Tax Cuts for Whom? Heterogeneous Effects of Income Tax Changes on Growth & Employment

Transcription

1 Tax Cuts for Whom? Heterogeneous Effects of Income Tax Changes on Growth & Employment Owen Zidar University of California, Berkeley NBER Summer Institute 2013: Public Economics July 23, 2013 Owen Zidar (UC Berkeley) Tax Cuts for Whom? July 23, / 48

2 Variation in Tax Policy & Structure of Income Tax Changes Average Change in Tax Liability as Share of Income Income Percentile Graphs by Year Owen Zidar (UC Berkeley) Tax Cuts for Whom? July 23, / 48

3 Research Questions How does the composition of income tax changes affect subsequent output & employment? Do tax cuts for high income taxpayers generate more employment & output growth than equivalently sized tax cuts for low and moderate income taxpayers? If so, why? 1 Traditional PF: Labor supply effects via marginal tax rates 2 Macro: Effects on Aggregate Demand 3 This Paper: Aggregate Demand but with focus on distributional effects Owen Zidar (UC Berkeley) Tax Cuts for Whom? July 23, / 48

4 Overview 1 Conceptual Framework: Aggregate demand effects due to redistribution from savers to constrained/less patient borrowers 2 Empirical Approach: National: Romer & Romer AER 2010 disaggregated by income group Regional: variation in income distribution across states 3 Data: Historical returns & counterfactuals from NBER TAXSIM 4 Results: Tax cuts for those with high incomes lead to substantially less employment growth and economic activity than similarly sized tax cuts for those with low and moderate incomes Owen Zidar (UC Berkeley) Tax Cuts for Whom? July 23, / 48

5 Motivation Why study the impacts of these tax changes and how they vary over the income distribution? Model: New Keynesian vs RBC Policy Design: Optimal stimulus design Effects of ending the Bush tax cuts for specific income groups Effects of mass refinancing Owen Zidar (UC Berkeley) Tax Cuts for Whom? July 23, / 48

6 Some Relevant Literature Little direct evidence likely due to empirical issues: endogeneity, simultaneity, and observability Macro Empirical: Romer & Romer (AER 2010). Mertens & Ravn (AER 2013) Theoretical: Monacelli and Perotti (2011), Heathcote (2005), Gali, Lopez-Salifo, and Valles (2007) Consumption responses to Taxes and Transfers Minimum Wage Aaronson, Agarwal, and French (AER 2012) MPC Jappelli and Pistaferri (2012 & 2010), Dynan Skinner and Zeldes (2001), McCarthy (1995), Parker (1999). Owen Zidar (UC Berkeley) Tax Cuts for Whom? July 23, / 48

7 I. Conceptual Framework Overview Agents with different MPCs because some constrained or myopic Consider lump sum redistribution τ b = τ s Increases aggregate consumption because c b,t and c s,t In standard new Keynesian framework, higher consumption increased output, L D, and employment Owen Zidar (UC Berkeley) Tax Cuts for Whom? July 23, / 48

8 II. Empirical Framework: Background (1/2) Romer & Romer (AER 2010) Y t = α + β Tax t + ɛ t (1) Types of Tax Changes 1 Counteract economic forces 2 Spending offsets 3 Address inherited deficit 4 Promote long run growth Owen Zidar (UC Berkeley) Tax Cuts for Whom? July 23, / 48

9 II. Empirical Framework: Background (2/2) Y t = α + M i=0 b i Tax t i + e t Owen Zidar (UC Berkeley) Tax Cuts for Whom? July 23, / 48

10 II. Empirical Framework: (1) National Approach Romer & Romer Specification Y t = α + M b i Tax t i + e t Decompose Romer Tax Shocks Tax B90,t for Bottom 90% Tax t is Tax T 10,t for Top 10% Tax NON,t for Non Income Changes i=0 Allow for different effects: b m vs β B90,m, β T 10,m, β NON,m Owen Zidar (UC Berkeley) Tax Cuts for Whom? July 23, / 48

