Taxable Income Response to Tax Rate Changes â Empirical Evidence ...
Taxable Income Response to Tax Rate Changes – Empirical Evidence from the German Taxpayer Panel
Daniela Witczak Peter Gottfried March 2009 – preliminary draft –
Abstract The elasticity of taxable income has gained increasing attention as a fiscal policy parameter. Whereas a number of studies exist for different countries, there has hardly been any empirical evidence for Germany so far. With a new panel data set which contains detailed information of all German taxpayers, we are now able to close this research gap. By means of microsimulation and instrumental variable regression we analyse the extent of taxable income response in Germany, interpreting an income tax reform which took place in 2004 as a quasi-natural experiment. We obtain rather large overall elasticities between 0.86 and 1.06. Income effects are small or insignificant. Taxpayers with a higher base year taxable income show stronger reactions to tax rate changes than those in lower income groups. Keywords: tax reform; elasticity of taxable income JEL classification: H24, H31
Institut fuer Angewandte Wirtschaftsforschung (IAW), Ob dem Himmelreich 1, 72074 Tuebingen (Germany), Tel.: +49 (0)7071 9896-22, Fax: +49 (0)7071 9896-99, Email: [email protected] The authors would like to thank Susan Kriete-Dodds (German Federal Statistical Office) for her support. We also gratefully acknowledge financial support from the Fritz Thyssen Stiftung.
1
Introduction
It is a well known fact that income taxes create behavioural distortion, thus the social costs of raising tax revenue might be much higher than the mere figure suggests. However, empirical studies revealed that labour supply response to tax rate changes was rather inelastic. In other words, a small increase in tax rates was assumed to considerably raise revenue without causing too much deadweight loss. But adjusting labour supply is only one possible form of tax induced behavioural response, and substantial gaps between assessed and realised post-reform revenues suggested that tax avoidance activities might be more widespread than expected, implying that the deadweight loss caused by income taxation has been underestimated. In applied public economics, this insight was taken into account by using a broader definition of tax induced response which captures not only changes in labour supply but also includes other income sources like self-employment income or capital gains, different forms of compensation and varying amounts of deductible expenses. All of these reactions eventually affect the tax base, so the central policy parameter to be estimated is the elasticity of taxable income with respect to a change in the marginal tax rate. Since the tax rate is a function of taxable income itself, an exogenous tax rate change is needed in order to identify behavioural responses. We interpret a major income tax reform which took place in Germany during the years 2000 to 2005 as a quasi-natural experiment. Marginal tax rates were lowered for all levels of taxable income, accompanied by several other institutional changes. On the whole the reform was designed to perceptibly reduce tax burden with particular emphasis on low- and middle-income families and employees as well as small and medium partnerships. Our study uses a new panel data set of individual tax returns which is now accessible for researchers via the German Federal Statistical Office for the years 2001 to 2004. Using this data set, we provide first empirical evidence for the elasticity of taxable income in Germany. Computations of marginal tax rates and adjustments of tax base definitions which are needed for the instrumental variable regressions are done with a previously developed microsimulation model which uses the detailed information on taxpayer level. The following section gives an overview of previous studies in this field of research. Section 3 outlines the theoretical model our empirical analysis is based on. The main items of the tax reform are presented in section 4. Section 5 introduces the data base we use and illustrates our econometric approach. The estimation results are presented and discussed in section 6, followed by a summary and conclusions. 1
2
Previous studies
Feldstein (1995) was the first to carry out a study based on panel data1 . He used a difference-in-differences approach and obtained taxable income elasticities between 1.1 and 3.05. These fairly high estimates were, among other things, due to a widening income distribution during the corresponding period in the U.S. which Feldstein did not account for. The (meanwhile quite numerous) studies following his article enhanced Feldstein’s approach and found smaller elasticities, as can be seen from Table 1. Table 1: Previous studies (based on panel data)
Author(s)
Country
Period
Central elasticity result(s)
Feldstein (1995)
United States
1985-1988
1.1 – 3.05
Auten/Carroll (1999)
United States
1985-1989
0.6
Goolsbee (2000)
United States
1991-1995
0 – 0.4
Aarbu/Thoresen (2001)
Norway
1991-1994
-0.03 – 0.2
Sillamaa/Veall (2001)
Canada
1986-1989
0.25
United States
1979-1990
0.4
Selén (2002)
Sweden
1989-1992
0.2 – 0.4
Saez (2003)
United States
1979-1981
0.3/0.4
Germany
1988-1990
0.4 – 0.6
Kopczuk (2005)
United States
1979-1990
0.123/1.063
Hansson (2007)
Sweden
1989-1992
0.37
Giertz (2007)
United States
1979-2001
0.26 – 0.4
Auten/Carroll/Gee (2008)
United States
1999-2005
0.4
Sweden
1991-2002
0.1 – 0.3
United States
1999-2005
0.3 – 0.4
Gruber/Saez (2002)
Gottfried/Schellhorn (2004)
Holmlund/Söderström (2008) Heim (2009)
1
The first to analyse taxable income reaction to tax rate changes was Lindsey (1987), but his study was based on repeated cross section data.
2
Table 1 shows that the majority of studies analyse taxpayer behaviour in the United States. Auten/Carroll (1999) were the first to explicitly account for mean reversion effects and subsequent studies have shown that it has considerable impact on the estimation results. Gruber/Saez (2002) and Saez (2003) decompose uncompensated income elasticity into substitution and income effect. This approach was adapted by Selén (2002) and Gottfried/Schellhorn (2004) and is also used in this paper. Kopczuk (2005) develops an interesting approach, including a term that interacts the change in the net-of-tax rate (NTR) with the share of deductible income in order to account for changes in the tax base. He obtains income-weighted elasticities for married filers of 0.123 (direct effect) and 1.063 (interaction term). The hypothesis that income elasticity depends partly on deduction rules is also supported by Giertz (2007) who finds that taxable income elasticity changes from 0.4 in the 1980s to 0.2 in the 1990s, but broad income elasticity changes only from 0.12 to 0.15, respectively. There are several studies that analyse taxpayer behaviour in non-US countries. Sillamaa/Veall (2001) examine the behaviour of Canadian taxpayers and find an average elasticity of 0.25. In Europe empirical evidence exists mainly for Scandinavian countries. Aarbu/Thoresen (2001) find results for Norway that are close to zero or even negative. Selén (2002) and Hansson (2007) find elasticities in Sweden ranging from 0.2 to 0.4. Holmlund/Söderström (2008) try to capture long-run effects by adding a lagged change in the NTR and obtain long-run elasticities for men between 0.1 and 0.3. Besides they are the first to explicitly address some econometric shortcomings of previous work like autocorrelation in the error term and correlation between the error term and base year income as well as the instrumented NTR. These methodical issues are not discussed in this paper, but will be of importance in future work. For Germany so far, hardly any empirical evidence has been provided, except Gottfried/Schellhorn (2004) who analyse income tax data for a south-western part of Germany and find an average elasticity of 0.4. However, due to regional specifics this result may not be representative for the rest of the country. It seems to be a central problem in all studies to find adequate controls for exogenous income trends and other factors that influence income growth. However, as can be seen from Table 1, data availability improves continuously and longer time periods of data allow to analyse and compare the effects of more than one tax reform and to control in a more accurate way for changes in the income distribution, as already done by Gruber/Saez (2002), Giertz (2007) and Heim (2009). Although there are still methodical issues to overcome, the empirical approach to identify tax induced income response has improved over time, and even the lower results of the post-Feldstein studies still indicate that the efficiency cost of income taxation is higher than previous labour supply elasticities suggested. 3
3
Theoretical framework
The basic model for our analysis is a consumption-income model C = Y − T (T I) = (1 − τ )T I + R.
