Chapter 6. Ex-Ante Evaluation of Policy Reforms using Behavioral Models. François Bourguignon and Francisco H.G. Ferreira

Transcription

1 Chapter 6 Ex-Ante Evaluaton of Polcy Reforms (Bourgugnon & Ferrera) Chapter 6 Ex-Ante Evaluaton of Polcy Reforms usng Behavoral Models Franços Bourgugnon and Francsco H.G. Ferrera Abstract. The tools for polcy evaluaton and analyss revewed n the precedng chapters are all ex-post, as they essentally compare the dstrbuton of welfare over a populaton before and after a gven polcy or a gven local program s mplemented. Whle ex-post evaluaton s essental for assessng the true mpacts of exstng polces and programs, there are many nstances n whch polcymaers would value the ablty to smulate ex-ante the effects of alternatve reforms n exstng polces and programs, or ndeed the ntroducton of entrely new ones. Ths chapter examnes the scope for ex -ante evaluaton technques relyng on the estmaton and smulaton of structural econometrc models of household behavor. The archetypcal example of such technques s the modelng of the labor supply effect of tax-beneft systems n developed countres and the smulaton of reforms n those systems. The development of condtonal transfer polces presently observed n developng countres s lely to mae these behavoral mcrosmulaton technques ncreasngly relevant there too Introducton The tools for ncdence analyss of taxaton and publc spendng revewed n the prevous chapters are fundamentally ex-post. Gven some tax or some publc expendture, these tools tell us: a) who pays the tax, or receves the benefts provded through publc spendng; b) how much does everyone pay or receve, n accountng terms; c) how much does everyone receve when tang nto account behavoral responses to taxes or the free delvery of publc servces; d) what are the ndrect effects of publc programs. Ths sort of ex-post analyss sheds very valuable lght on the actual dstrbuton of a tax or publc expendture, and thus allows one to mae a better udgment as to whether ndvdual spendng tems are 'worth ther cost', or whether a reform of the nstrument under analyss should be consdered. All ths s doubtlessly extremely useful. However, only exstng taxes or publc programs may be analyzed n ths way. In ths chapter we turn to technques desgned to shed lght on the potental Tool Kt Chapter 6 page 6-1

2 Chapter 6 Ex-Ante Evaluaton of Polcy Reforms (Bourgugnon & Ferrera) dstrbutonal mpacts of polces or polcy desgns - whch do not currently exst, but whch mght exst. Suppose ex-post analyss of some gven exstng program ponted to the necessty of reformng t. Suppose further that a number of alternatve desgns rather than a sngle one - are suggested for the reform. It mght be helpful for the government to have some estmate of how much each alternatve reform would cost, and of whch households would be affected and by how much, under each alternatve. If pure expermentaton on actual lve subects (such as people or communtes of people) of all possble reform desgns were not possble or would tae too much tme no actual data would be avalable to drectly evaluate these hypothetcal reforms wth. Some counterfactual must be generated, showng how each household n a sample survey would fare dependng on the reform beng undertaen and how much ths would cost. Snce households respond to polcy changes by changng ther own actons, ths counterfactual must rely on some representaton of household behavor. Essentally, ex-ante analyss s what f analyss. What f some features of the tax system or publc spendng were modfed? How dfferent would t be for ndvdual households from the ntal stuaton, or 'status quo'? Usng the concepts ntroduced n the ntroducton to ths part of ths volume, such an analyss s margnal because t s meant to capture dfferences from the status quo. Also, t s almost necessarly behavoral, because of the need to generate counterfactuals that tae agent responses nto account. 1 Le ex-post analyss, ex-ante polcy evaluaton whch s concerned wth dstrbutonal or poverty outcomes generally reles on household surveys. However, ex-ante evaluaton requres an addtonal and prelmnary use of the household survey data. Whereas the secret of good ex-post mpact evaluaton s to dentfy whch actual samples should be compared, ex-ante analyss requres the smulaton of a counterfactual sample, whch should represent the populaton characterstcs of nterest, as they would be under the counterfactual polcy n queston. Ths requres some model, whch transforms the actual sample nto the counterfactual one. At ts smplest, ths may be a smple arthmetc representaton of the ncdence of a tax or beneft, wthout smulatng any polcy response by the agents (.e. assumng that all relevant elastctes are zero). If one s uncomfortable wth that assumpton, because one beleves that the polcy change may have mportant prce or ncome effects on consumpton or labor supply behavor, for nstance, one 1 Pure accountng mcro-smulaton methods need not be totally dscarded, at least as a frst approxmaton. They may be useful to descrbe frst-round effects on a sample of households, at least when prevous evdence suggests that behavoral responses may be slow to come, or have small effects. We return to ths pont n the last secton of the chapter. Tool Kt Chapter 6 page 6-

