Cash Transfer Policies, Taxation and the Fall in Inequality in Brazil An Integrated Microsimulation-CGE Analysis

Transcription

1 INTERNATIONAL JOURNAL OF MICROSIMULATION (2016) 9(1) INTERNATIONAL MICROSIMULATION ASSOCIATION Cash Transfer Polces, Taxaton and the Fall n Inequalty n Brazl An Integrated Mcrosmulaton-CGE Analyss Samr Cury Fundação Getúlo Vargas (EAESP/FGV), São Paulo, Brazl Departamento de Planejamento e Análse Econômca (PAE) Av. 9 de Julho, 2029, Bela Vsta, PO code: São Paulo SP, Brazl e-mal: sacury@terra.com.br Eucldes Pedrozo Fundação Getulo Vargas (EESP/FGV) and Unversdade Paulsta, São Paulo, Brazl Rua Afonso Celso, 171, Ap. 163A Vla Marana, PO code: , São Paulo - SP, Brazl e-mal: eucldes.pedrozo@fgv.br Allexandro Mor Coelho Centro Unverstáro Álvares Penteado (FECAP), São Paulo, Brazl. Avenda Lberdade, Lberdade, ZIP: , São Paulo SP, Brazl e-mal: allexandro_coelho@yahoo.com.br ABSTRACT: Persstent and very hgh-ncome nequalty s a well-known feature of the Brazlan economy. However, from 2001 to 2014, the Gn ndex regstered an unprecedented fall of 13.5% percent, combned wth sgnfcant poverty reducton. Prevous studes usng partal equlbrum analyss have ponted out the mportant role of federal government transfer programs n ths nequalty reducton, partcularly n the frst perod ( ) when the new trend started. The am of ths research s to evaluate the effcency of the two most mportant cash transfer programs, Bolsa Famíla (PBF) and Benefíco de Prestação Contnuada (BPC), n achevng ther purpose of allevatng poverty and reducng nequalty n Brazl s ncome dstrbuton. The smulaton results, usng an ntegrated modelng approach, confrm the mportance of these programs n reducng nequalty from 2003 to However, the effect on poverty allevaton was not strong. Fnally, the methodologcal approach reveals some mportant mechansms that were not present n

2 INTERNATIONAL JOURNAL OF MICROSIMULATION (2016) 9(1) prevous analyses, such as the role of the tax structure that fnances these polces. KEYWORDS: poverty, nequalty, cash transfer program, fscal polcy, computable general equlbrum model, mcrosmulaton model, Brazl. JEL classfcaton: C68, D58, I38, D31, E62. Cash Transfer Polces, Taxaton and the Fall n Inequalty n Brazl - An Integrated Mcrosmulaton-CGE Analyss

3 INTERNATIONAL JOURNAL OF MICROSIMULATION (2016) 9(1) INTRODUCTION It s wdely known that Brazl has one of the most unequal ncome dstrbutons n the world, wth a Gn ndex around 0.60 untl the begnnng of ths century. 1 Ths hgh ncome nequalty has been one of the man causes of the country s equally hgh levels of poverty, despte often strong economc growth. However, from 2001 to 2014, there was an unprecedented and contnuous fall n ncome nequalty, wth the Gn ndex declnng 13.5% from to Ths has concded wth a sgnfcant reducton of poverty. Between 2001 and 2014, the percentage of poor declned from 38.7% to 15.3%, whle the percentage of extremely poor declned from 17.4% to 4.6%. 3 These evolutons concded wth the ntroducton and expanson of targeted ncome transfer programs as part of a natonal poverty allevaton strategy. There are many knds of ncome transfer programs n Brazl, such as Programa Bolsa Famíla (PBF), Benefíco de Prestação Contnuada (BPC) 4 and several retrement benefts and pensons. Ths research analyzes the frst two programs (PBF and BPC) because they are the man cash transfer programs specfcally desgned as socal polces wth the purpose of poverty (and nequalty) reducton. PBF was created n October 2003 and s presently the federal government s man cash transfer program. It s a cash transfer program wth condtons such as 85% school attendance for chldren n schoolng age, vaccnaton for chldren under sx years old, and regular vsts to a health center for both pregnant and breastfeedng women. In 2005, PBF had a total of 10.6 mllon benefcary households and R$6.9 bllon worth of transfers (equvalent to 0.3% of GDP n 2005). BPC s a socal assstance beneft snce It ams to ad the elderly who are not ncluded n the publc socal securty system and the dsabled who cannot support themselves despte ther households fnancal care. Both benefcary groups comprse 2.8 mllon benefcares, wth a budget of R$9.7 bllon (or 0.4% of GDP) for BPC n The beneft conssts of a cash transfer amountng to one mnmum wage (R$300 n 2005) where the benefcary s famly per capta ncome must be less than a quarter of the mnmum wage. These programs am to reduce poverty and ncome nequalty, hence the need for evaluatons of ther effects. The most common approach s partal equlbrum and decomposton analyss. Hoffmann (2006b) ponts out that 31% of the declne n Brazl s nequalty 87% n the country s Northeast regon from 2002 to 2004 were due to the above mentoned programs. Barros et al. (2007c) estmated that PBF and BPC nduced, respectvely, around 11.8% and 11.1% of the fall n Cash Transfer Polces, Taxaton and the Fall n Inequalty n Brazl - An Integrated Mcrosmulaton-CGE Analyss

