Product Quality and the Demand for Food: The Case of Urban China

Transcription

1 Product Qualty and the Demand for Food: The Case of Urban Chna Bran W. Gould Wsconsn Center for Dary Research and Department of Agrcultural and Appled Economcs Unversty of Wsconsn-Madson Dansheng Dong Department of Appled Economcs and Management Cornell Unversty Paper Presented at the 2004 Annual Meetng Amercan Agrcultural Economcs Assocaton August 1-4, 2004 Denver, CO A Quadratc Almost Ideal Demand System s used to examne the structure of food demand for a sample of urban Chnese households. The dual choce of product qualty and quantty s accounted for n the econometrc model va the ncluson of smultaneously estmated unt-value equatons. Correspondng Author: Bran W. Gould Wsconsn Center for Dary Research and Department of Agrcultural and Appled Economcs Unversty of Wsconsn-Madson 427 Lorch Street Madson, WI USA gould@aae.wsc.edu JEL Classfcaton: D12, I31, O12, Q17 Keywords: Chna, Product Qualty, QUAIDS The authors would lke to thank the Babcock Insttute for Internatonal Dary Research and Development for provdng research support.

2 1 Product Qualty and the Demand for Food: The Case of Urban Chna Introducton The Chnese populaton s undergong sgnfcant change n ts eatng patterns resultng from a varety of factors ncludng: change n the degree of government partcpaton n the food system, ncreased affluence of urban households and Chna s recent admttance to the WTO (Gale, 2003; Wang, 1997). As Chna contnues to develop there s general consensus that there wll be movement of the Chnese det away from the tradtonal (Gao, et al., 1996; Guo, et.al., 2000). Wth rsng ncomes t s projected that the Chnese populaton wll dversfy ther dets away from staples such as rce and wheat flour, to one contanng more lvestock products (Shono, et al., 2000; Gale, 2003). Besdes changes n the quantty of products purchases, Hsu, et.al (2001) note that t s mportant to recognze that the demand for product qualty wll become an ncreasngly mportant component of the food purchase process. For example, wth hgher ncomes, one would antcpate an ncrease n the demand for Aready-to-eat@ convenence, nutrtonally enhanced and/or alternatvely packaged foods. There are a varety of methods that can be used to quantfy the determnants of food demand n Chna. Banks, et al. (1996) assert that a demand system approach based on some underlyng utlty functon s preferred over sngle equaton approaches gven ther theoretcal consstency. In the present analyss we use the quadratc almost deal demand system (QUAIDS) of Banks, et. al. (1996, 1997) and Moro and Sckoka (2000). The data whch forms the foundaton of our emprcal model s a dataset that encompasses yearly food purchases by a sample of urban Chnese households. Our emprcal demand system s defned over 12 aggregate commodtes. Cross-sectonal surveys of food purchase behavor often contan purchase quantty and expendture nformaton. Dvson of observed expendtures by quantty (here referred to as unt-value) s often used as an estmate of a commodty=s prce (Gould 1996; Yen at al, 2003). Prevous analyses have recognzed that ths method of calculatng prce reflects not only dfferences n market prces faced by each household but also n endogenously determned commodty qualty (Thel, 1953; Houthaker, 1952; Deaton, 1997, 1988; Cox and Wohlgenant, 1986; and Nelson, 1991). For example,

3 2 observed dfferences n the prce pad for cheese may be reflectng not only local market condtons but also product form. That s, households purchasng cheese n block form would be expected to pay a lower prce than households purchasng cheese that s preslced or shredded, ceters parbus gven the addtonal value-added encompassed n the latter product forms. As Nelson (1991) notes, the porton of product prce determned by market forces s obvously beyond the control of the consumer whereas the qualty porton s endogenous to the purchase process. To assst n dfferentatng between these two forces, Nelson (1991) presents a revew of the consumer purchase process from the perspectve of both elementary goods (x ) and composte commodtes (Q j ) where an elementary good s relatvely homogeneous whle a composte commodty encompasses a set of elementary goods that vary accordng to characterstc(s) such as flavor, fat content, packagng, or product form. (p.1206). An example of an elementary good would be 2% mlk purchased n a half gallon sze package. In contrast, the commodty category flud mlk would represent a composte commodty that encompasses the above elementary good. We develop a model structure where we dfferentate between the market component of observed unt values and that porton that s due to endogenous qualty decsons of the household. Our econometrc model conssts of a system of expendture shares derved from the ndrect utlty functon assocated wth the QUAIDS model specfcaton. We augment ths share system wth a seres of unt value equatons that contan market level varables. Household characterstcs whch represent reduced form mpacts on endogenous unt values are also ncluded n the estmated unt value equatons. We use the results of Deaton (1988, 1997) to convert the estmated unt-value elastctes to the tradtonally nterpreted prce elastctes. Secton II of ths manuscrpt presents an overvew of the QUAIDS specfcaton of Banks, et al., (1997) whch s extended to ncorporate demographc characterstcs va the modfyng cost functon of Lewbel (1985). Restrctons mpled by economc theory are dscussed. We also provde the theoretcal framework by whch product qualty s made part of the purchase process. Secton III provdes an overvew of the household level survey data used n our analyss. Food categores and demographc varables are defned and descrptve statstcs provded. In secton IV the estmaton method s dscussed,