3 Chapter 6 Ex-Ante Evaluaton of Polcy Reforms (Bourgugnon & Ferrera) would need a behavoral model. Ths may be obtaned n two basc alternatve manners: ether through the estmaton of a structural econometrc model on the cross-secton of households provded by the household survey, or through the calbraton of a model wth a gven structure so as to mae t consstent wth what s observed n the survey and meant to correspond to the status quo. The archetypcal example of ex-ante analyss of ths nd s a tax-beneft model wth labor supply response, such as those commonly estmated for developed countres. Changes n the tax-beneft system n these models affect the budget constrants of households. They modfy ther dsposable ncomes wth unchanged labor supply, but through ths ncome effect and also through the prce effects nduced by changes n the after-tax prce of labor - they also modfy labor supply decsons. The quanttatve extent of these effects and ther net mpact - s determned through a behavoral model whch s generally estmated econometrcally across households observed n the status quo. There are many models of ths type n developed countres. By contrast, not much has been done along these lnes n developng countres, except perhaps for a few examples of the pure accountng mcro-smulaton approach. One reason may be that the cash element of the redstrbuton system s not usually mportant enough to warrant ths nd of analyss. It may also be that estmatng structural econometrc models of labor supply s made complex because of the nformalty of a large part of the labor maret. Nevertheless, the growng mportance of cash transfers and the ncreasng concern for dstrbutonal ssues s generatng great er need for ths nd of analyss. In addton, there are other dmensons than labor supply n household behavor that matter from a welfare pont of vew, and whch may be affected by transfers and other publc polces. Demand for schoolng or health care are some examples. The chapter s organzed as follows. In subsecton 6. we present the basc worhorse structural model used for ex-ante evaluaton of polcy reforms: the tax-beneft model wth labor supply response. Our emphass here s on ths model as an example of the sequence of steps needed to buld the model, generate the counterfactual dstrbuton, and compare t wth the actual n order to smulate and evaluate the reform. Secton 6. llustrates how a model change can be used to expand the set of polces whch can be analyzed, by ntroducng dscrete choce occupatonal or labor supply models. Secton 6.4 apples the methodology to the smulaton of a targeted condtonal cash transfer n a developng country, and llustrates the fnal steps of the ex-ante evaluaton approach. Extensons and lmtatons are consdered n the last secton of the chapter. See the survey of 'behavoral mcro-smulaton wth labor supply responses' by Creedy and Duncan (00). Tool Kt Chapter 6 page 6-

4 Chapter 6 Ex-Ante Evaluaton of Polcy Reforms (Bourgugnon & Ferrera) 6. The Basc Model: Accountng for Labor Supply Responses to Changes n Taxes Ths secton outlnes the logcal sequence a practtoner should follow when usng ex-ante evaluaton tools to smulate alternatve outcomes for a polcy or program reform. In order to mae the approach as concrete as possble, we couch the dscusson on a model of labor supply decsons whch are made when the budget constrants faced by ndvduals may be non-lnear, due to taxes and transfers. The basc reference s the frst model of ths type, due to Hausman (1980). We brea down the approach nto fve steps: defnton of the problem; dentfcaton of the requred data; specfcaton of the model; model estmaton; and polcy smulaton. Step 1: Identfy a well-defned, tractable polcy reform queston, whch the approach wll be used to shed lght on. In ths case, ths could be the effect of a rse n ncome taxes on (a) the dstrbuton of ncomes, and (b) the government s budget. The smulaton approach has often been used to address fscal questons of ths type because tax-experments (such as the applcaton of dfferent tax rates to comparable populaton subgroups) are generally dffcult to ustfy poltcally. Addtonally, a model whch smulated such tax changes wthout tang agents responses nto account would be lely to generate wrong revenue predctons. Step : Fnd a data set whch contans relable nformaton on the varables whch need to be ncluded n the model. In ths case, one would need a household or labor force survey, wth nformaton on earnngs and hours of wor for a representatve cross-secton of the populaton of nterest. Addtonally, one would deally need nformaton on whch taxes each ndvdual pays, and at what rates. If ths nformaton s not contaned n the best avalable surveys, one would nstead need a clear descrpton of tax rules, and would mae assumptons (such as 100% complance) about how those rules apply over the sample. Step : Wrte down the smplest economc model whch contans enough structure to capture the mechansms whch are lely to affect the agent responses to the polcy under consderaton. In ths case, the logcal economc structure s that of the textboo utlty maxmzng consumer. An economc agent wth characterstcs z chooses hs/her volume of consumpton, c, and hs/her labor supply, L, so as to maxmze hs/her preferences represented by the utlty functon u( ) under a budget constrant that ncorporates the tax-beneft system. The mportant thng s that the taxes under consderaton enter explctly nto the budget constrant, so that any changes n the parameters of the tax system wll affect the consumer s optmal choces. We wrte the model n a general form below, so that other tax or transfer changes can also be consdered: Tool Kt Chapter 6 page 6-4

