Econometric Input-output model for EU. countries based on Supply & Use tables: private consumption

Transcription

1 Econometrc Input-output model for EU countres based on Supply & Use tables: prvate consumpton Kratena, K. a, Mongell, I. b and Wueger, M. a a Österrechsches Insttut für Wrtschaftsforschung WIFO, Postfach 91 A-1103 Wen, AUSTRIA b European Commsson, Jont Research Centre, Insttute for Prospectve Technologcal Studes (IPTS) Edfco Expo, c/ Inca Garclaso 3, Sevlle, SPAIN Ths paper presents the results of an econometrc study on household prvate consumpton n EU15 amng at the estmaton of a household demand system ncludng prce and ncome parameters as well as other soco demographc explanatory varables. The econometrc model s estmated by consstently combnng a tme seres cross secton dataset of aggregated household expendture data for EU15 countres, publshed by Eurostat as part of the natonal accounts, and a cross secton dataset of household budget surveys for fve European countres. The estmated household demand system of equatons s ntegrated nto an Econometrc Input-Output (EIO) model based on Eurostat Supply and Use table, recently proposed by Kratena and Strecher (2009). That gves a sound representaton of the European household budget allocaton behavour n 1

2 response to change n commodty prce, total avalable ncome and household characterstcs and ts mpacts on output and employment n Europe. A set of polces and soco-demographc scenaros s analysed through the resultng EIO-modellng framework wth a demonstratve purpose. KEYWORDS: Household demand system, Mcro and Macro data, Econometrc Inputoutput model Correspondng author: I. Mongell, E-mal address: Ignazo.Mongell@ec.europa.eu, Tel.: , Fax: The vews expressed n ths paper belong to the authors and should not be attrbuted to the European Commsson or ts servces. 2

3 1. Introducton The Input-Output (IO) model has been the frst modellng framework based on a general equlbrum concept and wth a hgh ndustry detal to be mplemented for economc analyss and forecast. In the last two decades, IO tools have been extensvely used for envronmental analyss as a man tool or as a complement of other bottom up modellng framework (LCA, partal equlbrum, etc.). Stll n the era of fully-fledged Computable General Equlbrum (CGE) models, they are often preferred for the analyss of those scenaros (.e. short term, etc.) where consumers' preferences, combnaton of factors of producton or nternatonal trade patterns are not expected to play a relevant role and the stylzed representaton used for smple IO model s suffcent. However, whenever the level of complexty of the scenaro becomes hgher CGE models are used nstead, whch wth few specfc exceptons can anyway gve only comparatve statc type of results neglectng the tme dmenson and wth t the path through whch an economc system adjusts and acheves new steady state equlbrum. An alternatve both to smple IO models and CGE ones s the Econometrc IO model (EIO), constructed startng from smple IO model and ntegratng econometrcally estmated blocks such as a households demand system for a more realstc consumers' preferences representaton, a producton block allowng for factors substtuton and a module for trade representaton. Interestng overvews and comparsons of the three dfferent models brefly dscussed so far IO, CGE and EIO are gven already by West (1995) as well as Kratena and Strecher (2009) n ther papers, hence the scope of ths study s to present an econometrc analyss of the household prvate consumpton behavour n EU15 and to llustrate ts ntegraton n EIO modellng framework. A complete system of demand equaton s estmated by consstently combnng a tme seres and a cross secton dataset derved by respectvely the aggregated household 3

4 expendture data for EU15 countres, publshed as part of the Natonal Accounts, and a cross secton dataset based on the household budget surveys of fve European countres. The estmated household demand system ncludes prce and ncome elastcty as well as the nfluence on households' consumpton of other soco demographc characterstcs (household sze, age of the reference person or number of owned cars). The emphass of modellng s on consumpton of energy (.e. electrcty, heatng and prvate transportaton) whch s explaned as a functon of the 'servce prce', thereby measurng the households' response not only to prce changes of the consumpton tem tself, say euro for kwh of electrcty for nstance, but also to the effcency of the electrc household applances. The household demand system s also complemented wth an aggregate households' consumpton functon explanng the level of consumpton n terms of dsposable ncome. The demand system and the aggregate consumpton functon are then ntegrated n an Input-Output model based on Supply and Use tables, whch consttutes a frst step towards the setup of a complete EIO model. The ntegraton of household demand systems nto EIO or CGE models, usng both tme seres and cross secton data s descrbed n Jorgenson (1982), Bardazz and Barnabn (2001), Kennes (1983) as well as Labandera and Labeaga (1999). Smlarly to the exstng studes, our approach s based on establshed mcroeconomc theory of demand systems and applcaton to tme seres and cross secton data, and strves at overcome the often crtcsed aspect of the 'over-restrctve' structure of demand systems based on flexble functonal forms (as n Almon (1996)) and on the assumpton of the 'representatve household' by combnng economc varables (ncome, prces) wth household characterstcs. 4

