An African demand system? Welfare analysis using data from the International Comparison Program for Africa

Transcription

1 An Afrcan demand system? Welfare analyss usng data from the Internatonal Comparson Program for Afrca Abebe Shmeles Economc Development Research Department Revsed May 2010 Abstract: Ths paper uses data from the Internatonal Comparson Program 2005 to recover complete own and crossprce and ncome elastcty estmates for the Afrcan contnent usng the Extended Lnear Expendture System for 12 broadly defned commodtes. The results can be used for aggregate welfare comparson n such global models as GTAP (Global Trade Analyss Project) and exercses to nfer welfare mpact of prce shocks at the contnental level. In a heurstc way also, t s possble to derve a utltyconsstent global poverty lne from the demand functon that could be compared wth the popular nternatonal poverty lnes. 1

2 1. Introducton The Internatonal Comparson Program (ICP) was ntroduced n 1968 by the UN Statstcal Commsson, and housed ntally at the Unversty of Pennsylvana, to establsh a system of nternatonal comparson of natonal account aggregates free from dfferences n prce levels across countres 1. The Data collecton started n 1970 wth 10 countres and ths ncreased to 197 countres n The number of Afrcan countres covered n ths survey durng ths perod ncreased from 1 to 48. Snce 1985, the ICP has been managed by the ICP Global Offce and s housed at the World Bank 2. The Purchasng Power Party (PPP), the key output of the ICP s the most frequently used converter of natonal ncome statstcs nto an nternatonally comparable unts for decades. Most mportantly, global poverty measures are based on mean household per capta ncome or consumpton obtaned from natonal surveys expressed n natonal currences and then converted nto PPPs. The ICP 2005 was conceved manly to collect prce data on more than 1000 commodtes across more than 100 countres to provde a bass for nternatonal comparson of purchasng power so that global, regonal and subregonal poverty aggregates are measured consstently. The World Bank perodcally updated poverty estmates for the developng world by combnng basc data from household surveys wth PPP 3 that are the bass of regonal poverty fgures reported globally (Chen and Ravallon,2008, 2007, 2004). In a smlar ven, emprcal work on economc growth routnely uses ncomes expressed n PPPs to undertake crosscountry comparsons such as the popular SummersHeston data set (or PennTables as they are popularly called) whch n prncple provde summary aggregates of natonal accounts free of dfferences n prce levels across countres. 4 1 Ahmad (2006) 2 Detals on ICP are found at gepk: ~ppk: ~thestepk:270065,00.html 3 See 4 See for nstance Heston and Summers (1996) and Summers et al (1991, 1988) for detaled dscusson of the SummerHeston data sets. Useful crtques of ths data are also found n Knowles (2001), Dowrck (2005) and Dowrck and Quggn (1997) 2