4 3 estmated model parameters summarzed and, based on these estmated coeffcents, varous elastcty measures calculated. A general dscusson of the econometrc fndngs s ncluded wthn ths secton. The last secton of ths analyss provdes an overvew of our analyss and future research possbltes are dentfed. Descrpton of the Econometrc Model Overvew of the Quadratc AIDS Model When developng emprcal models of food (and nonfood) demand, prevous research has provded evdence of the mportance of allowng for complex nonlnear relatonshps between the level of total expendtures and such demand (Atknson, et al., 1990; Lewbel, 1991; Hausman, et al., 1995). To ths end, emprcal demand systems have been developed that allow for extended ncome effects. The QUAIDS s an example of such a system where expendture shares are quadratc n the logarthm of total expendtures. Ths specfcaton s based on a generalzaton of preferences represented by the Prce-Independent Generalzed Logarthmc (PIGLOG) structure. Under the orgnal specfcaton PIGLOG demand systems arse from ndrect utlty functons that are themselves lnear n the logarthm of total expendtures (Muellbauer, 1976). An examples of these PIGLOG specfcatons s the ubqutous Almost Ideal Demand System (AIDS) specfcaton of Deaton and Muellbauer (1980). The QUAIDS s based on the followng ndrect utlty (IU) functon: ln IU 1 lnm ln a( p) = + λ b( p) ( p) 1 where the frst term wthn the square brackets s the ndrect utlty functon of a PIGLOG demand system, m s total expendtures, and p, s observed prce. To assure that the homogenety property holds for ths ndrect utlty functon, a(p) s homogeneous of degree 1 n p, and the functons b(p) and λ(p) are dfferentable and homogeneous functons of degree zero n p. The functons ln[a(p)] and b(p) are the translog and Cobb-Douglas prce aggregator functons found n the tradtonal AIDS formulatons: ( ) 0 n 1 n n j j = 1 2= 1 j= 1 lna p = α + α lnp + γ ln p ln p (2) (1)

5 4 ( ) b p n p β = (3) where αj, β and γjare unknown parameters and n the number commodtes n the system. To complete the specfcaton λ(p) s defned as: ( p) n n p = 1 = 1 λ = λ ln where λ = 0 (Banks, et al., 1997). (4) From (1)-(4) and the ndrect utlty functon can be represented as: ln IU n 1 n n lnm α0 + α lnp + γj ln pln p j = 1 2 = 1 j= 1 n = λ ln p n p β + = 1 Applyng Roy s Identty to (5), the QUAIDS expendture shares (w ) can be represented va the followng: 1 1 (5) n m λ ln ln m w = α + γj pj + β + ln j= a( p) b( p) a( p) Incluson of Demographc Characterstcs n the Demand System Our use of household-level data requres that we recognze the heterogeneous nature of food preferences. The use of demand systems that account for such heterogenety has had a long hstory startng wth the efforts of Barten (1964) and extended by Pollack and Wales (1981), Heen and Pompell (1988), Gould, et al. (1991), Blundell, et al. (1993) and Peral (1993). Lewbel (1985) provdes a general conceptual framework for ncorporatng demographc characterstcs nto a demand system. As noted by Peral (1993), the use of demographc translatng has the effect of mpactng the underlyng cost functon va fxed or subsstence level costs whle demographc scalng changes the relatve slope of a household s budget constrant by modfyng the effectve prces va changes n demographc characterstcs. Followng Lewbel (1985) and Peral (1993), household expendture (m * ) can be represented as a functon of household utlty, U, prces, p, and an S-vector of demographc characterstcs, d: 2 (6)

6 * ( ( )) m f CUhpd,,, pd, (7) where C s a well-behaved expendture functon, h( ) and f ( )are contnuous functons that have frst and second dervatves that exst everywhere except possbly n a set of measure zero. The modfyng functon h( ) s assumed to generate non-negatve modfed prces for every commodty and a postve modfed prce for at least one. As noted above, there are a number of approaches that can be used n the specfcaton of h( ). For the present analyss and recognzng our use of unt-values nstead of market prces, we allow for demographc translatng va the system outlned by Peral (1993): * [,, ] M t ( d ) = 1 m f m p d = m p (8) where t (d) s a commodty specfc translatng functon. In our analyss, these translatng functons are specfed as: S ( ) ωsds s= 1 t d where = (9) ωs s the translatng parameter for the th commodty and the s th demographc characterstc. Substtutng (8) and (9) nto the ndrect utlty functon represented n (1) and applyng Roy s Identty, the resultng modfed system of quadratc budget shares of the QUAIDS specfcaton can be obtaned: n * * m λ m w = α + t( d) + γj ln pj + β ln + ln j= a( p) b( p) a( p) Ths specfcaton dffers from that of Abdula (2002) n that we allow more than just share equaton ntercepts to vary across household. Imposton of Theoretcal Restrctons wthn the QUAIDS Model As noted by Moro and Sckoka (2000), to allow for ntegrablty e.g., to be able to derve the underlyng expendture functon gven utlty and prces, a seres of parametrc restrctons need to be mposed. For example addng-up of expendture shares mples: n n n n n αk βk λ k ωks γkj k= 1 k= 1 k= 1 k= 1 k= 1 = 1, = 0, = 0, = 0,and = 0 (11) 2 5 (10)

7 6 The theoretcal restrcton of lnear homogenety wth respect to prce s satsfed va the followng parameter restrctons: n γ = 0 for j=1,2,,n. (12) k= 1 jk Symmetry s satsfed provded that γ j = γ for,j=1,2,,n. (13) j As shown by Banks, et al. (1997) commodty-specfc expendture elastctes, ξ, can be calculated as: ξ * Γ w 2λ m = + 1 where: Γ = + ln b( p) a( p) β w ln(m*) U uncompensated unt value elastctes ( j ) π :. (14) π U j Γ = w j κj j γj 2 m λβ * j m α γj j bp ( ) a( p j= ) where: Γ Γ + t(d) + lnp ln and κ j s the Kronecker delta. From the above uncompensated elastctes, compensated C elastctes ( j ) C U j j w j π can also be evaluated: π = π + ξ (16) Product Qualty and Food Demand The above overvew of the QUIADS model mplctly assumed that we knew market prces. In realty most expendture surveys have nformaton on quanttes purchased and expendtures. Dvdng expendtures by quantty generates estmates of household specfc unt values. The emprcal mplementaton of the QUAIDS model used n ths analyss replaces commodty prces wth unt values. We then adjust the calculated untvalue elastctes to account for the endogenous qualty effects. (15)

