WP/06/275 The Myh of Pos-Reform Income Sagnaion in Brazil Irineu de Carvalho Filho and Marcos Chamon
2006 Inernaional Moneary Fund WP/06/275 IMF Working Paper Wesern Hemisphere and Research Deparmens The Myh of Pos-Reform Income Sagnaion in Brazil Prepared by Irineu de Carvalho Filho and Marcos Chamon 1 Auhorized for disribuion by Sanjaya Panh and Paolo Mauro December 2006 Absrac This Working Paper should no be repored as represening he views of he IMF. The views expressed in his Working Paper are hose of he auhor(s) and do no necessarily represen hose of he IMF or IMF policy. Working Papers describe research in progress by he auhor(s) and are published o elici commens and o furher debae. This paper uses Engel curves o esimae real income growh in Brazil. The esimaed per capia household real income growh in meropolian areas during 1987 2002 is abou 4½ percen per year, well above he headline growh of 1½ percen obained by deflaing nominal incomes by he CPI. This suggess a subsanial CPI bias during ha period, likely owing o one-off effecs of rade liberalizaion and inflaion sabilizaion. The esimaed unmeasured gains are higher for poorer households, implying a marked reducion in real inequaliy. This finding challenges he convenional wisdom ha pos-reform real income growh in Brazil was low. JEL Classificaion Numbers: D12; E01; I32; O10; F15 Keywords: CPI bias, Trade Liberalizaion, Inflaion Sabilizaion, Economic Reform Auhors E-Mail Addresses: idecarvalhofilho@imf.org; mchamon@imf.org 1 We are graeful o Marin Cerisola, Dora Cosa, Angus Deaon, Marcello Esevão, André Faria, Francisco Ferreira, Michael Kremer, Timohy Lane, Paolo Mauro, Jeromin Zeelmeyer and seminar paricipans a he IMF, he World Bank Microeconomics of Growh Conference, EEA 2006, LACEA 2006, and IMF/WHD Deparmen Workshop 2006 for helpful commens. Any errors are our own.
2 Conens Page I. Inroducion... 3 II. Empirical Mehodology... 6 III. Daa... 8 IV. Resuls...10 A. Parameric Model...12 B. Semi-Parameric Model...12 C. Robusness...14 D. Disribuional Implicaions...15 V. Evidence From Durable Goods Ownership and Anhropomerics...15 VI. Conclusion...16 References...33 Tables 1. Mean Expendiure Shares, Median Toal Expendiure, by Expendiure Quinile in 1987/88...18 2. Descripive Saisics for Regression...19 3. Regression Resuls for Pooled Sample... 20 4. Bias Esimaes Across Differen Mehods and Samples... 21 5. Household Per Capia Expendiure and Ne Income: Headline and Correced...22 6. Expendiure Inequaliy Correced for CPI Bias: Expendiure Gini Coefficiens...23 7. Ownership of Durable Goods, and Sensiiviy o Income...24 8. Anhropomeric Measures for Children 0 60 Monhs Old...25 Figures 1. GDP Per Capia and Average Growh In Decade for Brazil...26 2. Food Budge Share by Expendiure Deciles...26 3A. Food Budge Share by Meropolian Area...27 3B. Food Expendiure Per Capia, by Meropolian Area...27 4. Changes in Relaive Prices...28 5. Non-Parameric Esimaes of Relaionship beween Food Shares and Household Expendiure...29 6. Esimaed Bias in 1987/88 1995/96 as a Funcion of CPI-Measured Real Expendiure in 1995/96...30 7. Esimaed Bias in 1995/96 2002/03 as a Funcion of CPI-Measured Real Expendiure in 2002/03...30 8A. Disribuion of Expendiure Deflaed by he CPI: 87/88, 95/96 and 02/03...31 8B. Disribuion of Expendiure Deflaed by he Esimaed True Cos of Living Index: 87/88, 95/96 and 02/03...31 9. Changes in Durable Goods Holdings and Sensiiviy o Income... 32
3 I. INTRODUCTION The pos-reform economic growh experience in Lain America has failed o fulfill he high expecaions held a he ime hose counries embarked on heir rade liberalizaion and privaizaion effors. As early as 1995, a number of observers, such as Krugman (1995) and Easerly, Loayza, and Moniel (1997) sared o wonder wheher Lain America s pos-reform growh had been disappoining. A decade laer, economiss remain disappoined and somewha puzzled a how economic growh failed o pick up despie he removal of so many disorions from hose economies. In his paper we presen household-level evidence ha convenional measures of real income growh grossly underesimae he rue pos-reform income growh for Brazil. I is possible ha he same applies o a number of oher Lain American counries. Afer several decades of very high growh in per capia GDP, he Brazilian economy came o a hal in he 1980s, wih growh during ha decade essenially zero (Figure 1). The convenional wisdom a ha ime was ha such economic sagnaion was he consequence of he failure of he impor-subsiuion developmen sraegy a coping wih adverse exernal shocks. Hence, expecaions were high when Brazil began o open up and liberalize is economy. For decades, Brazil had been one of he world s mos closed economies, wih very high ariff and non-ariff barriers, including ourigh bans on he imporaion of several goods. In he lae 1980s/early 1990s, Brazil began o cu ariffs subsanially and o dismanle mos non-ariff barriers; and by he mid-1990s i was already a relaively open economy (paricularly in comparison wih is former self). 2 Alhough impor peneraion remained relaively small as a percenage of GDP, compeiive pressures following rade liberalizaion led o dramaic improvemens in he produciviy of Brazilian manufacuring firms. 3 Inflaion was finally brough under conrol in 1994, wih he Plano Real, afer decades of chronic high inflaion. Bu, as Figure 1 indicaes, Brazil s pos-reform growh apparenly never lived up o reformers opimisic expecaions. 4 There is an exensive lieraure arguing ha changes in he Consumer Price Index (CPI) overesimae he increase in cos of living in he Unied Saes. 5 One of he main sources of CPI overesimaion is he lae inroducion of new goods ino he CPI baske. I is ofen he case ha new goods are iniially very expensive and only consumed by a few households. Bu over ime, new goods end o become cheaper and more widely consumed. The longer i akes 2 The (unweighed) average ariff for 53 final goods colleced by Kume, Piani, and de Souza (2000) declines from 55 percen in 1987 o 31 percen in 1990, and o only 11 percen in 1995. 3 Ferreira and Rossi (2003) aribue an increase in annual Toal Facor Produciviy (TFP) growh in Brazilian indusry of more han 5 percen during 1991 97 o he effecs of rade liberalizaion. Muendler (2001) finds ha increased foreign compeiion pressured Brazilian manufacurers o raise produciviy markedly during he same period. 4 Alhough he focus of his paper is on household income growh, GDP growh was used in Figure 1 and in he moivaion owing o is broader hisorical coverage. 5 For an overview of his lieraure, see Naional Research Council (2002) and Hausman (2003).
4 for a new good o be inroduced ino he CPI baske, he less of is price decline ha is capured by he CPI index. Anoher source of CPI overesimaion is subsiuion bias. Consumers end o economize on goods whose relaive prices have increased, while buying greaer quaniies of goods whose relaive prices have declined. The failure o accoun for his behavior has long been known as a cause of overesimaion of he increase in he cos of living, he more so when here are long lags beween adjusmens of he CPI baske and large changes in relaive prices. As Brazil gradually opened is economy o inernaional rade in he lae 1980s/early 1990s, many innovaions ha were gradually inroduced in he Unied Saes suddenly became available and more affordable o Brazilian households. However, he firs pos-liberalizaion revision of he Brazilian CPI ook place only in 1999, based on he 1995/96 expendiure survey (e.g. i was only hen ha compuers and cell phones were inroduced ino he CPI baske and large revisions ook place on he weighs of some goods). 6 This caching-up effec combined wih a relaively lae adjusmen in he CPI baske suggess ha, during he sample period, he CPI bias in Brazil may have been much larger han ha in he Unied Saes. 7 Anoher poenially imporan source of bias is he CPI s failure o accoun for qualiy improvemens in exising goods, which were also considerable in he pos-reform Brazil of he 1990s. The sabilizaion of inflaion achieved by he Plano Real in 1994 may also have ranslaed ino improvemens in living sandards no fully capured by sandard measures of real income. Neri (1995) calculaes ha bringing down inflaion from 40 percen o 1 percen per monh increases he purchasing power of hose wihou access o shor-erm financial invesmens by 9 percenage poins. Since he poor are less likely o have ineres-earning bank accouns, hey are worse-off under hyperinflaion relaive o a low-inflaion environmen even afer adjusing heir income by CPI inflaion. We calculae he real income growh in Brazil implied by Engel curves, following he mehod used by Nakamura (1997), Cosa (2001) and Hamilon (2001) for he Unied Saes. Engel s law saes ha he share of food in oal household expendiures declines as (real) income grows, and i is among he sronges empirical regulariies in economics. 8 We esimae a model for household-level budge share of food as a funcion of real income, relaive prices, and household characerisics, using daa from hree differen household surveys. By comparing he real income growh implied by he observed changes in he budge share of 6 For example, he weigh of auomobile purchases in he CPI baske doubled following his revision, going from 1.01 percen o 3.38 percen for new auomobiles and from 1.42 percen o 2.29 percen for used ones. 7 Alhough he ime inervals beween updaes in Brazil s CPI baske are no unusually long, heir iming was such ha he firs updae reflecing pos-liberalizaion consumpion paerns only ook place in 1999, a decade afer he liberalizing reforms began. 8 Of all empirical regulariies observed in economic daa, Engel s Law is probably he bes esablished; indeed i holds no only in he cross-secion daa where i was firs observed, bu has ofen been confirmed in ime-series analysis as well (Houhakker, 1987).