3 In ths paper, we extend the applcaton of ICP data to global welfare analyss of a representatve household n the Afrca regon by lookng at changes n demand for broad categores of consumpton tems n response to changes n prces and ncome. Ths s allowed by the fact that the ICP2005 data reports consumpton expendture for 12 broad commodty groupngs along wth ther relatve prces for 48 countres covered n the study. Certanly, these expendture tems are n prncple comparable and aggregaton s allowed by defnton keepng n mnd the basc assumpton used n the collecton of the prce and expendture data. Thus, a comparson of welfare changes followng prce or ncome movements can be nferred by lookng at the concentraton curve for dfferent commodty groupngs takng note of the fact that our observatonal unts are countres, not ndvduals, whch lmts the conventonal nterpretatons. One possble way of lookng at the country level nformaton s to thnk of polcy dalogues focused on regonal ssues, such as debt relef, development ad, trade lberalzaton and other ssues such as MDGs that requre the level of aggregaton mpled by our data. Our computatons ndcate that prce shocks that affect for nstance food tems may have the largest welfare loss at the contnental level than say shocks that lead to proportonal declne n per capta ncomes or ncrease n transport cost for example through a rse n energy prces. Smlar analogy can also be made about the welfare mpact of global transfers allowed by proportonal prce declnes through trade lberalzaton or subsdes (lke food ad) or any other mechansm. Smlarly, transfers that favor household expendture on accessng educaton are superor to any other means of transfer n terms of mprovng global welfare snce nearly households n all countres spent proportonately the same amount on educaton than on any other commodty. As a natural extenson of the analyss based on concentraton curves, we specfed and estmated the Extended Lnear Expendture System (ELES) usng personal savngs to dentfy all the parameters necessary to estmate own and cross prce elastctes as well as ncome elastctes (Lluch, 1975, Howe, 1975), the result of whch can be valuable nput to global model analyss such as GTAP whch uses ELES to model household behavor. Our results from the ELES generally are ntutve and also support the nference we obtaned from the concentraton curves. Our estmates of ncome elastctes for such broad consumpton categores as food (0.56), water (0.9), clothng ( 0.69), health (0.74) and educaton (0.24) suggest these are necesstes whle for the rest such as alcohol (1.0), recreaton (1.3), transport (1.4) and communcatons (2.2) are luxures. These elastcty estmates are strkngly smlar wth those often obtaned from large household surveys for ndvdual countres. Snce the ELES allows for the estmaton of subsstence consumpton expendture, t s nterestng to examne whether the utltyconsstent measures of a poverty lne s algned wth the popular one dollar a day nternatonal poverty lne. We were able to estmate a 1.12 dollar a day subsstence consumpton whch s very close to the conventonal poverty lne of 1.08 dollars a day. In addton, the margnal utlty of ncome, sometmes known as the nverse Frsch parameter, 3

4 whch measures level of development, suggest a relatvely hgher rato of subsstence component of consumpton n total expendture ndcatng low level of development of the regon. The rest of the paper s organzed as follows. Secton 2 outlnes the methodology we used to recover prce and ncome responses. Secton 3 descrbes the data and Secton 4 provdes the results wth some dscusson. Secton 5 concludes the paper. 2. Analytcal framework 2.1. Concentraton curves Concentraton curves are generalzed forms of the popular summary measure known as the Lorenz curve. In many plannng exercses, and ssues of economc growth, the dstrbuton of expendture on varous goods across a spectrum of household characterstcs renders valuable nsghts to polcy optons 5. The concept of concentraton curves were early llustrated and rgorously dscussed by Roy, et al (1959); and later Kakwan (1980) provded proof of some of the emprcal propertes, and Ythak and Slemrod (1991) used them to analyze ssues of margnal tax reform n a revenueneutral settng. As defned by Ytzhak and Slemrod (1991: 481), "the concentraton curve s a dagram smlar to the Lorenz curve. On the horzontal axs, the households are ordered accordng to ther ncome, whle the vertcal axs descrbes the cumulatve percentage of the total expendture on specfc commodty that s spent by the famles whose ncomes are less than or equal to specfed ncome level". Ths defnton of a concentraton curve embodes the ncome effects; and Rao et al (1959) ntroduced relatve concentraton curves to normalze the effects of dfferences n purchasng power so that the effect of dfferences n preferences for varous commodtes can be neatly captured. Kakwan (1980) 6 proved mportant theorems pertanng to concentraton curves of whch the followng may be reproduced for the purpose of ths paper:. If the ncome elastcty of commodty, E s greater than the ncome elastcty of commodty j, then, the concentraton curve for les above the concentraton curve for j; 5 see also Haggablade and Younger (2003), and Younger et al, (1999) for the applcaton of concentraton curves on Afrcan data. Early attempt on Ethopa usng the 1980/81 household ncome and consumpton survey was made by Shmeles (1993) 6 Kakwan (1980), op ct, pp