8 7 When usng observed unt-values as a proxy for prce, there s an mplct ncorporaton of not only the effects of exogenously determned product prce but also product qualty. To dfferentate these effects, one can defne the utlty (U) maxmzaton problem faced by a household n terms of elementary goods, x : 1 S* ( s* ) S * Max U x, x,, x st.. px = Y (17) 1 2 x,..., x = 1 where p s the th elementary good=s prce, S *, the number of goods and Y, household ncome. As an alternatve, Nelson (1991), usng Hcks= composte commodty theorem, shows that f we assume that wthn each composte commodty, the prces of all elementary goods vary proportonally, then: e j ; and e p = P p j Qj = px (18) j where p e s the base prce of elementary good x, whch Thel (1952) refers to as a qualty ndcator and P j s the j th composte good=s group specfc prce level proportonalty factor. From (17), we can represent the consumer=s optmzaton problem n terms of composte goods as: ( Q, Q,..., Q ).. P j Q = where Q = Q (, ) 1 2 M Max U st Y P Y (19) M j j j j = 1 and P s an M-vector of composte commodty prces. Expendtures on the j th composte commodty (E j ) and resultng unt values, V j can be related to expendtures on the assocated elementary goods va the followng: e e P j Q j E j= px = P j p x = Pj p x = P jq j, where V j =, (20) q j j j j and q = x, the observed physcal quantty of composte commodty j. These unt j j values wll not be exogenous gven that they depend on the endogenously determned qualty of each composte commodty (Deaton, 1997, 1988). As shown by Nelson (1991) and by Dong, Shonkwler and Capps (1998), a quantty-weghted sum of elementary goods base prces can be used as a measure of average qualty of a partcular composte commodty (? j ). That s:

9 8 ψ j x p e Q = q = q j j j j Combnng (20) and (21) the relatonshp between unt value (V j ), market prce (P j ) and qualty (? j ) s: ln V = lnp + lnψ. (22) j j j Usng the results of (22), Deaton (1988, 1997) shows that (21) p j jj j j Ψ lnψ π V lnv lnψ π jj = ηj and π jj = 1+ lnp ξ lnp lnp j j j j (23) where p π jj s the own-prce elastcty, ξ j s the j th commodty expendture elastcty, and? j s the j th qualty elastcty of expendture ( ln j ln E ) Ψ. As Deaton (1988) notes, we would expect that wth should market prces ncrease consumers can be expected to adjust both the quanttes purchase and the qualty of ther purchases (e.g., the composton of the underlyng goods that make up a partcular composte commodty wll be adjusted). We therefore would expect to see a degredaton n product qualty, that s, ln Ψ j < 0 (p.420). ln P j The above mples that f the researcher uses calculated unt-values obtaned from an expendture survey n the estmaton of commodty demand elastctes and nterprets them as such ths demand elastcty ncorporates both exogenous prce and qualty effects. Under the most usual condtons, such elastctes tend to overstate the prce elastcty n absolute magntude f the product of the prce and qualty elastctes s smaller than the expendture elastcty. The above relatonshps provde the framework for adjustng these demand elastctes to control for such qualty effects. From (23) the relatonshp between the own unt-value elastcty π and own-prce elastcty π p jj dln d ln ( Qj ) ( Pj ) s (Deaton, 1997, p.299): U jj d ln dln ( Qj ) ( Vj )

10 9 π U jj p π jj p π jj = π p jj = U 1 + η j π jj ξ j 1 π jj η j ξ j U (24) Dong, Shonkwler and Capps (1998) and Dong and (2000) ncorporate the above nto a model orgnally formulated by Wales and Woodland (1980) to account for selectvty bas n estmatng a condtonal commodty expendture equaton for a composte good whle at the same tme endogenzng unt-value. Under ther two-equaton model, a unt value regresson equaton s formulated along wth condtonal expendture functons where expendture and unt-value equaton error terms are assumed to be normally dstrbuted and correlated. Parameters of the expendture and prce equatons are estmated wthn a sngle lkelhood functon encompassng all observatons. Deaton (1997) provdes a general framework for ncorporatng product qualty wthn an emprcal demand system. Hs methodology nvolves augmentng a system of utlty-based share equatons wth an assocated set of unt-value equatons. In hs orgnal specfcaton, both sets of dependent varables (.e., expendture shares and commodty unt values) are dependent on unobserved prces. Gven the sze of our emprcal model and the complexty of the functonal form used, nstead of usng hs method for solvng for unobserved prces, we use observed unt-values n the share equatons and a set of market level varables n the unt value equatons as nstrumental varables for unobserved market prces. We then use the relatonshps shown n (24) to U P transform the estmated unt-value elastctes ( π j ) to prce elastctes ( j ) π. We augment the QUAIDS share equatons wth the followng unt value equatons: R D * j = τ0 j + τrj r + τdj d + υj r= 1 d= 1 lnv R D (25) where the t * s and t s are coeffcents to be estmated and R r s the r th regonal dummy varable used to capture the exogenous market prce effects on observed unt values. The vector of varables represented by D d correspond to the set of demographc characterstcs (ncludng the logarthm of total FAH expendtures) used as nstruments for the unobserved product qualty. The share equatons represented by (10) and the unt value equatons represented by (25) represent our complete smultaneous system. It should be noted that only the

