The contribution of government transfer programs to inequality. A net-benefit approach. *

Transcription

1 The contrbuton of government transfer programs to nequalty. A net-beneft approach. * June, 2006 Alvaro Forteza ** Ianna Ross ** * We are most grateful to the World Bank for fnancng ths study, but we are entrely responsble for ts specfc contents. ** Departamento de Economía, Facultad de Cencas Socales, Unversdad de la Repúblca, Uruguay. Emal: Alvarof@decon.edu.uy, Ianna@decon.edu.uy.

2 Resumen La contrbucón a la desgualdad de los programas de transferencas del goberno es usualmente determnada analzando en qué medda los benefcos pagados se focalzan en famlas de bajos ngresos. Varos analstas han encontrado que los benefcaros de algunas transferencas claves del goberno son en su mayoría hogares de ngreso medo y alto, contrbuyendo así a una mayor desgualdad. En el presente trabajo se dscute el hecho de que el mpacto de estos programas en la desgualdad debería ser evaluado consderando los benefcos recbdos netos de los mpuestos pagados por los hogares para fnancar los programas, puesto que los hogares de mayores ngresos recben benefcos por un monto mayor pero tambén pagan mpuestos más altos. Este enfoque es lustrado estmando el mpacto de cuatro programas de transferencas del goberno en la desgualdad en Uruguay y demostramos que las conclusones son dferentes dependendo de s se utlza benefcos brutos o netos en la estmacón. Palabras clave: Transferencas, desgualdad, redstrbucón. JEL: D31, H55, I38 2

3 Abstract The contrbuton of government transfer programs to nequalty s often assessed by analyzng to what extent the benefts pad go to lower ncome famles. Several analysts have found that some key government transfers actually go mostly to mddle and hgh ncome famles and thus contrbute to greater nequalty. We argue n ths paper that the mpact of these programs on nequalty should be evaluated consderng the benefts receved net of the taxes pad by households to fnance the programs, snce hgher ncome households receve hgher benefts but they also pay hgher taxes. We llustrate ths approach by estmatng the mpact of four government programs on nequalty n Uruguay and show that the conclusons are dfferent dependng on whether we use gross or net benefts n the estmaton. Key words: Transfers, nequalty, redstrbuton. 3

4 1. Introducton It s often argued that some government transfer programs lead to greater nequalty because hgh ncome famles receve a dsproportonately large percentage of the benefts (Feldsten, 1974, Brownng and Brownng, 1994, Mazza, 1999, Perry et al., 2006). Ths s usually the case of contrbutory programs le unemployment nsurance and contrbutory pensons, because the ndvdual beneft s lnked to the contrbuton wage. Better pad workers are enttled to hgher unemployment benefts and hgher pensons. Smlar results have been reported for publc spendng on hgher educaton n Latn Amerca. But analyss of the ncdence of publc expendture on dfferent groups of the populaton accordng to dstrbuton of ncome tells only half of the story about the contrbuton of publc programs to nequalty. We argue that the assessment of the contrbuton of government programs to nequalty should consder benefts pad net of taxes collected to fnance these programs. If the same households that receve hgher unemployment benefts, for example, tend to make the bggest contrbutons to fnance the unemployment nsurance program, then the program may actually contrbute to reducng dsposable ncome nequalty even f better pad workers receve hgher unemployment benefts. Ths dea can be llustrated usng the nequalty decomposton ndex proposed by Shorrocks (1982a and b). The contrbuton of the sources of ncome k to nequalty s measured regressng the sources of ncome k on total ncome across ndvduals. The coeffcent of total ncome n ths regresson s the contrbuton of the sources of ncome k to nequalty. Hence, the sources of ncome that have a postve coeffcent n these regressons contrbute to ncreasng nequalty and the sources that have a negatve 4