5 The emphass s on energy and on the envronmental mpact of households that consttutes a large part of the overall envronmental mpact of an economc system (Hertwch, 2008). Startng from these conclusons n the last years more and more polcy ntatves for sustanablty have mproved ther effectveness by encompassng measures lke ncentves or ecolabel to steer households' fnal consumpton towards a more envronmental frendly choce. These measures have actually mproved energy effcency of household applances, whch mpled a more productve use of energy for nstance (.e. less energy used per unt of fnal consumpton). Nevertheless, the envronmental loads assocated to fnal consumpton have ncreased as a consequence of hgher lvng standard, growng populaton and mght have ncreased also for a 'rebound effect' due to hgher effcency of the household devces. The rebound effect also known as 'Jevons paradox', from the name of the frst author that addressed ths ssue n 1865 (Jevons 1865), can undermne the efforts amng at reducng resources consumpton. Increased effcency of applances may ndeed dscourage 'cost-savng' behavour due to a lower prce of the resource nput and can also produce an ndrect scale effect on consumpton as a lower expendture on resource nputs makes avalable ncome for the purchase of addtonal electrcal and electronc equpments (Khazzoom 1980, Green 1992, Kratena and Wuger 2005, Greenng et al. 2000, Hertwch 2008). In ths paper the consumpton block ncludng the households' demand system plus the aggregate consumpton functon s coupled to an IO modellng framework based on the Supply-Use tables publshed for Denmark and the year The resultng EIO modellng framework s tested by analysng a scenaro called ncludng electrc effcency ncrease of both households' devces and ncrease of the share of households wth a reference person unemployed. 5

6 The paper s structured as follows secton two llustrate the theoretcal dervaton of the household demand system whch s then econometrcally estmated. Secton three ncludes the econometrc results of the tme seres and of the cross secton households' demand models and llustrates the procedure adopted to consstently combne the results of the two econometrc estmatons. Secton four sketches the aggregate consumpton functon. The ffth secton s dedcated to the scenaro analyss, whle the sxth to the conclusons. 2. Household demand system dervaton The tme seres model wll be set up for the sample of EU15 from 1995 on and a panelestmaton over countres wll be carred out. The structure of our model dstngushes between aggregate household consumpton, captal expendture of households, and expendture for heatng and transport energy as well as for other goods and servces. In prncple the consumers' decsons can be descrbed by utlty maxmzaton under constrants or by cost/expendture mnmzaton for a gven level of utlty (the dual model). In the followng a dual model of prvate consumpton s appled startng from the expendture functon of a demand system. The level of utlty u and the vector of commodty prces p are the arguments of an expendture functon for non-durables C(u, p ) whch together wth expendture for durables (nvestment I n applances wth prce ndex p I ) gves total expendture G: G C ( u, p p I (1) ) I Total expendture G could further be descrbed as a functon of dsposable ncome. For a gven savngs rate and a gven dsposable ncome, an ncrease n expendture for nvestment leads to lower expendture for non-durables C(u, p ). C( u, p G p I (2) ) I 6

7 In order to take these lnks nto account, a full model of prvate consumpton must be set up as n Kratena and Wüger (2008). Such a model must also nclude nvestment functons for durables and requres a dynamc cost mnmzaton or utlty maxmzaton model. Wllett and Naghshpour (1987) set up a model of dynamc utlty maxmzaton wth budget constrants from whch the optmalty condtons for nvestment are derved. In the present approach the consumer chooses a tme path of captal expendture, K, to mnmze dscounted costs for a gven level of utlty over a tme horzon for whch values for the exogenous varables are gven. We can derve two man optmalty condtons from such a cost mnmzaton problem, namely Shephard's Lemma (3) and the envelope condton for the captal stock (4): C ( u, p ) p x (3) C( u, p ) K ( r ) p I (4) Shephard's Lemma determnes the level of commodty demand x or n a logarthmc model the budget shares w accordng to: log C( u, p ) log p C( u, p ) p p C( u, p ) x p C( u, p ) w. The envelope condton states that the shadow prce of fxed assets must equal the user costs of captal,.e. the margnal beneft of a unt of captal must equal ts margnal cost. The shadow prce of captal s gven by the negatve of the term that measures the mpact of captal nputs on expendture. Energy commodtes are used by consumers for the 'producton' of servces (heatng, lghtng, communcaton, transport). These servces are demanded by households and requre nputs of energy flows, E and a certan captal stock, K. The man characterstc of ths stock s the effcency of convertng an energy flow nto a level of servce: 7

8 S E (5) ES In (5) E s the energy demand for a certan fuel and S s the demand for a servce nversely lnked by the effcency parameter ( ES ) of convertng the correspondng fuel nto a certan servce. For a gven converson effcency that allows to derve a servce prce p S (margnal cost of servce), whch s nfluenced by the energy prce and the converson effcency: p E p S (6) ES Ths s smlar to Khazzooms (1980, 1989) approach of dealng wth servces and shows the same property of a servce prce decrease wth an ncrease n effcency. These prces of servces (p S ) become arguments of the vector of commodty prces n the overall consumpton model (p ). The budget shares of energy demand can be defned as the tradtonal energy cost share or as the 'servce share': p E E C p S S C. We proceed by applyng the cost functon of the AIDS model (Deaton, Muellbauer (1980)) C(u, p ): log C ( u, p ) (1 u)log( a( p )) ulog( b( p )) (7) wth the translog prce ndex for a(p ): log a( p ) log p 0.5 log p log p approxmated n our case by the 0 k k k k j j k j Stone prce ndex: log P * w k log p k, the Cobb-Douglas prce ndex for b(p ): k 8