11 10 expendture share equatons have endogenous varables as arguments (.e., lnv j ). As noted by Greene (2003) expressng all equatons n a reduced form results n a lkelhood functon that s the same as the seemngly unrelated regresson equaton where we assume all error terms to be related va a multvarate normal dstrbuton (p.407). A Full-Informaton Maxmum Lkelhood estmator s used to obtan parameter values. Bref Descrpton of the Chnese Urban Household Food Expendture Data We evaluate the structure of food demand n Chna usng data obtaned from an annual household expendture survey conducted by the State Statstcal Bureau (SSB) for the year Ths data encompasses household expendtures for urban households n the provnces of Jangsu, Shandong, Guangdong, Helongjang and Henan. We use an urban sample to avod ssues assocated wth the consumpton of home produced foods. The frst three provnces are located on the Chna Sea. Jangsu and Quangdong are the most prosperous provnces gven ther locaton close to Shangha and Hong Kong, respectvely (Table 1). The latter two provnces are the poorest of the fve. Both are located n the nteror wth Helongjang beng the most north and east and Henan located west of Shandong. In addton to food purchase quantty and value, household and member demographc characterstcs are also ncluded n ths data set. A total of 3,650 households were used to estmate the parameters of our food system (Table 2). The unque aspect of ths expendture survey s that households are requred to mantan detaled daly expendture dares related to the purchase of food and nonfood tems over the entre survey year. The daly dares are then summarzed by county statstcal offces and aggregate results for each expendture tem and household reported to the SSB. The annual nature of these dares s n contrast to other developng country surveys that typcally encompass 1-2 weeks of purchases (Dong, et al., 2004; Sabates, et al., 2001). The brevty of the data collecton perod n these other settng usually result n the censorng of food expendtures. Such censorng represents a sgnfcant econometrc problem when estmatng dsaggregated food demand systems (Dong, et al., 2004; Peral and Chavas, 2000; Yen, et al., 2003). Gven the annual nature of the expendture dares used here, even wth the development of our 12 food demand system, commodty censorng was not a sgnfcant problem. For most commodtes less than 5% of the households dd not report purchasng a partcular commodty group.

12 11 For our analyss we adopt commodty group defntons smlar to the categores used by Gao, et al. (1996) n ther analyss of Chnese household food purchase behavor. These food categores nclude beef/mutton (BF), pork (PK), poultry (PLT), fsh/seafood (SFD), vegetables (VEG), fruts (FRT), rce (RIC), other gran products (OGR), dary products (DA), eggs (EGG), food fats and ols (FAT), and other foods for at-home consumpton (OTH). Table 2 provdes an overvew of food purchase patterns of our sampled households. Mean annual household ncome ranges from 15,884 yuan/household n Henan provnce to 47,446 yuan n Guongdong. In spte of ths wde range there s lttle varaton n the mean share of household ncome allocated to food. The mnmum mean food share s 24.6% for households n Shandong provnce. Ths compares to 28.0% for households n Jangsu provnce. In contrast, there are some sgnfcant dfferences n the allocaton of average household food budgets across specfc food categores. For example, Guongdong, the more affluent provnce, reles the least on gran-based commodtes (7.0% of total food expendtures) whle households n Henan rely the most (17.0% of expendtures). For the 5 provnces ncluded n ths analyss, the sample mean ncome s 16.1% greater than the mean household ncome across all urban Chnese households. Agan the relatve value of ths value vares across provnce, from slghtly more than 66% for our sample of households from Henan provnce to more than 230% for our sample of urban Guangdong households. Besdes data sorted by regon of resdence, we also present food purchase characterstcs for households contaned n the hghest and lowest total food expendture decles. Not surprsngly, the most sgnfcant dfference can be seen wth respect to the mportance of FAFH as a food source. Slghtly less than one-thrd of total food expendtures by households n the hghest expendture decle s spent on FAFH. Ths compares to only 7% for households n the lowest expendture decle. More than 35% of FAH expendtures s assocated wth beef, pork or poultry for the hghest decle households. Ths compares to less than 23% for households n the lowest decle. The lowest decle households also spend relatvely more on vegetables compared to hgh decle households.

13 12 We account for household heterogenety n the estmated expendture share equatons by ncludng a set of demographc characterstcs n the translatng functons, t (p). Table 2 provdes an overvew of these characterstcs along wth the household characterstcs used n the unt value equatons. Besdes the wde range n household ncome across regon noted n Table 1, there s a parallel pattern obtaned wth respect to the percent of households ownng refrgerators and/or freezer applances (D_REFRIG) and meal planner educaton (D_ADV, D_LHS). Applcaton of the Econometrc Model to Chnese Urban Households Gven expendture share addng-up and theoretcal symmetry condtons, estmaton of the share equaton parameters was acheved by droppng one share equaton e.g., other food, from the estmaton process. Parameters for ths omtted category were recovered from these condtons. We assume that the share equaton and unt value error terms are dstrbuted multvarate normal. We use the followng log-lkelhood functon to obtan parameter estmates: * * (2 1) 1 T M T T ε 1 ln ln(2 ) ln t ε L = π Ω t Ω (26) t= 1 υ t υ t where M s the number of aggregate commodtes ncluded n the system, T the number of households n the sample, e * t s the ([M-1] x 1) expendture share error terms used n estmaton,? t s the (M x 1) unt value error terms and O s the ([2M-1] x 1) error term covarance matrx. As noted above, after expressng the share equatons n reduced form a Full Informaton Maxmum Lkelhood estmator was used wthn the GAUSS software system to obtaned parameter estmates. In order to estmate the large system ncluded n the present applcaton we assumed that the covarance of the share and unt-value equaton error terms are zero. Even wth ths assumpton we were stll requred to estmate 519 parameters whch ncluded the remanng error covarance matrx elements. Overvew of Estmated Share Equaton Coeffcents Appendx A contans a lstng of the estmated coeffcents, assocated standard errors and equaton R 2 values for the 12 expendture share equatons. Appendx B contans smlar results for the unt value equatons. The share equaton R 2 values were of reasonable sze gven our use of cross-sectonal data rangng from for pork to for other grans. It was not surprsng that the R 2 values for the unt value