5 Chapter 6 Ex-Ante Evaluaton of Polcy Reforms (Bourgugnon & Ferrera) Max u(c, L; z; β, ε ) s.t. c y 0 + wl + NT(wL, L, y 0 ; z; γ), L 0 (1) In the budget constrant, y 0 stands for (exogenous) non-labor ncome, w for the wage rate and NT( ) for the tax-beneft schedule. Taxes and benefts depend on the characterstcs of the agent, hs/her non-labor ncome and hs/her labor ncome, wl. It may also depend drectly on the quantty of labor beng suppled, as n worfare programs. γ stands for the parameters of the taxbeneft system - varous tax rates, means-testng of benefts, etc Lewse, β and ε are coeffcents that parameterze preferences. The soluton of that program yelds the followng labor supply functon : L = F(w, y 0 ; z; β, ε; γ) Ths functon s non-lnear. In partcular, t may be equal to zero for some subset of the space of ts arguments. The set of restrctons on the vector γ whch ensure that L > 0 are nown as a partcpaton condton. Step 4: Estmate the Model. Suppose now that a sample of agents (ndexed by ) are observed n some household survey contanng relable nformaton on L, w, y 0 and z. The problem s now to estmate the functon F( ) above or, equvalently, the preference parameters, β and ε, snce all the other varables or tax-beneft parameters are actually observed. To do so, t s assumed that the set of coeffcents β s common to all agents, whereas ε s dosyncratc. It s not observed but some assumptons can be made on ts statstcal dstrbuton n the sample. Ths leads to the followng econometrc specfcaton: L = F(z, w, y 0 ; β, ε ; γ) () where ε plays the usual role of the random term n standard regressons. Estmaton proceeds as n standard models, mnmzng the role of the dosyncratc preference term n explanng cross-sectonal dfferences n labor supply. Ths leads to a set of estmates ߈ for the common preference parameters and εˆ for the dosyncratc preference terms. By defnton of the latter, t s true for each observaton n the sample that: L = F(z, w, y ; 0 ߈, ê ;?) Tool Kt Chapter 6 page 6-5

6 Chapter 6 Ex-Ante Evaluaton of Polcy Reforms (Bourgugnon & Ferrera) Whle the estmaton process ust descrbed s conceptually smple, ts mplementaton n practce s generally not so straghtforward. Ths s because of the non-lnearty of the budget constrant, and ts possble non-convexty due to the tax-beneft schedule, NT( ), and corner solutons at L=0. Functonal forms must be chosen for preferences, whch may ntroduce some arbtrarness n the whole procedure. Fnally, t may be feared that mposng full economc ratonalty and a functonal form for preferences severely restrct the estmates that are obtaned. There has been a debate on ths pont ever snce ths model frst appeared n the lterature see n partcular MaCurdy, Green and Paarsch (1990). For a survey of emprcal strateges sutable for estmatng a non-lnear labor supply functon such as (), see Blundell and MaCurdy (1999). In the next secton, we return to ths estmaton problem, and suggest a smple and robust alternatve, the cost of whch s dscreteness. Step 5: Smulate the Polcy Reform usng the Emprcal Estmate of the Model. It s now possble to smulate alternatve tax beneft systems. Ths smply requres modfyng the set of parameters γ. In the absence of general equlbrum effects, the change n labor supply due to movng to the set of parameters γ s s gven by : L s L = F(z, w, y 0 ; ߈, ê ; s? ) F(z, w, y 0 ; ߈, ê ;?) The change n the dsposable ncome may also be computed for every agent. It s gven by : C s C s = w (L L ) + NT(y 0 s s s,w L, L ;z ;? ) NT(y 0, w L, L ;z ;?) Then, one may also derve changes n any measure of ndvdual welfare, and construct from each of them a full counterfactual dstrbuton over the sample. 6.. Dscrete Models of Labor Supply or Occupatonal Choce We now return to the caveat made n the last paragraph of Step 4 above, where we noted that the man weaness of the approach outlned here so far was that, despte ts conceptual smplcty, estmaton of the non-lnear (but pecewse contnuous) labor supply functon was generally complex, often nvolvng maxmzaton of non-trval lelhood functons, and requrng the specfcaton of possbly arbtrary utlty functonal forms. Assumng ndeed a structural specfcaton of the NT( ) functon general enough for all reforms to be represented by a change n parameters γ. Tool Kt Chapter 6 page 6-6

7 Chapter 6 Ex-Ante Evaluaton of Polcy Reforms (Bourgugnon & Ferrera) It turns out that smpler and less restrctve specfcatons may be used, whch consderably reduce ths problem. In partcular, specfcatons used n recent wor consder labor supply as a dscrete varable that may tae only a few alternatve values, and evaluate the utlty of the agent for each of these values and the correspondng dsposable ncome gven by the budget constrant. As we wll see n ths subsecton, f ths dscreteness s not seen as too costly for the purposes of the polcy smulaton under consderaton, t can buy a great deal of smplcty n estmaton. As before, the behavoral rule s smply that agents choose the value that leads to the hghest level of utlty. However, the utlty functon may now be specfed n a very general way. In partcular, ts parameters may be allowed to vary wth the varous quanttes of labor that may be suppled, no restrcton beng mposed on these coeffcents. Such a representaton s therefore as close as possble to what s revealed by the data. Formally, a specfcaton that generalzes what s most often found n the recent tax and laborsupply lterature s the followng: L = D f U = f (z ;w, c ; ß,e ) f (z ; w, c ;ß,e ) for all () where D s the duraton of wor n the th alternatve and, U the utlty assocated wth that alternatve and c the dsposable ncome gven by the followng budget constrant: c = y 0 + wl + NT(wD, D, y 0 ; z; γ) When the functon f( ) s lnear wth respect to ts common preference parameters and when the dosyncratc terms are assumed to be..d. wth a double exponental dstrbuton, ths model s the standard multnomal logt. It may also be noted that t encompasses the ntal model (1). It s suffcent to mae the followng substtuton : f (z ; w, c ;ß,e ) = u(c, D ;z,ß,e ) Ths specfcaton, whch nvolves restrctons across the varous labor-supply alternatves s actually the one that s most often used. Tool Kt Chapter 6 page 6-7

