FUNGIBILITY, PRIOR ACTIONS AND ELIGIBILITY FOR BUDGET SUPPORT

Transcription

1 FUNGIBILITY, PRIOR ACTIONS AND ELIGIBILITY FOR BUDGET SUPPORT Abstract by Oliver Morrissey CREDIT and School of Economics, University of Nottingham Draft Jly 2005 A nmber of donors advocate providing general bdget spport as an alternative to aid-financing of projects or sector programmes. The basic principle is that if donors approve of a recipient s expenditre allocation plans it is appropriate to provide aid as bdget spport. This begs the qestion of how donors determine which recipients shold be eligible for bdget spport. Typically, recipients are reqired to demonstrate for a period that they can and do allocate spending in the way agreed with donors, and they then become eligible. This can be termed selectivity, where prior actions are sed as conditionality to determine eligibility so that bdget spport is only granted to those contries that allocate spending appropriately. This paper identifies some problems with selectivity to sggest that reliance on prior actions does not ensre that bdget spport is allocated optimally. A simple model is sed to inform the discssion by demonstrating that selectivity based on prior actions imposes excess costs on deserving recipients and may not generate the efficiency gains in aid allocation posited by proponents of targeting. This weakens the argment for selectivity (of the form commonly proposed), and we arge that it is not necessary to reqire prior actions. Opposition to general bdget spport reflects, at least in part, concerns over fngibility and over the effectiveness of aid-financed government spending. The second part of the paper provides a critiqe of fngibility as an argment against bdget spport. The paper concldes by arging that bdget spport can safely be granted if recipients allocate spending in a manner broadly agreed with donors. The isse that donors shold focs attention on is the effectiveness of spending, not on conditionality or fngibility. JEL Classification: F35, O20, O23 Keywords: Fngibility, General Bdget Spport, Prior Actions, Selectivity Contact: Professor Oliver Morrissey, CREDIT, School of Economics, University of Nottingham, Nottingham NG7 2RD, England. oliver.morrissey@nottingham.ac.k

2 1. Introdction Most aid is given to the governments of recipient contries and is sed to finance pblic expenditre. In the poorest contries, aid acconts for a large share of government spending. For example, in low income contries in 1997, aid acconted for almost a third of general government consmption on average, althogh it was over 100% for some African contries (e.g. Ghana, Senegal, Tanzania and Uganda), 1 whereas in lower middle income contries the share was abot six per cent (McGillivray and Morrissey, 2004: 74). Clearly, the impact of aid in developing contries depends on how it affects government spending and fiscal behavior more generally. Two isses are especially important: the effect of aid on the allocation of spending, and the effectiveness of that spending in delivering pblic goods and services (e.g. does health sector spending deliver improvements in health stats). For example, aid has a greater impact on growth if it increases spending on pblic investment, even thogh the prodctivity (effectiveness) of pblic investment in sb-saharan Africa is low (Gomanee et al, 2006). Recent research shows that aid increases spending on social sectors (health, edcation and sanitation) and this contribtes to poverty redction (Mosley et al, 2004) and improving aggregate welfare, althogh again in the poorest recipients the effectiveness of social spending in delivering welfare improvements is low (Gomanee et al, 2005). To enhance the effectiveness of aid, it is important to ensre that aid is allocated to the most appropriate areas of spending, and to increase the effectiveness of sch spending. Donors have tended to be most concerned abot the former isse, althogh recent work has addressed the effectiveness of spending (Devarajan and Reinikka, 2004; Reinikka and Svensson, 2004). A nmber of approaches have been adopted to restrict or inflence the areas of spending financed by aid. At one extreme is the idea of creating a special ring-fenced fnd that can only be sed for designated expenditres, e.g. nder HIPC-II the aid and savings from debt relief may be placed in a Poverty Action Fnd to be spent on headings identified in the PRSP (Morrissey, 2004). At the other extreme, donors may give aid in kind (e.g. medicines) or stiplate that recipients have to match aid-financed spending in specified areas, throgh matching grants or conterpart fnds (McGillivray and Morrissey, 2000, 2001). Underlying all of these measres to inflence the allocation of aid to expenditres is donor concerns with fngibility is aid allocated to the areas of spending that donors want to spport? 1 Aid may be eqivalent to more than 100% of government spending if the measre incldes aid that is not delivered throgh the bdget, sch as technical assistance or donor-financed projects. 1