11 II. Empirical Framework: (1) National Approach Output growth & exogenous tax changes for different income groups Y t =β 0 + β B90,0 ( Tax B90,t ) + β T 10,0 ( Tax T 10,t ) + β NON,0 ( Tax NON,t ) +... }{{} =b 0 Tax t + β B90,m ( Tax B90,t m ) + β T 10,m ( Tax T 10,t m ) + β NON,m ( Tax NON,t m ) }{{} =b m Tax t m + ɛ t Tax B90 and Tax T 10 are changes in income and payroll taxes as a share of GDP for the bottom 90% and top 10% respectively Assume Cov( Tax g,t, ɛ t ) = 0 g (BOT 90, TOP10, NONINCOME) following Romer & Romer AER 2010 Frisch Waugh Owen Zidar (UC Berkeley) Tax Cuts for Whom? July 23, / 48

12 II. Empirical Framework: (1) National Approach Add government transfer shocks for compositional concerns Y t =β 0 + β B90,0 ( Tax B90,t ) + β T 10,0 ( Tax T 10,t ) + β NON,0 ( Tax NON,t ) +... }{{} =b 0 Tax t + β B90,m ( Tax B90,t m ) + β T 10,m ( Tax T 10,t m ) + β NON,m ( Tax NON,t m ) }{{} =b m Tax t m M M + λ m ( Transfers t m ) + η m ( Y t m ) + ɛ t m=0 m=1 Owen Zidar (UC Berkeley) Tax Cuts for Whom? July 23, / 48

13 II. Empirical Framework: (2) Regional Approach Exploit variation in income distribution across states Heterogeneity: NJ & CT have 4X share of top 10% vs SD Idea: When national tax policy affects high income taxpayers, states with large shares of high income taxpayers will face larger shocks Labor literature: Bartik (1991), Card (1992), Katz & Murphy (1992), Moretti (2004) Test: If high income tax cuts have substantial effects, CT and NJ should grow faster following national high income tax cuts Value: Provides additional identifying variation Owen Zidar (UC Berkeley) Tax Cuts for Whom? July 23, / 48

14 II. Empirical Framework: (2) Regional Approach Income Tax Shocks by State Tax T 10,s,t for s = CT Tax Tax T 10,t is T 10,s,t for s = NJ... Tax T 10,s,t for s = SD Owen Zidar (UC Berkeley) Tax Cuts for Whom? July 23, / 48

15 II. Empirical Framework: (2) Regional Approach State emp growth & state tax shocks for different income groups M Y s,t = α + (β B90,m T B90,s,t m + β T 10,m T T 10,s,t m + X s,t m λ s,m ) m=0 + η s + φ t + ɛ s,t T B90,s,t is the exogenous change in taxes as a share of state GDP for taxpayers who are in the bottom 90 percent of AGI nationally Assume Cov ( Tax g,s,t m, ɛ s,t ) = 0 g, s, m < 3 Owen Zidar (UC Berkeley) Tax Cuts for Whom? July 23, / 48

16 III. Data Overview National Data: Dependent Variables: Employment (BLS) & macro aggregates(bea) 2 Independent Variables: SOI, NBER TAXSIM for 1960+, standard controls State Data: Dependent Variables: Employment data from BLS 2 Independent Variables: NBER TAXSIM and controls (government transfers, state taxes, population data from BEA) Owen Zidar (UC Berkeley) Tax Cuts for Whom? July 23, / 48

17 Data: Constructing tax changes Tax Change Measure is a function of three things: 1 Income and deductions from year prior to an exogenous tax change 1 2 Old tax schedule 3 New tax schedule 1 Preliminary tests suggest that results are robust to using two year lags and various inflation adjustments Owen Zidar (UC Berkeley) Tax Cuts for Whom? July 23, / 48

