Appendices for Online Publication Do Higher Corporate Taxes Reduce Wages? Micro Evidence from Germany

Transcription

1 Appendces for Onlne Publcaton Do Hgher Corporate Taxes Reduce Wages? Mcro Evdence from Germany Clemens Fuest Andreas Pechl Sebastan Segloch September 22, 2017 Fuest: fo Insttute, Unversty of Munch and CESfo Pechl: fo Insttute, Unversty of Munch, CESfo, IZA and ZEW Segloch: Unversty of Mannhem, IZA, ZEW and CESfo

2 A The theory of corporate tax ncdence The theoretcal lterature has produced varous models of corporate tax ncdence. These models lead to dfferent predctons, dependng on the assumptons made about factor and output markets, wage-settng nsttutons, the structure of the tax system and behavoral reactons to tax changes. In the semnal paper by Harberger (1962), the economy s closed, labor markets are compettve and captal s n fxed supply. At least for plausble parameter values, the corporate tax burden s almost fully borne by captal. 1 subsequent lterature has emphaszed the mportance of nternatonal captal moblty n open economes (Damond and Mrrlees, 1971a,b; Bradford, 1978; Kotlkoff and Summers, 1987; Harberger, 1995). 2 In these models, the share of the (source-based) corporate tax burden bore by domestc mmoble factors ncreases as the sze of the economy relatve to the rest of the world decreases. In the case of a small open economy that faces a perfectly elastc supply of captal, the burden of the corporate tax s fully borne by factors other than captal. If profts of a frm are the result of locaton specfc rents, the tax wll partly fall on these rents. By contrast, f rents are frm specfc and frms are moble, the tax burden wll be fully shfted to owners of mmoble factors lke land or labor. 3 However, complete mmoblty of labor s a strong assumpton, n partcular when consderng corporate taxes at the sub-natonal level. of standard models s that labor markets are compettve. The Another restrctve assumpton Relatvely lttle attenton has been pad to the role of wage-settng nsttutons and labor market frctons n the context of corporate taxaton, two exceptons the studes by Felx and Hnes (2009) and Arulampalam, Devereux and Maffn (2012). In ths Appendx, we dscuss the mplcatons of varous wage-settng models for the mpact of corporate tax changes on wages. As wll be explaned further below, the model wll be vared slghtly to ncorporate dfferent assumptons about wage settng and two aspects of the tax system relatng to formula apportonment and ncome shftng. Consder an economy whch conssts of n jursdctons. There s a large number of frms n the economy. To ease notaton we normalze the number of frms per jursdcton to unty. Frms use the followng factors of producton: captal (K) and labor of two 1 Feldsten (1974) and Ballentne (1978) study tax ncdence n models wth endogenous savngs and fnd that part of the tax burden s shfted to labor. 2 Other mportant extensons of the canoncal Harberger model focus on the sectoral composton (Shoven, 1976), savngs behavor (Feldsten, 1974; Bradford, 1978) and the presence of uncertanty n the economy (Ratt and Shome, 1977). 3 From a global perspectve, a tax ncrease n one jursdcton reduces the ncome of mmoble factors n that jursdcton but ncreases the ncome of mmoble factors and reduces captal ncome n the rest of the world. In prncple, the burden of corporate taxes may also fall on supplers or on customers, provded nput and output prces are not pnned down by nternatonal markets. 1

3 skll levels. Labor of skll type k, k = h, l, s denoted by L k. 4 We wll consder dfferent producton technologes. In the base verson of the model we consder a concave producton functon F (K, L h, L l ), whch s assumed to exhbt declnng returns to scale. One nterpretaton s that there s an mplct fourth factor, whch may be nterpreted as a locaton-specfc rent. Whle captal and both types of labor are moble across muncpal borders, frms are mmoble, due to the locaton-specfc rent. The after-tax proft of frm located n jursdcton j, j = 1...n, s gven by P j = p F (K, L h, L l )(1 τ j ) k wk L k (1 τ j ) (1 ατ j )r K (1) where p s the output prce, r s the non-tax cost of captal and w k s the wage labor of skll type k. In some varants of the model, labor markets may not clear. In these cases, we assume that unemployed workers of skll type k receve unemployment benefts denoted by w k. The tax rate on corporate profts n jursdcton j s denoted by τ j. The varable α, wth 0 α 1, s a tax base parameter representng the share of the captal cost whch can be deducted from the tax base. Ths parameter s the same n all jursdctons. A cash-flow tax would mply α = 1, that s full deductblty of all costs. Most exstng corporate tax systems are more restrctve, however. Costs of debt fnancng are usually deductble whle costs of equty fnancng are not and loss offset s typcally restrcted. These propertes of the corporate tax base are mportant for theoretcal predctons about the ncdence of the tax, as wll be shown below. In the followng we drop the ndex j for frm varables to ease notaton. Total dfferentaton of the proft equaton and usng the standard frst order condtons for proft-maxmzaton yelds dp = dτ j T + dp F (K, L h, L l )(1 τ j ) k dwk L k (1 τ j ) dr (1 ατ j )K (2) where T = p F (K, L h, L l ) k wk L k αr K s the proft tax base. Equaton (2) shows that a tax ncrease may lead to lower profts for frm owners, hgher output prces charged to customers, a declne n wages receved by workers, lower ncome for captal owners or a combnaton of these effects. It s also possble that some of these groups lose whle others gan. The dstrbuton of the tax burden depends on how the model s closed, that s, 4 To keep the notaton smple we abstract from other nput factors lke land, energy or other ntermedate goods. Clearly, the prces of these goods could also be affected by corporate tax changes and the supplers mght bear part of the corporate tax burden. Corporate tax changes could also be captalzed n house prces. 2