3 Three elements of fngibility can be distingished (McGillivray and Morrissey, 2004, provide a review). General fngibility arises where aid intended for a general prpose, investment spending, is actally sed for a different prpose, consmption spending. If donors believe that aid mst finance investment if it is to impact on growth, they will believe that aid redirected to recrrent spending ndermines the (growth) effectiveness of aid. A more specific case is categorical fngibility, where aid intended for a particlar spending heading, sch as health, is sed for a different heading, in particlar one that the donor does not intend to spport, sch as secrity or wages. Again, donors will believe that this ndermines the development effectiveness of their aid. The third isse is additionality: even if the aid is all allocated to the intended expenditre, government s own (tax) resorces previosly allocated to that expenditre may be reallocated elsewhere so that spending on the intended area does not increase by the fll amont of the aid (McGillivray and Morrissey, 2000). It is important to note that fngibility per se is not concerned with the misse of aid (sch as corrption), bt simply with misallocation, or specifically non-additionality. The misse of aid is a concern, bt it is relevant to the effectiveness of aid and spending rather than to fngibility. Concern abot fngibility is one reason why many donors are relctant to spport providing aid in the form of General Bdget Spport (GBS). Cordella and Dell Ariccia (2003) se a fngibility model to compare GBS, where donors inflence the overall allocation of resorces (throgh conditionality) bt fngibility may arise, to project aid, where donors can target an area of spending bt do not inflence overall allocation. In an environment where only some expenditres are sbject to conditionality, they arge that bdget spport is preferred if i) aid levels are small relative to the bdget, and ii) donor and recipient preferences are aligned. These conditions may not often hold together in practice for the poorest contries for which aid is most important. For example, aid is significant in Uganda (as shown above), so the first condition is not met even if the second is. Frthermore, as White and Morrissey (1997) show, the latter condition is only one of at least for possible cases (if preferences are defined over both aid levels and conditions). A core argment of this paper is that this fngibility critiqe of GBS is misplaced. If donor and recipient preferences on allocation are aligned, then fngibility is not an isse. Irrespective of the importance of aid in spending, recipients will allocate aid more or less in the way donors desire and GBS is appropriate. Frthermore, as White and Morrissey (1997) show, conditionality serves no sefl prpose in this case, and may be conter-prodctive (becase nintended non-compliance with conditions may appear as intentional). On the other hand, if 2

4 preferences are not aligned, conditionality is ineffective (White and Morrissey, 1997) and fngibility is less likely to ndermine GBS if aid is a large share of the bdget. The intition here has two elements: i) it is easier to monitor the allocation of spending over broad headings than actal spending on many particlar projects, and ii) if aid is a large share of the bdget recipients have fewer own resorces to reallocate. Ths, fngibility argments do not ndermine the case for GBS to poor contries; fngibility is a red herring, in the words of McGillivray and Morrissey (2000). A more important isse in choosing GBS over project aid relates to the effectiveness of pblic spending donor projects may be more effective than government spending in poor contries an isse we retrn to in the conclsion. Aid comprises a large share of the bdget in poor contries, and clearly the impact of aid on fiscal behavior is an important determinant of the development effectiveness of aid (McGillivray and Morrissey, 2004). The remainder of this paper concentrates on two isses. First, how can donors decide which recipients shold receive GBS? Donors cannot be certain of which recipients will allocate and spend aid effectively, i.e. make the best se of GBS, bt these are the ones they want to target for GBS. Donors often resolve this problem by selecting for GBS those recipients that implement reqired prior actions, a form of conditionality. This approach is assessed in Section 2. Second, is concern with fngibility a valid argment against GBS? What do we know abot the effects of aid on the allocation and efficacy of government spending, is fngibility and serios problem and what are the implications for aid delivery? These qestions are addressed in Section 3. Section 4 concldes that donors shold focs attention on how to increase the effectiveness of spending rather than on the allocation of aid across expenditre headings. 2. A Model of Selectivity throgh Prior Actions Assme donors only want to give GBS to those contries that will allocate the aid and government spending in the manner approved by donors and in which government spending is effective. Assme frther that some recipients are less able to monitor and garantee the allocation and effectiveness of spending than others becase of factors they cannot alter (at least in the medim term), sch as administrative inefficiency or low-level corrption, and that donors cannot identify those factors with certainty. In other words, among recipients that cold be given GBS there are at least some for which donors are ncertain if they wold se GBS appropriately and effectively. Or interest is in how donors can se prior actions (a form of ex ante conditionality) to acqire information on the distribtion of types of recipients so as to 3