5 coeffcent contrbute to reducng nequalty. Now consder a government transfer program that pays benefts and collects taxes that are both postvely correlated to total ncome across households (Fgure 1). If the contrbuton to nequalty s measured consderng only gross benefts, ths program ncreases nequalty. But the program represented n ths fgure reduces nequalty f ts contrbuton to nequalty s assessed consderng net benefts. Insert Fgure 1 Notce that the program assumed n Fgure 1 s also regressve n tax collecton,.e. rch households pay a lower share of ther ncome as taxes than poor households. Hence, the program seems to be regressve when evaluated by lookng separately at ether benefts or taxes. Nevertheless, ths program reduces dsposable ncome nequalty. After ths bref ntroducton, the paper contnues as follows. In secton 2 we present the methodology n detal. In secton 3 we present results for several government transfer programs n the case of Uruguay. Secton 4 concludes wth some fnal remarks. 2. The methodology In order to emprcally assess the mpact of government transfer programs on nequalty, we computed the nequalty decomposton ndex proposed by Shorrocks (1982a and b). We treated these programs as separate sources of ncome. Household surveys provde drect data on benefts receved by dfferent ndvduals from government programs, but they do not provde nformaton on taxes pad by ndvduals to fnance these programs. Because of 5

6 the lack of mcro-data on drect and ndrect contrbutons, we had to make some assumptons to compute net transfers. Let y be the ncome of household ( = 1,..., n) from sources of ncome k ( k 1,..., K ) =. Total ncome of household s y = k y. The dstrbuton of total ncome can be represented by y = ( y,..., ) and the dstrbuton of factor k ncome by y ( y,..., y ) 1 y n k =. Let s k ( I ) be the proportonal contrbuton of ncome k to total ncome nequalty measured k s k I wth ndex I, so that ( ) 1 =. Shorrocks (1982b) proposed the followng rule to decompose the contrbuton of each and every source of ncome to total ncome nequalty: 1k nk s ( I ) cov = Var ( yk, y) ( y) ( y y) k = ρ k, SD SD ( yk ) ( y) where cov ( y k, y) s the covarance and ( y k, y) ρ s the coeffcent of correlaton between factor k ncome and total ncome; Var ( y) s the varance of total ncome; and ( ) SD and SD ( y) are the standard devatons of factor k ncome and total ncome, respectvely. Notce that the contrbuton of sources of ncome k to total nequalty s just the slope coeffcent of the regresson of y k on y. Shorrocks (1982b) showed that ths s the only decomposton rule for any nequalty measure that comples wth a set of desrable propertes. y k Let us say that the last ncome source K corresponds to the government transfer program whose contrbuton to nequalty we want to evaluate. In ths last column of the matrx of 6

7 ncome we compute the net transfer the government program pays to each and every household. 1 Snce column K of the matrx of ncome regsters both benefts receved and contrbutons and taxes pad to fnance the program, the ncome regstered n other columns must be measured before taxes pad to fnance the transfer program. For the transfer program to be complete, the records n column K of the ncome matrx must add up to zero: someone else must pay for net benefts receved by any household. Formally, y K = 0 (1) We used mcro-data from households and expendture surveys and some aggregate data from admnstratve records of socal securty programs n Uruguay to buld the matrx wth elements y. The Natonal Insttute of Statstcs household survey provdes ndvdual data on several sources of ncome, ncludng benefts pad by some socal protecton programs, but t does not provde nformaton on contrbutons and taxes pad to fnance the program. We know from the socal securty nsttutons that these programs are fnanced wth a complex mx of payroll and general taxes. Among the latter, ndrect taxes are by far the bggest factor, wth value added tax accountng for a sgnfcant share of the whole package. Thus we dstngush payroll taxes a and ndrect taxes t collected to fnance the transfer program. 1 We talk about the transfer program to smplfy the presentaton, but t should be clear that the same prncples apply to more than one transfer program. 7

8 Labor earnngs n the household survey are reported after payroll taxes ( y ') added payroll taxes to get pre-tax labor earnngs:. Therefore we y = y ' + a ; k K (2) Naturally, a must be zero f the source of ncome k s non-labor ncome. Gven that other socal securty revenues are mostly ndrect taxes, we dd not need to add other taxes to the survey s reported ncome to get pre-tax ncome. Hence, taxes satsfy the followng condton: t 0 k K; t = t 0 (3) = K and the transfer program column was computed as: y K = b t a (4) where b stands for the beneft receved by household from the transfer program and a = k stands for total payroll taxes pad by household to fnance the program ( a ) a. Equatons (2) and (4) determne the matrx of ncome organzed to assess the contrbuton of the transfer program to nequalty, but we do not have drect data on some of the varables nvolved. The household survey does provde the after payroll tax earnngs ( y ') 8

