Consumer Demand System Estimation and Value Added Tax Reforms in the Czech Republic *

Transcription

1 JEL classfcaton: D1; H0; H31 Keywords: consumer behavor, demand system, QUAIDS, tax reforms, value added tax Consumer Demand System Estmaton and Value Added Tax Reforms n the Czech Republc * Petr JANSKÝ CERGE-EI and Insttute of Economc Studes, Faculty of Socal Scences, Charles Unversty, Prague (jansky.peta@gmal.com) Abstract The value added tax (VAT) rates have recently changed n the Czech Republc, and n ths paper I smulate the mpact of these reforms. They are an example of changes n ndrect taxes that change the prces of goods and servces, to whch households can respond by adjustng ther expendtures. I frst estmate the behavoral response of consumers to prce changes n the Czech Republc by applyng a consumer demand model of the Quadratc Almost Ideal Demand System (QUAIDS) on the bass of the Czech Statstcal Offce household expendture and prce data for the perod from 001 to 011. I derve estmates of own- and cross-prce and ncome elastctes for ndvdual households. I then use these elastctes to estmate the mpact of the changes n VAT rates that were proposed or mplemented between 011 and 013 on households quantty demanded and government revenues. One of the man fndngs s that the estmated ncreases n government revenues that take the consumer responses nto account are more than a quarter lower than the estmates that use the statc smulaton. 1. Introducton Value added tax (VAT) s one of the most mportant taxes n the Czech Republc, as well as n the rest of the developed world. The mpact of changes n VAT depends on the mcroeconomc behavor of consumers and my objectve here s to shed more lght on the behavoral responses of Czech consumers to tax rate changes. A rgorous analyss of the mpact s partcularly pertnent n the Czech Republc snce the reduced and standard VAT rates have recently gone through mportant changes. They were, respectvely, 10% and 0% n 011, 14% and 0% n 01, and after a last-mnute change from the prevously approved unfcaton of rates at 17.5% a one percentage pont ncrease n both rates to 15% and 1% n 013. The exstng mpact evaluatons of these VAT reforms have, at best, made use of mcroeconomc data and frst order approxmatons such as Dušek and Janský (01a) and Dušek and Janský (01b). However, these studes used a statc mcro- * I would lke to thank Mchal Bauer, Rchard Blundell, Lbor Dušek, Kaml Dybczak, Kaml Galuščák, Tomáš Holub, Karel Janda, Klára Kalíšková, Andrew Lecester, Peter Levell, Melane Lührmann, Danel Münch, Cormac O Dea, Davd Phllps, Phlppe Rusn and Ondřej Schneder for ther helpful comments and dscussons at varous stages of ths paper s preparaton. I am also grateful to partcpants at the Czech Mnstry of Fnance and Charles Unversty semnars for ther comments. Fnancal support from the Technology Agency of the Czech Republc (TACR TD010033), the CEFRES Fellowshp and the Czech Scence Foundaton (GACR 403/10/135) s gratefully acknowledged. The data used n ths study were collected and provded by the Czech Statstcal Offce. I have also benefted from nsttutonal support from the Insttute for Fscal Studes, Insttute of Economc Studes of the Faculty of Socal Scences at Charles Unversty and CERGE-EI, a jont workplace of Charles Unversty and the Economcs Insttute of the Academy of Scences of the Czech Republc. All errors are my own. 46 Fnance a úvěr-czech Journal of Economcs and Fnance, 64, 014, no. 3

2 smulaton wth no behavoral response and dd not properly account for the potental for consumers to substtute goods as relatve prces change, as dscussed by Banks et al. (1996). In the case of VAT rate ncreases, ths mght cause over-estmaton of the effects of VAT rate ncreases on government revenues. For a more rgorous analyss t s useful to have detaled knowledge of consumers ndvdual preferences, about whch, however, nformaton s not readly avalable. So as a frst step, ths paper derves second order approxmatons whch do not dsplay systematc bases as shown n Banks et al. (1996), but whch, n contrast to frst order approxmatons, requre knowledge of the elastctes. Specfcally, I estmate the Quadratc Almost Ideal Demand System (QUAIDS) as developed by Banks et al. (1997). The QUAIDS model was prevously estmated for the Czech Republc by Dybczak et al. (010) to derve elastctes and analyze the mpact of regulated prce changes on consumer demand. The QUAIDS model has also been appled n the analyss of VAT reforms n the UK by Crawford et al. (010) and Mexco by Abramovsky et al. (01). Therefore, to the best of my knowledge, ths s the second QUAIDS model for the Czech Republc and the frst QUAIDS model bult specfcally for the analyss of tax polcy n the Czech Republc. The model employs household expendture and demographc data from the Household Budget Survey (HBS) and prce data from the consumer prce ndex (CPI), both from the Czech Statstcal Offce (CZSO). Ths demand system dffers from the exstng models for the Czech Republc n terms of the consumer prce nformaton used. Regardng prces, I rely solely on the CPI, rather than on the HBS. I have chosen to classfy expendture so that t reflects not only functonal groupngs (e.g. food, clothes) but also dentfes goods and servces that ncur dfferent VAT rates. I thus apply an approach smlar to that of Abramovsky et al. (01), who poneered the use of the QUAIDS to analyze the mpacts of VAT changes. The estmated prce and ncome elastctes appear plausble n magntude and sgn. For nstance, food s found to be a necessty whle eatng out s found to be a luxury. Strong luxures nclude transport and recreaton and household goods. Ths categorzaton and these estmates of elastctes, together wth a smple mcroeconomc smulator, enable me to estmate how consumers respond to changes n VAT rates and the mplcatons for consumers spendng patterns, quantty demanded and government tax revenues. I then use ths model to smulate the recent VAT reforms. In lne wth Banks et al. (1996) and smlarly to prevous research by Crawford et al. (010) and Abramovsky et al. (01), I fnd that for the Czech Republc too, allowng for behavoral response makes a dfference to estmates of tax revenues, whch are lower n comparson to the estmates produced by a frst order approxmaton holdng behavor fxed, specfcally the statc mcro-smulaton model that dd not allow for any behavoral response and held the quantty of purchases fxed (nomnal rse or fall of expendtures n lne wth the rse or fall n VAT rates). The layout of the paper s as follows: Secton provdes a bref lterature revew. In Secton 3, I dscuss methodology ncludng the theoretcal consumer demand model. In the Appendx I show the dervaton of the formulae for the estmaton of elastctes. In Secton 3 I further descrbe the data and explan the smulaton Fnance a úvěr-czech Journal of Economcs and Fnance, 64, 014, no. 3 47

