State Taxes and Spatial Misallocation

Transcription

1 State Taxes ad Spatial Misallocatio Pablo D. Fajgelbaum UCLA& NBER Eduardo Morales Priceto & NBER Jua Carlos Suárez Serrato Duke & NBER Owe Zidar Chicago Booth & NBER February 2016 Abstract We study state taxes as a potetial source of spatial misallocatio i the Uited States. We build a spatial geeral-equilibrium framework that icorporates saliet features of the U.S. state tax system, ad the we use chages i state tax rates betwee 1980 ad 2010 to estimate parameters that determie how worker ad firm locatio respod to chages i state taxes. Our results strogly suggest spatial misallocatio from state taxes. Elimiatig spatial dispersio i taxes accoutig for 4% of GDP while keepig the distributio of govermet spedig costat would icrease worker welfare by 0.2%. The potetial losses from greater tax dispersio ca be large. Movig from o tax dispersio to twice as much dispersio as what is curretly observed would reduce worker welfare by 0.6%. We thak Costas Arkolakis, Lorezo Caliedo, ad Dave Doaldso for their discussios of the paper. We also thak David Atki, Araud Costiot, Klaus Desmet, Cecile Gaubert, David Lagakos, Erico Moretti, Pat Klie, Adrés Rodríguez-Clare, Esteba Rossi-Hasberg, Aleh Tsyviski, ad Jo Vogel for helpful commets. Pawel Charasz, Matt Pahas, Prab Upadrashta, Victor Ye, ad Joh Wieselthier provided excellet research assistace. Fajgelbaum gratefully ackowledges support from the UCLA Zima Ceter. Suárez Serrato gratefully ackowledges support from the Kauffma Foudatio. Zidar gratefully ackowledges support from the Kathry ad Grat Swick Faculty Research Fud ad Booth School of Busiess at the Uiversity of Chicago.

2 1 Itroductio Regioal fiscal autoomy varies cosiderably across coutries. I some coutries, such as Frace, Japa, ad the Uited Kigdom, regioal govermets do ot set tax policy. I other coutries, such as Australia, Caada, Germay, Italy, Spai, the Uited States, or Switzerlad, these govermets have varyig degrees of autoomy to itroduce or abolish taxes, set tax rates, ad grat tax breaks. As a result, tax rates applied to both workers ad firms vary across regios i these coutries. A fudametal cocer is that these regioal differeces i taxes have aggregate cosequeces. The stadard reasoig from recet research studyig dispersio i distortios across firms, as i Restuccia ad Rogerso 2008) ad Hsieh ad Kleow 2009), or across cities, as i Desmet ad Rossi-Hasberg 2013) suggests that regioal differeces i taxes may have egative aggregate effects through distortios i the spatial allocatio of resources. To the best of our kowledge, there is so far o quatitative evidece o the geeral-equilibrium trade offs betwee cetralized ad decetralized fiscal structures. 1 I this paper, we provide this evidece by quatifyig the aggregate effects of dispersio i tax rates across U.S. states. As we discuss i more detail i Sectio 3, the U.S. is a typical example of a coutry with a decetralized fiscal structure, both i terms of the share of total tax reveue collected by regioal govermets ad the degree of spatial dispersio i tax rates. 2 We develop a spatial geeral-equilibrium framework that icorporates saliet features of the U.S. state tax code, ad the we use the more tha 350 chages i state tax rates betwee 1980 ad 2010 to estimate key model parameters that determie how workers ad firms reallocate i respose to chages i state taxes. Usig the estimated model, we compute the aggregate effects of replacig the curret state tax distributio with couterfactual distributios with varyig levels of disparity i state tax rates. Our mai couterfactual evaluates the potetial welfare ad GDP gais from replacig the curret distributio of taxes across U.S. states with a couterfactual harmoised fiscal regime. I this harmoised regime, all states set the same tax rates but cotiue to provide the same amout of public goods as i the curret sceario thaks to a system of iter-state trasfers. Our results therefore isolate the allocative impact of tax dispersio without divig ito broader cosideratios of how resposibility over govermet spedig should be allocated across differet levels of govermet. 3 From a theoretical perspective, the questio of how taxes impact the allocatio of factors give a distributio of govermet spedig across regios is distict from the questio of which 1 I his essay o Fiscal Competitio or Harmoizatio? Some Reflectios, Oates 2001) writes: there is a huge literature o this topic, ad it is overwhelmigly theoretical i character. 2 Accordig to data for year 2011 from the OECD Fiscal Decetralizatio Database, the share of total tax reveue collected by U.S. states 20.9%) is very similar to that collected by regios i Germay 21.3%), Spai 23.1%), or Switzerlad 24.2%); oly sigificatly lower tha that i Caada 39.7%), ad sigificatly larger tha what is observed i Australia 15.3%) ad Italy 11.7%). 3 These broader cosideratios have bee discussed theoretically i a vast literature o fiscal federalism. See Gordo 1983) ad Oates 1999) for a review. As this literature has poited out, addig to our aalysis a empirical evaluatio of the welfare impact of chages i the distributio public spedig across states would crucially require data o variables that are either observed or easily iferred give the iformatio provided i stadard datasets e.g., heterogeeity across each level of govermet i both the iformatio they have about idividuals prefereces for public services ad their efficiecy i providig public goods). 1