4 INTERNATIONAL JOURNAL OF MICROSIMULATION (2016) 9(1) ncome nequalty from 2001 to Accordng to IPEA (2012), durng the perod from 2001 to 2011, these two programs contrbuted to reduce nequalty by 17% (and up to 24% between 2003 and 2005). The contrbuton of PBF and BPC s more mpressve f one takes nto account the lmted budget of these publc polces. In 2013, the total budget of the publc socal securty system represented 10.9% of GDP (6.8% for the general system INSS, and 4.1% for former publc workers) whle PBF and BPC represented only 0.5% and 0.65%, approxmately. 5 However, these partal equlbrum or decomposton approaches do not take nto account systemc (general equlbrum) effects nduced by these programs on the economy, as well as the feedback mpacts to household ncome and consumer prces. When poor households receve transfers, they tend to consume more, whch stmulates producton and employment, and so on through a multpler effect. These demand effects are enhanced when we take nto account dfferences n the expendture patterns of Brazlan households by ncome level. Among poor urban Brazlan households, food expendture s 40% of total consumpton, compared to only 12% for the rchest households (Cury et al., 2006). In addton, the relevance of the general equlbrum effects s justfed by the sze and evoluton of the transfer programs. Almost 14 mllon households (around 20% of all households) are PBF benefcares, coverng more than 50 mllon ndvduals (around 25% of the populaton). In 2014, the total value of cash transfers was R$27.2 bllon, representng more than 0.5% of the Brazlan GDP. In addton, we may expect that program effects are senstve to the mechansm chosen to fnance ths specfc publc expendture, whch generate drect and general equlbrum effects of ther own. Also, durng some mportant changes were ntroduced n the fscal system. For example, n the socal securty budget, the sharpest revenue ncrease came from PIS-COFINS taxes (a rse of 30% n ther GDP rato), whch n started to levy mports. Addtonally, cash transfers may cause some benefcares to reduce labor supply. Ths may counteract some of the benefts of these programs. Recently, several studes amed to estmate the effects of the PBF and other condtonal cash transfer programs on adult labor supply n Brazl (e.g., Soares et al., 2007; Ferro and Ncollela, 2007; Texera, 2008; Covre et al., 2008; and Foguel and Barros, 2008). These studes use dfferent emprcal strateges to compare benefcares aganst comparable non-benefcares. Cash Transfer Polces, Taxaton and the Fall n Inequalty n Brazl - An Integrated Mcrosmulaton-CGE Analyss

5 INTERNATIONAL JOURNAL OF MICROSIMULATION (2016) 9(1) There s some evdence that PBF can reduce labor market partcpaton manly among benefcary mothers. For example, Tavares (2008) found evdence of an adverse effect of PBF on benefcary mothers wllngness to partcpate n the labor market due, manly, to the program's condtonaltes (educaton and health of chldren). On the other hand, Frpo et al. (2014) fnd evdence that some ndvduals (partcularly women) delberately reduce ther labor ncome n order to qualfy for the PBF. Smlarly, Santos (2013) shows that the PBF has encouraged adult benefcares to offer nformal labor. In both cases, the purpose of the benefcary s hdng the real famly ncome to contnue recevng the cash grant. Provng that a specfc methodology s unequvocally superor to others s not an easy task. However, accordng to Cockburn et al. (2015), the analyss of the mpacts of large-scale polces on ncome dstrbuton requres the usage of tool that ntegrates mcrosmulaton (MS) technques and a computable general equlbrum (CGE) model. Gven the systemc consequences of the PBF and BPC programs and ther fnancng mechansms on the overall economy, we adopt a CGE model ntegrated wth a Mcrosmulaton model (CGE-MS model) approach to evaluate the mpacts of these programs on poverty and nequalty n Brazl. Ths paper s organzed nto fve sectons, ncludng ths ntroducton. The next secton descrbes the methodology: the CGE model, the MS model, and ther ntegraton. The research questons, smulaton scenaros and results are presented n secton 3. The last secton presents the concluson and the fnal remarks METHODOLOGY Ths study apples a CGE-MS combned teratve approach, where the startng model (.e. where the frst shock s appled) s the MS model. For ths reason, ths knd of approach s referred to as bottom-up/top-down n the lterature (see Cockburn et al., 2015, for a dscusson of the varous CGE-MS approaches) The CGE Model Ths secton brefly descrbes some key characterstcs of the CGE model. Further detals on ths model can be found n Cury et al. (2010, Appendx A.2). 7 The CGE model dstngushes 42 domestc sectors, 8 household types, 8 the Government, and the external sector. The model assumes that the Brazlan economy s an nternatonal prce taker but that the changes of ts export prces can affect the external demand for Brazlan goods through an export demand equaton. Producton Cash Transfer Polces, Taxaton and the Fall n Inequalty n Brazl - An Integrated Mcrosmulaton-CGE Analyss

6 INTERNATIONAL JOURNAL OF MICROSIMULATION (2016) 9(1) s a functon of 7 types of labor, 9 captal and ntermedate nputs and sold as mperfect substtutes n the domestc and nternatonal markets. 10 Households and frms demand domestc and mported goods accordng to the Armngton (1969) hypothess. In terms of closure, the nomnal exchange rate s exogenous, whle the prce ndex s endogenous. Foregn savngs s also exogenous, whch mples a fxed balance of trade. Government spendng s fxed exogenously, but the total publc defct s endogenous. Investment s determned by total savngs where the margnal propensty to save s fxed The Mcrosmulaton Model (MS) The database for the mcrosmulatons s the Natonal Survey by Household Samplng 2003 (PNAD 2003). It contans a natonally representatve sample of almost 384,834 ndvduals dstrbuted n 117,010 households. 11 As t does not contan nformaton about household expendtures, the MS model focuses on ndvdual labor supply. Each workng-age ndvdual (over 10 years old) was classfed accordng to the seven labor categores n the CGE model. Formal and nformal workers are assumed to have flexble wage. However, as publc servants n Brazl beneft from a job stablty clause, t s assumed that ther employment levels are fxed. 12 The sample ncludes 106,590 workng-age ndvduals, representng 48,742,853 ndvduals n the total populaton. The MS model s based on Savard (2003). The man adaptaton for ths model s the use of a segmented labor market. 13 For the unemployed, the reservaton wage of each ndvdual determnes the choce to supply (or not) labor n the market. One of the man dffcultes n makng CGE-MS ntegraton work s convergence. For ths convergence to be successful the two databases must have the same values. Thus, the ndvdual weghts were multpled by a factor (reweghtng) so that the PNAD 2003 database reflects the CGE model data. Table 1 presents the results of ths reweghtng for employed and unemployed people. Cash Transfer Polces, Taxaton and the Fall n Inequalty n Brazl - An Integrated Mcrosmulaton-CGE Analyss

