Lifetime versus Annual Tax Progressivity: Sweden, *

Transcription

1 Lfetme versus Annual Tax Progressvty: Sweden, * Nklas Bengtsson, Bertl Holmlund, and Danel Waldenström November 10, 2011 Abstract: Ths paper analyzes the evoluton of tax progressvty n Sweden from both an annual and a lfetme framework. Usng a rch mcro panel wth admnstratve records over ncomes, taxes and benefts over the perod , we compute measures of average tax rates across the ncome dstrbuton and account for the relatve mportance of dfferent tax bases as well as the role of transfers. The unquely long tme span allows us to estmate actual lfe cycle progressvty. Our man fndng s that Swedsh taxes are consderably less progressve over the lfetme than n any sngle year. In fact, lfe cycle taxes are close to proportonal bearng a redstrbutve effect of only a few percent. Ths result seems manly drven by ntrageneratonal ncome moblty, but also to some extent by the major tax reform of the The role of labor ncome taxes for progressvty has declned markedly whereas ncreased transfers to unemployed and old-age pensoners has pushed up progressvty. JEL: D31, H20 Keywords: Tax progressvty, Income dstrbuton, Lfetme ncome, Redstrbutve effect, Transfers. * We have receved valuable comments by are grateful to Andreas Pechl, Alan Trannoy and partcpants at the Natonal Conference n Economcs n Sweden, Lund Waldenström acknowledges fnancal support from the Jan Wallander and Tom Hedelus Foundaton and Rskbankens Jubleumsfond. Contact: Department of Economcs, Uppsala Unversty, P.O. Box 513, SE Uppsala, Sweden. E-mal: nklas.bengtsson@nek.uu.se, bertl.holmlund@nek.uu.se, danel.waldenstrom@nek.uu.se.

2 1. Introducton Tax reforms are often motvated by a poltcal wll to shft the burden of taxes accordng to the dfferent abltes to pay taxes across the ncome dstrbuton. But a fundamental problem wth conventonal assessments of tax burdens s that they typcally rely on annual crosssectonal outcomes. Incomes vary over the lfe cycle, wth young people often beng lowncome earners regardless of whether they are future hgh-pad surgeons or low-pad clerks. Old-age pensoners do not pay payroll taxes even though they may well earn more than younger ndvduals n the labor force. Captal ncomes are typcally observable and taxed when they are realzed rather than when they accrue, and such one-shot realzatons mght not truthfully depct the lfetme ncome status or lfetme tax burden. Accountng for lfetme varatons n both ncomes and ablty to pay taxes s mportant n order to make a balanced assessment of the trade-off between the equty and effcency of the tax system. In ths paper we analyze the mplcatons of studyng tax progressvty n an annual versus a lfetme perspectve. We do ths by explotng a rch data source wth regster nformaton on taxes pad and benefts receved by a large and natonally representatve sample of ndvduals. Usng a 42-year long panel we are able to compute measures of lfetme tax progressvty relatng nformaton about actual lfetme tax payments and actual lfetme ncomes for varous parts of the dstrbuton of lfetme ncomes. The use of such a long panel appears to be a unque contrbuton to the lterature. Prevous studes of the redstrbutve mpact of taxes over the lfetme have typcally been based on ether smulaton methods or much shorter panels. In partcular, our work extends the prevous work on tax progressvty usng mcrodata that started wth Pechman and Okner (1974). Prevous studes usng the lfe-cycle perspectve usng mcrodata nclude Daves, St-Hlare and Whalley (1984), Slemrod (1992), Fullerton and Rogers (1993), Cameron and Creedy (1994), Creedy and van de Ven (2001), and, on Swedsh data, Björklund, Palme and Svensson (1991). 1 These studes have ether reled on qute strong data assumptons or been restrcted to much shorter tme perods. The rchness and sze of our data a sample sze of about 200,000 ndvduals per year allow us to compare narrow ncome segments n the top of the ncome dstrbuton, such as percentles and 1 There s a related and more extensve lterature analyzng ncome nequalty over annual and lfetme horzons, payng lttle or no attenton to taxaton and tax progressvty (see Creedy, 1999, for an overvew). Studes of Sweden nclude Blomqust (1981), Björklund (1993), Hussenus and Selén (1994) and Pettersson and Pettersson (2003). 1

3 tenth of percentles. Such focus s of partcular relevance when pnpontng the dfferng mpacts of labor and captal taxaton. Another contrbuton of our paper s to provde a comprehensve assessment of how the redstrbutve propertes of the Swedsh tax system have evolved over the past decades. The Swedsh tax system has undergone major changes over the past 40 years. The overall tax burden has ncreased and government tax revenues have gradually become more dependent on socal securty contrbutons and value added taxes. Some specfc reforms are partcularly noteworthy. In 1971, the tradtonal system wth jont taxaton of marred couples was replaced by a system n whch each spouse pays taxes on hs or her own ncome. The tax reform of 1991 called the tax reform of century for ts groundbreakng mpact nvolved substantal cuts n margnal ncome taxes along wth the ntroducton of a dual ncome tax system where earned ncome and captal ncome are taxed at dfferent rates. More recent reforms nclude the abolshment of the nhertance tax, the gft tax and the wealth tax as well as the ntroducton of a system wth earned ncome tax credts. In lne wth conventonal termnology, taxes are referred to as progressve f the average tax rate ncreases wth ncome. We compute measures of average tax rates, typcally net of transfers, for varous segments of the ncome dstrbuton and examne how these rates change over tme, notng that the effectve degree of tax progressvty depends on the composton of the tax base as well as the statutory tax rates. We also compare the contrbuton of dfferent taxes such as the ncome tax, the payroll tax, the wealth tax and the value-added tax, to the overall level of tax progressvty. As n Pketty and Seaz (2007), we pay partcular attenton to tax progressvty at the top of the ncome dstrbuton. Several mportant results come out of the analyss. Frst, we fnd that lfetme tax progressvty s lower than tax progressvty n any sngle year. Ths has manly to do wth consderable wthn-lfe redstrbuton, where amounts receved as student or old-age support almost make up for the taxes pad for most ncome earners. 2 Our emprcal evdence at the ndvdual longtudnal level usng actual ncomes receved and taxes pad thus renforces much of the prevous smulaton evdence on lfe cycle progressvty. Second, we show that the dscrepancy between annual and lfetme tax progressvty reflects the transtory nature of low-ncome 2 Ths fndng gves emprcal backng to the notons of Bergh (2005) that the redstrbutve achevements of the welfare state cannot be assessed only from comparng pre- and post-fsc dstrbutons. 2