14 13 equatons tended to be larger than the share equaton values gven the use of lnv as the dependent varable. The range n unt value R 2 values was from for dary products to for the frut commodty. As shown n Appendx A, there are 11 demographc-related coeffcents n each share equaton resultng n a of 132 share equaton demographc related parameters. Approxmately two-thrds of these coeffcents were statstcally sgnfcant at the 0.01 level. There appears to be sgnfcant regonal dfferences n the structure of food demand as 40 of the 48 provncal dummy varable coeffcents were statstcally sgnfcant. Compared to the mpact of regon on food demand structure we found less of an mpact of household age composton. Less than a thrd of the age related dummy varable coeffcents were found to be statstcally sgnfcant. Not surprsngly, for dary products we see that havng chldren n the household postvely mpacts the share of total expendtures allocated to dary products. Havng adults over the age of 65 n the household only mpacted expendtures on fruts (-) and rce (+). We ncluded household head age (AGE) as an explanatory varable n the share equatons to capture some of the age-related cohort effects n food choce. From Appendx A, we see that head age s negatvely related to the share of total food expendtures allocated to beef, poultry, fruts and dary products. We also fnd some evdence that household ncome mpacts food choce. The estmated coeffcents assocated wth household ncome n the share equaton were statstcally sgnfcant except n for the poultry equatons. The level of food expendture shares assocated wth beef, frut, and dary products were postvely related to household ncome. We hypotheszed that refrgerator ownershp would have a postve mpact on the share of relatve expendtures spend on commodtes consdered to be pershable. Except for the other foods category, the commodtes where a postve mpact of refrgerator ownershp was for commodty groups that could be consdered beng composed of pershable foods (e.g., beef, seafood, frut, and dary products). From Appendx A we see that a majorty of prce related coeffcents are sgnfcantly dfferent from zero (.e., 65 out of 78). The beef and poultry commodtes had the least number of statstcally sgnfcant prce coeffcents. The prce coeffcents

15 14 assocated wth the dary products (10 out of 12) and other food categores (12 out of 12) had the largest number of statstcally sgnfcant coeffcents. Overvew of the Estmated Coeffcents n the Unt Value Equatons Appendx B contans the estmated coeffcents assocated wth the unt value equatons. As noted above n (22) we use the regonal dummy varables to provde an estmated of exogenous market prces (.,.e, lnp j ). There are sgnfcant regonal dfferences n commodty prces as reflected n 40 of the 48 estmated regonal dummy coeffcents beng statstcally sgnfcant from the base regon, Shandong. Market prces n Guangdong and Jangsu provnces are tend to be hgher versus those n Henan and Helongjang provnces. We ntally hypothesze a negatve relatonshp between household sze and observed unt values reflectng a household s ablty to obtan economes of scale when purchasng larger quanttes. Dong and Gould (2000) found such an effect n ther analyss of purchases of pork and poultry purchases. We do not fnd evdence of ths n the present applcaton wth sgnfcant negatve values obtaned n the fats and ols and other food commodty groupngs. There appears to be some relatonshp between educaton level and unt value. Ths suggests that more educated households exhbted an ncreased demand for qualty, ceters parbus. Household composton appears to matter n terms of the demand for food qualty. Unt-values for 5 of the 12 commodtes are mpacted by the percentage of household members that are older than 65 years. The beef commodty shows a postve unt value relatonshp compared to negatve values obtaned n the vegetable, frut, other gran and dary product unt value equatons. All of the unt value coeffcents assocated wth the PER_LT_6 and PER_6_14 varables were postve, ndcatng ncreased demand for qualty. Agan, ths may be reflectng a cohort effect where younger households beng accustomed to purchasng better qualty food. Evaluaton of the Structure of Food Demand U Form (14) and (15) expendture (? j ) and unt value elastctes ( πj ) are calculated usng the mean values of the exogenous varables. The resultng elastcty estmates are dsplayed n Table 3. Banks, et al. (1997) note that gven the expresson for the

16 15 expendture elastctes shown n (14) a postve β coeffcent n combnaton wth a negatve λ value mples for relatvely low expendture levels, the th commodty could be consdered a luxury and a necessty for relatvely hgh expendture levels (p.534). Only the pork commodty exhbts ths structure. It was surprsng that dary products dd not fall nto ths category gven the hstorcally low levels of dary product expendtures n the past and rapd ncreases n such expendtures recently. When evaluated at these mean values of the exogenous varables we fnd that beef, pork, dary products and fats and ols have expendture elastctes sgnfcantly greater than 1.0. Gven the relatve ncrease n per capta frut consumpton n recent hstory, t s surprsng that the estmated frut expendture elastcty s sgnfcantly less than 1.0. Table 3 also contans a summary of the estmated uncompensated (Marshallan) unt value elastctes. All own unt-value elastctes were found to be sgnfcantly less than zero and all except for the other grans commodty were nelastc. Surprsngly, dary products exhbted the most own unt-value nelastc structure. In terms of the cross prce effects we found a mxture of gross complements and substtutes. Of the 132 cross untvalue elastctes, 77 were found to be statstcally dfferent from 0. The large number of sgnfcant relatonshps renforces the need for us to dsaggregate our analyss of food purchases versus the more ad hoc sngle equaton approaches. Of the statstcally sgnfcant cross unt-value elastctes all exhbt nelastc relatonshps. The relatonshp between demand for other food commodtes due to a change n unt-value of the other grans commodty shows the largest substtuton relatonshp wth a cross-prce elastcty of The next largest substtute relatonshp was between the demand for rce n reacton to changes n the egg unt value. An example of a complementary relatonshp can be seen for the mpact on other food demand resultng from the pork unt value, Dfferentaton of Unt Value and Prce Elastctes Deaton (1988, 1997) provdes the theoretcal framework for dfferentatng the effects on food choce of changes n endogenously determned unt values versus exogenous market prces. As noted above, Table 3 contans elastcty measures that ncorporate qualty effects. To solate the qualty effects of prce changes, Table 4