3 of the model. Secton 4 presents the results especally n the form of the estmated elastctes. Secton 5 dscusses the applcaton of the model s results to the evaluaton of recent changes n VAT rates. It also shows the mpact of these on households and government revenues. Secton 6 provdes a concluson.. Lterature Revew The lterature on consumer demand and VAT s qute extensve and I wll therefore focus on only three areas. Frst, I brefly ntroduce the most mportant contrbutons to demand system estmaton. Second, I dscuss exstng artcles that estmate demand systems for the Czech Republc. Thrd, I provde an overvew of the lterature on the mpacts of VAT n the Czech Republc. Stone (1954) frst poneered the estmaton of a demand system based on consumer preferences theory; specfcally, he estmated lnear expendture systems as developed by Klen and Rubn (1947). A number of mprovements have been elaborated and proposed over the decades. One demand system that s often estmated nowadays s the Almost Ideal Demand System (AIDS) developed by Deaton and Muellbauer (1980). The Quadratc Almost Ideal Demand System (QUAIDS), developed by Banks et al. (1997), s essentally a verson of AIDS that allows Engel curves to be quadratc. Furthermore, Po (00) and Po (008) are useful ntroductons to estmatng the QUAIDS usng the STATA software, as I do here. Recent applcatons of the QUAIDS model smlar to those used n ths paper are Crawford et al. (010), who present estmates of the mpact of a hypothetcal unfcaton of VAT rates n the UK and also dscuss the mplcatons of VAT for labor market partcpaton based on UK data, and Abramovsky et al. (01), who evaluate Mexcan tax reform. Second, demand systems have recently been estmated for the Czech Republc, most notably n two research papers. Usng the AIDS n a modfcaton by Edgerton (1996), Janda et al. (010) estmated elastctes focusng on alcoholc beverages and found, for example, a very low own-prce elastcty of demand for beer. Dybczak et al. (010) were the frst to estmate the QUAIDS for the Czech Republc and dvde expendture nto eght categores food, clothng, energy, housng, health, transport, educaton and other that do not, however, algn wth VAT rates (as do the categores used n ths paper). They estmated own- and cross-prce and ncome elastctes and used them to analyze the mpact of changes n regulated prces on consumer demand. In addton, a number of studes, such as Dubovcka et al. (1997) and Janda et al. (000), have focused on estmatng Czech food demand elastctes usng flexble functon forms, to whch both the AIDS and QUAIDS also belong. Last, but not least, Crawford et al. (004) developed a new method for the estmaton of prce reactons and appled t to the Czech data. By estmatng the QUAIDS model wth the most recent data and for the purposes of ndrect tax polcy analyss, I contrbute to the exstng lterature on demand system estmaton n the Czech Republc, most notably Dybczak et al. (010). Thrd, a short overvew of VAT n the Czech Republc and the related lterature: the Czech Republc ntroduced VAT n 1993, and t apples to most household expendtures at one of ts two rates. In recent years the reduced and standard rates have been, respectvely, 10% and 0% n 011, 14% and 0% n 01, and n 013 after a last-mnute change from the prevously approved unfcaton of rates at 48 Fnance a úvěr-czech Journal of Economcs and Fnance, 64, 014, no. 3

4 17.5% the government ncreased both ts reduced and standard rates by one percentage pont to 15% and 1%, respectvely. VAT and changes thereof n the Czech Republc have been studed by Schneder (004), who analyzed the tax burden of households and found VAT to be relatvely regressve, and more recently by Klazar et al. (006), who use a mcro smulaton model to estmate ncdence of taxes but wthout the use of elastctes, and by Klazar et al. (007), who focused on the mpact of EU-accesson related harmonzaton of VAT rates. Klazar and Slntáková (010) studed VAT n the Czech Republc and ts mpact on households, and found VAT to be regressve when annual ncome s analyzed, though ther lfetme ncome analyss ndcated that VAT s progressve. Most recently Dušek and Janský (01a) and Dušek and Janský (01b) used a smple statc mcro-smulator wthout usng a demand system and accountng for behavoral response to VAT changes as I do n ths paper to provde the frst ndependent estmates of the mpact of the recently proposed VAT rate changes n the Czech Republc on the lvng standards of households as well as on the government s tax revenues. One objectve of ths paper s to compare the results of analyss of these VAT reforms accordng to whether behavoral change s taken nto account or not. By accountng for behavoral response n tax polcy analyss and showng the dfferences n the result when compared wth the statc mcro-smulaton methodology that holds fxed the quantty of goods and servces purchased, I contrbute to the exstng lterature on smulaton of VAT reforms n the Czech Republc (see Dušek and Janský, 01a and Dušek and Janský, 01b). 3. Methodology I estmate the demand system accordng to the Quadratc Almost Ideal Demand System form developed n Banks et al. (1997) and I further use ths for ndrect tax polcy analyss, as proposed by Banks et al. (1996) and appled n Crawford et al. (010) and Abramovsky et al. (01). The QUAIDS model allows me to take consumers substtuton responses nto account when relatve prces change due to VAT reforms, whereas ths s the frst such model bult n the Czech Republc specfcally for the analyss of tax polcy. 3.1 QUAIDS The model s based on an ndrect utlty functon from whch the shares of expendture on varous categores of goods and servces are derved, and these are then updated wth demographc characterstcs. Smlarly as n the only QUAIDS prevously estmated for the Czech Republc by Dybczak et al. (010), demand depends not only on prces and ncomes, but also on other household characterstcs such as the sze of the household or the employment status or age of the household s head. It s estmated by seemngly unrelated regresson equatons wth parameter restrctons so that the estmated demand system satsfes the condtons of addngup, homogenety, symmetry and negatvty (negatve sem-defnteness). 1 The QUAIDS model s a generalzaton of the Almost Ideal Demand System (AIDS) model that allows for quadratc Engel curves. The QUAIDS can therefore 1 The model does not allow for postve or negatve externaltes from expendture on certan goods and ths assumpton of no externaltes s a lmtaton on the usefulness of the QUAIDS when lookng at the effects of excse dutes on goods wth negatve externaltes such as fuel or tobacco. Fnance a úvěr-czech Journal of Economcs and Fnance, 64, 014, no. 3 49

