Analysis of Household Demand for Food in South Africa: Model Selection, Expenditure Endogeneity, and the Influence of Socio- Demographic Effects

Transcription

1 Analyss of Household Demand for Food n South Afrca: Model Selecton, Expendture Endogenety, and the Influence of Soco- Demographc Effects Lesba Bopape and Robert Myers Abstract Ths study analyzes food expendture patterns n South Afrca, takng nto account dfferences n demand behavor across rural and urban households, as well as across ncome groups. The analyss s carred out usng the QUAIDS model, accountng for demographc effects, structural change, and seasonalty effects. Expendture endogenety s also tested and controlled for. The study makes use of household food consumpton data, collected as part of the wazulu-natal Income Dynamcs Study. Demand behavor dffers sgnfcantly between rural and urban households, as well as across ncome groups, mplyng that an accurate analyss of expendture patterns n South Afrca requres a dsaggregated analyss that takes nto account these dfferences n demand behavor. ey words: food demand, expendture elastcty, prce elastcty JEL Classfcaton: D2, Q8 Selected paper prepared for presentaton at the Afrcan Econometrcs Socety Annual Conference, Cape Town, South Afrca, July 4-6, 2007 Lesba Bopape s an economst, Economc Statstcs Cluster, Statstcs South Afrca, Pretora, South Afrca; and Robert Myers s Unversty Dstngushed Professor, Department of Agrcultural Economcs, East Lansng, Mchgan, U.S.A. Ths work s part of Bopape s Ph.D. dssertaton at Mchgan State Unversty. We thank Jeffrey Wooldrdge, John Staatz, and elly Raper for ther helpful comments. All remanng errors are our own.

2 Analyss of Household Demand for Food n South Afrca: Model Selecton, Expendture Endogenety, and the Influence of Soco-Demographc Effects. Introducton South Afrca has the hghest per capta ncome n Sub-Saharan Afrca and s categorzed as a mddle ncome country. Furthermore, aggregate per capta avalablty data suggest that South Afrca s food secure n almost all basc foodstuffs (IFSS, 2002). These facts suggest that hunger and food securty should not be major polcy ssues n the country. However, these aggregate data mask a hghly unequal dstrbuton of ncome and a huge dvde between relatvely affluent urban areas and desttute condtons n many rural communtes. For nstance, a study to assess the household level food securty status ndcated that over 30% of the populaton s vulnerable to food nsecurty and over 20% of the chldren are estmated to be stunted and vtamn A defcent (HSRC, 2004). Polces desgned to reduce ncome nequalty, hunger, and malnutrton formed an mportant part of the socal and economc reforms ntroduced by the democratc government n 994. But ncome nequalty and household food nsecurty reman. One of the problems s that lttle s known about how food expendture patterns dffer across dfferent ncome groups, and across dfferent geographc regons. Wthout a thorough understandng of these dfferences n behavoral patterns n food expendtures, and how these patterns are changng over tme, t wll contnue to be dffcult to desgn polces that mprove food securty effectvely over a broad range of heterogeneous low-ncome households. Ths paper seeks to mprove knowledge and understandng of food expendture patterns n South Afrca, takng nto account dfferences n demand behavor across rural