5 . The concentraton curve for commodty wll be above (below) the egaltaran lne f, and only f E s less (or greater) than zero for all ncome level greater than zero.. The concentraton curve for commodty les above (below) the Lorenz curve f, and only f, E s less (greater) than unty for all ncome greater than zero. It follows therefore, that f the concentraton curve of a commodty les above the egaltaran, t s an nferor commodty, f the concentraton curve les between the Lorenz curve and the egaltaran lne, t s a necessary commodty, and f the concentraton curve les below the Lorenz curve, the commodty s luxury. Ytzhak and Slemrod (1991) made an nsghtful use of concentraton curves n the realm of publc economcs to analyze ssues of tax reform. It s rather becomng conventonal n the lterature to look nto the structure of ndrect tax systems, and the possblty of reform by maxmzng socalwelfare functon of the communty subject to a government revenue constrant 7. Ths approach presupposes the knowledge of Indrect Utlty Functon of the communty, and thus the respectve demand systems n order to be of any emprcal use. When one looks at the severe lmtatons that developng countres face to meet the data requrements of ths approach, then, the search for an alternatve method remans a very compellng one. In ths respect, the Margnal Condtonal Stochastc Domnance Rules (MCSD) developed by Ythak and Smlerod (1991) usng the concept underlyng concentraton curves can be consdered as a sgnfcant step to that end. MCSD s defned as a state where " f the (shfted) [due to tax ncdence] concentraton curve of one commodty s above the (shfted) concentraton curve of another commodty, then, the frst commodty domnates n the sense that a small tax decrease n the frst commodty accompaned by a tax ncrease n the second (wth revenue remanng unchanged) ncreases socal welfare functons. In other words, f and only f concentraton curves do not ntersect wll all addtve socalwelfare functons show that the tax change ncreases welfare. We refer to these rules as Margnal Condtonal Stochastc Domnance Rules" 8. Normally ths proposton would have requred the plottng of n(n1)/2 curves, whch for a suffcently large number of commodtes becomes cumbersome. The Gncoeffcent has 7 see Atknson (1970) for the specfcaton of a socalwelfare functon, Ahmad And Stern (1984), Kng (1983), Cragg (1991) for emprcal applcaton and Deaton (1979, 1981) for the mplcaton of addtve preferences to optmal commodty taxes. 8 Ytzhak and Smelord (1991), op ct, pp 482 5

6 been used to dentfy a class of easly computable necessary condtons for welfare domnance va the translaton nto ncome elastctes. Ths condton states that the ncome elastcty of commodty should be lower than that of commodty j n order for commodty to domnate commodty j n the event they are subject to an ndrect tax. We may show the above relatons explctly usng the concentraton rato or ndex concept, whch s defned as one half of the area below the 45 0 lne mnus the concentraton curve. That s, c Cov[ X, F( y)] m = (1) Where, c s onehalf of the concentraton rato, m s the mean expendture on commodty, X s total expendture on commodty, and F (y) s the cumulatve dstrbuton of ncome. Therefore, the area between the concentraton curve of commodty, and the concentraton curve of commodty j can be wrtten as: b b G (2) j c cj =[ ] S Sj y Where, b = Cov( X, F( y) Cov[ y, F( y)] S = m m y And m y stands for mean ncome or expendture. Here the revenue mplcaton of the polcy reform s assumed to be neutral that s there s no gan or loss to the government. We may nterpret b /S as the weghted average ncome elastctes of commodty, the weght beng here the Gncoeffcentmpled welfare functon, and s a nonparametrc estmator of the 6