9 and the benefts pad by the transfer programs (b ), but t does not provde drect data on payroll taxes ( a ) or ndrect taxes pad to fnance the transfer programs ( t ). We know from (1) that total taxes pad to fnance the program must be equal to total benefts pad by the program, but we need nformaton on ndvdual contrbutons. The socal securty nsttutons provde aggregate nformaton on ther sources of fnancng whch can be used to determne the shares of payroll and ndrect taxes n fundng the programs we are evaluatng. Let α K be the share n total spendng of government transfer program K fnanced wth payroll taxes. Estmated ndvdual payroll taxes and ndrect taxes should satsfy the followng condtons: ( α K ) a = = α K b ; t 1 b (5) In order to dstrbute these aggregates among ndvduals, we assumed that () payroll taxes are proportonal to labor ncome, provded the ndvdual does contrbute to socal securty (assumpton A1), and () ndrect taxes are proportonal to total expendture of the household (assumpton A2). More specfcally, we made the followng assumptons: (A1) Indvdual payroll taxes: a ay = 0 ' ; a > 0 f C otherwse and k LI 9

10 where C stands for the subset of workers who declared to the household survey that they do pay payroll taxes and LI stands for the subset of ncome sources that correspond to labor ncome. Notce that the rate of payroll taxes a s a weghted average of the rates pad by dfferent categores of workers, n accordance wth ther answers to the household survey. Also notce that ths rate multples post-tax labor ncome, whch s not the ordnary way of presentng the rates of payroll taxes n socal securty legslaton. 2 (A2) Indrect taxes: β1 ( y ' ) t = t ex = t * β * (6) * 0 k where ex stands for the total expendture of household. Because value added tax n Uruguay s hgh, we assumed that households pay ndrect taxes n proporton t of ther total expendture. But there s no nformaton on expendture n the household survey. So we approxmated household expendture as a (possbly non-lnear) functon of ncome, usng nformaton from the expendture survey of the Natonal Insttute of Statstcs. 3 The tax rates a and t can now be computed combnng equatons (5) and assumptons (A1) and (A2): 2 We chose ths notaton to avod the dstncton between employee and employer contrbutons, a dstncton we are not nterested n. Total payroll tax rates on post-tax labor ncome can be computed usng ordnary legal tax rates such as: (employer rate + employee rate)/(1-employee rate). The tax rate a s a weghted average of these transformed tax rates. 3 Other assumptons are of course possble and we dd some senstvty analyses, assumng for example that ndrect taxes are proportonal to total ncome rather than to total expendture. As mght have been expected, government programs look more progressve because taxes look less regressve wth ths alternatve assumpton, but the qualtatve results dd not change. These results are avalable on request from the authors. 10

11 a = t = α K b ( ) y C k LI ' ( 1 α ) b β ( y ') K β1 ( 0 ) k (7) Usng these tax rates and assumptons (A1) and (A2), we computed ndvdual tax payments a and t. We then computed ndvdual ncome y usng these estmated ndvdual tax payments n equatons (2) and (4). 3. Results We report the estmated decomposton of nequalty n Table 1. 4 The left panel was computed usng gross benefts and the rght panel was computed usng net benefts for the ncome sources correspondng to four government transfer programs: unemployment nsurance, pensons, famly allowances and a grouped set of other government programs. 5 Insert Table 1 The estmated contrbuton of these programs to nequalty s very dfferent dependng on whether gross or net benefts are used. The four government programs would have contrbuted to reducng nequalty n 2005 f ther contrbuton were evaluated usng benefts net of taxes pad to fnance the programs. But three of the programs would have contrbuted to ncreasng nequalty f only gross benefts were consdered n the computaton. Only famly allowances would have contrbuted to reducng nequalty 4 For the sake of brevty, we only report the estmatons for 2005, but we have smlar results for The other government programs nclude: donatons, subsdes, scholarshps, accdent compensaton, dvorce contrbutons and famly subsdy. 11