7 INTERNATIONAL JOURNAL OF MICROSIMULATION (2016) 9(1) Table 1 Employed and unemployed reweghng for L1 to L5 work factors Descrpton PNAD occupatonal condton CGE model data Unemployment rate Unemployment Reweghng Factor of the (n 1,000 persons) (n 1,000 persons) worker Employed Unemployed Total Employed Unemployed Total Employed Unemployed L1 L2 L3 L4 L5 Source: Unsklled nformal Sklled nformal Formal wth low skll Formal wth average skll Formal wth lgh skll 12,890 1,567 14, % 11,714 1,418 13, % , , % 5, , % ,923 1,349 15, % 12,274 1,184 13, % , , % 8, , % , , % 2, , % TOTALS 43,926 4,817 48, % 39,647 8,537 87, % PNAD 2003, CGE model data base. A pror concern regardng the ndvduals reservaton wage estmaton s the ssue related to labor supply dentfcaton. In prncple, the expanson of ncome transfers exogenously affects the wllngness to supply labor of varous demographc groups n dfferent ways. Thus, t s necessary to estmate an equaton for ndvdual labor supply, dentfed by the number of ndvdual work hours, as a functon of the ndvdual wage-hour after changes n ncome transfers for each demographc group has been consdered. It s also necessary to correct the potental auto-selecton bas to labor supply partcpaton. After applyng ths procedure, t s possble to properly dentfy the dfferent reactons of the labor supply to exogenous changes n the sze of transfers for ndvduals n each demographc group. Therefore, the estmaton procedure can be descrbed n two steps as follows: Step 1 At ths mcrosmulaton stage, we are nterested n the ndvdual mpact due an ncome transfers shock, especally for the demographc group of sngle mothers who are heads of household. Ths demographc group s the man benefcary of the PBF and deserves specal attenton because t s the most senstve to non-labor ncome from transfer programs, as found n our MS model results. Our emprcal strategy s based on a smpler verson n whch the worker makes an ndvdual decson. Due to the dentfcaton problem of the non-lnear budget constrant, we estmated a reduced-form hour equaton as a functon of the ndvdual s wage, the ncome from transfers Cash Transfer Polces, Taxaton and the Fall n Inequalty n Brazl - An Integrated Mcrosmulaton-CGE Analyss

8 INTERNATIONAL JOURNAL OF MICROSIMULATION (2016) 9(1) programs, other ncome, and a number of demographc controls. The other ncome varable combnes all sources of non-labor ncome, followng Blundell and McCurdy (1999). Ths last varable for marred women, for example, s calculated by takng the husband s actual earnngs nto account. On the other hand, we created another varable that represents the BPC and PBF programs n order to capture the effects of the ncome transfers on labor supply. 14 The predcted workng hours are obtaned from the observed and non-observed ndvdual and household characterstcs and hs own wage. Therefore, the worker s predcted hours of work ( h ) s estmated by the sem-log specfcaton accordng to Blundell and McCurdy (1999): 15 j h j Z u, 1,..., n and j 1,2, 3 log w logq log B (2.2.1) where,, ndvdual ;, and are the parameters to be estmated; w s the hourly wage rate for Q s the vector of the total household ncome net of the earnngs (ncludng ncome transfers) receved by the ndvdual ; ndvdual n 2003; B s the vector of benefts receved (PBF and BPC) by Z represents the ndvduals observable characterstcs; u s the random error term, whch captures the non-observable characterstcs that affect the ndvdual labor supply; and j s the ndvdual s demographc group, 1 beng for men, 2 for woman head of household wth chldren, and 3 for other women (who are not heads of households). The value of determnes the substtuton effect related to senstvty of ndvdual labor supply to changes n wages. The values of and represent the ncome effect, that s, the mpact of non-labor ncome on labor supply. The Z vector of ndvdual characterstcs was composed of the followng varables: Z educ,age,age, famsze, 2 D a where educ denotes the number of years of schoolng, age s a proxy to the level of experence; famsze s the famly sze n terms of number of ndvduals (excludng pensoners, domestc servants and ther parents), 1 for rural). D a s a dummy for the area where the famly s domcle s located (0 for urban and Indvdual workng hours are observed only for those who are already employed. Thus, the sample of ndvduals that present strctly postve work hours s not random. However, t s possble that the choce to work s related to ncome-dependent varables, ether from labor or non-labor (other Cash Transfer Polces, Taxaton and the Fall n Inequalty n Brazl - An Integrated Mcrosmulaton-CGE Analyss

9 INTERNATIONAL JOURNAL OF MICROSIMULATION (2016) 9(1) ncome sources). Therefore, the stuaton s typcally one of endogenous selecton, n whch there s a decson to partcpate or not n the labor market and, gven that the ndvdual had decded to work, t s necessary to determne how many workng hours he wll offer. In order to control for potental selecton bas, the procedure proposed by Heckman (1979) s appled, whch conssts of: Pr S 1 z Y Z (2.2.2) where s a functon of accumulated dstrbuton, where S s a qualtatve varable representng the occupatonal choce for an ndvdual : ths varable wll take the value 0, f the ndvdual does not supply work, or 1, f otherwse. The varable s a vector of estmated parameters that determne the probablty that the ndvdual takes part n the labor market. Y s the vector representng the varables related to labor and non-labor ncomes that affect the decson to supply labor by ndvdual. As before, of partcpatng n the labor market. Z are the ndvdual characterstcs that determne the probablty The equatons (2.2.1) and (2.2.2) are estmated by a two-stage method proposed by Heckman (1979). In ths model, equaton (2.2.2) corrects sample selecton bas by non-observables. The selecton varables used for dentfcaton are educ, age, age 2 and famsze. To test and correct the sample selecton bas we estmate a probt model of labor market partcpaton wth these selecton varables. These equatons are run separately for three demographc groups: men, female household heads wth chldren, and other women, to estmate the elastcty of labor supply. The nverse of Mlls rato z s extracted from equaton (2.2.2), whch s appled to equaton (2.2.1) to ensure that the parameters of these equatons are consstently estmated. After estmatng the coeffcents n (2.2.1) and the nverse of Mlls rato, t s possble to estmate the adjusted workng hours of each ndvdual, j h, based on observed and non-observed characterstcs. Adjusted workng hours are then appled to the ndvdual s observed wage, whch results n the adjusted ndvdual s wage ( w ). 16 Step 2 ŵ, The second part of the mcrosmulaton process s the computaton of the reservaton wages and the new employment rato. Indvdual labor supply s a functon of ndvdual market wage rates and non-labor ncome, among other varables. These wage rates can be observed for wage workers. For others there s an unobservable wage rate that an ndvdual could potentally receve. Cash Transfer Polces, Taxaton and the Fall n Inequalty n Brazl - An Integrated Mcrosmulaton-CGE Analyss

