Public education spending and poverty in Burkina Faso: A Computable General Equilibrium Approach 1

Transcription

1 Public education spending and poverty in Burkina Faso: A Computable General Euilibrium Approac 1 Lacina BALMA a, W. Francine Alida ILBOUDO b, Adama OUATTARA c, Roméo KABORE d, Kassoum ZERBO e, T. Samuel KABORE e a International Food Policy Researc Institute (IFPRI), balmalacina20@yaoo.fr, tel.: (00221) b Direction Générale de la Promotion de l'économie Rurale (DGPER) c Projet d'appui aux Filières Agro-sylvo-pastorales (PAFASP) d Banue Ouest Africaine de Développement (BOAD) e University of Ouagadougou Abstract Te paper analyses welfare and poverty implications of different options for creating and using fiscal space for public education expenditures. Te analysis uses a multi sectoral computable general euilibrium model calibrated for Burkina Faso. Education is demanded by ouseolds as an investment to transform unualified workers into ualified workers. Te simulations indicate tat a 40% across-te-board increase in public subsidies for primary education, financed by an increase in taxes on ouseold income and sales taxes, not only leads to an increase in welfare but also to a decline in te incidence of poverty for all ouseold types. Keywords: computable general euilibrium model, public education spending, Burkina Faso JEL Classification: C6, H5, I2, I3, J2, O5 October We acknowledge wit gratitude te financial and tecnical support of te Poverty and Economic Policy Researc Network, wic is financed by te Australian Agency for International Development (AusAID), and by te government of Canada troug te International Development Researc Centre (IRDC). We would like to tank Véroniue Robicaud, Hélène Maisonnave, Marie-Hélène Cloutier, Bernard Decaluwé, Jon Cockburn, Ismael Fofana, Kofi Nouve, Epipane Adjovi, François Bédia Aka, Mbaye Faye, Yousouf Koné and all te participants of te 7 t PEP conference in Manila (2008) and te 8 t conference in Dakar (2010) for teir support and various comments. 3

2 1. Introduction Education is often claimed as a source of uman capital accumulation and te engine to spur economic growt in te long term, and undeniably allow poor to escape poverty. It is terefore considered as a central concern for policy makers. Yet, often governments do not ave te appropriate tools to enable tem to establis priorities between various demands for funding subject to teir tigt fiscal constraints. Education is a major concern of te government of Burkina Faso as evidenced by its recent reform aiming to progressively implement compulsory and free education for te scool year. Te second reform covers te to periods and aims to move towards consolidation of universal primary education. According to te 2010 public expenditure review, total spending allocations to primary education amounted to 15.3% of te public budget in 2008 as opposed to 11.6% for ealt. Major reforms are also being carried out to improve te management of public spending in te framework of te Poverty Reduction Strategy Plan (PRSP), troug a Public Expenditure Review (PER). 2 Since 2000, te government as taken te initiative to carry out a number of sectoral PERs for ealt, primary education, infrastructure, rural development, secondary education and justice. Four major callenges are targeted via te PRSP: acceleration of euitable growt, improvement of social services, income creation for te poor and improved governance. In suc a context, te government must coose among numerous domestic policy options and state te effects tat tese coices ave on te provision of social services wic are essential for te population s welfare. Te main objective of tis study is to assess wat different options for creating and using fiscal space for public education spending imply for te ultimate objectives of reducing poverty and increasing welfare of te population. To acieve tis goal, we use a computable general euilibrium model. Te elements of tis metodology are presented in section III, were we focus attention on bringing education into te model. In sections IV and V, we paint a picture of te Burkinabé economy wit te data used for tis study and analyze te distributive effects of a simulated increase in public subsidies for education. To get tere, we 2 A PER responds to te government s expressed need to analyze canges in budget credits and expenditures since 1998 and to account for te results and sortcomings in order to improve links between te budgetary processes and te PRSP. It includes four capters wic address te following uestions: (a) study of te allocation of budget credits and budget execution over and teir link wit PRSP priorities; (b) institutional examination of program budgets and development of options to better integrate te budget and te budget review process; (c) review of budget execution in te ealt sector; and (d) review of budget execution in te education sector.