5 allocate GBS to the most deserving. We are not here concerned with how prior actions are decided and indeed the model abstracts from the isse of implementing actions; we assme contries will comply if they perceive a benefit, and won t otherwise. Ths, ndeserving recipients will not commit to prior actions and will not receive GBS and in this way they are selected ot so that GBS is targeted to the most deserving recipients. We sketch an interpretation of the model proposed in Bogheas et al (2005). Consider three types of developing contry recipient (D) abot which donors are ncertain, i.e. they do not know if expenditre wold be properly allocated and effective, bt they know there are three possible types. 2 The first type (s) have strong capacity (i.e. expenditre is allocated as planned and is effective), implying the ability to se GBS effectively. The second type (w) has the willingness to allocate GBS as planned bt weak capacity to ensre allocation and effectiveness. The third type () is nwilling to allocate spending as donors wish and expenditre tends to be ineffective; this type can be treated as ndeserving of GBS. Assme that the benefit of GBS aid is maximised only if spending is effective and allocation is as agreed with donors (one cold think of this as maximising joint tility from spending). Conceptally, at the beginning of the period, recipients face an expenditre allocation (e) financed by aid. If s types implement e they receive a retrn of X at the end of the period, whereas if w types invest e their retrn is X-α (where α is a measre of their expenditre inefficiency, in allocation and effectiveness, relative to s types). However, if w types ndertake adjstment by implementing a set of prior actions, which cost c, then they can increase the retrn from their expenditre. The parameter c measres both the extent and the cost of these actions, sch as costs of improving expenditre monitoring and management systems. Specifically, if w types spend e having implemented the prior actions they receive ( X α + g() c ), g ( ) = 0, g > 0, g < 0, g ( c ) = 1, g( c ) = α 0. Evidently the efficient level of adjstment is at cost c ; at this level, spending by w types yields the same retrn as spending by the s types. Adjstment activities have to be performed at the beginning of the period, prior to spending (hence, in this model, prior to receiving aid). Type economies receive zero benefit if they spend, regardless of whether they ndertake prior actions or not (zero benefit is the extreme; it is only necessary that the benefit of GBS is very low). 2 We assme only three types for convenience; adding more types increases complexity withot providing new insights. Note also that the discssion here is a specific interpretation of a qite general model in Bogheas et al (2005). 4

6 Each D economy s type is private knowledge. For any k {s, w, } the proportion of k types in the pool of D economies, assmed constant over time, is π k. N, X, and e are common knowledge, as is the fnction g( ), bt individal otpts are nobservable (e.g. even if expenditre allocation can be monitored, effectiveness is not observed). Adjstment cost, c, is only observable over intervals, either becase government actions exhibit indivisibilities or becase a donor s ability to measre adjstment activities is imperfect. As ( α > c ) for w types, adjstment is worthwhile, bt for s types prior actions impose a net cost (becase they are not necessary). The donor knows some pper bond for > 0 π, i.e., knows some vale ( 0,1) λ sch that π λ. This also implies knowledge of some lower bonds for the deserving types, i.e., that ( π π ) [ 1 λ,1) s + w. The donor does not know the actal proportions, bt has prior beliefs abot these proportions, given by a sbjective distribtion fnction H ( π, π ) correspond to the objective distribtion. Let π π s θ =, and let ( ) s, that need not F θ, π be the donor s sbjective distribtion fnction for θ, generated from H and assme that F is independent of π. In simple terms, donors can have separate views abot the proportion of types and abot the distribtion of types. This implies the spport of F mst be some sbset of 1 λ 0, λ. Sch 1 λ distribtions mst exist: for example, θ may take the vales 0, with eqal probability. λ 1 λ E( θ, π )= θ E 0,. (1) λ The donor has a total bdget of: =, where < κ = δ ( 1 π ) < 1 λ B κne 0. (2) Since the donor knows π λ, she knows δ ( 0,1), thogh she does not know its exact vale. Ths, the donor knows her bdget is not large enogh to award GBS to all effective 5