7 slope of the regresson lne of S on y. 9 Thus for commodty to domnate commodty j the weghted ncome elastcty of commodty should be larger than for commodty j. The weghtng scheme employed here s the Gnndex whch also mples a specfc form of socalwelfare functon. In fact, we can further broaden the weghtng scheme by usng the noton of the extended Gn ndex whch s gven by: Cov[ y,1 F( y)] G( α ) = α m y α 1 where, G (α) s a parameter chosen by the nvestgator. The Gn s a specal case of G (α) where, α s 2. The hgher s α the greater s the emphass on the bottom of the ncome dstrbuton Demand systems and household welfare A related approach would be to construct a smple demand system for the commodtes of nterest to recover ncome and prce elastctes that could be used for a wde range of ssues that requre dscusson of household consumpton behavour. In our case, the utlty functon that gves rse to the Lnear Expendture System (LES) s of partcular attracton. Frst, we work on a hghly aggregated data set whch has lost substantal nformaton n the process so that nonlnearty n Engel curves or flexblty n prce responses cannot be captured easly from the data. Secondly, the lnear expendture system s popular specfcaton n most global macro and CGE models allowng estmated parameters to have some practcal relevance. Thrd, the full parameters of the LES can be recovered from crosssecton data f nformaton on personal savngs s avalable. Fnally, nterestng welfare measures such as margnal utlty of ncome and drect lnk wth Gn coeffcent ncrease the attractveness of the LES. The utlty functon underlyng the LES s the StoneGeary utlty functon whch s specfed as follows: = ln( ) (3) where the vectors x and γ represent respectvely, consumpton of the th commodty and a subsstence component. Maxmzaton of (3) wth the usual budget constrant yelds the popular LES gven by equaton (4): 9 see Ytzhak and Slemrod (1991), op ct, pp

8 p t x th = p γ + β ( y p γ ) (4) t ht k= 1 k k Where p t s prce of commodty prevalng at perod t, x t s quantty of demanded by household n country h at perod t, y ht s total ncome of a representatve household n country h at perod t and γ and β are parameters to be estmated, representng respectvely the subsstence consumpton of commodty, andβ s the margnal budget share. The structure of the LES s motvated by the assumpton that regardless of ncome levels, each household allocates ts ncome frst on the purchase of rreducble quantty of each commodty deemed necessary for subsstence and the remanng s drven by consumpton preference. Estmaton of (4) s complcated by the nonlnear term lnkng margnal budget share wth the supernumerary ncome or consumpton expendture so that a numercal approxmaton s used n the context of nonlnear system of equatons. When data s lmted only to a crosssecton, t s possble to recover all parameters of the LES by usng addtonal nformaton on ncome, such as savngs under a certan assumpton. We note also that by constructon and propertes of demand functon, the margnal budget shares addup to unty and the sums of the ntercepts of the regresson for equaton (4) should be zero. Wth these condtons, then, Lluch (1973) proposed the Extended LES (ELES) where personal savngs s ncluded n the consumpton basket wth the subsstence component set to zero 10. The ELES lnks ncome elastcty values wth prce elastctes through the margnal utlty of ncome so that the full Slutzky matrx s recovered from a crosssecton data. We use the followng relatons to do that: (5) = (1+ ) f =j {6) = (1+ ) f j 10 For recent applcatons see Shmeles & Deleglegn (2009) and Burney and Kamal (2008) 8

9 Where s ncome elastcty of demand for commodty, s the crossprce elastcty and s the nverse of the Frsch parameter and s gven by = ϒ. Often the Frsch parameter s nterpreted to ndcate the level of development as t measures the proporton of subsstence expendture n total consumpton expendture. The hgher s the value of ths parameter, the greater the mportance of subsstence expendture, thus the lower the level of development. Thus the ELES can be used also to estmate the utlty consstent poverty lne usng the total subsstence expendture mpled by the demand model for each commodty. Another nterestng feature of the LES s that t establshes a drect lnk between expendture shares and Gn coeffcents (Kakwan, 1980) to quantfy the extent to whch the rse n prce has mpacted on the overall Gn coeffcent. From ths exercse t would be possble to tell whether the nflatonary process s aganst the poor or t s ncome neutral or n certan cases based aganst the well off households. Despte the wellknown lmtatons, the LES provdes a smple framework to capture the welfare mplcatons of changes n relatve prces. For nstance, t s possble to establsh whether nequalty of ncome rses, falls or remans the same due to only changes n relatve prces. To do that we use the result n Kakwan (1980) that lnks Gn coeffcent between two prce settngs on the assumpton that real ncome among households s held constant: G t = ( = 1 p p * * p p γ + ( yt pγ) = 1 = 1 = 1 p * β y 0 G 0 β ) (7) Where Gt s Gn coeffcent at perod t wth prce vector P*, y t s mean consumpton expendture at perod t and y 0 s mean consumpton expendture n perod 0. Usng estmated coeffcents from (7), t s possble to compute the Gn coeffcent at the new set of prces and examne whether or not t leads to a worsenng state. The LES s less attractve to nvestgate prce responses though. Emprcal estmaton of equaton (4) usng crosssecton data proceeds wth the assumpton of contemporaneous error components for the systems of lnear equaton for each commodty gvng rse to Seemngly Unrelated Regresson Estmator (SURE). All parameters are dentfed wth personal savngs allowed nto the consumpton bundle where by assumpton subsstence consumpton of savng s set to zero. 9