3 will first ave to present in section II, an overview of te teoretical and empirical literature on education, uman capital and teir effects on welfare. 2. Survey of te Literature A key study in te analysis of te impact of education on poverty, in a context of computable general euilibrium modeling, is tat of Savard and Adjovi (1998). Tis study introduces externalities to education and ealt markets witout accounting for eiter te benefits tat tese services provide to ouseolds or teir inability to sift demand for tese services in response to public policy. Agénor et al. (2002) also tackled tis uestion using a recursive dynamic general euilibrium model to analyze te impact of structural adjustment policy on poverty and income distribution. Te labour market is segmented into ualified and unualified labour, and unualified workers can become ualified. Te decision to acuire te ualification depends on tree factors: (i) te expected real wage differential between ualified and unualified labour markets; (ii) te stock of public capital in education; and (iii) te extent of te credit constraint expressed in terms of te average wealt of an unualified worker. However, te enormous complexity of tis model poses a callenge for advanced understanding of te underlying mecanisms at play. In anoter recursive dynamic general euilibrium model, Agénor et al. (2005) allow ouseolds to transform labour into skilled labour via a ualification production function wic is assumed to depend on te total volume of labour and te public stock of capital in te preceding period. However, teir model only includes one representative ouseold, wic means tat canges in te composition of labour in te economy are tose of te sole aggregate ouseold. Jung and Torbecke (2003) adopt a simpler dynamic model wit te goal of analyzing education-growt-poverty linkages in Tanzania and Zambia. Tey account for tree types of workers: uneducated, mid-educated (aving completed primary scooling) and igly educated (aving completed any level of education beyond primary). Tey specify a ualification production function as follows: MS m = AS m.eg βm 1. (s l.w l0 ) βm 2 were MS m is te level m of education in a given period (year). AS m is a scale parameter, EG is public education spending and s l W l0 is te opportunity cost, defined as te ouseold effort supplied (or time invested) s l in education, multiplied by te wage W l tey will receive if tey decide not to pursue an education.

4 Houseolds coose teir level of effort to maximize te value of wealt given current and expected future wages, te interest rate and te unemployment rate. Teir model as one important sortcoming toug: te constant sare of ouseold endowments in unualified, mid-ualified and ualified workers. In effect, altoug uman capital accumulation in te economy is a function of ouseold coices, te ualification acuisition process in teir model does not allow ouseolds wo wis to cange teir labour composition to do so. Tis limits te impact of education on te labour market, and results in an incomplete analysis of its distributive effects. Bourguignon et al. (2006) develop a model to evaluate te MDGs and to estimate te time and expense needed to reac tem. Tey introduce feedback effects into te economy via te labour market. Education is disaggregated by level of education, and student s beaviour during eac level is determined by a logistic function linking te performance of te education system to a group of identifiable factors. Te labour market is tus affected by te education system because labour evolves as a function of te education system and its performance in eac period. Applications of tis approac are provided by Logfren and Diaz-Bonilla (2006) for Etiopia and by Maisonnave and Decaluwé (2009) for Sout Africa. However, since te model is used to analyze te feasibility and costs of reacing te MDGs by 2015, it reuires accurate istorical data. Following a coort of students reuires data for eac level of education tat te coort passed troug. For example, if te base year is 2000 and completing a given level of education takes 10 years to complete, data is needed for eac level of education starting in Moreover, since te model seeks to capture interactions between public education and ealt spending, it is difficult to formulate ypoteses or assign elasticity values to te feedback effects. Dabla-Morris and Matovu (2002) developed a nested dynamic computable general euilibrium model wit eterogeneous agents for Gana. In teir model, all ouseolds attribute importance to bot family consumption and te uman capital of teir progeny. Parents beave altruistically, and determine te amount of time teir cildren sould spend in scool using a cost-benefit analysis. Education increases te cild s uman capital and future wages on te labour market, as well as te parent s utility. However, te fixed cost per level of education means tat suc expenditures reduce te ouseold s income and consumption. Te benefits of education depend on te ualities and skills of te cild, and acuisition of a ualification is endogenized to account for te decision to educate te ouseold in eac period. However, teir model only as a single representative sector, wic varies according to te intensity of eac type of labour and te associated wages. Tis is a

5 weakness of teir model because ouseold income is generally influenced by teir factor endowments and sectoral demand for tese factors. Finally, Cloutier et al. (2004) constructed a static computable general euilibrium model to study te impacts of public education spending on poverty, welfare and ineuality in Vietnam. Teir approac is particularly interesting because it introduces a ouseold endowment of ualified and unualified workers tat is flexible for eac ouseold category. Te ouseolds decision to invest in education results from a trade-off between future benefits (iger income) and te direct and indirect costs of education. Te government can tus influence te ouseold decision by reducing te ouseold cost of education, wit a resulting increase in education spending. By devoting more time and effort to education, ouseolds can alter te composition of teir internal labour force and te associated returns to labour. At te same time, tey influence teir labour income and tus teir total income, permitting an evaluation of te ultimate effects on poverty and welfare. Despite tese innovative caracteristics, te model is readily subject to critiue, given tat: (i) it is static and can only partially capture te intrinsically dynamic effects of an education policy and (ii) it is underpinned by te assumption of a representative agent, wic limits its significance. Te analytical framework in tis study is based on tat of Cloutier et al., described above. It is static and is centered on te assumption of a representative agent. However, it is distinguised by a greater level of disaggregation of representative ouseolds and differently defined labour market segmentation. In te Burkinabé context, ouseold education decisions regarding education are made on primary education more so tan iger education. Our definition of te levels of education or ualification conforms to tis specification. 3. Metodology Te metodology is addressed in tree steps: te first deals wit te CGE model, te second deals wit oter model parameters and te tird analyzes poverty, welfare and te income distribution. Step 1: Te CGE model Te study is based on an arcetypal model developed by Decaluwé et al., applicable for a small, open economy wit exogenous world prices. Te model includes 10 production sectors from te (aggregated) 2004 Social Accounting Matrix (SAM). Tese sectors 3 include tree agricultural sectors, tree industrial sectors, tree public services sectors and a private 3 Eac sector produces a single good.