4 on the assumed overall structure of the economy, n partcular the supply and demand elastctes n factor markets and the wage-settng nsttutons. In the followng, we dscuss the corporate tax ncdence on wages under dfferent assumptons about the labor market regme. As a benchmark, we start wth the case of compettve labor markets. We then turn to models wth wage barganng, far wage models, models where wages affect worker productvty and monopsonstc labor markets. In all of the followng cases, we assume that output markets are perfectly compettve and normalze the prce p of the output good to equal 1. A.1 Compettve labor markets Assume that nput markets are perfectly compettve, so that factor prces wll adjust to equate demand and supply. Factor demand functons are gven by the frm s frst order condtons and F (K, L h, L l ) K = (1 ατ ) (1 τ ) r (3) F (K, L h, L l ) L k = w k, k = h, l. (4) Equatons (3) and (4) mplctly defne the factor demand functons K D (w h, w l, R ) and L kd (w h, w l, R ) k = h, l where R = r 1 ατ 1 τ s the tax nclusve cost of captal. Whle the nterest rate r s assumed to be ndependent of captal demand n jursdcton j, wage rates are determned by equatng labor demand and labor supply. Labor supply s derved from worker utlty maxmzaton. Denote the utlty of a worker of skll type k by U k (C k, L k ). C k s the worker s consumpton and her budget constrant s gven by C k = w k (1 t)l k where t s the personal ncome tax rate. Standard utlty maxmzaton leads to a labor supply functon whch can be expressed as L ks (w k ), k = h, l. 5. Standard comparatve statc analyss of the labor market equlbrum condtons L kd (w h, w, l R ) = L ks (w k ), k = h, l yelds expressons for the mpact of a tax rate change on the skll-specfc wage. Consder 5 For notatonal convenence we express the labor supply functon as a functon of the wage rate before ncome taxaton 3

5 for example the effect on wages of skll type h: dw h dτ = LhD LhD ϕ ϕ L ld L ld ( ( ε lr ε hr ) ( ) 1 α 1 ε hh (1 α)(1 ατ ) w h ) ( 1 α (ε ll µ l ) 1 ) (1 α)(1 ατ ) w h (5) where ϕ s a postve parameter (the determnant of the matrx of coeffcents). Parameter ε st s the labor demand elastcty of skll group s wth respect to wage changes of skll type t and s defned as ε st = LsD of skll type k s gven by µ k = LkS (w h,wl,r ) w t w t L sd (w h,wl,r ) (w k ) w k w k L ks (w k ), s, t = h, l. The labor supply elastcty, k = h, l. Equaton (5) shows that, n general, the mpact of a tax change on the wage depends on demand and supply elastctes n the labor market. However, f the corporate tax s a cash-flow tax (α = 1), a change n the corporate tax rate wll be neutral for factor demand and, hence, wll leave wages unchanged. As a result, the corporate tax s a lump sum tax and the tax burden falls entrely on profts: P τ = [p F (K, L 1, L 2 ) 2 k=1 w k L k αrk ] < 0, wk τ = 0 k = 1, 2. Ths may be stated as Result 1: Compettve labor markets: The mpact of a tax change on wages depends on the demand and supply elastctes n the labor market. If all costs are perfectly deductble, the burden of the corporate ncome tax s fully borne by frm owners. Then a tax rate change does not affect the wage rate. Interestngly, the cash-flow tax result also carres over to varous (but not all) standard models of mperfect labor markets, as we wll show below. Most real world corporate tax systems devate from the polar case of a proft tax wth perfect cost deductblty, though. Accordngly, models of tax ncdence n the lterature typcally consder settngs where ether captal or labor costs are less than fully deductble. A.2 Wage barganng Varous labor market theores assume that wages are set va barganng between frms and ther employees. Barganng models mply that frm owners and employees share a surplus generated by the frm. If corporate taxes reduce ths surplus, t s straghtforward to expect that employees share part of the loss through lower wages. The magntude of these wage effects depends on the level where barganng takes place. We consder 4

6 ndvdual and collectve (frm and sector-level) barganng. A.2.1 Indvdual wage barganng Assume that the wage s set va barganng between the frm and the employee. most wdely used labor market model where ths happens s the job search model, where frms and ndvdual employees bargan over a matchng rent (see Rogerson, Shmer and Wrght, 2005, for a survey). Let the output a worker of type k n frm be gven by Q k (K k ). The addtonal proft the frm earns s P IB = Q k (K k )(1 τ ) w k (1 τ ) (1 ατ )rk k. The varable s the captal the frm nvests to equp the worker. The outcome of the barganng K k process s gven by The w k = arg max Ω w k where Ω = β k ln(w k w k ) + (1 β k ) ln P IB. The varable β k (0, 1) stands for the relatve barganng power of the employee. The frst order condtons of the barganng problem yeld w k = (1 β k )w k + β k Q k (K k )(1 τ ) (1 ατ )rk k (1 τ ). (6) The effect of a change n the corporate tax rate on the wage s w k τ = β k (1 α)rk k (1 τ ) 2 0. (7) A hgher corporate tax reduces the wage unless captal costs are fully deductble. Snce the employee s share of the surplus generated by the frm s ncreasng n the employee s barganng power, t s plausble that she also bears a larger loss f her barganng power s hgher. Ths may be stated as Result 2: Indvdual wage barganng: If captal costs are less than fully deductble, an ncrease n the local corporate tax rate reduces the wage. Ths wage change ncreases wth the barganng power of the employee. If the employee receves a large part of the surplus generated by the frm, she also bears a large 5

7 loss f the surplus declnes due to taxaton. A.2.2 Collectve barganng Assume that trade unon represent workers. We consder two cases: The frst case s frmlevel barganng, where frm-level unons bargan wth ndvdual frms. The second case s sector-level barganng, where sector-level unons bargan wth sector-level employer organzatons. Frm-level barganng. Denote the wage for a worker of skll type k employed by a frm located n jursdcton by w k = w k +s k, where s k s the wage premum generated by barganng at the frm-level. The barganng model we use for the frm-level s a standard effcent barganng model (McDonald and Solow, 1981), where unons and frms bargan over the wage premum, s k, and employment L k. Each skll type s represented by one trade unon and each frm negotates wth the two unons smultaneously (Barth and Zwemüller, 1995). The objectve functon of the trade unon representng the workers of skll type k n frm s gven by Z k = L k (w k w k ) = L k s k. In case of dsagreement, the rent of the unon Z k and the frm s proft P F B are equal to zero. After wages and employment levels are determned, frms set K to maxmze profts: where R denotes the cost of captal: F (K, L h, L l ) K = R (8) R = r (1 ατ ) (1 τ ). The outcome of the barganng process s gven by where s k, L k = arg max Ω k s k,lk Ω k = β k ln Z k + (1 β k ) ln P F B. 6