9 k log p and the level of utlty, u. As the level of utlty u s b( p ) log a( p ) 0 k k an argument of the expendture functon, an ndrect utlty functon can be derved: U log C( u, p) 0 k log a( p ) p k k (8) Applyng Shephard's Lemma to the cost functon (7), nsertng the ndrect utlty functon (8) and allowng for addtonal technologcal and soco-demographc factors captured n the vector of varables Z and D, gves the well known budget share equatons for the non-durable goods: w j j C C log p log Dlog Z j D (9) P P Note that ths formulaton allows for the pure nfluence of soco-demographc varables (Z) as well as for nteracton between soco-demographc factors and expendture (D). If the model s set up wthout any soco-demographc varables t reduces to: w T T j C P T T T T log p log j j T (10) U The followng expressons for ncome ( ) and uncompensated prce elastctes ( ) j wthn AIDS can be derved (Green and Alston, 1992): w D D 1 (11) U j j ( w D D)w j j (12) 9

10 Va the Slutsky equaton the followng general relatonshp holds between the compensated ( K ) and the uncompensated elastcty ( j U K U ): j j j j w. The compensated elastcty measures the pure prce effect and assumes that the household s compensated for the ncome effect of a prce change. Applyng the Slutsky equaton n the case of AIDS yelds for the compensated elastcty: K j j ( w D D)w j j w j (13) In (12) and (13) j s the Kronecker delta wth j = 0 for j and j = 1 for =j. The demand for energy-commodty E s determned by the level of servce demand S and energy effcency for the applance usng ths energy carrer ( ) as well as energy effcency for the other applances ( j ). Energy effcency for a dfferent aaapplance ( j ) has an mpact on energy demand for good due to cross prce effects, whch s a specal feature of our model of total household consumpton. We analyse the cross prce effects on a parwse base between the energy goods n our model. By totally dfferentatng the quantty demanded E (S, j) wth respect to t gves: de dt E j d dt j E S ds dt (14) In (14) the total change n E s descrbed as the sum of drect effects of effcency changes and of ndrect effects va servce demand. The drect effects of an effcency ncrease on energy demand (the frst term n (14)) s equal to -1. But an ncrease n effcency also leads to a decrease n the servce prce and thereby to an ncrease n servce demand. Dvdng both sdes of (14) by E rearrangng and takng nto account the prce elastcty of demand for energy servces ( j ) gves: 10

11 d log E d log j 1 j (15) Ths expresson s dentcal wth expressons of the total effect of effcency on energy demand ncludng the rebound effect derved by Berkhout, et.al. (2000) and Khazzoom (1980). The total mpact s therefore also determned by the own prce elastcty of energy demand or, more precsely, the (servce) prce elastcty of servce demand. Actually n our model energy commodtes enter as servce (wth correspondng servce prces) and therefore we can drectly derve servce prce elastctes. It mght be seen as an mportant advantage of a model for total household consumpton that dfferent feedbacks between dfferent energy commodtes can be analyzed. That gves a number of dfferent rebound effects,.e. effects of changes n the effcency of a certan applances on the dfferent energy demands. A change n the effcency of an applance mples an own prce-rebound effect on ths energy commodty, defned by the compensated own prce elastcty C. Besdes ths pure prce nduced effect there exsts also an ncome nduced rebound effect, defned by the dfference between the uncompensated and compensated prce elastcty: U C w. The same holds true for the mpact of the change n the effcency of an applance on the demand for another energy good. The pure prce nduced effect s agan gven by compensated cross prce elastcty C and the ncome nduced effect by the dfference j of the elastctes U C w. j j j 3. Econometrc estmaton: combnng cross secton wth tme seres nformaton The demand system descrbed n the prevous secton has been derved by consstently combnng two dfferent households' demand systems estmated wth two dfferent 11

12 datasets: a cross secton tme seres dataset for EU15 ncludng ncome, prce ndexes and budget shares for some consumpton categores, and a cross secton dataset constructed by consstently mergng the households survey of four European countres (Span, Italy, France and Austra). The reason for combnng such dfferent datasets n an econometrc study s that both contan useful and complementary nformaton. Tme seres data provde wth tme seres of prce ndexes by countres whch permt the estmaton and dervaton of accurate prce elastctes, but they are rather poor on socodemographc varables as these varables are normally not avalable as a tme seres, hence a households' demand system ncludng soco-demographc varables would be hardly possble to estmate just relyng on aggregated data. On the other hand households' surveys are the rchest source of data concernng fnal consumpton by households, ther lvng condton and soco-demographc status, but they do not nclude prce data. The combnaton of tme seres and cross secton data to estmate a household demand system has been already proposed n few other studes (Bardazz and Barnabn, 2001; Nchele aqnd Robn, 1995), whch nevertheless used data for a sngle country. Ths study goes n the same drecton and proposes a method that allows the combnaton of tme seres cross secton econometrc estmates to set up a household demand system for EU15 countres wth all the desrable characterstcs such as prce and ncome elastctes as well as soco demographc nfluence. The followng subsectons 3.1, 3.2 llustrate the data and econometrc methods used for the estmaton of the two separate models. Secton 3.3 explans the way the two models are combned. 3.1 Tme seres model and estmated coeffcents 12