3 level of govermet should be ivolved i the provisio of public goods. 4 This distictio is ot just a theoretical costruct; istead, it reflects the empirical reality that, i every coutry with subcetral govermets, there exist multiple mechaisms to trasfer tax reveues across govermet etities, makig it relevat to quatify the allocative impact from regios tax policy separately from their spedig policy. 5 The foudatio of our model icludes key features from caoical eviromets used i the fiscal competitio literature. 6 As i these models, we cosider a eviromet cosistig of may states, several factors of productio which may be fixed lad ad structures) or mobile workers ad firms), ad state govermets that use tax reveue ad trasfers from the federal govermet to supply public services which may be valued by workers or used as a itermediate i productio. However, providig a quatitative assessmet of the welfare effects of tax dispersio requires a framework that ca match saliet features of the spatial distributio of ecoomic activity i the U.S. ad that ca accout for how the resource allocatio ad public spedig respod to chages i the fiscal structure. Therefore, we build o this foudatio i several ways. First, we iclude each of the mai sources of tax reveue of U.S. state govermets: sales, persoal icome, ad corporate icome taxes apportioed through both firm sales ad factor usage. Secod, our model flexibly accouts for the fact that states may be heterogeeous i terms of productivity, ameities valued by workers, edowmet of fixed factors, factor itesities i productio, ad trade frictios with other regios, the modelig of which follows the stadard approach from quatitative trade models such as Eato ad Kortum 2002) ad Aderso ad Va Wicoop 2003). These igrediets eable the model to ratioalize the observed distributios of workers, wages, icome, ad trade flows across states as a equilibrium outcome for ay give distributio of tax rates. Third, we depart from perfect competitio, assumig istead that firms are moopolistically competitive as i Dixit ad Stiglitz 1977) ad Krugma 1980). 7 Fially, workers ad firms respectively draw idiosycratic prefereces ad productivities across states. This assumptio, which is commo i the local labor markets literature Moretti, 2011), gives the model eough flexibility to match the observed resposes of workers ad firms to chages i taxes. Workers i the model decide where to locate based o each state s taxes, wage, cost of livig, ad ameity level; while firms decide where to locate, how much to produce, ad where to sell based o each state s taxes, productivity, factor prices, ad market potetial a measure of other states market sizes discouted by trade frictios). Despite the complexity of this ecoomic geography model exteded with a real-world US state tax code, its features ca be mapped to a stadard quatitative models of trade ad ecoomic geography. Specifically, keepig costat the provisio 4 E.g., Wildasi 1980) refers to the coditios leadig to a efficiet outcome give ay arbitrary distributio of public spedig as locatioal efficiecy. 5 Accordig to data the for year 2011 from the OECD Fiscal Decetralizatio Database, the share of the total sub-cetral govermet expediture that is due to trasfers from the cetral govermet is: U.S., 18.9%; Germay, 14.0%; Spai, 55.6%; Switzerlad, 25.1%; Caada, 18.8%. 6 For a recet summary of these models see Kee ad Korad 2013). We discuss the coectio with this literature i more detail i the literature review sectio below. 7 Uder perfect competitio, the model would lack operatioal otios of firms or equilibrium profits. These are eeded to quatify the effects of U.S. corporate taxes ad to make cotact with observed data o firm mobility i respose to these taxes. 2

4 of public goods, higher icome or sales taxes i oe state are equivalet to a lower ameity level i that state, resultig i lower labor supply, lower pre-tax wages, ad, firm etry. Higher corporate taxes are equivalet to lower productivity, while a chage i the sales apportiomet of corporate taxes is show to be equivalet to a particular set of chages i trade costs. Additioally, publicgoods provisio depeds o state taxes ad tax bases, ad o trasfers from the federal govermet, through the budget costrait of each state govermet. As a result, firm ad worker decisios deped o taxes both i partial equilibrium give relative prices ad state spedig ad i geeral equilibrium through the impact of taxes o prices ad public-service provisio. 8 We use a simpler versio of the model to characterize how the result of our couterfactual may deped o parameters. We demostrate that our model does ot impose a aswer to whether the tax-harmoisatio couterfactual leads to positive or egative effects o either worker welfare or output per worker. Qualitatively, these resposes deped o the correlatio betwee tax dispersio ad the fudametals of each state e.g., productivity or ameities), ad o parameters that gover the degree of labor mobility, cogestio through the use of fixed factors i productio, or prefereces for public goods. For example, a reductio i tax dispersio is more likely to icrease worker welfare for sufficietly low labor mobility; ituitively, workers pick the best amog may locatio optios, implyig that higher variace i the appeal of available optios ca be beeficial whe workers are ot especially attached to specific locatios. Therefore, whether a harmoized tax scheme raises or lowers welfare ad real output give a arbitrary distributio of govermet spedig is a empirical questio that depeds o the model s parameters. We disciplie these parameters by matchig features from the cross-sectioal U.S. distributio of ecoomic activity as well as worker ad firm resposes to actual tax chages. Four structural parameters, i particular, are key for the results: the elasticities of worker ad firm mobility with respect to after-tax real wages ad profits, respectively, ad the weights of public services i worker prefereces ad firm productivity. To estimate these parameters, we use equilibrium relatioships from our model ad a logitudial dataset o the distributio of workers, establishmets, tax rates, ad govermet reveue across states from 1980 to Our model geerates a worker-locatio equatio that predicts each state s employmet share as a fuctio of after-tax real wages ad state govermet spedig, ad a firm-locatio equatio that predicts each state s share of establishmets as a fuctio of after-tax market potetial, factor prices, ad state govermet spedig. Ituitively, higher partial elasticities of employmet ad firm shares with respect to govermet spedig i the data correspod to higher weights of public services i worker prefereces ad firm productivity i our model. Our estimatio approach uses taxes i other states to istrumet for each state s factor prices ad govermet spedig. We estimate a partial elasticity of state employmet with respect to after-tax real wages slightly above 1, ad with respect to govermet spedig of about 0.2. Compared to these estimates, firms are relatively more resposive to taxes ad market size tha to 8 Eve though our key couterfactuals keep govermet spedig costat, we eed to iclude govermet spedig i the model i order to implemet these couterfactuals, i.e., to fid the levels of tax rates ad trasfers that guaratee that each state ca maitai the observed level of public spedig i the couterfactual sceario without tax dispersio. 3