10 3. Data and descrptve statstcs The ICP2005 data used n ths study covers 48 Afrcan countres (see Table 1 for the lst) for whch detaled prce data, household consumpton expendture for broadly defned categores were collected and estmated. Comparson between per capta consumpton computed usng PPP and offcal exchange rate by the Afrcan Development Bank AfDB (2009) ndcated sgnfcant dvergence, partcularly for poorer countres. The ICP2005 provded household consumpton expendture on 12 broad categores of consumpton goods whch we used to construct concentraton curves and estmate the parameters of the ELES. These are Food & Nonalcoholc drnks, Alcoholc drnk, Clothng, Water and Electrcty, Household utensls, Transport Servces, Educaton, Health, Recreaton, Communcaton, Restaurant & mscellaneous expendture. To dentfy the parameters of the ELES we also compled personal savngs for 2005 for the countres from ADB Data Platform. The descrptve statstcs ndcate that n Afrca the average share of household consumpton expendture on Food and Drnks s around 42%. Certanly there are outlers where food consumpton s close to or more than total food expendture due to negatve personal savng rates. In general however, the share of food expendture follows the well documented pattern that t declnes wth the level of economc development. Relatvely well off countres spend a small share of ther ncome on food whle poorer countres tend to spend a sgnfcant porton of ncome on food. Ths s dsplayed clearly n Fgure 1 wth the slght hump at the low level of ncome, but declnng smoothly afterwards. Expendture on Water & Electrcty comes next to Food and the shares for other commodty groups are less than 10% n general. Average personal savngs hover around 14% of dsposable ncome, whch n contrast to other developng regons s stll very low. Profle of consumpton expendture vares consderably across countres. Food consumpton expendture vares from the lowest ranges n Zamba and Botswana to Lesotho and Comoros that reported average expendture shares close to 90% to 100% due to low or negatve savng rates. Ths dversty s evdent also for other commodty groupngs. Household expendture shares on educaton and health are generally drven by polcy factors. Places where free prmary educaton or health care servces are not ntroduced generally experence hgh out of pocket expendture (such as Lesotho and Serra Leone). 4. Dscusson of results The concentraton curves provde nonparametrc comparson of welfare changes nduced by polcy or prce shocks on a wde range of commodtes. The Lorenz curve, whch s a specal case of concentraton curve can be used as a reference to compare for nstance proportonal tax or (drect budget support n the case of external transfers) to fnance certan publcly provded commodtes such as educaton, health, water & electrcty or 10