4 status rather than the transtory nature of hgh ncomes. Many of the ndvduals earnng low or zero market ncome do thus not belong permanently to the bottom of the ncome dstrbuton: they can be workers temporarly outsde the labor market, beng unemployed, n educaton or on sckness leave. These ndvduals appear to be greatly favored by the tax-cumbeneft system usng annual data as opposed to lfetme estmates. By contrast, transtory hgh ncome shocks, such as large realzed captal gans, fall manly on those who have already hgh permanent ncomes. At the top of the ncome dstrbuton, annual progressvty estmates therefore correlate hghly wth lfetme tax burdens. Thrd, we document that Swedsh tax progressvty has followed an nverted U-shape over the past four decades, ncreasng sharply n the 1970s and droppng n the 1990s and 2000s. The exact characterzaton of the experence depends on whether transfers are ncluded. When only consderng actual taxes, the prmary source of the varaton n progressvty appears to be changes n the tax system, n partcular the tax reforms of 1971 and 1991, rather than trends n the dstrbuton of market ncomes. The dramatc rse n unemployment and thereby assocated transfers durng the economc crss of the 1990s ncreased the tax-andtransfer progressvty. Comparng Sweden s progressvty experence wth that of Great Brtan and the U.S., taxes n Sweden appear less progressve, a fact prmarly due to the hgh levels of ncome and payroll taxes pad by low-ncome earners. Fourth, our decomposton of tax bases, n whch we not only nclude the standard ncome and wealth taxes but also payroll and consumpton taxes, reveals drastc restructurngs over the study perod. In partcular, payroll taxes have become ncreasngly mportant whereas the captal taxaton (whch ncludes taxes on captal ncome, real estate, and wealth) has dmnshed substantally. The remander of the paper s structured as follows. Secton 2 offers a bref overvew of some key features of the Swedsh tax system. In secton 3 we dscuss the bass of comparng annual and lfetme outcomes and dscuss varous measurement ssues. Secton 4 s devoted to a study of the evoluton of tax progressvty by means of cross-secton and panel data wth nformaton on ncomes and tax payments. Secton 5 concludes. 2. The Swedsh Tax System The total Swedsh tax-to-gdp rato stood at 40 percent n 1970 and has hovered around 50 percent durng the 2000s. Sweden used to have the hghest tax-to-gdp rato n the world, but 3

5 ths poston has n recent years been overtaken by Denmark. In fact, the Swedsh tax-to-gdp rato has fallen by four percentage ponts snce It should be noted that most transfers, such as pensons and socal nsurance benefts, are treated as taxable ncome n Sweden, a fact that tends to exaggerate dfferences relatve to countres where benefts are tax exempt. Taxes on labor are the most mportant source of tax revenues, accountng for more than half of the revenues. These taxes nclude personal ncome taxes to central and local governments as well as socal securty contrbutons pad by employers. The local government ncome tax rate s proportonal and stood at 20 percent of taxable ncome on average n 1970 and had ncreased to 31 percent n Snce 1991, the central government ncome tax only apples to ncomes above a threshold; some 20 percent of ncome earners are subject to the state tax whch nvolves two brackets, 20 and 25 percent. Payroll taxes, or socal securty contrbutons, are pad by employers n Sweden. Ther standard rate amounted to 11.6 percent of the wage bll n By 2009, the standard rate stood at 31.4 percent. These contrbutons are, however, not entrely equvalent to taxes snce they are partally matched by ncreased beneft enttlements, such as hgher retrement benefts. The socal contrbutons are proportonal to the wage bll snce 1982 (although wth some age varaton from 2006 and onwards). Before 1982, however, contrbutons were markedly nonlnear wth low rates for low earnngs, hgher rates for ntermedate earnngs (the standard rate), and agan much lower rates for top earnngs. The average payroll tax rate was therefore hgher than the margnal rate for workers wth hgh earnngs. Taxes on consumpton and nput goods nclude a value added tax (VAT). By 2008, these taxes contrbuted to more than a ffth of all tax revenues. The value added tax was ntroduced n 1969 and the standard rate was then set at 11 percent; numerous servces were however exempted from VAT. Snce 1991, the standard VAT rate has stood at 25 percent of the tax base. The effectve VAT rate s much lower than the standard rate snce some goods and servces are taxed at lower rates (6 or 12 percent) and some are exempt from taxaton. There are also a few excse taxes whch n total account for almost 8 percent of tax revenues n These taxes nclude sn taxes on alcohol and tobacco as well as taxes on energy, carbon and motor vehcles. 4