10 INTERNATIONAL JOURNAL OF MICROSIMULATION (2016) 9(1) Accordng to Heckman (1974) t s possble to express ths reservaton wage as a functon of ther ndvdual characterstcs as well as non-labor ncome and other constrants. Followng Savard (2003), the non-observed reservaton wage s obtaned from the observable and non-observable ndvdual s characterstcs, as well as household characterstcs. Due to the mportance of evaluatng the reservaton wage before and after an ncome transfer shock, we nclude non-labor ncome n the structural reservaton wage equaton and dentfy the ncome transfer varable separately. Therefore, the worker s reservaton log wage, equaton: where, Z w, s estmated by the log w log Q log B u, 1,..., n (2.2.3), and are the parameters to be estmated. The observed wage, wage adjusted by the procedure descrbed n step 1; 17 presented earler. Q, B and ŵ, s the hourly Z are the same varables Due to the mpossblty of observng the wage offer to the sample s unemployed ndvduals, we need to estmate a probt model that determnes the probablty of these ndvdual takng part n the labor market. Ths probablty, S 1, s estmated by the functon: S 1 z Y Z D Pr (2.2.4) g where: s a functon of accumulated dstrbuton; s a vector of estmated parameters that determne the probablty of the ndvdual takng part n the labor market; as before, Z and Y are, respectvely, the ndvdual characterstcs and the work and non-work ncome that determne the probablty of partcpatng n the labor market; and for female household heads wth chldren, 2 for other women). D g s a demographc dummy (0 for man, 1 As for equatons (2.2.1) and (2.2.2), equatons (2.2.3) and (2.2.4) are estmated by the Heckman two-stage method. Ths makes t possble to calculate the reservaton wage of each ndvdual, (k = 0,1) based on observed and non-observed characterstcs. If the ndvdual belongs to state k 1, the reservaton wage of worker wll be used n comparson wth the observed wage, to select among the potental employed or unemployed persons. If n state k 0 k w w,, the reservaton wage of ths ndvdual s obtaned to construct a rank of potental newly employed persons. Cash Transfer Polces, Taxaton and the Fall n Inequalty n Brazl - An Integrated Mcrosmulaton-CGE Analyss

11 INTERNATIONAL JOURNAL OF MICROSIMULATION (2016) 9(1) j If the estmated reservaton wage ( w ) s hgher than the earned wage ( w j ) observed n the database, then ths person s ndcated as potentally unemployed; otherwse, he remans employed,.e.: f w w otherwse k, ndvdual, s potentally unemployed he s potentally employed After makng ths comparson for each employed person, the model determnes the Heckman presmulaton occupatonal level by prvate labor type HLsl by summng up the number of people orgnally unemployed wth the number of people that would be unemployed accordng to the Heckman crteron. It deserves mentonng that ths occupatonal level s dfferent from the orgnal level n the database Lsl, gven that there are people n the database that work and earn wages lower than ther estmated reservaton wages. Ths occurs because the latter wages are estmates of what these people could earn n the market accordng to ther ndvdual and household characterstcs. Therefore, merely applyng the Heckman procedure to the database changes the occupatonal level for each labor type. As proposed by Savard (2003), the selecton of ndvduals who should be unemployed starts wth classfyng workers accordng to ther reservaton wages. Those wth the hghest reservaton wage wll be the frst to become unemployed f the real wage decreases. If there s a postve change n the real wages, the frst to be employed wll be those wth the lowest reservaton wage. 18 Step 3 In ths secton we evaluate the relatonshp between the condtonal cash transfer programs and ndvdual work decsons through substtuton and ncome effects. In table 2 we present the margnal effects n respect to hours of work, mpled by the estmates from steps 1 and 2. The wage compensated elastcty of labor supply reflects the strength of the substtuton effect from the perspectve of labor ncome. The wage elastctes are the coeffcents reported by the varable log w n equaton (2.2.1). For women wthout chldren (j = 3) ths elastcty s postve and hgher than for women who are household heads (j = 2), whch s to be expected accordng to the results of many emprcal studes. For men, the negatve elastcty s not usual, but t s statstcally non-sgnfcant. The magntude of the ncome effect s reflected by the ncome elastcty of labor supply. These Cash Transfer Polces, Taxaton and the Fall n Inequalty n Brazl - An Integrated Mcrosmulaton-CGE Analyss

