Labour Force Participation and Tax-Benefit Systems: A Cross-Country Comparative Perspective * PRELIMINARY AND INCOMPLETE

Transcription

1 Labour Force Partcpaton and Tax-Beneft Systems: A Cross-Country Comparatve Perspectve * Kaml Galuščák and Gábor Kátay PRELIMINARY AND INCOMPLETE Abstract Ths paper nvestgates the extent to whch cross-country dfferences n aggregate partcpaton rates can be explaned by dfferences n tax-beneft systems. We take the example of two countres, the Czech Republc and Hungary, whch despte a lot of smlartes dffer markedly n labour force partcpaton rates. Usng two comparable ndvdual-level labour supply estmates, we smulate how the aggregate partcpaton rate would change n one country f the other country s tax and socal welfare system were adopted. Estmaton results suggest that ndvdual preferences are essentally dentcal n the two countres. Smulaton results show that about one-half of the total dfference n the partcpaton rates can be explaned by dfferences n the tax-beneft systems. JEL Codes: C63, H24, I38, J22, P50. Keywords: Cross-country comparson, labour supply, mcrosmulaton, partcpaton rate, tax-beneft systems. * We thank Sanv Avouy-Dov, Martn Guz, Tomáš Konečný, Róbert Lel and Danel Münch for ther extensve revew of ths paper. We also thank the comments of the partcpants of the SOLE conference, the Czech Economc Socety Bennal Conference and the semnar partcpants at the Banque de France and the Czech Natonal Bank. The vews expressed n ths paper are those of the authors and not necessarly those of the nsttutons the authors are afflated wth. Czech Natonal Bank, kaml.galuscak@cnb.cz. correspondng author, Banque de France, gabor.katay@banque-france.fr.

2 2 Kaml Galuščák and Gábor Kátay 1. Introducton Cross-country comparson of labour force partcpaton rates s the most straghtforward way to dentfy labour supply problems n a partcular country or set of countres. Internatonal organsatons, central banks and economc thnk-tanks often motvate ther polcy recommendatons by comparng a country s partcpaton rate usually broken down by selected sub-populatons aganst an nternatonal standard or aganst other smlar countres statstcs. The resultng recommendatons frequently nvolve but are not lmted to reform of the country s tax and welfare system to restore or mprove work ncentves. 1 That s, the fnancal ncentve to work s mplctly the prmary canddate for explanng cross-country dfferences n the partcpaton rates of apparently homogeneous groups of ndvduals. Is ths really the most mportant factor n explanng cross-country dfferences, or are other aspects such as dfferences n preferences or non-fnancal ncentves even more mportant? To what extent can dfferences n fnancal ncentves explan the observed dfferences between countres n labour force partcpaton? Although the mpact of labour taxaton and the welfare beneft system on labour force partcpaton has been wdely nvestgated n the emprcal lterature, the extent to whch cross-country dfferences n partcpaton rates can be explaned by the dversty of tax-beneft systems has never been addressed drectly. To fll the gap, we take the example of two countres, the Czech Republc and Hungary. We use, for both countres, a very detaled, complex mcrosmulaton model and perfectly comparable mcro estmates of labour supply at the extensve margn (.e. the partcpaton decson) to quantfy the part of the dfference n the two countres partcpaton rates accounted for by dfferent tax and welfare beneft systems. More precsely, we frst estmate the same labour supply model for Czech and Hungaran ndvdual-level data. 2 The two entrely comparable estmated equatons for the Czech Republc and Hungary are then used to smulate how the aggregate partcpaton rate would change n one country f t adopted the other country s tax and socal welfare system. To the best of our knowledge, ths paper offers the frst tentatve explanaton of cross-country dfferences n labour force partcpaton rates by dfferences n taxaton and welfare benefts. The closest strand of lterature uses cross-country ndvdual-level or dsaggregated data ether to nvestgate the labour supply effects of a specfc polcy change or to explan developments n partcpaton rates from a cross-country comparatve perspectve. Wthout the sake of completeness, some recent cross-country emprcal studes analyse the effect of hypothetcal 1 For example, see OECD (2012), OECD (2014) and IMF (2013) for assessments of labour force partcpaton n Hungary and the Czech Republc the two countres ths paper focuses on and for resultng polcy recommendatons. By comparng labour force partcpaton rates n OECD countres, OECD (2012) concludes that structural reforms are also needed (n Hungary) to better explot exstng resources and rase one of the lowest actvty rates n the OECD. Smlarly, IMF (2013) urges Hungary to rase the exceptonally low labor partcpaton rate. OECD (2014) welcomes the fact that Czech labour force partcpaton has ncreased to the European average, although the relatvely low labour force partcpaton of women s dentfed as a bottleneck for the Czech Republc as well. In order to catch up wth countres wth hgher partcpaton rates, IMF staff recommend that Hungary mplement a more employment-frendly taxaton for low ncome earners and rase women s partcpaton n the labor market by reorentng publc spendng from cash benefts [ ] towards the development of hgh qualty early chldhood educaton and day care centers. Smlarly, the OECD recommends that Hungary preserve work ncentves when lowerng the tax wedge to promote the partcpaton of the elderly and to reform famly polces to enhance women s labour market partcpaton. 2 The labour supply model used n ths paper bulds on a prevous work by Benczúr et al. (2014). The Hungaran estmates are also presented n ths paper.