11 subsdes of food tems, etc. For Afrcan countres, the Lorenz curve s dstngushngly skewed reflectng the large varaton n ncome levels across the contnent. The mpled Gn coeffcent s around 45% whch s close to what s often reported for Afrca. Thus there s a large crosscountry nequalty perhaps more than some other developng regons. In Fgure 2 we compare two of the recently headlne catchng shocks experenced by households n Afrca: food and energy prce crss. One could pose the polcy problem for nstance as follows. In lght of hgher food and energy prces, would t make sense to subsdze food and energy tems fnanced say through proportonal ncome tax or borrowng externally (at may be concessonal rates). Fgure 1 ndcates that subsdzng food s clearly welfare mprovng at the contnental level whle focusng on energy prces may not. In fact, on the aggregate, subsdzng actvtes related to transport wll worsen welfare. What about transferrng resources to fnance socal sectors such as educaton and health? It s evdent from Fgure 3 that spendng on educaton leads to superor welfare mprovement n comparson to any other commodty, ncludng food tems. But, n general, any nternatonal transfers spent on food, educaton or health s much better than say transfers that mprove household ncome. Ths s the ntuton of most donors. Extendng the dscusson n the context of full demand system provdes some valuable parameters that are often used to calbrate global models. Based on the dscusson n secton 2, Table 3 reports parameter estmates of the Extended Lnear Expendture System where personal savngs s used to dentfy all the parameters. The results generally ndcate a well behavng lnear relatonshp between demand and personal ncome. Our estmaton method also adjusts for possble contemporaneous correlaton across equatons and the explanatory power of the lnear Engel curve s also reasonably hgh. Generally, the margnal budget share for some commodtes s lower than the average share so that demand s ncome nelastc. These are Food, Clothng, Water and Electrcty, Health and Educaton. It s nterestng to note that household per capta expendture on educaton s unformly dstrbuted n all countres across Afrca and the average budget share s also among the lowest. One s tempted to relate ths feature to efforts by governments to publcly provde educaton servces. The other nterestng dmenson s also the fact that average school attanment rate vary consderably despte proportonal effort by households n poor and rch countres to nvest on educaton. Luxury goods are the usual suspects: Alcohol, communcaton, recreaton, transport, restaurant related expenses, etc. Qute strkngly the elastcty values reported for ths broadly aggregated commodtes s more or less consstent wth what one often fnds from large household surveys of ndvdual countres. The prce responses ndcate (Table 5) qute a dampened feature largely because of the level of aggregaton (substtuton across commodtes s not feasble) as well as the structure of the ELES whch s based towards ncome elastcty. Another explanaton s also the hgh value of the Frsch parameter (or margnal utlty of ncome) whch lnks prce elastcty wth the ncome elastcty. Not surprsngly, most commodtes are prce nelastc as reported n Table 11

12 (5). The man factor drvng the own prce elastcty n ths set up s the margnal budget share (the hgher the household spends on partcular commodty for a one dollar ncrease n ncome, the larger the response to own prce shocks, vce versa) and the share of subsstence consumpton n total consumpton or the Frsch Parameter. The role of savngs to dentfy all the parameters of the ELES s crucal. Moreover, the mpled margnal budget share s also nterestng. A one dollar ncrease n per capta ncome could lead to a savng of 36 cents on the average mplyng that savngs s an ncome elastc commodty n Afrca. Fnally, our estmate of total consumpton expendture needed for subsstence s close to 407 n PPP whch s about 1.12 dollar a day, close to the 1.09 dollar a day conventonally used at a global level. To a certan degree, ths fndng gves credence to the global poverty lne whch has been under a lot of scrutny lately (Deaton, 2009) and Deaton & xxx (2010). 5. Conclusons Ths paper attempted to utlze the data generated by the ICPAfrca 2005 on 48 Afrcan countres to extract some welfare comparsons across a broadly defned group of commodtes. No doubt that such level of aggregaton may be consdered a bt stretchng the underlyng concept of choce theory whch essentally s bult on a number of restrctve assumptons. However, the whole dea of buldng the ICPAfrca 2005 data s to be able to compare standard of lvng across countres n a consstent framework. In that sense, then, aggregatng per capta consumpton expendture and some of ts extensons such as poverty, nequalty or any other measure of welfare s allowed. The 12 commodtes covered n the ICP2005 survey are also comprehensve allowng for some nterestng nferences that may help polcy dalogues. For nstance, should food be subsdzed at the expense of say fuel, or should drect ncome transfers (such as budget support) promote household welfare nstead of some targeted expendture say on health, educaton and other necesstes such as food. In dealng wth these ssues, certanly household level data at a country level would be more sensble because of the realsm n polcy actons. In a context where crosscountry polcy coordnaton s hardly observed our comparson may sound theoretcal. However there are nstances where well known global models requre such nputs for ther calbraton. One of the most frequently used global model f Global Trade Analyss Project whch focuses on crossregonal polcy smulatons such trade lberalzaton. One of the components modeled s household behavor where actually the Extended Lnear Expendture System s specfed to capture prce and ncome responses. Our computatons help dentfy these parameters for such exercses easly. Often, avalable country nformaton s mputed for the whole regon to run the models. Thus, our results may fll these gaps. The other nterestng dmenson of ths global demand analyss s that the estmated parameters strkngly are close to what one would obtan from household surveys n ths 12