(1)
which is similar to the one Gruber/Saez (2002) used in their study. Y is gross income, T I is taxable income, T is tax liability, τ is marginal tax rate, (1 − τ ) is the so-called net-of-tax rate (NTR) and R is virtual income. Taxable income (T I) is defined as gross income minus tax-deductible consumption. Scaling the price for non-deductible goods to 1, the price for deductible goods is then equal to the NTR. Virtual income R is a theoretical construct for progressive tax systems, reflecting the difference between the tax amount that would result from applying the marginal tax rate to taxable income and the actual tax amount, τ T I − T (T I). From eq. (1) it can be seen that a tax change affects both the NTR (1 − τ ) and virtual income R, so a taxpayer’s reaction of taxable income T I(1 − τ, R) is given by a change in the NTR and a change in virtual income: dT I =
∂T I ∂T I d(1 − τ ) + dR. ∂(1 − τ ) ∂R
(2)
A change in the NTR affects the relative prices of tax-privileged and nonprivileged consumption, and a change in virtual income affects the taxpayer’s budget, thus the tax change causes substitution as well as income effects. Using the Slutsky decomposition the term ∂T I/∂(1 − τ ) can be written as ∂T I c ∂T I ∂T I = + TI ∂(1 − τ ) ∂(1 − τ ) ∂R
(3)
with ∂T I c /∂(1 − τ ) capturing the substitution effect resulting from a change in the marginal tax rate and T I ∂T I/∂R capturing the income effect of a change in virtual income. Substituting eq. (3) into (2) and rearranging the terms then yields dT I d(1 − τ ) −dT = ζc +η TI 1−τ (Y − T I) + (τ T I − T )
4
(4)
where ζ c = ∂T I/∂(1 − τ ) (1 − τ )/T I is the substitution (or compensated price) elasticity and η = ∂T I/∂R R/T I is the income elasticity. Under the assumption that tax-deductible goods are normal, the compensated elasticity ζ c should be positive since a price increase caused by a cut in marginal tax rates should theoretically lower tax-privileged consumption, leading to a higher taxable income. The sign of the income elasticity η and the overall effect are theoretically undetermined. Since the marginal tax rate is a function of taxable income and thus endogenously determined by taxpayers, an external tax rate change is needed in order to analyse the effects on individual behaviour and income adjustment. The German income tax reform 2000 included such an external change in tax rates, and so we interpret it as a quasi-natural experiment.
4
The tax reform
The purpose of the major tax reform which took place in Germany during the years 2000-2005 was to stimulate economic growth and employment. The reform was mainly characterised by a change in the personal income tax schedule, although corporate taxation was also part of the reform programme. Various measures were implemented during the period 2001-2004, for which income tax panel data are now available (see section 5.1). The basic tax allowance was slightly increased during this period from 7206 e to 7664 e.2 The tax rate at the bottom of the tax schedule was reduced from 19.9 % to 16 %. At the top of the tax schedule the marginal tax rate was cut from 48.5 % to 45 % for taxable incomes exceeding 52 152 e, compared to 54 999 e in 2001. Tax rates in the medium range of the schedule were lowered as well. In contrast to most other countries who use a bracket system with constant marginal tax rates within a bracket, Germany uses a formula to compute tax liability which is of second order in income. As a consequence, marginal tax rates increase linearly in income. Both tax schedules, 2001 and 2004, are depicted in Figure 1. Taxpayers with a high taxable income and those with a taxable income slightly exceeding the basic tax allowance experienced the largest marginal tax rate cuts.
2
The German income tax schedule 2001 was still stated in DM amounts. For a comparison of the pre- and post-reform schedules we converted the values to e using the exchange rate 1 e = 1.95583 DM.
5
Figure 1: Marginal tax rates on taxable income 2001 and 2004
0,5
Marginal Tax Rate
0,4
0,3
Marginal Tax Rates 2001 Marginal Tax Rates 2004
0,2
0,1
0 20000
40000
60000
Taxable Income
Beside the change in the tax schedule several other reform measures were taken which altered the definition of the tax base. In order to support families, child allowances were raised from 2556 e to 2904 e per child. Loss offsetting restrictions (for single taxpayers as well as between spouses) that were in place until 2003 were cancelled in 2004, and another rather drastic change was implemented in 2002 when only 50 % of certain capital gains entered the tax base. On the other hand, allowable expenses for non-itemizing employees were cut from 1044 e to 920 e, allowances for single parents were cut from 2871 e to 1308 e and capital gains exemption was reduced from 1550 e to 1370 e. On the whole the vast majority of taxpayers experienced a perceptible tax relief.
6
5
Empirical strategy
5.1
Data
The data source for our study is the German Taxpayer Panel, a large panel data set of individual income tax returns which includes all German income taxpayers for the years 2001-2004. The data set faces the same advantages and disadvantages most tax return data face: On the one hand it contains detailed information on various types of income, deductions and allowances, for employees as well as for the self-employed and entrepreneurs, but also for persons with capital income or income from rent and lease. On the other hand the data set provides very little demographic information on the taxpayers. The approach we use is a classical panel data technique. We only analyse taxpayers who are subject to taxation in each year in order to exclude cases with discontinuous working biographies whose income level may vary for reasons not related to changes in the tax code.3 The balanced data allow us to observe about 17 000 000 tax payers over time. Since the aim of our analysis is to identify income changes which are directly attributable to reform incentives, we further eliminated cases from the data whose socio-economic characteristics changed significantly during the regarded period. More precisely taxpayers were excluded • whose marital status changed • who were under age 25 in 2001 • whose taxable income did not exceed the basic tax allowance in either 2001 or 2004 • who had their first child, • who received for the first time an allowance for single parents or an allowance for disabled persons and • who retired during the regarded period. A change in marital status implies a substantial change in the household’s financial situation that would influence estimation results. Individuals younger than 25 years may experience income increases at the beginning of their working life which are no behavioural response to the reform either. For lack of more 3 However, by doing so we neglect reactions at the margin like taking up an employment. Analysing repeated cross section data would be the preferable approach to include these reactions.
7
detailed information, we chose a threshold of 25 which is, of course, arbitrary. We further excluded taxpayers who had their first child during the regarded period since this circumstance strongly influences (generally the second earner’s) labour supply decisions. Finally taxpayers who retire are confronted with a considerable income cut which is not attributable to the reform either. After these adjustments 11 674 084 cases remain in the data set. Table 2 displays average taxable incomes, average pre-reform marginal tax rates and average percentage changes in marginal tax rates for taxpayers grouped by base year taxable income deciles. Table 2: Average taxable incomes, average pre-reform marginal tax rates and average percentage changes in marginal tax rates, in deciles
Decile
Average taxable income
Average prereform marginal tax rate
Average percentage change in statutory marginal tax rates
1
11 036
0,153
-18,32
2
16 724
0,244
-15,82
3
20 957
0,268
-7,66
4
24 739
0,285
-4,51
5
28 613
0,296
-3,98
6
33 437
0,306
-3,95
7
39 214
0,318
-3,95
8
46 756
0,333
-4,04
9
56 267
0,354
-4,76
10
126 160
0,431
-5,83
Note: 11 674 084 observations. The change in marginal tax rates is calculated by applying 2001 and 2004 tax law to 2001 income inflated to the 2004 level.