6 service sector. It also includes two categories of labour or education levels: ualified labour and unualified labour. Te model includes income and expenditures for six representative ouseolds from te ouseolds survey carried out by te National Institute of Statistics and Demograpy (NISD) in Most of te assumptions are tose of a standard CGE model. Some important particularities are wort mentioning toug. Te empasis is placed on euations wic bring uman capital and ualifications into te model. We also assume tat ouseold labour endowments are flexible, in te sense tat ouseolds can modify teir own endowments. Te ouseold education decision is entirely explained by te investment pattern, modelled as follows: ouseolds can alter te sare of ualified and unualified labour via education in order to maximize teir income. Te coice of tis sare determines te amount of time allocated to labour and education. Tis ten influences ouseold income, tus affecting consumption and te economy s total production, wic is itself a function of uman capital. 4 Houseold demand for education depends on relative wages for ualified and unualified workers, te opportunity cost of olding ualified work (i.e., tat of spending more time in education) as well as te direct cost of education. Only completion of primary education (or beyond), produced by te primary education sector, allows workers to become ualified. We suppose tat ouseolds must buy a given amount of primary education to old a ualified position. As for te number of units of iger education (secondary, post-secondary or university) consumed, tis is presumed fixed. 5 6 Te cost of producing primary education is paid in part by te government troug public education spending and in part by ouseolds. 3.1 Houseolds Te beaviour of eac ouseold is addressed in two steps: eac ouseold obtains utility from consumption of goods and attributes no value to leisure. 7 4 See Cauc and Micel (1996). 5 Higer education as a notable influence on wage remuneration (Lacaud, 2003). It is not, owever, a priority for ouseold education decisions. Te primary participation rate, despite its progress (72% in 2008/2009) wit te strong presence of te state and bilateral and multilateral partners, remains a preoccupation in terms of te goal of universal primary education. We can tus consider tat it is reasonable to assume tat primary education is te ultimate concern of ouseolds and te government. 6 Tis is a sortcoming of te model because a public policy wic allows ouseolds to invest in basic education sould continue beyond te primary level. In effect we could suppose tat among te large number of students wo complete primary following te increase in te government subsidy, some continue on to secondary and postsecondary studies. Endogenizing te secondary and post-secondary scooling coices, all te wile allowing ouseolds to invest in eac level of education, would be an important extension. 7 Incorporating leisure into tis model is neverteless an interesting extension. In effect, if te consumption aspect of education is account for, i.e., if education directly contributes to te ouseold s utility, ten education and leisure sould become substitutes because ouseolds obtain utility from leisure and sould be considered jointly. Eac ouseold s available time sould tus be divided into tree parts: leisure time, work time and study time.

7 (,...,,..., ) U = U C C C [1] 1 i n were U represents a Cobb-Douglas- or Stone-Geary-type utility function maximized by eac ouseold in order to determine its consumption constraint, yielding a linear demand system. C i, of eac good, subject to its budget In te standard model (no education) of Decaluwé, Cockburn and Robicaud, eac ouseold possess fixed endowments of ualified and unualified labour, capital and land. Te ouseold tus as no control over its income (or te resulting budget constraint). Its income is comprised of returns to tese production factors (labour, capital and land), dividends and transfers (governmental and oter). In oter words, te ouseold cannot react to a cange in returns to tese assets. Te consumers coice (satisfying utility maximization) is tus te only decision tat is modelled. Here, te ouseold faces an additional decision because education is also considered. It must coose te sare of adult members it wises to keep in eac category of labour (ualified and unualified). 8 Houseolds can transform unualified labour into ualified labour by consuming a predetermined uantity of units of primary education. Oterwise stated, in te long run, "consuming" enoug primary education leads to a larger sare of ualified labour in te ouseold. Tis additional coice allows ouseolds to influence teir labour income and to determine teir investment expenditures in primary education. As in te standard case, owever, te ouseold as no impact on oter sources of income (factor payments to capital and land, dividends, transfers, etc.). To start wit, since income only depends on endowments of ualified and unualified labour, and te time (or effort) invested in education, ouseold cooses te sare of ualified ( δ ) and unualified (1-δ ) workers to maximize teir labour income subject to te constraint of imperfect transformation between ualified and unualified labour. In oter words, and in conformity wit te Cloutier et al. (2004) model: Max YH = W (1 δ ) LS + W (1 s) δ LS sβ Pc δ LS δ n, edb edb + non labor income - Pc ED eds, eds [2] s. l. c. LS = B 1 l l β [ δ LS ] k (1 β )[(1 δ ) LS ] k + k [3] l l l l YH Net education spending by ouseold ; 8 All workers must be eiter ualified (active or studying) or unualified.