12 INTERNATIONAL JOURNAL OF MICROSIMULATION (2016) 9(1) ncome elastctes descrbed by the varables log B (publc transfer benefts) and log Q (all other non-labor ncome) n equaton (2.2.1) are all negatve, as expected. The hghest sensblty s related to the group formed by female household heads wth chldren, whch s n lne wth the great majorty of emprcal work on ths subject. Also, the results are consstent wth standard theory and show that cash benefts may have partcpaton effects on specfc populaton groups. Table 2: Elastctes - Margnal effects for groupng demographcs Varable J + 1 (Men) J = 2 (Women wth chldren) J = 3 (Women) Elastcty S.E. Elastcty S.E. Elastcty S.E. Wage elastcty (log w) (0.0506) ** (0.0070) ** (0.0047) Income elastcty (log B) (0.0010) ** (0.0014) ** (0.0008) Income elastcty (log Q) ** (0.0002) ** (0.0006) ** (0.0004) Note: ** sgnfcant at 1%; * sgnfcant at 5%. Source: Authors estmates Integraton of the CGE and MS models The mpacts of the PBF and BPC programs on welfare ndcators are assessed wth an ntegrated CGE-MS modelng framework wth a b-drectonal lnkage to guarantee convergence of solutons for both models. The followng llustrates how ths b-drectonal procedure: Step 1 j The MS model contans data on thousands of ndvduals, estmates the reservaton wage ( w ) for each person n the database, and defnes occupatonal levels for each category of prvate labor by means of the equatons (2.2.3) and (2.2.4), as mentoned n secton 2.2. Frst, for each smulaton, the benefts receved from the ncome transfer programs n 2003 ( B ) are replaced by ther new values ( B ) n equatons (2.2.3) and (2.2.4), whose re-estmaton generate * the Heckman post-smulaton occupatonal level for each prvate labor type ( HLsl * MS ). Then, the dfference between the Heckman post and pre-smulaton occupatonal levels by prvate labor type, ( * HLsl MS HLsl ), 19 are added to the orgnal occupatonal level n the database ( Lsl ), generatng a post-smulaton occupatonal level by prvate labor type calculated by the MS model ( Lsl ). Step 2 * MS The occupatonal levels compatble wth the new values of benefts ( Lsl * MS ), as well as the new Cash Transfer Polces, Taxaton and the Fall n Inequalty n Brazl - An Integrated Mcrosmulaton-CGE Analyss

13 INTERNATIONAL JOURNAL OF MICROSIMULATION (2016) 9(1) * amount of gven benefts ( B ), are then appled to the CGE model, where B * t B, 1,..., n; t BF BPC (2.3.1) t, and t B s the amount of benefts that ndvdual receved from PBF and BPC. * The new values of taxes used to fnance the changes n transfer programs ( B ) are also appled to the CGE model to smulate the changes they nduce n the economc envronment. All these changes nduce the economc system to acheve a new general equlbrum and, as part of ths process, the labor market reach equlbrum wth new real wage values ( W of prvate labor. Step 3 * CGE ) for each knd * The percentage change n the average real wage ( W CGE ) for each knd of prvate labor from the CGE smulaton s appled to the wages earned by each person n the MS model database ( w ), defnng after-shock values for earned wages ( w ). 20 After ths, we compare these new ndvdual wages * * w wth ther respectve reservaton wage ( w ) by means of the Heckman procedure. Usng the prevously mentoned crteron for ths procedure, we have that: j * j f w w, ndvdual s unemployed, otherwse, ndvdual s employed. After classfyng the workers by ther reservaton wages, those wth the hghest reservaton wage are the frst to become unemployed f the real wage decreases. However, f the real wage rses, the frst to be employed wll be those wth the lowest reservaton wage. Addng the number of people to be employed or unemployed accordng to ths crteron to the ntal occupatonal level defnes * a new occupatonal level for each prvate labor type Lsl. Step 4 These new occupatonal levels are then transmtted to the CGE model as shown n Fgure 1: MS Cash Transfer Polces, Taxaton and the Fall n Inequalty n Brazl - An Integrated Mcrosmulaton-CGE Analyss

14 INTERNATIONAL JOURNAL OF MICROSIMULATION (2016) 9(1) Fgure 1 MS CGE Integraton Frms CGE model Goods & servces supply Labor demand Goods & servces markets Government B*, T * Labor market Goods & servces demand * W CGE Wages Households Beneft transfers B* Households b, B Labor supply Lsl * MS MS model If the occupatonal levels calculated by the MS model are dfferent from those n the CGE model, they change the equlbrum of the labor markets, whch wll then change wages and the economc envronment as a whole untl the CGE model reaches a new equlbrum. In ths sense, step 2 restarts, but wthout changes n benefts and ther fnancng sources, and ths ntegrated soluton procedure loops untl the dfferences n the occupatonal level for all prvate labor types calculated by the MS model ( Lsl *) approaches zero. 21 Ths assocaton s done consstently wth the equlbrum of aggregate markets n the CGE model, whch requres that: () relatve changes n average earnngs n the MS model must be equal to changes n wage rates from the CGE model for each prvate wage group n the labor market; () relatve changes n the number of prvately waged workers by labor market segment n the MS model must match those n the CGE model; and () changes n the consumpton prce vector, p, must be consstent wth the CGE. 22 Accordng to the above procedure, the prvate labor supply s beng modfed wth each smulaton teraton; for example, some ndvduals wll be losng ther former jobs. If ths happens, the share of each household n the total ncome of each labor category can also change (parameter hl n equaton 2.1.2). In order to capture these varatons, we ncorporate the dfferences among the parameter hl, along the smulaton rounds as a shock n the CGE as well, whch performs smultaneously wth the procedures descrbed n ths secton. 23 Cash Transfer Polces, Taxaton and the Fall n Inequalty n Brazl - An Integrated Mcrosmulaton-CGE Analyss

15 INTERNATIONAL JOURNAL OF MICROSIMULATION (2016) 9(1) Non-Labor Income Procedures After the CGE and MS models solutons converge, t s stll necessary to treat non-labor ncomes before calculatng poverty and nequalty ndcators. Bascally, the varables related to these sources of ncome n the MS model follow the CGE varatons or mantan the same value as n the household survey, as shown n Table 3. In the former case, the changes from the CGE model are transmtted to the correspondng varables n the MS model n a undrectonal way. Table 3: Integraton of CGE-MS model for non-labor ncome (Base 2003) Household Income Source Procedure n the Mcrosmulaton (PNAD 2003) Self Employed Income Interest, Dvdends and Others and House Rental Retree and Penson Publc Benefts Retree and Penson Prvate Benefts CGE results varatons of these ncome sources are appled to the mcrosmulaton model vectors. 24 CGE results varaton of these ncome flows ndvdualzed to the 8 household categores n the model are appled to the mcrosmulaton model vectors. 25 The same vector values as n the mcrosmulaton base year model. The same vector values as n the mcrosmulaton base year model. Donaton receved The same vector values as n the mcrosmulaton base year model. Note: For each famly, the above sources are deflated by a famly specfc prce ndex (after smulaton). 26 Source: Authors elaboraton. 3. SIMULATIONS AND RESULTS 3.1. Descrpton of smulatons The drect amng of the smulatons s to assess the effects of changng values and benefcares of the programs PBF and BPC from the ones presented n 2003 to the ones presented n Transfer Programs. We addressed the changes between 2003 and 2005 wth smlar procedures adopted by Barros et al. (2007c). 27 However, we construct a specfc mputaton methodology for the 2005 addtonal benefts. 28 Gven ths nformaton, we then took the beneft share for the eght CGE household categores and apply the actual amounts gven by Brazl s budget data, ensurng consstency wth our SAM data. The values are shown n Table 4. Cash Transfer Polces, Taxaton and the Fall n Inequalty n Brazl - An Integrated Mcrosmulaton-CGE Analyss