5 govermet spedig. 9 We calibrate the remaiig parameters productio techologies ad state fudametals) such that the model exactly reproduces, as a equilibrium outcome, the distributio of factor shares, wages, employmet, ad trade flows across states i 2007, the most recet year i which all the data we eed are available. As a over-idetifyig check, we compare the model s predictios for variables that are ot targeted by the parametrizatio with the data. We fid that the distributios across states of GDP ad tax reveue share i GDP implied by the estimated model are highly correlated with those observed i the data. Our results strogly suggest spatial misallocatio from state taxes. Keepig govermet spedig costat through a system of cross-state trasfers, elimiatig tax dispersio would icrease worker welfare by 0.2% relative to a 4% share of state taxes i GDP). Accoutig for the progressivity of state icome taxes yields cosiderably larger worker welfare gais from tax harmoisatio, which icrease to 0.4%. The gais i terms of aggregate real icome ad cosumptio are approximately half as large. We also fid that the potetial losses from greater tax dispersio ca be large: movig from a sceario with o tax dispersio to oe with twice as much dispersio as what is curretly observed while also keepig govermet spedig costat through cross-state trasfers) would reduce worker welfare by about 0.6%. Without cross-state trasfers, a reveue-costat elimiatio of the curret tax dispersio leads to a 0.8% gai, ad movig from o tax dispersio to twice as much dispersio as what is curretly would lead to a 1.7% loss. We establish the robustess of these fidigs i several ways. The model predictios for the impact of a tax harmoisatio that keeps govermet spedig costat are robust to differet approaches to estimate the parameters that gover the mobility of firms ad workers ad the weight of govermet services i prefereces ad productivity. For istace, the predictio that this chage i taxes will icrease welfare by 0.2% holds both i the extreme case i which we assig zero weight to public services i prefereces ad productivity, ad i the case i which we assume that the observed govermet size i each state reflects workers prefereces for public services. We also show that our results are robust to usig differet methods for measurig state tax rates. For example, the coclusio that there are positive allocative gais from harmoisig taxes across U.S. states is robust to usig progressive tax rates as well as adjustig our measures of state corporate tax rates for state subsidies ad for the differet fiscal treatmet of C-corporatios ad S-corporatios. The rest of the paper is structured as follows. Sectio 2 relates our work to the existig literature. Sectio 3 describes the features of the U.S. state tax system that motivate our aalysis. Sectio 4 develops the model ad describes its geeral-equilibrium implicatios. Sectio 5 studies theoretically how dispersio i taxes affects welfare ad aggregate productio i a simplified versio of our model. Sectio 6 presets the estimatio, ad Sectio 7 presets the couterfactuals. Sectio 8 cocludes. Detailed derivatios, additioal figures ad details o both estimatio ad data sources are show i a Olie Appedix. 9 These estimates are i the rage of existig work that has estimated similar elasticities; e.g., Boud ad Holzer 2000), Notowidigdo 2013), Suárez Serrato ad Wigeder 2014), Diamod 2015), Suárez Serrato ad Zidar 2015), ad Giroud ad Rauh 2015). See Sectio 6.3 ad Appedix D.4 for details. 4

6 2 Relatio to the Literature Misallocatio Our paper cotributes to the literature o the aggregate effects of misallocatio. A commo approach cosists of measurig distortios across firms as a implied wedge betwee a observed allocatio ad a model-implied udistorted allocatio, as i Hsieh ad Kleow 2009), ad the udertakig couterfactuals to ispect the aggregate effects of dispersio i these wedges. Recet papers have adopted a similar methodology to aalyze misallocatio across geographic uits, such as Desmet ad Rossi-Hasberg 2013) ad Bradt et al. 2013). 10 These wedges capture distortios that may be due to multiple sources. Rather tha iferrig distortios from wedges, we focus o the spatial misallocatio geerated by state taxes that we directly observe i the data. We use observed variatio i these taxes to estimate key model parameters. Trade ad Ecoomic Geography Our framework combies igrediets from quatitative ecoomic-geography models that itroduce labor mobility ito quatitative trade models, such as Alle ad Arkolakis 2014), Ramodo et al. 2015), Reddig 2015), Gaubert 2015), ad Mote et al. 2015). Our research questio the impact of state taxes o the U.S. ecoomy distiguishes our paper from this previous literature. This focus drives our modelig choices, estimatio approach, ad couterfactuals. Our model combies a umber of igrediets appearig i these ad i previous studies, 11 plus a few ew oes dictated by our questio. Importatly, we icorporate a real-world tax code ecompassig the mai taxes used by U.S. states ad by the federal govermet, ad a govermet sector that uses these taxes to fiace public services valued by workers ad firms ad trasfers to the states. 12 Relative to this literature, a cetral feature of our aalysis is the focus o performig couterfactuals with respect to policy variables that are directly observed U.S. state tax rates), ad the use of observed variatio i these same policies to idetify the key model parameters. 13 Fiscal Competitio ad Fiscal Federalism The literature o fiscal competitio i the traditio of Flatters et al. 1974), Wilso 1986), ad Zodrow ad Mieszkowski 1986), summarized amog others by Oates 1999), Wilso ad Wildasi 2004), ad Kee ad Korad 2013), typically cosiders static ad perfectly competitive ecoomies with two or more regios ad two factors of productio, oe immobile ad oe perfectly mobile, which may be used to produce a cosumptio good ad a o-traded public good. These basic igrediets are icluded i our model. As have 10 See also Behres et al. 2011) ad Hsieh ad Moretti 2015) for eviromets with spatial distortios across cities. A related literature o spatial misallocatio cosiders rural-urba icome gaps; e.g., Golli et al. 2013) ad Lagakos ad Waugh 2013) fid productivity gaps betwee agricultural ad o-agricultural sectors which are suggestive of misallocatio, ad Brya ad Morte 2015) study whether these icome gaps reflect spatial misallocatio. 11 Our model icludes a edogeous umber of moopolistically competitive firms i each locatio similarly to Krugma 1991) ad Helpma 1998), the use of differetiated products as itermediates as i Krugma ad Veables 1995), ad workers with idiosycratic prefereces for locatio as i Tabuchi ad Thisse 2002). 12 Aother ovel igrediet is imperfect firm mobility i the form of idiosycratic productivity draws across states. For a quatitative setup also featurig imperfect mobility of several factors of productio see Galle et al. 2015). 13 Bartelme 2015) estimates labor ad wage elasticities with respect to market potetial usig Bartik istrumets. I a iteratioal-trade cotext, Caliedo ad Parro 2014) estimate trade elasticities usig variatio i tariffs. 5