8 Chapter 6 Ex-Ante Evaluaton of Polcy Reforms (Bourgugnon & Ferrera) Even under ths more general form, one mght be tempted to argue that the precedng specfcaton s stll too restrctve, because t reles on some utlty maxmzng assumpton. It turns out that ex-ante ncdence analyss or polcy evaluaton cannot dspense wth such a basc assumpton. The ex-ante nature of the analyss requres that some assumpton to be made about the way agents choose between dfferent alternatves. Gven that, assumng that agents maxmze some crteron defned n a dfferent way for each alternatve s not really that restrctve. Second, t should be clear that f no restrcton s mposed across alternatves, then the utlty maxmzng assumpton s compatble wth the most flexble representaton of the way n whch labor supply choces observed n a the survey are related to ndvdual characterstcs, ncludng the wage rate and the dsposable ncome defned by the tax-beneft system, NT( ). The fact that model () can be nterpreted as representng utlty maxmzng behavor s to some extent secondary, although ths permts mplementng counterfactual smulatons n a smple way. More mportant s that ths model fts the data as closely as possble. Interestngly enough, the only restrcton wth respect to that obectve n the general expresson () s the assumpton that the ncome effect n each alternatve.e. the c argument n f( ) - depends on dsposable ncome as gven by the budget constrant whch ncorporates the tax-beneft schedule, NT( ). The economc structure of ths model thus les essentally n the ncome effect. If t were not for that property, t would smply be a reduced-form model amed at fttng the data as well as possble. In effect, the restrcton that the ncome effect must be proportonal to dsposable ncome seems to be a mnmal assumpton to ensure that ths representaton of cross-sectonal dfferences n labor-supply behavor may at the same tme represent a ratonal choce among varous laborsupply alternatves. After all, wthn ths framewor, the smulated effect of a reform of the taxbeneft system, NT( ), on ndvdual labor supply s estmated on the bass of the cross-sectonal dsposable ncome effect n the status quo. The role of dosyncratc terms, εˆ or εˆ, n the whole approach should not be downplayed. They represent the unobserved heterogenety of agents' labor supply behavor. Thus, they wll be responsble for some of the heterogenety n responses to a reform of taxes and benefts. It can be seen n () that agents who are otherwse dentcal mght react dfferently to a change n dsposable ncomes, despte the fact that these changes are the same for all of them. It s enough that the dosyncratc terms, εˆ, be suffcently dfferent among them. Tool Kt Chapter 6 page 6-8

9 Chapter 6 Ex-Ante Evaluaton of Polcy Reforms (Bourgugnon & Ferrera) Estmates of the dosyncratc terms result drectly from the econometrc estmaton of the common preference parameters, βˆ or βˆ 4. Note, however, that t s possble to us e a 'calbraton' rather than an estmaton approach. Wth the former, some of the coeffcents βˆ or βˆ would not be estmated but gven arbtrary values deemed reasonable by the analyst. Then, as n the standard estmaton procedure, estmates of the dosyncratc terms would be obtaned by mposng that predcted choces, under the status quo, and actual choces concde. Before closng ths secton, t s mportant to emphasze that there s some ambguty about who the 'agents' behnd the labor supply model (1) should be. Tradtonally, the lterature consders ndvduals, even though the welfare mplcatons of the analyss concern households. Extendng the model to households requres consderng smultaneously the labor supply decson of all members at worng age. Ths maes the analyss more complex. Applcatons of the precedng model are numerous n developed countres. Surveys are gven n Blundell and MaCurdy (1999) and n Creedy and Duncan (00). The dscrete approach underlned above s best llustrated by van Soest (1995), Hoynes (1996) or Keane and Mofftt (1998). A nce applcaton of ths approach for predctng the lely effect of the ntroducton of the Worng Famles Tax Credt n UK s Blundell et al. (000). 6.4 A developng country applcaton: cash transfers, demand for schoolng and labor supply For a number of reasons, the precedng framewor has not been appled to developng countres very frequently. Frst, drect transfers to households, whether postve or negatve, have usually been less mportant n developng countres. Second, the functonng of the labor maret may mae the concept of labor supply somewhat artfcal or nsuffcent n several nstances. In partcular, the dstncton between formal and nformal employment s mportant, wth the former often beng ratoned. Overall, t s not clear that the ncome effect, so crucal n the precedng model, s of sgnfcant sze n developng countres. Both lmtatons apply more strongly to the poorest segment of socety. Nevertheless, the broad ssue of agent response to polcy changes - whch motvated the precedng models - s becomng ncreasngly relevant n the developng world, as both tax and transfer systems develop. For nstance, t was observed n South Afrca that the payment of lump- 4 They would be standard resduals wth specfcaton () and most lely pseudo resduals n the dscrete formulaton ( ). Tool Kt Chapter 6 page 6-9