3 and urban households, as well as across ncome groups. The study makes use of an unusually rch dataset on household food consumpton, collected as part of the wazulu- Natal Income Dynamcs Study (IDS). The IDS dataset contans detaled nformaton on food expendtures and prces as well as household socoeconomc and demographc characterstcs. The IDS ntervewed households over a ten-year perod, wth surveys n 993, 998, and Ths paper uses the IDS data to estmate demand functons for seven food groups grans, meat and fsh, fruts and vegetables, dary, ols and fats, sugar, and all other foods. The Quadratc Almost Ideal Demand System (QUAIDS) of Banks et al. (997) s used to estmate prce and expendture elastctes, takng nto account the mpact of household demographc characterstcs on food expendture patterns. Prevous studes on food demand n South Afrca have typcally been based on hghly aggregate data and have ether been lmted to examnng only one commodty (e.g., Taljaard, 2003; Neuwoudt, 998) or gnored any mpact of demographc factors on food demand (Bowmaker and Neuwoudt, 990; Lebenberg and Groenewald, 997). The excepton s the study by Agbola (2003) whch s based on mcro data and ncorporates household demographcs. However, Agbola uses cross-sectonal data that was collected n 993, one year pror to the major reforms ntroduced by the democratc government. Furthermore, Agbola s study s based on a restrctve lnearzed Almost Ideal Demand System (LA/AIDS) model, whch does not allow for adequate curvature n the Engel curves. In a related study usng the IDS dataset, Bopape and Myers (2006) explctly test for whether the demand model should be specfed wth a quadratc (QUAIDS) or a lnear (AIDS) expendture term and found evdence aganst AIDS. Ths study also tests 2

4 for expendture endogenety and controls for t where necessary. The problem of expendture endogenety s typcally gnored n food demand studes. 2. Theoretcal Framework 2.. The QUAIDS Model The almost deal demand system (AIDS) of Deaton and Muellbauer (980) has been a popular functonal form to model demand behavor durng the past two decades. The AIDS model has budget shares that are lnear functons of log total expendture. AIDS s a member of the Prce-Independent Generalzed Logarthmc (PIGLOG) class of demand models (Muellbauer, 976), whch are derved from ndrect utlty functons that are themselves lnear n log total expendture. However, there s a growng body of lterature provdng evdence on the mportance of allowng for nonlnearty n the budget share equatons (Lewbel, 99; Banks et al., 997). The quadratc almost deal demand system (QUAIDS) model developed by Banks et al. (997), whch has budget shares that are quadratc n log total expendture, s an example of the emprcal demand systems that have been developed to allow for ths expendture nonlnearty. The QUAIDS model s a generalzaton of PIGLOG preferences based on the followng ndrect utlty (V) functon: ln x ln a( p) ln V = + ( ) ( ) λ p () b p where x s total expendture, p s a vector of prces, a(p) s a functon that s homogenous of degree one n prces, and b(p) and (p) are functons that are homogeneous of degree 3

5 zero n prces. As n the orgnal AIDS model, ln a(p) and ln b(p) are specfed as the translog and Cobb-Douglas equatons: ln a( p ) = α 0 + α ln p + γ j ln p ln p j (2) 2 ( ) = = β = = j= b p p (3) where =,, denote commodtes. The functon (p) s specfed as: ( p) = λ λ ln where λ = 0. (4) = p Applcaton of Roy s dentty to () gves the QUAIDS budget share equatons. To control for varyng preference structures and heterogenety across households, we ncorporate demographc varables (z) nto the QUAIDS model through the lnear demographc translatng method (Pollak and Wales, 98). Ths leads to the followng emprcal specfcaton of the QUAIDS budget share equatons: = w = α + j= γ j ln p j + β x ln a ( ) + ln + ( ) ( ) p b p a p λ x 2 L s= δ s z s (5) where ( z..., ) z s a set of demographc varables. Formulas for the QUAIDS s =, z L expendture and prce elastctes are derved by dfferentatng the budget share equatons wth respect to ln x and ln p j, respectvely. Followng Banks et al. (997), we smplfy the expressons for the elastcty formulas by usng the ntermedate results: λ x ln ( p) a( p) w 2 µ = β + (6) ln x b w λ β j x µ j = γ j µ α j + γ jl ln pl ln ln p ( ) ( ). (7) j l= b p a p 2 4