16 INTERNATIONAL JOURNAL OF MICROSIMULATION (2016) 9(1) Table 4: Total benefts by CGE household category and changes between 2003 and 2005 (R$ thousands) Households Bolsa Famíla BPC Bolsa Famíla BPC Total Increase Share of Benefts n Total Household Income F1 777, ,171 1,829,805 1,418,757 1,796, % F2 35,269 19,741 88, , , % F3 616, ,187 1,250, , , % F4 810,877 2,203,557 1,861,258 4,346,372 3,193, % F5 131, , , , , % F6 319, , , , , % F7 336, , , ,837 12, % F8 157,558 50, ,481 25,328 99, % Total 3,185,000 5,132,934 6,871,361 7,838,638 6,392, % Source: Author s elaboraton based on data from the federal budget and SAM (2003). The table above shows the dfferences n benefts between 2005 and Total transfers ncreased by R$6,392 mllon, whch represents 0.57% of total household ncome. By program, the ncrease was approxmately 116% for BF and 53% for BPC. Also, there was an overall mprovement n the targeted group. The poorest households n the CGE model (F1, F2 and F3) ncrease ther total BF share from 44.9% (2003) to 46.1% (2005). However, despte some mprovements, the data show that the BPC targetng was much worse than the BF program (19.4% n 2003 and 26.6% n 2005). We carred out two smulatons, dstngushed by whether the ncreased transfers are fnanced by publc defct (whch reduces total savngs and nvestment, accordng to the model s macro closure) or ncreased taxes: SIMU A and SIMU B. SIMU B: In ths smulaton, the ncrease n benefts s fnanced by ncreased federal government taxes. Ths choce was made n order to hold the nomnal publc defct almost constant. The man justfcaton for ths polcy arrangement s the Brazlan Fscal Responsblty Law, whch states that new expendtures must be explctly fnanced. In order to replcate what happened n that perod, we dd an extensve analyss of the 2005 federal budget data to dentfy the specfc tax sources that fnanced the PBF-BPC programs durng that year. Table 5 summarzes the amounts of each federal tax source, ther percentage composton, and the equvalent CGE tax as presented n the CGE model. Cash Transfer Polces, Taxaton and the Fall n Inequalty n Brazl - An Integrated Mcrosmulaton-CGE Analyss

17 INTERNATIONAL JOURNAL OF MICROSIMULATION (2016) 9(1) Table 5: Programs tax sources n 2005 (R$ thousands) Brazl Tax Source Value Composton Equvalent tax n the CGE model Contrbução para Fnancamento da Segurdade Socal (COFINS: budget code 153) Contrbução Provsóra sobre Movmentação Fnancera (CPMF: budget code 155) Outros Impostos Dretos (Income Tax and other drected taxes) Impostos sobre Produtos (Mx of Indrect Taxes) Contrbução Socal sobre o Lucro das Pessoas Jurídcas (CSLL: budget code 151) Operações de Credto Externas - Em Moeda (budget code 148) 7,570, % 5,265, % 993, % 445, % 418, % 15, % COFINS tax and ts value added reform Drect taxes on frms and households Drect taxes on frms and households Indrect taxes on Revenue Drect taxes on frms and households Total 14,710, % Source: Authors elaboraton. From ths table, 29 we collected the share of each tax n the total ncrease of program expendture (R$6,392,065,000). Thus, the followng adjustments were made to the CGE model: () 2.2% ncrease n drect taxes on the gross ncome of the eght CGE household categores and frms; () ntroducton of the Cofns tax through the replcaton of the PIS-COFINS tax reform, whch was mplemented by the federal government n the same perod. From the total revenue generated by ths reform, 27.5% was approprated as fundng for the cash transfer programs Macroeconomc mpacts Table 6 presents selected macro results for SIMU A and SIMU B. We frst analyze SIMU B, whch captures the combned effects of changes n transfers and taxes, whle the results from SIMU A capture only the mpacts of changes n transfers (and the publc defct). In general, the macro mpacts are adverse snce they nduce a real GDP fall of 0.46% and an aggregate employment decrease of 0.48%, whle generatng a prce ndex ncrease of 0.65%. These adverse effects can manly be attrbuted to the partal reform of PIS-COFINS that were the man tax sources of the transfer programs. Cash Transfer Polces, Taxaton and the Fall n Inequalty n Brazl - An Integrated Mcrosmulaton-CGE Analyss