7 economies. 3 Note that the donor wold not know this nless she knew some pper bond for π. The donor s objective is to distribte the bdget among D economies, so as to maximize their expected total (or average) benefit. Sppose the donor offers GBS of e provided the recipient carries ot prior actions costing c. All types will reject this bt all s and w types wold be willing to implement the adjstment, accept the aid, and se it for spending with a retrn [ c] X. As δ < 1, not all potential recipients can be fnded. Assming fnds are allocated randomly, average conditional benefit (G C ) therefore is: G C ( 1 )( X )] = δ [ π c. (3) Ths, prior actions scceed in screening ndeserving applicants ot of the application process, and thereby eliminate leakage. This efficiency in targeting GBS (implementing selectivity, albeit of a self-selection form) comes at the cost of nnecessary adjstment by s types, which leads to a total wastage of δπ scn. If e is distribted non-selectively all D economies will then wish to receive GBS and the eqilibrim involves pooling. Given the donor s bdget constraint (2), only δ ( 1 π ) proportion of each type will receive the aid. Average realised benefit (G R ) is: G R ( π )( [ π )( X c ) + cπ eπ ] = δ (4) s Ths, non-selective GBS diverts resorces away from some s or w types, bt eliminates waste from nnecessary adjstment. Note that (3)-(4) yields: G R G C π s = δπ ( 1 π )[ θc ( X e c )], where θ =. (5) π If the term in sqare brackets is zero, G R = G C ; if positive G R > G C, and if negative G R < G C. We can define the vale of θ for which the term in brackets is zero sing (5) to get: 3 This simplifies exposition, bt we only need the bdget to be insfficient to cover all D recipients, i.e. κ < 1. 6

8 Proposition 1. Let θˆ = X e c c. Given the assmptions: (i) there exist distribtion fnctions for θ which wold make a risk netral donor prefer prior action selectivity, (ii) if λ is believed to be high, then θ E is expected to be low (π high relative to π s ), therefore θ ˆ E < θ (implying G R < G C ) and the distribtion fnction for θ wold make a risk netral donor prefer prior action selectivity (iii) if λ is believed to be low, then θ E is expected to be high (π low relative to π s ), therefore θ ˆ E > θ (implying G R > G C ) and there exist distribtion fnctions for θ which wold make a risk netral donor prefer non-selective allocation. By Proposition 1(i), regardless of the vales of the known parameters X, e, c and λ, there will always exist prior beliefs which wold jstify selectivity. Intitively, the donor need only expect the proportion of s agents to be sfficiently small. If the pper bond for the proportion of types is sfficiently high, then it is not possible to hold beliefs involving very high proportions of s types. Conseqently, all possible prior beliefs regarding the distribtion of θ, which are independent of the proportion of agents, mst jstify selectivity (Proposition 1(ii)). If it is believed that the proportion of agents is relatively low, i.e. λ is low, then one may expect the proportion of s agents to be high. Sch beliefs wold jstify non-selectivity (Proposition 1(iii)). Proposition 1(i) implies there wold exist other beliefs in this sitation that wold jstify selectivity. Under plasible restrictions on the prior distribtion of beliefs, when individal otpts are nobservable and recipient costs are sfficiently indivisible, donors favoring prior action selectivity wold not receive any additional information from the eqilibrim. They may learn which recipients are types, bt they wold not get information on the distribtion of s and w types. Conseqently, they wold have no reasonable gronds for revising their beliefs and reqiring prior actions becomes self-perpetating. In the context of this paper, the implication is that allocation according to prior actions reveals no objective information on the distribtion of types that wold facilitate more efficient selection of eligibility for GBS. 7