7 metioed, our model geeralizes this structure to a multi-regio settig i which the distributio of state characteristics ca be disciplied by usig data o the distributio of ecoomic activity. A cetral questio i this literature has bee whether jurisdictios settig tax policies accordig to the equilibrium of a o-cooperative game deliver a socially efficiet allocatio. It is well uderstood from this previous literature that the aswer to this questio depeds o the specific features of the eviromet i which it is cosidered, such as: which set of tax istrumets jurisdictios may cotrol e.g., taxes o mobile or fixed factors), how agets value govermet spedig e.g., whether there is cogestio i access to public services), what the objective fuctio of policymakers is e.g., ret-seekig, maximizig social welfare, or maximizig the welfare of oly fixed or mobile factors), or what iformatio each level of govermet has e.g., how much each govermet kows about local prefereces). 14 Our approach is to perform a couterfactual exercise i which all key ecoomic forces at play ca be disciplied by data. Cosequetly, we focus i our aalysis o a couterfactual that, coditioal o our estimates of the key model parameters, does ot require takig a stad o the objective fuctio or the iformatio sets of policy-markers, or o the process through which observed taxes are determied. Assumptios o how taxes are determied are oly relevat at the momet of estimatig the parameters, ad we discuss these idetificatio assumptios i detail i Sectio I a semial paper of the local public fiace literature, Tiebout 1956) illustrates how heterogeeous prefereces for govermet services ca play a cetral role i determiig the locatio of workers. Quatifyig these heterogeeous prefereces for a large set of worker types ad states is empirically challegig ad, therefore, our model assumes that all workers located i the same state have the same valuatio for public spedig but this valuatio may vary across states). However, we show that our couterfactual results are robust to several approaches to measurig the valuatio that workers have for govermet services. Specifically, sice our mai couterfactuals hold real govermet spedig fixed, worker locatio decisios i our couterfactuals deped oly o how each worker respods to chages i after-tax real wages ad ot o the chages i govermet services provided i each state). Factor Mobility i Respose to Tax Chages Empirically, the geeral-equilibrium effects implied by our aalysis deped o the elasticities of firm ad worker locatio with respect to taxes. Evidece o the effects of taxes o worker mobility icludes Bartik 1991) ad, more recetly, Moretti ad Wilso 2015). I terms of firm mobility, Holmes 1998) uses state borders to show that maufacturig activity respods to busiess coditios, ad a large literature studies the 14 E.g., Kee 1989) ad Lockwood 1997) establish coditios uder which some forms of commodity-tax harmoisatio may be superior to a observed tax distributio resultig from competitio across govermets uder alterative assumptios o whether govermets supply public goods. 15 A recet example of the literature o fiscal competitio is Ossa 2015). He uses a ecoomic-geography model with home-market effects to compute the Nash equilibrium of a game where states use lump-sum taxes to fiace firm subsidies. State subsidies are i fact ofte fiaced through exemptios from corporate taxes, ad their value amouts to about 10% of the state tax reveue that we cosider. Our results barely chage whe we adjust corporate state taxes by effective subsidy rates by state see Sectio 7.6.1). Ossa 2015) does ot icorporate state taxes or uses observed variatio i policy to estimate parameters, which are two of the key features of our aalysis. 6

8 impact of local policies o busiess locatio. 16 Suárez Serrato ad Zidar 2015) provide evidece o the impact of corporate taxes o worker ad firm mobility, ad Suárez Serrato ad Wigeder 2014) show that local ecoomic activity respods to public spedig. While these papers quatify the local effects of actual policy chages, our framework allows us to quatify how couterfactual policy chages i oe state or i may states simultaeously, such as a tax harmoisatio, impact geeral-equilibrium outcomes i every state idividually ad i the U.S. ecoomy as a whole Backgroud o the U.S. State Tax System Our bechmark aalysis focuses o three sources of tax reveue: persoal icome, corporate icome, ad sales taxes. The reveue raised by these taxes accouted, respectively, for 35%, 5%, ad 47% of total states tax reveue i 2012, ad collectively amouted to 4% of U.S. GDP. I this sectio, we first describe how we measure each tax rate. We the preset statistics that summarize the dispersio i tax rates across states. We coclude with evidece o the relatioship betwee state tax reveue ad govermet spedig. Appedix F details the sources of the data discussed i this sectio. 3.1 Mai State Taxes Persoal Icome Tax States tax the persoal icome of their residets. The base for the state persoal icome tax icludes both labor ad capital icome. I our bechmark aalysis, we use a flat state icome tax rate, ad we the explore how our couterfactual results chage if we accout for the progressivity of icome taxes at both the state ad federal levels. 18 We compute a icome tax rate for each state usig the average effective tax rate from NBER TAXSIM, which rus a fixed sample of tax returs through differet tax schedules every year ad accouts for most features of the tax code see Appedix F.1 for details). I 2010, the average across states was 3%; the states with the highest icome tax rates were Orego 6.2%), North Carolia 5.2%), ad Hawaii 5.0%), while seve states had o icome tax. 16 E.g., Devereux ad Griffith 1998) estimate the effect of profit taxes o the locatio of productio of U.S. multiatioals, Goolsbee ad Maydew 2000) estimate the effects of the labor apportiomet of corporate icome taxes o the locatio of maufacturig employmet, Hies 1996) exploits foreig tax credit rules to show that ivestmet respods to state corporate tax coditios, ad Giroud ad Rauh 2015) show that C-corporatios reduce their activity whe states icrease corporate tax rates. Chiriko ad Wilso 2008) ad Wilso 2009) also provide evidece cosistet with the view that state taxes affect the locatio of busiess activity. 17 Our paper is also related to the literature that has aalyzed the geeral equilibrium effects of tax chages. Shove ad Whalley 1972) ad Ballard et al. 1985) poit out the importace of geeral equilibrium effects whe aalyzig large chages i policy. See Nechyba 1996) for a early CGE model of local public goods. Albouy 2009) studies how federal tax progressivity impacts the allocatio of workers. A large literature i macroecoomics also studies the dyamic effects of taxes i the stadard growth ad real-busiess cycle model; Medoza ad Tesar 1998), amog others, study dyamic effects of taxes i a iteratioal settig. 18 The schedule of state icome tax rates teds to be progressive, but it is typically much flatter tha the federal icome tax schedule. We compare the progressivity of state ad federal icome tax rates whe we itroduce progressive icome taxes i Sectio