8 W W n eds Qualified wage; Unualified wage; Pc Consumer price of a unit of iger education; Pcedb LS Consumer price of a unit of primary education; Houseold s potential labour, i.e., volume of labour; δ Sare of ualified workers in ouseold ; 1 Sare of unualified workers in ouseold ; δ β, edb Sare of unit price of education paid by ouseold ; l B l β Scale parameter for te constant elasticity of transformation (CET) of labour; Distributive parameter for te labour CET function; l k Transformation parameter for te labour CET function; s Sare of active life allocated to primary studies to become ualified; ED, eds Volume of iger education demanded by ouseold. Income from unualified labour is represented by W.(1 δ ). LS, te product of te n unualified wage and te number of unualified wage earners in te ouseold. Income from ualified labour, W.(1 s) δ. LS, for its part, is a function of te ualified wage, te potential number of workers δ. LS and te sare s of active life te worker must invest in primary education. We tus also observe a long-term euilibrium 9 were s % of te ouseold s potential ualified labour is in scool in any given year. Tis allows te ouseold to maintain its desired sare δ of ualified workers. Oterwise stated, in order to increase its ualified labour endowment by δ, ouseold must increase te number of units of primary education it consumes by s δ LS education increases in te process.. Te amount of potential working time invested in Tis leads to te first cost of pursuing education (or of aving more ualified labour): te opportunity cost W. sδ. LS. Tis is a function of te ualified wage because a few additional years of education increases wages but decreases te sare of teir active life 9 We analyze te problem from a long term perspective in a static framework witin wic te volume of students is proportional to te volume of paid ualified labour in eac year.

9 spent on te ualified labour market. 10 At te aggregate level, te uantity of ualified labour is tus lower tan its potential level due to te amount of time spent at scool. Paid ualified labour is tus (1 s) δ. LS and te number of students is sδ. LS costs tus influence te ouseold s coice.. Wages and opportunity Te ouseold is also influenced by te direct cost of education, sβ, edbpcedbδ LS. Since primary education is partially subsidized by te government, te ouseold only pays a sare β, edb of te total cost of its primary education, wic as a unit price of Pc edb. Oterwise said, eac unit of primary education provided as a total cost of: CEDT, edb =, edb were β +TED [4] CEDT, edb Total unit cost of primary education; TED Public unit cost of primary education; β, edb Private unit cost of primary education. Tis cost is sared (euation 4) between a public cost or subsidy and a private cost wic includes scool fees paid by ouseolds, contributions to parent committees, transportation and ousing costs if te student must move or needs tutoring, etc. Te total cost and te government subsidy (public cost) are presumed exogenous wile te private cost is endogenous. As a result, an increase (decrease) in government subsidies for primary scooling, all else eual, leads to a decrease (increase) in bot private costs β, edb and direct costs. Te sare of active life spent in studies, te proportion of te unit cost covered by te ouseold and te price of eac unit of education tus influence te coice of δ and 1-δ via te direct cost of education. Te income maximization constraint (euation 3) plays an 10 In te case of a standard (dynamic) uman capital accumulation model (ex: Ben Porat, 1967), te opportunity cost of education is expressed in terms of te wage for unualified workers. Tis results from te fact tat, in order to increase a ouseold s level of uman capital, unualified workers ave to be witdrawn from te unualified labour market and sent to scool. In te case of te present study (comparative static) te opportunity cost takes a somewat different meaning. Te ouseold s trade-off is between a longer period of work (witout investing time in education) at a lower wage and a sorter duration of work at a iger (ualified) wage. Since ouseolds determine teir optimal sare of ualified labour, rater tan te optimal time to invest in education, we are tus faced wit an opportunity cost of aving more ualified labour rater tan te classic opportunity cost of education. As a result, tis cost sould express te fact tat aving more ualified labour implies a smaller sare of active ualified workers. In effect, if te wage is iger, aving more ualified labour is more beneficial (a iger wage for eac year in work) but is also more costly (more years at study, fewer working years working for tis iger wage).

10 important role in modeling te education decision. It represents te limited opportunities to acuire ualifications. Witout tis constraint, even te sligtest cange in te benefits or costs of education would encourage te ouseold to allocate all of its labour to one or te oter of te types of labour. Te ease wit wic ouseolds are in a position to complete tis transformation depends on te transformation elasticity 11 associated wit te CET function and, as a result, of te value of te transformation parameter Wen coosing δ and 1-δ, te ouseold makes a trade-off between te benefits of aving more ualified labour in te ouseold and te costs of education (bot direct and opportunity costs). Te resulting coice function is: l k. δ. LS = τ l W s s Pc relative gain relative cost l (1 ) β τ, edb edb l l β (1 β) (1 δ ) LS Wn W n [5] δ 1 δ = net gain W sw sβ, edbpc edb l l β (1 β) Wn Wn W n { { skills premium opportunity cost direct cost τ l τl [6] were l τ is te constant elasticity of transformation, l τ = 1 ( k l 1). Te benefit of getting educated (accounting for te time invested in education) in tis euation is W (1 s) Wn wile te cost is β, Pc W s edb edb n. Tus, if te benefit increases by substantially more tan te direct cost, relative to te unualified wages, all else eual we sould expect ouseolds to increase teir endowment of ualified labour according to te elasticity and teir initial labour endowment. Once δ and 1- δ are known, te ouseold supplies a uantity (1 δ ) LS of unualified labour to production activities and a uantity (1 s) δ LS of ualified labour (non- students, i.e., te active sare). Remaining potential ualified labour, sδ. LS individuals wo are in scool and are tus inactive on te labour market., refers to 11 Cf. step 3 from te metodology section for te coice of tis elasticity.