As already shown in section 4, taxpayers with lower and upper taxable incomes experienced the highest tax rate cuts, the changes in the middle of the income distribution are more moderate and quite homogeneous. The large percentage change in the first decile also results from the increase in the basic tax allowance what implies that some taxpayers are no longer subject to taxation after the reform. 8
5.2
Econometric approach
The estimation equation is derived from the model introduced in section 3, based on the Slutsky decomposition to estimate substitution and income effect separately. To approximate the growth rate of taxable income with respect to a percentage change in the NTR we use a logarithmised specification, as it is widely done in this field of research: ln(
(1 − τ1 ) T0 − T1 T I1 ) = ζ c ln +η + u. T I0 (1 − τ0 ) (Y0 − T I0 ) + (τ0 T I0 − T0 )
(5)
As already mentioned the marginal tax rate is determined by the level of taxable income.4 Regressing the change in the tax base, T I1 −T I0 , on the observed change in the NTR, (1 − τ1 ) − (1 − τ0 ), would thus cause simultaneity problems since the observed NTR in 2004 already contains tax induced adjustments. To avoid this problem we use a microsimulation model and create a counterfactual scenario: We calculate synthetic marginal tax rates τ˜0 and τ˜1 (and synthetic tax liabilities T˜0 and T˜1 , respectively), applying the pre- and the post-reform tax code to 2001 income which is inflated to the 2004 level for every taxpayer in order to exclude nominal income growth effects: τ˜0 = τ (T˜I0 , z0 )
(6a)
τ˜1 = τ (T˜I0 , z1 )
(6b)
T˜I0 is taxable income based on 2001 income inflated to 2004 level, and z0 and z1 denote 2001 and 2004 tax law respectively. With (1 − τ˜0 ) we obtain the NTR the taxpayer would face in 2004 if tax law and his real income did not change over time. (1 − τ˜1 ) is the NTR on inflated income under post-reform tax law, but before any behavioural response. The difference between these two NTRs serves as an instrument variable for the truly observed difference between the NTRs. The same procedure applies to the tax liability: The taxpayer would have to pay T˜0 in 2004 did his real income stay constant over time and did tax law not change. T˜1 would be tax liability under 2004 tax law. The difference between T˜1 and T˜0 serves analogously to the NTRs as an instrument variable for the observed difference between tax liabilities in the income effect term. 4
The same problem applies to the tax amount T1 which is needed to calculate the income effect term. Other factors which influence marginal tax rate and tax amount like joint filing do not put aside the endogeneity problem.
9
For the change in taxable income, we use our microsimulation model to adjust inflated base year income for other changes in the tax code that were part of the reform, e. g. concerning expenses and allowances. We thus eliminate income changes that are not directly attributable to the reform but due to differences in deduction possibilities over time. As before, inflating incomes excludes changes that are due to nominal growth. The difference between the observed taxable income in 2004 and the created counterfactual income is regressed on the NTR, the income effect variable and other control variables in a two-stage least squares setting. Including other control variables yields the following estimation equation:
ln(
T0 − T1 (1 − τ1 ) T I1 +η + ρ ln(T I0 ) + γ’X + u. (7) ) = ζ c ln (1 − τ0 ) (Y0 − T I0 ) + (τ0 T I0 − T0 ) T˜I0
The logarithm of base year income is used to control for reversion-to-the-mean effects. Taxpayers with very high or low transitory incomes in the base year are likely to experience large income changes over time. Those changes are often rather deviations from the individual long-run income equilibrium than true behavioural responses and are therefore controlled for.5 X is a vector of covariates that includes marital status, number of children, age (and age squared) and two dummy variables. The first dummy indicates whether the taxpayer declares mainly wage income or income from self-employment and is used to control for the fact that taxpayers with income from self-employment have more possibilities to rearrange different components of their income than employees. The second dummy indicates whether the main type of income (dependent or non-dependent income) changed between 2001 and 2004 and accounts for significant changes that go along with starting a business or switching from self- to regular employment. The dummy for marital status addresses the fact that joint filers are treated differently from single filers. The number of children controls for the time constraint parents face. Age and age squared serve as proxy variables for life-cycle effects.
5
Since we compare only two years of data which comprise one tax rate change, the inclusion of a 10-piece spline in base year income as is done by Giertz (2007) or Gruber/Saez (2002) in order to control for mean reversion and exogenous income trends more accurately would probably destroy identification.
10
6
Results
The basic results are presented in Table 3. All estimates are based on 2SLSRegressions. The F statistics for the coefficients of the NTR and the income effect instruments in the first stage are very high (not shown here). The elasticity of taxable income with respect to the NTR for all taxpayers is 1.06. This is remarkably high compared to the results in recent studies which mainly range from 0.2 to 0.6 (see Table 1). Since these are first results from ongoing work, they have to be regarded as preliminary and are somewhat likely to change when identification methodology is further developed. We discuss the magnitude of this coefficient in more detail below. The income effect is not only insignificant but also practically zero which corresponds to the results of Gruber/ Saez (2002) who do not find significant income effects either, whereas in the study carried out by Selén (2002) income effects are positive and significant. The parameter for the log in base year income has the expected negative sign and controls for income changes that are rather convergences towards the individual long-run equilibrium than tax induced responses. The other demographic control variables are all significant (except age squared), but their values are rather small. Joint filers experience a slightly higher income growth than single filers. Interpreting age as a life-cycle proxy, the negative sign implies that income is more stable when taxpayers get older. One could expect the number of (tax relevant) children to have a negative impact, but it might reflect life-cycle effects as well, being to a certain degree correlated with taxpayers’ age. In this respect, the positive sign of the child variable seems plausible. The negative sign of the switch-dummy is conform to the theoretical argumentation that starting a business or else giving up business in favour of a safe job or fixed working hours often goes along with a loss of income at the beginning. Surprisingly, the dummy for self-employment has a negative sign as well. One reason for this may be that mean reversion effects are not adequately controlled for and are partly captured by this variable. Another reason may be the definition of the dummy variable itself: We assigned taxpayers to be self-employed when they declared income from a closely held business or professional earnings, but we did not include income from participations and investments or capital gains that are attributable to business income according to income tax law. Since this is the more flexible part of these income types, an alternative definition of this dummy might lead to different results. When looking at the results per income group, the impact of the variables differs. In the lowest income group, the NTR elasticity becomes negative and the income effect increases, whereas in the middle and upper income groups the elasticity almost doubles, the income effect being negligible. Mean reversion effects are strongest in the lower and upper group. 11
Table 3: Basic estimation results
All taxpayers Variable
By income group < 30 000 e
30 000 e – 60 000 e
> 60 000 e
NTR elasticity (ζ c )
1.0601∗∗∗ (0.0064)
-0.0971∗∗∗ (0.0083)
1.7196∗∗∗ (0.0194)
1.9101∗∗∗ (0.0205)
Income elasticity (η)
0.0000 (0.0001)
-0.0123∗∗∗ (0.0006)
-0.0001∗∗∗ (0.0000)
-0.0001∗∗ (0.0000)
ln(base year income)
-0.1880∗∗∗ (0.0002)
-0.3663∗∗∗ (0.0004)
-0.1256∗∗∗ (0.0011)
-0.2237∗∗∗ (0.0012)
Joint filing
0.0395∗∗∗ (0.0003)
0.0498∗∗∗ (0.0003)
0.0205∗∗∗ (0.0006)
0.1231∗∗∗ (0.0017)
Number of children
0.0180∗∗∗ (0.0001)
0.0153∗∗∗ (0.0001)
0.0234∗∗∗ (0.0002)
0.0195∗∗∗ (0.0005)
Age
-0.0056∗∗∗
-0.0065∗∗∗
-0.0053∗∗∗
-0.0059∗∗∗
(0.0000)
(0.0000)
(0.0000)
(0.0001)
0.0000 (0.0000)
0.0000 (0.0000)
0.0000 (0.0000)
0.0000 (0.0000)
Switch
-0.0889∗∗∗ (0.0007)
0.0479∗∗∗ (0.0008)
-0.215∗∗∗ (0.0014)
-0.3205∗∗∗ (0.0023)
Self-employed
-0.0833∗∗∗ (0.0005)
-0.0501∗∗∗ (0.0005)
-0.1587∗∗∗ (0.0008)
-0.1072∗∗∗ (0.0015)
Constant
2.1788∗∗∗ (0.0022)
3.9798∗∗∗ (0.0045)
1.4957∗∗∗ (0.0119)
2.5140∗∗∗ (0.0133)
11 674
5 600
4 640
1 410
Age squared
No. of obs. (in 1000)
Note: Estimates from 2SLS regressions. Standard errors in parentheses. denote significance on the 1 %, 5 % and 10 % level respectively.