13 The econometrc estmaton of the tme seres model uses data on consumpton expendture from Natonal Accounts for EU-15 and data on the effcency of the stocks of energy consumng durables of households, ncludng prvate cars, electrcty usng applances and heatng applances. A specal feature of ths model s the dervaton of a servce prce (margnal cost of servce), whch s defned by the relaton of the energy prce to converson effcency for a certan fuel. We treat ths converson effcency as emboded n the stock of captal goods and applances. Ths approach would, n a further step, allow us to drectly lnk converson effcency to the path of captal accumulaton resultng n a comprehensve descrpton of emboded technologcal change. The data on converson effcency comprse effcency ndces of captal stocks for major energy-usng applances, dfferentated by heatng and electrcty. For electrcal applances,.e. only electrcty usng applances, we use data for refrgerators, freezers, washng machnes, dsh washers, TVs and dryers. For heatng, water heatng and cookng we drectly use the aggregate effcency ndces for households. The man data source on specfc energy consumpton of these captal stocks s the ODYSSEE database ( for the hstorcal sample from 1990 to The ODYSSEE database s the result of a project on "energy effcency ndcators n Europe" comprsng n total the EU 27 members plus Norway and Croata. We use the varable 'specfc consumpton' from the ODYSSEE database, whch s defned as a hypothetcal energy consumpton gven by the technologcal characterstcs of the applance and some base year unt consumpton. In order to calculate an aggregate effcency ndex for all electrcal household applances we derve a weghted average effcency ndex accordng to the share of each applance n total electrcty consumpton. Ths share was taken from the varable unt consumpton of electrcty by 13

14 applances, also contaned n the ODYSSEE database. A major problem n the constructon of our data set was fllng the gaps for certan countres concernng specfc applances. Here we used a country groupng methodology, so that mssng data were flled by takng over the data of a representatve country n the same group (e.g.: data from UK for Ireland or data from Span for Greece). In the area of energy for heatng several prmary energy carrers are affected next to electrcty ths s manly gas, ol, coal and dstrct heatng. In total, the effcency ndex for household heatng (the techncal ODEX ndex) comprses elements of effcency n the heatng equpment as well as n the outer shell of the buldng, ncludng data on specfc energy consumpton from sngle famly houses and mult famly flats. Therefore we drectly used ths aggregate ndex as the varable for effcency n heatng. For effcency n prvate transport we calculated the average consumpton per vehclekm of the prvate car fleet n EU 15 countres, basng ourselves on the results of the TREMOVE-project (documented at: We drectly used the data on vehcle-km drven together wth the energy consumpton of prvate cars to calculate the average fleet consumpton, the nverse of whch s our measure for effcency. In the logcs of our model the drect nfluence of effcency mprovements can be seen from the devaton of servce prces from energy prces. As wll be explaned below ths dataset has been complemented by data sets on household characterstcs that have also been used n the cross-secton model. One data set of household characterstcs ntroduced n the cross secton model represents dummy varables at the level of ndvdual households. In the tme seres model these characterstcs have been transformed nto aggregate varables of shares of households wth certan characterstcs wthn total households. Ths has been done for a common 14

15 subset of household characterstcs data that s avalable both n the cross-secton dataset (consumer surveys) and n the tme seres data set (from EUROSTAT). Due to lmtatons n the latter dataset the cross-secton estmaton comprses more household characterstcs than the tme seres model. The other data set of household characterstcs conssts of contnuous varables (number of cars per household, persons per household) that are avalable at the ndvdual household level (consumer surveys) as well as at the country level for the perod The followng tables dsplay the results of our panel-regresson of the tme seres model wth data for EU15 from 1995 to We estmate the demand system derved from the AIDS model as a panel wth fxed country-effects and applyng the SUR estmator. The contnuous varables comprsed n the vector D, that enter the nteracton term between household characterstcs and expendture are: () the stock of cars per household, and () the household sze (persons per household). The soco-demographc varables captured n the vector Z comprse: () the populaton structure of professonal actvtes (employed, unemployed, other, unknown), and () the age structure of populaton. In secton 3.2 we descrbe these varables as found n the cross-secton dataset. TABLE 1 As has been mentoned above the soco-demographc varables n the vector Z are defned as shares of households wth the correspondng characterstc wthn the total of all households. A shft n the composton of the household structure compared to some base year therefore changes the expendture pattern of households. Ths s equvalent to the treatng of these varables as dummes n the cross-secton analyss (see below), where the expendture pattern of a household wth a certan characterstc ceters parbus (.e. for same expendture level) dffers from another household wth dfferent 15

16 characterstcs. In the tme seres model the parameters lnked to Z therefore measure the dfference brought about by the dfference n the household structure compared to some 'base case'. The results for these estmatons are also shown n terms of own and cross prce elastctes and ncome elastctes. We can use the uncompensated prce elastcty as a drect measure of the rebound effect of energy effcency mprovements. Accordng to our result ths would gve a rebound effect for gasolne/desel (automotve fuels) between 56% and 71%, for heatng fuels between 50% and 63% and for electrcty between 24% and 47%. Comparng these results wth other studes referred n the surveys of Greenng, Greene (1997) and Greenng, et.al. (2000) they can be characterzed as lyng at the upper bound of the range found n the lterature. For heatng (ncludng water heatng) rebound effects found n the lterature are between 10% and 30% (Greenng, et.al., 2000). They are slghtly hgher for coolng and lower for prvate car transport. Therefore, the rebound effect for prvate car transport dentfed here s sgnfcantly above the results found n the lterature. The rebound effect of 24% for electrcty reflects the average results n the lterature. The compensated cross prce elastctes between the energy commodtes have a postve sgn n all four models, ndcatng a substtutve relatonshp wth the excepton of the cross prce elastctes between gasolne/desel and electrcty, whch show a negatve sgn. Changes n effcency lead to changes n the prce system and therefore to demand reactons n all energy categores. 3.2 Cross secton: model, data and estmated coeffcents For Engel curve s ntended a functon descrbng how a consumer allocates the overall avalable budget on some goods or servces holdng prces fxed. Engel curve can be also 16