13 settngs. There s nothng strange or out of the ordnary n our estmates of ncome and prce elastcty values. Fnally, even our poverty lne estmate from subsstence expendture mpled by the model s very close to the nternatonal poverty lne whch was computed usng a completely dfferent approach. 13

14 References Ahmad,E., and Stern, N., (1984), The Theory Of Indrect Tax Reform and Indan Indrect Taxes, Journal Of Publc Economcs, 25, Atknson, A.B., 1970, The Measurement of Inequalty, Journal of Economc Theory, 2, Ahmad, S. (2006), Hstorcal overvew of Internatonal Comparson Program, mmeo World Bank Chen, Shaohua & Ravallon, Martn, "The developng world s poorer than we thought, but no less successful n the fght aganst poverty," Polcy Research Workng Paper Seres 4703, The World Bank. Chen, Shaohua & Ravallon, Martn, "The changng profle of poverty n the world:," 2020 vson brefs BB01 Specal Edton, Internatonal Food Polcy Research Insttute (IFPRI). Chen, Shaohua & Ravallon, Martn, "Absolute poverty measures for the developng world, ," Polcy Research Workng Paper Seres 4211, The World Bank Chen, Shaohua & Ravallon, Martn, "How Have the World's Poorest Fared Snce the Early 1980s?," Polcy Research Workng Paper Seres 3341, The World Bank: Cragg, M., 1991, Do we care; A study of Canada's Indrect Tax System Canadan Journal of Economcs, XXIV, No.1. Feb, Davdson. R., Duclos, J.Y "Statstcal Inference for Stochastc Domnance and for the Measurement of Poverty and Inequalty,"Econometrca, Vol. 68 (6) pp Deaton, A.S. (2009) Purchasng Power Party Exchange Rates for the Global Poor, memo, Prnceton Unversty Deaton, A.S., 1981, Optmal Taxes and The Structure of Preferences, Econometrca, Vol.49, Deaton, A. and J. Mullbeauer (1980): Economcs and Consumer Behavour, Cambrdge Unversty Press. Cambrdge Dowrck, Steve (2005). Errors n the Penn World Table demographc data. Economcs Letters, 86,

15 Dowrck, Steve and Quggn, John (1997). True measures of GDP and convergence. Amercan Economc Revew, 87, March, Ferrera, Francsco H.G. & Ravallon, Martn, "Global poverty and nequalty : a revew of the evdence," Polcy Research Workng Paper Seres 4623, The World Bank Hagenaars, A., 1987, A Class of Poverty Indces, Internatonal Economc Revew, vol 28, no. 3, Haggablade, S and Younger. S Indrect Tax Incdence n Madagascar: Updated Estmates Usng the InputOutput Table. Mmeo. Heston, Alan and Summers, Robert (1996). Internatonal prce and quantty comparsons: potentals and ptfalls. Amercan Economc Revew, 86(2), May. Kakwan, N., 1980 Income Inequalty, and Poverty: Methods of Estmaton and Polcy Applcatons, Oxford Unversty Press Kng., M., 1983, Welfare Analyss of Tax Reform, Journal of Publc Economcs, 21, Knowles, Stephen (2001) Are the Penn World Tables data on government consumpton and nvestment beng msused? Economcs Letters, May, 71(2), Kravs, Irvng B. (1984) Comparatve studes of natonal ncome and prces. Journal of Economc Lterature, 22, March, 139. Marrs, Robn (1984) Comparng the ncomes of natons: a crtque of the nternatonal comparson project. Journal of Economc Lterature, 22, Nuxoll, Danel A. (1994) Dfferences n relatve prces and nternatonal dfferences n growth rates. Amercan Economc Revew, 84, December, Summers, Robert and Heston, Alan (1988). A new set of nternatonal comparsons of real product and prce levels estmates for 130 countres, , Revew of Income and Wealth, 34(1), March, 125. Summers, Robert and Heston, Alan (1991). The Penn World Table (Mark 5): an expanded set of nternatonal comparsons, , Quarterly Journal of Economcs, 106(2), May, Thomas W. Hertel and Marnos E. Tsgas (1997) Structure of GTAP n n T.W. Hertel (ed.), Global Trade Analyss: Modelng and Applcatons, Cambrdge Unversty Press 15