∗∗∗ /∗∗ /∗
12
An elasticity that exceeds one is much higher than what has been found in most studies. Aside from other exogenous income trends between 2001 and 2004 we may not have adequately controlled for in our regressions, descriptive statistics show that especially income from rent and lease underwent a change that surely is not entirely attributable to tax incentives. When we look at Table 4, we see that in 2003 the total balance switched from a negative to a positive value, with an overall change from around -2,5 bn e in 2001 to over 5 bn e in 2004.6 The relatively constant number of taxpayers that report income from rent and lease (increasing only by 2,5 %) suggest that this development is not due to additional taxpayers declaring high positive incomes from rent and lease but that it is in fact individual income that increases over time. This is supported by mean income which changes from -630 e in 2001 to 1283 e in 2004. Table 4: Income from rent and lease 2001-2004
Mean
Median
Sum
N
2001
-630
-198
-2 435 611 965
3.868.881
2002
-117
-69
-458 820 609
3.906.534
2003
373
53
1 473 236 964
3.951.914
2004
1283
258
5 088 235 807
3.964.539
An attempt to explain this development may be that these are cumulative effects of changing individual cost-revenue ratios. Deductible acquisition or maintenance costs which initially exceeded revenue from rent and lease decline over time, e. g. when using declining-balance depreciation or as a consequence of rent increases or betterments in real property. Amortised mortgages also necessarily result in positive income in the course of time. A coincidence of these factors may in total lead to an overbalance of positive income. In order to see how much this trend influences the estimation results, we exclude income from rent and lease from our calculations, although we risk losing tax induced response to some extent since this income type often serves the purpose to reduce the tax base. The estimation results are presented in Table 5. 6
These are the values calculated only for taxpayers who appear continuously in the panel. The cross section totals differ in magnitude, but show the same tendency (see Statistisches Bundesamt (2008)).
13
Table 5: Estimation results without income from rent and lease
All taxpayers Variable
By income group < 30 000 e
30 000 e – 60 000 e
> 60 000 e
NTR elasticity (ζ c )
0.863∗∗∗ (0.0061)
0.8389∗∗∗ (0.0061)
0.2459∗∗∗ (0.0089)
0.5896∗∗∗ (0.0155)
1.2967∗∗∗ (0.0188)
Income elasticity (η)
-0.0002∗∗∗ (0.0000)
-0.0001∗∗∗ (0.0000)
-0.0059∗∗∗ (0.0002)
-0.0008∗∗∗ (0.0001)
-0.0001∗∗∗ (0.0000)
ln(base year income)
-0.174∗∗∗ (0.0002)
-0.1738∗∗∗ (0.0001)
-0.2932∗∗∗ (0.0004)
-0.0863∗∗∗ (0.0009)
-0.2033∗∗∗ (0.0011)
Joint filing
0.0356∗∗∗ (0.0003)
0.0352∗∗∗ (0.0003)
0.0446∗∗∗ (0.0003)
-0.0071∗∗∗ (0.0005)
0.0973∗∗∗ (0.0015)
Number of children
0.0184∗∗∗ (0.0001)
0.0184∗∗∗ (0.0001)
0.0154∗∗∗ (0.0001)
0.0245∗∗∗ (0.0001)
0.0224∗∗∗ (0.0004)
Age
-0.0066∗∗∗
-0.0065∗∗∗
-0.0078∗∗∗
-0.0056∗∗∗
-0.0071∗∗∗
(0.0000)
(0.0000)
(0.0000)
(0.0000)
(0.0000)
0.0000 (0.0000)
0.0000 (0.0000)
0.0000 (0.0000)
0.0000 (0.0000)
0.0000 (0.0000)
Switch
-0.1104∗∗∗ (0.0008)
–
0.0249∗∗∗ (0.0010)
-0.2342∗∗∗ (0.0012)
-0.3331∗∗∗ (0.0022)
Self-employed
-0.0984∗∗∗ (0.0005)
-0.1103∗∗∗ (0.0004)
-0.0686∗∗∗ (0.0006)
-0.1742∗∗∗ (0.0007)
-0.1129∗∗∗ (0.0013)
Constant
2.0706∗∗∗ (0.0021)
2.0668∗∗∗ (0.0021)
3.2991∗∗∗ (0.0043)
1.1285∗∗∗ (0.0101)
2.3861∗∗∗ (0.0123)
5 510
4 620
1 450
Age squared
No. of obs. (in 1000)
11 674
Note: Estimates from 2SLS regressions. Standard errors in parentheses. ∗∗∗ /∗∗ /∗ denote significance on the 1 %, 5 % and 10 % level respectively.
14
The results are now quite different. The elasticity for all taxpayers falls from 1.06 to 0.86 (and even slightly more when leaving out the switch-dummy). But more remarkable are the changes in estimates when distinguishing between different income groups: In contrast to the situation when income from rent and lease was included, the elasticity for taxpayers with base year taxable income less than 30 000 e is now positive. In the middle income group, elasticity falls by more than one percentage point from 1.72 to 0.59, and for taxpayers with taxable income exceeding 60 000 e it falls from 1.91 to 1.3. Income growth that is driven by the striking increase in income from rent and lease seems to be concentrated in the middle and upper part of the (taxable) income distribution, whereas in the lower income group losses predominate. The influence of the income effect is still negligible. The coefficients for the demographic control variables do not change very much, so the distribution of income from rent and lease does not seem to be highly correlated with those socio-economic characteristics. An elasticity of 0.86 is still relatively large. A possible reason for this may be the fact that our data subset is rather restrictive. Since we exclude taxpayers who had their first child or who retired during the period of the study, we systematically exclude taxpayers who are likely to experience income cuts – these cases are included in most other studies. Another interesting point is made by Slemrod/ Kopczuk (2002) and Kopczuk (2005) who argue that taxable income elasticity is not a fixed parameter but depends on the deduction possibilities defined by tax law and thus can to some extent be controlled by policy makers. Since German income tax law offers various possibilities to narrow the tax base over the whole income distribution, it seems possible that behavioural response may be more pronounced in Germany than in other countries.
7
Conclusions
This paper presents first results from ongoing work to identify the effects of the tax rate changes in 2004 on taxable income in Germany. We find an overall taxable income elasticity of 1.06 when including all types of income and an elasticity of 0.86 when leaving out income from rent and lease. The estimates differ greatly between income groups, with high income taxpayers showing stronger responses. The magnitude of the results leads to the assumption that the estimates are currently still driven by factors not yet fully controlled for and therefore they have to be handled with care. A more thorough analysis of the data (e. g. concerning income composition and distribution) is needed to better identify exogenous income trends that are not (at least entirely) tax related. Nevertheless we obtain a first insight in the dimension of taxpayer behaviour in Germany which is nontrivial and has to be taken into account when considering further tax reforms. 15
References Aarbu, K. O./Thoresen, T. O.: Income Responses to Tax Changes - Evidence from the Norwegian Tax Reform. In: National Tax Journal LIV (2001), Nr. 2, 319–335 Auten, C./Carroll, R./Gee, G.: The 2001 and 2003 Tax Rate Reductions: An Overview and Estimate of the Taxable Income Response. In: National Tax Journal 61 (2008), Nr. 3, 345–364 Auten, G./Carroll, R.: The Effect of Income Taxes on Household Income. In: The Review of Economics and Statistics 81 (1999), Nr. 4, 681–693 Feldstein, M.: The Effect of Marginal Tax Rates on Taxable Income: A Panel Study of the 1986 Tax Reform Act. In: Journal of Political Economy 103 (1995), Nr. 3, 551–572 Giertz, S. H.: The Elasticity of Taxable Income over the 1980s and 1990s. In: National Tax Journal 60 (2007), Nr. 4, 743–768 Goolsbee, A.: What Happens When You Tax the Rich? Evidence from Executive Compensation. In: Journal of Political Economy 108 (2000), Nr. 2, 352–378 Gottfried, P./Schellhorn, H.: Empirical Evidence on the Effects of Marginal Tax Rates on Income - the German Case. IAW Discussion Paper, Februar 2004 Gruber, J./Saez, E.: The elasticity of taxable income: evidence and implications. In: Journal of Public Economics 84 (2002), 1–32 Hansson, A.: Taxpayers’ responsiveness to tax rate changes and implications for the cost of taxation in Sweden. In: International Tax and Public Finance 14 (2007), Nr. 5, 563–582 Heim, B. T.: The Effect of Recent Tax Changes on Taxable Income: Evidence from a New Panel of Tax Returns. In: Journal of Policy Analysis and Management 28 (2009), Nr. 1, 147–163 Holmlund, B./Söderström, M.: Estimating dynamic income responses to tax reforms: Swedish evidence. Working Paper 28, 2008, Institute for Labour Market Policy Evaluation (IFAU) Kopczuk, W.: Tax bases, tax rates and the elasticity of reported income. In: Journal of Public Economics 89 (2005), 2093–2119 16
Lindsey, L.: Individual taxpayer response to tax cuts: 1982-1984, with implications for the revenue maximizing tax rate. In: Journal of Public Economics 33 (1987), 173–206 Saez, E.: The effect of marginal tax rates on income: a panel study of ’bracket creep’. In: Journal of Public Economics 87 (2003), 1231–1258 Selén, J.: Taxable Income Responses to Tax Changes - Panel Analysis of the 1990/91 Swedish Reform. FIEF Working Paper No. 177, August 2002 Sillamaa, M.-A./Veall, M. R.: The Effect of Marginal Tax Rates on Taxable Income: a Panel Study of the 1988 Tax Flattening in Canada. In: Journal of Public Economics 80 (2001), 341–356 Slemrod, J./Kopczuk, W.: The optimal elasticity of taxable income. In: Journal of Public Economics 84 (2002), 91–112 Statistisches Bundesamt: Finanzen und Steuern - Jährliche Einkommensteuerstatistik 2004. 2008