8 The varable β k (0, 1) stands for relatve barganng power of the skll type k unon n frm. The frst order condtons of the barganng problem yeld s k = (1 βj )βk Π F B (1 β kβj ) L k (1 τ ) k, j = h, l, k j (9) where Π F B = F (K, L h, L l )(1 τ ) k wk L k (1 τ ) (1 ατ )rk. For employment we fnd F (K, L h, L l ) L k = w k k = h, l. (10) The wage premum s k s equal to a share of the surplus per employee generated by the frm. The sze of ths share s ncreasng n the relatve barganng power of the skll group and decreasng n the barganng power of the other group of employees. Employment s set so that the margnal productvty of labor s equal to the skll-specfc reservaton wage. Dfferentatng (9) yelds where ds k L k + s k dτ dl k dτ = β 0 ((1 α)rk ) 0 (11) β 0 = (1 β j )βk (1 β k βj )(1 τ ) 2 > 0. The left-hand sde of (11) s equal to the change n the rent accrung to the workers of skll type k employed by frm. Ths rent unambguously declnes because of the tax change. Whether the wage rate declnes depends on how employment changes n response to the tax change. Equatons (8) and (10) mplctly defne the factor demand functons K (w k, w j, τ,..), L k (w k, w j, τ,..). Standard comparatve statc analyss shows that the mpact of a tax change on demand for labor of type k may be postve or negatve, dependng on whether the dfferent producton factors are complements or substtutes. The effect on wages s therefore also ambguous. Ths may be summarzed as: Result 3: Frm-level barganng: If captal costs are less than fully deductble, an 7

9 ncrease n the local corporate tax rate reduces the rent of each skll group. The effect on the wage rate s ambguous and depends on potental changes n employment. Ths result s smlar to that of ndvdual barganng. Hgher taxes reduce the rent that can be shared between the frm and ts employees. For gven levels of employment, wages unambguously declne n response to a tax ncrease. In the lterature, ths effect has been referred to as the drect effect of a corporate tax change on wages n frms where wages are set va collectve barganng (Arulampalam, Devereux and Maffn, 2012; Fuest, Pechl and Segloch, 2013). Takng nto account changes n employment may change the wage effect ( ndrect effect ). If the number of employees declnes n response to a tax ncrease, the rent generated by the company s shared among a smaller number of employees and the overall wage effect can be postve or negatve. Sector-level barganng. We now assume that barganng takes place at the sectorlevel. To ease notaton we normalze the number of sectors n the unonzed part of the labor market to unty. Ths mples that there are n frms n the sector. An employer organzaton bargans wth sector-level unons over the sector wde wage. We contnue to assume that each skll group s represented by ts own trade unon. The employer organzaton has the objectve of maxmzng aggregate profts of the frms n the sector. Followng the senorty model proposed by Oswald (1993), we assume that each unon wshes to maxmze the premum over the reservaton wage for the skll group t represents, whch s gven by v k = w k w k. For gven wages, frms set proft-maxmzng employment. The outcome of the sector-level barganng process s gven by where v k = arg max Ω Sk v k Ω Sk = γ k ln v k + (1 γ k ) ln m =1 P SB. The varable γ k (0, 1) stands for the relatve barganng power of the sector-level skll type k unon. Rearrangng the frst order condton of the barganng problem yelds v k = γ 0 n =1 Π SB n L k (1 τ ) =1 k, j = h, l, k j (12) 8

10 where γ 0 = (1 γj )γ k (1 γ j γ k ) > 0. The sector wde wage premum s equal to a share of the average surplus per worker generated by the frms n the sector. Employment and nvestment decsons are now gven by F (K, L h, L l ) L k = w k k = h, l (13) and F (K, L h, L l ) K = R. We now analyze the effect of a corporate tax change n jursdcton m, m (1,..., n), on v k. Total dfferentaton of equaton (12) yelds dv k = γ 0 where [ L k mdτ m n =1 ] n dl k (1 τ ) [ n Π SB + n =1 =1 L k (1 τ ) =1 n dπ F B L k (1 τ ) =1 ] 2 k, j = h, l, k j (14) n =1 dπ F B = [ F (K m, L h m, L l m) ] k wk L k αrk dτm ( ) + v h n dl h (1 τ ) + v l n dl l (1 τ ). =1 =1 In general, the mpact of a tax change on the wage s ambguous. The wage effect converges to zero f the frm n the jursdcton where the tax change occurs s small, relatve to the sector as a whole. The condtons for the wage effect to be neglgble dv k 0, whch mples dl k = 0 for all m, k = h, l follow from (14) and 9

11 are gven by [ L k m n L k (1 τ ) =1 ] Lk m τ m(1 τ m) 0, F (K m, L h m, L l m) k wk L k m αrk m n L k (1 τ ) =1 0. (15) The effect s thus neglgble f employment (ncludng the tax nduced change n employment) as well as the tax base n jursdcton m are small, relatve to the number of employees n the sector as a whole, weghted wth the tax factors (1 τ ). Ths may be summarzed as Result 4: Sector-level barganng: If captal costs are less than fully deductble, an ncrease n the tax rate may ncrease or decrease wages. The wage effect converges to zero f the actvty of the sector n the jursdcton where the tax change occurs s small, relatve to the rest of the sector. Result 4 suggests that local tax changes wll have a smaller or neglgble effect on wages f wage barganng takes place at the sector level, rather than the frm level, because the sector wll usually nclude many jursdctons. 6 A.3 Far wage models In far wage models (Akerlof, 1982) the wage s usually assumed to be a functon of ) wages of other employees of the same frm, ) an external reference wage 7 and ) profts of the frm (Amt and Davs, 2010; Egger and Kreckemeer, 2012). Consder a frm wth two types of workers. Assume that the far wage for type k workers employed by frm s gven by the functon w kf = f k (w k, w k, P ), where w k are unemployment benefts, w k are wages of the other skll group n the frm and profts P F W are gven by P F W = F (K, L h, L l )(1 τ ) k wk L k (1 τ ) (1 ατ )rk. 6 Some labor markets are characterzed by two ter barganng, where sector-level barganng sets a mnmum wage and wage premums on top of the mnmum wage are negotated at the frm-level (Boer, 2014). In such a settng, one would expect local tax changes to have a more sgnfcant mpact on local wages than n the case of pure sector-level wage barganng. 7 We assume that the reference wage, whch can be the average wage level pad n other frms, a statutory mnmum wage or a transfer to the unemployed, s gven. It may of course be the case that the reference wage s affected by local tax changes. Ths would not alter the result that hgher taxes lead to lower wages and vce versa. 10