6 From 1991 and onwards, Sweden has practced a dual ncome tax system wth separate rules for earned ncome and captal ncome. Indvdual captal ncome (nterest, dvdends and captal gans) s subject to a proportonal ncome tax of 30 percent. 3 The corporate ncome tax s 28 percent of net taxable proft as of Inhertance and gft taxes were abolshed n 2004 and the wealth tax has been abolshed as from As from 2008, the property tax on real estate was abolshed and replaced by a muncpal property charge, a reform that substantally reduced property taxes for the more expensve dwellngs. By 2008, captal taxes accounted for 11 percent of the total tax revenues. A recent nnovaton of the Swedsh tax system s the ntroducton of an earned ncome tax credt from 2007 and onwards. Ths reform reduces margnal taxes on low ncomes by rasng the threshold for ncome tax lablty. A noteworthy feature of the reform s that the lower taxes apply only to earned ncome, not to transfers such as unemployment or sckness benefts. The reform thus ncreases the after-tax dfference between ncome from work and ncome durng non-work. 3. Measurng Annual and Lfetme Incomes, Taxes and Progressvty The ratonale behnd makng lfetme estmates of ncome nequalty and tax progressvty s that the same ndvduals tend to appear n dfferent tals of the ncome dstrbuton n dfferent years. Annual snapshots of tax progressvty therefore capture both redstrbuton between ndvduals wthn years and redstrbuton across years for the same ndvduals. Ths means that the dfference between annual and lfetme progressvty bols down to two dmensons: how transtory annual market ncome s and the extent of nsurance elements n the tax system. 4 Regardng the transtory nature of ncomes, t s a well-known fact that annual ncomes are more volatle than longer-run ncomes and that lfetme ncomes are more equally dstrbuted than annual ncomes (for an early treatment on Swedsh data, see Blomqust 1981). Part of ths s because annual ncomes reflect a cross-sectonal snapshot of the current populaton, capturng dfferences between younger ndvduals, who have recently entered the labor mar- 3 The effectve captal gans tax rates have vared somewhat, e.g., wth cuts n the early 1990s and a generally lower effectve rate on real estate sales (only two thrds of such gans taxable). 4 See also Fullerton and Rogers (1993) for a comprehensve dscusson. 5

7 ket, experenced workers, and senors who have left the labor market. In welfare states, a tax on labor ncome effectvely dstrbutes ncome from those who are n the labor force to those who have not yet entered t and those who have left t. A related ssue s the role of captal ncome gans. On tax records, captal gans appear n a qute lumpy fashon, placng some ndvduals n the top of the ncome dstrbuton although they do not regularly show up there. From a permanent ncome perspectve, captal ncome gans should be measured when they accrue rather than when they are realzed. If captal ncome realzatons place ndvduals n the top of the ncome dstrbuton temporarly and f captal ncome gans are taxed less heavly than labor earnngs, the annual estmate of tax progressvty would appear regressve at the top compared to the lfetme perspectve. 5 The second dmenson of lfetme progressvty pertans to the role of socal securty. In any gven year, the bottom of the ncome dstrbuton ncludes ndvduals who are temporarly out of work (beng, e.g., sck, unemployed or on parental leave). Transfers targeted at these ndvduals wll renforce the effect of a strong degree of annual tax progressvty. The annual average tax rate at the bottom of the annual ncome dstrbuton would be zero, whereas the effectve tax rate, whch defnes transfers as negatve taxes, would be negatve (see further below for a dscusson of these concepts). However, to the extent that socal securty transfers are pad and receved by the same ndvduals, the tax and transfer system wll appear less progressve n a lfetme perspectve. In Sweden, replacement ratos have occasonally been about 90 percent, whch means that every unt of currency SEK collected n socal securty mples close to a one-to-one ncrease n expected beneft. Some analysts therefore regard the payroll tax as an actuarally far nsurance premum rather than a tax, wth no effect on vertcal redstrbuton at all. In realty, however, the socal securty system contans caps, exceptons and elgblty constrants, makng t much less far n an actuaral sense and more taxlke. Annual estmates of tax progressvty therefore need to make an ex ante stand on how much of the payroll tax s really a tax, whch s a very delcate task (see, e.g., Sørensen, 2010, p. 211). In contrast, lfetme estmates of tax progressvty nets out taxes that are transferred to the same ndvdual automatcally. 6 Notably, t s not certan that a tax system that exhbts tax progressvty from an annual per- 5 For ths reason, Pketty and Saez (2007) rank tax unts based on ncome net of captal gans (and then add back captal ncome gans when calculatng the tax rates). 6 See also the dstncton between nsurance and redstrbuton n Hoynes and Luttmer (forthcomng). 6

8 spectve wll exhbt progressvty from a lfetme perspectve. Imagne two ndvduals, A and B, n two perods. A has a frst-perod ncome of 40 and second-perod ncome of 60, whereas B receves 50 n both perods. Suppose A pays 10 percent n taxes the frst year and 60 percent n taxes the second year, whereas B always pays 50 percent. Ths system s clearly progressve n any sngle year. However, by summng ndvdual ncome and taxes over tme, B faces a hgher lfetme tax rate (50 percent) than A (40 percent), despte the fact that they are equally rch (both have an lfetme ncome of 100). The above example makes t clear that annual and lfetme estmates of tax progressvty can dffer substantally. The dstncton between horzontal redstrbuton (across tme) and vertcal redstrbuton (across ndvduals) s necessary to understand ths dfference, n partcular n understandng the rse (and presumed demse) of the modern welfare state. From a poltcal economy perspectve, nsurance elements n the tax-and-transfer system can justfy publc sector spendng; socal nsurance, f properly desgned, can be less dstortonary than redstrbutve taxes wth hgh margnal tax rates. Drawng partly on ths dea, Lndert (2004) notes that countres wth a hgh share of publc spendng relatve GDP (lke Sweden) have tradtonally had a more proportonal tax system, at least compared to the Anglo-Saxon countres. Our vew s that such a story needs to consder data on the lfetme dstrbuton of the tax burden n order to be consdered complete. 3.1 Average Tax Rates We wsh to descrbe how average tax rates vary by ncome. To serve ths am, we consder two measures of average tax rates. The frst measure, arguably the most conventonal, relates total tax payments to the tax base, pre-tax total ncome. In Sweden, the tax base ncludes market ncome as well as taxable transfers such as unemployment benefts and sckness benefts. We thus have T (1) t Y B, where T s total tax payments, Y s pre-tax market ncome (also referred to as gross ncome) and B taxable transfers; subscrpt refers to ndvdual. We shall refer to the sum of Y and B as pre-tax total ncome or taxable ncome. 7