5 allow a good to be a luxury at one level of ncome and a necessty at another, a property that Banks et al. (1997) found to be of emprcal relevance for the UK and they also showed that t s suffcent for the nonlnear term to be a quadratc n log ncome. Ths was documented for the Czech Republc by Dybczak et al. (010). Here I brefly ntroduce the QUAIDS model, essentally as developed and presented by Banks et al. (1997), but I do so n a more complete way, ncludng the dervaton of the elastctes, n the Appendx. The QUAIDS model s based on the followng ndrect utlty 3 : 1 ln x ln a( p) ln V = + λ( p) b( p) Where x s total expendture, p stands for prces, a(p), b(p) and λ (p) are defned as: 1 1 ln a( p) = α0 + α ln p + γ ln p ln p ( ) j ( ) ( j ) j ln λ ( ) b p n p β = 1 = n ( p) λ ln ( p ) = = 1 where = 1,..., n denotes a good, ln a(p) s the translog prce aggregator functon, and b(p) s defned as the smple Cobb-Douglas prce aggregator. Applyng Roy s dentty to the equaton for ln a(p), I have the followng equatons for w, the share of expendture on good n total expendture for each household: x λ x w = α + γ ln ( p ) + β ln + ln a( p) b( p) a( p) j j j These equatons are the subject of the estmaton by seemngly unrelated regresson equatons. 3. Data To estmate the QUAIDS model I employ the best avalable data for the Czech Republc n the form of two datasets from the Czech Statstcal Offce (CZSO). The Household Budget Survey (HBS) s a representatve sample of approxmately 3,000 Czech households collected on a yearly bass. 4 For each of them, the HBS contans nformaton on how much they spend on varous goods and servces (around 50 expendture tems), who they are (around 60 demographc Some studes dentfed the mportance of further terms n ncome for some, but not all, expendture share equatons before Banks et al. (1997); see for example Atknson et al. (1990) or Blundell et al. (1993) and Banks et al. (1997) for further dscusson and references. 3 The same ndrect utlty functon defnes the AIDS model, but wth the λ(p) term set to zero. 50 Fnance a úvěr-czech Journal of Economcs and Fnance, 64, 014, no. 3

6 varables) and how they earn ther ncome (around 30 ncome tems). 5 The HBS has been appled to the estmaton of demand systems by both Janda et al. (010) and Dybczak et al. (010) and t was also used by Crawford et al. (004). 6 In terms of the years, there s a trade-off between the amount of data and ther qualty and consstency; I resolve ths ssue by usng 11 years. I employ data for the perod between 001 and Therefore I have data from approxmately 33,000 households n total and I assume ths to be a representatve sample for the Czech Republc. I use CZSO prce data, gathered for the purpose of the consumer prce ndex (CPI), that are classfed nto approxmately 150 categores accordng to the classfcaton of ndvdual consumpton by purpose (COICOP). 8 The prce nformaton s avalable for the Czech Republc as a whole and separately for the country s captal cty, Prague. I rely on the CPI as the sole source of prce nformaton; ths s n contrast to Janda et al. (010), who used the HBS not only as a source of expendture nformaton, but also as a source of prce nformaton. Specfcally, they dvded the expendtures by the quantty of purchased goods and servces. In ths way they derved unt values and used them as prces. Ther approach has some advantages of relatvely easly obtanng expendture- and household-specfc prces and of havng these n a very detaled form,.e. as detaled as the HBS expendture data, whch mght 4 The CZSO gathered the HBS data n such a way that only the data for the perod can be consdered, after applyng the suppled weghts, to be representatve of the overall Czech populaton. However, as some prelmnary robustness checks dd not fnd sgnfcant dfferences between the two perods and snce Dybczak et al. (010) showed the same, I conclude that there s no sgnfcant rsk n usng the data for the longer perod for the purpose of estmatng a demand system. Stll, some concerns reman and although treatment of the potental problems of ths nature s largely beyond the scope of ths paper, I wll brefly revew them here. The CZSO was crtczed for falng to provde the full HBS dataset to European Unon representatves (Eurostat 009). Furthermore, as Crawford and Smth (00) dscuss, systematc over- or under-reportng of expendtures can occur n ths type of data collecton due to forgetfulness (e.g., consumpton outsde the home), actve concealment (e.g., a recept from a beauty salon) and gult (e.g., cgarettes). Ths knd of problems s also dscussed n Svátková (006) and Klazar et al. (006). 5 One of the problems mentoned by Crawford and Smth (00) that may occur n the Czech HBS s that, snce some expendtures are recorded monthly, large and nfrequent purchases may be underestmated. Ths s one of the reasons why I exclude housng from the expendture data, both purchase and rent (the correspondng HBS codes n the 011 data are 4010, 4080, 5440, 5550), wth a further reason beng that relatvely hgh and nfrequent housng expendtures would lkely dstort the model estmaton. By excludng housng expendtures, I am also excludng ther potental mplcatons for wealth and, consequently, purchasng decsons of the households. Ths s a common approach n the exstng lterature, for example Banks et al. (1997). 6 One of the drawbacks of the HBS s that the data are gathered by purposve quota samplng and although the desgners of the survey have made efforts to make t as smlar to a random sample as possble, ts characterstcs are not the same as those of a random sample. One mplcaton of ths suboptmal samplng s that results based on the data mght be based. However, my use of the data s n lne wth the efforts of the CZSO to make the sample resemble a random sample as closely as possble and wth prevous uses of the data, such as by Dybczak et al. (010). 7 Ths perod ncludes the recent economc crss, whch could nfluence consumpton decsons. I was able to do only a very prelmnary analyss of the crss s nfluence (also due to data lmtatons) and I therefore leave the related questons for further research, lkely smlar to that conducted by Crossley et al. (013) for the Unted Kngdom. 8 Some prces are regulated n the Czech Republc, ether natonally (healthcare fees) or locally (waste dsposal servce charges) and would therefore arguably requre specal treatment, whch I however do not provde here due to the complexty of such an exercse and because the share of regulated prces s relatvely low. Fnance a úvěr-czech Journal of Economcs and Fnance, 64, 014, no. 3 51