12 We assume that the far wage functon has the followng standard propertes: f k w k f k w k, f k w k, f k P > 0, (16) f k P L k (1 τ ) > 0, (17) 1 f k w k f k w k > 0. (18) The far wage s ncreasng n unemployment benefts w k, n the wage of the other skll group employed by the frm and n the frm s profts. Equaton (17) mples that the far wage for skll group k ncreases f the wage of the other skll group k ncreases. Ths does not follow drectly from the frst dervatves, as an ncrease n the wage of the other skll group reduces profts. The effect on profts reduces the far wage. Equaton (18) mples that an ncrease n any of the reservaton wages rases the far wages of both groups. In equlbrum, the frm pays far wages to both types of employees and sets factor nputs to maxmze after-tax profts. Optmal factor nputs are gven by the standard margnal productvty condtons. Equlbrum wages are gven by w k = f k (w k, w k, P ) k = h, l. (19) Equaton (19) mplctly defnes the equlbrum wage rates w h and w l as functons of, among other thngs, the corporate tax rate τ. Standard comparatve statc analyss shows that the effect of a change n τ on wages s gven by w kf τ = T ξ [1 + f k P L k (1 τ ) + f k w k f ] k L k (1 τ ) < 0 P where T = F (K, L h, L l ) k wk L k αrk s the proft tax base and ξ = 1 f k + w k ( f k P (L k + f k w k f k w k L k ) + k f (L k P + f k w k ) L k ) (1 τ ) > 0. 11

13 Ths may be summarzed as Result 5: Far wage model: An ncrease n the local corporate tax rate reduces the wages of all skll groups. The ntuton behnd Result 5 s that f hgher corporate taxes reduce after-tax profts, farness consderatons would suggest that employees wll bear part of ths burden and vce versa. Ths effect s ndependent of whether or not wage and captal costs are fully deductble from the tax base. The neutralty property of cash-flow taxes does not hold here because wage farness s assumed to depend drectly on after-tax profts. A.4 Models where wages affect labor productvty Some labor market models emphasze that frms may want to rase wages because hgher wages lead to hgher labor productvty and, hence, hgher output. These models nclude effcency wage models, where hgher wages lead to more effort or lower worker fluctuaton, and models of drected job search, where hgher wages lead to better matches between workers and frms. 8 Followng Acemoglu and Shmer (1999), we assume that output s uncertan and depends on the qualty of frm worker matches. 9 There s only one type of labor. If a frm offers a hgher wage, more workers wll apply for the job and the chances of a good match ncrease, gven the wages offered by other frms. Wth probablty ρ (w, q) the addtonal output produced by fllng a vacancy n a frm located n jursdcton j equals Q (K ), wth probablty 1 ρ (w, q) t s equal to zero. The wages pad by other frms as well as other factors whch may be relevant for the lkelhood of success are summarzed by the vector q. The functon ρ (w, q) has the followng propertes: 10 ρ w > 0, 2 ρ w 2 < 0, 2 ρ = 0. (20) q 8 The key dfference to the far wage model dscussed n the precedng secton s that the latter emphaszes the drect lnk between the profts of a frm and the wage that s perceved to be far. No such drect lnk exsts here. However, far wage models may also be consdered as models where wages affect labor productvty because wages deemed as unfar would reduce worker effort or ncrease costly fluctuaton. 9 The results would be smlar n an effcency wage model followng Solow (1979) wth contnuous effort. In shrkng models wth dscrete effort (such as Shapro and Stgltz, 1984), we would not expect a drect effect on wages (for gven employment) but only an ndrect effect through changes n unemployment rates and hence the shrkng constrant. 10 The assumpton that all cross dervatves are equal to zero s made to smplfy the exposton, t s not necessary for the results. w k 12

14 Expected profts are now gven by P e = ρ (w, q)q (K )(1 τ j ) w (1 τ j ) (1 ατ j )rk. (21) The frst order condtons for the optmal wage and optmal nvestment are gven by and ρ w Q (K )(1 τ j ) (1 τ j ) = 0 (22) ρ (w, q)q (K )(1 τ j ) (1 ατ j )r = 0. (23) Equatons (22) and (23) mply that we can wrte the equlbrum wage rate as a functon w = w (τ, φ, α, r). Standard comparatve statc analyss leads to where w τ j = [ ] r ρ Q (1 τ j ) 2 w (K )(1 α) 0 (24) [ ] 2 = ρ (w, q)q (K ) 2 ρ w Q ρ 2 (K ) Q w (K ) > 0. Note that > 0 follows from the second order condtons for proft maxmzaton. A hgher corporate tax rate thus reduces the wage f there s lmted deductblty of captal costs. Ths may be summarzed as Result 6: Models where wages affect productvty: If captal costs are less than fully deductble, an ncrease n the local corporate tax rate reduces wages. The optmal wage trades off hgher output aganst the cost of hgher wages. The ncrease n output acheved through a wage ncrease s hgher, the hgher the captal stock of the frm. In the presence of mperfect deductblty of captal costs, nvestment declnes when the tax rate ncreases. Therefore the frm s margnal productvty gan from a wage ncrease falls. As a result, t s optmal for the frm to adjust ts wage polcy towards lower wages and a lower qualty of worker frm matches. 13