3 Labour Force Partcpaton and Tax-Beneft Systems: A Cross-Country Comparatve Perspectve 3 welfare reforms (e.g. Immervoll et al., 2007); others study the effects of tax reforms (see e.g. Aaberge et al., 2000, or Paulus and Pechl, 2008, for a cross-country analyss of flat-tax reforms); cross-country mcrosmulaton models are also used to estmate governments redstrbutve preferences (Blundell et al., 2009; Bargan et al., 2011). Some other papers use ndvdual-level or dsaggregated data to descrbe the evoluton of labour supply n some selected countres, ether for the whole populaton (Balleer et al., 2009, or Blundell et al., 2011) or for a selected subpopulaton (e.g. Cpollone et al., 2013, for women). However, none of these papers or any other exstng emprcal ones have so far provded drect evdence to explan dfferences n partcpaton rates between countres. The Czech Republc and Hungary provde an nterestng case for comparson. The two countres exhbt a lot of smlartes: ther economes are geographcally close, are smlar n sze and level of economc development, and partly share a common hstory. In partcular, both economes experenced full employment untl the regme change at the end of the 1980s. Despte all ther common factors, ther partcpaton rates dffer markedly nowadays. In 2013, wth only 57.5% of the years old populaton workng or actvely seekng employment, Hungary recorded the fourth lowest partcpaton rate among the EU member states (behnd Croata, Italy and Malta), whle the Czech partcpaton rate was close to the EU15 average (64.3% n the Czech Republc and 64.8% n the EU15). 3 These dfferences are typcally explaned by dfferent labour market polces adopted durng the frst few years of the transton process and kept at least partly unchanged snce then. Both countres mplemented fast-track reforms n the early 1990s, wth two man dfferences. Frst, Hungary opted for case-by-case prvatsaton as opposed to the voucher prvatsaton method adopted by the former Czechoslovaka. Second, the Hungaran government ntroduced a strct bankruptcy regulaton n After an ntal collapse, the wdespread changes to the economc system shfted both economes to a relatvely fast growng path. Nevertheless, the dfferences n the prvatsaton methods and the draconan bankruptcy regulaton mplemented n Hungary provoked a much larger drop n employment n Hungary as compared to Czechoslovaka. Between 1990 and 1993, total employment declned by 9% n Czechoslovaka and by 22% n Hungary. Moreover, n Hungary, a number of polcy measures such as allevated condtons for enterng the old-age or dsablty penson systems contrbuted to pushng people out of the labour force rather than nto unemployment. As a result, the partcpaton rate declned contnuously untl 1997, when t reached only 50.6% of the years old populaton, about 10 percentage ponts lower than the EU15 average and more than 13 percentage ponts lower than n the Czech Republc. 4 As we wll see, n 2008, the Hungaran transfer system could stll be vewed as generous relatve to the Czech system. After presentng our methodology (n the next secton), we descrbe the key dfferences between the two countres tax and socal welfare systems (Secton 3). Our results (presented n Secton 4) show, frst, that the estmated labour supply elastctes for the Czech Republc are very close to the results for Hungary, suggestng that, at least n ths dmenson, ndvdual preferences are smlar n the two countres. Ths holds true even for sub- 3 Eurostat data 4 For a detaled analyss of the evoluton of the Hungaran partcpaton rate, see Kátay and Nobls (2009).

4 4 Kaml Galuščák and Gábor Kátay populatons dependng on the level of educaton, gender and martal status: n both countres, lower educated people, the elderly and marred women (of chldbearng age) are the most responsve to tax and transfer changes. Gven the large dsperson of the labour supply elastctes found n the emprcal lterature for dfferent countres and tme perods, the smlartes of the estmated elastctes for the two countres we analyse are n lne wth the evdence n Bargan et al. (2012) suggestng that a consderable part of the cross-country varaton n the elastctes found n the lterature s drven by methodologcal dfferences. Second, the smulaton results suggest that about one-half of the total dfference n the partcpaton rates of the years old populaton can be explaned by dfferences n the taxbeneft systems. The results are quas-symmetrc, meanng that f the Czech system was adopted, the Hungaran partcpaton rate would ncrease by about the same number of percentage ponts that the Czech partcpaton would decrease by f the Hungaran system were mplemented. The hghest responses are obtaned for marred women and women of chldbearng age. Ths s related to the much more generous maternty beneft system n place n Hungary as compared to the Czech Republc. 2. Methodology We frst estmate, for both Czech and Hungaran ndvdual-level data, a structural labour supply model where both taxes and transfers are treated n a unfed framework. In the second part, we use the two estmated equatons to smulate how each ndvdual s probablty of beng actve and hence the aggregate labour supply would change n one country f t adopted the other country s tax-beneft system. More precsely, we perform a mcrosmulaton exercse: we replace the net ncome varables (wages, transfers and non-labour ncomes) n the estmated equatons by ther values that would result from adoptng the other country s tax and transfer system. As a consequence of the benefts ganed or losses suffered due to changes n the effectve tax rates and the amount of (potental) transfers, ndvduals are expected to adjust ther labour supply accordng to the estmated probt equatons. The weghted average of the ndvdual changes n the partcpaton probabltes corresponds to the aggregate labour supply shock nduced by the hypothetcal reform. Under certan assumptons detaled later, the result of ths exercse reveals the extent to whch the dfference between the two countres partcpaton rates can be explaned by the dfferences n the taxaton and welfare beneft systems. In ths framework, the part of the cross-country dfferences n partcpaton rates that can be attrbuted to the dfferences n tax and welfare systems s assessed usng counterfactual smulaton scenaros. That s, the effects of the cross-country dfferences n tax-beneft systems are only ndrectly dentfed. Ideally, to address ths queston drectly, one would need to observe a country whch adopts the entre tax-beneft system of another country. Such quas-experment would allow us to use a more transparent dentfcaton strategy and to drectly estmate the mpact of such large-scale reform. 5 In the absence of such experment, the approach adopted n ths paper provdes a vable alternatve strategy. 5 See e.g. Mofftt (2002) for a lterature revew on program evaluaton methodology to assess behavoural responses to taxes and transfers.