6 In terms of the µ, the formula for expendture elastctes can be wrtten as: e µ w = +. (8) The expresson for the Marshallan or uncompensated prce elastctes can be wrtten as: e u j µ j = δ j (9) w where j s the ronecker delta takng the value δ = f = j and δ = 0 f j. The Hcksan or compensated prce elastctes are obtaned from the Slutsky equaton: j j e = e + w e. (0) c j u j j The theoretcal restrctons of addng-up, homogenety, and symmetry are mposed on the parameters to ensure ntegrablty of the demand system (Moro and Sckoka, 2000). Adddng-up smply requres that the household does not spend more than ts total budget (.e., w = ), and can be expressed n terms of model parameters as follows: = β = 0 λ = 0 α γ = 0 j. () = = = = Marshallan demands are homogenous of degree zero n ( p,m), and ths property s satsfed by mposng the parametrc restrctons: j= j γ = 0 j. (2) j Slusky symmetry requres that j= γ = 0 j. (3) j 5

7 2.2. Expendture Endogenety Most emprcal demand analyses do not cover all products and servces that households purchase. Data lmtatons, fnte computer memory, and the ncreased complexty and tme requred for estmatng large models make t necessary to abstract from a completely specfed demand system contanng a dfferent equaton for each of the myrad goods avalable n the market (LaFrance, 99). The practce s typcally to assume that preferences are separable and estmate a set of condtonal demands for the goods of nterest as functons of prces and total expendture on these goods (Pollak and Wales, 969). However, such a practce rases questons regardng the possblty of smultanety bas n the budget share equatons of the demand model. Total expendture may be determned jontly wth the expendture shares of the ndvdual commodtes that enter the demand model, makng t endogenous n the expendture share equatons. Estmaton gnorng expendture endogenety may lead to nconsstent demand parameter estmates. In ths study, we follow Bundell and Robn (999) and control for endogenety usng an extenson of the lmted nformaton augmented regresson technque suggested by Hausman (978). To llustrate how the augmented regresson technque works, consder the regresson of y, the dependent varable, on a set of exogenous explanatory varables, z, and an endogenous explanatory varable, y 2,.e., y = z + y 2. Also, suppose an nstrumental varable, z 2, exsts for y 2. Correcton for the endogenety of y 2 usng the augmented regresson technque proceeds n two steps. The frst step nvolves estmatng a reduced form regresson of the endogenous varable on a set of nstrumental varables, where the set of nstrumental varables nclude all the other exogenous explanatory varables (.e., 6

8 regress y 2 on z and z 2 ). The resduals, v, from ths frst-stage regresson are then ncluded as an addtonal explanatory varable n the orgnal y equaton. The OLS estmator of the parameters and n ths augmented regresson s dentcal to the Two-Stage Least Squares (2SLS) estmator (Blundell and Robn, 999). Moreover, testng for the sgnfcance of the coeffcent on v s a test for the exogenety of y 2. We use total household ncome (ln m) as an nstrument for total expendture (ln x). For ncome to be a good nstrumental varable for expendture, t must meet two condtons: the relevance condton, whch requres that ncome be suffcently correlated wth expendture (the endogenous varable), and the exogenety condton, whch requres that ncome must not be correlated wth the error term n the demand model. The former condton s testable whle the latter condton s mantaned. A test for the relevance condton nvolves determnng whether ncome s partally correlated wth expendture. We conduct ths test followng Wooldrdge (2002, pp. 8-22) by determnng whether ln m s statstcally sgnfcant n the reduced form regresson for ln x. Because total household ncome s exogenous n the household food expendture allocaton model, t s not unreasonable to assume that t (total household ncome) satsfes the exogenety condton. 3. Data Data used n ths study comes from the wazulu-natal Income Dynamcs Study (IDS). IDS s a panel dataset comprsng three surveys: the 993 Project for Statstcs on Lvng Standards and Development (PSLSD) survey, and the 998 and 2004 surveys 7