11 Finally, aving defined te budget constraint, te ouseold must ten maximize teir utility function as usual in order to determine teir consumption of various goods and services (oter tan te two education services, wic do not bring any utility to te ouseold). 3.2 Government Te government s income is comprised of taxes collected and transfers from te rest of te world (troug bilateral and multilateral cooperation). It allocates between public spending and transfers, and tus determines its deficit or its current surplus. In te case of a surplus, additional funds can ten be used to finance investment. Since primary education is partially subsidized by te government, a sare of its spending is devoted to tis sector. Suc spending is endogenous because it depends on ouseolds demand for education. In effect, te government subsidy for primary education demanded by tese ouseolds acts as a per-unit subsidy. Te volume of government consumption ( G edb ) of primary education is defined as: G edb = TED s LS δ [7] Public spending on education does not directly produce uman capital. Rater, it reduces te private cost (and tus te direct cost) faced by ouseolds, encouraging tese ouseolds to invest more in primary education. Higer education is also subsidized. However, since ouseold demand for iger education is presumed fixed, te amount of tis level of education covered by te government is exogenous, as opposed to te case for te volume of primary education, wic is endogenous. Te government budget constraint can be written as: SG = YG Pc G Pc G Pc G transfers [8] SG eds edb edb eds ser Government savings G eds Government consumption of iger education YG Government income G ser Government consumption of services oter tan education P c Consumer price of a unit of service. s e r Government savings ( SG, te current budget surplus), revenues and spending (oter tan on primary education) are eld constant by using sales taxes on consumer products to cover spending. We also compare te results wit alternative compensation mecanisms. ser

12 3.3 Production factors Labour, land and capital are te production factors. Land is specific (immobile) and exclusive to te agricultural sector, wit returns to land depending on demand from agricultural sectors. Since te economic effects of education, te central concern of tis study, are largely felt over te long run, we model capital as mobile between production sectors. Tis results in a single rate of return to capital across te economy. Labour is divided into unualified (not aving completed primary scool) and ualified (aving completed te primary scool or beyond) labour. Workers are mobile between sectors, resulting in a single wage for eac type of work in eac sector. Sector-specific wage variations after te sock are te result of sector-specific labour demand and te relative scarcity of eac type of labour. Wage flexibility allows for te labour market clearing. Algebraically, te euilibrium conditions are: LNQ = LQA = LDNQi [9] i LDQi [10] i were LNQ Volume of unualified labour in ouseold and LNQ = (1 δ ) LS ; [11] LQA Volume of ualified labour in ouseold and LQA = (1 s) δ LS ; [12] LDNQ i Demand for unualified labour by sector i; LDQ i Demand for ualified labour by sector i. Given te flexibility of education and te ouseold s labour endowment, te relative scarcity of tese two categories is affected by ouseold education decisions. For example, an increase in ouseold education demand, all else eual, would: (i) increase te supply of ualified labour; (ii) decrease te supply of unualified labour and (iii) decrease te volume of active labour supplied. Te expected effects on wages would tus be an increase in wages for unualified workers relative to ualified workers. 3.4 Production sectors Eac production sector uses a constant returns tecnology and is perfectly competitive. Te output results from te combination of a fixed proportion of value added and intermediate consumption. As for value added, it is represented by a nested constant elasticity of substitution (CES) function of composite labour and capital. Composite labour is itself a

13 CES function of skilled and unskilled labour. As a result, a cange in te composition of ouseold labour supply resulting from a new policy would influence te volume of composite labour and te composite factor as well as tat of value added and final output. 3.5 Euilibrium and macroeconomic closure Te macroeconomic closure of te model is as follows: - Markets for goods and services are perfectly competitive; prices are determined by supply and demand adjustments. Consumption prices, owever, are modified by state intervention via fiscal policy. - We assume full employment 12 for labour, remuneration for wic is set by supply and demand adjustments on eac market (for ualified and unualified labour). - Government savings and te current account balance are eld fixed by endogenous adjustment of taxes (on consumption and ouseold income) and te consumer price index. - Real investment (in volume) is financed by savings of ouseolds, te government and te rest of te world. Step 2: Oter model parameters Te model uses parameters from external sources, notably: te income elasticity of demand for products (oter tan demand for te two types of education); te Frisc parameter; te elasticity of substitution between imported and domestic products; te elasticity of substitution between ualified and unualified labour; te elasticity of transformation between external sales (exports) and domestic sales; te elasticity of transformation between te two categories of labour, etc. Tese parameters are eiter calibrated using te SAM or are borrowed from te literature on CGE models and empirical studies in oter developing economies. 13 Specifically, it is assumed tat ouseolds ave a limited ability to transform unualified labour into ualified labour (euation 3). Tis is reflected by a constant elasticity of transformation function, proposed to be eual to 2.5. Transformation of labour is tus elastic. Returns to education on te labour market, i.e., te impact of a complete additional year on income, can be estimated using a Mincer euation (Mincer, 1958). In our model, te returns associated wit te cange from being unualified (not aving completed primary scool) to being ualified (aving completed primary scooling or more) sould lie somewere between te returns to completion of primary scool and te returns to completion of 12 We could also introduce unemployment. If te unemployment rate for ualified workers is ig, we may assume tat tese workers will not tolerate te additional education effort to acuire a iger skill level. 13 Details on CGE parameters are addressed by Annabi et al. (2006).