5 Labour Force Partcpaton and Tax-Beneft Systems: A Cross-Country Comparatve Perspectve Modellng Labour Supply at the Extensve Margn An ndvdual s labour-supply decson s usually modelled as a trade-off between the utlty ganed from consumpton or lesure. Startng from a standard utlty functon charactersed by strctly postve margnal utltes wth respect to consumpton and lesure, the partcpaton decson s determned by the dfference between the market wage and the reservaton wage. In the standard textbook approach based on margnal calculus, non-partcpaton follows smply from the corner soluton of the model (Hausman, 1981). The theory yelds a bnary dependent varable specfcaton n whch the probablty of workng or actvely seekng work s modelled as a functon of the net market wage, net non-labour ncome and ndvdual preferences. In ths framework, taxes nfluence labour supply by affectng net market wages and net non-labour ncomes. 6 The early generaton Hausman model proved to be too restrctve n many respects. Frst, relyng on tangency condtons, the Hausman model s restrcted to the case of (pecewse) lnear and convex budget sets. Ths assumpton s partcularly restrctve f certan transfers get lost mmedately on takng up a job whch s usually the case and the wage earned durng the frst few hours worked does not compensate for the dscrete loss n benefts. As a consequence, the standard reservaton wage defned as the lowest wage rate at whch a worker would be wllng to work nonzero hours s undefned. 7 Second, quas-concavty of the utlty functon s mplctly mposed a pror. As dscussed by MaCurdy et al. (1990), the Hausman model requres the Slutsky condton to hold at all nternal ponts of the budget constrant. Ths restrcton s, however, typcally rejected by emprcal studes (see e.g. Blundell and MaCurdy, 1999) and may lead to based estmates by mplctly lmtng the range of elastctes that can be obtaned. Thrd, predctons usng the standard approach do not ft the data well, as the model does not account for the fact that very few observatons exst wth a small postve number of hours worked. Fnally, the model makes t dffcult to handle household members jont labour supply decsons. In response to these shortcomngs, the dscrete choce approach to modellng labour supply provdes a convenent alternatve to the contnuous hours methodology. Orgnally proposed by van Soest (1995), ths approach has become ncreasngly popular and qute standard n recent years. In ths framework, utlty-maxmsng ndvduals are supposed to choose between a few alternatve dscrete sets of hours of work, such as nactvty (zero hours worked), part-tme or fulltme. Reducng the maxmsaton problem to choosng among a dscrete set of possbltes yeldng dfferent utltes consderably smplfes the problem and provdes a smple yet rather general way of representng labour supply decsons n the presence of nonlnear and non-convex budget constrants. The model s n prncple very flexble. The only restrcton to the model s the mposton of ncreasng monotoncty n consumpton. Mathematcally, the utlty (U) that ndvdual (or household) derves from choosng alternatve j from J possble dscrete choces s represented by a random utlty model of the form: 6 For a comprehensve presentaton of the dfferent varants of the standard modellng approach and ther dentfcaton strateges, see e.g. Blundell and MaCurdy (1999) and Blundell et al. (2007) 7 To overcome ths defcency, fxed costs of work can be ntroduced, yet ths addtonal source of non-convexty s also dffcult to handle (see e.g. Bourgugnon and Magnac, 1990). Van Soest and Das (2001) argue that nonconvextes n general mply rather restrctve and mplausble forms for preferences.

6 6 Kaml Galuščák and Gábor Kátay U j cj 1 l j Vcj,1 l j, Z, j, (1) where Vcj, 1lj, Z, Vj s a postve determnstc term that represents the mean utlty across observatonally dentcal agents when alternatve j s chosen and j s a random term that s not correlated wth the determnstc part of the utlty functon. Vj s condtonal on the alternatvespecfc consumpton ( c j ) and lesure ( 1 l j, wth total tme endowment normalsed to 1), a set of ndvdual characterstcs Z ) and preference parameters. ( In our emprcal work, we model ndvdual labour supply decsons. We assume that there are only two labour-market states, actve and nactve. Indeed, n both the Czech Republc and Hungary, part-tme work s relatvely rare: before the outbreak of the current crss, the share of part-tme employees among all workers was less than 5% n both countres. By 2013, ths share had ncreased to 6.6% n the Czech Republc and to 6.7% n Hungary. For comparson, the share of part-tme employees n the EU15 reached 21% n 2008 and ncreased further to 23.6% n The partcpaton decson bols down, therefore, to a comparson of the utlty ganed from workng full-tme (j = 1) to the utlty ganed from stayng out of the labour force and gettng the full amount of transfers (j = 0). Indvdual therefore chooses to work full-tme f U 1 U0. Gven the budget constrant statng that consumpton must equal total ncome, the probablty of partcpaton s gven by: P V P, j1 PU 1w T T,1 l1 U0T,1 w T T,1 l, Z, V T,1, Z, 1 where P, j1 represents the probablty that ndvdual s economcally actve; w beng the net market wage of ndvdual f workng full-tme ( l l ); 1 T s the sum of all the hypothetcal socal transfers the ndvdual gets (or would get) at zero hours worked and loses when workng full-tme; and T s the total net non-labour ncome, ncludng socal transfers ( T ) and other non-labour ncome (pensons, dvdend payments, ncome of other members of the household, etc.) If the random components j exp exp j wth fxed varance, McFadden (1974) derves a formal expresson of the probablty that alternatve j s chosen and shows that the parameters can be estmated by maxmum lkelhood technques. 9 1 are..d. extreme value dstrbuted wth c.f.d. In ths paper, we take a very smlar, but somewhat dfferent approach. As n our case, only two labour market states exst, we can further smplfy the model to obtan a smple probt or logt. Let W w T be the gans to work from acceptng a full tme job, defned as the dfference between the net wage and the amount of transfers lost when workng. Snce U 0T,1 s ndependent of U W T, l s monotoncally ncreasng n consumpton and thus n W and (2) 8 Source: Eurostat 9 Ths model, however, exhbts the unpleasant property of Independence of Irrelevant Alternatves (IIA). To overcome ths defcency, emprcal studes usually ntroduce some unobserved heterogenety n preferences, that s, some of the parameters are assumed to be randomly dstrbuted. The estmaton technque for ths model, referred to as mxed logt or random parameter logt, s descrbed n detal n McFadden and Tran (2000).

