TAXABLE INCOME RESPONSES. Henrik Jacobsen Kleven London School of Economics. Lecture Notes for MSc Public Economics (EC426): Lent Term 2014

Transcription

1 TAXABLE INCOME RESPONSES Henrik Jacobsen Kleven London School of Economics Lecture Notes for MSc Public Economics (EC426): Lent Term 2014

2 AGENDA The Elasticity of Taxable Income (ETI): concept and policy relevance. The long-run evolution of top marginal tax rates and top income shares. Estimating the ETI: empirical strategies, identification problems, and findings. The high-income Laffer curve.

3 MOVING BEYOND LABOR SUPPLY A large literature estimates the elasticity of labor supply. Estimated labor supply elasticities are close to zero, which suggests that the efficiency cost of taxation is very small. But there are many other dimensions of behavioral response, which might create efficiency losses. A shift in focus from labor supply responses to taxable income responses, which capture the full range of responses to taxation.

4 CHANNELS OF TAXABLE INCOME RESPONSE (1) Quantitative labor supply responses: hours worked, participation. (2) Qualitative labor supply responses: effort on the job, type of job, training, education. (3) Changes in savings and portfolio choice. (4) Legal shifting of income into untaxed or lower-taxed form [tax avoidance]. (5) Illegal under-reporting of income [tax evasion].

5 ELASTICITY OF TAXABLE INCOME (ETI) Feldstein (1995, 1999) first pointed out the potential importance of the ETI: he argued that the ETI provides a sufficient statistic for revenue effects, deadweight loss, and optimal taxation. Joel Slemrod (1998): "recently... much attention has been focused on an elasticity that arguably is more important than all others, because it summarizes all of what needs to be known for many of the central normative questions of taxation. This is the elasticity of taxable income with respect to the tax rate." Is the ETI really that important?

6 ETI AND DEADWEIGHT LOSS The deadweight loss is given by =,where is the utility loss from taxation (in monetary units) and is collected tax revenue. The marginal DWL is given by =. We have = + where is the mechanical revenue effect and is the behavioral revenue effect. Wehave = using the envelope theorem. = ( + ) =. marginal DWL equals behavioral revenue loss, which is determined by tax base elasticities (ETI in the context of income taxation).

7 REAL RESPONSES VS. AVOIDANCE/EVASION Model makes sense for real responses, but what about avoidance/evasion? Key assumption: fiscal externality arising from the tax wedge is the only externality from taxable income responses. This requires that the private costs of avoidance/evasion equal the social costs. This is not satisfied for e.g. fines for evasion, but may be for real tax sheltering costs and moral costs. Fully including evasion/avoidance in DWL calculations may overstate efficiency effects estimate both real income elasticities and ETIs.

8 ETI LITERATURE Lindsey (1987) and Feldstein (1995) were first to estimate the ETI, using the Reagan 81-reform (Lindsey) and Reagan 86-reform (Feldstein) as natural experiments. Since then, a large literature has estimated the ETI using US data. More recently, there has been work on other countries. Excellent surveys and critical discussions of the ETI literature: Slemrod (1998), National Tax Journal. Saez (2004), Tax Policy and the Economy. Saez,Slemrod,&Giertz(2012), Journal of Economic Literature.

9 ACENTURYOFU.S.INCOMETAXATION US income tax starts in Marginal tax rates (MTRs) are very low initially, but increase sharply in the interwar period. Large exemption levels implied that less than 10% of the population paid income tax. After 1942, exemption levels were lowered and more people included in the tax net. Top MTR was extremely high (94% in ). Since then, the top MTR has been changed as follows: 91% to 70% Kennedy: RA64 70% to 50% Reagan: ERTA81 50% to 28% Reagan: TRA86 28% to 31% Bush Sr: OBRA90 31% to 39.6% Clinton: OBRA % to 35% Bush Jr: EGTRRA01

10 TOP INCOME SHARE ANALYSIS Denote by the top income share and by the top MTR at time. If a legislated change in occurs between time 0 and 1, theetican be estimated as ˆ = ln 1 ln 0 ln(1 1 ) ln(1 0 ) Identifying assumption: absent the tax change, the top income share would have remained constant. This is a dif-in-dif using the whole population as a control group: ln = %-change in top income %-change in population income.