14 secondary scool. In te case of Burkina Faso, returns to completion of primary education are assumed to be 9% (World Bank, 2004). Step 3: Analysis of poverty, welfare and income distribution Te indicators of poverty and te income distribution are constructed using te DAD software package under Stata (Duclos, Araar and Fortin (2009)). Te euivalent variation (EV) is used to measure te impact on welfare. Te impact of education on welfare is tus measured indirectly via its effects on te net income used to purcase utility-enancing goods and services. On te one and, education contributes to te growt in te stock of (better paid) labour. On te oter and, investment spending on education reduces resources available to spend on te goods and services tat yield utility or welfare. We use an absolute poverty line for all ouseold categories in te survey. Te consumer price index (CPI) in te simulation varies according to te consumption patterns. Tese canges in te CPI are ten applied to te poverty line, making tis tresold endogenous: variations in prices of goods tus alter te poverty line. Te FGT indices in te initial situation are first calculated using te total level of real consumption per capita. After te simulation, eac ouseold category s net cange in income as generated by te CGE model is ten expressed as per capita consumption for eac ouseold category in te survey. Moreover, te poverty line is adjusted to account for canges in te CPI resulting from te simulation. Te new FGT indices are calculated using tis new tresold and new levels of real per capita consumption. Finally, te impact on te FGT indices is evaluated. 4 Basic data and description of te structure of te economy Calibration of our model is carried out using te 2004 SAM constructed by Nouve and Ouattara (2004) for te World Bank. It contains 119 accounts grouped into six categories: factors, agents, sectors, domestic demand, external demand and accumulation. Te SAM includes six ouseold categories wit seven sources of income: ualified wages, unualified wages, gross operating surplus, land rents, transfers from oter ouseolds, public transfers and transfers from abroad. It also as four types of expenditures: consumption, 14 transfers to oter ouseolds, taxes and saving. We specify four levels of education: primary, te first level of secondary, te second level of secondary and post-secondary or university. For te purposes of tis study, te first level is te primary education sector, wile te tree iger levels are aggregated to form te iger education sector. Analysis of labour begins wit a muc more detailed picture of te 14 We distinguis education consumption (expenditure) from te oter ones wic are obtained by maximizing te utility function.

15 level of ouseold ualifications. We tus specify five levels of ualification: not aving completed primary, aving completed primary, lower secondary, upper secondary and postsecondary. In conformity wit te objectives of te study, we consider te first level as unualified labour and aggregate te last four as ualified labour. Te poverty analysis is based on data from te Houseold Living Conditions Survey (HLCS) carried out by te National Institute of Statistics and Demograpy in Te HLCS includes 8500 ouseolds across te 13 regions of te country. It provides information, notably on income, education expenditures and ealt spending. Te sources of ouseold income by socio-professional category are sown in table 4.1. We find tat all ouseold categories earn a large sare of teir wage income from unualified labour, wit tis figuring surpassing 50% of income for private informal wage earners. Tis explains te poor sowing for ualifications in te reference situation. Returns to capital are te majority (52%-63%) of ouseold income for all ouseold types oter tan wage-earning ones. All ouseold types receive some income from land, altoug tis sare is negligible for public and private formal wage earners (0.4%). Te largest sare of income for public and private formal employees comes from wages paid to ualified labour (43.2%). Table 4.1: Houseold source of income according to socio-professional category (% of net income) a Public and private formal wage earners Private informal wage earners Cotton farmers Food farmers Livestock farmers Independent and inactive All ouseolds Unualified labour Qualified labour Capital Land Intra ouseold transfers Formal wage-earners public and private Informal wage-earners private Livestock farmers Independent and inactive Total intra ouseold transfers Public transfers

16 Transfers wit ROW Gross income Source: SAM. a ROW stands for rest of world Public and private employees (weter formal or informal) and livestock farmers do not receive transfers from oter ouseolds. A sare of teir income (respectively 0.5%, 0.1% and 1.7%), is tus allocated to transfers to tese ouseolds witout compensation. Government transfers to cotton farmers are marginal, at 0.3% of net ouseold income, and are also uite weak among public and private informal workers. However, among oter ouseold categories, tey remain fairly evenly distributed and vary between 2.2% and 2.7% of ouseold income. All ouseold groups depend on remittances. Table 4.2: Houseold investment in education (% of net income) a Public and private formal wage earners Private informal wage earners Cotton farmers Food farmers Livestock farmers Independent and inactive All ouseolds Primary education spending Higer education spending Investment in education Income net of education spending Savings Direct taxes Source: SAM. a Net income is calculated after deducting education expenditures. Te sare of income allocated to primary education spending among public or private (informal or formal) wage earners is less tan spending on iger education due to te substantial presence of te government at te primary and prescool levels 15 (table 4.2). Moreover, we observe tat oter ouseold types allocate a iger sare of teir net incomes to primary education tan tey allocate to iger education. Tis confirms tese ouseolds low sare of members in iger education and tus teir poor level of ualification in te initial situation. 15 Te government of Burkina Faso attaces a particular interest in te primary education sector, upeld by development partners and actualized by te 10-Year Primary Education Development Plan (PDDEB). Tis brougt te gross scool enrolment rate from 44 % in 2000/01 to 74.9 % in 2009/10.