7 Labour Force Partcpaton and Tax-Beneft Systems: A Cross-Country Comparatve Perspectve 7 W (for a gven res res W W T, 1 l1 T and 1 l 1 ), there exsts a unque reservaton gans to work such that: W W res T 1 l, U W T,1 l U T,1, (3) The ndvdual therefore chooses to partcpate f the gans from acceptng a full-tme job (the gans to work) exceed a certan threshold, the reservaton gans to work 10 : P res W T, 1l U T,1 PW W T, l, j1 P U (4) res Approxmatng W T, 1 l of beng actve: logt ' 1 by Z t yelds a formal expresson for the probablty P, j 1 G logw Z' logt (5) where G s the cumulatve dstrbuton functon of the standard normal dstrbuton (n what follows denoted by ) f t follows a centred normal dstrbuton or the cumulatve dstrbuton functon of the logstc dstrbuton f the errors are dstrbuted as logstc. 11 In ths framework, nstead of drectly estmatng the parameter of the utlty functon, we approxmate the reservaton gans to work by usng a set of observed ndvdual characterstcs ( Z ) and the total net non-labour ncome ( T ). 2.2 Estmaton Method In practce, not all the varables of eq. (5) are observed for all ndvduals. For employed ndvduals, we drectly observe ther gross wages, so we can compute ther net wages ( w ). 12 However, we do not observe the hypothetcal amount of transfers the ndvdual would get f he or she decded to stop workng. Usng ndvdual characterstcs and the detals of the welfare system, we calculate the amount of lost transfers ( T ) the ndvdual would be enttled to beng non-employed. Consequently, the gans to work are calculated as W w T, whle the amount of lost transfers ( T ) s added to the observed non-labour ncome to obtan T. For the non-employed, the total amount of transfers s observed, whether ncome-tested or not. 13 We do not observe, however, the market wage w. In the emprcal lterature, a Heckman model s usually used to estmates wages for the non-employed. The only dfference compared to the standard Heckman procedure s that nstead of wages, we drectly estmate the gans to work. We then use the predcted gans to work to estmate the 10 In the smple case when margnal utlty s constant, ( +, 1 ) (, 1 ) + (, 1 ) and therefore the reservaton gans to work s: = (, 1) (, 1 ) (, 1 ) 11 For brevty reasons, we only present the results for the probt equaton. Logt results are very smlar and are avalable upon request. 12 Gross wages nclude employer contrbutons as well. Accordngly, we take nto account both employer and employee-sde contrbutons and taxes when computng net wages. 13 We can also determne ΔT the same way we computed t for employed ndvduals by applyng the welfare rules. We used the observed values for the estmaton part; however, we used our own calculaton for the smulaton part. See Secton 2.3 for detals.

8 8 Kaml Galuščák and Gábor Kátay structural probt or logt eq. (5). 14 We used the predcted values also for employed ndvduals nstead of the observed ones n order to avod selecton bas assocated wth dfferent treatment of wages for employed and non-employed. 15 The major challenge for estmatng structural labour supply models s to deal wth endogenety. The dentfcaton usually reles on excluson restrctons that may be dffcult to justfy. Indeed, the estmaton procedure corresponds n fact to an IV probt or logt where the frst step s a Heckman model to predct labour ncomes. The key element of the dentfcaton s the presence of labour demand shfters n the wage equaton whch are mssng from the structural equaton and act lke excluded nstruments, that s, varables whch determne the market wage but only affect labour supply ndrectly through wages. Assumng that such vald nstruments are avalable, the endogenety and measurement error bases assocated wth logw can be corrected. 16 In our emprcal model, Z ncludes educaton dummes, household head, mother wth nfant (< 3 years old), attendng full-tme educaton, household sze (number of persons), pensoner, famly status (husband, wfe, chld, sngle, dvorced, etc.) and age-group (15 24, 25 49, 50+) dummes and year dummes. The excluded nstruments (.e. labour demand shfters) are age, age squared and educaton dummes and ther nteractons, and county dummes. The county dummes are proxes for local labour market condtons, whle age and age square and ther nteractons wth the educaton dummes are proxes for experence that nfluence market wages. In other words, once we control for ndvduals lfecycle poston (student, famly status, age group, retred, etc., all varables ncluded n Z ), we assume that an extra year has a neglgble mpact on labour supply drectly, but t strongly affects the wage and hence nfluences labour supply ndrectly. Once estmated, the structural probt estmaton gves us the coeffcents of logw and log T (γ and ψ respectvely) as well as the condtonal margnal effects of these varables evaluated at a certan vector, e.g. at the sample means on the probablty of beng actve. To obtan the mpact of the net wage ( w ) tself, t s easy to see that the margnal effect of logw has to be multpled by w w T. Smlarly, the effect of condtonal transfers on actvty s gven by log W 1 w w T logt T T. We report these margnal effects n Secton Smulatons The smulatons are carred out symmetrcally usng both the Hungaran and the Czech results. As a frst step, we predct the partcpaton probabltes usng the probt eq. (5) estmated for the Hungaran and Czech data as follows: 14 Note that we used the uncondtonal predcton, that s, the effect of the nverse Mlls rato s not ncluded. 15 see e.g. MaCurdy et al. (1990) and Löffler et al. (2014) for a dscusson on ths ssue. 16 Note that logw s subject to measurement error for varous reasons. For example, observed wages are based downward for part-tme workers or for ndvduals who work less than ~20 days per month. Lost transfers ( T) whch are part of logw may also be msmeasured due to possble lower take-up rates of some benefts and due to the fact that some transfers may be provded wth dscreton by socal offces. It s possble, however, to compare the smulated T wth the observed transfers for ndvduals who worked for some part of the year and were non-employed for the rest of the year, as both ther wages and ther transfers are observed. Ths comparson revealed that for both countres and for most ndvduals, the smulated transfers were close to the observed ones.