10 Chapter 6 Ex-Ante Evaluaton of Polcy Reforms (Bourgugnon & Ferrera) sum pensons to elderly people wthout other resources was accompaned by changes n the labor supply of the households they belonged to see Bertrand et al. (00). Because of ths, the change n monetary ncome n poor households was dfferent than had been expected. The ampltude of ths phenomenon mght be measured ether through ex -post dfferences n dfferences technques, or by ex-ante models of the type shown above. Of course, the ex-ante approach would become relevant n order to assst, say, n desgnng reforms to the exstng mnmum penson system. Progresa n Mexco, Bolsa Escola n Brazl and smlar 'condtonal cash transfer programs' n several other countres offer a second example of applcaton of the ex-ante evaluaton approach. The rest of ths secton s devoted to a summary presentaton of an ex -ante evaluaton of the effects of Bolsa Escola n Brazl and of potental changes n the format of that program. Ths evaluaton may be seen as an extenson or a varaton of the framewor dscussed n the prevous secton. The remander of ths secton s based on Bourgugnon, Ferrera and Lete (00), where the exercse s descrbed and analyzed n greater detal. The Bolsa Escola program conssts of transfers to households whose ncomes per capta are below 90 Reas approxmately US$ 0 - per month, provded that all ther chldren aged 6 to 15 are enrolled n a publc school, and that ther ndvdual attendance rates do not fall below 85% n any gven month. The monthly transfer s equal to 15 Reas per chld gong to school, up to a maxmum of 45 Reas per household. Ths may be consdered as a 'condtonal cash transfer program' because t combnes cash transfers based on a means-test and some addtonal condtonalty.e. havng chldren n school age actually gong to school. As the man occupatonal alternatve to school s wor, ths really s a labor supply problem smlar to the one analyzed above. The dscrete approach outlned n Secton 6. s used for each chld aged 10 to 15, wth the followng three labor-supply alternatves, ndexed by. =1 f the chld has some maret earnngs and does not go to school; = f the chld has some earnngs and goes to school; and fnally = f he/she does not wor on the maret and goes to school. Followng equaton (), the utlty of the household whch chld belongs to s specfed as : U = z ) β + α.( Y + + y ε for = 1,,. As before, z stands for characterstcs of both the chld and the household; Y s household ncome wthout the chld's earnngs; y s the ncome earned by the chld n alternatve ; and ε stands for dosyncratc preferences. Tool Kt Chapter 6 page 6-10

11 Chapter 6 Ex-Ante Evaluaton of Polcy Reforms (Bourgugnon & Ferrera) y s clearly the ey varable for descrbng the condtonal cash transfer program, snce transfers depend on ncome per capta n the household and on the schoolng status of the chld, whch tself affects her earnngs. Under the status quo, that s before the program was launched - and n the household survey beng used for estmaton - ths varable s defned as follows. In alternatve 1, y equals the observed maret earnngs of the chld, w. In alternatve, the chld s assumed to wor only a proporton M of the tme avalable when not gong to school. Hs/her observed earnngs therefore are Mw on average. In the thrd alternatve, the chld does not brng home any maret earnngs. But ths does not prevent hm/her from contrbutng to domestc producton. Ths contrbuton mght be assumed to be on average some fxed proporton of the earnngs obtaned from maret wor by ds wth the same observable characterstcs. Let Λw be the correspondng amount. A dfference wth the prevous case s that Λ s not observed. Substtutng the precedng values of y nto (4) leads to : U = z. ε (5) β + α Y + ρ w + where ρ s gven by : 1 ρ = α ρ = α. M, ρ = α. Λ (6) 1, Expresson (5) s comparable to the dscrete choce labor supply model () above. It can be estmated by a multnomal logt model. A potental problem mght be that ths model allows estmatng the coeffcents correspondng to some alternatve only as a devaton from those of some other alternatve, whch s taen as a reference. In ths applcaton, snce the chld s earnngs varable dffers across alternatves, t turns out to be necessary to estmate all three α, =1,,. In ths case ths s acheved through the restrctons gven by (6) whch allow the dentfcaton of the three coeffcents α and Λ. 5 It s those restrctons, and the fact that M can be estmated as a coeffcent on a dummy varable for school attendance n an earnngs equaton for chldren wth postve earnngs - that permts estmatng the whole model. See Bourgugnon, Ferrera and Lete (00) for detals. 5 Ths s not true of coeffcents β, but t may be seen from (7) below that only the nowledge of dfferences of these coeffcents across alternatves matters when determnng the utlty maxmzng alternatve. Tool Kt Chapter 6 page 6-11

12 Chapter 6 Ex-Ante Evaluaton of Polcy Reforms (Bourgugnon & Ferrera) Estmates ˆ, αˆ and εˆ of the preferences may thus be obtaned from the observaton of adult β and chld ncomes, varous household and chld attrbutes, and of the demand for schoolng and supply of chld labor, n a household survey taen pror to the launchng of the program. Once these estmates have been obtaned, t s easy to smulate the effect of Bolsa Escola on the decsons about chldren's occupatons. The alternatve wth the hghest utlty s chosen, wth the utlty of each alternatve beng now gven by : U U U U U 1 = Z = Z = Z = Z = Z βˆ. 1. βˆ. βˆ. βˆ. βˆ 1 + αˆ Y + αˆ + αˆ + αˆ ( Y + αˆ 1 + αˆ ( Y + γ γ ) + αˆ Mw Y + αˆ Λw Y + αˆ w + εˆ Mw + εˆ + εˆ ) + αˆ Λw + εˆ + εˆ f f f f Y + Mw Y + Mw Y γ Y > γ γ > γ In ths system, γ 1 and γ stand for the parameters of the 'tax-beneft' system beng modeled. The former stands for the Bolsa Escola transfer for each chld n school and the latter s the means test. Together, these condtons ncorporate the fact that Bolsa Escola can mae the schoolng alternatves and more attractve for poorer famles. Movng from alternatve 1 to (resp. alternatve ) ncreases or reduces monetary ncome accordng to whether the transfer γ 1 s above or below (1-M) w (resp. w ). In all cases, however, these moves potentally mean a hgher future ncome for the chld. Ths model was estmated on all chldren aged 10 to 15 n the Brazlan household survey sample, PNAD Below 10 years of age, the number chldren nvolved n maret actvtes was not suffcent to estmate the model. Results turned out to be qute consstent. In partcular, the value of M derved from the comparson of earnngs among those worng chldren not gong to school and those who go was found to be around 70%. Lewse, the coeffcent measurng the maret equvalent of the domestc producton of ds gong to school but not actve n the labor maret, Λ, was found to be 75%. Fnally, t turned out that, as n several other studes of the demand for schoolng, the ncome effect as measured by dfferences α - α 1 and α - α 1 s rather wea. After estmatng all the coeffcents of the model and the dosyncratc preference terms, ε, the Bolsa Escola program and alternatve formats of that program were smulated on each of the households n PNAD. Focusng on all chldren between 10 and 15 led to a sample comprsng Tool Kt Chapter 6 page 6-1