7 Table 1 Demographc and Other Varables Used n the Demand System Varable Chld Members Age Sex Employment status Educaton low Educaton mddle Cty sze Praha Tme trend Descrpton = number of chldren = number of household members = age of the head of household = 1 f the head of the household s female, 0 otherwse = 1 f the head of the household s employed, 0 otherwse = 1 f the head of the household has prmary educaton or less, 0 otherwse = 1 f the head of the household has secondary educaton, 0 otherwse = 1 for regonal captals, = for ctes, 3 = for vllages = 1 f the household resdes n the captal cty, Praha, 0 otherwse = year of the survey (mplemented as 01 mnus the year of the survey) mprove the precson of elastcty estmates. However, t can be naccurate n some cases, whle another drawback s that the HBS data for the quantty of purchased goods and servces s ncomplete. Dybczak et al. (010) combne the two datasets, whereas they fll n mssng prce nformaton n the HBS data from the CPI. There s an obvous trade-off between these two sources of prce nformaton and there are three man reasons why I opt to use the CPI as my source of prces. Frst, dfferences n unt values can be caused by product qualty dfferences, rather than by the prce dfferences that I am to study. Wth these unt values t s almost mpossble to dstngush between the nfluence of changes n prce and n qualty, snce rsk observed prce varaton may nstead reflect varatons n qualty. By usng the CPI data, I lmt the extent of ths problem. Second, the HBS only ncludes nformaton on the quantty of purchased goods and servces for a lmted number of expendture tems. Unt values can thus only be constructed for those goods for whch quantty nformaton s avalable. Therefore, f I used the HBS data for prces, I would need to lmt my analyss to a small subset of overall expendtures, as Janda et al. (010) dd, or alternatvely fll n the HBS unt values whenever these are not avalable usng the CPI prces, as was done by Dybczak et al. (010). In contrast to Janda et al. (010), I prefer to analyze as hgh a share of overall household consumpton as possble, and ths s made possble by applyng the CPI data. In contrast to Dybczak et al. (010), I prefer the consstency of usng only one complete source for nformaton on prces, namely the CPI. In the expendture share equatons estmated n the QUAIDS, I nclude a tme trend and a number of demographc varables to take account of preference varaton that may be correlated wth total expendture or prces n a way that s consstent wth the model. Table 1 provdes a lst of these varables. I classfy the HBS expendture data accordng to the VAT rates, reduced and standard, presented n the appendces to the VAT Act as of January 013. When HBS classfcaton s not detaled enough to allow accurate dvson accordng to the VAT rate or when some expendtures are exempted from VAT, I assgn the HBS expendtures to a group accordng to the VAT rate. I merge the HBS and CPI data usng the HBS codes and COICOP codes and, although these two classfcatons do not 5 Fnance a úvěr-czech Journal of Economcs and Fnance, 64, 014, no. 3

8 Table Expendture Groups and Ther Average Shares n 011 Group Expendture VAT rate 1 Food Reduced 4.4 Eatng out and other luxures Standard Household goods Standard Clothng Standard 6. 5 Other servces Reduced Transport and recreaton Standard Energy Standard Other goods Standard 11.9 Average share n the sum of these expendtures (%) always match perfectly and both of them have undergone revsons over tme, no substantal compromses had to be made durng the matchng process. In order to estmate the QUAIDS, I dvded the detaled expendture tems nto eght groups. I followed three prncples whle groupng the expendture tems, and n ths I dffer from the prevously estmated demand systems for the Czech Republc. Frst, the dvson should correspond to natural categores as people mght thnk about them, whch was essentally the case n Dybczak et al. (010). Second, the expendture groups should be smlar n sze, whch s advantageous both for the estmaton of the model and for the nterpretaton of the results. Thrd and most mportantly for my analyss, the expendture groups should be dvded accordng to the VAT rate as the greatest extent possble. A number of compromses had to be made when followng these three prncples, and when consderng these, I have gven hghest prorty to the thrd prncple. I calculate the prce ndces of aggregated commodtes as weghted arthmetc averages of the prce ndces of the ndvdual goods and servces makng up the aggregated commodty for each year, and then aggregate them for each household to arrve at overall expendture group-, household- and year-specfc prce ndces. Table provdes the names and shares n total expendtures of the eght expendture groups for 011. A more detaled descrpton can be found n Table A1 n the Appendx and basc summary statstcs n Table A n the Appendx. I use ths categorzaton of expendtures nto groups n my estmaton. 3.3 Estmaton The estmaton of QUAIDS s made n two stages and apples the approach of Abramovsky et al. (01). 9 At the frst stage of the estmaton, the values of a(p) and 9 There are a number of dfferent ways by whch to estmate QUAIDS and smlar models and t would be desrable to explore these optons for the Czech Republc n future research. These optons nclude varatons n the expendtures ncluded and excluded n the demand system (such as the excluson of housng n most demand systems, ncludng ths estmaton), the number of expendture groups and dvson of goods among these groups, whether and how to deal wth the outlers, whether to nclude taste shfters and demographc characterstcs and f so whch ones, what tme perod and frequency to cover, whether to use unt value prces or other, external prces, and what methods of computng elastctes (usng a weghted average, as n ths estmaton, or a representatve or average household) to employ, and whether to use a homogeneous or representatve sample of the populaton. Fnance a úvěr-czech Journal of Economcs and Fnance, 64, 014, no. 3 53