9 whch ntervewed households from the 993 PSLSD survey who resde n wazulu- Natal Provnce. The PSLSD s a naton-wde survey undertaken n the last half of 993 by a consortum of South Afrcan survey groups and unverstes, ncludng the South Afrcan Labor and Development Research Unt of the Unversty of Cape Town and the World Bank. The man nstrument was a comprehensve household survey collectng data on a broad array of soco-economc condtons of households, ncludng ther food and nonfood expendtures. In addton to the household questonnare, a communty questonnare was admnstered n each cluster to collect data on prces for a detaled lst of food products purchased by households. In 998, households surveyed by the PSLSD n wazulu-natal Provnce were rentervewed by the IDS survey. The thrd IDS survey was undertaken n wazulu-natal s the most populous provnce n South Afrca, consttutng approxmately 20% of South Afrca s populaton. The economc, socal, and racal stratfcaton of wazulu-natal mrrors that of the country as a whole: the provnce ncludes a wealthy metropoltan area, Durban, poor townshps surroundng t and a poor and largely rural former homeland, wazulu. Also, poverty and nequalty n the provnce are relatvely smlar to those at the natonal level (Woolard et al., 2002), so that results of the analyss should provde mportant nsghts about the condtons n other provnces. The recall perod n all the three surveys s one month. The 998 and 2004 surveys re-ntervewed the orgnal 993 households, and tracked and ntervewed households that had moved, as well as new households that splt from the orgnal 993 households. After removng observatons deemed unusable for the current purpose (such 8

10 as households reportng zero consumpton on all food tems, or those mssng crtcal nformaton, such as expendture nformaton), a total sample of 28 observatons across the three surveys s used. The problem of reported zero expendtures s not sgnfcant n ths dataset, manly because of the broad commodty group defntons used here. Table presents a summary of the food expendture shares of the sampled households, ncludng the dfferences across ncome groups, and rural and urban areas. Grans consttute the largest share of household total food expendture, rangng from about 26 and 24% among the hgh ncome and urban households to 37% among the low ncome and rural households. The share of grans n the households budgets s lower at hgher ncome levels. The budget share of meat products, a more expensve source of calores, s hgher among the hgh ncome and relatvely affluent urban households. The mean monthly ncome of hgh ncome households s seven tmes more than that of low ncome households, reflectng the generally hgh wealth nequalty n South Afrca. A detaled lstng of the food tems contaned n the food categores are 9

11 Table. Average Expendture Shares by Income Group and Regon Income groups Regons Entre sample Low Mddle Hgh Rural Urban Grans Meat, fsh Fruts, vegetables Dary products Ols, butter, fats Sugar, sugar products Other food Total household food expendture Total household ncome Food expendture as % of ncome Table 2. Varable Summary Statstcs of Household Characterstcs by Income Group and Regon Income Groups Regons Entre sample Low Mddle Hgh Rural Urban Household sze Educaton of head Age of household head Proporton male headed Proporton rural Proporton black (race) Summary statstcs of household demographc characterstcs are presented n Table 2. On average, urban households are of smaller sze, headed by younger males wth hgh levels of educaton. Most of these characterstcs are shared by hgh-ncome households, except the latter have larger famly szes. The rural and low-ncome groups comprse manly black households. 0

12 4. Demand Model Results The QUAIDS model s estmated usng maxmum lkelhood (ML), wth theoretcal restrctons of addng-up, homogenety, and symmetry mposed durng estmaton. Table 3 presents the estmated coeffcents for the QUAIDS model. The quadratc expendture term s statstcally sgnfcant n most of the expendture share equatons. It s n the expendture share equatons for meat and fsh, fruts and vegetables, and ols and fats that the null hypothess of expendture lnearty s not rejected. However, as the results to be presented below ndcate (Table 4), the hypothess that the quadratc expendture term s zero across all equatons s strongly rejected. Only 9 of the 28 prce effects are sgnfcantly dfferent from zero at the 0% sgnfcance level, suggestng that there s not much quantty response to movements n relatve prces, possbly due to the level of aggregaton n the commodty groups. Most (34 out of 49) of the coeffcent estmates on the demographc varables are statstcally dfferent from zero. Households wth large szes consume more grans and dary products whle ther small-szed counterparts consume more meat and fsh and fruts and vegetables. These results are as expected, gven that grans provde a relatvely cheap source of calores compared to such foods as meat and fsh, and that household sze s negatvely correlated wth ncome. Also, large households are lkely to have more chldren who consume mlk and other dary products.