13 Chapter 6 Ex-Ante Evaluaton of Polcy Reforms (Bourgugnon & Ferrera) 4,000 persons. However, only 6 per cent of them could pass the means-test n Bolsa Escola and were thus potentally drectly affected by the program. Table 1 shows the effect of the Bolsa Escola program on the schoolng of chldren between the ages of 10-15, lvng n poor households 6, as smulated wth the model setched above. It can be seen that the program s ndeed qute effectve n reducng the number of poor chldren not gong to school. Ther proporton n the populaton of poor year-olds falls from 8.9%, wthout the program, to.7% under the smulated program. Interestngly enough, the proporton of chldren both gong to school and havng some actvty on the labor maret tends to ncrease, whch suggests that the program has lttle effect on chld labor when chldren are already gong to school. Alternatve specfcatons of the program scenaros have the expected effect on schoolng. In partcular, rasng the amount of the transfer or mang t age-progressve further reduces the proporton of chldren not gong to school. However, mang the program more generous by rasng the means test has very lttle effect. Another nterestng fndng s that the schoolng condtonalty s extremely mportant to acheve the obectve of unversal schoolng. As shown by the last scenaro, the ncome effect due to the cash transfer would have practcally no effect on schoolng wthout the condtonalty of Bolsa Escola. Table shows the expected effect of the Bolsa Escola program on poverty, defned here on the whole populaton and on the bass of monetary ncome only. Ths effect turns out to be more muted. The poverty headcount goes down by only 1. percentage ponts, reflectng the moderate sze of the program shown n the last row of the table the substantal nequalty wthn the poor segment of the populaton, and the negatve (chld) labor supply effect of the program. The comparson of the Bolsa Escola smulaton wth the results obtaned under scenaro 5 no condtonalty suggests that ths last effect s small, however. Indeed, there s almost no dfference n the poverty measures reported under both scenaros. Other scenaros have the expected effect on poverty, wth ncreases n transfer amounts beng once agan more potent than rses n the level of the means-test. A much more detaled analyss than the results shown n these two tables would of course be possble. Some addtonal detal s provded n Bourgugnon et. al. (00). As the purpose of ths chapter s mostly to present and dscuss the methodology of ex-ante evaluaton approaches to publc redstrbuton programs, we shall go no further here. The few results presented above are suffcent to llustrate both the prncples and the potental usefulness of the approach. 6 Note that the defnton of poverty used here does not concde wth the means -test n Bolsa Escola. Tool Kt Chapter 6 page 6-1

14 Chapter 6 Ex-Ante Evaluaton of Polcy Reforms (Bourgugnon & Ferrera) Despte the appeal of ths methodology, however, few applcatons are avalable for developng countres. In most cases, applyng t requres only a structural model of some dmenson of household behavor that permts smulatng a change n one or many polcy parameters. Thus, models of demand for schoolng along the lnes of Gertler and Glewwe (1990) could be used to smulate the mpact of several polces n the feld of educaton, le reducng the drect cost of schoolng, provdng school lunches, changng the qualty of schoolng, etc For nstance, Younger (00) analyzes wth ths nd of approach the consequences of reducng unformly the dstance to school n Rural Peru Conclusons, Extensons and lmtatons Although a relance on the multnomal logt model of dscrete occupatonal choce (or labor supply) consderably facltates the estmaton of ths nd of structural models, whch can then be used to smulate the mpact of polcy or program reforms tang behavoral responses nto account, t s stll probably true that startng such an exercse from scratch s not the easest tool n ths toolt. In ths context, t s worth consderng pure accountng ex-ante margnal ncdence technques, at least as an ntal step. Even ths (much smpler) approach s not yet of generalzed use, and t can be plausbly argued that n many cases the frst-order approxmaton generated by such a non-behavoral smulaton mght be nformatve n ts own rght, and serves as a learnng stage before attemptng the behavoral part of the model. It may be worth notng, n ths respect, that a pure accountng approach to the evaluaton of the Bolsa Escola program, assumng 100 percent enrolment n the program would have led to a poverty smulaton whch would be a bt but not massvely off The most obvous feld of applcaton of the ex-ante margnal ncdence approach has to do wth all the consequences of a change n the budget constrant faced by households, especally through all types of means-tested or condtonal transfers. Behavoral features whch are the most senstve to ncome changes are the safest to study, because strong estmates for cross-sectonal ncome effects are the most lely to translate nto actual responses to a program that modfes the ncome of agents. Thus, labor supply and related behavor le schoolng at one end of the actve lfe cycle and nactvty at the other end are the frst canddates for behavoral analyss. Thngs are less easy wth other components of publc fnance because they generally nvolve a prce and a qualty dmenson whch are very mperfectly observed. The use of publc servces for whch user fees apply s a case n pont. If some health care faclty s made avalable n a localty, then t s most lely that users wll frst be dscrmnated by ncome. If there s some cost recovery, then regular users wll be those agents wth an ncome above some threshold. Extendng health Tool Kt Chapter 6 page 6-14