15 A.5 Monopsonstc labor market Consder a frm facng the labor supply functon by L s = L s (w), L s (w) > 0. Output s produced usng a standard, strctly concave producton technology F (K, L ) wth complementarty between labor and captal: 2 F (K,L s (w )) K L > 0. Profts are gven by P M (K, w ) = F (K, L s (w ))(1 τ j ) w L s (w )(1 τ j ) (1 ατ )rk The frst order condtons for proft maxmzaton are F (K, L s (w )) L L s (w )(1 τ j ) (L s (w )w + L s (w ))(1 τ j ) = 0 (25) F (K, L s (w )) K (1 τ j ) (1 ατ j )r = 0 (26) Equatons (25) and (26) mplctly defne the proft-maxmzng wage rate w and the captal stock set by the monopsonst, as functons of the tax corporate rate. Standard comparatve statc analyss leads to w τ j = 1 Γ where the second order condtons mply Ths mples Γ = 2 P M (K, w ) K 2 [ ] 2 F (K, L s (w )) L s (w )(1 α) < 0. K L [ 2 P M (K, w ) 2 P M w 2 (K, w ) K w ] 2 > 0. Result 7: Monopsonstc labor market: If captal costs are less than fully deductble, an ncrease n the local corporate tax rate reduces wages. A hgher corporate tax rate reduces nvestment so that the margnal productvty of labor falls. As a result, frms employ less labor. In a monopsonstc labor market ths mples a lower wage. 11 A.6 Extensons In ths subsecton, we consder two extensons of the model that are both related to partcular aspects of corporate taxaton. The frst extenson takes nto account that frms 11 Ths result holds n models of monopsonstc wage settng wth constant labor supply elastctes. If ths assumpton s relaxed, the result on tax shftng s theoretcally ambguous. 14

16 may operate n more than one jursdcton. Many countres use formula apportonment to allocate corporate profts to dfferent jursdctons for taxaton purposes. The second extenson s to allow for tax avodance through dfferent types of ncome shftng. A.6.1 Frms operatng n multple jursdctons wth formula apportonment Consder a company wth plants n two jursdctons, 1 and 2. As a frst step, we assume that there s just one type of labor. 12 Employment (captal) n jursdcton j s denoted by L j (K j ), j = 1, 2. The wage rate s the same n both plants. After-tax profts of the company are P F A = F (K 1, K 2, L 1, L 2 )(1 τ ) (1 τ )w[l 1 + L 2 ] (1 ατ )r[k 1 + K 2 ] Assume that the tax apportonment formula s based on payroll as the only apportonment factor. 13 Gven that there s a unform wage rate n the two plants, the proft tax rate s gven by τ = τ 1L 1 + τ 2 L 2 L 1 + L 2. (27) The effect of a tax rate change n one jursdcton on the frm s effectve proft tax rate τ, gven the level of employment, s τ τ j = where τ j s the tax rate of jursdcton j. L j L 1 + L 2, j = 1, 2 Assume that wages are set va collectve barganng whch takes place at the frmlevel, not at the plant-level, and that wages pad to workers of a gven skll group are the same n the two plants. The objectve functon of the skll type k unon s now gven by Z F A = (L 1 + L 2 )(w w) = (L 1 + L 2 )s F A. 12 The case for two skll types s dscussed below. 13 Ths s the case for the LBT n Germany. In the US, apportonment for state taxes s based on payroll, sales, and assets, see Suárez Serrato and Zdar (2016). 15

17 The outcome of the barganng process s gven by s F A, L 1, L 2 = arg max Ω F A s F A,L 1,L 2 where Ω F A = λ ln Z F A + (1 λ) ln P F A. The varable λ (0, 1) stands for the relatve barganng power of the unon. The frst order condton for the wage rate yelds where Π F A s F A = λ [(L 1 + L 2 )(1 τ )] Π F A = F (K 1, K 2, L 1, L 2 )(1 τ ) (1 τ )w[l 1 + L 2 ] (1 ατ )r[k 1 + K 2 ]. For gven levels of employment, the change n the wage premum caused by a change n the tax rate s gven by s F A τ j = λ (1 α)r(k 1 + K 2 )L j (L 1 + L 2 ) 2 (1 τ j ) 2 0. Ths mples: Result 8: Formula apportonment and frm-level barganng: In frms wth plants n many jursdctons and homogeneous labor, where corporate taxaton s based on formula apportonment, wages are set va collectve barganng at the frm-level, and captal costs are less than fully deductble, an ncrease n the corporate tax rate n one jursdcton decreases wages n the entre frm. If employment n the jursdcton that changes the tax rate s small, relatve to employment n the frm as a whole, the tax effect s also small. Consder next the case of two skll types, k = h, l. After-tax profts of the company are now ( P F A k ) = F (K 1, K 2, L h 1, L l 1, L h 2, L l 2)(1 τ ) j k wk L k j (1 τ ) (1 ατ )r[k 1 + K 2 ] 16

18 wth obvous notaton. The proft tax rate s gven by τ = j j k τ jw k L k j. k wk L k j For gven employment, the effect of a tax rate change n one jursdcton on the frm s effectve proft tax rate τ s τ τ j = j k wk L k j k wk L k j. s: The effect of a wage change for workers of skll type h on the effectve proft tax rate [ τ L h w = [τ h 1 τ 2 ] 1 L l 1 ] Lh 2 L l L 1L l 1 l 2 2 σ where σ = [ ] 1 + wh L h 1 + w l L l 2 1 [w h L h w h L h 2 + w l L l 2 + w l L l 2] 2 > 0. 2 Assume, for nstance, that muncpalty 1 has a hgher tax rate than muncpalty 2. The effect of an ncrease n the wage of the hgh sklled w h on the tax burden wll depend on whether ths ncreases the payroll share of the hgh tax muncpalty, or that of [ the low ] tax muncpalty. If the share of hgh sklled s hgher n jursdcton 1, so that L h 1 > 0, the tax rate τ wll ncrease, and vce versa. The effect of a wage change L l 1 Lh 2 L l 2 on the proft tax rate a frm effectvely pays s therefore generally ambguous. Once agan assumng frm-level collectve barganng and homogeneous wages for a skll group across plants, the objectve functon of the skll type k unon s now gven by Z F Ak = (L k 1 + L k 2)(w k 1 w k ) = (L k 1 + L k 2)s F Ak. The outcome of the barganng process s gven by s F Ak, L k 1, L k 2 = arg max Ω F Ak s k,l k 1,Lk 2 17