17 16 provdes a summary of alternatve elastcty measures used to correctly evaluate the ownprce mpacts on commodty demand. Smlar to the nterpretatons of Pras and Houthaaker (1955), Cramer (1973) and Deaton (1997) we nterpret the elastcty of unt value to a change n total expendtures lnψ j lnvj (FAH), as a measure of the qualty elastcty, e.g. η =. These lne ln FAH elastcty values are dsplayed n the frst column of Table 4. They are relatvely small but much larger than those obtaned by Myrland et al. (2003) n ther analyss of the demand for salmon (0.002) whch was based on a smlar methodology as used here but appled to sngle commodty. Dong and Gould (2000) n ther double-hurdle model of pork and poultry purchases by Mexcan households obtan smlar values as those reported n Table 4, and 0.10, respectvely. In the analyss of food demand n urban households n Pakstan ( ), Deaton (1997) obtaned a relatvely large value for hs aggregate meat category, when compared to the estmated values obtaned for our beef, pork and poultry commodtes. The second column of Table 4 repeats the value of the own unt value elastctes prevously reported n Table 3. The next 3 columns use the results of (24) to summarze our estmates of the own-prce elastctes, P jj π under alternatve expendture elastcty scenaros. Smlar the results of Deaton (1997) the relatvely small qualty elasttes result n small dfferences n U jj π versus P jj π values. The P π jj values shown under the Base column correspond to the values obtaned usng the qualty elastctes obtaned from the estmated unt value equatons. Only for the frut and other grans commodtes do we see a major dfference between the unt-value versus prce based elastcty measures. Next to the Base model we present our estmates of the own prce elastcty f the qualty elastcty had been double (2X) or trple (3X) the estmated value. Gven our estmates of P jj π we (23) to generate estmates of the relatonshp between observed unt values and unobserved market prces, e.g., π V jj lnv ln p j j. These estmates are provded n columns 7-9 of Table 4 usng the same qualty elastcty values used n the evaluaton of the prce elastctes. We would expect the followng range of values:

18 17 0< π V jj <1. The hgher the number the lower the role qualty plays n determnng observed unt values. The range n values was from for eggs to and for the other grans and frut categores, respectvely. The relatvely large value for eggs s not surprsng gven the egg commodty s composed of mostly a sngle fundamental good, chcken eggs whch s farly standardzed. In contrast, the other grans category s composed of a number of dfferent types of fundamental goods some wth very lttle valued added characterstcs such as raw small grans versus others that embody sgnfcant value added such as breads and bakery tems that are gran based. The relatvely hgh V jj π values for the meat categores and for dary products were surprsngly gven the varety of product forms and degree of processng assocated wth composte commodtes. Smlar to the analyss of alternatve qualty elastcty values. P π jj we also examne the senstvty of V π jj to Gven the defnton of unt values and the relatonshp between observed unt values, product prce and qualty noted n (22), the qualty own prce elastcty π Ψ jj Ψ exhbts the opposte pattern of the above unt-value own prce elastctes. That s, the qualty of the Other Gran and Frut commodtes responded the most to changes n unt values. We verfy the theoretcal results of Deaton (1988, 1997) by fndng that for all commodtes, a prce ncrease ceters parbus wll result not only reducton n quantty purchased but also a reducton n the qualty of the composte commodty bundle. If the qualty elastcty for the Frut commodty had been versus the actual value of 0.254, the qualty own-prce elastcty s projected to ncrease n abosolute value from to Summary Ths research utlzes a generalzaton of the Quadratc Almost Ideal Demand System (QUAIDS) to quantfy the structure of food demand of urban households n 5 Chnese provnces. We extend prevous research va the estmaton of a dsaggregated demand system whle at the same tme accountng for the endogenous decson as to product qualty. By capturng the jont nature of the quantty and qualty decsons we are able to p jj j

19 18 examne the mpact of qualty when market prces change. As predcted, ncreases n market prces result n changes n purchase patterns such that overall commodty qualty s reduced. Ths paper found that there exst statstcally sgnfcant substtuton effects between foods. The utlzaton of the QUAIDS specfcaton was mportant as a lkelhood rato test ndcated that ths specfcaton added sgnfcant explanatory power to the model versus the tradtonal AIDS specfcaton. Usng the QUAIDS specfcaton we obtan reasonable unt value and prce elastcty estmates when compared to other analyses of food demand by Chnese households. Our method for examnng the mpacts of product qualty results n a set of elastcty measures that ndcate that product qualty s relatvely nelastc wth respect to changes n market prces. The analyss presented here s based on demand characterstcs for the sample as a whole. A more detaled analyss should examne how the demand for qualty vares across households n dfferent ncome decles, expendture decles, regon of resdence, etc. We also need to provde estmated cross-prce elastctes usng the estmated cross unt-value elastctes. These values wll enable use to answer the queston as to whether the smlarty of the own prce and own unt-value elastctes carres over to cross prce effects. An extenson of the current model would be to examne the senstvty of our results to changes n the method by whch we endogenze product qualty. That s, n the estmated QUAIDS share equatons we use predcted smultaneously determned unt values nstead of unobserved commodty market prces as explanatory varables. Ths s n contrast to the methodology proposed by Deaton (1997) whereby both the system of share equatons and unt value equatons use the unobserved commodty market prces as explanatory varables. He apples ths methodology to a Lnear Approxmate AIDSbased system. Our use of the much more nonlnear QUAIDS model greatly complcates the mplementaton of Deaton s (1997) framework. In the future, we wll undertake the task of ncorporatng hs methodology wthn the QUAIDS structure.