9 Labour Force Partcpaton and Tax-Beneft Systems: A Cross-Country Comparatve Perspectve 9 ˆ CZ, CZ CZ CZ CZ CZ ˆ j logw Z ˆ, 1 ˆ' P ˆ HU, HU HU HU HU HU ˆ j logw Z ˆ, 1 ˆ' P logt CZ logt HU (6a) (6b) ˆ C1, C2 P, j 1 denotes the predcted probablty of ndvdual n country C 1 = {CZ, HU} beng actve and usng the tax and transfer system of country C 2 = {CZ, HU}. Parameters wth superscrpts CZ and C1 C1 HU are estmated parameters for the Czech and Hungaran data respectvely. W and T are the gans-to-work and the non-labour ncome (ncludng condtonal transfers) varables calculated usng country C 1 s tax and transfer system. For example, n eq. (6a), we took the observed gross wages for employed Czech ndvduals and the estmated gross market wages for non-employed CZ CZ ndvduals and we appled the Czech tax-beneft system to compute W and T for each ndvdual. In eq. (6a) and (6b), we set the constant term so that the weghted average of the predcted probabltes exactly matches the aggregate partcpaton rate for the whole populaton. In addton to labour ncome taxes, we took nto account ndrect taxes (that s, the average effectve VAT rate, denoted as ) n our smulaton exercse. The gans to work therefore become w T W. Smlarly, non-labour ncomes are also dvded by As a second step, we convert all ncome varables n the Czech database to Hungaran fornt (HUF) usng the 2008 average exchange rate (CZK 1 = HUF 8.04). 18 Then we apply the Hungaran tax-beneft system to the Czech data to compute net ncome varables and thus to HU HU obtan W and T. Before calculatng the mpled predcted probabltes of beng actve, we convert the varables back to CZK. We perform symmetrcally the same exercse by applyng the Czech tax-beneft system to the Hungaran data. The predcted probabltes are gven by: ˆ CZ, HU CZ HU CZ CZ ˆ j logw Z ˆ, 1 ˆ' P ˆ HU, CZ HU CZ HU HU ˆ j logw Z ˆ, 1 ˆ' P logt HU logt CZ (7a) (7b) The dfferences between these predcted probabltes and those obtaned n the prevous step CZ HU CZ CZ that s, Pˆ, j Pˆ, HU CZ HU HU, 1, j1 and Pˆ, j Pˆ,, 1, j1 correspond to the ndvdual labour supply shocks nduced n one country by adoptng the other country s tax-beneft system. The weghted averages of these expressons gve the aggregate labour supply shocks trggered by the hypothetcal reforms. Whether the ntal labour supply shock fully translates nto partcpaton n the long run depends on the general equlbrum nteractons of the entre economy. For example, n a textbook neoclasscal general equlbrum long-run model of a small open economy, the overall effect prmarly depends on the assumpton made on the long-run elastcty of captal supply. In the perfectly elastc case whch s a standard assumpton n the lterature the expected rate of 17 In the estmaton part, the effect of VAT s captured by year dummes. 18 Alternatvely, measures that also reflect dfferences n prces or economc development, such as real exchange rates or the rato of the two countres nomnal GDP per capta, could be used. We also performed the smulatons usng these alternatve measures, but do not present them n ths paper. The results avalable from the authors upon request changed only margnally.

10 10 Kaml Galuščák and Gábor Kátay return on captal s pnned down by nternatonal benchmarks, whch mples that the captal-tolabour rato and the rato of factor prces stay constant n the long run. It follows that, for example, after a postve labour supply shock nduced by a tax cut, captal accumulaton wll follow untl the new equlbrum s reached wth an unchanged captal-labour rato. After the adjustment perod, gross wages wll also reach ther pre-reform level, so factor prces reman unchanged n the long run. In ths framework, the ntal labour supply shock fully translates n the long run nto partcpaton. Consequently, under these assumptons, the results of our smulaton exercses can be nterpreted as the effect of dfferences n the tax-beneft systems n the two countres on the dfferences n the two countres partcpaton rates Key Elements of the Czech and Hungaran Tax and Transfer Systems The Czech and Hungaran tax-beneft systems are both charactersed by ndvdual personal ncome taxaton and a broad range of welfare benefts. In what follows, we brefly descrbe and compare the two systems effectve n 2008, the reference year we used for our smulaton. 20 In ths secton, we only consder the man characterstcs of the systems that apply to regular wage earners; the self-employed and other specfc cases are not consdered. In our emprcal work, more detals are taken nto consderaton. For a more comprehensve descrpton of the two systems, see, for example, the country chapters of the OECD s Benefts and Wages publcaton. 21 For a detaled presentaton of the calculaton methods for net ncomes and (hypothetcal) transfers, see Galuščák and Pavel (2012) for the Czech case and Benczúr et al. (2014) for the Hungaran case. Addtonal detals on the transfer systems of the two countres are presented n the Appendx. In both countres, labour ncomes are subject to personal ncome tax and employer and employee contrbutons. In Hungary, the tax schedule comprses three brackets, wth tax rates of 18%, 36% and 40%. 22 Employees wth annual earnngs under a gven threshold are elgble for an earnng ncome tax credt (EITC) of 18% of ther wage ncome, whch s phased out at a rate of 9%. A chld tax credt (fxed amount) s avalable for famles wth three or more chldren. Socal securty contrbutons related to sckness, unemployment and pensons are appled. In the Czech Republc, famly taxaton for marred couples wth chldren was ntroduced n The jont taxaton, however, became rrelevant when a flat tax rate of 15% was ntroduced n It s appled to so-called super-gross ncome, whch ncludes employer socal securty contrbutons. Tax credts are avalable per person, per spouse under a gven ncome and per dependent chld. If the tax credt per chld s negatve, t s provded to the household as a tax bonus. 19 For a detaled explanaton of the long-run general equlbrum effects of a tax reform and the mplcatons of captal supply elastcty for the new equlbrum, see a closely related paper: Benczúr et al. (2012). 20 In the datasets we use, 2008 s the most recent wave avalable for both countres. Although snce we started to work on ths project, more recent waves became also avalable, we decded to stck to the last pre-crss year as the assumpton that the economes are close to the steady state s more plausble. Moreover, several large-scale tax and beneft reforms have been adopted n Hungary snce the outbreak of the crss. It s reasonable to assume that full adjustment to the changes takes tme, especally n crss perod The thrd bracket s n fact the second bracket plus a 4% surtax a.k.a. the soldarty levy whch was ntroduced n 2007.