15 Chapter 6 Ex-Ante Evaluaton of Polcy Reforms (Bourgugnon & Ferrera) care to poorer people could be done ether by lowerng cost recovery or by subsdzng low ncome users on the bass of some permanent ncome crteron. Such systems are used n several countres see for nstance the SISBEN system for health care n Colomba. Under the assumpton of fxed prce and qualty, they could be studed wth the same technques as above. The problem s that qualty and prce are hghly unlely to stay fxed, and smulatng the effects of varatons n these two dmensons adds consderable complexty. Although some household surveys record vsts to health centers, the prce and qualty characterstcs are not recorded or there may not be enough varaton across households to estmate demand elastctes n any convncng way. The analyss may thus be only partal. 7 Consumpton responses to changes n prces through taxes or subsdes also face ths problem of zero or unnown varaton n prces. From a welfare pont of vew, t s well-nown that t s unnecessary to now the demand response because, as a frst approxmaton, the effect of a prce change s smply the change n total expendtures that t causes at constant consumpton behavor. Behavoral responses are mportant only to fgure out the change n net tax recepts at the aggregate level. But then, aggregate demand analyss based on tme seres may be used for ths. 8 The same apples to changes n the prces of goods that are produced by households, that s the prce of the output of self-employed households or smply the wage rate of those employed n the formal labor maret. Overall, ndrect tax analyss, ncludng labor taxes, thus dffers substantally from changes n drect taxes and transfers, or n the supply of publc servces. The latter are more prone to ex-ante evaluaton or margnal ncdence analyss - than the former. To conclude, t s worth stressng some of the lmtatons of the precedng approach whch have not been mentoned explctly n ths short presentaton. Frst, ths approach may be dffcult to mplement because t generally requres the estmaton of an orgnal behavoral model that fts the polcy to be evaluated or desgned, and of course the correspondng mcro data. Because of ths, t s unlely that an analyss conducted n a gven country for a partcular polcy can be appled wthout substantal modfcaton to another country or another type of polcy. The methodologcal nvestment behnd ths approach may thus be mportant. Because of ths t can usefully be preceded by a pure accountng mcro-smulaton based on smpler assumptons. Second, the fact that the behavoral approach reles necessarly on a structural model that requres some mnmal set of assumptons s to be emphaszed. In general, there s no way these assumptons may be tested. In the labor supply model wth a dscrete choce representaton, the basc assumpton s that net dsposable ncome, as gven by the tax-beneft system, s what 7 Note that ths problem s present n the demand for schoolng problem too. Although publc schools n Brazl do not charge user fees, the analyss descrbed above mplctly assumes that qualty s mantaned constant everywhere. Effects of qualty varatons n the country are smply summarzed by the dosyncratc preference terms. 8 See chapter by Sahn and Younger n ths volume. Tool Kt Chapter 6 page 6-15

16 Chapter 6 Ex-Ante Evaluaton of Polcy Reforms (Bourgugnon & Ferrera) matters for occupatonal decsons. A reduced form model would say that the exogenous dosyncratc determnants of the budget constrant are what matters. Econometrcally, the dfference may be tenuous but the mplcatons n terms of smulaton are huge. For nstance, the modelng of Bolsa Escola s based on the mplct assumpton s that there s ncome poolng at the household level. Ths s certanly not granted and smulaton results would be dfferent f the model were specfed otherwse. Fnally, the strongest hypothess s that cross-sectonal ncome effects, as estmated on the bass of a standard household survey, concde wth the ncome effects that wll be produced by the program under study or reforms n t. In other words, tme ncome effects for a gven agent should concde wth the effect of cross-sectonal ncome dfferences. Here agan, ths s an hypothess that s hard to test and yet rather essental for ex-ante analyss. The only test we can thn of would be to combne ex-ante and ex-post analyss. For nstance, one could run some ex-ante analyss on a Mexcan household survey taen pror to the launchng of Progresa and then compare wth the results obtaned for schoolng and ncome n the ex-post evaluatons that have been made of that program. Ths sort of combnaton of ex -ante and ex-post evaluatons - such as those undertaen by Todd and Wolpn (00) and Attanaso, Meghr and Santago (00) s at the cuttng edge of the appled mcroeconomcs of development. In the absence of such valdaton exercses whch are only possble n rather specal crcumstances - some uncertanty about the predctons generated by ex-ante evaluatons based on structural behavoral modelng s bound to reman. Ths beng sad, ths can be a very valuable tool to vsualze the dstrbutonal mpact of alternatve desgns of polces whch are lely to generate strong behavoral responses. A pure accountng approach to margnal ncdence analyss s often a useful frst step. But, because people do generally change ther behavor when the polcy envronment around them changes, ntroducng behavor on an ex-ante bass s ultmately desrable for smulatng polcy reform n most realms. Tool Kt Chapter 6 page 6-16