5 food wih he headline real income growh obained by deflaing nominal income by he CPI, we can esimae he mismeasuremen of he laer. Assuming nominal income is measured accuraely, we aribue he difference beween he real income growh based on our esimae of he rue cos of living and he headline real income growh o measuremen error in he CPI. This approach can capure he overall mismeasuremen of real income ha may resul from facors such as he lae inroducion of new goods ino he CPI bundle, consumer subsiuion, increases in he durabiliy of goods, improvemens in disribuion neworks, and mismeasuremen of prices. I can also capure bias owing o improvemens in he qualiy of goods o he exen here is subsiuion away from luxury foods oward he goods whose qualiy has improved. This paper also focuses on disribuional issues as i uses a semiparameric mehodology ha allows for variaion in he esimaed bias a differen poins in he income disribuion. Household income and expendiure surveys were conduced in Brazil in 1987/88, 1995/96, and 2002/03. During he ime span of hose surveys, he budge share of food declines from poorer o richer expendiure quaniles, and has also declined over ime wihin each quanile and across he income disribuion. Cosa (2001) and Hamilon (2001) esimae he CPI bias in he Unied Saes a roughly 1 percenage poin per year since he 1980s. Their esimaes are similar o hose of he Boskin Commission, which esimaed he bias a 1.1 percenage poins per year in 1995 96 (Boskin and ohers, 1996). Comparison wih hose esimaes suggess a much more sizable undersaemen of real income growh in Brazil. Our preferred bias esimaes imply ha for households in he meropolian areas covered by he CPI, here was a discrepancy beween he gross change in he rue cos of living and he gross change in he headline CPI of abou 3 percen per year during 1987/88 2002/03. 9 This implies ha real income growh was subsanially higher han he headline growh obained by deflaing nominal incomes by he CPI. When our esimaes allow he bias o vary wih household expendiure levels, we find larger unmeasured improvemens in real income for poorer households, implying a subsanial reducion in income inequaliy when measured in real erms. This paern is a leas parly driven by high inflaion disproporionaely huring he poor, as suggesed by Neri (1995). The aggregae bias esimaes of 3 percen per year are obained by averaging he bias esimaes a differen levels of expendiure, weighing by household expendiure (since he CPI iself is based on an aggregae consumpion bundle where individual household bundles are weighed by heir expendiure). The bias facing he ypical household in he sample (corresponding o he average bias esimaed weighed by populaion and sampling weighs), is abou 5 percen per year. The remainder of his paper is organized as follows. Secion II describes he mehodology. Secion III describes he daa. Secion IV presens he main resuls, based on Engel curves for 9 Tha is, if he change in he CPI in a given period were 10 percen and he esimaed bias 3 percen, he esimaed change in he rue cos of living would be 6.7 percen, since (1-0.03) 1.1=1.067.
6 food consumpion. Secion V presens alernaive corroboraing evidence based on ownership of durable goods and anhropomerics. Secion VI concludes. II. EMPIRICAL METHODOLOGY Our parameric esimaes of real income growh follow he mehod developed in Cosa (2001) and Hamilon (2001), building on an insigh by Nakamura (1997). We sar wih he demand funcion for food ha emerges from Deaon and Muellbauer s (1980) Almos Ideal Demand Sysem: w = φ+ γ(ln P ln P ) i, j, F, j, N, j, + β (lny ln P ) x i, j, G, j, + θ X + µ x i, j, i, j, (1) where he subscrips refer o household i, region j, and period ; w is he share of food in oal household expendiures; P F, P N and P G are he rue bu unobservable price indices of food, nonfood and he general index for all goods; Y is he household's nominal expendiure; X is a vecor of household characerisics; and µ is he residual. A negaive β characerizes a necessiy good while a posiive β characerizes a luxury or superior good. The rue price index P G is measured wih CPI error. Le Π G,j, denoe he percen cumulaive increase in he CPI measured price and E G,j, denoe he percen cumulaive measuremen error from period 0 o period, for food, nonfood or all goods, as indicaed by he subscrip. Equaion (1) can be rewrien as: w = φ + γ (ln(1 +Π ) ln(1 +Π )) i, j, F, j, N, j, + β (lny ln(1 +Π )) i, j, G, j, + γ(ln P ln P ) β ln P F, j,0 N, j,0 G, j,0 + γ(ln(1 + EF, j, ) ln(1 + EN, j, )) βln(1 + EG, j, ) + θ X x i, j, + µ i, j, x (2) We assume ha he CPI measuremen error does no vary geographically, and rewrie (2) as: w = φ+ γ(ln(1 +Π ) ln(1 +Π )) i, j, F, j, N, j, + β (lny ln(1 +Π )) + δ D + j x i, j, G, j, δ D j j + θ X + µ x i, j, i, j, (3) where D j and D are regional and ime dummies and:
7 δ = γ(ln P ln P ) βln P j F, j,0 N, j,0 G, j,0 δ = γ(ln(1 + E ) ln(1 + E )) βln(1 + E ) F, N, G, All he erms in Equaion (3) are observable and once he equaion above has been esimaed, we are ready o compue he cumulaive CPI bias. If food and nonfood are equally biased, hen: ln(1 + E ) = δ / β (4) G, I seems very likely ha mismeasuremen is less of a problem for food han for nonfood goods. As a resul, o he exen ha food is a necessiy ( β < 0 ) and food shares increase wih he relaive price of food ( γ > 0 ), one can show ha equaion (4) in general undersaes he bias for small posiive values of γ as in our esimaes. The parameric specificaion discussed above assumes ha all households a a given dae face he same bias. In he conex of a high inequaliy counry, i is paricularly relevan o inquire abou differenial effecs across he income disribuion. The esimaion of (3) hrough minimizaion of squared errors yields an esimae of he bias for he average household. However, he acual CPI index is based on an aggregae consumpion bundle ha by design disproporionaely represens richer households as hey accoun for a disproporionae share of aggregae consumpion. Thus, o he exen ha he discrepancy beween he rue cos of living index and he headline CPI varies across he income disribuion, here migh be subsanial differences beween he bias facing he average household and he bias for he aggregae consumpion bundle, which is he one ha maps o he CPI. The model in equaion (3) can be exended o address his concern, by assuming he bias is a linear funcion of he log of real expendiure: ln(1 + E ) = a + b(lny ln(1 +Π )) G, i, i, j, G, j, Mainaining he same assumpion ha food and non-food are equally biased and ha he bias does no vary by region, one can esimae: w = φ + γ (ln(1 +Π ) ln(1 +Π )) i, j, F, j, N, j, + ( β + λd )(lny ln(1 +Π )) + δ D + j x i, j, G, j, δ D j j + θ X + µ x i, j, i, j, and obain he following expression for CPI bias a differen poins in he expendiure disribuion.