9 In this context, whether a donor advocates prior actions will depend on beliefs abot the proportion of recipients for which they imply significant net (excess) costs, and abot the proportion that wold not tilise GBS effectively. If a donor believes the first proportion is low (i.e., prior actions impose few excess costs on the prodctive), and/or the second is high (i.e., potential gains from targeting are large), they wold advocate prior actions. In this way, it is beliefs regarding the distribtion of types of recipient that lead donors to prefer selectivity to non-selective allocation of GBS (where non-selective here means that eligibility is not based on reqired prior actions). Evidently, the first best allocation involves ensring that as many of the s types are covered as possible, withot any adjstment. Remaining resorces shold be sed to cover as many w types as possible, with adjstment, while no type shold receive any aid (in the form of GBS). However, as type is private information, the first best allocation cannot be attained. The simple point is that reqiring prior actions may deliver the targeting benefit of exclding types bt does so by imposing an excess cost on s types. A soltion is possible if donors cold identify some observable recipient characteristic that is common and specific to each type. Of corse, if niqe indicators cold be identified for each type, donors wold face no ncertainty regarding recipient type and this model wold not be applicable. Bogheas et al (2005) arge that it is not generally possible to identify sch indicators for aid or charity in general, bt perhaps it is easier in the case of GBS. What information on performance and characteristics of recipients wold allow donors to identify those that can make the greatest se of GBS? This is the isse we now trn to by considering fngibility. Note that if one has sch information, this determines eligibility and prior actions (conditionality) are not reqired. 3. Government Spending: Fngibility, Allocation and Effectiveness In the context of broad conditionality and policy reform, it may be reasonable to assme that donors are relatively ncertain abot the type of (most) recipients. In the case of bdget spport, donors may have sfficient information to identify recipient type (as sitability for GBS), at least to classify nder one of the three types we consider. Of corse, if they can do so prior action conditionality is not reqired. Donors do not start from a blank canvas. Althogh aid modalities and the natre of donor-recipient relations have changed over the past two decades, donors have a long and well-established working relationship with the majority of recipients. Frthermore, while governments may change, the same senior officials sally remain in place, at least in Ministries of Finance. While policy reform is a slow and difficlt 8

10 process inflenced by many economic and political factors, bdget and expenditre processes are more clearly defined, both in terms of what is done and of who is responsible. Sch processes have also been improved over the past ten years or so, for example with widespread adoption of Medim Term Expenditre Frameworks (MTEFs). In principle, donors are reasonably well informed abot how well recipients monitor and allocate expenditre, and are party to the monitoring process; they know where aid goes, even if they don t know how effective spending is. In practice, it may only be the lead donors, the World Bank and IMF, and those large bilateral donors that engage with Ministries of Finance (sch as the British and Dtch) that feel well informed abot expenditre allocation and monitoring. Even if donors are collectively fairly well informed abot expenditre management, many have reservations abot bdget spport. Typically, as arged above, these reservations relate to concerns abot fngibility. In the remainder of this section we arge that sch concerns are misplaced, and may even be misleading (there may be other reasons to be concerned abot GBS, bt here we are only concerned with fngibility). There are two elements to the argment. First, fngibility is a very static concern; it ignores the dynamics of the evoltion of spending, and the broader context of the effects of aid on fiscal behavior (spending, taxation and borrowing). This literatre is reviewed in McGillivray and Morrissey (2004), and we only consider the core isses here. Second, fngibility is assmed to be an action of recipients they decide to se aid to achieve an allocation of expenditres that differs from what donors intend. McGillivray and Morrissey (2001) demonstrate that what appears to be fngibility can arise for a variety of reasons and may not be de to prposefl actions of recipients. We sketch their model as they specifically consider ways of delivering aid, sch as in kind or matching grants, that are intended to minimise fngibility. There are many problems with stdies of fngibility (McGillivray and Morrissey, 2000, 2004). Typically, analysts focs on the sector allocation of aid in a particlar year, ignoring dynamic (or lagged) effects. Given the problems of measring sector aid (there is rarely good data on how donors intended the aid to be allocated), slow disbrsement and the many implementation problems that will occr in spending ministries, it is not srprising that stdies find a weak, if any, correlation between sector aid and sector expenditre in any given year. The stdies do not explicitly allow for the effect of aid on tax effort or borrowing (the fiscal effects) or on the evoltion of total or sector spending. It may well be the case that in any given year expenditre otcomes by sector do not correspond closely to aid allocations. If, over time, spending on the headings favored by donors do increase, then how mch of a concern is 9