13 Table 3 Constant 0.85 (0.050) Estmated Prce, Income, and Demographc Effects Grans Meat/fsh Fruts/Veg Dary products (0.0234) (0.067) (0.04) Ols/butter/ fats (0.0084) Sugar 0.05 (0.007) Other foods (0.024) PGRA (0.08) (0.0089) (0.006) (0.0054) (0.0039) (0.0032) (0.0083) PMF (0.0089) (0.02) (0.0060) (0.0052) (0.0037) (0.003) (0.0078) PFRUVE (0.006) (0.0060) 0.0 (0.0060) (0.0038) (0.0027) (0.0023) (0.0059) PDAIRY (0.0054) (0.0052) (0.0038) (0.0050) (0.0027) (0.0023) (0.0052) POBF (0.0039) (0.0037) (0.0027) (0.0027) (0.0039) (0.0024) (0.0037) PSUG (0.0032) (0.003) (0.0023) (0.0023) (0.0024) (0.0030) (0.0033) POTH (0.0083) (0.0078) (0.0059) (0.0052) (0.0037) (0.0033) (0.009) ln x (0.09) (0.03) (0.0084) -0.0 (0.0068) (0.0038) (0.0032) (0.004) (ln x) (0.004) (0.0038) (0.0029) (0.0023) (0.003) (0.00) (0.0036) Household sze (0.0007) (0.0007) (0.0005) (0.0004) (0.0002) (0.0002) (0.0006) Race (= black) (0.0093) (0.0089) (0.0065) (0.0053) (0.003) (0.0026) (0.0080) Rural (= rural) (0.0066) (0.0062) (0.0045) (0.0036) (0.002) (0.008) (0.0055) Educ. of head (0.0006) (0.0005) (0.0004) (0.0003) (0.0002) (0.000) 0.00 (0.0005) Dummy (0.059) (0.056) (0.0) (0.007) (0.006) (0.005) (0.040) Dummy (0.0204) (0.023) (0.057) (0.044) (0.0094) (0.0078) (0.085) Survey month (0.002) (0.0020) (0.004) (0.002) (0.0007) (0.0006) (0.008) v (0.02) (0.03) (0.0082) (0.0067) (0.0040) (0.0034) (0.00) Notes: () Estmated standard errors n parentheses, (2) All prces are n logarthms, P = prce, GRA = grans, MF = meat and fsh, FRUVE = fruts and vegetables, DAIRY = dary products, OBF = ols, butter and fats, SUG = sugar, and OTH = other foods. 2

14 Table A n the appendx presents the parameter estmates of the reduced form regresson for ln x. The results of the F test for the jont sgnfcance of ln m and (ln m) 2 are presented n the bottom row of Table A. These results provde evdence of a strong partal correlaton between ln m and (ln m) 2 and ln x, and hence, provde evdence that ncome s a relevant nstrument for expendture. Results of the test for expendture exogenety are presented at the bottom row of Table 3. The t values for the sgnfcance of the v resduals n the ndvdual budget share equatons are clearly low, mplyng that the exogenety of log expendture cannot be rejected n each of these equatons. The mddle column of Table A n the appendx presents the parameter estmates of the reduced form regresson for ln x. The results of the F test for the jont sgnfcance of ln m and (ln m) 2 are presented n the bottom row of Table A. These results provde evdence of a strong partal correlaton between ln m and (ln m) 2 and ln x, and hence, provde evdence that ncome s a relevant nstrument for expendture. Results of the formal test of the hypothess that log expendture s exogenous across all the budget share equatons are presented n Table 4. As these results clearly show, expendture exogenety cannot be rejected. Hence, n estmatng the demand system usng ths sample of households, there s no need to control for expendture endogenety. Table 4 also presents other specfcaton tests. The AIDS model s rejected n favor of QUAIDS. Demographc varables clearly nfluence demand behavor. Aggregate tme effects, captured by year dummy varables, are sgnfcant, whch provdes some evdence of structural change durng the sample perod. The month of the survey s also sgnfcant, whch ndcates the 3