8 δ λ ln(1 + EGi,,) = + (lnyi, j, ln(1 +Π G, j,)) (5) β β Finally, we can use a flexible non-parameric approach for he bias by esimaing a semiparameric version of he demand funcion (1), allowing for esimaion of he bias a differen levels of expendiure. This semi-parameric approach faciliaes he expendiure-weighing of esimaes so as o map ino he acual CPI and also enables us o idenify he implicaions of CPI bias for he evoluion of he (real) income disribuion. Sill under he assumpion ha food and non-food are equally biased and ha he bias does no vary by region, we can rewrie (3) as: w = φ + γ (ln(1 +Π ) ln(1 +Π )) i, j, F, j, N, j, + f (lny ln(1 +Π ) ln(1 + E )) x i, j, G, j, G, i, + θ X + µ x i, j, i, j, We esimae f(lnyi, j, ln(1 +ΠG, j, ) ln(1 + EG, i, )) non-paramerically using he differencing mehod described in Yachew (1997). 10 The bias a a given CPI-measured real income level a ime is hen esimaed as he increase in CPI-measured real income ha would have prevened he Engel curves from shifing. Tha is, we solve a each expendiure level for he value of E Gi,, ha saisfies: f (lny ln(1 +Π ) ln(1 + E )) = f (lny ln(1 +Π )) i, j, G, j, G, i, 0 i, j, G, j, III. DATA The Pesquisa de Orçamenos Familiares (henceforh, POF) is he household income and expendiure survey carried ou by he Insiuo Brasileiro de Geografia e Esaísica (IBGE) and based on probabilisic sample and sraified design. The POF surveys were conduced in 1987/88, 1995/96, and 2002/03. The firs survey covers a ime when Brazil was sill a very closed economy and ariffs choked off mos rade in manufacured goods. By he ime he second survey was conduced, Brazil was already a relaively open economy (paricularly in comparison wih is former self) and inflaion had already been brough down o single digi levels. Hence, whereas he 1987/88 and 1995/96 surveys were fielded in subsanially differen economic environmens, he changes beween 1995/96 and 2002/03 were smaller. The 1987/88 and 1995/96 POF surveys are represenaive of he household populaion in 9 10 In a nushell, he mehod consiss of esimaing he parameric erms afer differencing he daa (under he assumpion ha f() is a smooh funcion), and esimaing f(), he non-parameric erm, on he difference beween he dependen variable and he esimaed parameric erms in he equaion, using locally-weighed linear regressions using quaric kernel weighs.
9 meropolian regions (Belém in he Norhern region; Foraleza, Recife and Salvador in he Norheas; Rio de Janeiro, Belo Horizone and São Paulo in he Souheas; Curiiba and Poro Alegre in he Souh), plus he Disrio Federal and he municipaliy of Goiânia, in he Cener- Wes region. The 2002/03 POF was he firs o cover a naionally represenaive sample. In order o make he samples comparable over ime, we limi he 2002/03 sample o he geographical sub-sample also covered by he previous surveys. Goiânia is excluded from our sample because i did no have is own region-specific price index prior o 1991. The POF conains household expendiure informaion on individual goods a a disaggregaed level, households invenory of durable goods, and also demographic, schooling and income characerisics of each household member. IBGE produces and disseminaes he Índice de Preços ao Consumidor Ampliado (IPCA), which is considered Brazil s official consumer price index, and is based on he baske consumed by he POF families. The IPCA is also he broades measure of consumer prices available, covering all families earning up o 40 imes he minimum wage in he geographical areas covered by he POF. A few years afer each POF is colleced, he IPCA index is reweighed, hence caching-up wih changing consumpion paerns of Brazilian households. Throughou he paper, we will be referring o he IPCA whenever he CPI in Brazil is menioned. The POFs are conduced over a 12-monh period. The field work ook place in: March 1987 February 1988, Ocober 1995 Sepember 1996, and July 2002 June 2003. Households are asked o provide informaion on heir expendiures, which are colleced based on differen reference periods depending on he ype of expendiure and is frequency. Four reference periods are used: 7 days, 30 days, 90 days, and a longer recall period of eiher 12 monhs (2002/03 POF) or 6 monhs (1987/88 and 1995/96 POFs). 11 The POF uses six collecion insrumens for is daa: Housing condiions quesionnaire, collecive expendiures quesionnaire, collecive expendiures noebook, individual expendiures quesionnaire, individual income quesionnaire and living condiions quesionnaire. Expendiures on iems of frequen use by he household were recorded on he collecive expendiures noebook. Those iems include mainly food, beverages, cleaning and personal hygiene producs. Tha noebook was kep during a 7-day period by he person who manages hose expendiures in he household budge. During ha period, he expendiures on hose frequen use producs were recorded, as well as he amoun and locaion of purchase. This noebook is he source of informaion for expendiures on food and beverages o be consumed a home. Expendiures on food and beverage consumpion ouside he home were recorded on he individual expendiures quesionnaire, which was based on recall over a seven-day reference period. Given he lengh of he collecion, reference and recall periods, each POF conains expendiure daa spanning a 24-monh period. In order o make he daa comparable across households surveyed over differen periods, IBGE deflaes or inflaes all resuls o a same 11 Because here are differences in he recall window across POF surveys, inference on expendiure disribuion and inequaliy across surveys could be biased (Deaon and Kozel 2005).
10 reference dae. The reference daes for he POFs used are: Ocober 15, 1987; Sepember 15, 1996; and January 15, 2003. The deflaion of expendiure daa o he reference dae uses iemspecific deflaors. 12 For example, if a household bough rice and beans on March 4 h 2003, hose wo expendiures are deflaed o January 15, 2003 values using wo differen deflaors. The deflaed expendiures are also annualized in he POF daabase, by muliplying he expendiure by a facor equal o 12 monhs divided by he reference period for he expendiure (for example, a facor of 52 for he seven-days reference period). In he hree surveys, here are respecively 1.0, 3.9 and 4.8 percen of he households in our full sample ha failed o repor any expendiures in he collecive expendiures noebook (which covers mainly food for home consumpion and cleaning and hygiene producs). I is possible ha some of hese households did no have any expendiures o repor. Bu i is also likely ha some of hem may no have complied wih filling he collecive expendiures noebook, urning i blank regardless of heir expendiure behavior. Our preferred esimaes will be based on he subse of households wih a leas one expense record in he collecive expendiures noebook, henceforh referred as he complian sample, bu we will also presen resuls for he full sample. IV. RESULTS Since our empirical sraegy relies on Engel s law, i is useful o documen ha food is indeed a necessiy good. Table 1 shows he budge share of differen consumpion groups in 1987/88 for each quinile of he oal expendiure disribuion. The budge share of food indeed srongly declines wih oal expendiure. While households in he boom quinile of oal expendiure spen on average over 40 percen of heir budge on food, he ones in he op quinile spen only 16 percen. In Table 1, budge shares end no o vary much wih oal expendiure for mos oher groups of goods, wih he excepion of educaion and ransporaion, boh superior goods, i.e., goods whose budge share increases wih oal expendiure. 13 Households in he boom half of he disribuion spend less han 10 percen of heir budge on ransporaion, whereas households in he op quinile spend over 20 percen, mainly because hey usually own a privae vehicle. Figure 2 shows he budge share for food by decile of household expendiure for he hree surveys. The budge share of food declines from poorer o richer deciles, as well as over ime wihin each decile. Figure 3A shows he evoluion of he average food budge share by meropolian area. For each area, and for each subsequen pair of years, he budge share of food declines over ime. Figure 3B shows per capia food expendiure by meropolian area for he hree survey years. On average, here was an increase in per capia real food expendiure (in 1996 reais) for each subsequen survey. Four meropolian areas showed improvemens beween each survey (Recife, Salvador, Curiiba, and Poro Alegre), while for he oher areas 12 If prices are no colleced for a given iem, he price for is sub-group is used. 13 Goñi, López, and Servén 2006 find ha ransporaion and communicaion are superior goods for households in Brazil, Colombia, Peru, and Mexico, and educaion is a superior good for Brazil and Peru, bu no for Colombia and Mexico.
11 here was a leas one episode of decline. However, real per capial food expendiure was higher in 2002/03 han in 1987/88 for all meropolian areas. Figure 4 shows he relaive prices for differen groups of goods (vis-à-vis he aggregae price index). Noe he persisen increase in he relaive price of mos nonradable goods (e.g., housing and healh care), and he large decline in he relaive price of mos radable goods (e.g., clohing and household supplies) over ime. Similar rends are experienced in oher counries, and fi he prior of greaer echnological progress in he radable secors. This rend seems o sharpen around he early 1990s, suggesing ha rade barriers were previously dampening hese secular changes in he relaive price of radable and nonradable goods. The divergence in relaive prices shown in Figure 4 highlighs he poenial for CPI mismeasuremen from oudaed weighs in he consumpion baske and subsiuion bias. We firs esimae rue real income growh (i.e., growh in income deflaed by he rue cos of living index) based on equaion (3). Following Cosa (2001), we use oal expendiure insead of income because expendiures beer reflec permanen income. Our specificaion allows for regional variaion in relaive prices. The conrols for household characerisics include dummies for home ownership and renal; gender of he head of he household; presence of a spouse; wheher he head of he household, he spouse, or boh have labor income; and he number of household members in age groups: 0 o 4; 5 o 9; 10 o 14; 15 o 19; and 20+ years old. Table 2 provides summary saisics for our complian sample. The average budge share of food was 30½ percen, 26½ percen, and 23¼ percen in he 1987/88, 1995/96, and 2002/03 surveys, respecively. The relaive price of food declined beween he firs wo surveys bu remained sable aferwards. Average household size declined considerably, as indicaed by he sagnan average oal household expendiures despie a sizable increase in per capia expendiure. Noe ha he per capia expendiure on food deflaed by he food CPI has increased (even hough he food share in he budge has declined). This rules ou possible concerns ha he declining budge shares may be due o households being forced o cu back on food in order o mee oher expendiures. Family composiion changed over ime, wih a reducion in he households where a spouse was presen and an increase in he households no headed by a male. The surveys also indicae an increase in he likelihood of he head of he household no having labor income, perhaps due o a combinaion of igher labor marke condiions, shifing demographics, and expansion of enilemen programs. 14 Spousal labor remained more sable. The expendiure daa do no include he renal equivalen value of owner-occupied housing. If ha value changed over ime a a differen rae han he general price index, failure o ake i ino accoun could bias our esimaes for he mismeasuremen of he rue cos of living. In order o address his problem, we presen resuls for wo samples: one including only reners and he oher including all households. Noe ha no all non-reners are owners (some households live in a housing uni provided by ohers, for example employers or relaives, 14 For evidence on a sizeable effec of cash benefis on labor paricipaion in Brazil, see de Carvalho Filho (2005).