11 fngibility? (McGillivray and Morrissey, 2004: 80). If one takes a longer term view, it is the evoltion of expenditre that is important, and whether this is in the direction (and at a speed) acceptable to donors. Donors only need to monitor expenditre, sch as the MTEF, rather than exercise concerns over fngibility. Bilding on the pblic choice literatre on fiscal illsion, McGillivray and Morrissey (2001) propose the concept of aid illsion whereby failres in the transmission of information from donors throgh government policy officials to spending officials (often at the local government level) weaken the links between aid and spending. They propose two levels, negotiation of the parameters of aid (e.g. block grants verss matching grants) between donors and recipient policy-makers, and implementing associated spending (a relationship between recipient policy and spending officials). Aid illsion can be shown to generate, inter alia, apparent ex post fngibility even where preferences are aligned (i.e. policy officials did not intend fngibility) or over spending (total spending increases by more than the amont of aid, analogos to a flypaper effect, an effect observed in some stdies of the fiscal impact of aid). Conventional treatments of fngibility consider interaction between donors and recipients, where the latter incorporates both the officials who make spending plans, and in doing so may treat aid as flly fngible, and those who implement the plans. A richer analysis is provided by treating both levels separately. McGillivray and Morrissey (2001) focs on the actions of the implementing official with ltimate responsibility for expenditre on specific headings. These officials are responsible for allocating a given amont of aid within a spending category, bt need not know the share of aid in the total revene pool at their disposal and may not be flly informed of the bdget terms on which aid was granted. The finance ministry, or policy officials, draw p the bdget and set expenditre plans; they may, or may not, wish to treat aid as flly fngible. The paper allows for possibilities that have been observed in empirical research on the impact of aid on pblic sector fiscal behavior. Aid can lead to greater than proportional increases in total expenditre, tax and other recrrent revene can rise or fall simltaneosly, and the conditions for fll fngibility can be present. They consider aid in kind and two scenarios for matching grants: 1) Donors give aid in kind (e.g. medicines) to ensre it is allocated in the intended area. Policy officials intend fngibility, by planned reallocation of their own resorces. However, spending officials perceive aid in kind as redcing the price of a good they reqire or distribte, increase their demand and overall spending on the sector increases by at least the amont of the aid. Ex post there is no fngibility, althogh it was intended. 10

12 2) Donors offer a matching grant and policy officials agree with the donors spending allocation. However, spending officials misperceive the matching grant as a block grant and therefore do not spend as mch as intended. Ex post there is apparent fngibility althogh it was not intended; apparent fngibility can arise from implementation rather than intentions. 3) Donors offer a matching grant bt policy officials intend fngibility. However, spending officials perceive the matching grant correctly and spending increases by more than the vale of the aid. intended. Again, ex post there is no fngibility, althogh it was intended. This highlights the fact that it is the action of the spending officials that really matters. The aim of the analysis is to demonstrate that nintended otcomes can reslt from misperceptions or illsions regarding either the real or nominal vale of the aid inflow, or the way in which the aid is delivered. The analysis is not intended to describe what does happen, rather to highlight the complexity that arises when many levels of officials are involved, 4 specifically that observed fngibility may not be de to prposefl actions. The implications of the argments in this section are that a focs on conditionality may i) lead donors to incorrectly infer that aid is misallocated becase it detracts from considering the evoltion of expenditre allocation over time and ii) lead donors to attribte misallocation to the intentions of officials when it is actally de to the complexity of expenditre management and implementation. This does not sggest that intentional fngibility never occrs, bt does imply that is sfficient for donors to monitor the evoltion of expenditre plans and otcomes. 4. Conclsions and Policy Implications This paper addresses two isses related to providing aid in the form of general bdget spport. First, is conditionality (reqiring prior actions) an effective mechanism for selecting which recipients shold be eligible for GBS? Second, are concerns abot fngibility solid reasons for not providing GBS? The model in section 2 was simply sed to demonstrate that, when donors are ncertain abot recipient types, the reqirement for prior actions (conditionality) is an imperfect selection mechanism. It may exclde the least deserving recipients bt does so by imposing an excess cost on the most deserving recipients, and does not help in identifying the 4 Note that complexity is increased when there are many donors with different preferences and reqirements for aid delivery and monitoring, and also when spending is more decentralised. 11