11 60% 50% 40% 30% 20% 18% 16% 14% Top 1% Income Share TOP 1% INCOME SHARE AND MARGINAL TAX RATE IN THE US 12% 10% 8% 6% Top 1% Marginal Tax Rate 10% Top 1% Marginal Tax Rate Top 1% Income Share 0% Source: Saez, Slemrod, and Giertz (2009) 4% 2% 0%

12 60% 50% 40% 30% 20% 30% 25% Next 9% Income Share NEXT 9% INCOME SHARE AND MARGINAL TAX RATE IN THE US 20% 15% 10% Next 9% Marginal Tax Rate 10% Next 9% Marginal Tax Rate Next 9% Income Share 0% Source: Saez, Slemrod, and Giertz (2009) 5% 0%

13 INSIGHTS FROM TOP INCOME SHARE ANALYSIS 1. The top-percentile share started to increase precisely in 1981 when the top MTR started to decline. 2. A sharp jump in the top-percentile share in corresponds exactly to the sharp drop in the top MTR enacted by TRA The top-percentile share continues to increase in the 1990s despite increasesinthetopmtr. 4. The income share for the 90th-99th percentile is very smooth and displays no correlation with the MTR for this group. Circumstantial evidence that the ETI will be strongly heterogeneous across reforms/time periods and across income groups.

14 FELDSTEIN (1995): TRA86 Studies TRA86 as a natural experiment. TRA86 is the most fundamental tax reform in the US since WWII. TRA86 lowered the top MTR from 50% to 28% phased in over two years, with smaller tax cuts further down the distribution. The reform also involved substantial base-broadening by repealing exemptions and preferential tax treatment.

15 FELDSTEIN (1995): EMPIRICAL STRATEGY Uses a panel of individual tax returns. Compares years 1985 and Exploit differences in marginal tax cuts across the income distribution: Treatment group = highest-income taxpayers in 1985: 85 = 49-50% Control group 1 = high-income taxpayers in 1985: 85 = 42-45% Control group 2 = medium-income taxpayers in 1985: 85 = 22-38% A dif-in-dif is then used to estimate the ETI: ˆ = ln ln ln 1 ln 1 where is taxable income.

16 Source: Feldstein (1995) CHANGES IN TAX RATES AND REPORTED INCOMES BETWEEN 1985 AND 1988

17 DIFFERENCE-IN-DIFFERENCES: THE ELASTICITY OF TAXABLE INCOME Source: Feldstein (1995)

18 PROBLEMS WITH FELDSTEIN S APPROACH (1) If inequality increases for non-tax reasons, theetiis biased upwards as the dif-in-dif attributes all of the differential increase in top incomes to the tax reform. (2) Defining treatment and control by pre-reform income level creates a mean-reversion problem. For tax cuts at the top, this biases the ETI downwards. (3) When both treatments and controls are affected, dif-in-dif requires homogeneous elasticities. If elasticities are increasing in income, the ETI is biased upwards.

19 THE HOMOGENEOUS ELASTICITY ASSUMPTION Our estimate: ˆ = ln ln ln(1 ) ln(1 ) Suppose true elasticities are for the -group and zero for the -group. Then, given parallel trends, we have ln ln = ln(1 ) If ln(1 )= ln(1 ) where [0 1), weobtain ˆ = 1 1 In Feldstein, we have '

20 PROBLEMS WITH FELDSTEIN S APPROACH (4) A very small sample. Results driven by very few observations. (5) TRA86 changed both the tax rate and the tax base a concurrent definition of taxable income would confound behavioral and definitional effects use a constant pre- or post-definition, but this is not without bias. (6) Increase in top incomes partly driven by income shifting from corporate to personal tax bases (Gordon-Slemrod, 2000). As corporate income is also taxed (albeit at a lower rate), the ETI overstates the revenue and efficiency effects. (7) Short-term vs long-term responses and the timing of income.

21 GRUBER AND SAEZ (2002) Panel data from 1979 to 1990, including federal (ERTA81, TRA86) and state tax reforms. Relate changes in taxable income to changes in marginal tax rates in three-year intervals ( ,..., ). Basic specification without income effects: ln ( )= ln (1 )+controls + Nonlinear tax system endogenous to simulate instrument for ln (1 ) using mechanical tax changes. Control for the (possibly nonlinear) relationship between income changes and base-year income levels (inequality trends, mean reversion).

22 GRUBER AND SAEZ (2002) Find an ETI of around 0.4 and an elasticity of "broad income" of Larger elasticities at the top than at the bottom. Identifying assumption: the relationship between income changes and base-year income levels is not changing over time in a way which is correlated with the tax reforms. Problem: estimates are very sensitive to specification, especially the form of base-year income controls (Kopczuk 2005). We do not know which specification adequately controls for non-tax related income trends andmeanreversion.

23 ANATOMY OF BEHAVIORAL RESPONSE If part of the change in taxable income is driven by income shifting between corporate and personal tax bases, the ETI is not a sufficient statistic to calculate revenue and welfare effects. In a tax system with several tax bases, we have to know the elasticity for each tax base as well as cross-elasticities between bases (tax shifting) in order to evaluate revenue, welfare, and optimal taxation. Illuminating to consider the composition of the top 0.01% income share in the US over time.