9 b(p) are unknown to me and therefore I approxmate b(p) as 1 and lna(p), usng * the Stone prce ndex named after Stone (1954)as ln p w ln p. The QUAIDS s lnear n parameters condtonal upon the prce ndces and therefore I can and do employ a lnear seemngly unrelated regresson (SUR) method to estmate the model. Addng-up s mposed by excludng the equaton for the n-th good from the estmated system of equatons; the parameters for ths equaton are calculated usng the parameters from the other (n 1) equatons and the addng-up restrctons (the results are not senstve to the choce of the nth good). Homogenety (by expressng all prces relatve to the prce of the other remanng goods) and symmetry are mposed usng lnear restrctons on the parameters. The parameters estmated at the frst stage are then used to calculate values for a(p) and b(p). The model s then re-estmated usng the same specfcaton as n the frst stage, except that * p s replaced wth a(p) and λ wth λ ( ) b p. The new parameter values are used to update a(p) and b(p), and the model s then re-estmated for a thrd tme. Ths updatng of prce ndces and re-estmaton s terated eght tmes, by whch tme the parameter values have converged to four decmal places. I nstrument for expendture usng monetary ncome because t may be endogenous. I do so usng a control functon approach as s common n the lterature and as appled n Banks et al. (1997) and Abramovsky et al. (01). 10 I calculate standard errors usng bootstrappng wth 1,300 teratons, wth clusterng at the household level. 4. Demand System Results Ths secton dscusses the results of the estmated demand system. Table 3 below presents the parameter estmates for the QUAIDS and the estmated parameters correspond to equaton number 1 above. The table uses astersks to ndcate the statstcal sgnfcance of the estmates, whch s rather low and I dscuss ths problem below wth regard to elastctes. It s dffcult to nterpret the parameters of the QUAIDS drectly and I therefore manly dscuss the elastctes, as s also common n the exstng lterature. Specfcally, I present the estmates of ncome, own- and cross-prce elastctes. I calculate the elastctes for each household ndvdually and I subsequently construct a weghted average, wth the weghts beng equal to the household s share of the total expendture and to the total sample expendture for the relevant good, for the ncome and prce elastctes, respectvely. Table 4 presents the ncome elastctes, estmated usng the total expendture varable. For comparson, the table shows also the own-prce elastctes, both the Marshallan (uncompensated) and the Hcksan (compensated), whch are shown agan together wth the cross-prce elastctes shown n Tables 5 and 6. Sectons n the Appendx nclude the elastctes formulas used n Tables 4 to Specfcally, I regress the log of total expendture (lnx) and the square of the log of total expendture (lnx) on the prces and demographc varables ncluded n the demand system and on the log of household monetary ncome and the square of the log of household monetary ncome, and I nclude lnear, square and cubc terms of the resduals from these regressons n our demand system equatons. 54 Fnance a úvěr-czech Journal of Economcs and Fnance, 64, 014, no. 3

10 Table 3 QUAIDS Parameter Estmates for the Czech Republc Food Eatng out Household goods Clothng Other servces Transport, recreaton Energy Other goods α *** *** *** *** *** *** αage *** *** *** *** *** *** *** *** αmember *** *** *** *** *** *** *** αchld *** *** *** *** *** *** * αempstat *** *** *** *** ** *** *** αeduclow *** *** ** *** *** αeducmd *** *** * *** *** *** *** αpraha *** *** *** *** *** *** αcty sze *** *** *** *** *** *** *** γ *** ** *** *** γ γ γ ** * *** ** γ γ *** ** γ *** *** *** γ ** *** Fnance a úvěr-czech Journal of Economcs and Fnance, 64, 014, no. 3 55

11 Food Eatng out Household goods Clothng Other servces Transport, recreaton Energy Other goods β *** *** *** *** *** *** *** λ *** *** * *** * Tme trend *** *** V *** ** *** V * *** *** *** *** V *** *** *** *** *** V ** *** *** V *** *** *** *** V *** *** *** *** *** Notes: The parameters V1 6 relate to the lnear, square and cubc terms of the resduals from two regressons as descrbed n footnote number 9. The cells wth parameters are complemented wth astersks n lne wth ther sgnfcance: *** mples sgnfcance at the 1% level, ** mples sgnfcance at the 5% level and * mples sgnfcance at the 10% level. 56 Fnance a úvěr-czech Journal of Economcs and Fnance, 64, 014, no. 3