12 wihou ren being charged). Since expendiures are poenially measured wih error, we also presen esimaes where income is used as an insrumen. This is paricularly imporan because aenuaion bias would end o drive down he absolue value of he coefficien on he log of expendiure, hence increasing he esimae of he CPI bias in equaion (4). A. Parameric Model Table 3 repors he linear regression resuls for a sample pooling he hree surveys. The coefficiens on he ime dummies are negaive and saisically significan, as expeced given he average decline in he budge share of food repored in Figure 2. The esimaed coefficiens on log expendiure range from -0.07 o -0.09 for he full sample, which implies income elasiciies in line wih previous esimaes for Brazil (e.g., Asano and Fiuza, 2001). The coefficiens on he log of he relaive price of food are no saisically significanly differen from zero (which may reflec compeing income and subsiuion effecs, or limied regional variaion in relaive prices of food). 15 Table 3 also repors he resuling esimae for he cumulaive CPI bias, E G,, which is negaive as expeced (he values repored correspond o is absolue value). Given ha negaive bias, he implied gross change in he rue cos of living is (1- E G, ) imes he gross change in he CPI, and he resuling gross rue real income growh is 1/(1- E G, ) imes he gross real income growh obained by deflaing nominal income by he CPI. For 1987/88 and 1995/96 he implied cumulaive bias for our preferred specificaion, which uses insrumenal variables o reduce aenuaion bias, was 45.8 percen for he enan sub-sample and 34.0 percen for he full sample. The corresponding annualized raes are: 6.9 percen and 4.7 percen, respecively. The annualized raes were compued based on he midpoins of he surveys. For 1995/96 and 2002/03 he cumulaive bias in he IV specificaion was 0.3 percen for enans and 18.1 percen for he full sample. The corresponding annualized raes are 0.0 percen and 2.9 percen, respecively. These esimaes indicae a sizable bias, bu declining over ime. The paern of iniially large and declining esimaed bias migh be due o he lae pos-liberalizaion revision in he CPI baske in 1999 and one-off gains from disinflaion beween 1987/88 and 1995/96. The cumulaive bias beween 1987/88 and 2002/03 is very similar o he one in Table 3 if esimaed based on wo separae regressions using consecuive pairs of surveys. Table 4 summarizes he esimaes above, presens esimaes based on differen samples, and esimaes when he bias is as a linear funcion of he log of real expendiure, as in equaion (5) of he previous secion and he semi-parameric specificaion discussed below. B. Semi-Parameric Model We now urn o he semi-parameric esimaion of he model. Figure 5 shows he nonparameric esimaes of he relaionship beween he food budge share and he log of real expendiure. 16 As expeced, he food budge share declines wih real expendiure and he 15 This is no likely o affec he hrus of our resuls since he relaive price of food remained roughly consan beween he second and hird surveys, ye we find subsanial CPI bias during ha period. 16 The esimaes are based on locally-weighed linear regressions wih quaric kernel weighs.
13 curves have shifed downward wih each survey. For low levels of expendiure, his relaionship is non-monoonic, bu he proporion of households in ha expendiure range is relaively small. I is anicipaed ha heir conribuion o esimaed expendiure-weighed aggregae CPI bias. The overwhelming majoriy of he households are in he expendiure range where he food budge share declines as real expendiure increases. As discussed in Secion 2, he food share-real expendiure profiles in Figure 5 map ino a CPI bias-real expendiure profile by compuing he necessary change in real expendiure, a each real expendiure level ha would mainain he Engel curves in he same posiion. For example, for each level of expendiure in he 1995/96 survey, CPI bias is deermined by is difference o he real expendiure level in 1987/88 ha was associaed o he same food budge share. For CPI measured real expendiure R we can solve for he corresponding bias in E ( ) 1995/1996 R as he soluion o: f (ln R ln(1 + E ( R)) = f (ln R) 1995/96 1995/96 1987 /88 where f 1995/96 and f 1987 /88 are he esimaed Engel curves for 1995/96 and 1987/88. 17 Since we rely on Engel s Law, such mapping is only meaningful in he range where he food share is declining on income. The daa, however, someimes shows non-monooniciy of he Engel curve for low levels of expendiure. To address his issue, a he lef ail of he L expendiure disribuion, we se he bias o equal ha of he firs expendiure level R in he declining range of he curve for which he food share is below he one for he lowes level of expendiure R. In more precise erms, we impose he consrain: min L L E ( R) = E ( R ) if R< R, 1995/96 1995/96 1 L where ( ( )), L R = f f R f ( R ) < 0 1995/96 1995/96 min 1995/96 A he righ-ail of he expendiure disribuion, here are levels of he food share in 1995/96 for which here are no counerpars in he 1987/88 Engel curve. In hose cases we se he bias U o equal ha of he highes expendiure level R in he 1995/96 curve for which he mapping ino he 1987/88 curve is sill possible (o he highes value of expendiure R max ). Tha is: U U E ( R) = E ( R ) if R> R, 1995/96 1995/96 U U where R = R (1 + E ( R )) max 1995/96 17 The Engel curves f are obained afer conrolling for all he righ-hand side variables used in he linear regressions repored in Table 3.
