1 European Dary Industry Model DEMAND FOR DAIRY PRODUCTS IN THE EUROPEAN UNION Claudo Soregarol (UNICATTT) Audrey Trévsol (INRA) Workng Paper 02 /2005
2 Demand for dary products n the European Unon Claudo Soregarol, Unverstà Cattolca del Sacro Cuore (UNICATT), Pacenza, Italy Audrey Trevsol, Unversty of Toulouse (INRA), Toulouse, France 89 th EAAE Semnar, Modellng agrcultural polces: state of the art and new challenges. 3th-5 th February 2005, Parma, Italy Ths research s developed under the EDIM (European Dary Industry Model) project supported by the European Commsson wthn the 6 th framework program for Research and Development ( Abstract The objectve of ths project s to develop complementary modellng tools able to smulate the mpact of alternatve polcy scenaros for the dary sector over the medum term. Because the analyss of the European Unon (EU) dary polcy n the future strongly depends on the evoluton of the demand for dary products and on the elastcty of demand, we have developed a specfc analyss to study demand for dary products n the European Unon. We focus on two ssues relatve to the European Unon demand for dary products. Frst, we present estmates of the autonomous trend n consumpton of dary products n the ffteen EU member countres. Second, we estmate demand prce and expendtures elastctes for dary products n France and for Italy. Keywords: demand, dary, trend, elastcty, AIDS. 1. Introducton Ths study s part of the European Dary Industry Model (EDIM) project. The objectve of ths project s to develop complementary modellng tools able to smulate the mpact of alternatve polcy scenaros for the dary sector over the medum term. A EU dary sector model has been developed by Bouamra et al. (2002). It ntegrates the whole channel of the EU dary ndustry from the supply of mlk to the demand for fnal commodtes through an ntermedate step of processng mlk nto fnal commodtes. It s an hedonc (mlk characterstcs), spatal equlbrum model whch ntegrates an agrcultural product (cow mlk), 2 mlk components (fat and proten), the dfferent member states (14) and the rest of the world, and 14 fnal dary products. The model ntegrates the EU dary polcy nstruments that nclude mlk producton quota, nterventon prces as for butter and SMP, domestc subsdes for ndustral uses of butter and SMP, a producton subsdy for casen, export subsdes and mport tarff rate quotas for each fnal dary product as well as drect payments. Fnally, GATT mport and export commtments are explctly modelled. Smulaton results from ths model are very senstve to the evoluton of domestc demand. Because, aggregate demand for dary products s rather nelastc, any change n aggregate demand generates large prce effects (gven that supply s fxed due to mlk quota). It s thus
3 very mportant to precsely analyze how demand for the dfferent dary products s lkely to change n the future. The vew behnd ths analyss s that annual consumpton of a product depends on prces and on other factors such as ncome, populaton growth, food habts or health concerns. In the followng study, we focus our attenton on the mpact of these nonprce factors that affect consumpton. Because n the exstng model assumptons on prce elastcty of demand for dary products comes from varous studes, we also estmate a demand system for dary products n France and Italy. We use an AIDS model of mult-stage budgetng allocaton. Usng the results we compute condtonal and uncondtonal elastctes for the dfferent dary products. 2. Methodology 2.1 Estmaton of consumpton trends We estmate the autonomous change n consumpton, that s excludng the changes n consumpton that are due to prce changes. We frst correct the observed data from varatons that are due to prce changes. We thus calculate the theoretcal consumpton of every past year assumng that the prce of the product s the prce observed n a gven year. We use the followng expresson to determne the adjusted consumpton: Pref Pt Ctc = Ct 1 + ε D (1) Pt wth Pt and Pref demand. Ctc the adjusted consumpton for year t, t the prce for the perod t and the reference perod, ε D C the observed consumpton for year t, the prce elastcty of Then, usng these adjusted consumpton data, we estmate a consumpton trend functon that depends on the followng possble explanatory varables: tme, GDP, populaton. Several forms of econometrc or statstcal models are used. The functonal forms used n ths study are the followng: - Lnear regresson. We estmate varous lnear regressons (usng the varables or the logarthm of varables, ncludng or not squares of the ndependent varables). - Lnear regresson usng a Box-Cox transformaton of the explaned varable: C λ ' Ct 1 t = = a x, t λ + ε t (2) - Models n dfferences, that s explanng the dfference of the explaned varable C t by the dfferences of explanatory varables. wth C t = Ct Ct 1 and x t = xt x( t 1) =1,,n C = a x, + b + ε (3) t t t