18 INTERNATIONAL JOURNAL OF MICROSIMULATION (2016) 9(1) Table 6: Macroeconomc ndcators (percentage change)* Macroeconomcs ndcators SIMU A (%) SIMU B (%) GDP Consumpton Investment Publc Sector Defct Exports (**) 0.84 Imports (**) 1.07 Employment Prce Index Note: (*) Real percentage change from the CGE base year. (**) Lower than 0.01%. Source: Authors elaboraton. PIS-COFINS reform ncreases the taxaton of frms value-added (VA), whch leads to ether hgher margnal revenues or lower margnal costs through the reducton of VA components. Snce captal s fxed by sector, ths mples a lower labor demand, whch nduces a decrease n wages and, ultmately, household ncome. Specfcally, the fall n aggregate consumpton s due to the decrease n overall household ncome despte the rse of ncome among the poorest households who beneft most from the ncrease n transfers. The taxaton of mports ncreases ther prces n the domestc market and nduces another adverse effect on aggregate consumpton, through hgher domestc prces. Exports fall due to the prce-responsve behavor of external agents and the model s external closure characterstcs. Frst, the smulaton nduces an ncrease n domestc prces, whch causes a decrease n external demand for Brazlan export goods. Second, snce external macro closure mples a constant trade balance, the fall n mports must be accompaned by a fall n exports. The government defct ncreases by 7.88%, whch shows that the ex-ante smulated taxaton was not enough to completely fnance the total transfer costs. Despte the total fnancng ntenton of SIMU B, the government defct was not held constant because the taxes dead weght losses turn out to be very strong durng the smulaton. 30 The comparson between SIMU A and SIMU B demonstrates the solated effects of transfers wthout the tax ncreases. GDP n SIMU A s practcally stable. However, the shock nduces a trade-off between consumpton and nvestment, wth the former ncreasng by 0.5% and the latter decreasng by 1.42%. Ths fact can be explaned by the ncrease n ncome transfer and by the hgher publc defct (+17.89%), whch reduces total savngs. The consequent fall n nvestment wll reduce economc growth n the near future, engenderng addtonal negatve effects not captured n ths analyss. Therefore we can conclude that the adverse mpacts of SIMU B are due Cash Transfer Polces, Taxaton and the Fall n Inequalty n Brazl - An Integrated Mcrosmulaton-CGE Analyss

19 INTERNATIONAL JOURNAL OF MICROSIMULATION (2016) 9(1) to the smulated program s fnancng structure Impacts on Labor Market SIMU B has an adverse effect on aggregate employment (-0.48%, accordng to Table 6). In SIMU A, the effect s small (-0.11%) but t s one of the few macro aggregates wth negatve mpacts. Table 7 shows that employment falls for all categores of workers n the prvate sector. The number of publc worker does not change because, by model assumpton and legal status, the quanttes are fxed and the adjustments are done through flexble real wages. Table 7: Change n employment from the base-year (%) L1 L2 L3 L4 L5 L6 L7 SIMU A ,00 0,00 SIMU B ,00 0,00 Note: Source: L1-unsklled nformal; L2-sklled nformal; L3-low-sklled formal; L4-average-sklled formal; L5- hgh-sklled formal; L6- low-sklled publc servants; L7- hgh-sklled publc servants. Authors elaboraton. In the SIMU B, among the prvate sector, we can see two patterns. The effects are more pronounced n the nformal markets (L1 and L2) and among less sklled workers. In our nterpretaton, wth lower mports and a fxed trade balance, there s pressure to overvalue the real exchange rate makng exports more expensve. The sectors n whch exports are more senstve to prce changes are the most tradtonal ones, whch are more ntensve n nformal and low-sklled workers. On the other hand, the fall of consumpton n hgh ncome households affect the employment of more sklled workers n the sectors that produce goods wth hgher technologcal content, such as automobles, auto parts, electronc, electrcal, and pharmaceutcal ndustres. Table 8 show generally negatve mpacts on real wages by labor type. In the SIMU B, the wage of less sklled workers (L1 and L3) fall more than the other types. The reducton n publc servants earnngs s due to the model hypothess stressed above. Table 8: Change n the average real wage from the base-year (%) L1 L2 L3 L4 L5 L6 L7 SIMU A + 0,32 0,12 0,04 0,07 0,09 0,04 0,01 SIMU B 1, ,52 0,90 1,61 1,66 1,62 Note: Source: L1-unsklled nformal; L2-sklled nformal; L3-low-sklled formal; L4-average-sklled formal; L5- hgh-sklled formal; L6- low-sklled publc servants; L7- hgh-sklled publc servants. Authors elaboraton. Cash Transfer Polces, Taxaton and the Fall n Inequalty n Brazl - An Integrated Mcrosmulaton-CGE Analyss

20 INTERNATIONAL JOURNAL OF MICROSIMULATION (2016) 9(1) Table 9 shows the effects on the total labor ncome, combnng wage and employment effects. These are stronger among the less sklled workers, especally for those n the nformal market, and due manly to real wage effects. Table 9: Changes n real payroll from the base-year (%) L1 L2 L3 L4 L5 L6 L7 SIMU A + 0,19 0,25 0,21 0,13 0,14 0,04 0,01 SIMU B Note: Source: L1-unsklled nformal; L2-sklled nformal; L3-low-sklled formal; L4-average-sklled formal; L5- hgh-sklled formal; L6- low-sklled publc servants; L7- hgh-sklled publc servants. Authors elaboraton. Agan, SIMU A has practcally no sgnfcant adverse effects. In contrast to SIMU B, nformal unsklled workers (L1) fare better, due to the fact that there s a reallocaton of producton toward more labor-ntensve sectors. Fnally, t s mportant to stress that the convergence procedures affect the fnal labor market equlbrum. Table 10 llustrates ths process for SIMU B. The frst lne show changes n real wages, the prce ndex and GDP n the frst smulaton round. In the second lne, the same varables are presented for the last round. The convergence soluton shows that changes n the transfer programs nduce general equlbrum effects that ntally affect wages prmarly and then, due to the teratve process, are partally reallocated to employment mpacts (quantty effects). Table 10: Dfferences between frst and last SIMU B rounds selected varables (%) wage L1 wage L2 wage L3 wage L4 wage L5 pndex GDP Frst round smu B 2,16 1,39 1,76 1,29 1, Last round smu B 1, ,52 0,90 1,61 0,65 0,46 Source: Authors elaboraton Impacts on Income Dstrbuton Table 11 shows the mpacts of the changes n the transfer programs on nequalty ndcators. In general, the results confrm the mportant role of transfer programs n the fall n Brazl s nequalty durng that perod. 31. Cash Transfer Polces, Taxaton and the Fall n Inequalty n Brazl - An Integrated Mcrosmulaton-CGE Analyss