18 Data: Constructing tax changes Example: 1993 Omnibus Budget Reconciliation Act Owen Zidar (UC Berkeley) Tax Cuts for Whom? July 23, / 48

19 Data: Constructing tax changes Example: 1993 Omnibus Budget Reconciliation Act Suppose a taxpayer made $180K in 1992 Based on the 1992 schedule & her income and deductions in 1992, she would have paid $50,500 Based on the 1993 schedule & her income and deductions in 1992, she would have paid $54,000 My measure assigns her a $3,500 tax increase in 1993 Owen Zidar (UC Berkeley) Tax Cuts for Whom? July 23, / 48

20 Data: Constructing tax changes 1000 Change in Tax Liability I do this calculation for entire sample of NBER returns 0 Owen Zidar (UC Berkeley) Adjusted Gross Income Tax Cuts for Whom? July 23, / 48

21 Comparison of Aggregate Changes w/ Romer Changes Tax Change as a Share of GDP Year Romer Tax Measure Income & Payroll Tax Changes Owen Zidar (UC Berkeley) Tax Cuts for Whom? July 23, / 48

22 Disaggregated Income & Payoll Tax Changes Income Only Percent of GDP Year Tax Change: Bottom 20% Tax Change: 21 40% Tax Change: 41 60% Tax Change: 61 80% Tax Change: Top 20% Owen Zidar (UC Berkeley) Tax Cuts for Whom? July 23, / 48

23 State Bartik Statistics TAXSIM has states for those with income<$200k, so I (1) use obs below cutoff and (2) extrapolate shares based on state shares of $150 to $200K. Top States Bottom States Rank State Top 10 Share Rank State Top 10 Share 1 NJ TN CT AL MD SC AK WV VA AR MA ME CA MT NY MS CO ID DC SD 3.8 Owen Zidar (UC Berkeley) Tax Cuts for Whom? July 23, / 48

24 IV. Results Overview National Data: 1 Output and Employment growth 2 Mechanisms: Consumption and Investment State Data: 1 Similar specification at state-level 2 Effects across the income distribution Owen Zidar (UC Berkeley) Tax Cuts for Whom? July 23, / 48

25 National Data: Employment & Romer Tax Shocks Employment Growth over 2 Years Sum of Tax Romer Changes as % of GDP (from T 2 to T) Owen Zidar (UC Berkeley) Tax Cuts for Whom? July 23, / 48

26 National Data: Employment & Top 10% Employment Growth over 2 Years Sum of Tax Changes for Top 10% as % of GDP (from T 2 to T) Owen Zidar (UC Berkeley) Tax Cuts for Whom? July 23, / 48

27 National Data: Employment & Bottom 90% Employment Growth over 2 Years Sum of Tax Changes for Bottom 90% as % of GDP (from T 2 to T) 1963 Owen Zidar (UC Berkeley) Tax Cuts for Whom? July 23, / 48

28 Effects of Romer Tax s on GDP & Emp. Growth Dependent Variable Growth Y Growth E (1) (2) (3) (4) Tax ROMER,t (0.8) (0.8) (0.4) (0.4) Tax ROMER,t 1-1.4*** -1.4** -0.6** -0.6* (0.5) (0.5) (0.3) (0.3) Tax ROMER,t *** -0.6** (0.5) (0.5) (0.3) (0.3) Control for Growth Y lags N Y N N Control for Growth E lags N N N Y R-squared Tax Change: β t + β t 1 + β t ** -2.17** -1.41** -1.27** (1.00) (0.89) (0.62) (0.50) Notes: Newey-West standard errors with lag of 2 in parentheses in Column (1) & (3). I allow for serial correlation by including Growth E,t k or Growth Y,t k for k (1, 2) in regressions Columns (2) & (4). Robust standard errors in parentheses for Column (2) & (4). *** p<0.01, ** p<0.05, * p<0.1. There are 61 observations for all columns. Owen Zidar (UC Berkeley) Tax Cuts for Whom? July 23, / 48