9 Ths measure, referred to as the average tax rate (or ATR for short) s conventonally referred n nternatonal comparsons, but t suffers from a number of conceptual shortcomngs concernng progressvty measurement. Frst, the ATR does not drectly account for the redstrbutve effects of transfers, whch can be dsproportonally targeted to poor ndvduals. As such, t s a conceptually dffcult to regard transfers B as part of the tax base, snce t s n turn a functon of Y. Taxable transfers and market ncome are perfect substtutes n the short run, so n the annual perspectve ths need not be a severe ssue. But from a lfetme perspectve, calculatng average tax rate n ths fashon s not optmal. By addng taxable transfers to market ncome Y, pre-tax total ncome effectvely double-counts transfers wthn the socal securty system. B A second, more comprehensve measure, s the effectve tax rate (ETR for short), whch not only ncludes taxes pad but also transfers receved, notng that they are n prncple negatve taxes. 7 The ETR s n ths way a functon of dsposable ncome relatve to gross ncome and defned as (2) t Y D e Y where dsposable ncome ( D ) ncludes after-tax ncome from labor and captal as well as transfers after tax (ncludng transfers that are not taxed). ETR takes nto account that the taxcum-beneft system becomes more progressve f transfers are dsproportonately targeted at ndvduals wth low ncomes. Effectve tax rates wll typcally be negatve for ndvduals wth very low pre-tax ncomes. Snce no upper age lmt s mposed, old age ndvduals who receve pensons (whch are transfers, not ncluded n market ncome) wll thus usually have negatve ETRs when usng annual fgures. ETR goes to nfnty as market ncome approaches zero and s not defned for ndvduals wthout any market ncomes. Gross (market) ncome s obtaned as 7 Analyses of progressvty of taxes net of transfers s not new, but has prevously been made n a more general framework n whch all of government s redstrbutve expendtures are consdered (see Lambert, 2001, ch. 11). 8

10 (3) 1 Y Y t Y w p k where as: p t s the payroll tax rate and k Y s ncome from captal. Dsposable ncome s wrtten (4) D 1 t w Y w 1 t k Y k t c D (1 t w ) B Z where w t s the ncome tax rate appled to earned ncome, k t the tax rate on captal, c td VAT-payments, B taxable transfers, and Z transfers that are not taxed. From (4) we derve: 1 w w k k D t Y c B t Y Z 1 t (5) 1 ( ) 1 Usng (2), (3) and (5) yelds ETR as: (6) t 1 ( ) 1 w w k k 1 t Y B t Y Z 1 1 t Y 1 t Y e c w p k whereas ATR s obtaned as: (7) t Y t Y B t Y t D p w w w k k c w p k Y 1 t Y B t where dsposable ncome s defned by (5). All else equal, an ncrease n transfers mples a fall n ETR snce transfers ncrease dsposable ncomes. An ncrease n untaxed transfers ncreases ATR by ncreasng dsposable ncome and thereby VAT-payments. An ncrease n taxable transfers has n general ambguous effects on ATR as B appears both n the numerator and the denomnator of (7). 8 8 These exercses take the varous tax rates as fxed. We have t B 1 w p w k k / w c / w sgn sgn t t t t Y Y t Z Y 9

11 Lfetme equvalents to (6) and (7) are obtaned by takng the dscounted lfetme sum of ncomes and taxes before calculatng the tax rates (that s, the denomnator and nomnator n (6) and (7) are summed separately before takng the rato). 3.2 Measurng Tax Progressvty There are alternatve ways of characterzng a progressve tax system. A central noton s that progressvty reflects the relatonshp between the pre-tax ncome dstrbuton, the tax schedule and, fnally, the post-tax ncome dstrbuton outcome. In hs overvew of the tax progressvty lterature, Lambert (2001) states that a common dstncton s made between the redstrbutve capacty of a tax and the departure from proportonalty. The former aspect refers to the redstrbutve effect comng from the dfference between pre- and post-tax ncomes and s captured by the well-known Reynolds-Smolensky measure (Reynolds and Smolensky, 1977). The latter part relates to the dstrbuton of tax burdens and s typcally assocated wth Kakwan (1977). 9 Both of these measures are founded on Lorenz curve representatons and thus a sold lnkage to welfare theory. Both are also commonly used n emprcal assessments (Urban, 2009), but as Lambert (1985) notes the Kakwan measure s less useful when faced wth the negatve taxes arsng from ncorporatng transfers nto the tax scheme. For that reason, and bearng n mnd that our man nterest les n de facto redstrbuton of ncomes as a result of both taxes and transfers, we wll n ths analyss put man emphass on the Reynolds- Smolensky measure. Ths measure, L RS, s defned as the dfference between the pre-tax ncome Gn coeffcent G Y and the post-tax Gn coeffcent wth ncome earners ranked accordng to pre-tax ncomes, G Y-t : (8) L RS = G Y G Y-t. The Reynolds-Smolensky measure tells us how much, n percentage ponts, ncome nequalty s reduced. Note that t captures the redstrbutve capacty of taxes over the populaton as one entty. If one wshes to treat the redstrbutve outcomes separately across dfferent parts of the ncome dstrbuton, ths can be done by usng a undmensonal verson of the measure called A negatve mpact s most lkely except for ndvduals wth a very large fracton of ncome as ncome from captal. 9 As was orgnally shown by Kakwan (1984) and later extended by Aronson, Johnson and Lambert (1994) and Lambert and Urban (2008), the Reynolds-Smolensky redstrbute effect can be decomposed nto a vertcal and a rerankng effect. 10