14 Similarly, he bias from 1995/96 o 2002/03 is esimaed by compuing he real expendiure growh ha would make he 2002/03 Engel curve mach he one for 1995/1996, subjec o he same adjusmens a he ails of he expendiure disribuion. Figure 6 shows he esimaed annual bias from 1987/88 o 1995/96 as a funcion of headline real expendiure, as well as he esimaed densiy funcion of he log of CPI-deflaed real expendiure in 1995/96. The bias is higher for he poores households, and gradually declines as real expendiure increases. The annualized bias for he average household is 5½ percen per year, whereas he expendiure weighed aggregae bias is only 2 percen per year. Figure 7 shows he esimaed annual bias from 1995/96 o 2002/03. The profile is flaer han he one in Figure 6, suggesing ha he differences in CPI bias across he income disribuion have narrowed. The bias for he average household is abou 4½ percen while he expendiure weighed aggregae bias is 4¼ percen. C. Robusness Table 4 repors he esimaed biases under differen mehods and samples. The firs se of esimaes are based on he linear regression model as in Table 3. The second se of esimaes is based on a parameric specificaion for he bias as a linear funcion of he log of real expendiure. Finally, he hird se of esimaes is based on he semi-parameric approach discussed above. I presens resuls for four differen samples: he preferred complian sample, he full sample, he enans sample, and a winsorized sample, where he value of food and oal expendiures for observaions below he 5 h and above he 95 h percenile are se o he value a ha percenile. For he firs period, from 1987/88 o 1995/96, he populaion sample weighed bias esimaes are always above 3¾ percen per year, and as high as 9½ percen per year for he full sample and OLS esimaes. Since he full sample is likely conaminaed by observaions ha did no repor acual expendiure in he survey, hese esimaes should be read wih cauion. For he complian sample, IV bias esimaes are smaller han OLS ones, which is suggesive of aenuaion bias due o mismeasuremen of he real expendiure variable. The evidence from he mehods ha allow for variaion in he bias esimaes across he expendiure disribuion suggess ha CPI during he 1987 1996 period was more oversaed for poorer han for richer households. Expendiure-weighed bias esimaes range beween 2 and 3 percen, wih he excepion of he sample of enans where one canno rejec zero bias. For he second period in sudy, from 1995/96 o 2002/03, he populaion sample weighed bias esimae remained above 2¾ percen for all samples bu for he enan sample for which no saisically significan bias was found. For he expendiure weighed bias, which maps o he CPI aggregae baske, esimaes range from 2¾ o 4½ percen again wih he excepion of he sample of enans for which no saisically significan bias was found. In summary, he differen esimaes are broadly comparable. In our preferred specificaions, he bias for he average household is 4½ 6 percen per year in 1987/88 1995/96 and 3 5 percen per year in 1995/96 2002/03. The esimaes for he aggregae bias, where
15 households are weighed by expendiure, are 2 3 percen per year in 1987/88 1995/96 and 2½ 4 percen per year beween 1995/96 2002/03. 18 D. Disribuional Implicaions The bias esimaes shown in Figures 6 and 7 imply a subsanial decline in real inequaliy. Figure 8a plos he disribuion of CPI-deflaed real expendiures, while Figure 8b plos he disribuion of our esimaes of he rue real expendiure implied by acual food budge shares in 1995/96 and 2002/03 and he Engel curve in 1987/88. While he disribuions in Figure 8a are virually overlapping, Figure 8b indicaes a marked improvemen, wih he disribuion moving o he righ and becoming more equal. Table 5 repors he CPI-deflaed headline and he bias adjused real expendiure and ne income per capia for he average and median households, as well as for he boom and op quiniles of he expendiure disribuion. When expendiure and ne income are deflaed by he CPI, he larges gains over he sample period are experienced by he op quinile and he gains for he average household are larger han for he median household and he boom quinile. Afer correcing for CPI bias, he larges gains are now experienced by he boom quinile, followed by he median household, he average and he op quinile (bu growh is higher for all groups). Table 6 repors Gini coefficiens based on CPI-deflaed expendiures and he bias correced real expendiures (in 1996 reais). The bias correcion indicaes a subsanial decline in he Gini coefficien from 1987/88 o 1995/96, bu no from 1995/96 o 2002/03. Much of he improvemens in inequaliy may sem from he inflaion sabilizaion ha ook place in 1994. Since he poor had more limied access o shor-erm financial insrumens, hey were disproporionaely burdened by he inflaionary ax. Our finding ha CPI bias was greaer for he poor beween 1987/88 and 1995/96 is consisen wih he view ha he poor were he ones wih more o gain from he reducion in inflaion. The decline in he relaive prices of radable goods as shown in Figure 5 also has conribued o he narrowing in real expendiure inequaliy since he poor consume relaively more of ha class of goods. V. EVIDENCE FROM DURABLE GOODS OWNERSHIP AND ANTHROPOMETRICS Table 7 presens he percenage of households ha own differen durable goods using daa from he POF. I suggess a subsanial improvemen in he living condiions of Brazilian households. For example, while only 29 percen of households in he POF sample owned a washing machine in 1987/88, 53 percen did by 2002/03. During his same period, he ownership of color TVs increased from 57.4 percen o 93.4 percen. The number of 18 Bias esimaes are similar when he sample is consrained o households whose oal income is beween 1 and 40 imes he minimum wage, he arge populaion for which he IPCA index and he relaive prices used in our regressions are calculaed. Such consrain excludes from he regression sample 2.7 percen, 2.7 percen and 4.7 percen of he weighed households respecively in he 1987/88, 1995/96, and 2002/03 original samples due o oal income lower han 1 minimum wage, and 6.6 percen, 5.3 percen and 5.5 percen, respecively, because household oal income is greaer han 40 minimum wages.
16 households ha owns eiher a car or a moorcycle increased from 34.2 percen o 38.1 percen. A firs, one may be inclined o dismiss his ype of evidence as being driven mainly by declines in he price for hese goods and no necessarily by income growh. However, Figure 9 shows ha he increase in he ownership of durable goods was skewed oward hose goods ha are more of a luxury. Figure 9A shows he change in average holdings of a given good beween 1988 and 2002 and he sensiiviy o income of holdings of ha good in he 2002/03 cross-secion. Tha sensiiviy is obained by regressing he household s holdings of ha good on he log of oal expendiure (insrumened by he log of income) and he same conrols as in he previous regression. Figure 9B is similar o 9A bu based on a dummy for ownership of a leas one uni of ha durable good. Boh indicae a disproporionae increase in he demand for goods ha are more of a luxury, as expeced following he large increase in real income implied by our esimaes. The implied increase in real income (conrolling for relaive price changes) ha would explain he change in durable goods holdings and ownership shown in Table 7 and Figure 9 is even higher han he one based on Engel-curves for food consumpion (alhough he esimaes are no saisically significanly differen from each oher). Table 8 presens he evoluion of anhropomeric measures of he heigh-for-age of Brazilian children under 5 years of age. Naional anhropomeric figures are available for 1975, 1989 and 1996. While heigh has a geneic componen, i is also driven by he qualiy of nuriion, which becomes an increasingly imporan deerminan he furher down a household is in he income disribuion. As able 8 indicaes, here has been a marked decline in he share of children 2 sandard deviaions or more below he inernaional reference median, which implies a subsanial improvemen in he nuriion of he poorer households. Table 8 also presens he real minimum wage. As in he case of food demand and durable good ownership, he anhropomeric improvemens conradic he sagnan real income daa. Unforunaely, he recen POFs did no have anhropomeric daa for children. 19 I would be ineresing o compue he elasiciy of hese anhropomeric measures wih respec o measured real income, so as o consruc a quaniaive esimae of he implied rue real income growh (similarly o our analysis for durable goods). This is a very ineresing exension once such daa becomes available. VI. CONCLUSION This paper uses household-level daa o presen evidence ha headline real household income growh figures in Brazil grossly underesimae he growh in real income. Our esimaes indicae ha aggregae per capia income during ha period grew by 4½ percen per year. These figures are subsanially higher han he 1½ percen annual headline growh obained by deflaing nominal per capia household income by he CPI. I is possible ha similar biases occurred in oher Lain American counries, and ha such mismeasuremen has misguided much of he policy discussion on he effeciveness of markeoriened reforms. 19 The 2002/03 POF did have anhropomeric daa, bu only for people 20 years or older.
17 Our findings also indicae a subsanial improvemen in he disribuion of income when measured in real erms, paricularly beween 1987/88 and 1995/96. The real per capia income growh for he ypical household was even higher han he aggregae household income growh during ha period. This improvemen is parly due o he decline in he inflaion ax, which likely burdened disproporionaely he poor. This paper focuses on he mismeasuremen of households real income, bu similar sources of bias apply for he measuremen of producion price deflaors (e.g., new and beer goods). I is unlikely ha he GDP deflaor was accuraely measured if he CPI bias was large, paricularly because household consumpion, which is deflaed by he CPI, is a large componen of GDP. 20 One should no assume ha measuremen problems associaed wih he CPI a presen are similar o he large bias esimaed for 1987 2002. Relaive price changes (which amplify subsiuion bias) were unusually srong in he early 1990s. Qualiy improvemens, a furher source of bias, were likely concenraed in he years immediaely afer ariff reducions. The pooled regression esimaes sugges ha he bias has declined over ime, and much of he esimaed bias beween he 1995/96 and 2002/03 surveys may sem from he period prior o he 1999 revision of he CPI baske. A new revision will be made in July 2006 (based on he 2002/03 POF), which should furher improve he measuremen precision. A quanificaion of he size of he CPI bias in he more recen period wih he mehod used in his paper will only be possible when he resuls of a new POF become available for comparison wih he 2002/03 expendiure paerns. 20 The naional accouns sysem compues GDP from he producion side and from he expendiure side, reconciling he wo measures. Household consumpion accouns for abou 55 percen of he expendiure GDP.
18 Table 1. Mean Expendiure Shares, Median Toal Expendiure, by Expendiure Quinile in 1987/88 (In percen, excep where oherwise indicaed) Expendiure Quiniles Expendiure Shares 1 2 3 4 5 Food 41.4 37.3 31.2 25.3 16.0 Housing 18.0 16.2 16.7 17.2 18.9 Clohing 8.5 10.9 12.1 11.6 10.4 Transpor, including own vehicle 7.9 8.8 10.5 15.0 21.9 Healh 5.9 5.0 5.5 6.1 6.2 Elecrical appliances and furniure 5.7 7.4 8.0 7.1 7.2 Educaion 1.4 2.0 2.3 2.9 3.4 Hygiene 1.7 2.0 2.0 1.9 1.4 Recreaion 6.6 6.5 6.8 6.9 6.2 Oher 2.8 3.9.7 6.1 8.4 Toal expendiure (in 1996 reais) 5,251 11,419 18,868 32,166 94,370.