13 latter. The argment in section 3 was that nde focs on fngibility can detract attention from what is really happening. It is sefl to draw a distinction between fngibility, where donor and recipient spending preferences differ (by definition), and aid illsion where both donors and recipients have imperfect processes for allocating and monitoring aid expenditres. Most importantly, the analysis highlights that what really matters is ensring that the officials responsible for spending have fll and correct information. Taken together, the argment is that donors need only be concerned that recipients share their spending allocation preferences. This can be determined by observing expenditre allocation plans and otcomes, a monitoring process that has become almost a rotine featre of donorrecipient relations. If preferences are aligned, GBS can be granted: conditionality is not reqired (and is not even helpfl) and fngibility concerns are not relevant. If preferences are not aligned, then GBS is not appropriate, and again conditionality is nlikely to serve a sefl prpose (as it will not align preferences). If donors want to deliver aid throgh the bdget, they can se mechanisms sch as aid in kind or matching grants to minimise fngibility (bt it is important to ensre that spending officials are flly aware of the aid modalities). Alternatively, donors cold choose to fnd projects directly. The argment in section 2 is an example of a more general critiqe of conditionality for selection. Morrissey (2005b) ses the model to discss conditionality and selectivity more generally, and bilds on Morrissey (2005a) to arge that a partnership bilt on dialoge and monitoring the se of aid is preferable to selectivity based on conditionality. The World Bank is also adopting more flexible, simple and transparent approaches to conditionality, advocating ownership and partnership and an approach based on reptation and reslts (Koeberle, 2005: 66). However, prior actions are still viewed as an important element of conditionality, if only to signal commitment to a particlar direction of policy. Emphasising ownership does not necessarily resolve the inefficiency of prior actions as a basis of conditionality otlined in Section 2. For example, a recipient may commit to a particlar policy action not becase it believes it is the best option bt becase it knows the donor favors that action. Frthermore, althogh monitoring is seen as important, it is not sally interpreted simply as monitoring the way in which aid is sed bt rather monitoring has typically focsed more on compliance with ex ante conditionality than on progress, otcomes and poverty impacts (Koeberle, 2005: 74). Conditionality may have a role to play, bt or argment is that it shold not have a role in determining eligibility for GBS. 12

14 Even if preferences are aligned, GBS is not necessarily better than project aid as this depends on the effectiveness of pblic spending donor projects may be more effective than government spending in poor contries. For example, Gomanee et al (2005) show that althogh aid increases government spending on social sectors in low income contries, government social spending is not effective in delivering welfare improvements, whereas aid is effective (perhaps throgh projects). On the other hand, government social spending is effective in middle income contries, bt aid has no discernable effect on the level of social spending (perhaps becase aid is a small share of spending). This is an empirical qestion, and we are not aware of mch evidence comparing the effectiveness of donor project spending verss government spending. The practical implication is that donors cold contine with project aid (where they can demonstrate that it is effective) even where mch aid, or some donors aid, is channelled throgh GBS. This may appeal to some small donors who are relctant to commit to bdget spport. Belgim, for example, prefers project aid becase that is where the aid agencies experience lies and they feel nable to play an inflential role on general bdget allocation (Holvoet and Renard, 2005: 143). In conclsion, the argment of this paper is that it is appropriate to give aid as GBS in those contries where expenditre allocation is monitored and seen to be broadly aligned with donor preferences. Donors shold have sfficient experience and information to be able to identify sch contries. Prior actions or conditionality for receiving GBS are nnecessary, and there is no need to be concerned with fngibility. However, it is important that donors focs attention on improving the effectiveness of government spending to ensre that GBS is sed effectively. There will contine to be a role for donor project aid in contries where government spending has low effectiveness, althogh donors shold be able to demonstrate that their project spending is effective. In contries where donor and recipient expenditre preferences are not aligned, there is little jstification for GBS; the implication is that in these contries more effort and dialoge is needed to align spending and, most likely, improve expenditre planning and monitoring. Conditionality will not necessarily improve this process. There are still ways that donors can inflence where aid is spent, sch as by providing aid in kind and matching grants, bt the sccess of this will depend on getting correct information to officials responsible for spending. Bdget spport is not always and everywhere appropriate, bt increasing the effectiveness of spending is a priority in all recipients. 13