15 mportance of seasonalty n the food purchase and consumpton patterns of these households. Table 4 Results of the Wald Tests for AIDS specfcaton, Demographc effects, Structural Change, and Seasonalty 2 Degrees of freedom p-value Expendture endogenety (all equatons) AIDS specfcaton Demographc effects Structural change (aggregate tme effects) Month/seasonalty effects It s easer to nterpret prce and ncome effects n terms of elastctes. Estmates of expendture elastctes ( s) are presented n Table 5. The frst column presents expendture elastcty estmates for the entre sample, whle the other columns reports those for the rural, urban, and ncome groups sub-samples. All expendture elastcty estmates are postve, as would be expected for broadly defned food aggregates lke the ones consdered here. Most (39 out of 42) of the estmates are statstcally dfferent from zero at less than % sgnfcance level. Grans are expendture nelastc across all household groups, wth substantal rural-urban and ncome group dfferences. These estmates for expendture elastcty for grans dffer sgnfcantly from the earler estmates ( =.250) reported by Agbola (2003) usng the 993 natonwde crosssectonal data. Meat and fsh are luxures across all household groups; expendture on meat and fsh s more elastc among rural and low ncome households than among urban 4

16 and hgh ncome households. The expendture elastcty estmates for meat and fsh are close to the.027 reported by Agbola. Expendtures on fruts and vegetables and ols, butter and fats are relatvely more responsve to ncreases n total household expendture among urban and hgh ncome households than among rural and low ncome households. Table 5. Estmated Expendture Elastctes Expendture elastctes ( ) Commodty Entre sample Rural (n = 446) Urban (n = 735) Low Mddle Hgh Grans (0.254) Meat, fsh.363 (0.326) Fruts, vegetables (0.095) Dary products (0.58) Ols, butter, fats (0.0989) Sugar (0.0886) Other (0.88) Note: Standard errors n parentheses (0.85).496 (0.343) (0.062) (0.692) (0.0989) (0.0804) (0.2075) (0.430).56 (0.34) (0.62) (0.3) (0.0989) (0.20) (0.566) (0.39).528 (0.309) (0.07) (0.760) (0.0926) (0.0760) (0.2082) (0.22).420 (0.329) (0.078) (0.545) (0.0990) (0.0830) (0.2032) (0.436).84 (0.359) (0.94) (0.370) (0.034) (0.60) (0.598) Table 6 presents estmates of the Marshallan and Hcksan own prce elastctes. Own prce elastctes are all negatve as expected. Based on the uncompensated prce elastcty estmates, fruts and vegetables and dary products are prce nelastc across all household groups, and all other food groups are prce elastc. Hence, households respond more than proportonately to changes n the prces of these foods. However, when the substtuton effects are consdered, grans and meat and fsh become prce elastc, wth compensated own-prce elastcty estmates of less than unty. It s only n the case of hgh ncome households that grans reman prce elastc when both the uncompensated 5