12 relatve share adjustment (RSA). The RSA s defned as the after-tax ncome share for a partcular ncome group (or ndvdual) dvded by the pre-tax ncome share of the same group (see Baum, 1987; 1998). It s straghtforward to show that RSA s equvalent to (9) RSA 1 t 1 t where t s the average tax rate for group and t s the overall average tax rate. Here, the average tax rate can be defned ether as the effectve tax rate (ETR) or the average tax rate (ATR). A strctly proportonal tax system would thus mply RSA 1 for all. The rate at whch RSA declnes wth ncome provdes nformaton about the progressvty of the tax system. The undmensonal verson of ths measure focuses on pre- and post-tax ncome shares of two groups, vz. the top quntle (P80 100) and the lower four quntles (P0 80). Ths dstncton between the top and the rest of the ncome dstrbuton, based on mposng a dvdng lne at the 80th ncome percentle, follows prevous conventons used by the European Unon and many others to analyze ncome nequalty n terms of ncome quntles. In fact, n ther nvestgaton of how to defne the mddle class, Atknson and Brandoln (2011) land at the 80th ncome percentle beng a sensble boundary for separatng the top ncome earners from the rest of the dstrbuton. The rato between ncomes at the top and at the bottom the top quntle rato s computed before tax as well as after tax. The measure of progressvty (Q) captures the rato between pre- and post-tax ncome shares for the top and the bottom and bols down to comparng the tax rates met by the top and bottom groups,.e., (10) 1 t Q 1 t top b RSA RSA top b where subscrpts refer to top and bottom, respectvely. A strctly proportonal tax system would mply Q=1. Progressve tax systems generate measures below one whereas measures above one reflect a regressve tax system. 3.3 Data Our man data source s a longtudnal data base, LINDA, whch s a 3.35 percent random sample of the Swedsh populaton. LINDA s based on a combnaton of dfferent publc reg- 11

13 sters, such as ncome tax regsters and populaton censuses. 10 These data are avalable from 1968 and onwards and we focus on the perod In addton to rch nformaton on ncomes, the data nclude nformaton on tax payments and for most tax bases and deductons. 11 There s also nformaton on some human captal attrbutes and demographc characterstcs such as age, gender, educaton and martal status. We focus on ndvduals aged 20 and over. Usng ndvduals as the unt of analyss s delberate as the Swedsh tax system has been ndvdual-based snce Furthermore, our focus on lfetme ncomes and taxes complcates the use of household unts snce they tend to change over the lfe cycle. The sample s representatve for the total populaton each year except for the perod due to mssng nformaton on resdence (see below). The data consst of about 180,000 to 240,000 observatons per year. We dropped the three rchest ndvduals from the sample each year. 12 A small number of duplcate errors n the early perod ( ) were dropped. Our analyss provdes a lfetme approach on tax progressvty by focusng on ndvduals aged n We follow ths cohort for 42 years and rank the ndvduals on the bass of ther lfetme real ncomes,.e., ncomes over the perod Ths means that n 2009 the ndvduals are between 62 and 82 years old. 13 Lfetme tax payments are appled to lfetme ncomes so as to obtan lfetme average tax rates by lfetme ncome category. When calculatng lfetme taxes and ncomes, we dscount the fgures by a rate of three percent annually. 14 A few addtonal remarks concernng the analytcal approach are needed. Frst, taxes nfluence peoples behavor and ths s reflected n the dstrbuton of market ncomes as well as tax burdens. Unlke studes usng smulaton data, whch need to model explctly (or assume away) such responses, our use of actual ncome and tax outcomes mples that we for good and for bad ncorporate behavoral adjustments spurred by the tax system. Second, we as- 10 For a descrpton of LINDA, see Edn and Fredrksson (2000). 11 We use supplementary hstorcal tax rate nformaton for the calculaton of payroll taxes and VAT (see further below). 12 For the early 1980s, the ncluson of two extremely rch ndvduals has had dramatc mpact on the results when usng annual data. 13 We do not requre survval of all ndvduals n the panel up to 2009; a deceased person contrbutes zero to lfetme ncome. 14 The dscount rate of three percent s chosen so as to reflect the average real nterest rate durng the perod of analyss, and s also a standard rate n prevous studes for Sweden (see, e.g., Blomqust, 1981). 12

14 sume that tax ncdence s the same for annual and lfetme outcomes by consderng annual ncomes as beng snapshots of long-term outcomes. Thrd, the unt of analyss s the ndvdual. Although the households forms an economcally more relevant unt at the annual level, n a lfe cycle perspectve t s hard to thnk n terms of a household unt gven the large varaton of household structure over the lfe path. From a more practcal vewpont, most taxes n Sweden are leved on the ndvdual. 15 Fourth, captal ncome,.e., nterest earnngs, dvdends and captal gans, s ncluded n both annual and lfetme ncomes. Our man concept of ncome s pre-tax market ncome defned as the sum of earnngs from employment and self-employment and ncome from captal, ncludng captal gans. We do not nclude socal nsurance transfers and pensons n our defnton of earnngs (although employer-pad benefts, such as sck pay for short spells of absence, are ncluded n our defnton of earnngs). 16 However, earnngs from employment are defned as nclusve of an mputed payroll tax, reflectng the assumpton that the burden of payroll taxes leved on employers s ultmately borne by workers. Thus, transfers n the socal nsurance system are thus counted as market ncome when they are collected and not when they are pad out. Captal ncome s defned as the sum of nterest payments, ncome from owner-occuped housng, dvdends and captal gans. 17 Captal ncome s never negatve (thus, we do not subtract losses from total ncome). 18 Fnally, pre-tax market ncome also ncludes the mputed value of owner-occuped housng for those who pay real estate taxes (.e. those who own ther houses). We also elaborate wth two other ncome concepts. Pre-tax total ncome s pre-tax market ncome plus taxable socal transfers. Pre-tax total ncome reflects the conventonal tax base defnton, and s used when calculatng average, rather than effectve, tax rates (see above). Dsposable ncomes nclude all transfers (ncludng non-taxable transfers) net of VATpayments. Dsposable ncomes are reported n the ncome regsters only from 1978 onwards. 15 Snce 1971, all ncome taxes (and deductons) and most transfers are ndvdual. Wealth and consumpton taxes and a few transfers (chld allowances) are household-based. In the regster databases, however, these taxes and transfers are splt between the adults n a household. 16 The employer has to provde sck pay for the frst weeks of sckness absence. Such sck pay s regarded as wage payments and s subject to payroll taxaton. Longer perods of sckness absence are covered by sckness benefts provded by the natonal socal nsurance system. 17 Income from owner-occuped housng s mputed by tax authortes before 1991 ( schablonmässg nkomst av annan fastghet ) and by us thereafter, usng the real estate tax payments as bass. 18 Ths s manly for comparatve reasons, as captal ncomes before 1991 s never negatve whereas they can be so after 1991 due to a separate reportng of earnngs/gans and losses. 13