17 Te government draws a major sare of its revenues from returns to capital (39.1%) and taxes on products (22.5%) (Table 4.3). Taxes on production and direct taxes are less important sources of revenues, respectively amounting to 5.9% and 6.2% of public revenues. Table 4.3: Sources of revenue and government budget items Revenue Value (billions CFA) Value Sare (%) (billions CFA) Sare (%) Expenditures Capital Administration Taxes net of subsidies on production 40 Primary 5.9 education VAT and oter taxes on products Higer 22.5 education Taxes on imports 79.5 Houseold 11.6 transfers Direct taxes Savings ROW Total Total Source: SAM. Tis income is allocated among current and investment spending (education and oter administration), transfers to ouseolds and savings (te current budget surplus) respectively accounting for 85.6%, 7.2% and 7.3% of total public expenditures. More specifically, te education expenditures found in te SAM represent 10% of public spending. Well over alf, or 73%, of tese expenditures are devoted to primary education. Te SAM data sow tat te total cost of a unit of education differs by ouseold category (table 4.4). We could imagine tat a unit of education is more expensive for a rural ouseold because, for example, students must go furter to attend classes or pay additional costs for a private tutor. Tis is also te case for agricultural ouseolds living in urban areas on te edge of large cities. Tis situation would tus definitely affect te simulation results. Moreover, te sare of te cost of education covered by cotton-farming ouseolds, food farmers, livestock farmers and te independent or inactive is greater tan tat paid by te government (see te private sare of unit cost variable). Wage-earning ouseolds, for teir part, pay a relatively low sare of te unit cost.

18 Table 4.4: Distribution of education costs a Private unit cost Public unit cost Total unit cost Private sare of unit cost (%) Public and private formal workers Private informal workers Cotton farmers Food farmers Livestock farmers Independent and inactive All ouseolds Source: SAM, calculated using GAMS. a Cf. euation 4. Notwitstanding tis cost difference between ouseolds, it is assumed ere tat te government subsidy for eac unit of education is te same. Since te government subsidy is te same for eac unit regardless of its total cost, tis as te effect of advantaging some ouseolds to te detriment of oters. 5. Scenario and simulation results In order to analyze te impacts of public spending on education in Burkina Faso, we simulate a 40% increase in public primary education subsidies, te same for eac socioeconomic ouseold type. Wen te state wises to maintain a fixed current balance, an increase in spending reuires an increase in taxes to rebalance te budget. Tis budgetbalancing strategy can create distortions and compromise te goal of ouseold poverty reduction. Te impact of te increase in taxes on ouseolds would be a function of te nature of tis measure. It is recognized tat direct taxes or taxes on income and wealt create smaller distortions tan indirect taxes. Tis means tat a consumer s level of utility is iger wit an

19 additional direct tax tan wit an indirect tax, even if te two generate te same levels of tax receipts. In tis study, we adopt two alternative tax mecanisms, summarized in table 5.1, to maintain a constant level of government revenues following its discretionary policy to increase education spending. In te following sections, te results of te first scenario (SIM1) are first presented in terms of te impacts on education and labour demand, te distribution of ouseold income and consumption prices and welfare, before moving to te results of te alternative scenario (SIM2). Table 5.1: Scenarios and financing metod Scenario SIM1 SIM2 40% increase in public primary education spending 40% increase in public primary education spending Financing mecanisms 7.2% income tax increase 2% sales tax increase State budget balance After Base simulation First scenario (SIM1) In tis scenario, we assume tat te government looks to domestic financing to cover te additional expense 16 troug taxes on ouseold income. Te overall impact of tis policy on poverty and welfare is difficult to predict a priori. a) Education demand and te labour market We ave tree mecanisms wereby te public education spending sock is transmitted to ouseolds: Te direct cost effect: As per euation 4, a 40% increase in te government subsidy ( TED ) would ave te effect of reducing te private unit cost of primary education β (, edb ) by te same amount for eac student. Tis directly increases te net gain from investment in primary education in euation 6. Te relative income effect: Houseolds respond to a decline in te direct cost of education by increasing teir investment in primary education wic, in turn, increases teir supply of ualified labour as well as te time devoted to primary education. Since teir overall labour endowment is fixed, tis implies tat te supply of unualified labour 16 Te state could adopt a counter-cyclical policy. In tis case it could finance te additional spending tanks to te bilateral and multilateral partners aid to minimize te distortions as muc as possible.