17
Daniela Witczak Peter Gottfried March 2009 – preliminary draft –
Abstract The elasticity of taxable income has gained increasing attention as a fiscal policy parameter. Whereas a number of studies exist for different countries, there has hardly been any empirical evidence for Germany so far. With a new panel data set which contains detailed information of all German taxpayers, we are now able to close this research gap. By means of microsimulation and instrumental variable regression we analyse the extent of taxable income response in Germany, interpreting an income tax reform which took place in 2004 as a quasi-natural experiment. We obtain rather large overall elasticities between 0.86 and 1.06. Income effects are small or insignificant. Taxpayers with a higher base year taxable income show stronger reactions to tax rate changes than those in lower income groups. Keywords: tax reform; elasticity of taxable income JEL classification: H24, H31
Institut fuer Angewandte Wirtschaftsforschung (IAW), Ob dem Himmelreich 1, 72074 Tuebingen (Germany), Tel.: +49 (0)7071 9896-22, Fax: +49 (0)7071 9896-99, Email: [email protected] The authors would like to thank Susan Kriete-Dodds (German Federal Statistical Office) for her support. We also gratefully acknowledge financial support from the Fritz Thyssen Stiftung.
1
Introduction
It is a well known fact that income taxes create behavioural distortion, thus the social costs of raising tax revenue might be much higher than the mere figure suggests. However, empirical studies revealed that labour supply response to tax rate changes was rather inelastic. In other words, a small increase in tax rates was assumed to considerably raise revenue without causing too much deadweight loss. But adjusting labour supply is only one possible form of tax induced behavioural response, and substantial gaps between assessed and realised post-reform revenues suggested that tax avoidance activities might be more widespread than expected, implying that the deadweight loss caused by income taxation has been underestimated. In applied public economics, this insight was taken into account by using a broader definition of tax induced response which captures not only changes in labour supply but also includes other income sources like self-employment income or capital gains, different forms of compensation and varying amounts of deductible expenses. All of these reactions eventually affect the tax base, so the central policy parameter to be estimated is the elasticity of taxable income with respect to a change in the marginal tax rate. Since the tax rate is a function of taxable income itself, an exogenous tax rate change is needed in order to identify behavioural responses. We interpret a major income tax reform which took place in Germany during the years 2000 to 2005 as a quasi-natural experiment. Marginal tax rates were lowered for all levels of taxable income, accompanied by several other institutional changes. On the whole the reform was designed to perceptibly reduce tax burden with particular emphasis on low- and middle-income families and employees as well as small and medium partnerships. Our study uses a new panel data set of individual tax returns which is now accessible for researchers via the German Federal Statistical Office for the years 2001 to 2004. Using this data set, we provide first empirical evidence for the elasticity of taxable income in Germany. Computations of marginal tax rates and adjustments of tax base definitions which are needed for the instrumental variable regressions are done with a previously developed microsimulation model which uses the detailed information on taxpayer level. The following section gives an overview of previous studies in this field of research. Section 3 outlines the theoretical model our empirical analysis is based on. The main items of the tax reform are presented in section 4. Section 5 introduces the data base we use and illustrates our econometric approach. The estimation results are presented and discussed in section 6, followed by a summary and conclusions. 1
2
Previous studies
Feldstein (1995) was the first to carry out a study based on panel data1 . He used a difference-in-differences approach and obtained taxable income elasticities between 1.1 and 3.05. These fairly high estimates were, among other things, due to a widening income distribution during the corresponding period in the U.S. which Feldstein did not account for. The (meanwhile quite numerous) studies following his article enhanced Feldstein’s approach and found smaller elasticities, as can be seen from Table 1. Table 1: Previous studies (based on panel data)
Author(s)
Country
Period
Central elasticity result(s)
Feldstein (1995)
United States
1985-1988
1.1 – 3.05
Auten/Carroll (1999)
United States
1985-1989
0.6
Goolsbee (2000)
United States
1991-1995
0 – 0.4
Aarbu/Thoresen (2001)
Norway
1991-1994
-0.03 – 0.2
Sillamaa/Veall (2001)
Canada
1986-1989
0.25
United States
1979-1990
0.4
Selén (2002)
Sweden
1989-1992
0.2 – 0.4
Saez (2003)
United States
1979-1981
0.3/0.4
Germany
1988-1990
0.4 – 0.6
Kopczuk (2005)
United States
1979-1990
0.123/1.063
Hansson (2007)
Sweden
1989-1992
0.37
Giertz (2007)
United States
1979-2001
0.26 – 0.4
Auten/Carroll/Gee (2008)
United States
1999-2005
0.4
Sweden
1991-2002
0.1 – 0.3
United States
1999-2005
0.3 – 0.4
Gruber/Saez (2002)
Gottfried/Schellhorn (2004)
Holmlund/Söderström (2008) Heim (2009)
1
The first to analyse taxable income reaction to tax rate changes was Lindsey (1987), but his study was based on repeated cross section data.
2
Table 1 shows that the majority of studies analyse taxpayer behaviour in the United States. Auten/Carroll (1999) were the first to explicitly account for mean reversion effects and subsequent studies have shown that it has considerable impact on the estimation results. Gruber/Saez (2002) and Saez (2003) decompose uncompensated income elasticity into substitution and income effect. This approach was adapted by Selén (2002) and Gottfried/Schellhorn (2004) and is also used in this paper. Kopczuk (2005) develops an interesting approach, including a term that interacts the change in the net-of-tax rate (NTR) with the share of deductible income in order to account for changes in the tax base. He obtains income-weighted elasticities for married filers of 0.123 (direct effect) and 1.063 (interaction term). The hypothesis that income elasticity depends partly on deduction rules is also supported by Giertz (2007) who finds that taxable income elasticity changes from 0.4 in the 1980s to 0.2 in the 1990s, but broad income elasticity changes only from 0.12 to 0.15, respectively. There are several studies that analyse taxpayer behaviour in non-US countries. Sillamaa/Veall (2001) examine the behaviour of Canadian taxpayers and find an average elasticity of 0.25. In Europe empirical evidence exists mainly for Scandinavian countries. Aarbu/Thoresen (2001) find results for Norway that are close to zero or even negative. Selén (2002) and Hansson (2007) find elasticities in Sweden ranging from 0.2 to 0.4. Holmlund/Söderström (2008) try to capture long-run effects by adding a lagged change in the NTR and obtain long-run elasticities for men between 0.1 and 0.3. Besides they are the first to explicitly address some econometric shortcomings of previous work like autocorrelation in the error term and correlation between the error term and base year income as well as the instrumented NTR. These methodical issues are not discussed in this paper, but will be of importance in future work. For Germany so far, hardly any empirical evidence has been provided, except Gottfried/Schellhorn (2004) who analyse income tax data for a south-western part of Germany and find an average elasticity of 0.4. However, due to regional specifics this result may not be representative for the rest of the country. It seems to be a central problem in all studies to find adequate controls for exogenous income trends and other factors that influence income growth. However, as can be seen from Table 1, data availability improves continuously and longer time periods of data allow to analyse and compare the effects of more than one tax reform and to control in a more accurate way for changes in the income distribution, as already done by Gruber/Saez (2002), Giertz (2007) and Heim (2009). Although there are still methodical issues to overcome, the empirical approach to identify tax induced income response has improved over time, and even the lower results of the post-Feldstein studies still indicate that the efficiency cost of income taxation is higher than previous labour supply elasticities suggested. 3
3
Theoretical framework
The basic model for our analysis is a consumption-income model C = Y − T (T I) = (1 − τ )T I + R.