29 Dependent Variable: Growth E (1) (2) (3) Tax Bottom90,t (1.1) (1.3) (0.9) Tax Bottom90,t 1-2.7** -2.7** -2.5** (1.3) (1.1) (1.0) Tax Bottom90,t 2-2.4** -1.6* -1.3 (1.1) (0.9) (1.0) Tax Top10,t (1.5) (1.7) (1.1) Tax Top10,t (1.5) (1.9) (1.2) Tax Top10,t (0.8) (0.6) (0.5) Control for Growth Y lags N Y Y Control for Transfers to GDP t & lags N N Y R-squared Bottom90 Tax Change: β t + β t 1 + β t ** -4.91** -4.57* (2.44) (2.00) (2.35) Top10 Tax Change: β t + β t 1 + β t (2.65) (3.15) (1.94) Notes: OwenNewey-West Zidar (UC Berkeley) standard errors with lag of 2 intax parentheses Cuts for Whom? in Column (1). I allow for for serial correlation July 23, 2013 by including 29 / 48

30 Owen Zidar (UC Berkeley) Tax Cuts for Whom? July 23, / 48

31 State Data: Employment & Top 10% State Employment Growth Sum of Tax Changes for Residents in Top 10% as % of GDP (from T 2 to T) Owen Zidar (UC Berkeley) Tax Cuts for Whom? July 23, / 48

32 State Data: Employment & Bottom 90% State Employment Growth Sum of Tax Changes for Residents in Bot. 90% as % of GDP (from T 2 to T) Owen Zidar (UC Berkeley) Tax Cuts for Whom? July 23, / 48

33 Event Study: Tax Shocks and State Employment Percent Year Note: Top 10% in Red and Bottom 90% in Blue Owen Zidar (UC Berkeley) Tax Cuts for Whom? July 23, / 48

34 Dependent Variable: Growth E,s (1) (2) (3) (4) Tax Bot90,s,t (0.9) (1.0) (0.8) (0.7) Tax Bot90,s,t 1-3.2** -1.6** -2.2*** -1.4** (1.2) (0.7) (0.7) (0.6) Tax Bot90,s,t 2-2.1** (0.9) (0.6) (0.7) (0.6) Tax Top10,s,t (0.4) (0.2) (0.2) (0.3) Tax Top10,s,t (0.3) (0.2) (0.2) (0.3) Tax Top10,s,t (0.2) (0.2) (0.2) (0.2) Control for Growth E lags N Y Y Y Control for GovTrans PERCAP,s,t & lags N N Y Y R-squared Bottom 90: β t + β t 1 + β t * * -2.59** (2.53) (1.58) (1.50) (1.07) Top10: β t + β t 1 + β t * * (0.75) (0.35) (0.35) (0.28) Notes: All results are weighted by state population. Robust standard errors clustered by state are in parentheses. 1,297 Obs. *** p<0.01, ** p<0.05, * p<0.1. Owen Zidar (UC Berkeley) Tax Cuts for Whom? July 23, / 48

35 Testing Identifying Assumption with Artificial Tax Shocks CDF Estimated Cumulative Effect of Artifical Tax Change for Bottom 90% Owen Zidar (UC Berkeley) Tax Cuts for Whom? July 23, / 48

36 Effects for more income groups than Bottom 90 & Top 10 How does effect β vary over the income groups? A second order approximation of the β(g) function β(g) = θ 0 + θ 1 g + θ 2 g 2 Plug into estimating equation Growth E,t = α + β 1 τ 1,t + β 2 τ 2,t β 10 τ 10,t + X t λ + ɛt Growth E,t = α + (θ 0 + θ 1 + θ 2 ) τ 1,t + (θ 0 + θ θ ) τ 2,t X t λ + ɛt ( 10 ) ( 10 ) ( 10 ) Growth E,t = α + θ 0 τ g,t + θ 1 g τ g,t + θ 2 g 2 τ g,t + X t λ + ɛt g=1 g=1 g=1 Owen Zidar (UC Berkeley) Tax Cuts for Whom? July 23, / 48