15 We therefore compute them for earler years back to 1968 by addng chld allowances usng statutory allowance levels. Personal ncome taxes for each tax base are observed as they appear on tax records between 1971 and For the perod , only the total sum of all personal tax payments s shown. In order to separate muncpal and state ncome tax payments for these three years, we use the muncpal statutory tax rates to mpute the muncpal ncome tax and then calculate the state ncome tax as a resdual from the total ncome tax payment. Place of resdence was not observed untl 1971, and the statutory muncpal tax rates for 1969 was used for 1970 due to lack of orgnal data. Thus, for the years , the sample s only representatve for those lvng n Sweden n Payroll taxes are leved on employers and mputed (see equaton 3) usng nformaton on ndvdual labor ncome and statutory tax rates n Söderberg (1996). The payroll tax rate s typcally the same for all ndvduals, but as already noted t has vared by earnngs before 1982 and t vares by age from 2006 and onwards. Before the tax reform n 1991, earnngs, captal ncome and the mputed value of owneroccuped housng were taxed at the same rate. A specal real estate tax was ntroduced n 1985 and the taxaton of labor ncome and captal ncome was separated n Net marketable wealth,.e., the sum of most real and fnancal assets less debts, was taxed at the household level untl ts abolshment n We thus observe the value of owner occuped housng up untl From 1991 and onwards, we mpute the value of owner-occuped housng for those who pay real estate tax. Value added taxes (VAT) are mputed usng nformaton on effectve VAT rates from Statstcs Sweden. Effectve VAT rates are lower than the standard rates because of tax dfferentatons and tax exemptons. To obtan effectve VAT rates, we use data on overall VAT revenues along wth data on prvate consumpton. For now, we treat the VAT rate as equal across ndvduals, thus gnorng the fact that dfferentated VAT rates may result n varatons n effectve VAT rates as consumpton patterns vary across ndvduals wth dfferent ncomes. 19 Corporate ncome s manly not taxed at the personal level n Sweden and s therefore excluded from the analyss. Small busnesses and some partnershps, however, are regstered at the ndvdual level and thus taxed along wth other ncomes. Inhertance and gft taxes exsted n Sweden up untl 2004, but were taxed separately from ncomes and were therefore never ncluded n the database by Statstcs Sweden. Parts of nhertances should, however, be reflected n the realzed captal gans (when hers sell ther nherted assets). 14

16 Moreover, we mpute ndvdual VAT payments by applyng the VAT rate to dsposable ncome, a procedure that underestmates progressvty snce t gnores that top ncome ndvduals typcally consume much less than ther dsposable ncomes. 4. The Evoluton of Taxaton n Sweden, Average Tax Rates We begn our analyss by provdng a seres of annual snapshots of the Swedsh tax progressvty. In Fgure 1 the evoluton of average tax rates (ATR) s shown across ncome categores. Several features stand out. There s marked ncrease n tax rates for all categores up to 1990, but the tax rates at the top ncrease faster than at the bottom. The early 1990s nvolves a sharp fall n progressvty, a development that s bound to be manly drven by the tax reform. Fnally, we note a regressve element n the tax system prevalng snce the early 1990s: average tax rates are often lower at the very top than at ncome levels just below the top. Ths feature reflects n part the fact that the ncome from captal a major ncome source at the top s relatvely low compared to earnngs taxes. 20 [Fgure 1 about here] 4.2 The Composton of Tax Bases What type of taxes contrbutes to progressvty? Fgure 2 shows how the composton of taxes (dsregardng transfers) has evolved over the tme. Throughout the 42-year perod of analyss, the ncome tax represents the most progressve element n the Swedsh tax system. The wealth tax and the real estate tax have but a trval mpact on the overall pcture, despte beng referred to as mportant redstrbutve elements n the Swedsh transfer system. The payroll tax and the muncpal ncome tax are proportonal except n the bottom (where socal nsurance becomes a much more common source of taxable ncome) and n the top (where captal ncome becomes relatvely mportant). 20 Another reason for the lower tax rates n the very top s the relatvely large deductons for nterest expenses among the ncome rch. Note also that we rank ndvduals based on the expresson n the denomnator n (6) and (7). Past work has subtracted captal ncome gans from total ncome before rankng (Pketty and Saez 2007), precsely because they wsh to avod to have transtory rch ndvduals n the top. To ease comparson wth lfetme measures of ncome, where such manpulatons are unwarranted, we nclude captal ncome gans n the annual rankng. 15

17 [Fgure 2 about here] 4.3 Internatonal Comparson of Average Tax Rates n 1970 and 2004 To get a more complete pcture of the Swedsh experence we also wsh to examne how Sweden s ncome and wealth tax progressvty compares to the same tax progressvty n other Western countres. The comparson s based on the recent study of the progressvty of the U.S. federal tax system by Pketty and Saez (2007), n whch the tax progressvty n France, Great Brtan and the U.S. n 1970 and around 2004 are compared. Fgure 3 dsplays an nterestng pattern. In 1970 (panel a) Sweden had the by far hghest average tax rate over the entre dstrbuton, wth excepton for the top 0.1 percentle. As we have shown above, ths was due to the hgh ncome and payroll taxes pad by low-ncome earners. The graphcal evdence suggests that progressvty (heurstcally defned as the curvature of the graph) s lowest n Sweden and n France. Swedsh top margnal tax rates ncreased rapdly after the 1971 tax reform up untl unprecedented levels n the latter half of that decade. [Fgure 3 about here] In 2004 (panel b), the pcture has changed dramatcally. Average tax rates are sgnfcantly lower across the board n all countres, ncludng Sweden. Moreover, the degree of progressvty has also decreased n all countres. The decreased s most pronounced n the Anglo- Saxon countres, but also qute notable n Sweden. Stll there are dfferences n the composton of taxes pad, e.g., as regards the relatvely mportance of payroll taxes n France and Sweden but not n the other countres. The man message s that ncome and wealth taxaton n Sweden seems to have converged wth that of other Western countres over the last decades. In the 1970s, Swedsh ctzens both pad markedly hgher taxes than elsewhere and experenced a hghly progressve tax schedule. One generaton later ths s no longer true, and Swedsh taxes are algned wth those n France, Great Brtan and the U.S. Although Sweden famously reformed ts tax system n 1991, the convergence n tax schedules s drven as much by reforms n the UK, US and France as n Sweden. 16