20 19 Bblography Abdula, A., Household Demand for Food n Swtzerland: A Quadratc Almost Ideal Demand System. Swss Journal of Economcs and Statstcs. 138, Atknson, A.B., J. Gomulka, and N.H. Stern, Spendng on Alcohol: Evdence from the Famly Expendture Survey Economc Journal. 100, Banks, J., R. Blundell, and A. Lewbel, Tax Reform and Welfare Measurement: Do We Need Demand System Estmaton. Economc Journal. 106, Banks, J., R. Blundell, and A. Lewbel, Quadratc Engel Curves and Consumer Demand. Revew of Economcs and Statstcs. 79, Barten, A.P., Famly Composton, Prces and Expendture Patterns. In: Hart, P., G. Mlls, and J.K. Whtaker (eds.), 16 th Symposum of the Colston Socety, London, UK. Econometrc Analyss for Natonal Economc Plannng. Blundell, R.W., and C. Meghr, Bvarate Alternatves to the Tobt Model. Journal of Econometrcs. 34, Blundell, R. P. Pashardes, and G. Weber, What Do We Learn About Consumer Demand from Mcro Data? Amercan Economc Revew. 83, Cox, T.L.,and M.K. Wohlgenant, Prces and Qualty Effects n Cross-Sectonal Demand Analyss. Amercan Journal Agrcultural Economcs, 68, Cramer, J.S., Interacton of Income and Prce n Consumer Demand, Internatonal Economc Revew, 14(2): Deaton, A., Qualty, Quantty, and Spatal Varaton of Prce, Amercan Economc Revew, 78(3): Deaton, A., The Analyss of Household Surveys: A Mcroeconometrc Approach to Development Polcy, Johns Hopkns Unversty Press, Baltmore. Deaton, A. and J. Muellbauer, An Almost Ideal Demand System. Amercan Economc Revew. 70, D. Dong and B.W. Gould, Qualty Versus Quantty n Mexcan Household Poultry and Pork Purchases. Agrbusness. Summer: Dong, D., B.W. Gould and H.M. Kaser, Food Demand n Mexco: An Applcaton of the Amemya-Tobn Approach to the Estmaton of a Censored Food System, Amercan Journal of Agrcultural Economcs, forthcomng.

21 20 Dong, D., J.S. Shonkwhler, and O. Capps, Estmaton of Demand Functons Usng Cross-Sectonal Household Data: The Problem Revsted. Amercan Journal Agrcultural Economcs, 80, Gale, F., Chna s Growng Affluence: How Food Markets are Respondng Amber Waves, Economc Research Servce, USDA, Washngton, D.C. Gao, X M, E.J.Wales, G.L. Cramer, A Two-Stage Rural Household Demand Analyss: Mcrodata Evdence from Jangsu Provnce, Chna. Amercan Journal of Agrcultural Economcs. 78, Gould, B.W Factors Affectng U.S. Demand For Reduced-Fat Flud Mlk. Journal of Agrcultural and Resource Economcs, 21, Gould, B.W., and H.J. Vllarreal, Adult Equvalence Scales and Food Expendtures: An Applcaton to Mexcan Beef and Pork Purchases. Appled Economcs. 34, Gould, B.W., T.L. Cox, and F. Peral, Demand for Food Fats and Ols: The Role of Demographc Varables and Government Donatons. Amercan Journal of Agrcultural Economcs. 73, Greene, W., Econometrc Analyss, 5 th ed., Prentce Hall, New York, Guo, X, T.A. Mroz, B.M. Popkn, and F. Zha., Structural Change n the Impact of Income on Food Consumpton n Chna, Economc Development and Cultural Change. 48, Hausman, J.A., W.K. Newey, and J.L.Powell, Nonlnear Errors n Varables: Estmaton of Some Engel Curves. Journal of Econometrcs 65, Heen, D. and G. Pompell, The Demand for Beef Products: Cross-Secton Estmaton of Demographc and Economc Effects. Western Journal of Agrcultural Economcs. 13, Houthaker, H.S Compensated Changes n Quanttes and Qualtes Consumed. Revew of Economc Studes, 19, Hsu, H.H., W.S.Chern and F. Gale, How Wll Rsng Income Affect the Structure of Food Demand?. In: F. Gale (Ed.), Chna s Food and Agrculture: Issues for the 21 st Century, Economc Research Servce, USDA, Washngton, D.C. Lewbel, A, A Unfed Approach to Incorporatng Demographc or Other Effects

22 21 nto Demand Systems. Revew of Economc Studes. 70, Moro, D. and P. Sckoka, Heterogeneous Preferences n Household Food Consumpton n Italy. European Revew of Agrcultural Economcs. 27, Muellbauer, J., Communty Preferences and the Representatve Consumer. Econometrca. 44, Myrland, O., D. Dong and H.M. Kaser, Prce and Qualty Effects of Generc Advertsng: The Case of Norwegan Salmon, unpublshed manuscrpt, Department of Appled and Manageral Economcs, Cornell Unversty. Natonal Bureau of Statstcs, webste, Nelson, J. (1991). Qualty Varaton and Quantty Aggregaton n Consumer Demand for Food. Amercan Journal Agrcultural Economcs, 73, Perall, F., Consumpton, Demographcs and Welfare Measurement: Metrc and Polcy Applcatons to Colomba. Ph.D. Dssertaton, UW-Madson. Peral, F., and J.P. Chavas, Estmaton of Censored Demand Equatons from Large Cross- Secton Data. Amercan Journal of Agrcultural Economcs, 82(4), , Pollak, R. and T. Wales, Demographc Varables n Demand Analyss. Econometrca. 49, Pras, S.J., and H.S. Houthakker, The Analyss of Famly Budgets, Cambrdge Unversty Press, London. Sabates, R., B.W. Gould, and H.J. Vllarreal, Household Composton and Food Expendtures: A Cross Country Comparson. Food Polcy. 26, Shono, C. N. Suzuk, and H.M. Kaser, Wll Chna s Det Follow Western Dets. Agrbusness. 16, Thel, H Qualtes, Prces and Budget Inqures, Revew of Economc Studes, 19, Wang, Z., Chna and Tawan Access to the World Trade Organzaton: Implcatons for U.S. Agrculture and Trade, Agrcultural Economcs, 17, Yen, S.T., B.H.Ln and D.M. Smallwood, Quas- and Smulated Lkelhood Approaches to Censored Demand Systems: Food Consumpton by Food Stamp Recpents n the Unted States, Amercan Journal of Agrcultural Economcs, 85: Gale and Lohmar