20 must decline. Te result is a relative decline in wages for tose wit ualifications and a decline in te net benefit of primary education. Te opportunity cost effect: Te decline in relative wages also results in lower opportunity costs. Tis decline in te opportunity costs is, owever, proportionally lower tan te decline in relative wages because te sare, s, of te active life devoted to studies in order to acuire ualifications is less tan one. Since te opportunity cost reinforces te direct cost effect, te result is an increase in te net gain from primary education. All else eual, te public subsidy s sare of te cost increases from 0.62 to (table 5.2). Since te total unit cost of education 18 is fixed, tis leads to a decline in te private unit cost of education ( β, edb ) from 0.38 to 0.13 for all ouseolds togeter. As a result, te direct cost of education declines, from a sare of to Moreover, te results indicate tat te increase in te public subsidy leads to an identical decline in te relative wage and opportunity costs for all ouseolds. As we will see later on, tis can be explained by te fact tat ouseolds increase teir investments in education, and as a result, teir supply of ualified labour. Te net effect of tese elements is an increase in net gain 19 of 2.78%. Te ouseolds react by increasing demand for education by 1.2% (variable Students in table 5.2). Tis increase in te demand for education brings about an identical increase in te volume of labour supplied by ouseolds. 20 Since eac ouseold s total endowment is fixed, tis increase is compensated for by a small decline in te supply of unualified labour (- 0.09%). Altoug it occupies a greater sare of te total amount of labour in te economy, te decline in te supply of unualified labour did not lead to a decline in te number of active workers in te economy. 21 Tis figure actually increases by 0.27%. As table 5.2 sows (cf. te active labour variable), 98.34% and 98.55% of cotton- and food-farming ouseolds are active in te reference situation. Among tese ouseolds, te increase in te volume of ualified labour, togeter wit te related decrease in te volume of unualified labour, leads to an increase in te volume of active labour. However, tese results sow tat an increase in ualified labour and a decrease in unualified labour following te 17 We find some deviation relative to te initial goal of 40% because ouseolds adjust teir beaviour wen facing lower costs. 18 In tis part of te study, te word education is often used to refer to primary education. 19 Note tat all of tese elements are components of te net gain expressed in euation In fact, te volume of ualified labour, (1 ) s δ LS and ouseold investment in education, sδ. LS, (te volume of students) are proportional because s (te sare of active life) is exogenous (section 4.1). 21 Note tat total active ouseold labour is composed of active ualified labour (26.24%) and unualified active labour (66.04%).

21 sock, all else eual, leave te volume of active labour of wage-earning ouseolds uncanged. Since agricultural ouseolds are a large sare of te Burkinabé population (about 80% in 2006 according to te last population census), an increase in te volume of active labour for tis group implies te same for te economy as a wole. On factor markets, te increase in te supply of ualified labour and te decrease in tat of unualified labour leads to a decrease in ualified wages (-7.19%) and an increase in te unualified wage (0.31%), as presented in table 5.3. Te results sow tat te impact of te public education subsidy on education varies according to te ouseold category and te variation in net gain. All ouseolds face te same absolute cange in benefits because te public subsidy is uniform. Furtermore, te relative wage and opportunity cost are te same for all ouseolds because tey observe te same ualified and unualified wages and also spend te same sare of active life in study. Terefore, te cange in te net benefits of getting an education is only dictated by tat of te direct cost. Given teir ig initial direct cost and teir weak initial net benefits, cotton and food farmers see a larger increase in te net benefit of getting an education after te subsidy sock. However, wage-earnings ouseolds wit marginal direct costs 22 and tus wit te igest net benefit initially see teir net benefits decrease after te subsidy sock. In addition to te benefit sown in table 5.2, te ouseold s education coice also l depends on te distributive parameter β and te transformation elasticity 23 τ l (euation 6). Te weaker te initial level of te volume of ualified labour, te more te volume of students and ualified workers increases after te sock. Agricultural ouseolds ave a very low sare of ualified ouseold members and tus see a larger increase in te sare of ualified members. Tis contrasts wit wage-earning ouseolds, wo lower teir investment in education and tus teir supply of ualified labour. We can conclude tat tis public primary education subsidy policy improves te benefits of getting an education. Tis leads to an increase in te supply of ualified labour and, ultimately, to a decline in te supply of unualified workers in te economy, wic brings an increase in unualified wages. 22 Note tat te small cost for tis category of ouseold is more tan neutralized by te 40% increase in te public subsidy suc tat it becomes negative after te sock (cf. bloc simulation: primary education cost, table 5.2). Tis can be explained by te assumption of fixed unit costs suc tat 40% of te increase in public spending per unit turns out to be very important for wage earners. 23 Cf. Section 3 for te elasticity coice.

22 Table 5.2: Structure and impact on education and labour supply Public and private formal workers Private informal workers Cotton farmers Food farmers Livestock farmers Independent and inactive All ouseolds Base: cost of primary education Private unit cost Public subsidy Total unit cost Sare (%) private cost Simulation: Cost of primary education Private unit cost Public subsidy Total unit cost Sare (%) of private cost Base: income and education cost Relative income Opportunity cost Direct cost Net benefit Simulation: income and education cost Relative income Opportunity cost Direct cost Net benefit % Net benefit Base: Sare (%) of total ouseold labour Unualified labour Qualified labour