19 Table 2. Descripive Saisics for Regression Sample Complian sample Tenans 1987/88 1995/96 2002/03 1987/88 1995/96 2002/03 Share of food 0.306 [0.162] 0.264 [0.172] 0.233 [0.157] 0.279 [0.14] 0.212 [0.138] 0.207 [0.131] Share of food a home 0.251 [0.157] 0.218 [0.166] 0.185 [0.15] 0.223 [0.135] 0.165 [0.127] 0.153 [0.123] Share of food a resauran 0.054 [0.067] 0.046 [0.069] 0.048 [0.068] 0.056 [0.069] 0.046 [0.066] 0.053 [0.067] Relaive price of food 100 [0] 74.39 [5.1] 72.66 [5.73] 100 [0] 74.5 [4.88] 72.64 [5.46] Ln (Relaive price of food) 4.6 [0] 4.3 [0.06] 4.28 [0.07] 4.6 [0] 4.3 [0.06] 4.28 [0.07] Expendiure on food, in 1996 R$ 2902 [2440] 2909 [3331] 2742 [2838] 2705 [2261] 2880 [2999] 2700 [2653] Per capia expendiure on food, in 1996 R$ 816 [846] 914 [1182] 940 [1117] 835 [985] 989 [1236] 1051 [1279] Expendiure, in 1996 R$ 17410 [25363] 15897 [23704] 16841 [24244] 16710 [26398] 17056 [21241] 15258 [16203] Per capia expendiure, in 1996 R$ 5002 [7585] 5218 [9982] 5984 [10713] 5317 [8287] 5913 [8150] 6202 [8811] Ln (Expendiure/CPI), in 1996 R$ 9.241 [1.012] 9.113 [1.041] 9.196 [1.009] 9.252 [0.911] 9.317 [0.885] 9.275 [0.824] Ln (Afer-ax income/cpi), in 1996 R$ 9.291 [1.037] 9.07 [1.157] 9.099 [1.234] 9.273 [0.975] 9.143 [1.033] 9.137 [1.164] Ln (Household size) 1.280 1.200 1.123 1.204 1.129 1.005 Number ages 0 o 4 0.439 0.339 0.287 0.512 0.377 0.300 Number ages 5 o 9 0.466 0.355 0.307 0.430 0.314 0.250 Number ages 10 o 14 0.396 0.388 0.319 0.303 0.315 0.241 Number ages 15 o 19 0.392 0.377 0.347 0.302 0.322 0.292 Number ages 20 and up 2.396 2.280 2.231 2.250 2.149 2.035 Male head 0.791 0.752 0.681 0.765 0.736 0.661 Spouse presen 0.739 0.697 0.656 0.702 0.662 0.594 Head of household has some income from work 0.815 0.766 0.753 0.888 0.864 0.839 Spouse has some income from work 0.284 0.286 0.292 0.318 0.300 0.298 Head and spouse have income from work 0.262 0.259 0.256 0.300 0.284 0.275 Renal uni 0.301 0.193 0.179 Owner occupied 0.578 0.695 0.715 Oher living arrangemen (ceded housing) 0.121 0.112 0.105 Sample size 12290 13955 6436 3385 2298 1070
20 Table 3. Regression Resuls for Pooled Sample (Dependen variable = expendiure share of food in 1987, 1996, and 2003) (1) (2) (3) (4) Tenan -OLS Tenan -IV Complian-OLS Complian-IV Dummy for 1996-0.049 [0.016] -0.047 [0.017] -0.043 [0.012] -0.039 [0.013] Dummy for 2003-0.05 [0.019] -0.047 [0.020] -0.064 [0.013] -0.057 [0.013] Ln(Relaive price of food) 0.031 [0.050] 0.035 [0.054] 0.022 [0.038] 0.041 [0.038] Ln (Expendiure/CPI) -0.056 [0.005] -0.076 [0.005] -0.072 [0.003] -0.093 [0.004] Ln(Household size) 0.043 [0.019] 0.05 [0.020] 0.014 [0.007] 0.022 [0.008] Number ages 0 o 4 0.004 [0.006] 0 [0.006] 0.012 [0.003] 0.007 [0.003] Number ages 5 o 9 0.006 [0.005] 0.004 [0.005] 0.012 [0.003] 0.009 [0.003] Number ages 10 o 14 0.003 [0.006] 0.001 [0.007] 0.011 [0.003] 0.008 [0.003] Number ages 15 o 19-0.001 [0.006] -0.002 [0.006] 0.009 [0.002] 0.008 [0.002] Number ages 20 and up -0.001 [0.005] 0.003 [0.006] 0.006 [0.002] 0.009 [0.002] Male head 0.035 [0.007] 0.039 [0.007] 0.032 [0.005] 0.033 [0.006] Spouse presen -0.022 [0.007] -0.026 [0.007] -0.014 [0.005] -0.014 [0.005] Head of household has some income from work 0.002 [0.006] 0.006 [0.007] -0.009 [0.003] -0.004 [0.003] Spouse has some income from work 0.011 [0.013] 0.014 [0.013] -0.019 [0.006] -0.016 [0.006] Head and spouse have income from work -0.018 [0.013] -0.016 [0.012] 0.014 [0.007] 0.017 [0.007] Housing uni ceded by family, employer 0.051 [0.006] 0.041 [0.005] Owner occupied uni 0.024 [0.003] 0.023 [0.003] Consan 0.62 [0.208] 0.775 [0.223] 0.853 [0.166] 0.93 [0.166] Observaions 6753 6753 32681 32681 R-squared 0.224 0.21 0.295 0.283 Cumulaive bias 87-96 (%) 58.82 [13.46] 45.729 [13.448] 44.405 [9.463] 34.039 [9.353] Annual equivalen 87-96 (%) 9.82 6.87 6.61 4.73 Cumulaive bias 96-03 (%) 1.081 [14.803] 0.398 [11.635] 25.298 [5.653] 18.07 [4.405] Annual equivalen 96-03 (%) 0.16 0.06 4.23 2.91 Mean dependen variable: 0.237 0.237 0.264 0.264 Noe: Robus sandard errors in brackes. Sample excludes observaions wih no expendiure records in he collecive expendiures noebook. Conrols also include regional dummies. Toal income is used as an insrumen o oal expendiure in he IV regressions. Cumulaive bias repored corresponds o E G,. The implied gross change in he rue cos of living is (1- E G, ) imes he gross change in he CPI, and he resuling gross rue real income growh is 1/(1- E G, ) imes he gross real income growh obained by deflaing nominal income by he CPI.
21 Table 4. Bias Esimaes Across Differen Mehods and Samples Parameric Esimaes Bias Consan Across Households OLS, Full Sample 9.54 4.89 [7.2 11.7] [3.7 6.1] OLS, Complian 6.61 4.23 [4.6 8.6] [3.2 5.3] OLS, Tenan 9.82 0.16 [4.6 14.9] [-2.5 3.1] IV, Full Sample 6.40 3.02 [4.6 8.0] [2.1 4.0] IV, Complian 4.73 2.91 [3.0 6.4] [2.1 3.7] IV, Tenan 6.87 0.06 [2.7 10.7] [-1.9 2.2] Bias Linear Funcion of Real Expendiure Populaion Weighed Expendiure Weighed 87 96 96 03 87 96 96 03 OLS, Complian 5.58 4.69 1.96 3.93 [4.1 7.1] [3.6 5.7] [0.4 3.8] [2.6 5.4] IV, Complian 4.52 3.21 2.43 3.94 [3.2 6.0] [2.4 4.0] [0.9 4.2] [2.6 5.3] Semi-Parameric Esimaes Tenan 3.84-0.46-0.20-0.48 [-1.4 8.1] [-5.2 4.2] [-3.7 4.7] [-4.2 3.3] Full Sample 6.14 4.84 2.83 4.32 [4.3 8.2] [2.9 6.3] [0.9 5.3] [2.4 6.2] Complian 5.68 4.59 2.17 4.33 [4.3 7.9] [3.4 6.2] [0.5 4.9] [2.8 6.5] Winsorized 4.94 3.58 3.04 2.83 [3.4 6.6] [2.3 5.1] [1.7 4.7] [1.9 3.8] Noe: 90 percen confidence inerval in square brackes. Full sample sands for all he meropolian areas for which CPI is available for he hree surveys. Complian sample corresponds o hose observaions ha urned heir collecive expendiure noebook wih a leas one expendiure recorded. Tenan sample corresponds o he complian households ha ren heir dwellings. Winsorized sample corresponds o he full sample households where he value of food and oal expendiures for observaions below he 5 h and above he 95 h percenile are se o he value a ha percenile.