(1)
which is similar to the one Gruber/Saez (2002) used in their study. Y is gross income, T I is taxable income, T is tax liability, τ is marginal tax rate, (1 − τ ) is the so-called net-of-tax rate (NTR) and R is virtual income. Taxable income (T I) is defined as gross income minus tax-deductible consumption. Scaling the price for non-deductible goods to 1, the price for deductible goods is then equal to the NTR. Virtual income R is a theoretical construct for progressive tax systems, reflecting the difference between the tax amount that would result from applying the marginal tax rate to taxable income and the actual tax amount, τ T I − T (T I). From eq. (1) it can be seen that a tax change affects both the NTR (1 − τ ) and virtual income R, so a taxpayer’s reaction of taxable income T I(1 − τ, R) is given by a change in the NTR and a change in virtual income: dT I =
∂T I ∂T I d(1 − τ ) + dR. ∂(1 − τ ) ∂R
(2)
A change in the NTR affects the relative prices of tax-privileged and nonprivileged consumption, and a change in virtual income affects the taxpayer’s budget, thus the tax change causes substitution as well as income effects. Using the Slutsky decomposition the term ∂T I/∂(1 − τ ) can be written as ∂T I c ∂T I ∂T I = + TI ∂(1 − τ ) ∂(1 − τ ) ∂R
(3)
with ∂T I c /∂(1 − τ ) capturing the substitution effect resulting from a change in the marginal tax rate and T I ∂T I/∂R capturing the income effect of a change in virtual income. Substituting eq. (3) into (2) and rearranging the terms then yields dT I d(1 − τ ) −dT = ζc +η TI 1−τ (Y − T I) + (τ T I − T )
4
(4)
where ζ c = ∂T I/∂(1 − τ ) (1 − τ )/T I is the substitution (or compensated price) elasticity and η = ∂T I/∂R R/T I is the income elasticity. Under the assumption that tax-deductible goods are normal, the compensated elasticity ζ c should be positive since a price increase caused by a cut in marginal tax rates should theoretically lower tax-privileged consumption, leading to a higher taxable income. The sign of the income elasticity η and the overall effect are theoretically undetermined. Since the marginal tax rate is a function of taxable income and thus endogenously determined by taxpayers, an external tax rate change is needed in order to analyse the effects on individual behaviour and income adjustment. The German income tax reform 2000 included such an external change in tax rates, and so we interpret it as a quasi-natural experiment.
4
The tax reform
The purpose of the major tax reform which took place in Germany during the years 2000-2005 was to stimulate economic growth and employment. The reform was mainly characterised by a change in the personal income tax schedule, although corporate taxation was also part of the reform programme. Various measures were implemented during the period 2001-2004, for which income tax panel data are now available (see section 5.1). The basic tax allowance was slightly increased during this period from 7206 e to 7664 e.2 The tax rate at the bottom of the tax schedule was reduced from 19.9 % to 16 %. At the top of the tax schedule the marginal tax rate was cut from 48.5 % to 45 % for taxable incomes exceeding 52 152 e, compared to 54 999 e in 2001. Tax rates in the medium range of the schedule were lowered as well. In contrast to most other countries who use a bracket system with constant marginal tax rates within a bracket, Germany uses a formula to compute tax liability which is of second order in income. As a consequence, marginal tax rates increase linearly in income. Both tax schedules, 2001 and 2004, are depicted in Figure 1. Taxpayers with a high taxable income and those with a taxable income slightly exceeding the basic tax allowance experienced the largest marginal tax rate cuts.
2
The German income tax schedule 2001 was still stated in DM amounts. For a comparison of the pre- and post-reform schedules we converted the values to e using the exchange rate 1 e = 1.95583 DM.
5
Figure 1: Marginal tax rates on taxable income 2001 and 2004
0,5
Marginal Tax Rate
0,4
0,3
Marginal Tax Rates 2001 Marginal Tax Rates 2004
0,2
0,1
0 20000
40000
60000
Taxable Income
Beside the change in the tax schedule several other reform measures were taken which altered the definition of the tax base. In order to support families, child allowances were raised from 2556 e to 2904 e per child. Loss offsetting restrictions (for single taxpayers as well as between spouses) that were in place until 2003 were cancelled in 2004, and another rather drastic change was implemented in 2002 when only 50 % of certain capital gains entered the tax base. On the other hand, allowable expenses for non-itemizing employees were cut from 1044 e to 920 e, allowances for single parents were cut from 2871 e to 1308 e and capital gains exemption was reduced from 1550 e to 1370 e. On the whole the vast majority of taxpayers experienced a perceptible tax relief.
6
5
Empirical strategy
5.1
Data
The data source for our study is the German Taxpayer Panel, a large panel data set of individual income tax returns which includes all German income taxpayers for the years 2001-2004. The data set faces the same advantages and disadvantages most tax return data face: On the one hand it contains detailed information on various types of income, deductions and allowances, for employees as well as for the self-employed and entrepreneurs, but also for persons with capital income or income from rent and lease. On the other hand the data set provides very little demographic information on the taxpayers. The approach we use is a classical panel data technique. We only analyse taxpayers who are subject to taxation in each year in order to exclude cases with discontinuous working biographies whose income level may vary for reasons not related to changes in the tax code.3 The balanced data allow us to observe about 17 000 000 tax payers over time. Since the aim of our analysis is to identify income changes which are directly attributable to reform incentives, we further eliminated cases from the data whose socio-economic characteristics changed significantly during the regarded period. More precisely taxpayers were excluded • whose marital status changed • who were under age 25 in 2001 • whose taxable income did not exceed the basic tax allowance in either 2001 or 2004 • who had their first child, • who received for the first time an allowance for single parents or an allowance for disabled persons and • who retired during the regarded period. A change in marital status implies a substantial change in the household’s financial situation that would influence estimation results. Individuals younger than 25 years may experience income increases at the beginning of their working life which are no behavioural response to the reform either. For lack of more 3 However, by doing so we neglect reactions at the margin like taking up an employment. Analysing repeated cross section data would be the preferable approach to include these reactions.
7
detailed information, we chose a threshold of 25 which is, of course, arbitrary. We further excluded taxpayers who had their first child during the regarded period since this circumstance strongly influences (generally the second earner’s) labour supply decisions. Finally taxpayers who retire are confronted with a considerable income cut which is not attributable to the reform either. After these adjustments 11 674 084 cases remain in the data set. Table 2 displays average taxable incomes, average pre-reform marginal tax rates and average percentage changes in marginal tax rates for taxpayers grouped by base year taxable income deciles. Table 2: Average taxable incomes, average pre-reform marginal tax rates and average percentage changes in marginal tax rates, in deciles
Decile
Average taxable income
Average prereform marginal tax rate
Average percentage change in statutory marginal tax rates
1
11 036
0,153
-18,32
2
16 724
0,244
-15,82
3
20 957
0,268
-7,66
4
24 739
0,285
-4,51
5
28 613
0,296
-3,98
6
33 437
0,306
-3,95
7
39 214
0,318
-3,95
8
46 756
0,333
-4,04
9
56 267
0,354
-4,76
10
126 160
0,431
-5,83
Note: 11 674 084 observations. The change in marginal tax rates is calculated by applying 2001 and 2004 tax law to 2001 income inflated to the 2004 level.