9 Corporate Icome Tax States also tax busiesses. The tax base ad tax rate o busiesses deped o the legal form of the corporatio. The tax base of C-corporatios is atioal profits. 19 State tax authorities determie the share of a C-corporatio s atioal profits allocated to their state usig apportiomet rules, which aim to capture the corporatio s activity share withi their state. To determie that activity share, states put differet weight o three apportiomet factors: payroll, property, ad sales. Payroll ad property factors deped o where goods are produced ad typically coicide; the sales factor depeds o where goods are cosumed. 20 I 2012, the average corporate icome tax rate across states was 6.4%; the states with the highest corporate tax rates were Iowa 12%), Pesylvaia 10%), ad Miesota 10%), while six states had o corporate tax. Apportiomet through sales teds to be more prevalet: ietee states exclusively apportio through sales, while roughly half of the remaiig states apply either a 50% or 33% apportiomet through sales. Sice C-corporatios accout for the majority of et icome i the Uited States, i our bechmark aalysis we treat all busiesses as C-corporatios. 21 We also explore how our results chage whe we apply alterative corporate tax rates that adjust for the fractio of C-corporatios i total reveue i each state, or that accout for tax subsidies that some states grat to firms, reducig their effective corporate tax rate. Sales Tax Sales taxes are usually paid by the cosumer upo fial sale, ad states typically do ot levy sales taxes o firms for itermediate iputs or goods that they will resell. 22 I 2012, the average sales tax rate was 5%; the states with the highest sales tax rates were New Jersey 10%), Califoria 7.5%), ad Idiaa 7%), while five states had o sales taxes. I our bechmark aalysis, we defie the sales tax rate as the statutory geeral sales tax rate applied oly to fial cosumer sales. 3.2 Dispersio i Tax Rates ad i Tax Reveue across States Both tax rates ad tax bases vary cosiderably across states. Pael a) of Figure 1 shows the 2010 distributio of sales, icome, corporate, ad sales-apportioed corporate tax rates. 23 For each tax, rates vary across states, corporate tax rates beig the most dispersed; the percetiles of the distributios of sales, average persoal icome, ad corporate icome tax rates are 7%-1%, 5%-0%, ad 9%-0%, respectively. For each type of tax, there are at least five states with 0% rates. 19 Most states limit the tax base to profits eared withi the water s edge, i.e., profits from domestic activity. 20 For example, a sigle-plat firm j located i state i with export share s j i to each state pays a corporate tax rate of t j = t corp fed + tl i + sj i tx, where t corp fed is the federal tax rate, tx = θt x corp is the corporate tax apportioed through sales i state where t corp is the corporate tax rate of state ad θ x is its sales apportiomet), ad t l i = 1 θi x ) t corp i is the corporate tax apportioed through property ad payroll i state i. 21 C-corporatios accouted for 66% percet of total busiess receipts i 2007 PERAB, 2010). 22 Most states make some kid of exceptio of sales tax for firms purchasig goods. These exemptios vary widely across states, but geerally, if a firm purchases material ad uses it as a iput i productio, it is exempt from the sales tax. For example i Alabama, property that becomes a igrediet or compoet part of products maufactured or compouded for sale costitutes a exempt wholesale sale. Ala Code Sec a)6); Ala Code Sec a)9b); Ala Code Sec )b); Ala Admi Code r ; Ala Admi Code r ). 23 The sales-apportioed corporate tax rate is the product of the sales apportiomet factor which is betwee 0 ad 1) ad the corporate rate; i.e., it is t x = θt x corp defied i footote 20. Table A.2 i Appedix F.2 shows the state tax rates i 2007 i all 50 states. Table A.1 shows the federal icome, corporate, ad payroll tax rates i