4 In the second step, we check the valdty of models (valdty of the estmated models s judged by checkng resdual propertes, model sgnfcance and sgnfcance of explanatory varables). To choose among the vald models, we use two selecton crtera: the adjusted R² and the mean squared error of the predctor. 2.1 Estmaton of prce elastcty of demand As t s now very common, we used the Almost Ideal Demand System (Deaton and Muellbauer, 1980). The general specfcaton of the AIDS model s: w n = α + γ log p + β log( Y / P) (4) j= 1 j j where: w s the budget share of the th good, α s the constant coeffcent n the th share equaton, γ j s the slope coeffcent assocated wth the j th good n the th share equaton, p j s the prce of the j th good, Y s the total expendture for goods. P s a general prce ndex. It s defned by: n n n 1 log P = α + α log p + γ j log p log p 0 j 2 = 1 = 1 j= 1 (5) As commonly used n emprcal work, P s replaced by the Stone prce ndex (P*). However, ths ndex does not satsfy the propertes of ndex numbers snce t vares wth changes n the unt of measurement of prces. As Moschn (1995) suggests, ths problem can be solved by usng a corrected Stone ndex defned as n S p log( P ) = w log( ) (6) 0 p = 1 0 where p s a base perod mean value for the th product. Hence, scalng prces by ther means s a practcal soluton to solve the problem of the unts of measurement. We use ths method n the emprcal applcaton. Moreover, the model s taken n dfferences, wth a term for the trend effect: Y w = α + γ j log( p j ) + β log( ) + η log( year) j=1,...,n (7) j P Theoretcal restrctons are mposed: addng-up, homogenety, symmetry, and negatvty. Concavty was locally mposed usng a semflexble approach of the Cholesky decomposton as proposed by Moschn (1998). For France, we estmate an AIDS model for each step of a four-stage budgetng allocaton (see Fgure 1). In the frst stage, consumer chooses between food and non food expenses.
5 Then, among food expenses, he chooses between dary products (except butter), fat products, meat and others. Because butter drectly competes wth other fat products, we desgned a fat product category. Moreover, n order to analyse the mpact of demand for meat on demands for dary product, we specfy a meat aggregate. Then n a thrd stage, we focus on the one hand on the fat system and on the other hand on the dary products. Fnally, n a fourth stage, because there are a lot of dfferent cheeses n France, we focus on the demand for cheese. Fgure 1: Utlty tree used to estmate a demand system for dary products n France For Italy, we use a smplfed utlty tree (Fgure 2). A three-stage budgetng allocaton s defned as cheese consumpton s not detaled as t was for France. However, the general deas for decomposton are smlar except the choce for butter consumpton. Here, butter s assumed to compete drectly wth other dary products. Fgure 2 - Utlty tree used to estmate a demand system for dary products n Italy
6 3. Data 3.1 Data used to compute estmates and projectons of consumpton We use annual data from Eurostat, CNIEL (Centre Natonal Interprofessonnel de l Econome Latère, France) and ZMP (Zentrale Markt- und PresberchtstelleZentrale Marktund Presberchtstelle) for total human consumpton (household consumpton + ndustral consumpton) by country. We used UNEP (Unted Natons Envronnemental Program) data and Eurostat data for Gross Domestc Product (GDP), data from the European Councl for populaton. We use prce data from INSEE (Insttut Natonal de la Statstque et des Etudes Economques, France), ZMP, UNICATT (Unversta Cattolca del Sacro Cuore, Italy), FAL (Bundesforschungsanstalt für Landwrtschaft, Germany ) and UK statstcs. When possble, data cover the perod To compute forecasts of consumpton, we assume varous growth rates of GDP. In ths paper, we present the results of consumpton forecastng for a GDP growth rate equal to 1% per year from 2004 to For populaton, we use forecasts computed by Unted Natons and Eurostat (avalable for each country). 3.2 Data used to compute elastcty estmates. For France, we use data about quantttes and values of dary products purchased n France from INSEE (for the 1 st, 2 nd and the fat products 3 rd stage) and SECODIP (for the dary products 3 rd stage and for the 4 th stage). Data from INSEE are annual data and cover the perod Data from SECODIP are four-week perod and cover the perod For Italy, we use data from Natonal Accountng Data from ISTAT (for the frst and second stages) and retal data from NIELSEN (for the 3 rd stage). Data cover respectvely the perods (annual data) and (by semester). 4. Results 4.1 Estmates of consumpton trends The EU dary model that was developed consders 14 dfferent dary products (butter, flud mlk, soft cheese, hard cheese, sem-hard cheese, processed cheese, blue cheese, fresh cheese, cream, fresh dary products, whole mlk powder, skmmed mlk powder, condensed mlk and casen). When possble, that s when tme seres data are avalable at ths level of dsagregaton, we estmate consumpton trends for each of the 13 fnal consumpton dary products (all except casen). Moreover, because the model represents the dary ndustry n each EU member states, we also estmate trends at the EU member state level. More precsely, we estmate trends for the 4 man dary consumer countres (Germany, France, Italy and the Unted Kngdom) and for the group of the other 11 EU member countres. In table 1, we provde the man results. We provde results for three countres plus the trend calculated at the EU-15 level. EU-15 trend s calculated usng the trends estmated for the dfferent member states. It s nterestng to note that butter and flud mlk consumpton are lkely to decrease at the EU-15 level. On the contrary, the demand for the other dary products s lkely to ncrease. Some ncreases n consumpton are rather hgh (cheese, fresh dary products). These reflect past evoluton. Obvously, t s questonable to know whether past