37 Aggregate Effects Across the Income Distribution 0 Employment Growth AGI Decile This figure shows the third order approximation of the β(g) function, i.e., ˆθ 1 g + ˆθ 2 g 2 + ˆθ 3 g 3. Owen Zidar (UC Berkeley) Tax Cuts for Whom? July 23, / 48

38 Testing for Heterogeneous Effects: A 3 Piece Linear Spline Dependent Variable Growth E,s Tax Bottom50,s,t -2.8*** (0.5) Tax UpperMiddle50to90,s,t -1.1*** (0.3) Tax Top10,s,t -0.2* (0.1) Constant 0.8*** (0.1) Observations 1,297 R-squared Bottom50 vs Top10: β B90 βt T *** (0.39) Upper Middle vs Top10: β UM βt T ** (0.38) Notes: Controlled for serial correlation by including Growth E,t k for k (1, 2) in regressions. Squared and cubic lags (Growth E,t 1 ) j for j (2, 3) with lagged Gov Transfers also included. All results are weighted by state population. Robust standard errors clustered by state in parentheses. *** p<0.01, ** p<0.05, * p<0.1. Owen Zidar (UC Berkeley) Tax Cuts for Whom? July 23, / 48

39 Conclusion Summary 1 Construct a new measure of income tax changes 2 Show substantial heterogeneity in effects of fiscal policy 3 Find stimulative effect of income tax cuts are largely from bottom 90% and empirical link between employment growth and tax changes for upper income earners seems weak to negligible 4 Suggest letting Bush tax cuts expire for $250K won t have substantial employment consequences over the business cycle Owen Zidar (UC Berkeley) Tax Cuts for Whom? July 23, / 48

40 APPENDIX: National Summary Stats: Table: National Summary Statistics: Variable Mean Std. Dev. Min. Max. N Year Growth E,t Growth Y,t Tax ROMER,t Tax Bottom90,t Tax Top10,t P[1] p1 Tax NONINCOME,t lnconsumption t lndurables t lnnondurables t lninvestment t lnresidentialinv t Transfers to GDP t Fed Funds Rate t PCE Inflation t Unemployment t Notes: The Tax variables are percent of Nominal GDP (i.e. 100 GDPt τ Owen Zidar (UC Berkeley) Tax Cuts for Whom? July 23, / 48

41 APPENDIX: State Summary Stats: Table: State Summary Statistics: Variable Mean Std. Dev. Min. Max. N Year Growth E,s,t Tax B90,s,t Tax T 10,s,t Bartik Tax Shock T 10,s,t Unemployment s,t GovTransfers PERCAP,s,t lngovtransfers PERCAP,s,t lnstatelocaltaxes PERCAP,s,t EPOP s,t Notes: Units for the Tax variables are percent of Nominal State GDP (i.e. 100 GDPs,t τ Owen Zidar (UC Berkeley) Tax Cuts for Whom? July 23, / 48

42 APPENDIX: Frisch Waugh -Tax s for Top vs Bottom Change in Tax Liability for Top 10% as % of GDP Change in Tax Liability for Bottom 90% as % of GDP Owen Zidar (UC Berkeley) Tax Cuts for Whom? July 23, / 48