16 Table 1: Afrcan countres covered under ICP2005 Country Per capta consumpton expendture n PPP Angola Benn Botswana Burkna Faso Cameroon Cape Verde Central Afrcan Republc Chad Comoros Congo Congo, Democratc Republc Côte d'ivore Djbout Egypt Equatoral Gunea Ethopa Gabon Gamba Ghana Gunea GuneaBssau Kenya Lesotho Lbera Madagascar Malaw Mal Maurtana Maurtus

17 Morocco Mozambque Namba Nger Ngera Rwanda Sao Tome and Prncpe Senegal Serra Leone South Afrca Sudan Swazland Tanzana Togo Tunsa Uganda Zamba Zmbabwe Mean Table 2: summary statstcs Average expendture ratos Obs Mean Std. Dev. Mn Max Food expendture Savng rato Water Transport servces Clothng Household utensls Mscellaneous goods Health Alcohol & tobacco Educaton Recreaton Restaurant Communcaton

18 Table 3: Seemngly Unrelated Regresson estmate for the ELES parameters: Dependent varable s dsposable ncome Broad consumpton categores Coeffcent SD Zvalue Food and Nonalcohol _cons Alcohol _cons Clothng _cons Water and electrcty _cons Household utensls _cons Health _cons Transport _cons Communcatons _cons Recreaton _cons Educaton _cons Restaurant _cons Mscellaneous goods _cons BreuschPagan test of ndependence: ch2(66) = , Pr =

19 Table 4: Subsstence consumpton and ncome elastcty of demand for broad commodtes Commodty groupngs Subsstence consumpton Margnal budget share Average budget share Income e Food & Nonalchold drnk Alchol & tobacco Clothng Water & electrcty Furnture (nondurable) Health Transport Communcatons Recreaton Educaton Restauant Mscellaneous goods Total subststence expendture

20 Table 5: cross and own prce elastcty values from the ELES Food & Nonalchold drnk Alchol & tobacco Clothng Water Food & Nonalchold drnk Alchol & tobacco Clothng Water & electrcty Furnture (nondurable) Health Transport Communcatons Recreaton Educaton Restauant Mscellaneous goods Furnture (nondurable) Health Transport Communcatons Recreaton Educaton Restauant Mscellaneous goods

21 Fgure 1: Share of food n total consumpton expendture n Afrca: 2005 Share of food n total expendture Per capta consumpton expendture Fgure 2: Engel functon for Food expendture Food & drnks; dsposable ncome dsposable ncome Food Engel curve 45 degree lne 21

22 Fgure 3: Concentraton curve for selected commodtes n Afrca usng data from ICP 2005 L(p) Percentles (p) lne_45 Per capta exp on transport Per capta dsposable ncome Food per capta expendture Total per capta expendture Fgure 4: Concentraton curve for selected commodtes n Afrca usng data from ICP 2005 C(p) Percentles (p) lne_45 Per capta health expendture Per capta dsposable ncome Food per capta expendture Per capta expendture on educaton 22