23 Table 1. Allocaton of Food Expendtures Across Food Commodty Group, Expendture Decle and Regon Entre Lowest Hghest Sample Decle Decle Jangsu Shandong Guangdong Henan Helongjang Beef Pork Poultry Seafood/Fsh Vegetables Rce Other Gran Frut Dary Eggs Fats/Ols Other Food Total Food Expend. (Yuan) 6,356 2,043 16,476 6,426 5,071 13,083 4,128 4,435 Food-at-Home (FAH) Expend. (Yuan) 5,135 1,790 11,604 5,335 4,242 9,234 3,660 3,983 Household Income(Yuan) 23,661 10,931 56,990 22,961 20,588 47,446 15,884 16,612 Income as % of All Urban Chnese Food Expend. As % of Income FAH* as % of Income Sample Sze *Note: Durng June 2001, 1 $U.S.=8.28 Yuan. In 2001, the average urban household ncome across all Chnese provnces was 20,378 yuan (Natonal Bureau of Statstcs, 2003). The decle categores n the above table are defned relatve to total food expendtures.

24 Table 2. Summary of Household Characterstcs Used n the Demand System Entre Varable Descrpton Unt Equaton Sample HH_INC Total Household Income S 23,661 Yuan (18,546) HH_SIZE a Number of Resdent Members S,V 3.1 # (0.83) Bottom Decle 10,931 (5,795) 2.9 (0.86) Top Decle 56,990 (30,163) 3.5 (0.92) Jangsu Shandong Guangdong Henan Helongjang 22,961 (13,575) 2.9 (0.92) 20,588 (10,336) 3.1 (0.64) 47,446 (28,630) 3.3 (0.77) D_REFRIG Own Refrgerator/Freezer 0/1 V AGE Age of Household Head Year S Meal Planner Educaton Status D_ADV Completed More Than Hgh School 0/1 V D_LHS Completed Less Than Hgh School 0/1 V Household Composton D_LT_6 Chld < 6 Years Old Present 0/1 S D_6_14 Chld Between 6-14Years Old 0/1 S D_SENIOR Adult > 65 Years Old Present 0/1 S PER_LT_6 % Members <6 Yrs. Old % V PER_6_14 % Members Between 6-14 Yrs. Old % V PER_SENIOR % Members > 65 Yrs. Old % V Household Provnce D_JS Jangsu* 0/1 S,V D_SD Shandong 0/1 S,V D_GD Guangdong 0/1 S,V D_HN Henan 0/1 S,V D_HLJ Helongjang 0/1 S,V Sampled Households Note: The values n parentheses are standard devatons. The symbol * dentfes the omtted regon. The meal planner s assumed to be the female head, f present, otherwse t s the male head. The S and V symbols dentfy use of the assocated varable n the expendture share and unt value equatons respectvely. a In the share equatons, the nverse of household sze s used as an explanatory varable. In the unt value equatons, the natural logarthm of household sze s used. 15,884 (8,393) 3.2 (0.93) 16,612 (8,579) 3.0 (0.79)

25 24 Table 3. Uncompensated Unt-Value and Expendture Elastctes Quantty Change BF PK PLT SFD VEG FRT RI OG DA EG FAT OTH BF PK * PLT * SFD * VEG * FRT * RI * OG DA * EG * FAT * OTH * Exp.Elas Prce Change Note: The above elastctes were evaluated at the mean values of the exogenous varables. For the unt value elastctes, shaded values dentfy those statstcally dfferent from 0 at the 0.01 level of sgnfcance. For expendture elastctes a shaded value dentfes elastctes statstcally dfferent from 1.0. For own unt value elastctes, a * ndcates a value statstcally dfferent from -1.0 at the 0.01 sgnfcance level.

26 25 Commodty Qualty (? j ) [1] Table 4. Comparson of Varous Elastcty Measures Unt Value P U ( π jj ) [2] ( ) Own Prce V π jj Unt-Value Own Prce ( π jj ) Qualty Own Prce ( π Ψ jj ) Base 2X 3X Base 2X 3X Base 2X 3X [3] [4] [5] [6] [7] [8] [9] [10] [11] BF PK PLT SFD VEG FR RIC OGR DA EGG FAT OTH Note: The terms 2X and 3X refer to estmates based on values of the Qualty Elastcty beng 2 and 3 tmes as large as the values dsplayed n column (1).

27 Appendx A. Summary of Share Equaton Estmated Coeffcents, Assocated Standard Errors and Equaton R 2 BF PK PLT SFD VEG FR RIC Varable Coeff. S.E. Coeff. S.E. Coeff. S.E. Coeff. S.E. Coeff. S.E. Coeff. S.E. Coeff. S.E. Prce Coeffcents ( γ j ) BF PK PLT SFD VEG FR RIC Expendture Coeffcents ( β, λ ) ln(m) (ln(m)) (contnued)