As already shown in section 4, taxpayers with lower and upper taxable incomes experienced the highest tax rate cuts, the changes in the middle of the income distribution are more moderate and quite homogeneous. The large percentage change in the first decile also results from the increase in the basic tax allowance what implies that some taxpayers are no longer subject to taxation after the reform. 8
5.2
Econometric approach
The estimation equation is derived from the model introduced in section 3, based on the Slutsky decomposition to estimate substitution and income effect separately. To approximate the growth rate of taxable income with respect to a percentage change in the NTR we use a logarithmised specification, as it is widely done in this field of research: ln(
(1 − τ1 ) T0 − T1 T I1 ) = ζ c ln +η + u. T I0 (1 − τ0 ) (Y0 − T I0 ) + (τ0 T I0 − T0 )
(5)
As already mentioned the marginal tax rate is determined by the level of taxable income.4 Regressing the change in the tax base, T I1 −T I0 , on the observed change in the NTR, (1 − τ1 ) − (1 − τ0 ), would thus cause simultaneity problems since the observed NTR in 2004 already contains tax induced adjustments. To avoid this problem we use a microsimulation model and create a counterfactual scenario: We calculate synthetic marginal tax rates τ˜0 and τ˜1 (and synthetic tax liabilities T˜0 and T˜1 , respectively), applying the pre- and the post-reform tax code to 2001 income which is inflated to the 2004 level for every taxpayer in order to exclude nominal income growth effects: τ˜0 = τ (T˜I0 , z0 )
(6a)
τ˜1 = τ (T˜I0 , z1 )
(6b)
T˜I0 is taxable income based on 2001 income inflated to 2004 level, and z0 and z1 denote 2001 and 2004 tax law respectively. With (1 − τ˜0 ) we obtain the NTR the taxpayer would face in 2004 if tax law and his real income did not change over time. (1 − τ˜1 ) is the NTR on inflated income under post-reform tax law, but before any behavioural response. The difference between these two NTRs serves as an instrument variable for the truly observed difference between the NTRs. The same procedure applies to the tax liability: The taxpayer would have to pay T˜0 in 2004 did his real income stay constant over time and did tax law not change. T˜1 would be tax liability under 2004 tax law. The difference between T˜1 and T˜0 serves analogously to the NTRs as an instrument variable for the observed difference between tax liabilities in the income effect term. 4
The same problem applies to the tax amount T1 which is needed to calculate the income effect term. Other factors which influence marginal tax rate and tax amount like joint filing do not put aside the endogeneity problem.
9
For the change in taxable income, we use our microsimulation model to adjust inflated base year income for other changes in the tax code that were part of the reform, e. g. concerning expenses and allowances. We thus eliminate income changes that are not directly attributable to the reform but due to differences in deduction possibilities over time. As before, inflating incomes excludes changes that are due to nominal growth. The difference between the observed taxable income in 2004 and the created counterfactual income is regressed on the NTR, the income effect variable and other control variables in a two-stage least squares setting. Including other control variables yields the following estimation equation:
ln(
T0 − T1 (1 − τ1 ) T I1 +η + ρ ln(T I0 ) + γ’X + u. (7) ) = ζ c ln (1 − τ0 ) (Y0 − T I0 ) + (τ0 T I0 − T0 ) T˜I0
The logarithm of base year income is used to control for reversion-to-the-mean effects. Taxpayers with very high or low transitory incomes in the base year are likely to experience large income changes over time. Those changes are often rather deviations from the individual long-run income equilibrium than true behavioural responses and are therefore controlled for.5 X is a vector of covariates that includes marital status, number of children, age (and age squared) and two dummy variables. The first dummy indicates whether the taxpayer declares mainly wage income or income from self-employment and is used to control for the fact that taxpayers with income from self-employment have more possibilities to rearrange different components of their income than employees. The second dummy indicates whether the main type of income (dependent or non-dependent income) changed between 2001 and 2004 and accounts for significant changes that go along with starting a business or switching from self- to regular employment. The dummy for marital status addresses the fact that joint filers are treated differently from single filers. The number of children controls for the time constraint parents face. Age and age squared serve as proxy variables for life-cycle effects.
5
Since we compare only two years of data which comprise one tax rate change, the inclusion of a 10-piece spline in base year income as is done by Giertz (2007) or Gruber/Saez (2002) in order to control for mean reversion and exogenous income trends more accurately would probably destroy identification.
10
6
Results
The basic results are presented in Table 3. All estimates are based on 2SLSRegressions. The F statistics for the coefficients of the NTR and the income effect instruments in the first stage are very high (not shown here). The elasticity of taxable income with respect to the NTR for all taxpayers is 1.06. This is remarkably high compared to the results in recent studies which mainly range from 0.2 to 0.6 (see Table 1). Since these are first results from ongoing work, they have to be regarded as preliminary and are somewhat likely to change when identification methodology is further developed. We discuss the magnitude of this coefficient in more detail below. The income effect is not only insignificant but also practically zero which corresponds to the results of Gruber/ Saez (2002) who do not find significant income effects either, whereas in the study carried out by Selén (2002) income effects are positive and significant. The parameter for the log in base year income has the expected negative sign and controls for income changes that are rather convergences towards the individual long-run equilibrium than tax induced responses. The other demographic control variables are all significant (except age squared), but their values are rather small. Joint filers experience a slightly higher income growth than single filers. Interpreting age as a life-cycle proxy, the negative sign implies that income is more stable when taxpayers get older. One could expect the number of (tax relevant) children to have a negative impact, but it might reflect life-cycle effects as well, being to a certain degree correlated with taxpayers’ age. In this respect, the positive sign of the child variable seems plausible. The negative sign of the switch-dummy is conform to the theoretical argumentation that starting a business or else giving up business in favour of a safe job or fixed working hours often goes along with a loss of income at the beginning. Surprisingly, the dummy for self-employment has a negative sign as well. One reason for this may be that mean reversion effects are not adequately controlled for and are partly captured by this variable. Another reason may be the definition of the dummy variable itself: We assigned taxpayers to be self-employed when they declared income from a closely held business or professional earnings, but we did not include income from participations and investments or capital gains that are attributable to business income according to income tax law. Since this is the more flexible part of these income types, an alternative definition of this dummy might lead to different results. When looking at the results per income group, the impact of the variables differs. In the lowest income group, the NTR elasticity becomes negative and the income effect increases, whereas in the middle and upper income groups the elasticity almost doubles, the income effect being negligible. Mean reversion effects are strongest in the lower and upper group. 11
Table 3: Basic estimation results
All taxpayers Variable
By income group < 30 000 e
30 000 e – 60 000 e
> 60 000 e
NTR elasticity (ζ c )
1.0601∗∗∗ (0.0064)
-0.0971∗∗∗ (0.0083)
1.7196∗∗∗ (0.0194)
1.9101∗∗∗ (0.0205)
Income elasticity (η)
0.0000 (0.0001)
-0.0123∗∗∗ (0.0006)
-0.0001∗∗∗ (0.0000)
-0.0001∗∗ (0.0000)
ln(base year income)
-0.1880∗∗∗ (0.0002)
-0.3663∗∗∗ (0.0004)
-0.1256∗∗∗ (0.0011)
-0.2237∗∗∗ (0.0012)
Joint filing
0.0395∗∗∗ (0.0003)
0.0498∗∗∗ (0.0003)
0.0205∗∗∗ (0.0006)
0.1231∗∗∗ (0.0017)
Number of children
0.0180∗∗∗ (0.0001)
0.0153∗∗∗ (0.0001)
0.0234∗∗∗ (0.0002)
0.0195∗∗∗ (0.0005)
Age
-0.0056∗∗∗
-0.0065∗∗∗
-0.0053∗∗∗
-0.0059∗∗∗
(0.0000)
(0.0000)
(0.0000)
(0.0001)
0.0000 (0.0000)
0.0000 (0.0000)
0.0000 (0.0000)
0.0000 (0.0000)
Switch
-0.0889∗∗∗ (0.0007)
0.0479∗∗∗ (0.0008)
-0.215∗∗∗ (0.0014)
-0.3205∗∗∗ (0.0023)
Self-employed
-0.0833∗∗∗ (0.0005)
-0.0501∗∗∗ (0.0005)
-0.1587∗∗∗ (0.0008)
-0.1072∗∗∗ (0.0015)
Constant
2.1788∗∗∗ (0.0022)
3.9798∗∗∗ (0.0045)
1.4957∗∗∗ (0.0119)
2.5140∗∗∗ (0.0133)
11 674
5 600
4 640
1 410
Age squared
No. of obs. (in 1000)
Note: Estimates from 2SLS regressions. Standard errors in parentheses. denote significance on the 1 %, 5 % and 10 % level respectively.