22 Table 5. Household Per Capia Expendiure and Ne Income: Headline and Correced (in 1996 R$) Annual percen 1987-88 1996-95 2002-03 change Using official CPI as he deflaor Household per capia expendiure Mean 5,003 5,219 5,985 1.2 Median 2,748 2,646 2,976 0.5 Boom 20% 790 701 850 0.5 Top 20% 14,928 16,463 18,874 1.5 Household per capia ne income Mean 5,461 4,935 6,610 1.3 Median 2,912 2,569 2,821-0.2 Boom 20% 1,145 1,151 1,152 0.0 Top 20% 15,379 13,560 20,325 1.8 Correcing for esimaed CPI bias Household per capia expendiure Mean 4,067 5,219 7,914 4.4 Median 1,723 2,646 3,942 5.5 Boom 20% 335 701 1,143 8.3 Top 20% 14,604 16,463 24,940 3.6 Household per capia ne income Mean 4,125 4,935 8,359 4.7 Median 1,773 2,569 3,711 4.9 Boom 20% 462 1,151 1,548 8.2 Top 20% 14,500 13,560 26,219 3.9 Noe: Based on esimaes of he semi-parameric specificaion in he complian sample ha reurned heir collecive expendiure noebook wih a leas one expendiure recorded. The boom and op 20 percen refer o quiniles of expendiure per survey year in his complian sample.
23 Table 6. Expendiure Inequaliy Correced for CPI Bias: Expendiure Gini Coefficiens Panel I: Gini coefficiens based on CPI deflaed expendiures 1987/88 0.533 1995/96 0.550 2002/03 0.542 Panel II: Gini coefficiens based on expendiures correcing for he CPI bias 1987/88 0.533 1995/96 0.414 2002/03 0.398 Noe: Based on semi-parameric bias esimaes from he complian sample ha reurned heir collecive expendiure noebook wih a leas one expendiure recorded..
24 Table 7. Ownership of Durable Goods and Sensiiviy o Income Owns a leas one (percen) Average number of unis/household Regression coefficien income sensiiviy 1987/88 1995/96 2002/03 1987/88 1995/96 2002/03 A leas one Number unis TV 89.0 94.6 95.2 1.22 1.47 1.54 0.05 [0.003] 0.514 [0.027] Color TV 57.4 83.2 93.4 0.70 1.20 1.48 0.07 [0.004] 0.529 [0.027] Fan 52.3 64.5 67.9 0.72 0.99 1.14 0.061 [0.008] 0.263 [0.027] Sove 98.7 99.3 99.3 1.04 1.04 1.03 0.003 [0.001] 0.007 [0.003] Iron 90.2 92.0 91.3 1.07 1.08 1.04 0.078 [0.005] 0.173 [0.028] Refrigeraor 87.6 91.1 94.4 0.92 0.95 0.98 0.062 [0.004] 0.087 [0.005] Blender 83.5 84.7 85.4 0.89 0.88 0.90 0.08 [0.006] 0.091 [0.02] Sound sysem 45.8 64.1 66.8 0.49 0.72 0.76 0.135 [0.008] 0.228 [0.01] Bicycle 30.2 40.7 39.6 0.43 0.60 0.58 0.023 [0.008] 0.067 [0.014] Washing machine 29.3 46.8 52.7 0.30 0.48 0.53 0.284 [0.008] 0.291 [0.008] VCR n.a. 37.9 47.9 n.a. 0.41 0.52 0.26 [0.008] 0.325 [0.009] Car or moorcycle 34.2 36.9 38.1 0.46 0.48 0.49 0.292 [0.007] 0.439 [0.01] Car 33.0 35.6 36.1 0.42 0.45 0.45 0.294 [0.007] 0.422 [0.01] Cake mixer 34.8 39.1 42.1 0.35 0.40 0.43 0.215 [0.008] 0.217 [0.008] Hair dryer 39.6 36.1 31.0 0.48 0.42 0.35 0.224 [0.007] 0.289 [0.009] Waer filer n.a. n.a. 36.3 n.a. n.a. 0.37-0.008 [0.008] -0.006 [0.009] Microwave oven n.a. 16.0 30.1 n.a. 0.16 0.30 0.263 [0.007] 0.266 [0.007] Sewing machine 47.2 35.4 24.4 0.50 0.39 0.29 0.071 [0.007] 0.062 [0.021] Desk radio 39.5 36.0 20.8 0.46 0.44 0.25 0.049 [0.007] 0.099 [0.01] Porable radio 38.0 28.4 20.7 0.47 0.36 0.25 0.066 [0.007] 0.11 [0.01] Personal compuer n.a. 7.0 22.2 n.a. 0.07 0.25 0.258 [0.006] 0.299 [0.007] Freezer 6.9 18.8 18.8 0.07 0.19 0.19 0.131 [0.006] 0.137 [0.007] Vacuum cleaner 22.4 21.7 19.0 0.23 0.22 0.19 0.22 [0.006] 0.226 [0.006] Air condiioner 6.4 8.6 11.0 0.10 0.13 0.16 0.103 [0.005] 0.196 [0.009] Toaser 9.6 9.2 12.4 0.10 0.09 0.13 0.131 [0.005] 0.135 [0.005] Tape recorder 19.4 17.7 10.3 0.22 0.20 0.11 0.007 [0.005] 0.014 [0.006] Floor waxer 41.4 23.0 9.9 0.42 0.23 0.10 0.059 [0.005] 0.061 [0.005] Ozonizer filer n.a. 6.3 9.3 n.a. 0.06 0.09 0.088 [0.005] 0.088 [0.005] Dryer 4.0 9.7 6.9 0.04 0.10 0.07 0.065 [0.004] 0.065 [0.004] CD player n.a. 7.8 7.2 n.a. 0.09 0.08 0.047 [0.004] 0.058 [0.005] Dish anenna n.a. 1.7 4.5 n.a. 0.02 0.05 0.015 [0.003] 0.015 [0.003] Black and whie TV 57.4 25.6 5.9 0.70 0.27 0.06-0.015 [0.004] -0.016 [0.004] DVD player n.a. n.a. 6.6 n.a. n.a. 0.07 0.103 [0.004] 0.108 [0.004] Dishwasher 2.0 6.6 5.8 0.02 0.07 0.06 0.084 [0.004] 0.085 [0.004] Moorcycle 3.3 2.8 3.8 0.04 0.03 0.04 0.014 [0.003] 0.016 [0.003] LP player 12.8 5.1 n.a. 0.13 0.05 n.a. n.a. n.a. Noes: Based on he full sample. Regression coefficiens are he sensiiviy o log of oal expendiure for he dummy for owning a leas one uni of he durable in quesion, and for he number of unis of ha durable. Coefficien was esimaed using he 2002/03 POF, wih log of income as an insrumen for log of oal expendiure, conrolling for relaive prices and for he household characerisics used as conrols in he regressions shown in Tables 3 and 4. Sandard errors in brackes. The abbreviaion n.a. indicaes he relevan daa were no available.
25 Table 8. Anhropomeric Measures for Children 0 60 Monhs Old Percenage Below 3 Sd. Dev. from U.S. Median Heigh for Age Percenage Below 2 Sd. Dev. from U.S. Median Real Minimum Wage (in 2006 R$) Year Sample 1975 Naional 14.2 32.0 310.78 Urban 10.0 25.9 1989 Naional 4.2 15.4 238.48 Urban 3.0 12.3 1996 Naional 2.5 10.5 212.68 Urban 1.6 7.8 Noe: Anhropomeric Daa from he World Healh Organizaion Global Daabase on Child Growh and Malnuriion. Real minimum wage daa from IPEADATA.
26 Figure 1. Brazil: GDP Per Capia and Average Growh In Decade (Consan 2004 reais) 2004 R$ 12000 10000 8000 6000 4000 2000 0 7 6 5 4 3 2 1 0-1 1950 1955 % 1960 1965 1970 1975 1980 1985 1990 1995 2000 Year Source: Insiuo de Pesquisa Econômica Aplicada (IPEA). Noe: Solid line indicaes level while columns indicae average growh rae in decade. Figure 2. Food Budge Share by Expendiure Deciles 1987/88 1995/96 2002/03 45 40 35 30 % 25 20 15 10 5 0 1 2 3 4 5 6 7 8 9 10 Sources: IBGE; POF; and auhors calculaions.
27 Figure 3A. Food Budge Share by Meropolian Area 1987/88 1995/96 2002/03 45 40 35 30 25 % 20 15 10 5 0 Belém Foraleza Recife Salvador Belo Horizone Rio de Janeiro São Paulo Curiiba Poro Alegre Disrio Federal Toal Figure 3B. Food Expendiure Per Capia, by Meropolian Area (in 1996 reais) 1987/88 1995/96 2002/03 1100 1000 R$ of 1996 900 800 700 600 500 Belém Foraleza Recife Salvador Belo Horizone Rio de Janeiro São Paulo Curiiba Poro Alegre Disrio Federal Toal Sources: IBGE; POF; and auhors calculaions.