15 References Bogheas, S., I. Dasgpta and O. Morrissey (2005), Togh Love Or Unconditional Charity? University of Nottingham, School of Economics, CREDIT Research Paper 05/08. Cordella, T. and G. Dell Ariccia (2003), Bdget Spport vs Project Aid, IMF Working Paper WP/03/88. Devarajan, S. and Reinikka, R. (2004), Making services work for the poor, Jornal of African Economies, 13 (spplement 1), Gomanee, K., Girma, S. and Morrissey, O. (2006), Aid and Growth in Sb-Saharan Africa: Acconting for Transmission Mechanisms, Jornal of International Development (forthcoming). Gomanee, K., Morrissey, O. Mosley, P. and Verschoor, A. (2005), Aid, Government Expenditre and Aggregate Welfare World Development 33:3, Holvoet, N. and R. Renard (2005), Belgian Aid Policies in the 1990s, in P. Hoebink and O. Stokke (eds), Perspectives on Eropean Development Co-operation, London: Rotledge (Research EADI Stdies in Development), pp Koeberle, S. (2005), Contitionality Under What Conditions, chapter 5 Koeberle, et al (eds), pp Koeberle, S., H. Bedoya, P. Silarszky and G. Verheyen (eds) (2005), Conditionality Revisited: Concepts, Experiences and Lessons, Washington, DC: The World Bank, McGillivray, M. and O. Morrissey (2000) Aid Fngibility in Assessing Aid: Red Herring or Tre Concern?, Jornal of International Development, 12:3, McGillivray, M. and O. Morrissey (2001), Aid Illsion and Pblic Sector Fiscal Behavior, Jornal of Development Stdies, 37:6, McGillivray, M. and O. Morrissey (2004), Fiscal Effects of Aid in T. Addison and A. Roe (eds), Fiscal Policy for Development: Poverty, Reconstrction and Growth, Basingstoke: Palgrave Macmillan in association with UNU-WIDER, pp Morrissey, O. (2004), Making Debt-Relief Conditionality Pro-Poor chapter 12 in T. Addison, H. Hansen and F. Tarp (eds), Debt Relief for Poor Contries, Basingstoke: Palgrave Macmillan in association with UNU-WIDER, pp Morrissey, O. (2005a), Alternatives to Conditionality in Policy-Based Lending, chapter 25 in Koeberle, et al (eds), pp Morrissey, O. (2005b), Whither Conditionality? Selectivity verss Monitoring, paper presented at the UNU-WIDER Jbilee Conference, WIDER Thinking Ahead: The Ftre of Development Economics, Helsinki Jne 2005 ( Mosley, P., Hdson, J. and Verschoor, A. (2004), Aid, poverty redction and the new conditionality, The Economic Jornal, 114, F217-F243. Reinikka, R. and Svensson, J. (2004), Efficiency of pblic spending: new microeconomic tools to assess service delivery in T. Addison and A. Roe (eds), Fiscal Policy for Development: Poverty, Reconstrction and Growth, Basingstoke: Palgrave Macmillan in association with UNU-WIDER, pp White, H. and O. Morrissey (1997), Conditionality when Donor and Recipient Preferences Vary, Jornal of International Development, 9:4,