12 Table 4 Income and Own-Prce Elastctes Group Expendture Income elastcty Marshallan ownprce elastctes Hcksan own-prce elastctes 1 Food * Eatng out and other luxures *** Household goods *** *** *** 4 Clothng Other servces 0.74 * * Transport and recreaton *** Energy Other goods *** -0.5 *** *** Notes: The cells wth parameters are complemented wth astersks n lne wth ther sgnfcance: *** mples sgnfcance at the 1% level, ** mples sgnfcance at the 5% level and * mples sgnfcance at the 10% level. Table 5 Marshallan (Uncompensated) Prce Elastctes Group Food Eatng out and other luxures Household goods Clothng Other servces Transport and recreaton Energy Other goods Food * * Eatng out and other luxures Household goods *** Clothng ** Other servces Transport and recreaton * * *** Energy -0.17* Other goods *** Notes: The cells wth parameters are complemented wth astersks n lne wth ther sgnfcance: *** mples sgnfcance at the 1% level, ** mples sgnfcance at the 5% level and * mples sgnfcance at the 10% level. Only half of the ncome elastctes are statstcally sgnfcant at least at the 10% level. More optmstcally, the estmated ncome elastctes seem reasonable. Other servces, whch nclude publc servces, are necesstes and the same holds for other goods, although these have an ncome elastcty of just below 1. Food and energy are both necesstes, albet not statstcally sgnfcant. So both expendture groups wth the reduced VAT rate food and other servces are necesstes. Eatng out, clothng, household goods and transport and recreaton all have ncome elastcty greater than 1 and are therefore consdered luxures. The patterns of ncome elasttctes are relatvely comparable to those estmated by Dybczak et al. (010). Food and energy are n both cases expendture groups wth the lowest ncome elastcty, whle transport has the hghest ncome elastcty. Fnance a úvěr-czech Journal of Economcs and Fnance, 64, 014, no. 3 57

13 Table 6 Hcksan (Compensated) Prce Elastctes Group Food Eatng out and other luxures Household goods Clothng Other servces Transport and recreaton Energy Other goods Food * Eatng out and other luxures Household goods * *** ** Clothng * Other servces Transport and recreaton ** Energy Other goods *** Notes: The cells wth parameters are complemented wth astersks n lne wth ther sgnfcance: *** mples sgnfcance at the 1% level, ** mples sgnfcance at the 5% level and * mples sgnfcance at the 10% level. Tables 5 and 6 show the Marshallan (uncompensated) and the Hcksan (compensated) prce elastctes, respectvely. Own-prce elastctes are on the dagonal and cross-prce elastctes off the dagonal. 11 Statstcal sgnfcance s depcted by astersks and the elastctes are largely statstcally nsgnfcant at the standard levels. 1 Somewhat more optmstcally, most Marshallan own-prce elastctes are statstcally sgnfcant at least at the 10% level and ths s one of the supportng reasons why I use these for smulatons n the next chapter. Fortunately, the confdence ntervals of the smulated mpacts of tax changes on demand and revenues are relatvely narrow despte the low statstcal power of the elastctes. The reason for ths s that the restrctons the QUAIDS model mposes on the elastctes causes them to be negatvely correlated, whch results n lower varance when I combne them n the smulaton estmates. Both Tables 5 and 6 show that all own-prce elastctes are negatve and ths observaton s n lne wth basc economc ntuton. As far as the own-prce elastcty of demands s concerned, food and energy are among the least elastc, whereas other servces and goods and transport and recreaton are among the most elastc. The patterns of cross-prce elastctes and therefore substtuton and complemen- k k j 11 The tables show the elastctes of a good n the column wth respect to prce changes of the good n the row. For example for Hcksan (compensated) cross-prce elastctes, the number of the column corresponds to the and the number of the row to the k n the followng equaton: ε = ε + ε w. An c u elastcty provdes nformaton on the percentage change n quantty demanded for a good n the row. 1 Regretfully, there are no easy remedes. One straghtforward, but not currently applcable, soluton to low statstcal power seems to be to get hold of better data, specfcally watng untl more years of data, whch should deally be more detaled, become avalable. There s more potental for greater detal n the data. For example, regonal prce ndces and quarterly expendture data would ease the estmaton by allowng t to be carred out for 14 regons on a quarterly bass n contrast to two regons on an annual bass. 58 Fnance a úvěr-czech Journal of Economcs and Fnance, 64, 014, no. 3

14 tarty seem reasonable. For example, food and eatng out are substtutes. Not surprsngly, cross-prce elastctes are mostly relatvely small and smaller than ownprce elastctes. Despte dfferent defntons of expendture groups and some dfferences between the estmated Marshallan and Hcksan elastctes, there s some scope for comparson between these elastctes and those earler estmated for the Czech Republc by Dybczak et al. (010) also usng the QUAIDS model. Food and other goods are n both cases among the expendture groups wth the lowest own-prce elastcty. There s a dfference n prce elastctes for energy wth my very low estmates n contrast wth the relatvely hgh values determned by Dybczak et al. (010) and ths mght sgnal the need for further, more focused research n energyrelated products and related prces. 5. Smulaton of Value Added Tax Reforms Ths secton apples the demand system results for tax polcy smulaton. In the case of the Czech Republc, there are a number of recent changes n VAT rates sutable for smulaton. The reduced and standard VAT rates were, respectvely, 10% and 0% n 011, the last year for whch data are avalable. Then n 01, the rates were 14% and 0% and these were ncreased by one percentage pont to 15% and 1% n 013. Furthermore, there was another proposal, ntally legslated and then postponed n late 01 untl 016, that n 013 the two rates would be unfed at 17.5%. 13 The objectve of ths smulaton s to evaluate the latest reform, namely the ncrease n both rates by one percentage pont n 013, and compare ths wth the unfcaton proposal. The last avalable year of data s from 011, when the VAT rates were 10% and 0%, and therefore I frst smulate the expendtures under the VAT rates of 14% and 0% that were n place n 01 and establsh that as the status quo. Based on ths, I then smulate the outcome n 013 n order to estmate the changes between 01 and 013 and compare these wth the unfcaton proposal. I use the estmated elastctes to smulate the mpacts of VAT reforms on consumer spendng patterns and tax revenues. For ths I use the Marshallan elastctes. Specfcally, I estmate the mpact of the mpled change n prces for each household and expendture group ndvdually. Estmates ncorporatng such behavoral response are then compared to those based on the no-behavoral-response statc mcro-smulaton results. When usng the demand system results for these purposes, I model changes n VAT rates as changes n the prces of the eght aggregate expendture groups used n the QUAIDS demand system. Also on the bass of 13 A more detaled descrpton of recent changes n VAT rates follows. Untl 011, the reduced VAT rate was 10% and the standard VAT rate was 0%. The 011 law ntroduced two changes to the VAT rates to be mplemented n 01 and 013. Frst, there was an ncrease n the reduced rate to 14% n 01, whle the standard rate remaned at 0%. Second, the unfcaton of both VAT rates at 17.5% was due to be mplemented n 013. However, n November and December 01 another new law was approved whch changed the VAT rates for 013 so that both VAT rates ncreased from the 01 levels by one percentage pont, to 15% and 1% for the reduced and standard VAT rates, respectvely. Accordng to the law and the Mnstry of Fnance of the Czech Republc (011), the unfcaton of VAT rates at 17.5% should eventually happen, but only from 016. Furthermore, VAT rates have become an mportant poltcal ssue snce 010 and t seems relatvely lkely that further changes wll be made after the last general electon held n October 013. Fnance a úvěr-czech Journal of Economcs and Fnance, 64, 014, no. 3 59