28 Figure 4. Changes in Relaive Prices 250 200 Food Household supplies Transpor and communicaion Personal expenses Housing Clohing Healh and personal care 150 100 50 0 1980 1981 1982 1983 1984 1985 1986 1987 1988 1989 1990 1991 1992 1993 1994 1995 1996 1997 1998 1999 2000 2001 2002 2003 2004 2005 Sources: IBGE; and auhors calculaions. Noe: Relaive price defined as he raio of he price level for he group o he level of he overall index.
29 Figure 5. Non-Parameric Esimaes of Relaionship Beween Food Shares and Household Expendiure Food share, by expendiure level 0.1.2.3.4.5 6 8 10 12 Ln(Expendiure/CPI) 1987/88 1995/96 2002/03 Noes: Esimaes based on he full sample. Curves obained from locally weighed linear regressions using quaric kernel weighs.
30 Figure 6. Esimaed Bias in 1987/88-1995/96 as a Funcion of CPI-Measured Real Expendiure in 1995/96 Bias in percen per year -10 0 10 20 30 Annual bias 87/88-95/96, by expendiure level 90 percen boosrap confidence inerval 6 8 10 12 Ln(Expendiure/CPI) 0.005.01.015.02.025 P.d.f. 95-96 Figure 7. Esimaed Bias in 1995/96 2002/03 as a Funcion of CPI-Measured Real Expendiure in 2002/03 Bias in percen per year -10 0 10 20 30 Annual bias 95/96-02/03, by expendiure level 90 percen boosrap confidence inerval 6 8 10 12 Ln(Expendiure/CPI) 0.005.01.015.02.025 P.d.f. 02-03 Noes: Bias esimaes based on he shif of semi-paramerically esimaed Engel curves for he full sample (using he same household characerisics and relaive price conrols as he regression in Table 3).
31 Figure 8A. Disribuion of Expendiure Deflaed by he CPI: 87/88, 95/96 and 02/03 Disribuion of Ln(Expendiure/CPI) 0.2.4.6.8 6 8 10 12 Ln(Expendiure/CPI) PDF 87/88 PDF 95/96 PDF 02/03 Figure 8B. Disribuion of Expendiure Deflaed by he Esimaed True Cos of Living Index: 87/88, 95/96 and 02/03 Disribuion of Ln(Correced Real Expendiure) 0.2.4.6.8 6 8 10 12 Ln(Expendiure/Esimaed True Cos of Living Index) PDF 87/88 PDF 02/03, Correced PDF 95/96, Correced Noes: Bias correced expendiures based on semi-parameric esimaes for he complian sample.
32 Figure 9. Changes in Durable Goods Holdings and Sensiiviy o Income Panel A. Average Holdings -1 -.5 0.5 1 Fan SoundSysem Washer Bike Freezer Microwave PC Dishwasher AirCon CakeMixer VCR DishAnenna CDPlayer DryerRefrigeraor Sove Moorcycle Ozonizer Blender Toaser Iron Vacuum TapeRecorder Hair Dryer SewingMachine DeskRadio PorableRadio Waxer BW-TV Car ColorTV 0.1.2.3.4.5 Sensiiviy o income Chg avg holdings 1988-02 Linear predicion Panel B. Average Dummy for Ownership -.6 -.4 -.2 0.2.4 ColorTV Fan Bike Refrigeraor DishAnenna Dryer AirCon Blender Dishwasher Sove CDPlayer Iron Ozonizer Moorcycle TapeRecorder DeskRadio PorableRadio SewingMachine Waxer BW-TV SoundSysem Freezer Toaser CakeMixer Vacuum Hair Dryer Washer PCMicrowave VCR Car 0.1.2.3 Sensiiviy o income Change in % own 1988-02 Linear predicion Noe: Esimaes based on he full sample. Sensiiviy o income defined as he coefficien obained when average ownership or ownership dummy is regressed on log of oal expendiure (insrumened by he log of income). The regression also includes conrols for relaive prices and for he household characerisics described in Tables 2 and 3.
33 REFERENCES Asano, Seki, and Eduardo P. S. Fiuza, 2001, Esimaion of he Brazilian Consumer Demand Sysem IPEA Texos para Discussão No. 793 (Rio de Janeiro: Insiuo de Pesquisa Econômica Aplicada). Beay, Timohy, and Erling R. Larsen, 2005, Using Engel Curves o Esimae Bias in he Canadian CPI as a Cos of Living Index, Canadian Journal of Economics, Vol. 38, No. 2, pp. 482 99. Bils, Mark, and Peer Klenow, 2001, Quanifying Qualiy Growh, American Economic Review, Vol. 91, No. 4, pp. 1006 30. Boskin, Michael, Ellen Dulberger, Rober Gordon, Zvi Griliches, and Dale Jorgenson, 1996, Toward a More Accurae Measure of he Cos of Living, Final Repor o he U.S. Senae Finance Commiee (December). Cosa, Dora, 2001, Esimaing Real Income in he Unied Saes from 1888 o 1994: Correcing CPI Bias Using Engel Curves, Journal of Poliical Economy, Vol. 109, No. 6, pp. 1288 1310. De Carvalho Filho, Irineu, 2004, Old-Age Benefis and Reiremen Decisions of Rural Elderly in Brazil, unpublished (Washingon: Inernaional Moneary Fund). Deaon, Angus, and Valerie Kozel, 2005, Daa and Dogma: The Grea Indian Povery Debae, The World Bank Research Observer Vol. 20, pp.177 99. Deaon, Angus, and John Muellbauer, 1980, An Almos Ideal Demand Sysem, American Economic Review, Vol. 70, No. 3, pp. 312 26. Easerly. W., Norman Loayza, and Peer Moniel, 1997, Has Lain America s Pos-Reform Growh Been Disappoining? Journal of Inernaional Economics, Vol. 43, pp. 287 311. Ferreira, Pedro, and José Rossi, 2003, New Evidence from Brazil on Trade Liberalizaion and Produciviy Growh, Inernaional Economic Review, Vol. 44, pp. 1383 1405. Goñi, Edwin, Humbero López and Luis Servén, 2006, Geing Real Abou Inequaliy: Evidence from Brazil, Colombia, Mexico, and Peru, World Bank Policy Research Working Paper No. 3815, January (Washingon: The World Bank Group). Hamilon, Bruce, 2001, Using Engel s Law o Esimae CPI Bias, American Economic Review, Vol. 91, No. 3, pp. 619 30. Hausman, Jerry, 2003, Sources of Bias and Soluions o Bias in he Consumer Price Index, Journal of Economic Perspecives, Vol. 17, No. 1, pp. 23 44.
34 Houhakker, Hendrik, 1987, Engel s Law, in John Eawell, Murray Milgae, and Peer Newman, eds., The new Palgrave: A dicionary of economics, Vol. 2 (London: Macmillan), pp. 143 44. Krugman, Paul, 1995, Duch Tulips and Emerging Markes: Anoher Bubble Burss, Foreign Affairs, Vol. 74, No. 4, pp. 28 44. Kume, Honorio, Guida Piani and Carlos de Souza, 2000, A Políica Brasileira de Imporação no Período 1987-98: Descrição e Avaliação, unpublished (Rio de Janeiro: Insiuo de Pesquisa Econômica Aplicada). Muendler, Marc-Andreas, 2001, Trade, Technology, and Produciviy: A Sudy of Brazilian Manufacurers, 1986 1998, unpublished (Berkeley: Universiy of California). Nakamura, Leonard, 1997, Is he U.S. Economy Really Growing Too Slowly? Maybe We re Measuring Growh Wrong, Federal Reserve Bank of Philadelphia Business Review (March April), pp. 3 14. Naional Research Council (2002). A Wha Price? Concepualizing and Measuring Cos-of- Living and Price Indices, Charles L. Schulze and Chrisopher D. Mackie, eds. Neri, Marcelo (1995): "Sobre a Mensuração dos Salários Reais em Ala Inflação", Pesquisa e Planejameno Econômico, 25, no. 3, pp.497 525. Sachs, Jeffrey D. and Andrew Warner, 1995, Economic Reform and he Process of Global Inegraion, Brookings Papers on Economic Aciviy, 1995, (1), pp. 1 95. Thomas, Duncan, 1986, The Food Share as a Welfare Measure, Ph.D. Disseraion, Princeon Universiy, mimeo. Yachew, A., 1997, An Elemenary Esimaor of he Parial Linear Model," Economics Leers, Elsevier, vol. 57 (2), pages 135 43 (December).