21 INTERNATIONAL JOURNAL OF MICROSIMULATION (2016) 9(1) Table 11: Inequalty ndcators from household per capta ncome (base year 2003) Inequalty Indcators Base Year SIMU A SIMU B Orgnal Results** Change Results** Change Gn Index % % Thel-T Index % % Source: from the CGE-MS ntegraton model. (base year: 2003 PNAD survey) Analyzng the Gn ndex, the fall of 0.48% (SIMU B) s slghtly lower than the ones reported by other studes that evaluated the contrbuton of transfer programs to the fall n nequalty usng partal equlbrum or decomposton analyss. Barros et al. (2007c) found that 22.9% of the total Gn decrease between 2001 and 2005 was due to BF and BPC. In the same perod, these authors reported a total decrease n the Gn ndex of 2.6%. Therefore, the decrease dsplayed n table 10 accounts for approxmately 14% (SIMU A) and 19% (SIMU B) of the total fall n nequalty durng that perod. Here, t s mportant to stress some dfferences between our smulatons and the study by Barros et al. (2007c). The smulatons capture the effects of changes n a shorter perod, from 2003 to 2005, n a general equlbrum envronment. Although the perod s dfferent, we found evdence that the transfer programs alone (SIMU A) had lower effects on nequalty than those reported by other studes. However, n the case of SIMU B, the effect s smlar. Table 12 shows the mpacts on household ncome. The changes n programs have a slghtly adverse effect on natonal average household ncome n SIMU A (-0.18 %), whch s magnfed (-0.81%) n SIMU B. In both, the postve effects on the three poorest household categores are prmarly due to the ncrease n transfer amounts to them. On the other hand, the effects of the programs expanson on the rchest households (F7 and F8) are already negatve n the frst smulaton (SIMU A) and are magnfed when the changes n taxaton were consdered. There are two mportant causes of these negatve mpacts. The frst s related to labor ncome reported n the former secton. The other can be attrbuted to the general prce ncrease and ts effect on non-labor ncome such as socal securty benefts and nterest on fxed ncome nvestments. Cash Transfer Polces, Taxaton and the Fall n Inequalty n Brazl - An Integrated Mcrosmulaton-CGE Analyss

22 INTERNATIONAL JOURNAL OF MICROSIMULATION (2016) 9(1) Table 12: Change n household ncome from the base-year (%) Average household ncome Orgnal SIMU A SIMU B Values (R$) Values (R$) Change Values (R$) Change Natonal average % % Household 1 (F1) % % Household 2 (F2) % % Household 3 (F3) % % Household 4 (F4) % % Household 5 (F5) % % Household 6 (F6) % % Household 7 (F7) % % Household 8 (F8) 3, , % 2, % Note: Source: F1 poor urban households headed by actve ndvduals; F2 poor urban households headed by non-actve ndvduals; F3 poor rural households; F4 urban households wth low average ncome; F5 urban households wth medum ncome; F6 rural households wth medum ncome; F7 households wth hgh average ncome; F8 households wth hgh ncome. Authors elaboraton. The postve mpacts of the program s expanson on the poorest household category are better reflected by SIMU A. Ths smulaton also captures systemc effects nduced from these programs, as shown n sectons 3.2 and 3.4. Besdes that, SIMU B also shows the addtonal negatve mpact of the ncrease of taxaton. Ths helps to understand the mprovement of the Gn ndex n SIMU B n relaton to the SIMU A. Asde from capturng the ncome ncrease of the poorest households, t also captures the fall n ncome of the rchest households due to the ncreased taxaton Impacts on Poverty The effects on poverty are presented n table 13. Based on observed and smulated ncome per household head, we calculate three poverty ndcators: Headcount ndex (P0), Poverty Gap (P1) and Severty of Poverty (P2). These ndcators are calculated usng the mcrosmulaton model before and after smulatons. The reference lne values are from September 2005 estmated by Barros et al. (2007b) and deflated to 2003 wth the IPCA (Índce de Preços ao Consumdor Amplo) prce ndex. Cash Transfer Polces, Taxaton and the Fall n Inequalty n Brazl - An Integrated Mcrosmulaton-CGE Analyss

23 INTERNATIONAL JOURNAL OF MICROSIMULATION (2016) 9(1) Table 13: Poverty ndcators - PNAD 2003 Poverty Indcators Base year SIMU A SIMU B Results* Results Change Results Change Poverty Lne (Lne = R$ 143,70) P % % P % % P % % Extreme Poverty Lnes (Lne = R$ 71,84) P % % P % % Source: P % % Authors elaboraton. The general reducton n poverty ndcators (P0, P1 and P2) shows that the transfer programs alone (SIMU A) had postve effects on poverty and on extreme poverty. From the results n the table 13 we also see that the mpacts on poverty were reduced by the changes n taxaton (SIMU B). As dscussed above, tax changes generate some adverse mpacts that affect the poor populaton and, n a more ntensve way, the extremely poor ndvduals. As we have seen prevously n secton 3.4, the mpacts on labor ncome were stronger among the less sklled workers (L1 and L3) and the nformal workers (L1 and L2). These workers are predomnantly from the poorest households, whch are hghly dependent on labor ncome. Therefore, despte the ncrease n the receved benefts, some households fall nto poverty due to ncreased taxaton. In the case of SIMU B, the program expanson dd not have an mpact on extreme poverty. However, the poverty gap and severty of extreme poverty worsened. One explanaton s the deteroraton of non-labor ncome due to the prce ncrease, whch especally affected the famly F2 located, before shock, just above the poverty lne and whose ncome s bascally derved from socal securty benefts. 4. CONCLUSIONS The am of the smulatons presented here was to nvestgate the role of the two most mportant Brazlan cash transfer programs n reducng nequalty. Ths was done through a CGEmcrosmulaton (CGE-MS) teratve approach. Two smulatons were run, where the cost of ncreased transfers was fnanced by ncreased taxes or a rse n the publc defct. The macro results showed that, n general, the mpacts were adverse for several macro ndcators, ncludng GDP, employment, and prce ndex. However, mportantly, the study s CGE-MS Cash Transfer Polces, Taxaton and the Fall n Inequalty n Brazl - An Integrated Mcrosmulaton-CGE Analyss