15 Table 7 Smulated Average Expendture Shares after Changes n VAT rates (%) Group Expendture 01 (14% and 0%) 013 realsed reform (15% and 1%) 013 proposed reform (17.5% and 17.5%) 1 Food R Eatng out and S other luxures 3 Household goods S Clothng S Other servces R Transport and S recreaton 7 Energy S Other goods S Total Notes: The superscrpt R and S denote the reduced and standard VAT rates, respectvely. the dscusson n Dušek and Janský (01a), I assume that changes n VAT rates are fully reflected n prces. Ths assumpton s more lkely to be fulflled n the long term rather than the short term, and ths should be taken nto account when nterpretng the results. Table 7 shows smulated estmates of average expendture shares after the 01 as well as 013 ncludng the earler proposal VAT rate changes. Overall, nether the 01 nor 013 approved VAT changes have a very substantal mpact on spendng patterns; the expendture shares change only n terms of tenths of a percent. The smulaton of the ntroducton of a unform 17.5% VAT rate suggests a larger and more vared mpact, whch s not surprsng due to the fact that ths mples a greater change n both rates n terms of percentage ponts and the fact that the two rates move n opposte drectons. In partcular, the share of food, on whch the reduced VAT rate s currently leved, ncreases substantally by more than half a percentage pont, reflectng ts low own-prce elastcty of demand, whle the share of other goods generally falls. 14 Table 8 shows estmates of smulated changes n the quantty demanded n terms of the percentage of total expendture for the two VAT rate proposals n 013. Table 8 ncludes confdence ntervals, whch I estmated on the bass of bootstrapped estmates of elastctes. These confdence ntervals are relatvely narrow, especally for the mplemented 013 reform, and generally narrower than the statstcal sgnfcance of the elastctes would suggest. I fnd that ncreasng both VAT rates by one percentage pont to 15% and 1%, respectvely, results n relatvely small decreases n the quantty demanded, wth the extent of the mpact mostly correspondng wth the ncome as well as ownprce elastctes. For example, the groups that decreased the most both household 14 However, the approxmately.4% (0.6 percentage pont) ncrease n the share of food from the 01 level s less than the 3.1% ncrease n the prce of food followng the mposton of VAT, mplyng the quantty of food purchased would be lower f VAT was mposed at the unfed rate. 60 Fnance a úvěr-czech Journal of Economcs and Fnance, 64, 014, no. 3

16 Table 8 Smulated Average Percentage Changes n the Quantty Demanded n 013 after changes n VAT rates from the 01 baselne (%) Group Expendture 013 realsed reform (15% and 1%) QUAIDS 95% confdence nterval 013 proposed reform (17.5% and 17.5%) QUAIDS 95% confdence nterval 1 Food R Eatng out and S other luxures Household S goods 4 Clothng S Other servces R Transport and S recreaton 7 Energy S Other goods S Total Notes: The superscrpt R and S denote the reduced and standard VAT rates, respectvely. goods and transport and recreaton, each by around % also have the hghest ncome elastctes and some of the hghest own-prce elastctes. The opposte pattern also holds: the groups that decreased the least both food and energy, by 0.4% also have the lowest ncome elastctes and lowest own-prce elastctes. The estmated mpacts for the unfcaton proposal are once agan more vared. The overall mpact on quantty demanded s only slghtly negatve, wth the largest declne, of 3.%, for other servces and the largest ncrease, of 4.6%, for eatng out. Table 9 shows smulaton results for government revenues correspondng to the VAT rate changes, whch are estmated on the bass of a sample of Czech households extrapolated for the whole populaton of the Czech Republc. These estmates necessarly reflect only the VAT leved on household consumpton and, snce the mpact n other sectors s not taken nto account, do not reflect the overall mpact on government revenues. Furthermore, I do not model any admnstratve savngs or lower levels of tax evason that the unfcaton of rates mght brng about. The frst part of Table 9 shows the results for the mplemented proposal (15% and 1%), the second part for the repealed 013 proposal (unfed VAT rate of 17.5%). In each of the two parts of Table 9, the frst column shows the estmated revenues from the reforms usng a statc mcro-smulaton not allowng for any behavoral response and holdng the quantty of purchases fxed. Ths s a smlar approach to that used n Dušek and Janský (01a),.e. no results from the QUAIDS model were employed to smulate these mpacts. The second column uses the QUAIDS results to allow for spendng patterns changng n response to the prce changes, smlarly to Tables 7 and 8. The thrd and fourth columns report ther confdence ntervals, whch I estmated on the bass of bootstrapped estmates of elastctes. The confdence ntervals are relatvely narrow for most estmates, especally for the mplemented 013 reform. Ths s somewhat more encouragng from the pont of Fnance a úvěr-czech Journal of Economcs and Fnance, 64, 014, no. 3 61