11 Labour Force Partcpaton and Tax-Beneft Systems: A Cross-Country Comparatve Perspectve 11 Comparng the two personal ncome tax (PIT) systems, Fgure 1 reveals that, frst, the Hungaran system s more progressve and, second, average taxes are hgher, especally but not only for hgher ncome earners. Moreover, the Czech average tax rate can be negatve for famles wth chldren f the calculated ncome tax s lower than the amount of famly beneft the taxpayer s elgble for. The hgher average PIT rates n Hungary are also reflected n hgher tax revenues: the Hungaran government n 2008 collected 7.9% of GDP as PIT tax revenue, compared to 3.7% n the Czech Republc (see Table 1). Fgure 1: Average Earnng Tax Rate (AETR) n CZ and HU (2008) AETR t 4t 6t 8t 1m gross annual ncome AETR m 4m 6m 8m 10m gross annual ncome (a) CZ (b) HU Note: The average earnng tax rate s the total amount of tax pad on labour earnngs as a percentage of the tax base f all forms of ncome taxes, contrbutons, tax credts and bonuses are taken nto account. Both countres provde a wde range of socal benefts n order to at least partly compensate for temporary loss of labour ncome, to reduce and prevent poverty or to acheve other goals. Overall, f we look at the share of GDP spent on varous transfers n 2008, the Hungaran beneft system seems more generous than the Czech one (Table 1). 23 As a general rule, n Hungary, maternty benefts and unemployment benefts ether are fxed amounts or depend on prevous ncome and are subject to ncome taxaton. Other socal benefts and pensons are, on the other hand, exempt from taxaton. In the Czech Republc, all benefts are non-taxable (except very hgh pensons). In the case of maternty benefts, the amount s calculated based on prevous gross ncome, whereas unemployment benefts are calculated based on prevous net ncome. These dfferences make drect comparson of the two systems rather challengng. Some detals of the two welfare beneft systems are lsted n Table 1. For a more detaled descrpton, see the Appendx or the references presented n the frst paragraph of ths secton. 23 The Hungaran beneft system s often cted as beng relatvely generous compared to the systems of neghbourng countres based on the share of GDP redstrbuted, on the elgblty crtera for some benefts and on the amount of welfare transfers an ndvdual mght be elgble for. At the same tme, Hungary s not successful n ensurng access to transfers by the most vulnerable. In comparson wth the Czech Republc, the share of people at rsk of poverty or socal excluson s more than two tmes hgher n Hungary than n the Czech Republc accordng to Eurostat data (33.5% vs 14.6% n 2013). The Hungaran redstrbuton s arguably far from the egaltaran prncple.

12 12 Kaml Galuščák and Gábor Kátay Table 1: Key Dfferences n the Tax and Transfer Systems (2008) PIT Descrpton % of GDP CZ HU CZ HU Taxes on labour one tax rate (15% on supergross) + allowances (chldren), negatve AETR possble three tax rates (18%, 36%, 40%) wth ETC + allowances, no negatve AETR 3.7% 7.9% employee contr. 12.5% 17% 3.5% 3.2% employer contr. 35% Other taxes 32% HUF/month 9.9% 9.9% eff. tax on cons. 15.4% 18.2% 10.6% 15.6% eff. tax on sales % % eff. tax on captal 9.8% 7.3% 4.2% 2.6% maternty beneft UB other socal benefts old-age penson dsablty penson 6 months: 69% of prevous gross ncome (capped) frst 3 months: 50% of prevous net ncome; next 3 months: 45% of prevous net ncome. Capped; longer enttlement perod for the elderly Benefts frst 2 years: 70% of prevous ncome (capped), thrd year: fxed amount; longer enttlement perod for dsabled chld, twns or three chldren max 3 months: 60% of prevous ncome (capped); then max 6 months: fxed amount; then max 3 more months: lower fxed amount; longer enttlement perod for the elderly 0.2% 0.7% 0.2% 0.4% fxed amount fxed amount 1.2% 1.2% offcal: 61 years and 10 months for men, 60 years for women (less wth chldren); effectve: 62.6 for men, 58.8 for women offcal: 62 years; effectve: 59.5 for men, 58.5 for women 5.8% 9.7% 1.5% 2.4% Source: Natonal Accounts (HU and CZ), ESCB Publc Fnance Report (2014), Czech Mnstry of Labour. Notes: The effectve tax on consumpton s the sum of ndrect taxes dvded by fnal consumpton expendture; the effectve tax on captal s corporate tax revenues dvded by the share of captal revenues based on the Natonal Accounts. Other socal benefts nclude: (CZ) total state socal support (ncludng parental allowance and brth grant) and materal need benefts and (HU) regular socal assstance beneft, socal supplement, other benefts (e.g. retranng), care allowances and other rregular socal benefts.

