition of aid dependence. I take any country with an average nominal aid to nominal GDP ratio of % or greater to be aid dependent. Once again thi i omewhat arbitrary but % i a large lice of any economy and the mean acro the full ample wa approximately %. Figure 9 and how the reult. The top et are for the aid dependent countrie and the bottom et for the remainder. For the mot part the reult in term of aid and economic growth are very imilar to what we aw in the full ample: an aid hock induce economic growth in the early period in both cae followed by lower growth in the next few period with a final poitive pike before tapering off. There are ome difference worthy of note. Compared to both the full ample and the le dependent ample, the level of the early poitive repone in the highly dependent countrie are omewhat higher and the ubequent
Figure : Low Aid Dependence IRF Impule repone. 7. -.. - -. - -.. 6.7 Repone of.7... -. -. -.7 -. -... -. -........ -. -. -.6 -. -. -.6 -.8 -.... -.8 -.. -. -. -. Note: Each IRF depict the repone of the row variable to a one tandard error hock in the etimate of the column variable. The upper and lower line are 9% error band generated uing Monte Carlo imulation with draw. The ample i compoed of thoe countrie with an average nominal aid to nominal GDP ratio of le than %. negative repone are not a evere. Perhap mot interetingly, the repone after t= in the highly dependent ample are negative. In the baeline cae, there wa no further repone after thi point. In the low dependence group, we can ee ome mall poitive induced growth periting until the th period. While the model offer no explanation for thee difference, it i intereting nonethele to ee them in the data. The higher repone in the dependent ample ugget that the firt hypothei above may hold. Likewie the negative effect that become apparent only in the dependent ample could be conitent with behavioural change (uch a corruption or potponed reform). With regard to human development, in the highly dependent ample, an aid hock increae the growth rate of human development in a imilar way a in the full ample, while aid hock tend to lower the growth of human development in the le dependent ample. However, the error band are in both cae are too broad for firm inference.
7 Concluion Thi paper ha been an attempt to look at whether aid i effective uing a different approach to that found in the exiting literature. Given the range of contradictory finding on thi quetion, no ingle work can claim to be definitive. However, the approach taken here i free from many of the difficultie encountered by the uual panel approach. In particular, the PVAR method doe not ret on the oft maligned intrumentation trategie ued in much of the previou work. The reult lie omewhere between the finding and argument of aid peimit and aid optimit. Aid doe eem to work in term of generating economic growth but not to an extent that could be called tranformative. The time path of the repone in mot ample howed that the early ucce of aid i mitigated by ome later negative repone before a recovery and eventual petering out. Evidence wa alo preented that ugget that thi impact i more pronounced in countrie with good economic policy. Aid doe not eem to be more effective at promoting economic development in democracie but doe appear to work better, at leat initially, in what would commonly be regarded a poor intitutional environment. Economie that are relatively dependent on aid how lower negative repone for the mot part but mall negative repone in later period that one doe not oberve in the le dependent ample. The reult for human development were generally ambiguou but ugget that aid may induce mall increae in the variable ued a a proxy for human development, the growth rate of life expectancy at birth. Thi increae wa larger and unambiguou in democracie and we alo aw a fairly unambiguou poitive repone in good intitutional environment. Aid dependent countrie ee poitive repone while the ret how a negative impact. The major failing of the PVAR method i that it offer no inight into what the underlying mechanim generating thee reult might be. That aid, much of thi literature i baed on ad-hoc peculation and econometric pecification. The benefit of the approach i that, a other VAR advocate point out, it let the data peak for itelf. In thi cae the data eem to be making two point:. Aid work in general but not all that well.. Aid eem to work better in term of economic growth in ome environment, but not to uch an extent that it would be enible to focu aid only on countrie with thoe characteritic. It i often the cae that when we ee aid failing to timulate economic growth in one of our grouping, we ee it cauing human development growth and vice vera.
Reference The World Bank. Doing Buine. The International Bank for Recontruction and Development / The World Bank, a. The World Bank. World Development Indicator, b. Craig Burnide and David Dollar. Aid, Policie, and Growth. American Economic Review, 9():87 868, September. Charle C. Chang, Eduardo Fernandez-Aria, and Lui Serven. Meauring Aid Flow : A New Approach. Policy Reearch Working Paper Serie, The World Bank, February 999. Paul Collier and David Dollar. Aid Allocation and Poverty Reduction. European Economic Review, 6(8):7, September. C-J. Dalgaard and H. Hanen. On Aid, Growth and Good Policie. The Journal of Development Studie, 7(6):7, Augut. Carl-Johan Dalgaard, Henrik Hanen, and Finn Tarp. On The Empiric of Foreign Aid and Growth. Economic Journal, (96):F9 F6, 6. Angu Deaton. Intrument, Randomization, and Learning about Development. Journal of Economic Literature, 8():, June. William Eaterly. Can Foreign Aid Buy Growth? Journal of Economic Perpective, 7(): 8, Summer. William Eaterly, Ro Levine, and David Roodman. Aid, Policie, and Growth: Comment. The American Economic Review, 9():77 78,. Walter Ender. Applied Econometric Time Serie. Wiley, nd edition,. Karuna Gomanee, Sourafel Girma, and Oliver Morriey. Aid, Public Spending and Human Welfare: Evidence from Quantile Regreion. Journal of International Development, 7 ():99 9, a. Karuna Gomanee, Oliver Morriey, Paul Moley, and Arjan Verchoor. Aid, Government Expenditure, and Aggregate Welfare. World Development, (): 7, March b. Henrik Hanen and Derek Dewey Headey. The hort-run macroeconomic impact of foreign aid to mall tate: An agnotic time erie analyi. The Journal of Development Studie, 6():877 896,. Stephen Koack. Effective Aid: How Democracy Allow Development Aid to Improve the Quality of Life. World Development, ():, January.