7 trends wll contnue. However, the estmated models always ntegrate the possblty of slow down n the consumpton (for example usng polynomal expressons). It should also be stressed that these results correspond to the total consumpton of dary products. That s the drect consumpton from consumers at home and outsde home plus the ndrect consumpton va the agro-food system. Table 1: Estmated annual average growth rate of dary product consumpton between 2000 and 2010 Product EU-15 Germany France Italy Butter -1.4% -1.8% -0.4% +0.7% Cheese +1.7% +1.4% +0.7% +1.2% Processed cheese +1.4% +0.4% +1.9% +1.5% Flud mlk -0.6% -1.1% -0.2% -1.4% Cream +1.3% +0.1% +2.1% +2.1% Fresh dary products +2.4% +1.0% +2.3% +3.7% Cheese = all knds of cheese except processed cheese Wth respect to mlk powders and condensed mlk, consumptons greatly vary from year to year and t s dffcult to estmate trends. We estmate that the annual average growth rate of consumpton for whole mlk powder s +1.7%. For condensed mlk, t s +2.3%. For skmmed mlk powder, t was not possble to estmate a trend as the evoluton of consumpton s not explaned by any of the varables we have. It seems that skm mlk powder can be substtuted by whey powder and thus the consumpton hghly depends on the relatve prce between these two products. Gven these dfferent trends of consumpton, usng the composton of the dfferent products n fat and proten, we compute an estmated annual average growth rate of fat and proten consumpton. We fnd that both fat and proten consumpton wll ncrease. It s estmated that EU consumpton of fat (from mlk) wll ncrease by 0.5% a year and that EU consumpton of proten (from mlk) wll ncrease by 1.1% a year. 4.2 Estmates of prce demand elastctes In the followng, we report our man fndngs wth respect to own and cross prce elastctes of dary products n France and Italy. We computed both the condtonal and the uncondtonal uncompensated elastctes. The several stage budgetng allocaton mples weak separablty of consumer preferences. The practcal mplcaton of ths assumpton s that estmaton s more smple and parsmonous n degree of freedom: n fact, each step can be estmated ndependently. However, the elastctes drectly estmated usng demand systems are condtonal to the assumed separablty tree and therefore to the assgned group of expendture. Whether condtonal or uncondtonal elastctes should be more relevant depends on the specfc objectve of the emprcal applcaton. For the polcy objectves of the EDIM project, uncondtonal elastctes are more relevant snce they better measure the detaled reacton of dary consumpton to a change n macroeconmc varables such as GDP and polcy measures. Therefore, to compute uncondtonal elastctes from the condtonal counterpart we adopted the method suggested by Edgerton (1997).
8 France In Tables 2 and 3, we present the condtonal and uncondtonal uncompensated prce elastctes of dary products. Uncondtonal elastctes are lower than condtonal ones. All expendtures elastctes are hghly sgnfcant and of the same order of magntude (around 0.6). Own prce elastctes are generally sgnfcant (except for cream) and as t s frequently found n the lterature demand for dary products s nelastc. However, we found a larger value for the demand for flud mlk. Ths result was not expected as t s generally thought that demand for flud mlk s very nelastc. Only a small number of cross prce elastctes are sgnfcant. Wth respect to uncondtonal elastctes, only one cross-prce relaton s sgnfcant: the one between cheese and cream whch reveals a complemantarty between these two products. Table 2: Condtonal uncompensated elastces for dary products Product Fresh Products Cream Cheese Flud Mlk Expendture Fresh Products *** * *** Cream *** *** Cheese ** *** *** *** Flud Mlk *** 1.027*** Fresh products (yoghurts + dary desserts + fresh cheese). *** sgnfcant at 1%, ** sgnfcant at 5%, * sgnfcant at 10% Table 3: Uncondtonal uncompensated elastctes for dary products Fresh Product Products Cream Cheese Flud Mlk Expendture Fresh Products ** *** Cream * *** Cheese * ** *** Flud Mlk *** 0.620*** Fresh products (yoghurts + dary desserts + fresh cheese). *** sgnfcant at 1%, ** sgnfcant at 5%, * sgnfcant at 10% Table 4 provdes the results wth respect to fat products. The expendture elastcty of demand for butter s hghly sgnfcant and of the same order of magntude as the ones obtaned for the other dary products. The prce elastcty of butter s weakly sgnfcant and consumpton s nelastc. Fnally, we do not get a sgnfcant cross-prce effect between butter and ols and margarne. Table 4: Uncondtonal uncompensated elastctes for fat products Product Butter Ols&Margarne Expendture Butter * *** Ols&Margarne ** 0.466*** *** sgnfcant at 1%, ** sgnfcant at 5%, * sgnfcant at 10% Fnally, Table 5 provdes the results for cheese whch are derved from the fourth stage. Agan, expendture elastctes are hghly sgnfcant and vary between 0.43 and Own prce elastctes are all negatve and they are hgh (n absolute term) for some cheese. We also fnd some substtuablty between the dfferent cheese categores (soft cheese and sem-hard cheese, fresh cheese and hard cheese, processed cheese and hard cheese).