10 These differeces i tax structures across states are associated with differeces i total tax reveue collected. Pael b) of the same figure shows the distributio i tax reveue as share of GDP across states. The share of the sum of icome, sales, ad corporate tax reveue i GDP varies across states betwee 2% ad 7%. While most states collect both icome ad sales taxes, some rely almost exclusively o sales tax reveue, such as Texas ad Nevada, while others are sales-tax free, like New Hampshire ad Orego. Local sub-state) govermets also tax residets. Overall, state taxes amout to roughly 60% of state ad local tax reveue combied. 24 Heterogeeity i tax rates across states is also preset whe both state ad local taxes are take ito accout. Figure A.1 i the olie appedix reproduces Pael a) of Figure 1 usig the sum of state ad local tax rates. It shows that cross-state differeces i tax rates icrease whe local tax rates are take ito accout. Figure 1: Dispersio i State Taxes i 2010 a) Distributio of Tax Rates Across States b) Tax Reveue as Share of GDP Across States i 2010 Desity State Tax Rates i 2010 Sales Corporate Idividual Icome Sales Apportioed Corporate State Tax Reveue as Share of GDP i AK DE NH WY TX SD CO LA NV OR GA TN VA OK MT IL MO WA ND FL AZ NE UT SC AL OH IA NM MD NC PA KS IN ID NJ MA RI CA KY MI CT NY WI VT MN AR MS ME Icome Sales Corporate WV HI 3.3 Relatioship Betwee State Tax Reveue ad Govermet Spedig Besides the three types of taxes discussed i Sectio 3.1, a major source of reveue of U.S. state govermets is the trasfers that they receive from the federal govermet. O average, these trasfers amout to roughly 6% of state GDP. Oce these federal govermet trasfers are take ito accout, state govermets typically have balaced budgets Poterba, 1994). Federal trasfers therefore allow state spedig to exceed state tax reveue. The actual process determiig the level of trasfers ejoyed by each state i each year is complex. However, empirically, for the period 1980 to 2010, the size of the total direct expeditures of each state is very well approximated as a state-specific multiplier of state tax reveue. That is, lettig Et G be state s expeditures i year t, ψ > 0 be a state-specific multiplier of state tax reveue, ad R t be state tax reveue, the 24 Local govermets rely more heavily o property taxes tha icome, corporate, ad sales taxes. State tax reveue make up roughly 90%, 85%, ad 80% of cosolidated state ad local reveue from icome, corporate, ad sales taxes, respectively, but oly 3% of cosolidated property tax reveue. 9

11 estimates of the regressio le G t) = l 1 + ψ ) + lr t ) + ε t 1) yield a R 2 of Therefore, our model assumes the relatioship Et G = R t + T fed st, where T fed st = ψ R t is the part of state spedig fiaced through federal trasfers Quatitative Geography Model with State Taxes ad Public Goods 4.1 Model Overview We model a closed ecoomy with N states idexed by or i. A mass M of firms ad L of workers respectively receive idiosycratic productivity ad preferece shocks, which gover how they sort across states. We let M ad L be the measure of workers ad firms that locate i state. Chagig M or L does ot affect the allocatio, resultig oly i a scalig up of M or L everywhere; therefore, we ormalize M ad L to 1, implyig that M ad L are the fractios of firms ad workers located i state. Each state has a edowmet H of fixed factors of productio lad ad structures), a ameity level u, ad a productivity level z. There is a iceberg cost τ i 1 of shippig from state i to state if oe uit is shipped from i to, 1/τ i uits arrive). Firms are sigle-plat ad sell differetiated products. To produce, they use the fixed factor, workers, ad itermediate iputs usig techologies that may vary across states. Workers receive oly labor icome, which they sped i the state where they live. Firms ad fixed factors are owed by immobile capital owers exogeously distributed across states. State govermets collect persoal icome taxes t y, sales taxes t c, ad corporate icome taxes apportioed through sales, t x, or through payroll ad fixed factors, t l. Each state uses the tax reveue to fiace the provisio of public services, which eter as shifters both of that state s ameity ad of the productivity of firms that locate i that state. The weight of public services i prefereces may vary across states. The federal govermet collects persoal icome taxes t y fed, payroll taxes tw fed, ad corporate taxes t corp fed. Federal taxes are icluded because they affect effective state tax rates ad are used to fiace both federal trasfers to the state govermets ad federal public goods that are equally valued by cosumers idepedetly of where they locate e.g., we assume that federal spedig i atioal defese equally beefit ay U.S. worker regardless of where that worker is located). 25 We measure the variable E G t as the state direct expeditures from the Cesus of Govermets. The mai direct-expediture items icluded i this measure are educatio, public welfare, hospitals, highways, police, correctio, atural resources, parks ad recreatio, govermet admiistratio, ad utility expediture. 26 It is worth otig that the fit of a regressio that assumes that the amout of federal trasfers ejoyed by each state i every year ca be approximated as a state-specific costat rather tha a state-specific multiplier of tax reveue) yields a worse fit. Specifically, the estimates of the regressio that assumes that E G t = ψ + R t + ε t yield a R 2 of

12 4.2 Productio Techologies I each state, a competitive sector assembles a fial good from differetiated varieties through a costat elasticity of substitutio CES) aggregator with elasticity σ, ˆ Q = i j J i ) σ ) q j σ 1 σ 1 σ i dj, 2) where J i deotes the set of varieties produced i state i ad q j i is the quatity of variety j produced i state i ad used for productio of the fial good i state. Lettig p j i be the price of this variety i state, the cost of producig oe uit of the fial good i state ad also its price before sales taxes) is ˆ P = i j J i ) 1 ) 1 σ p j 1 σ i dj. 3) Each variety j is produced by a differet firm; to produce q j i i regio i, firm j combies its ow productivity i that locatio z j i, the fixed factor hj, workers l j, ad itermediate iputs i j, through a Cobb-Douglas techology: [ ) q j i = 1 h j βi ) l j 1 βi ] γi ) i j 1 γi zj i, 4) γ i β i 1 β i 1 γ i where γ i is the value-added share i productio of every firm i state i, ad 1 β i is the labor share i value added i state i. The existece of a fixed factor is oe of the sources of cogestio i the model; the higher the umber of firms ad workers located i a give state, the higher the relative price of this fixed factor. Productio fuctios are allowed to vary by state; this flexibility is eeded to match the heterogeeity i the shares of total paymets to labor ad itermediate iputs expeditures i states GDP observed i the data. 27 The fial good Q is o-traded ad ca be used by cosumers workers ad capital-owers) for aggregate cosumptio C ), by firms as a itermediate iput i productio I ), ad by state govermets G ) ad the federal govermet G fed ) as a iput for the supply public services: 4.3 Workers Q = C + I + G + G fed. 5) A cotiuum of workers l [0, 1] decide i which state to work ad cosume. The idirect utility of worker l i state is v l = v ɛ l, where the vector { ɛ l N } captures worker l s idiosycratic =1 prefereces for livig i each state ad v is commo to all workers who locate i. This commo 27 This heterogeeity i the productio fuctio may be thought of as a way of capturig differeces i sectoral compositio across states; i the presece of multiple sectors, the labor ad itermediate-iput shares of each state would be edogeous ad chage i the couterfactuals, but we abstract from this margi i our aalysis. 11