43 Dependent Variable Growth Y (1) (2) (3) (4) Tax Bottom90,t (1.7) (1.8) (1.4) (1.3) Tax Bottom90,t ** -2.3** (2.6) (2.5) (1.1) (1.1) Tax Bottom90,t (2.2) (2.0) (1.3) (1.1) Tax Top10,t (2.3) (2.8) (1.7) (1.5) Tax Top10,t (2.5) (2.9) (1.3) (1.3) Tax Top10,t (1.2) (1.5) (0.7) (0.8) Constant 2.9*** 2.3*** 3.8*** 8.8*** (0.4) (0.8) (1.2) (1.6) Control for Tax NONINCOME,t Y Y Y Y Control for Growth Y lags N Y Y Y Control for Transfers to GDP t & lags N N Y Y Control for Debt to GDP t N N N Y Control for Fed Funds Rate t & Inflation t N N N Y Observations R-squared Bottom90 Tax Change: β t + β t 1 + β t * t-stat p-val Top10 Tax Change : β t + β t 1 + β t t-stat p-val Notes: Newey West standard errors with lag of 2 in parentheses in Column (1). I allow for for serial correlation by including Growth E,t k for k (2, 3, 4) in regressions. Robust standard errors in parentheses for Column (2) & (4). *** p<0.01, ** p<0.05, * p<0.1. Owen Zidar (UC Berkeley) Tax Cuts for Whom? July 23, / 48

44 Unanticipated Bartik Top10 Dependent Variable Growth E Growth E,s Growth E,s (1) (2) (3) Tax Bot90,t -0.9 (1.3) Tax Bot90,t 1-1.6* (0.9) Tax Bot90,s,t (1.7) Tax Top10,t 0.8 (1.2) Tax Top10,t 1 1.7* (0.9) Tax Top10,t (1.1) Tax Bot90,s,t (0.8) (0.6) Tax Bot90,s,t 1-4.0*** -1.7** (1.1) (0.7) Tax Bot90,s,t (0.9) (0.6) BartikTax Top10,s,t (0.3) (0.7) BartikTax Top10,s,t (0.2) (0.9) BartikTax Top10,s,t *** (0.2) (0.6) Constant 5.8*** 1.4** -1.1 (1.3) (0.5) (0.8) Observations 57 1,297 1,271 R-squared Bottom90 Tax Change: β t + β t 1 + β t * *** *** t-stat p-val Owen Zidar (UC Top10 Berkeley) Tax Change : β t + β t 1 + Tax β t 2 Cuts for Whom? 3.489* * July 23, / 48

45 APPENDIX: FAVERO & GIAVAZZI Orthogonality Test Top 10% Tax Shock vs. Residual Residual is Tax Shock after Partialing Out Lagged Macro Aggregates Tax Change as a Share of GDP Year Top 10% Tax Shock Residual Top 10% Tax Shock Owen Zidar (UC Berkeley) Tax Cuts for Whom? July 23, / 48

46 APPENDIX: FAVERO & GIAVAZZI Orthogonality Test Bottom 90% Tax Shock vs. Residual Residual is Tax Shock after Partialing Out Lagged Macro Aggregates Tax Change as a Share of GDP Year Bottom 90% Tax Shock Residual Bottom 90%Tax Shock Owen Zidar (UC Berkeley) Tax Cuts for Whom? July 23, / 48

47 APPENDIX: Tax Changes by Group with Top 1% back Percent of GDP Year Tax Change: Bottom 20% Tax Change: 21 40% Tax Change: 41 60% Tax Change: 61 80% Tax Change: Top 1% Tax Change: 81 99% Owen Zidar (UC Berkeley) Tax Cuts for Whom? July 23, / 48

48 Testing for Heterogeneous Effects (without Transfers) Dependent Variable Growth E,s Tax Bottom50,s,t -1.8*** (0.5) Tax UpperMiddle50to90,s,t -1.4*** (0.3) Tax Top10,s,t -0.2* (0.1) Constant -0.2 (0.1) Observations 1,297 R-squared Bottom50 vs Top10: β B90 βt T p-val Upper Middle vs Top10: β UM βt T t-stat p-val Notes: Controlled for serial correlation by including Growth E,t k for k (1, 2) in regressions. Squared and cubic lags (Growth E,t 1 ) j for j (2, 3) also included. All results are weighted by state population. Robust standard errors clustered by state in parentheses. *** p<0.01, ** p<0.05, * p<0.1. Owen Zidar (UC Berkeley) Tax Cuts for Whom? July 23, / 48