9 Table 5: Uncondtonal uncompensated elastctes for the dfferent cheese categores Sem- Product Soft Hard Hard Blue Fresh Processed Expendture - Soft 0.586*** *** *** Hard *** ** 0.139*** 0.688*** Sem-Hard 0.527*** *** *** Blue *** ** 0.752*** Fresh ** *** *** Processed *** ** ** 0.428*** *** sgnfcant at 1%, ** sgnfcant at 5%, * sgnfcant at 10% Italy In Tables 6 and 7, we present the condtonal and uncondtonal uncompensated prce elastctes for dary products n Italy. Table 6: Condtonnal uncompensated elastces for dary products n Italy Lqud mlk Fresh dary PDO cheese Other cheese Butter EXP Lqud mlk *** *** *** 0.205*** 0.874*** Fresh dary *** *** *** 1.036*** PDO cheese *** *** *** *** *** 1.204*** Other cheese *** *** *** *** Butter 0.595*** *** *** *** 0.938*** Table 7: Uncondtonnal uncompensated elastctes for dary products n Italy Lqud mlk Fresh dary PDO cheese Other cheese Butter EXP Lqud mlk *** *** 0.628*** Fresh dary *** *** 0.347*** *** 0.744*** PDO cheese *** *** Other cheese *** *** *** Butter 0.242*** -0.22*** *** 0.674*** As n the case of France, the absolute value of the drect prce elastcty for lqud mlk s hgh compared to a pror expectatons. Qute nelastc appear to be the prce elastcty for PDO cheeses that also show the hghest total expendture elastcty. Among the other results, more nvestgaton s needed to explan the relatvely hgh substtutablty between lqud mlk and butter. Fnally, expendture elastctes are of the same order of magntude as compared to the results for France. 5. Conclusons Snce the results of the dary model, that ams at analyzng the mpact of the CAP reform, are very senstve to the assumptons on demand trend and snce trends may be dfferent among EU member countres, t was mportant to estmate these demand trends. Accordng to these estmates, consumpton trends of dary products are dfferent accordng to the product consdered. Consumpton of basc products such as butter and flud mlk wll decrease
10 whereas products such as cheese and fresh dary products wll ncrease. On the whole, t s expected that demand for proten ncreases more than the demand for fat n the future. Snce the results of any analyss of dary polcy reforms depend on the assumptons on prce elastcty of dary products, we estmate a demand system n France and Italy. We generally fnd that demand s rather prce nelastc and that expendture elastctes are sgnfcant and vary between 0.4 and 0.7. However, the absolute value of prce elastctes s sometme hgher than expected. It s specally the case for the demand for flud mlk both n France and Italy. For France, we also show that aggregate demand for cheese s relatvely prce nelastc. However, the demand of a specfc cheese s more prce elastc as substtuton between cheese s frequently sgnfcant. 6. Reference lst Bouamra-Mechemache Z., J.P. Chavas, T. Cox, V. Réqullart, 2002, EU dary polcy reform and future WTO negotatons: a spatal equlbrum analyss, Journal of Agrcultural Economcs. 53(2):4-29. Deaton, A., J. Muellbauer. Economcs and consumer behavor. Cambrdge Unversty Press (1980). Edgerton, D.L.. Weak Separablty and the Estmaton of Elastctes n Multstage Demand Systems. Amercan Journal of Agrcultural Economcs 79: (1997). Moschn, G.. Unts of measurement and the Stone ndex n demand system estmaton. Amercan Journal of Agrcultural Economcs 77: (1995). Moschn, G.. The Semflexble Almost Ideal Demand System. European Economc Revew, 42 (2): (1998).