24 TOP 0.01% INCOME SHARE, COMPOSITION, AND MTR IN THE US, % 90% 3.0% 80% Top 0.01% share and composition 2.5% 2.0% 1.5% 1.0% Other Interest Dividends Sole Prop. Partnership S-Corp. Wages MTR 70% 60% 50% 40% 30% 20% Marginal Tax Rate for the top 0.01% 0.5% 10% 0.0% 0% Source: Saez, Slemrod, and Giertz (2009)

25 ANATOMY OF TOP INCOMES OVER TIME 1. The top income share remains flat in the 1960s and well into the 1970s despite very substantial tax cuts. 2. About 1/3 of the surge in taxable income around TRA86 is driven by S-corporation income. This suggests shifting from C-corp to S-corp. 3. Partnership income rises dramatically after TRA86 as partnership loss tax shelters are closed. 4. Dramatic shift in the composition of top income from dividends to wage and S-corp [rentiers working rich] 5. Dramatic secular growth in top wage income since the early 1970s, with short-term spikes around 1988 and 1992 due to timing effects.

26 OVERALL CONCLUSIONS FROM U.S. STUDIES Generic problem: identifying variation comes from tax cuts to the top, which are correlated with non-tax factors driving top incomes. Tax return data does not offer direct controls for non-tax factors. ETI estimates seem to be driven mostly by two phenomena: 1. Dramatic secular growth in top wage incomes. Question: what are the deep reasons for this phenomenon? Possible answers: (i) tax policy, (ii) skill-biased technical progress, (iii) international trade, (iv) superstar markets, (v) social norms. 2. Shift from C-corporations to S-corporations. Question: how important is income shifting relative to income creation?

27 EVIDENCE FROM OTHER COUNTRIES United Kingdom: Brewer, Saez, and Shepard (2010) Thatcher tax cuts: top MTR reduced from 83% to 60% in 1979, and to 40% in Top income share analysis UK setting yields qualitatively similar results and poses the same keyproblemsasussetting.

28 EVIDENCE FROM OTHER COUNTRIES Denmark: Kleven and Schultz (2013) 1. Full-population administrative data over 25 years [link tax return information with detailed socioeconomic information] 2. A stable income distribution avoids bias from non-tax changes in inequality 3. Tax reforms that create large and compelling tax variation across different income levels and different income forms 4. Clear graphical panel evidence of taxable income responses around large 1987-reform 5. Findings are very robust to specification

29 EVOLUTION OF TOP INCOME SHARES IN DENMARK Source: Kleven and Schultz (2013)

30 MECHANICAL TAX VARIATION FOR LABOR INCOME AROUND 1987-REFORM ( DIFFERENCE) Source: Kleven and Schultz (2013)

31 GRAPHICAL EVIDENCE ON TAXABLE INCOME RESPONSES TO THE DANISH 1987-REFORM Labor Income Source: Kleven and Schultz (2013)

32 GRAPHICAL EVIDENCE ON TAXABLE INCOME RESPONSES TO THE DANISH 1987-REFORM Labor Income: Large vs Small Tax Cuts Source: Kleven and Schultz (2013)

33 GRAPHICAL EVIDENCE ON TAXABLE INCOME RESPONSES TO THE DANISH 1987-REFORM Capital Income Source: Kleven and Schultz (2013)

34 KLEVEN AND SCHULTZ (2013): KEY FINDINGS 1. Modest labor income elasticities ( ) 2. Larger capital income elasticities ( ) 3. Larger elasticities when estimated from larger reforms (frictions) 4. Modest income shifting between labor and capital income Policy implication: broad tax bases and strong enforcement (small avoidance/evasion opportunities) ensures modest behavioral responses even under very high marginal tax rates.

35 HIGH-INCOME LAFFER RATE Top marginal tax rate is and applies to income above. Denoteby the average income for the tax payers above. Tax revenue raised by the top MTR is given by = ( ). Consider a marginal change,, inthetoprate. The mechanical revenue effect is given by = ( ) The behavioral revenue effect is given by = The high-income Laffer rate is determined by = + =0.

36 HIGH-INCOME LAFFER RATE The high-income Laffer rate: = 1 1+ where (1 ) (1 ) is the ETI and 1 reflects how close top-rate taxpayers are to the bracket threshold on average. Top tail is approximately Pareto distributed does not vary with and is equal to the Pareto parameter. In the U.S., we have 1 6. The different ETI estimates we ve seen imply huge differences in Laffer rates and policy conclusions. But this overstates the importance of the ETI as it is not a structural parameter.

37 ETI AS A POLICY INSTRUMENT The ETI is not a structural parameter. It depends on avoidance and evasion, which depend on the tax and enforcement system (Slemrod and Kopczuk 2002). The ETI will be low under (i) a broad tax base that offers limited opportunity for income shifting, (ii) rigorous tax enforcement that offers limited opportunity for evasion. If the ETI is very high (Laffer rate very low), what is the best policy response? Two possibilities: (i) reduce the MTR, (ii) reduce the ETI. Optimal policy depends on the marginal costs and benefits of (i) and (ii).