17 defned as Marshallan demand functons holdng the prces of all goods fxed. A general form of an Engel curve s: q g ( Y, z) (16) where q s the quantty consumed of good or servce, Y s total avalable wealth, usually proxed by total expendtures on goods and servces, and z s a vector or matrx of other characterstcs of the household, such as age and household sze and composton. Engel curves are commonly expressed n the budget share form: w f ( Y, z) (17) where w s the part of Y spent on good. The goods are typcally aggregate commodtes such as total food, clothes, or transportaton. In ths study the analysed consumpton categores of non durables are: food wth alcohol and tobacco, clothes and footwear, operaton of prvate transportaton, fuels for heatng, coolng and cookng, electrcty and others. As shown n (16) a comprehensve system of Engel curves expresses household's budget allocaton for consumpton purposes not only as a pure functon of total avalable wealth but also as a functon of other relevant household's characterstcs nfluencng quantty and type of purchased goods and servces. The addtonal varables cover a set of quanttatve and qualtatve soco-demographc characterstcs ntroduced n the model respectvely as dummy varables, so as an ntercept shft, or as slope shft or nteracton term wth the ncome parameters. The functonal form selected for ths estmaton s as shown n (18). Here we swtched from the AIDS model used for the 17

18 tme seres model to the specfcaton of the quadratc AIDS model proposed by Banks, Blundell and Lewbel n w k k ( * n n )log( Y) ( * n n )[log( C)] 2 (18) Where w stands for the consumpton share of the th consumpton category, α s the constant for the th consumpton category, δ s a set of dummy varables that capture the effect of k specfc demographc and socal varables ncluded n the analyss as ntercept shft. The β's are the coeffcents for the lnear term of the total expendture C and the λ's are the coeffcents for the squared ncome term log[c] 2. ε s the error term. The terms β * φ and λ * φ capture the effects of a set of n household's characterstcs ntroduced n ths case as a slope shft for each of the analysed consumpton categores. The household characterstcs are ntroduced both n ths study ether as contnuous or dscrete varables. The qualtatve are ntroduced as a set of k dummy varables, whch are namely: a) professonal actvty of the reference person' classfed n three aggregated categores: 'employed' (d _prof1), 'unemployed' (d_prof2), 'other' (d_prof3) and 'unknown' (d_prof4); b) level of educaton of the reference person' classfed n the followng three categores: 'PhD or unversty degree' (d_edu1), 'secondary educaton' (d_edu2) and 'prmary educaton or none' (d_edu3); c) age of the reference person' classfed n the followng three groups: 'from 0 to 44 years old' (d_age1), 'from 45 to 59 years old' (d_age2) and 'over 60' (d_age3); d) dummy varable for each of the country ncluded n the database: Italy (d_t), Span (d_es), Austra (d_at) and France (d_fr); 18

19 e) set of dummy varables for ntervals of year of the constructon of the buldng: older than 1946 (d_cons1), between 1947 and 1960 (d_cons2), between 1961 and 1980 (d_cons3), between 1981 and 1995 (d_cons4), beyond 1996 (d_cons5), non declared (d_cons6). The other varables whch are avalable n the contnuous form are ntroduced n the model as a slope shft. These varables are 'household sze' (l_hs and l_hs2) and 'number of owned cars' (l_car and l_car2) whch s lkely to nfluence the household's consumpton of 'prvate transportaton' and 'others' whch ncludes the publc transportaton and the nsurance servces. As the model n (18) s for budget shares whch sum up to one by constructon, the addtvty condton ensurng that the sum of the shares resultng from a change n one of the explanatory varable of the model s always one s automatcally fulflled wthout the need of addtonal regresson constrants. The addtvty condtons are as follows: k * k * k (19) The dataset s constructed by usng the household budget survey for the year 2004 of the followng 3 countres: Span (total number of observatons 8881), Austra (total number of observatons s 7349), Italy (total number of observatons s 24853) and France (total number of observatons s 10240). The surveys requred some adjustments to be used together consstently. For nstance, the consumpton expendture are reported at a very detaled level of classfcaton (COICOP for all the countres), whch requred a reclassfcaton and groupng of the 19

20 data to form the sx consumpton categores: 'Food, alcohol & tobacco', 'Clothes & footwear', 'Housng-sold & lqud fuels', 'Housng-electrcty', 'Prvate transportaton' and 'Others'. Some of the soco-demographc varables ncluded n the estmaton were not reported n the four surveys usng the same classfcaton. The varable "age of the reference person" for nstance s reported n years n three of the four surveys therefore for consstency classes have been constructed and the varable s therefore used as a dummy. The same s true for the varables "level of educaton of the reference person" and "type of the professonal actvty of the reference person"; also n ths case classes have been constructed consstently. To deal wth the zero entres, some observatons have been truncated on the bass of the followng crtera: 1. f the consumpton share for both 'Housng-electrcty' and 'Housng-sold & lqud fuels' s zero. The observatons truncated are 493; 2. f the consumpton share for 'Others' s zero. The number of truncated observatons s 874; 3. f both the consumpton share of 'Prvate transportaton' and 'Others' s zero. Number of dropped observatons s 0; 4. f the consumpton share of 'Prvate transportaton' s zero and the number of cars owned by the household s larger than 1. Number of dropped observatons s 1530; 5. f the consumpton share for food s zero. Number of truncated observatons s 83. TABLE 2 As shown n Table 2, the total number of observatons s and the dataset offers large varance of the man varables of the analyss as shown by the standard devaton. 20