17 and compensated elastcty estmates are consdered. Ths ndcates the greater substtuton possbltes that hgh ncome households have n respondng to changes n the prces of grans compared to the other household groups. Table 6. Estmated Own-Prce Elastctes Marshallan/uncompensated own-prce elastctes Commodty Entre sample Rural (n = 446) Urban (n = 735) Low Mddle Hgh Grans (0.0742) Meat, fsh (0.586) Fruts, vegetables (0.0957) Dary (0.0788) Ols, butter, fats (0.0745) Sugar (0.0749) Other (0.6605) (0.0628) -.79 (0.679) (0.0909) (0.0926) (0.0755) (0.065) (0.8068) (0.086) (0.448) (0.06) (0.065) (0.0726) (0.042) (0.5037) (0.0574) (0.624) (0.0848) (0.0976) (0.0720) (0.0597) (0.7706) (0.0673) -.2 (0.69) (0.0933) (0.082) (0.0752) (0.0682) (0.7880) ( ( ( ( ( ( ( Hcksan/compensated own-prce elastctes Grans (0.0408) (0.0435) (0.0350) (0.047) (0.0423) ( Meat, fsh (0.0306) (0.0283) (0.0354) (0.027) (0.0295) ( Fruts, vegetables (0.093) (0.092) (0.094) (0.089) (0.092) (0.096 Dary (0.000) (0.0094) (0.03) (0.009) (0.0098) (0.02 Ols, butter, fats (0.0053) (0.0052) (0.0054) (0.0052) (0.0052) ( Sugar (0.0043) (0.0044) (0.004) (0.0044) (0.0044) (0.004 Other (0.08) (0.062) (0.0222) (0.057) (0.066) ( All elastctes are computed from endogenety-adjusted QUAIDS model 2. These are the expendture and own-prce elastctes computed from endogenety-unadjusted QUAIDS model 6

18 Agbola s uncompensated and compensated prce elastcty estmates for grans are.730 and.394, respectvely. Grans comprse manly of tems that are staple foods n South Afrca, so that one can consder Agbola s estmates to be too hgh. The expendture and prce elastctes estmated n ths study correct for many of the problems assocated wth prevous estmates n South Afrca. Our estmates are based on the flexble QUAIDS model that allows for more curvature n the Engel curves, wth account taken for the nfluence of demographc, seasonalty and aggregate tme effects. Hence, the prce and expendture elastcty estmates presented n ths study are more accurate and can be used more relably for polcy analyss. 5. Conclusons Ths paper presents an analyss of the food expendture patterns of households n South Afrca. Varous tests are performed to decde on the most approprate model specfcaton. The tradtonal AIDS model s rejected n favor of the more flexble QUAIDS model. Ths s the frst study to apply the QUAIDS model to analyze food demand n South Afrca; prevous studes were based on restrctve functonal forms LA/AIDS and log-lnear models. The QUAIDS model s estmated accountng for demographc composton of households, structural change, and seasonalty effects. Meat and fsh are luxures across all household groups. Expendture on meat and fsh s more elastc among rural and low ncome households than among urban and hgh ncome households. Sugar products and ols, butter and fats are own-prce elastc based on both the compensated and uncompensated prce elastcty estmates. Grans, whch comprse 7

19 tems that are manly staple foods n South Afrca, are prce elastc among hgh ncome households. There are substantal dfferences n the consumpton patterns of rural and urban households, as well as across households n dfferent ncome groups. The polcy mplcaton of these dfferences s that the desgn of ant-poverty and nutrent enhancement programs needs to be regon-specfc and take nto account these behavoral dfferences n food expendtures. 8

20 Appendx Table A The estmated reduced form for ln x Varable ln x std. Err. Constant (0.3453) Household sze (0.003) Race ( f black) (0.0427) Rural ( f rural) (0.0308) Educaton 0.08 (0.0026) Dummy for (0.0904) Dummy for (0.4) Survey month (0.0094) Total household ncome (0.0706) Total household ncome (0.0049) lnprces Jontly sgnfcant R F (p-val) 6.90 (0.0000). Standard errors n parentheses 9