17 Chapter 6 Ex-Ante Evaluaton of Polcy Reforms (Bourgugnon & Ferrera) 6.6. References Atnson A. and F. Bourgugnon (1991), Tax-Beneft Models for Developng Countres: Lessons from Developed Countres, n J. Khallzadeh-Shraz and A. Shah, eds., Tax Polcy n Developng Countres, Washngton, The World Ban, p Attanaso, O., C. Meghr and A. Santago (00): Educaton Choces n Mexco: Usng a Structural Model and a Randomzed Experment to Evaluate Progresa, Mmeo, Unversty College London. Bertrand, M., S. Mullanathan and D. Mller (00), Publc polcy and extended famles: evdence from South Afrca, Mmeo, Unversty of Chcago Blundell, R. A. Duncan, J. McCrae and C. Meghr (000), Evaluatng In-Wor Beneft Reforms : The Worng Famles' Tax Credt n the UK, Dscusson Paper, Insttute for Fscal Studes, London Blundell, R. and T. MaCurdy (1999), Labor Supply: A Revew of Alternatve Approaches, Ch. 7 n O. Ashenfelter and D. Card (eds), Handboo of Labor Economcs, Volume A, Elsever, Amsterdam Bourgugnon, F., F. H. G. Ferrera and P. Lete (00), Ex-ante Evaluaton of Condtonal Cash Transfer Programs: the Case of Bolsa Escola, World Ban Polcy Research Worng Paper #916, Washngton, DC. Creedy, J. and A. Duncan ( 00), Behavoral mcro-smulaton wth labor supply responses, Journal of Economc Surveys, 16(1), p.1-9 Gertler, P. and P. Glewwe (1990): The Wllngness to Pay for Educaton n Developng Countres: Evdence from Rural Peru, Journal of Publc Economcs, 4(): p Hausman, J. (1980), The effect of Wages, Taxes and Fxed Costs on Women's Labor Force Partcpaton, Journal of Publc Economcs, 14, P Hoynes, H. (1996), Welfare Transfers n Two-Parent Famles: Labor Supply and Welfare Partcpaton under AFDC-UP, Econometrca, 64(), p. 95- Keane, M. P. and R. Mofftt (1998): A Structural Model of Multple Welfare Porgram Partcpaton and Labor Supply, Internatonal Economc Revew, 9(), p MaCurdy, T., D. Green and H. Paarsch (1990), Assessng Emprcal Approaches for Analyzng Taxes and Labor Supply, Journal of Human Resources, 5(), p Todd, P. and K. I. Wolpn (00): Usng Expermental Data to Valdate a Dynamc Behavoral Model of Chld Schoolng and Fertlty: Assessng the Impact of a School Subsdy Program n Mexco, Mmeo. Van Soest, A. (1995), A Structural Model of Famly Labor Supply: a Dscrete Choce Approach, Journal of Human Resources, 0, p Younger, S. (00), Benefts on the Margn: Observatons on Average vs. Margnal Beneft Incdence, Mmeo, Cornell Unversty. Tool Kt Chapter 6 page 6-17

18 Chapter 6 Ex-Ante Evaluaton of Polcy Reforms (Bourgugnon & Ferrera) Table 1: Smulated effect on schoolng and worng status of alternatve specfcatons of condtonal cash transfer program (all chldren years old) Poor Households Orgnal Bolsa escola's program Scenaro 1 Scenaro Scenaro Scenaro 4 Scenaro 5 Not gong to school 8.9%.7% 1.9% 0.6% 1.8%.6% 8.9% Gong to school and worng.1% 4.7% 5.1% 5.4% 5.% 4.9%.0% Gong to school and not worng 68.1% 71.6% 7.9% 74.0% 7.0% 71.4% 68.% Total 100.0% 100.0% 100.0% 100.0% 100.0% 100.0% 100.0% Source: PNAD/IBGE 1999 and author's calculaton note: Scenaro 1: transfer equal R$0, maxmum per household R$90 and means test R$90 Scenaro : transfer equal R$60, maxmum per household R$180 and means test R$90 Scenaro : dferent values for each age, no household celng and means test R$90 Scenaro 4: transfer equal R$15, maxmum per household R$45 and means test R$10 Scenaro 5: Bolsa escola wthout condtonalty Table. Smulated dstrbutonal effect of alternatve specfcatons of the condtonal cash transfer program Orgnal Bolsa escola's program Scenaro 1 Scenaro Scenaro Scenaro 4 Scenaro 5 Poverty measures Poverty headcount 0.1% 8.8% 7.5% 4.6% 7.7% 8.8% 8.9% Poverty gap 1.% 11.9% 10.8% 8.8% 10.9% 11.9% 1.0% Total square devaton from poverty lne 7.9% 6.8% 5.9% 4.6% 6.0% 6.8% 6.8% Annual cost of the program (mllon Reas) Source: PNAD/IBGE 1999 and author's calculaton note: Scenaro 1: transfer equal R$0, maxmum per household R$90 and means test R$90 Scenaro : transfer equal R$60, maxmum per household R$180 and means test R$90 Scenaro : dferent values for each age, no household celng and means test R$90 Scenaro 4: transfer equal R$15, maxmum per household R$45 and means test R$10 Scenaro 5: Bolsa escola wthout condtonalty Tool Kt Chapter 6 page 6-18