19 where Ω F Ak = λ k ln Z F Ak + (1 λ k ) ln P F At. As above, the varable λ k (0, 1) stands for relatve barganng power of the skll type k unon. The frst order condton for the wage rate yelds s F Ak = λ k P F At, k = h, l (28) (1 λ k ) [(L k 1 + L k 2)(1 τ ) Φ k w] where Φ k w = P F A τ τ w k. The key dfference between ths case and that wth homogeneous labor s that a wage change now affects the effectve tax rate. It thus nfluences the outcome of unon-frm barganng. For nstance, f a hgher wage ncreases the effectve tax rate, whch mples Φ wk < 0, the wage premum acheved by the unon wll be smaller, other thngs equal, and vce versa. Equaton (28) mplctly defnes the two frm-specfc wage premums emergng from the barganng process as functons of the type s F Ak = s F Ak (τ, τ j, T, L k, L k j...). Dfferentatng (28) shows that the change n the local corporate tax rate on wages s, n general, ambguous. A.6.2 Income shftng Income shftng to avod taxes may occur n dfferent forms. Multnatonal frms can use debt fnancng or transfer prcng to shft profts across natonal borders. Income shftng may also occur between dfferent tax bases wthn a country. For nstance, frm owners may shft ncome between the corporate and the personal ncome tax base by changng wages pad to famly members. We dscuss the two cases n turn. Internatonal ncome shftng. Assume that the frm s profts are gven by P S j = p F (K, L h, L l )(1 τ j ) k wk L k (1 τ j ) (1 α j τ j )r K + θ j S c(s ). (29) The varable S s ncome shfted from the proft tax base to the personal ncome tax base of the frm owners, whch may be postve or negatve, θ j s the tax beneft per unt of 18

20 ncome shfted and c(s ) s a convex shftng cost functon. 14 Proft maxmzaton factor nput decsons lead to the usual margnal productvty condtons, and optmal ncome shftng mples c (S ) = θ j so that the proft-maxmzng amount of shfted ncome S can be expressed as a functon of the tax beneft S = S (θ j ), wth S > 0. Consder frst the case of a multnatonal company whch s able to shft ncome abroad. If the frm can do so, for nstance, through a foregn subsdary chargng a fully deductble cost to the domestc parent company, the tax advantage from ncome shftng s gven by θ j = τ j τ f, where τ f s the foregn proft tax rate. Assume that wages n the multnatonal frm are determned by frm-level barganng. In ths case, the wage premum generated by unon frm barganng s gven by z k = (1 βj )βk (1 β k βj ) Π S L k (1 τ ) k, j = h, l, k j (30) where Π S = F (K, L h, L l )(1 τ ) k wk L k (1 τ ) (1 ατ )rk + (τ j τ f )S c(s ). Dfferentatng (30) yelds dz k L k + z k dτ dl k dτ = β S 0 [(1 α)rk (S (1 τ f ) c(s ))] 0 (31) where β S 0 = (1 β j )βk (1 β kβj )(1 τ > 0 k, j = h, l, k j. 2 ) The rght-hand sde of (31) s ncreasng n S (gven that S = S ), whch mples that the declne n the rent accrung to labor s smaller, the hgher the equlbrum level of ncome shftng. Ths yelds Result 9 Internatonal ncome shftng: If frms engage n nternatonal ncome shftng and wages are set by frm-level barganng, the declne n the rent accrung to 14 Here we assume that proft shftng s carred out by changng the wages of frm owners workng n the frm or famly members of the frm owner. Ths mples that s would be reported as wage ncome. Another way of shftng ncome s to provde captal n the form of debt, rather than equty. Many countres have ntroduced ant-tax-avodance legslaton, whch lmts ncome shftng. We therefore take nto account costs of ncome shftng. Ths can be nterpreted as the cost of hrng tax consultants or the cost of concealng ncome shftng. For notatonal smplcty we assume that shftng costs are not tax deductble. 19

21 labor caused by a hgher corporate tax decreases as the equlbrum level of ncome shftng ncreases. Natonal ncome shftng. We now consder the possblty of domestc ncome shftng between the proft tax base and wage ncome. In ths case the tax advantage from ncome shftng s gven by θ j = φ j τ j t p, where t p s the margnal tax rate on wage ncome of the relevant employee. Ths s relevant n settngs where the wages of some employees are effectvely proft dstrbutons, so that wage barganng plays no role for them. Assume that the wages pad n the absence of ncentves for ncome shftng, that s for equal taxes on profts and labor ncome, would be gven by the functon w ks (τ j,...). Then the dw ks dτ j observed change n the wages pad out by the frm would equal k L k + ds dτ j. Whle true wages are lkely to declne n response to hgher taxes, albet by less than they would n the absence of ncome-shftng possbltes, we now have the addtonal effect that the ncome-shftng effect ds dτ j > 0 ncreases reported wages. Thus f ncome shftng s mportant, we would expect observed wages to declne less, or even ncrease, n response to hgher corporate taxes. Ths may be summarzed as Result 10 Natonal ncome shftng: If frms shft ncome between the proft tax base and the labor ncome tax base, a hgher corporate tax rate wll lead to a smaller declne n reported wages than n the absence of ncome shftng. Wages may even ncrease. Unfortunately, we cannot test ths mechansm drectly wth our data because we do not know whether there are employees who are members of the owner famly. 20