17 Table 9 Effect of Consumer Demand Response on Revenues from Changes n VAT Rates (CZK bllon) Group Expendture 013 realsed reform (15% and 1%) No behavoural response 013 realsed reform (15% and 1%) QUAIDS 013 realsed reform (15% and 1%) 013 realsed reform (15% and 1%) QUAIDS 95% confdence nterval 1 Food R Eatng out and other S luxures 3 Household goods S Clothng S Other servces R Transport and S recreaton 7 Energy S Other goods S Total Group Expendture 013 proposed reform (17.5% and 17.5%) No behavoural response 013 proposed reform (17.5% and 17.5%) QUAIDS 013 proposed reform (17.5% and 17.5%) 013 proposed reform (17.5% and 17.5%) QUAIDS 95% confdence nterval 1 Food R Eatng out and other S luxures 3 Household goods S Clothng S Other servces R Transport and S recreaton 7 Energy S Other goods S Total Notes: The superscrpt R and S denote the reduced and standard VAT rates, respectvely. vew of relevance of ths smulaton than the low statstcal sgnfcance of elastctes would suggest. 15 The magntude of dfference between the two estmaton methods s n lne wth expectatons: allowng for consumer spendng patterns to change n response to VAT changes has a relatvely large mpact on the resultng change n VAT revenues. Wth the statc mcro-smulaton model the estmated mpact on VAT revenues s, roundng these fgures, CZK 10 bllon for the mplemented proposal (15% and 1%) and CZK -1 bllon for the repealed 013 proposal (unfed VAT rate of 17.5%). The correspondng estmates usng the QUAIDS are CZK 7 bllon and around zero, 15 The reason s that the restrctons the QUAIDS model mposes on the elastctes causes them to be negatvely correlated, whch results n lower varance when I combne them n the smulaton estmates. 6 Fnance a úvěr-czech Journal of Economcs and Fnance, 64, 014, no. 3

18 .e. lower n total by almost CZK 3 bllon and approxmately CZK 1 bllon, respectvely. Nevertheless, as the confdence ntervals show, the dfferences are more mportant n the case of the mplemented 013 reform (15% and 1%) than the repealed 013 reform (17.5% and 17.5%),where the dfference does not seem statstcally sgnfcant. The estmated tax revenue after allowng for consumers behavor to adjust (n accordance wth QUAIDS preferences) s, as expected, lower n magntude than the estmate usng the statc mcro-smulaton methodology that holds fxed the quantty of goods and servces purchased. For the mplemented 013 proposal one percentage pont ncrease n both VAT rates the estmated ncreases n government revenues that take the consumer responses nto account are 8% lower than the estmates wth no behavoral response. These dfferences are mportant and have polcy mplcatons. 16 The 8% dfference s hgher than the estmates for smlar VAT smulaton n Mexco presented n Abramovsky et al. (01), whose fndngs mply a dfference of only about 7% and 16% for proposed and approved VAT reforms, specfcally. The relatvely hgh varaton n these estmates suggests a need for further research as to the extent of ths dfference, whch should be, at least before the estmates converge more closely, both country- and reform-specfc. These dfferences show that behavoral responses can be quanttatvely mportant and that they should be consdered n order for revenue projectons to be more precse. 6. Concluson In ths paper I have employed detaled data from the Czech Statstcal Offce to estmate a consumer demand model of the quadratc almost deal system (QUAIDS), derved the own- and cross-prce and ncome elastctes and smulated the mpacts of recent VAT reforms n the Czech Republc on households and on government revenues. Accountng for behavoral response yelded substantally dfferent estmates of the mpacts of the most recently mplemented reform on government revenues. The estmated tax revenue after allowng for behavor adjustment n lne wth QUAIDS preferences s substantally lower (by as much as 8%) than the estmate usng the statc mcro-smulaton methodology that holds fxed the quantty of goods and servces purchased. Ths contrbutes to the exstng smulaton of VAT reforms as well as more generally to the qualty of evdence-based polcymakng n the Czech Republc. Specfcally, these fndngs hghlght the need for the Mnstry of Fnance of the Czech Republc to take these behavoral responses rgorously nto account by, for example, ncorporatng the best practce n the form of the QUAIDS model nto ts own models. 16 For example, these dfferences mght partly explan the past cases of over-estmaton of VAT revenues by the Mnstry of Fnance of the Czech Republc and mght lead to further future cases. On p. 1 and 13 of the government s budget report for 013 (Mnstry of Fnance of the Czech Republc, 01), the overall ncrease (mostly due to a one percentage pont ncrease n both VAT rates and an expected ncrease n consumpton) n VAT revenues s estmated at 1.% and at CZK 9.8 bllon for the central government, and snce approxmately one-thrd of VAT revenues goes to the regonal governments, ths corresponds to around CZK 14 bllon n total. Although my estmates of approxmately CZK 10 bllon and CZK 8.5 bllon are lmted n the sense that they cover only the VAT revenues rased from household consumpton and are therefore not drectly comparable wth the mnstry s numbers, my expert estmate s that the mnstry s numbers are around the upper bounds of (my) realstc estmates and that dependng on the presumed growth of household consumpton n 013 ther numbers mght be overestmated, possbly partly as a result of falng to properly account for behavoral response. Fnance a úvěr-czech Journal of Economcs and Fnance, 64, 014, no. 3 63