21 The negatve fgure n Table 2 for the mnmum value of the categores prvate transportaton and fuels s so because few famles receve subsdes for these two categores. The surveys' data on goods and servces household expendtures have been aggregated to the fve categores, subject of the study, from the very detaled level usually COICOP wth fve dgts. The aggregaton has been straghtforward for all the categores as based on a common classfcaton code. From the consumpton categores constructon the energy expendture for the second dwellng has been excluded to avod dstortons. The model has been estmated usng the SUR estmator (Zellner 1962). TABLE 3 By lookng at the prevous results what we can quckly observe s that only 13 parameters out of 110 have a statstcal sgnfcance below 10%. As the estmated system Engel curves ncludes the squared ncome term, t can classfy the commodty groups as necesstes (.e. negatve slope, so a smaller consumpton share wth an ncreasng ncome) or luxures (.e. postve slope, so a larger consumpton share wth an ncreasng total expendture) at dfferent levels of total expendtures. All the total expendtures parameters are statstcally sgnfcant. The categores 'Food, alcohol and tobacco', 'Prvate transportaton' and 'Fuels for housng' are luxures at low level of total expendtures and become necesstes at hgher levels of total expendtures (.e. nverted U-shape). Whle the remanng categores 'Others' and 'Electrcty' are necesstes at low level of total expendtures and become luxures at hgher levels of expendtures (.e. U shape). The household sze varable, whch has been ntroduced n the model as nteracton term wth the natural logarthm of total expendture, has a dfferent nfluence over the consumpton categores. Household sze 'slopes up' the consumpton of 'Food, alcohol and tobacco', the consumpton of 'Prvate transportaton' and of 'Electrcty', whle t 'slopes down' the consumpton of 'Fuels for housng' and of 'Others'. The next 21

22 ndependent varable n the lst s the number of cars owned. Ths varable s expected to explan a substantal part of the varaton of the consumpton share for 'Prvate transportaton', wth a postve nfluence on the slope of ths curve. The estmated parameters not surprsngly confrm ths expectaton, ndeed an ncreasng number of cars owned postvely nfluence the slope of ths curve. The varable number of cars owned also should have an nfluence on the slope of the curve for 'Others' that ncludes both the consumpton of publc transportaton and of nsurance servces related to the ownershp of a car. The estmated results for ths varable ndcate a negatve nfluence on the slope of ths curve and mplctly a substtuton between publc and prvate transportaton. The effect of ths varable on all the remanng categores despte statstcally relevant, can not be explaned clearly. All the remanng explanatory varables are ntroduced n the model as ntercept shft (.e. dummy varables). 3.3 Combnng cross secton and tme seres estmaton va calbraton In prncple the lnk between the tme seres model descrbed n secton 3.1 and the cross secton model presented n secton 3.2 could be done by combnng both model approaches n one comprehensve estmaton procedure or by smply calbratng an unque model usng the estmaton results of both models. Recently Kratena, Meyer and Wueger (2009) have proposed a methodology of combnng estmaton results for calbratng some parameters that are then a pror fxed n a fnal estmaton process. Ths methodology has been appled for Austra, where one aggregate tme seres has been combned wth one cross secton-consumer survey. Ths methodology also heavly reles on calbraton, but determnes the cross prce parameters stll by econometrc estmaton. 22

23 In our case the data bases used for the EU model are not a one-to-one correspondence of pure tme seres and cross secton data for a certan sample of countres. Instead we have produced pooled tme seres regresson results for the EU 15 and cross secton results for 4 countres. As a fnal step of our research we want to derve a model of prvate consumpton for the EU 15 that can be lnked to nput-output models. One opton therefore s to lnk the tme seres and the cross secton model by choosng best fttng elastcty values and parameters for soco-demographc varables from both approaches and calbratng a full consumpton model for a new data set (a sngle country or the EU 15, etc.). Income s the man lnk varable of both models and we can use the advantage of the cross secton over the tme seres nformaton concernng number of observatons and hgher varance across dfferent household types. As results from the estmatons we use the elastctes representng a relatve measure of the propertes of each demand system (cross secton and tme seres). The elastctes of both models are used together wth the budget shares to derve parameter restrctons. The full model set up n terms of budget share equatons can be wrtten as: w k prof k D j prof k j log p k j age k D age k log k C P edu k D hs edu k log C P log hs car log C P log car (20) The varables prof D k, age D k edu D k and represent the shares of the household groups wth the correspondng characterstcs (professonal actvty, age and educatonal level of the reference person) wthn the total of households. The household characterstc 'constructon of the dwellng' has been excluded, as aggregate data at the country level usually are not avalable for ths varable. 23