22 B Insttutonal background B.1 German busness taxes In 2007, proft taxes accounted for about 6.2% of total tax revenue (ncludng socal securty) n Germany (OECD, 2015). 15 In terms of tax revenues, the LBT s the most mportant proft tax, accountng for about 60 70% of total proft tax revenues from corporate frms. Overall, the share of proft tax revenues from local taxes s relatvely hgh n Germany compared wth other countres. In the US, for nstance, state and local corporate taxes together account only for about 20% of total corporate taxes (NCSL, 2009). In addton, the LBT s the most mportant source of fnancng at the dsposal of muncpaltes, generatng roughly three quarters of muncpal tax revenue. As mentoned n Secton I, there are two other proft taxes n Germany, the corporate ncome tax (CIT), whch apples to corporatons, and the personal ncome tax (PIT), whch apples to non-corporate frms. We dscuss the most mportant features of these two taxes n turn. Corporate ncome tax. The rate of the natonwde corporate ncome tax, τ CIT, has undergone several changes n recent years. Untl 2000, a splt rate mputaton system exsted n Germany, where retaned profts were subject to a tax rate of 45% n 1998 and 40% n 1999 and Dstrbuted profts were taxed at a rate of 30% from 1998 to As of 2001, retaned and dstrbuted profts were taxed equally at 25% (26.5% n 2003). In 2008, τ CIT was lowered to 15%. In all years, a so-called soldarty surcharge (to fnance the costs of reunfcaton), sol, of 5.5% of the corporate tax rate was added. There are two steps to calculatng the total statutory tax rate for corporate frms. Frst, LBT and CT rates are added. Second, the deducton of the LBT payments from the tax base has to be taken nto account. The statutory tax rate for corporate frms, τ corp, from 1998 to 2007, s τ corp = τ CIT (1+sol)+t fed LBT θmun LBT. Snce 2008, the denomnator of 1+t fed LBT θmun LBT the equaton s equal to 1, as the LBT can no longer be deducted from the tax base. Personal ncome tax. Non-corporate frms (Personengesellschaften) are subject to the progressve personal ncome tax (on operatng profts assgned to the propretor). Non-corporate frms have an LBT allowance of 24,500 euros and a reduced t fed LBT for small 15 Ths s below the OECD average of about 10.6% (US: 10.8%, UK: 9.4%). Part of ths relatvely low share of proft taxes s due to the rather hgh share of socal nsurance contrbutons (SIC) n Germany. If SIC are excluded, the share n total taxes s about 11.5%. A hgh share of unncorporated frms n Germany s a second factor. These frms pay PIT, n addton to the LBT, and the OECD does not classfy PIT as proft taxes. 21

23 non-corporate frms pror to 2008: for every 12,000 euros exceedng the allowance of 24,500 euros, t fed LBT was rased by one percentage pont so that the full basc federal rate of 5.0% had to be pad only for taxable ncome exceedng 72,500 euros. The tax rate for a non-corporate frms τ non corp from 1998 to 2007, s τ non corp = τ P IT (1+sol)+t fed LBT θmun LBT. 1+t fed LBT 1.8 The denomnator of the equaton shows that a fxed share of the LBT labltes can be deducted from the personal ncome tax base. Ths share amounted to t fed LBT 1.8 Y from 2001 to 2007 and t fed LBT 3.8 Y from 2008 onwards. B.2 German labor market nsttutons Tradtonally, German labor unons have been very nfluental. 16 Collectve barganng agreements (CBAs) at the sector-level are the most mportant mechansm for wage determnaton. Nevertheless, there has been a sgnfcant declne n barganng coverage. In West (East) Germany, CBA coverage decreased from 76% (63%) n 1998 to 65% (51%) n The share of workers covered by sectoral agreements fell from 68% (52%) to 56% (38%) (Ellguth, Gerner and Stegmaer, 2012). 17 In addton to sector-level CBA, some frms have frm-level agreements, whle other frms are not covered by a CBA and rely on ndvdual contracts wth each employee. The average duraton of a CBA ncreased from 12 months n 1991 to 22 months n Usually, negotatons take place n the frst half of a year. Frms may pay wages above those negotated n CBAs. Except for a few ndustres, there was no legal mnmum wage n Germany durng our perod of analyss. However, the socal securty and welfare system provdes an mplct mnmum wage and CBAs ensure that wages are above that level. 16 See Dustmann et al. (2014) for an overvew and analyss of the development of German labor market nsttutons durng our perod of nvestgaton. 17 Coverage rates vary by ndustry: collectve barganng s slghtly above average n the manufacturng sector, whle the hghest coverage s n the publc sector and the lowest n ICT, agrculture and restaurant ndustres. Overall, unon coverage rates n Germany are lower than n other European countres except the UK and some Eastern European countres but hgher than n the US (Du Caju et al., 2008). 22

24 C Descrptve Statstcs Jursdctonal changes Analogously to Fgure 1, Fgure C.1 shows the cross-sectonal and tme varaton n LBT rates for the full sample of muncpaltes, ncludng muncpaltes that underwent a jursdctonal change. The rght panel clearly shows that the number of tax changes for these merged muncpaltes s relatvely hgh. However, the varaton n tax rates s artfcal and related to the way we mpute tax rates. As descrbed n Secton I.B, the wage data contans geographcal nformaton for the jursdctonal boundares as of December 31, In order to match the tax data, we have to brng t to the same boundares. Ths generates artfcal varaton n tax rates, as we need to calculate populaton weghted average tax rates for those merged jursdctons. Consequently, we fnd a large number of (small) tax changes for East German muncpaltes. Table C.3 shows that on average 12.4% of the muncpaltes change ther tax rate per year. Among the merged muncpaltes, however, the share s 33% (wth a much smaller average change). Gven ths measurement error n tax rate changes, we focus on non-merged muncpaltes n our baselne analyss (and check whether results for merged and non-merged muncpaltes dffer). Due to ths restrcton, we are left wth about 10,000 muncpaltes and 18,000 tax changes for dentfcaton (nstead of 11,441 muncpaltes wth about 27,000 partly artfcal tax changes). 23

25 Fgure C.1: Cross-sectonal and tme varaton n local tax rates Source: Statstcal Offces of the Laender. Maps: GeoBass-DE / BKG Notes: Ths fgure shows the cross-sectonal and tme varaton n muncpal scalng factors of the German LBT. Ths fgure ncludes both non-merged and merged muncpaltes. The left graph depcts the cross-sectonal varaton n LBT rates (n %) nduced by dfferent scalng factors for 2003 (the md-year of our sample). The rght graph ndcates the number of scalng factor changes per muncpalty between 1993 and Jursdctonal boundares are as of December 31,