13 Labour Force Partcpaton and Tax-Beneft Systems: A Cross-Country Comparatve Perspectve Results The estmatons are carred out usng SILC ( Survey of Income and Lvng Condtons ) data for the Czech Republc for the years The SILC s a yearly cross-secton survey of households whch provdes detaled nformaton on demography and ncomes. For each year, the survey contans broadly 10,000 households and 23,000 ndvduals. All ncome varables (labour and non-labour ncomes, transfers) are expressed n total annual terms. To compute net ncomes and potental transfers that ndvduals are enttled to or would be enttled to n the absence of any ncome from work, we used a modfed verson of the tax-beneft model developed by Galuščák and Pavel (2012). We used the same codes to compute net labour ncomes, but we extended the orgnal verson of the model n order to take nto account more complex features of the Czech tax and transfer system (e.g. taxes on captal revenues, maternty beneft, etc.). As for the transfers, we smulated the total yearly amount of benefts the ndvdual gets f he s not workng or could get f he decded to stop workng. 24 The estmatons for Hungary were carred out usng the Hungaran Household Budget Survey (HBS) database for the years The Hungaran HBS s more detaled but otherwse perfectly comparable to the CZ-SILC database. Frst, several ncome varables n the Czech database are broken down nto several categores n ts Hungaran counterpart. Second, some varables such as ncome from abroad, ncome of mnors, lfe annuty payments, scholar grants and severance ndemntes are reported n the Hungaran database but no nformaton s avalable from the Czech database. For the smulaton exercse, we carefully lnked all the varables to ther counterparts n the other database. If one-to-one correspondence could not be acheved, the closest possble several-to-one match was establshed. 4.1 Estmaton Results Table 2 reports the baselne probt estmates, followng the methodology presented n Secton 2. As expected, hgher gans to work ncrease the probablty of partcpaton n the labour market, whle non-labour ncome has the opposte effect. Overall, the Czech results are remarkably close to the Hungaran ones. The pont estmates and the condtonal margnal effects of gans to work evaluated at the populaton averages are bascally the same n the two countres. Ths s also true for the condtonal margnal effect of net wages, suggestng that, on average, a smlar change n the average tax rate has the same effect n both countres. As for non-labour ncome, the estmated coeffcent and the margnal effect are lower n absolute value n the Czech Republc than n Hungary, but the results for the condtonal margnal effect of transfers are very close to each other. Ths seemngly confusng result s a reflecton of the more detaled non-labour ncome (ndependent of workng status) data used for the Hungaran estmaton. As explaned n Secton 2.2, the margnal effect of condtonal transfers ( T ) s obtaned by usng the followng formula: log W 1 w w T logt T T. Gven that the Hungaran non-labour ncome contans more nformaton (T ncludes certan types 24 The orgnal tax-beneft model n Galuščák and Pavel (2012) nstead smulates the amount of monthly benefts an ndvdual can get at the moment of the survey.

14 14 Kaml Galuščák and Gábor Kátay of ncomes mssng from the Czech database), the share of the potentally lost transfers n total non-labour ncome T T s hgher n the Czech database. As a consequence, the (lower, n absolute value) Czech result for the margnal effect of non-labour ncome s multpled by a relatvely hgher postve value, whch leads to a smlar estmate for the condtonal margnal effect of transfers. Table 2: Man Results CZ (1) (A) Estmaton results HU coef. std. err. coef. std. err. gans to work non-labour ncome (B) Condtonal margnal effects dy/dx std. err. dy/dx std. err. gans to work non-labour ncome net wage transfer Source: CZ-SILC database, and Hungaran HBS database, The Hungaran results are also avalable n Benczúr et al. (2014). Notes: Probt estmates, as descrbed n the paper, usng a sample of ndvduals aged Standard errors are bootstrapped wth 200 replcatons. The structural probt equaton ncludes: log of gans to work, log of non-labour ncome, mother wth nfant (less than 3 years old), full-tme student and educaton (less than elementary school, elementary school, vocatonal, secondary educaton, tertary educaton) dummes, age-group dummes (15 24, 25 49, 50+), pensoner, gender and head of household dummes, household sze and famly status dummes (sngle, marred lvng together, marred lvng separately, wdow(er), dvorced), household membershp status dummes (husband, wfe, companon, sngle parent, chld, ascendant, other relaton, non-relaton, sngle), year dummes. The controls ncluded n the Heckman wage model whch are mssng from the structural probt are: county dummes, nteracton of age and age square wth educaton dummes. (2) We are ultmately nterested n the effects of tax and transfer changes, that s, the condtonal margnal effects of the net wage and transfers. On average, a 10% ncrease n the net wage leads to a 3.7 percentage pont ncrease n the probablty of beng actve n the Czech Republc and a 3.9 percentage pont ncrease n Hungary. A 10% ncrease n transfers decreases the average ndvdual s probablty of beng actve by 1.2 percentage ponts n the Czech Republc and by about 1.4 percentage ponts n Hungary. These strkngly smlar results suggest that ndvdual preferences are quas-dentcal n the two countres. We report n Table 3 the condtonal margnal effects by selected subgroups dependng on level of educaton, gender and martal status. Agan, the Czech results are very close to the Hungaran ones. Panel A shows that the responsveness of partcpaton to changes n both net wages and transfers s lower for persons wth hgher (tertary) educaton. The condtonal margnal effects calculated on the prme-age populaton averages (25 54 years old, reported n Panel B) are lower