12 accordng to the gross-benefts measure. In the four cases, the contrbuton to nequalty s smaller when net rather than gross benefts are used. These dfferent results are drven by dfferent correlaton coeffcents of net and gross benefts to total ncome. Whle net benefts are negatvely correlated to total ncome n the four programs consdered n ths study, gross benefts are postvely correlated to total ncome n three of them (Table 1, columns 2 and 5). Famly allowances s the only program that presents a negatve correlaton of gross benefts to total ncome. In all programs, the correlatons to total ncome are more negatve when net rather than gross benefts are used. 4. Concludng Remarks Accordng to these results, assessng the mpact of government programs on nequalty lookng only at the benefts they pay could be msleadng, because taxes also count and the famles that receve hgher benefts also tend to pay more taxes. We showed a real world case n whch both expendtures and taxes looked regressve when analyzed separately, and yet the programs reduced nequalty when both sdes of the balance were ncorporated nto the analyss at the same tme. The dea that government s contrbuton to nequalty depends on both expendture and taxes s of course not new. In the words of Perry et al. (2006, p 96), The overall mpact of the government budget depends on the combned effect of taxes and expendture. And ndeed, ths pont s usually well taken care of n studes that compare pre- and post-transfer ncom e nequalty (see for nstance Beblo and Knaus, 2001; Atknson, 2004, Perry et al., 12

13 2006). However ths same pont s often overlooked n the analyss of ndvdual government programs. The usual clam that some programs are regressve because they spend more on hgh than on low ncome households s just one example of ths practce. One possble explanaton for ths usual shortcut to a one-sded approach to the evaluaton of ndvdual programs s the dffculty nvolved n estmatng how these programs are fnanced. But as we have shown n ths paper, ths shortcut can be rather msleadng. We propose nstead to make smple assumptons to approxmate benefts net of taxes, and to use them to assess the contrbuton of government programs to nequalty usng the framework of nequalty decomposton developed by Shorrocks (1982a). Lndert et al. (2005) share wth us concern for deductng contrbutons made to fnance socal nsurance programs to evaluate ther redstrbutve mpact. But ther approach departs from ours n that they only net out socal nsurance contrbutons. They do not extend ths reasonng to other publc programs nor do they compute all sources of fnancng for the programs for whch they estmate net benefts. In ther vew, the publc transfers nature of socal nsurance stems from the fact that these programs often spend more than they collect through socal securty contrbutons, and have thus to be partally fnanced out of general taxes. Accordngly, these authors assess the redstrbutve mpact of these programs lookng just at the porton of benefts that s fnanced by general tax revenues due to defcts n the penson system (Lndert, et al. 2005, p 105). Such an approach s not only partal, but t also nevtably gets tangled up wth the not-very-meanngful controverses about how the defcts of socal securty and other government programs should be computed. 6 We 6 One example s the clam that not all the assstance of the central government to socal securty should be computed as actual defct n Brazl, because part of t accounts for the employer contrbuton that the 13