18 4.4 Effectve Tax Rates As ponted out n Secton 3, a substantal fracton of tax payments are returned to tax payers as transfers. Some of these transfers are ntertemporal redstrbutons, e.g., socal securty taxes returned as pensons later n lfe. Transfers can offset non-progressve elements n the tax system f they are targeted at the relatvely worse off. For nstance, earler evdence ndcates that the ncrease n the chld allowance n the early 1990s largely offset the regressve mpact of the tax reform (Björklund, Palme and Svensson, 1995). These consderatons call for a more comprehensve measure of tax progressvty, whch regards transfers as negatve taxes and correctly adjust the denomnator to nclude only market ncome. The effectve tax rate (ETR) s equal to one mnus the rato between dsposable ncome (ncludng transfers) and market ncome (not ncludng taxable transfers). In Fgure 4, the evoluton of effectve tax rates, ETR, are shown for selected percentles. The dsperson of ETR between the top and the mddle of the dstrbuton reaches a maxmum n the early 1980s. As expected, the use of ETR does not dffer from ATR measures (Fgure 1) at the top, where transfers consttute a neglgble part of total ncome. However, the more comprehensve ETR measure produces lower tax rates even among the upper mddle class. Notably, the effectve tax rates n the mddle and the bottom fell dramatcally n the early 1990s; a development trggered by a macroeconomc crss that brought about a rse n unemployment from 2 to 10 percent between 1990 and 1993 and a fall n the employment-to-populaton rate by more than 10 percentage ponts. These events reduced effectve tax rates va sharp falls n market ncomes n conjuncton wth substantal ncreases n transfers to the non-employed. [Fgure 4 about here] 5. Annual vs. Lfetme Tax Progressvty Fgure 5 encapsulates the central emprcal fndngs of our study. It shows the evoluton of annual tax progressvty over the whole cross-secton of Swedsh taxpayers and also the lfetme progressvty of the cohort of ncome earners lvng ther adult lves durng the entre study perod. The left panel s based on ETR progressvty, consderng both taxes and transfers, and the rght panel wth ATR progressvty consders tax payments only. We note that the level of progressvty dffers across the two panels, wth the redstrbutve effect beng 17

19 markedly larger n the ETR case. Ths reflects the role of transfers. 21 Lfetme taxes are conceptually more coherent n the ETR case snce the ATR case ncludes both payroll and penson ncomes n the ncome concept, whch generates a double count. Swedsh annual tax progressvty has followed an nverse-u shaped pattern over the past forty years. The 1970s saw a sharp ncrease, wth tax progressvty almost trplng. In the case of ETR progressvty, the reducton of the Gn coeffcent ncreased from beng percent n the begnnng of the 1970s to almost 30 percent around For ATR progressvty, the same ncrease was from 3 to about 10 percent. Thereafter the two concepts exhbt some dfferences although the overall curvature s roughly the same. When only consderng tax payments n the ATR-based measures (rght panel), progressvty fell n a stepwse fashon frst n the md-1980s and then, more drastcally, n the early 1990s. As noted n Secton 2, these epsodes when the poltcal system explctly amed at lowerng the hghest margnal tax rates. When nstead consderng the combned effect of taxes and transfers n the ETR progressvty (left panel), progressvty remans relatvely hgh throughout the 1980s, and declned only slowly from the md-1990s although not all the way back to the low level around Here we clearly observe the mutng effect of the transfer system, whch channels funds to the bottom of the dstrbuton and thereby ncrease effectve progressvty. The most strkng example of ths should be the spke n unemployment and unemployment nsurance of the early 1990s. [Fgure 5 about here] Lfetme tax progressvty s depcted as straght lnes n the two panels of Fgure 5. The most mmedate result comng out of comparng the annual and lfetme progressvty lnes s that lfetme progressvty appears to be markedly lower than annual progressvty. In fact, t s lower than annual progressvty n just about any sngle year of the analyss. 22 In Table 1 a 21 The levels of ETR progressvty are n lne wth a cross-country analyss of tax-and-transfer redstrbuton n other Western countres (Immervoll et al. 2006). 22 A potental concern wth ths comparson between annual and lfetme tax progressvty s that there are dfferent sample populatons n the two cases. The annual outcomes are based on the ncdence among the entre populaton whereas the lfetme estmatons derve from a specfc cohort aged n A reasonable queston to ask would therefore be whether the dfferent annual and lfetme outcomes reman when one studes the annual progressvty of the specfc cohort. Such a comparson holds the sample constant, but of course suffers from the problem that the annual seres wll now be constantly changng wth respect to the age profle. We have studed the annual and lfetme tax progressvty of both the full populaton and the cohort. The man message s that also for the cohort, lfetme progressvty s markedly lower than annual progressvty. There are sx out of 42 years, predomnantly durng the begnnng of the perod, where the cohort s annual levels are somewhat lower 18