26 Table C.1: Muncpal scalng factors, mean mn p5 p50 p95 max Average Source: Statstcal Offces of the Laender. Notes: The table provdes descrptve statstcs on the muncpal scalng factors for all non-merged muncpaltes (N=10,001) n Germany over tme. 25

27 Table C.2: Muncpal scalng factors changes per muncpalty, any ncrease large ncreases changes muncpaltes n % muncpaltes n % all muncpaltes (N=11,441) all non-merged muncpaltes (N=10,001) non-merged muncpaltes n LIAB (N=3,522) all merged muncpaltes (N=1,440) Source: Statstcal Offces of the Laender. Notes: The table summarzes the number of tax ncreases and large tax ncreases for all, non-merged and merged muncpaltes from 1993 to Large ncreases are defned as the top 25% of the tax ncrease dstrbuton, that s an ncrease of the busness tax rate of 1.1 percentage ponts or more. 26

28 Table C.3: Tme varaton n muncpal scalng factors, muncp. wth a(n) change... ncrease... decrease share mean change share mean ncrease share mean decrease all muncp non-merged muncp. (all) non-merged muncp. (LIAB) merged muncp by year (all non-merged muncpaltes) Source: Statstcal Offces of the Laender. Notes: The top part of the table summarzes the frequency, sgns, and szes of muncpal scalng factor changes for all muncpaltes (N=11,441), non-merged muncpaltes (N=10,001), and merged muncpaltes (N=1,440). The bottom part of the table shows the frequency, sgn and sze of muncpal scalng factor changes for non-merged muncpaltes over tme. Table C.4: Percentles of the share of non-wage-censored workers across frms p1 p5 p10 p25 p50 p75 p90 p95 p99 obs manuf , 137 servce , 490 total , 627 Source: LIAB. Notes: Ths table shows the dstrbuton of the share of non-wage-censored workers across frms n dfferent sectors. Workers are defned as wage-censored f they earned more than the socal securty contrbutons earnngs celngs at least once n the sample. In ths table, manufacturng ncludes constructon and servces nclude trade. 27

29 Table C.5: Descrptve statstcs, plant sample, non-merged muncpaltes, mean p50 sd Wage 2,733 2, Local scalng factor LBT rate (n %) Muncpal spendng (n mllons) 2, ,155 Muncpal populaton 436,255 49, ,957 Dstrct unemployment rate Dstrct GDP (n mllons) 18,977 6,758 28,198 Share: West German muncpaltes Number of employees ,136 Share: Lable plants Share: Sector level barganng Share: Frm level barganng Share: No collectve barganng Share: Manufacturng Share: Constructon Share: Trade Share: Servces Share: Publc/Utltes Share: Hgh proftablty Share: Medum proftablty Share: Low proftablty Share: Sngle plant frms Share: German owner Source: LIAB and Statstcal Offces of the Laender. Notes: Total number of plantyear observatons: 69,249. Number of plants: 21,253. All monetary varables n 2008 euros. 28

30 Table C.6: Descrptve statstcs, worker sample, non-merged muncpaltes, mean p50 sd Wage 3,491 3,363 1,092 Local scalng factor LBT rate (n %) Muncpal spendng (n mllons) 2, ,667 Muncpal populaton 470, , ,344 Dstrct unemployment rate Dstrct GDP (n mllons) 22,233 9,211 28,541 Share: West German muncpaltes Number of employees 5,802 1,138 10,345 Share: Lable frms Age Share: Male Share: Hgh-sklled Share: Medum sklled Share: Blue collar Share: Never censored ndvduals Source: LIAB and Statstcal Offces of the Laender. Notes: Number of person-year observatons: 12,673,576. Number of ndvduals: 4,091,932. All monetary varables n 2008 euros. 29

31 D Addtonal Results 2 Log real wages (relatve to pre-reform perod t=-1) no restrcton 0 drops n sample (base) 0 drops, 1 hke n wndow 0 drops, 1 hke n sample Fgure D.1: Event study graphs: wage effects by event wndow cut Source: LIAB and Statstcal Offces of the Laender. Notes: The graph plots event study estmates (ˆγ j, j [ 3, 4]) and correspondng 95% confdence bands of dfferent specfcatons of equaton (1). Dependent varable s the log medan frm wage (observed on 30 June for each year). Event dummes are equal to one for tax ncreases greater than or equal to the 75th percentle of the tax ncrease dstrbuton. The tax change occurred for the treatment group on 1 January n event year t = 0, as ndcated by the vertcal red lne. All regresson models nclude muncpal, frm and state year fxed effects. The estmaton sample comprses all establshments lable to the LBT n non-merged muncpaltes. Dependng on the specfcaton, we addtonally restrct the sample to muncpaltes wthout a tax decrease durng the observaton perod, not more than one ncrease n the event wndow, and/or only one tax ncrease n the whole observaton perod (see legend). Standard errors are clustered at the muncpal level. Estmates are reported n Table D

32 4 Log real wages (relatve to pre-reform perod t=-1) All frms Lable Non Lable Fgure D.2: Event study graphs: wage effects by frm lablty Source: LIAB and Statstcal Offces of the Laender. Notes: The graph plots event study estmates (ˆγ j, j [ 3, 4]) and correspondng 95% confdence bands of dfferent specfcatons of equaton (1). Dependent varable s the log medan frm wage (observed on 30 June for each year). Event dummes are equal to one for tax ncreases greater than or equal to the 75th percentle of the tax ncrease dstrbuton. The tax change occurred for the treatment group on 1 January n event year t = 0, as ndcated by the vertcal red lne. All regresson models nclude muncpal, frm and state year fxed effects. The estmaton sample comprses all establshments n non-merged muncpaltes that dd not experenced a tax decrease durng the observaton perod. Dependng on the specfcaton, we addtonally restrct the sample to frms that are lable to or exempt from the LBT (see legend). Standard errors are clustered at the muncpal level. Estmates are reported n Table D