13 compoet is ) αw, G v = u L χ 1 T W ) w ) 1 αw,, 6) P where we defie the workers tax keep-rate i.e., the fractio of real icome, w /P, kept by workers after payig sales ad icome taxes) as 1 T 1 ty fed )1 ty ) t w fed 1 + t c. 7) Equatios 6) ad 7) imply that workers have prefereces over ameities, public goods, ad fial cosumptio goods. 28 First, the coefficiet u captures both atural characteristics, like the weather, ad the rate at which the govermet trasforms total real spedig ito services valued by workers; this rate icludes the fractio of the state budget used to fiace public services valued oly by workers. 29 For simplicity, we refer to this term as simply ameities. Secod, the appeal of state also depeds o real govermet spedig, G, ormalized by L χ W. The parameter χ W captures rivalry i public goods, ad rages from χ W = 0 o-rival) to χ W = 1 rival). 30 Third, workers care about the quatity of) fial goods that ) they ca cosume i state. This quatity equals after-tax wages, 1 t y fed 1 t y ) t w fed w, ormalized by the after-tax price, 1 + t c ) P. 31 As a result, real cosumptio equals the pre-tax wage, w /P, adjusted by icome ad sales taxes, 1 T. The parameter α W, captures the weight of state-provided services i prefereces. The weight of public services i prefereces, α W,, may vary across states, reflectig complemetarities betwee state-specific features such as the weather or atural ameities ad govermet services. The idiosycratic taste draw ɛ l is assumed to be i.i.d. across cosumers ad states, ad it follows a Fréchet distributio, Pr ɛ l < x ) = e x ε W, with ε W > 1. A worker l locates i a state if = arg max v ɛ l. Remidig the reader that we have ormalized the mass of workers to 1, the fractio of workers located i state is L = v ) εw, 8) v 28 The framework could easily be geeralized to allow for direct cosumptio of the fixed factor by workers i the form of housig. I that specificatio, the price of lad would also eter as part of the cost of livig. Additioally, the effective tax keep-rate could be modified to also accout for average property taxes, ad housig supply could be allowed to be elastic. While extedig the model with these forces would be straightforward, quatifyig them would be less so because property taxes are largely imposed at the local sub-state) level, ad housig supply elasticities vary cosiderably across cities withi states, as documeted by Saiz 2010). 29 The coefficiet u may also capture utility from a atioal public good provided by the federal govermet. ) αw, ) 1 αw,, 1 T ) w P where u,0 is the amout of atioal public services provided by the federal govermet, More specifically, 6) is cosistet with first defiig v = u,0g fed G W L χ W are atural characteristics, G fed G W = z W θ W G are govermet services valued by workers, z W is the efficiecy or the quality of real spedig i services valued by workers, ad θ W is the fractio of the state budget dedicated ) to services valued by workers. Startig from this iitial defiitio, 6) correspods to defiig u = G W u,0 z W θ W αw,. 30 This is a similar modelig approach to existig papers i the fiscal competitio literature, e.g. see Boadway ad Flatters 1982). 31 Note that equatio 7 takes ito accout that state icome taxes ca be deducted from federal taxes. I our bechmark aalysis, we abstract from the progressivity of both federal ad state icome taxes. I Sectio 7.5, we relax this assumptio ad allow the federal ad the state persoal icome tax rates to be a fuctio of state wage. 12

14 where v v ε W ) 1/εW. 9) Uder the Fréchet distributio, both the ex-ate expected utility of a worker before drawig { ɛ l } N =1 ad the average ex-post utility of agets located i ay state are idetical ad proportioal to v; hece, we adopt it as our measure of worker welfare. 32 A larger value of ε W implies that the idiosycratic taste draws are less dispersed across states; as a result, locatios become closer substitutes ad a icrease i the relative appeal of a locatio a icrease i v /v) leads to larger respose i the fractio of workers who choose to locate there. From the defiitios of v ad L i 6) ad 8), it follows that ε W 1 α W, ) is the partial elasticity of the fractio of workers who locate i state with respect to after-tax real wages, 1 T )w /P ), while ε W α W, is the partial elasticity with respect to real govermet services per worker, G /L χ W. We rely o these relatioships to estimate {ε W, α W, } i Sectio Capital Owers Immobile capital owers i state ow a fractio b of a portfolio that icludes all firms ad fixed factors, idepedetly of the state i which they are located. We do ot eed to specify the umber of capital owers located i each state for our computatios. I the model, a larger owership rate relative to other states results i larger trade imbalaces. Therefore, we will calibrate the owership shares b to match the observed trade imbalaces across states. 33 Capital owers sped their icome locally, pay sales taxes o cosumptio, ad pay both federal ad state icome taxes o their icome Firms A cotiuum of firms j [0, 1] decide i which state to produce ad how much to sell to every state. Each firm j produces a differetiated variety ad is edowed with a vector of productivities {z j i }N i=1 across states. Firms are moopolistically competitive; whe a firm j located i state i sets its price p j i i state, the quatity exported to state is qj i = Q p j i /P ) σ. We first describe the profit maximizatio problem faced by firms located i a give state, ad the solve the firms locatio problem. We fially discuss some of the aggregatio properties of our model. 32 The costat of proportioality equals Γ εw 1 ε W ), where Γ ) is the gamma fuctio. 33 See Sectio 6.2 for details. Two alterative modelig approaches would be to assume that all workers ow equal shares of the atioal portfolio, or that the returs of that portfolio are spet outside of the model. Uder these approaches, the model would lead to empirically icosistet predictios for trade imbalaces across states. I cotrast, our curret approach allows us to disciplie the assumptios o factor owership with observed data o trade imbalaces. 34 Whe cosiderig progressive federal ad state icome i Sectio 7.5, we will assume that capital owers are subject to the highest margial federal ad state icome tax rates. Cooper et al. 2015) show that the majority of busiess icome accrues to very high icome earers. 13