20 subset of the estmates of Swedsh ncome nequalty and tax progressvty over the perod s presented. A second result comng out of both Fgure 5 and Table 1, almost equally strkng, s the low level 0.105, or about 10 percent Gn reducton of the lfetme ETR progressvty. To put ths number nto perspectve, t s half of the 20 percent redstrbutve effect that Björklund (1993) found for Sweden durng the perod Possbly ths reflects the effect of ncreasng the tme wndow for measurng people s ncomes. It s larger than the almost proportonal 4 6 percent redstrbutve effect found for the U.S. by Slemrod (1992). 23 But when comparng wth the lfetme ATR progressvty, a measure that s arguably more comparable to the Björklund and Slemrod estmates (that do not address the role of transfers), the lfetme redstrbutve effect s only 4.5 percent,.e., exactly the U.S. level found by Slemrod. [Table 1 about here] What explans the fndng that Swedsh tax progressvty s so much lower over the lfe cycle than at the annual level? We can thnk of at least two plausble channels. The frst concerns a dfference n ncome nequalty between the annual and lfetme horzons. Consder fgure 6, whch shows Gn coeffcents for market ncomes before and after effectve taxes over the annual and lfetme. Regardless of ncome concept, lfetme nequalty s lower than annual nequalty but the dfference s most pronounced for pre-tax ncomes. Ths s bound to reflect substantal ncome moblty so that low ncome n one year s partly offset by hgher ncome n another year, thus mplyng a more even ncome dstrbuton n the long run than n the short run. The lfe cycle contans perods durng whch people may be low ncome earners almost regardless of ther lfetme ncome status, e.g., when beng students or retrees. Ths result s largely n lne wth prevous studes comparng annual and longer-run ncome nequalty, such as Blomqust (1981) and Björklund (1993) for Sweden and Slemrod (1992) for the Unted States. [Fgure 6 about here] The second channel puts more emphass on the role of the tax system. Specfcally, we wsh than over ts entre lfe cycle. But that s the excepton, and when notng the stark ncreases n progressvty n later years, especally when usng effectve tax rates, the overall fndng remans ntact

21 to know whether annual tax burdens are the same as lfetme tax burdens across the ncome dstrbuton. A part of the answer s provded by Fgure 7, whch presents both annual and lfetme effectve tax rates of the top ncome quntle and bottom four ncome quntles. Its man message s that whle top taxes are roughly at the same level n annual and lfetme horzons, taxes pad by the low and mddle ncome earners dffer qute markedly dependng on tme horzon. In partcular, the lfetme tax rate for the bottom quntles s almost twce as hgh as the annual tax rate. Ths pattern suggests, just as for ncome nequalty, that the roots of the dfferental experence of annual and lfetme tax progressvty emanates from the volatle ncomes and resultng volatle tax burdens n the bottom of the ncome dstrbuton. [Fgure 7 about here] Fgure 7 strengthens the evdence that t s transtory poverty rather than transtory rchness that causes the dscrepancy between annual and lfetme estmates of tax progressvty. The nterpretaton s consstent wth the hstorcal labor market development n Sweden, along wth ts large emphass on socal securty. The huge ncrease n unemployment n the early 1990s mpled that large segments of the populaton earned zero market ncome. As the socal securty system kcked n, ndvduals remanng employed were effectvely taxed to smooth the dsposable ncomes at the bottom of the dstrbuton. In contrast, the lfetme tax burden at the top s remarkably close to annual tax burden at the same ncome segment. Ths pattern s consstent wth prevous research on the role of captal ncome gans n Swedsh ncome nequalty (Rone and Waldenström 2012); captal ncome gans seem to fall on the same persons from year-to-year. Further evdence of the mportance of the tax-and-transfer system for the fndng of a lower lfetme tax progressvty s presented n Fgure 8. It shows how market and dsposable ncome shares evolve across annual and lfetme horzons. In the left panel, market ncome shares over the lfetme are lower for all ncome fractles except for the bottom four decles where the opposte holds true. By contrast, the dsposable ncome shares are strkngly smlar for annual and lfetme outcomes across the entre dstrbuton. These patterns renforce the result that the socal nsurance element n the Swedsh tax system creates an annual dstrbuton of dsposable ncomes that are largely smlar to the lfetme dstrbuton of dsposable ncomes. 20

22 [Fgure 8 about here] As a fnal pece of evdence, we nqure the role of the dfferent tax bases for annual and lfetme progressvty. Table 2 presents average annual tax rates for each tax base usng both ETR and ATR concepts. 24 The dfferences between lfetme and annual outcomes are strkngly consstent across the two tax concepts. Frst, ncome taxes have relatvely smlar progressvty profles for annual and lfetme progressvty, a fndng that s largely n lne wth prevous results for the U.S. (Daves, St-Hlare and Whalley, 1984; Fullerton and Rogers, 1991). Although muncpal taxes devate somewhat n the bottom ncome fractles, annual and lfetme tax rates are almost dentcal for both state ncome tax rates (pad predomnantly by hgher ncome earners) and payroll tax rates. Captal (.e., wealth, property and captal ncome) taxes and consumpton taxes become more progressve or less regressve over the lfe cycle than at the annual level. In the case of captal taxes, ths fndng reflects the fact that among the transtory low-ncome earners there are people wth postve net wealth (e.g., old-age pensoners ownng ther house). The large annual regressvty of consumpton taxes s partcularly nterestng snce even at the annual level sales taxes can be argued to reflect lfetme ncome tax ncdence, due consumpton smoothng. Studes of lfetme vs. annual tax ncdence n the U.S., usng annual sales and excse taxes as measure of the former, have found lfetme taxes to be less progressve, or more regressve, than annual taxes (see, e.g., Poterba, 1989, Metcalf, 1994; Caspersen and Metcalf, 1994). Our fndngs of a qute regressve annual consumpton tax, caused prmarly by the low annual ncomes n the bottom of the dstrbuton, are thus not entrely n lne wth ther results. Movng to lfetme consumpton tax burden over lfetme ncomes,.e., an analyss more smlar ro that of Fullerton and Rogers (1991), however, we arrve at lfetme consumpton taxes ndeed beng more progressve, or less regressve, than annual taxes as was also found n the prevous lterature.. Fnally, payroll taxes and transfers are more progressve n the annual context over the lfe cycle, whch fts ncely wth the paper s man fndng. Payroll taxes are not pad by jobless people and snce few people are jobless ther entre lfe, ths wll prmarly affect annual pay- 24 Recall that the bottom four decles n the annual framework consst of retred, students and unemployed wth vrtually zero market ncomes, makng effectve tax and transfer rates very hgh. Snce ther transfer ncomes are taxable (as n equaton 1), they stll record a far amount of ncome taxes. Ths s why recorded payments of ncome tax need to be netted out aganst the transfers. 21