24 In general the calbraton procedure uses the elastcty formulas and combnes them wth the data set of the country n the base year chosen for the calbraton n order to derve the parameter values. For the prce parameters ( j ) we drectly apply the own and cross prce elastctes of the tme seres model and combne them wth the budget share data. That yelds the followng expressons for the and the j :: j j w w w w (21) j j w w 2 j w (22) Agan j s the Kronecker delta wth j = 0 for j and j = 1 for =j. For the expendture parameters the procedure becomes more complcated, as the nteracton terms also nfluence the expendture elastctes. Startng pont of dervng all the parameters s the calculaton of elastctes n the cross secton model, consstng of the three dfferent terms that can be rearranged to yeld: 1 w (23) 2 log C hs log( hs) 2 hs log( hs)log C car log( car) 2 car log( car) log C Note that ths representaton contans the explct representaton of all quadratc expendture terms from the cross secton estmaton results, whch are added to the lnear part captured n. We concentrate the sx terms on the rght hand sde of (23) to three terms, each one for the mpact of C, hs and car and assume that the weghts of these three terms n determnng the expresson on the left hand sde are also gven from the cross secton estmaton. That allows us calculatng the sngle parameters for a 24

25 gven ncome elastcty (from the cross secton estmaton) and the data for budget shares as well as log(hs) and log(car) of the country n the base year chosen for the calbraton. Fnally the parameters k for the soco-demographc varables are calculated takng as a startng pont also the cross secton estmaton results. The parameter values from these results are taken together wth the constant term to arrve at the total mpact of the dummy varables. The specfc parameters for the nfluence of each household group can then be calculated as the dfference between the mean of ths total mpact and each sngle total mpact value. Ths methodology guarantees that the sum of the so calculated mpact parameters s zero and s multpled wth a varable D that expresses shares of households n the populaton. Therefore these terms always have a zero total mpact on consumpton, but change the consumpton patterns, when the household structure changes. In a frst exercse we have desgned calbraton fles for 4 countres (Belgum, Denmark, Fnland and Germany) for the benchmark year 2000 that can be mmedately adjusted for other countres and years. Comparng the results for the parameters n these countres followng from ths calbraton exercse we fnd small dfferences n j as well as n, whch are due to dfferent budget shares n the benchmark year (2000). On the other hand all parameters lnked to the soco-demographc varables k are dentcal, as these are drectly derved from the cross secton estmaton results and budget shares play no role. For the parameters hs and car of the nteracton terms of hs and car wth total expendture the results are mxed. One mportant and strkng result s that the mpact of cars per household has postve as well as negatve mpacts on gasolne/desel demand, although the budget shares closely spread around the value of 6%. Small changes n the budget share obvously can lead to a change n the sgn of ths mpact 25

26 parameter ( car ). Ths result of the calbraton procedure must be seen crtcal and mght call for a fnal adjustment procedure n the calbraton methodology n order to correctly represent the tme seres and cross secton-model estmaton results. 4. Aggregate consumpton functon The households' demand system descrbed n sectons 2 and 3 represents the households' behavour decson on the allocaton among certan goods and servces of a gven overall expendture. However, the ntegraton n the modellng framework of an aggregate consumpton functon permts the aggregated level of expendture to be determned as an endogenous varable. An aggregate consumpton functon explans the level of households' consumpton n terms of dsposable ncome and for the modellng framework presented n ths paper t has been estmated n the form of an Error Correcton Model (ECM). The ECM s an econometrc procedure appled to explan the relatonshp between two contegrated tme seres varables, such as dsposable ncome and consumpton n ths partcular case, whch converge to equlbrum n the long run but exhbt an ndependent random walk n the short term. In these cases the ECM explans both the short term reacton and the long term path toward equlbrum. Equaton (24) shows the specfcaton for the ECM model, where Δlog(c t ) s the dfference between logarthm of total expendture at tme t and t-1, α 0 a constant, β the parameter for the short term reacton of overall consumpton as a functon of dsposable ncome, γ s the correcton parameter and measures the speed at whch pror devatons from equlbrum are corrected, υ the error term. n log( c t ) ( ) log( xt ) (log( ct ) log( xt )) (24) t 0 26

27 The ECM consumpton model specfcaton s a convenent form of representng aggregate consumpton n terms of dsposable ncome as t ncludes both short term reactons and long term adjustments of consumpton as a functon of dsposable ncome. The ECM has been estmated usng tme seres from 1990 to 2008 for Gross dsposable ncome and Households' consumpton obtaned from Dansh sectoral natonal accounts as ths paper mplements the Econometrc IO modellng framework wth the Dansh Supply and Use tables and Household demand system. Pror to the estmaton of an ECM s the estmaton of a pure Keynesan long-term consumpton functon of the form of: log( c ) 0 log( ) (25) t x t t to test for the exstence of a contegrated relatonshp between the two varables dsposable ncome and consumpton. In general, a hgh R 2 value and a low Durbn- Watson statstc (DW statstc) ndcate the presence of a contegrated relatonshp as the hgh R 2 reflects the presence of a common long term trend n the data whle the low DW statstc ndcates non statonary resduals. In the present case R 2 and DW for the model n (25) s respectvely and An Augmented Dckey-Fuller test on the resduals ε confrms the presence of a contegrated relatonshp though n a weak form, whch mght depend on the relatvely short tme seres avalable for these varables. TABLE 4 The estmated parameters exhbt a good overall sgnfcance except for the ntercept. The ECM s n the form of log-log therefore the estmated parameters can have a quck nterpretaton as elastctes. The β s larger than 1 whch means that n the short term consumpton reacts more than proportonally to an ncrease of dsposable ncome. The γ has the expected negatve sgn and ndcate the yearly rate at whch a devaton of consumpton from dsposable ncome approaches the long term equlbrum. In 27