15 Profit Maximizatio give Firm Locatio Cosider a firm j located i state i whose productivity is z j i. The, its profits are π i z j i ) = max 1 t j {qi} j i ) N =1 x j i c i z j i ) N τ i q j i, 10) where t j i is the corporate tax rate of firm j if it were to locate i state i, xj i = P Q 1 σ q j i )1 1 σ =1 are its sales to state, ad c i = w 1 β i i r β i i ) γ i P 1 γ i i is the the cost of the cost-miimizig budle of factors ad itermediate iputs, where r i stads for the cost of a uit of lad ad structures i state i. 35 All firms face corporate taxes apportioed through sales, payroll, ad lad ad structures. 36 A firm j located i state i whose share of sales to state is s j i pays sj i tx times the pre-tax atioal profits i corporate taxes apportioed through sales to state. Firms located i i also pay t l i times the pre-tax atioal profits i corporate icome taxes apportioed through payroll ad lad ad structures to state i, ad a rate t corp fed i federal corporate icome taxes. As a result, the corporate tax rate of firm j is: t j i = tcorp fed + tl i + N t x s j i. 11) Due to the sales apportiomet of corporate taxes, the decisio of how much to sell to each state i 10) is ot separable across states as i the stadard CES maximizatio problems with costat margial productio costs i Krugma 1980) or Melitz 2003). Whe a firm icreases the fractio of its sales to state i.e., whe s j i icreases), the average tax rate chages depedig o the sales-apportioed corporate tax i state, t x, relative to that i other states. Sice the corporate tax base is atioal profits, firms trade off the margial pre-tax beefit of exportig more to a give state agaist the potetial margial cost of icreasig the corporate tax rate o all its profits. Pricig Distortio Through Corporate Taxes =1 Despite the o-separability of the sales decisio across markets, the solutio to the firm optimizatio problem retais coveiet properties from the stadard CES maximizatio problem that allow for aggregatio; we describe these properties here ad refer to Appedix B.1 for details. Specifically, all firms located i a state i have the same sales shares across destiatios irrespective of their productivity, i.e., s j i = s i for all firms j located i i; from 11), this leads to a commo corporate tax rate across firms, t j i = t i. Additioally, firms set idetical, costat markups over margial costs, but these markups vary bilaterally depedig o corporate taxes. The price set i by a firm with productivity z located 35 Note that the defiitio of c i accouts for the fact that, ulike cosumers, firms do ot face the sales tax whe purchasig the fial good to be used as a itermediate. See footote 22). 36 This assumptio implies that we treat all compaies as C-corporatios. I practice, may compaies are set up as S-corporatios ad parterships. These compaies are ot subject to corporate icome taxes. We igore them i our baselie model because they represet a small fractio of U.S. busiess reveues see our previous discussio i Sectio 3.1. However, i Sectio 7.6.1, we perform a robustess check where corporate tax rates are adjusted by the actual share of C-corporatios i each state. 14

16 i state i is: where σ σ c i p i z) = τ i σ t i σ 1 z, 12) t i tx tx s i 1 t i. 13) The term t i is a pricig distortio created by heterogeeity i the sales-apportioed corporate tax rates. The pricig distortio icreases with the sales tax i the importig state, t x, relative to other states, implyig higher prices for states with higher sales-apportioed corporate taxes. If there is o dispersio i the sales-apportioed corporate tax rates t x = t x for all ), the pricig decisio becomes the same as i the stadard CES maximizatio problem t i = 0 for all i ad ). Firm Locatio Choice We assume that firm-level productivity z j i ca be decomposed ito a term z 0 i commo to all firms that locate i i ad a firm-state specific compoet ɛj i : zj i = z0 i ɛj i. The commo compoet of productivity is: z 0 i = Gi M χ F i ) αf z 1 α F i. 14) As i the case of ameities, this commo compoet has a edogeous part that depeds o the amout of public spedig ad a exogeous part, z i. The edogeous part equals real govermet spedig G i ormalized by M χ F i, where the parameter χ F captures rivalry amog firms i access to public goods. The exogeous part captures both atural characteristics that impact productivity, like atural-resource availability, the rate at which the govermet trasforms real spedig ito services valued by firms, ad the share of public goods provided by state govermets that icrease the productivity of the firms located i their states. 37 Firm j decides to locate i state i if i = arg max i π i z j i ). The idiosycratic compoet of productivity, ɛ j i, is i.i.d. across firms ad states ad is draw from a Fréchet distributio, Prɛ j i < x) = e x ε F. This distributio implies that firm-level profits, π i z j i ), are also Fréchet-distributed with shape parameter ε F / σ 1) > As a result, ad remidig the reader that we have ormalized the mass of firms to 1, the fractio of firms located i state i is M i = πi z 0 i ) π ) ε F σ 1, 15) ) 37 More specifically, 14) is cosistet with first defiig zi 0 G F αf = i M χ F z 1 α F i,0, where z i,0 are atural characteristics, G F i = zi F 1 θ i) G are govermet services valued by firms, z F is the efficiecy or the quality of real i spedig is the fractio of the state budget dedicated ) to services valued by firms. Startig z F i θi F αf. Remidig the reader that θi W is the fractio of the state budget dedicated to services valued by workers see Footote 29), we ote that θi F + θi W may be greater tha 1. I.e., it is possible that the same govermet spedig is valued by both workers ad firms. 38 This follows from the fact that, combiig 10) ad 14), we ca express the profits of firm j whe it locates i state i as the product of a commo ad a idiosycratic compoet: π iz j i ) = πiz0 i )ɛ j i )σ 1. i services valued by firms, ad θi F from this iitial defiitio, 14) correspods to defiig z i = z 1 α F i,0 15