14 advocate a more comprehensve approach that takes nto account all taxes collected to fnance the programs, for the mpact of a government program on nequalty depends not only on the taxes that are conventonally defned to fnance that specfc program but on all the sources of ncome that the government makes use of to fnance the program. Several authors have convncngly argued that fscal polces should deally be assessed consderng lfe-tme rather than just current ncome and transfers (Auerbach et al., 1999; Hardng et al., 2002; Mason et al., 2005). Estmatng lfe-tme ncome s not an easy task though, partcularly when only cross-secton data s avalable. Usng lfe-tme ncome to analyze redstrbuton s partcularly complcated for t requres performng mcrosmulatons. Also, the dynamc estmatons are senstve to the choce of the dscount rate and there s no smple rule to choose among dfferent rates. We dd not do detaled dynamc mcrosmulatons to llustrate the mportance of lookng at net benefts when assessng the contrbuton of transfer programs to nequalty, but t goes wthout sayng that the same pont llustrated here wth a statc example fully apples n a lfe-tme framework. If anythng, we expect ths pont to be more crucal n a dynamc than n a statc perspectve, because households that receve larger transfers n some perods of ther lves tend to be the same that pay more taxes n other perods. 5. References Atknson, Anthony (2004): Increased Income Inequalty n OECD Countres and the Redstrbutve Impact of the Government Budget. Inequalty, Growth and Poverty n an Era of Lberalzaton and Globalzaton. Oxford Unversty Press. government has to pay for publc employees. The lterature on socal securty s full of endless dscussons le ths. Furthermore, f the very concept of total fscal defct could be ll defned, as Auerbach et al. (1999) among others have argued, the defnton of the defct of one agency of the government s even more so. 14

15 Auerbach, Alan; Kotloff, Laurence y Lebfrtz, Wll (1999): Generatonal Accountng around the World. Natonal Bureau of Economc Research. The Unversty of Chcago Press. Beblo, Mram y Knaus, Thomas (2001): Measurng Income Inequalty n Euroland. The Revew of Income and Wealth, 47, Brownng, Edgar y Brownng, Jacquelne M. (1994): Publc Fnance and the Prce System. Englewood Clffs, New Jersey 09632, Prentce Hall. Feldsten, Martn (2001): Unemployment Compensaton: Adverse Incentves and Dstrbutonal Anomales. Natonal Tax Journal, 27, Hardng, Ann; Percval, Rchard; Schofeld, Deborah y Walker, Agnes (2002): The Lfetme Dstrbutonal Impacts of Government Health Outlays. The Australan Economc Revew, 35, Lndert, Kathy; Skoufas, Emmanuel y Shapro, Joseph (2005): Redstrbutng Income to the Poor and the Rch: Publc Transfers n Latn Amerca and the Carbbean. The World Bank, 123. Mason, Andrew; Lee, Ronald; An-Ch, Tung; La, Mun-Sm y Mller, Tm (2005): Populaton Agng and Intergeneratonal Transfers: Introducng Age nto Natonal Accounts. Unversty of Calforna at Berkley, 30. Mazza, Jacquelne (1999): Unemployment Insurance: Case Studes and Lessons for Latn Amerca and the Carbbean. IADB, 85. Perry, Gullermo; Aras, Omar S; López, J. Humberto; Maloney, Wllam F y Servén, Lus (2006): Poverty Reducton and Growth: Vrtuous and Vcous Crcles. The World Bank, Washngton DC. Shorrocks, Anthony (1982a): The Impact of Income Components on the Dstrbuton of Famly Incomes. Quarterly Journal of Economcs, 98, Shorrocks, Anthony (1982b): Inequalty Decomposton by Factor Components. Econometrca, 50,

16 Fgure 1: The Contrbuton of Government Transfer Programs to Inequalty Taxes Benefts, Taxes Benefts Total ncome Benefts mnus Taxes Total ncome 16

17 Table 1: The Contrbuton of Several Sources of Income to Total Inequalty (2005) Sources of Income Contrbuton to Inequalty a/ (n percent) Estmaton 1: Gross Benefts Correlaton Rato of Standard Coeffcents Devatons Contrbuton to Inequalty a/ (n percent) Estmaton 2: Net Benefts Correlaton Rato of Standard Coeffcents Devatons Labor Income, dependent formal workers Labor Income, dependent nformal workers Labor Income, self employed Unemployment Insurance Pensons Famly Allowances Other Government Transfers b/ Other Sources of Income c/ TOTAL Notes: a/ The contrbuton of the ncome source (y k ) to total ncome (y) nequalty s the slope n the regresson of y k on y. See Shorrocks (1982b) b/ Donatons, subsdes, scholarshps, accdent compensaton, dvorce contrbutons, famly subsdy. c/ Interests, rents, profts, severance payments, remttances from abroad, other sources of ncome. Source: Own computatons on the household survey 17