21 References Agbola, F.W. Estmaton of Food Demand Patterns n South Afrca Based on a Survey of Households. Journal of Agrcultural and Appled Economcs 35(3)(December 2003): Banks, J.; Blundell, R., and A. Lewbel. Quadratc Engel Curves and Consumer Demand. The Revew of Economcs and Statstcs 79(4)(November 997): Blundell, R.W. and J.L. Powell. Endogenety n Semparametrc Bnary Response Models. Revew of Economc Studes 7(2004): Blundell, R.W. and J.M. Robn. Estmaton n Large and Dsaggregated Demand Systems: An Estmator for Condtonally Lnear Systems. Journal of Appled Econometrcs 4(999): Blundell, R.W., P. Pashardes, and G. Weber. What Do We Learn About Consumer Demand Patterns from Mcro Data. Amercan Economc Revew 83(3) (June993): Bowmaker, P.A., and W.L Neuwoudt. Demand equatons for selected South Afrcan agrcultural commodtes. Agrekon 29()(990): Deaton, A., and J. Muellbauer. Economcs and Consumer Behavor. Cambrdge Unversty Press, 980a. Deaton, A., and J. Muellbauer. An Almost Ideal Demand System. Amercan Economc Revew 70(3)(June 980b): Dhar, T., J.P. Chavas, and B.W. Gould. An Emprcal Assessment of Endogenety Issues n Demand Analyss for Dfferentated Products. Amercan Journal of Agrcultural Economcs 85(3)(August 2003): Döckel, J.A., and J.A. Groenewald. The demand for food n South Afrca. Agrekon 9(4)(970):5-20. Ellot, M.B., and J. van Zyl. The effect of ncome on the market for carbohydrates n South Afrca. Agrekon 30(4)(99): Hausman, J.A. Specfcaton Tests n Econometrcs. Econometrca 46(978):25-27 Human Scence Research Councl (January 2004). Food Securty n South Afrca: ey Polces Issues for the Medum Term (Onlne). Avalable from: 20

22 Integrated Food Securty Strategy (IFSS). Natonal Department of Agrculture Polcy Document. Pretora, South Afrca, LaFrance, J.T. When s Expendture Exogenous n Separable Demand Models? Western Journal of Agrcultural Economcs 6(99):49-62 Lebbrandt, M., and I. Woolard. A comparson of poverty n South Afrca s nne provnces. Development Southern Afrca 6()(March 999): Loubser, M. (990). Income Elastctes of the demand for consumer goods and servces. Bureau of Market Research No. 75, Unversty of South Afrca, Pretora. Lubbe, W.F. Decomposton of prce tme seres components of the red meat ndustry for effcent polcy and marketng strateges. Agrekon 29(4)(990): May, J., and I. Woolard. Measurement and experence. In J. May (Ed.), Poverty and nequalty n South Afrca: report prepared for the offce of the executve deputy presdent and the nter-mnsteral commttee on poverty and nequalty. Durban, Praxs Publshng, 998. Neuwoudt, W.L. The demand for lvestock products n South Afrca for 2000, 200 and 2020: Part." Agrekon 37(2)(998). Po, B. P. From the Help Desk: Demand System Estmaton. Stata Journal 2(4)(2002): Pollak, R. A., and T.J. Wales. Estmaton of the Lnear Expendture System. Econometrca 37(4)(October 969): Poonyth, D., R. Hassan, and J.F. rsten. Random Coeffcents Analyss of Changes n Meat Consumpton Preferences n South Afrca. Agrekon 40 (200): Taljaard, P.R., Z.G. Alemu, and H.D. van Schalkwyk. The Demand for Meat n South Afrca: An Almost Ideal Estmaton. Agrekon 43(4)(December 2004): Woolard, I., S. lasen, and M. Lebbrandt. Income moblty and household dynamcs n South Afrca: The case of wazulu-natal. Labour Markets and Socal Fronters 2(2002):5-. Wooldrdge, J.M. Econometrc Analyss of Cross Secton and Panel Data. Cambrdge, MA: MIT Press, World Bank. GNI Per Capta 2004, Atlas Method and PPP. At: World Development Report (2002), The World Bank. 2