15 Labour Force Partcpaton and Tax-Beneft Systems: A Cross-Country Comparatve Perspectve 15 (n absolute value) than the results for the whole populaton, ndcatng that prme-age ndvduals are less responsve to tax and transfer changes than the rest of the populaton. Wthn the prmeage populaton, however, the educaton level strongly nfluences ndvduals responsveness: the lower educated ( elementary school or less ), probably hghly transfer-dependent, group s hghly responsve, whle the estmated elastctes for secondary and tertary educated people are much smaller. The remanng rows n panel B suggest that women s partcpaton s more responsve that that of men, partcularly for marred women and women of chld-bearng age. Fnally, the results n panel C show that the reacton of elderly persons to changes n the net wage and transfers s even hgher. The estmaton results and the mpled wage elastctes are qualtatvely n lne wth exstng results n the emprcal lterature. However, drect comparson of the elastctes s not straghtforward. There s a large consensus n the lterature that women (and especally marred women) respond more to changes n ther net market wage than men. Lower educated and elderly ndvduals are also usually found to be more responsve than the average. The varaton n the magntude of the estmated labour supply elastctes found n the lterature s nonetheless consderable and these reported elastctes are rarely drectly comparable for several reasons. Frst, wth some exceptons, early labour supply models usng the contnuous Hausman approach usually do not dstngush between the decson to partcpate (the extensve margn) and the decson regardng hours worked (the ntensve margn) and only report uncompensated and/or compensated (Hcksan) hours elastctes, not the partcpaton elastcty. 25 Second, although the dscrete choce approach always ncludes, by constructon, the partcpaton elastcty, estmates from varous studes are not easly comparable due to dfferences n specfcatons and defntons. A recent comprehensve comparson of the nternatonal evdence on labour supply elastctes by Bargan and Pechl (2013) shows that the large varance n the reported elastctes s at least partly explaned by dfferences n specfcatons (most mportantly, whether wages are predcted for all ndvduals or only for non-workers) and the tme perod consdered. A meta-analyss by Chetty et al. (2013) provdes a consensus elastcty: accordng to the authors, exstng mcro and macro evdence ponts towards an aggregate steady-state elastcty of labour supply at the extensve margn of For a comparatve overvew of the hours elastctes stemmng from the Hausman approach, see e.g. Pencavel (1986) for marred men, Kllngsworth and Heckman (1986) for marred women, or the more recent survey by Everest et al. (2008). Other lterature surveys nclude Blundell and MaCurdy (1999), Blundell, MaCurdy and Meghr (2007), Keane (2011) and Meghr and Phllps (2010), whch focus mostly on analyses based on the standard contnuous approach, but also nclude some more recent evdence from dscrete choce methodology.

16 16 Kaml Galuščák and Gábor Kátay Table 3: Condtonal Margnal Effects by Selected Subgroups CZ all ndvduals elementary school or less secondary educaton tertary educaton prme-age subsample elementary school or less secondary educaton tertary educaton sngle men sngle women marred men marred women women of chld-bearng age (25 49) elderly (50+) HU dy/dx std. err. dy/dx std. err. (A) Full sample net wage transfer net wage transfer net wage transfer net wage transfer (B) Prme-age subsample net wage transfer net wage transfer net wage transfer net wage transfer net wage transfer net wage transfer net wage transfer net wage transfer net wage transfer (C) The elderly net wage transfer Notes: Condtonal margnal effects computed usng the estmaton on the full sample and evaluated at the subgroup-specfc mean values of the covarates; Hungaran results are also avalable n Benczúr et al. (2014).

17 Labour Force Partcpaton and Tax-Beneft Systems: A Cross-Country Comparatve Perspectve 17 Overall, the margnal effects obtaned for the Czech Republc and Hungary le n the (rather large) range obtaned n prevous studes, especally studes usng the dscrete choce approach. 26 Gven the large dsperson of the labour supply elastctes estmated for dfferent countres and tme perods and usng varous methodologes, the smlartes of the results we obtan for two dfferent countres s n tself nterestng and we tentatvely suggest that methodologcal dfferences mght be responsble for a consderable part of the cross-country varaton n the elastctes found n the lterature. The fndngs of Bargan et al. (2012), who estmate a comparable labour supply equaton for a large set of countres and show that cross-country dfferences n genune work preferences are rather small, corroborate ths vew. 4.2 Smulaton Results The smulatons were carred out as explaned n Secton 2.3. The results are shown n Table 4. The fgures reported ndcate the partcpaton rates for 2008 (the reference year chosen for our smulatons) for both countres on the overall sample of ndvduals aged 15 to 74 (rows (a) to (g)), for the prme-age subsample (rows (h) to (m)) and for elderly people (row (n)). The partcpaton rates are consderably hgher n the Czech Republc than n Hungary accordng to offcal Eurostat statstcs (see columns (1) and (6)). Columns (2) and (7) show that the smple aggregate partcpaton rates obtaned from our datasets are very close to the offcal Eurostat data. The dfference n partcpaton rates s 8.5 percentage ponts accordng to the Eurostat data and 6.6 percentage ponts n our datasets. The smulaton results usng each country s own tax and beneft systems are presented n columns (3) and (8). The reported numbers are weghted averages of the partcpaton probabltes predcted by eq. (6a) and (6b). As mentoned n Secton 2.3, the constant terms n eq. (6a) and (6b) are set so as to match the aggregate statstcs. The smulated full-sample aggregate partcpatons (cells (a3) and (a8)) are therefore equal to the observed partcpaton rates (cells (a2) and (a7)). The dfferences n the partcpaton rates for the selected subgroups (columns (3) and (8)) come from dfferences n preferences (captured by the labour supply controls n eq. (6a) and (6b)), dfferences n productvty and thus n gross wages, dfferences n average tax rates, whch nfluence net wages, and/or dfferences n potental transfers. For both countres, the smulated partcpaton rates for the selected subgroups are close to the observed ones. Ths suggests that overall, the estmated equatons capture ndvduals heterogenety qute well. The predcted aggregate partcpaton rates usng eq. (7a) and (7b) are reported n columns (4) and (9). In columns (5) and (10) we show how Czech and Hungaran labour force partcpaton would change (n percentage ponts) f one country adopted the other country s tax-beneft system. The results of our smulaton exercse show that, overall, the Czech partcpaton rate would decrease by 2.9 percentage ponts f the country adopted the Hungaran system, whle the Hungaran partcpaton rate would ncrease by 3.0 percentage ponts f the Czech system were adopted. These results are hghly symmetrc and ndcate that the dfferences n the two tax-beneft systems explan about half of the total dfference n the two countres partcpaton rates. The dfference s 26 As noted n Bargan and Pechl (2013), the tradtonal contnuous approach seems to overstate the elastctes compared to dscrete choce models. For example, the own-wage elastctes for marred women are relatvely large usng the contnuous technque (n several cases, larger than 1), whereas the recent dscrete choce approach ponts towards smaller elastctes (rangng between 0.1 and 0.5 wth some exceptons).