28 partcular ts value of means that an ntal devaton s almost halved n one year and n total t takes a bt more than two years to return to equlbrum. The t test of sgnfcance for the γ parameter s a further test confrmng the presence of a contegrated relatonshp between dsposable ncome and consumpton. 5. Scenaro analyss The model presented n ths paper s appled for the analyss of a set of two scenaros concernng some of the varables of the household demand system, namely: Effcency rate of utlzaton of electrcty for housng purposes; Share of households wth non employed person of reference; The ncrease of effcency use of electrcty for housng purposes s analysed to look and quantfy the 'rebound effect' assocated to an ncrease of the effcency rate and a consequent decrease of the servce prce. All the smulatons have a pure demonstratve scope, as the soco-demographc varables and of the effcency rate are assumed to follow an arbtrary path of development. Table 5 shows the assumed path of development underlyng the scenaros. TABLE 5 The scenaros are analysed ndvdually usng as ndcator the electrcty consumpton. The scenaros are analysed from 2000, the base year, untl FIGURE 1 In both Fgure 1 and 2, the plotted lnes exhbt a growng trend dependng on the aggregate consumpton functon, whch has a long term equlbrum equal to 0.95, the β 1 coeffcent shown n Table 4; the long term equlbrum between overall consumpton 28

29 and ncome means that consumpton grows untl achevng a level equal to 95% of avalable ncome. For ths reason also at the baselne, fnal consumpton of Electrcty n ths partcular case exhbts a growng trend. Fgure 1 shows the results for the frst of the analysed scenaros, called 'Effcency'. The scenaro 'Effcency' s analysed by lookng at the consumpton of the category 'Electrcty'. The graph shows three lnes one for the consumpton of electrcty for the baselne, one consderng only the ncrease of the effcency rate and a last lne correspondng to electrcty consumpton takng nto account the rebound effect. The rebound effect s an ncrease of the demand for electrcty as a consequence of a decrease of the servce prce (secton 2 for detals). An ncrease of the effcency rate of utlzaton of electrcty n household devces, ndeed, has a postve effect as t mples a decrease of the consumpton of electrcty, however the decrease of the prce of the servce, calculated as the rato between the prce and the effcency rate (secton 2 for detals), undermnes ths postve effect as a lower prce for electrcty nduces a hgher consumpton. The dstance between the two lnes represents the calculated rebound effect. FIGURE 2 The second scenaro assumes an ncrease of 1% per year of the share of the households wth a non employed reference person, and of course a decrease of the share of the households n the populaton wth an employed reference person. Ths scenaro s modelled by exogenously assumng a change of the structure of the households nfluencng the consumpton patterns through the ntercept shft coeffcents estmated n secton 3.2 and calbrated for Denmark. The results plotted n Fgure 2 show lower 29

30 electrcty consumpton as a consequence of an ncrease of the non employed reference person n the Dansh households. 6. Conclusons Ths paper has presented the results of an econometrc study amng at the estmaton of a household demand system sutable to be ntegrated n an Econometrc IO model. The household demand system ncludes prce and ncome parameters as well as other soco demographc explanatory varables and s estmated by consstently combnng aggregated tme seres data from Natonal Account, wth household survey cross secton dataset. The combnaton of these two dfferent types of data permts the estmaton of a household demand system wth prce elastctes, as the aggregated tme seres data offer enough varety for prce ndexes, as well as wth a rch set of soco-demographc explanatory varables derved from the household surveys. The paper also presents the results of the estmaton of an aggregated consumpton functon as an error correcton model relatng overall avalable household ncome to total expendture. The estmated household demand system, along wth an aggregated consumpton functon, are then ntegrated nto an Econometrc Input-Output (EIO) model based on Eurostat Supply and Use table of Denmark, as recently proposed by Kratena and Strecher (2009). The modellng framework ntegratng econometrc models for consumpton and supplyuse table s used for the analyss of two scenaros. The frst scenaro analyss calculates the rebound effect assocated to a decrease of the servce prce for electrcty nduced by an ncrease of the effcency rate of electrcty utlzaton for housng purposes. The results show that a lower servce prce for electrcty nduces a hgher demand for ths good, whch undermne the postve effects of hgher effcency. 30

31 The second scenaro s about the nfluence of a hgher share of the household wth an unemployed reference person and shows ths nduces a lower electrcty consumpton. The Econometrc IO model s proposed as an alternatve both to smple IO models and CGE; t s constructed startng from a supply and use model and ntegratng econometrcally estmated blocks such as a households demand system for a more realstc consumers' preferences representaton and an aggregate consumpton functon for the relatonshp between ncome and total expendture. TABLES AND FIGURES TABLE1 Parameters standard errors FOFO *** FOCL *** FOF ** FOH FOH_E CLCL ** CLF ** CLH CLH_E ** FF *** FH * FH_E * HH *** HH_E *** H_EH_E *** FO *** CL ** F *** H H_E * 31

32 Quantles Varable n Mean S.D. Mn.25 Mdn.75 Max w_food w_others w_transp_prv w_electrcty w_hous_en w_clot_ftw l_exp exp e+05 food_al_tob others e+05 clot_ftw transp_prv electrcty hous_en household sze n. cars TABLE 2 Prvate Food Electrcty Heatng transportaton Others β *** *** *** *** *** (0.0013) (0.0047) (0.0104) (0.0134) (0.0016) λ *** *** *** -- (0.0003) (0.0006) (0.0007) -- β * sze *** *** *** *** *** (0.0002) (0.0004) (0.0009) (0.0012) (0.0002) λ * sze *** *** *** (0.0000) (0.0001) (0.0001) -- 32