∗∗∗ /∗∗ /∗
12
An elasticity that exceeds one is much higher than what has been found in most studies. Aside from other exogenous income trends between 2001 and 2004 we may not have adequately controlled for in our regressions, descriptive statistics show that especially income from rent and lease underwent a change that surely is not entirely attributable to tax incentives. When we look at Table 4, we see that in 2003 the total balance switched from a negative to a positive value, with an overall change from around -2,5 bn e in 2001 to over 5 bn e in 2004.6 The relatively constant number of taxpayers that report income from rent and lease (increasing only by 2,5 %) suggest that this development is not due to additional taxpayers declaring high positive incomes from rent and lease but that it is in fact individual income that increases over time. This is supported by mean income which changes from -630 e in 2001 to 1283 e in 2004. Table 4: Income from rent and lease 2001-2004
Mean
Median
Sum
N
2001
-630
-198
-2 435 611 965
3.868.881
2002
-117
-69
-458 820 609
3.906.534
2003
373
53
1 473 236 964
3.951.914
2004
1283
258
5 088 235 807
3.964.539
An attempt to explain this development may be that these are cumulative effects of changing individual cost-revenue ratios. Deductible acquisition or maintenance costs which initially exceeded revenue from rent and lease decline over time, e. g. when using declining-balance depreciation or as a consequence of rent increases or betterments in real property. Amortised mortgages also necessarily result in positive income in the course of time. A coincidence of these factors may in total lead to an overbalance of positive income. In order to see how much this trend influences the estimation results, we exclude income from rent and lease from our calculations, although we risk losing tax induced response to some extent since this income type often serves the purpose to reduce the tax base. The estimation results are presented in Table 5. 6
These are the values calculated only for taxpayers who appear continuously in the panel. The cross section totals differ in magnitude, but show the same tendency (see Statistisches Bundesamt (2008)).
13
Table 5: Estimation results without income from rent and lease
All taxpayers Variable
By income group < 30 000 e
30 000 e – 60 000 e
> 60 000 e
NTR elasticity (ζ c )
0.863∗∗∗ (0.0061)
0.8389∗∗∗ (0.0061)
0.2459∗∗∗ (0.0089)
0.5896∗∗∗ (0.0155)
1.2967∗∗∗ (0.0188)
Income elasticity (η)
-0.0002∗∗∗ (0.0000)
-0.0001∗∗∗ (0.0000)
-0.0059∗∗∗ (0.0002)
-0.0008∗∗∗ (0.0001)
-0.0001∗∗∗ (0.0000)
ln(base year income)
-0.174∗∗∗ (0.0002)
-0.1738∗∗∗ (0.0001)
-0.2932∗∗∗ (0.0004)
-0.0863∗∗∗ (0.0009)
-0.2033∗∗∗ (0.0011)
Joint filing
0.0356∗∗∗ (0.0003)
0.0352∗∗∗ (0.0003)
0.0446∗∗∗ (0.0003)
-0.0071∗∗∗ (0.0005)
0.0973∗∗∗ (0.0015)
Number of children
0.0184∗∗∗ (0.0001)
0.0184∗∗∗ (0.0001)
0.0154∗∗∗ (0.0001)
0.0245∗∗∗ (0.0001)
0.0224∗∗∗ (0.0004)
Age
-0.0066∗∗∗
-0.0065∗∗∗
-0.0078∗∗∗
-0.0056∗∗∗
-0.0071∗∗∗
(0.0000)
(0.0000)
(0.0000)
(0.0000)
(0.0000)
0.0000 (0.0000)
0.0000 (0.0000)
0.0000 (0.0000)
0.0000 (0.0000)
0.0000 (0.0000)
Switch
-0.1104∗∗∗ (0.0008)
–
0.0249∗∗∗ (0.0010)
-0.2342∗∗∗ (0.0012)
-0.3331∗∗∗ (0.0022)
Self-employed
-0.0984∗∗∗ (0.0005)
-0.1103∗∗∗ (0.0004)
-0.0686∗∗∗ (0.0006)
-0.1742∗∗∗ (0.0007)
-0.1129∗∗∗ (0.0013)
Constant
2.0706∗∗∗ (0.0021)
2.0668∗∗∗ (0.0021)
3.2991∗∗∗ (0.0043)
1.1285∗∗∗ (0.0101)
2.3861∗∗∗ (0.0123)
5 510
4 620
1 450
Age squared
No. of obs. (in 1000)
11 674
Note: Estimates from 2SLS regressions. Standard errors in parentheses. ∗∗∗ /∗∗ /∗ denote significance on the 1 %, 5 % and 10 % level respectively.
14
The results are now quite different. The elasticity for all taxpayers falls from 1.06 to 0.86 (and even slightly more when leaving out the switch-dummy). But more remarkable are the changes in estimates when distinguishing between different income groups: In contrast to the situation when income from rent and lease was included, the elasticity for taxpayers with base year taxable income less than 30 000 e is now positive. In the middle income group, elasticity falls by more than one percentage point from 1.72 to 0.59, and for taxpayers with taxable income exceeding 60 000 e it falls from 1.91 to 1.3. Income growth that is driven by the striking increase in income from rent and lease seems to be concentrated in the middle and upper part of the (taxable) income distribution, whereas in the lower income group losses predominate. The influence of the income effect is still negligible. The coefficients for the demographic control variables do not change very much, so the distribution of income from rent and lease does not seem to be highly correlated with those socio-economic characteristics. An elasticity of 0.86 is still relatively large. A possible reason for this may be the fact that our data subset is rather restrictive. Since we exclude taxpayers who had their first child or who retired during the period of the study, we systematically exclude taxpayers who are likely to experience income cuts – these cases are included in most other studies. Another interesting point is made by Slemrod/ Kopczuk (2002) and Kopczuk (2005) who argue that taxable income elasticity is not a fixed parameter but depends on the deduction possibilities defined by tax law and thus can to some extent be controlled by policy makers. Since German income tax law offers various possibilities to narrow the tax base over the whole income distribution, it seems possible that behavioural response may be more pronounced in Germany than in other countries.
7
Conclusions
This paper presents first results from ongoing work to identify the effects of the tax rate changes in 2004 on taxable income in Germany. We find an overall taxable income elasticity of 1.06 when including all types of income and an elasticity of 0.86 when leaving out income from rent and lease. The estimates differ greatly between income groups, with high income taxpayers showing stronger responses. The magnitude of the results leads to the assumption that the estimates are currently still driven by factors not yet fully controlled for and therefore they have to be handled with care. A more thorough analysis of the data (e. g. concerning income composition and distribution) is needed to better identify exogenous income trends that are not (at least entirely) tax related. Nevertheless we obtain a first insight in the dimension of taxpayer behaviour in Germany which is nontrivial and has to be taken into account when considering further tax reforms. 15
References Aarbu, K. O./Thoresen, T. O.: Income Responses to Tax Changes - Evidence from the Norwegian Tax Reform. In: National Tax Journal LIV (2001), Nr. 2, 319–335 Auten, C./Carroll, R./Gee, G.: The 2001 and 2003 Tax Rate Reductions: An Overview and Estimate of the Taxable Income Response. In: National Tax Journal 61 (2008), Nr. 3, 345–364 Auten, G./Carroll, R.: The Effect of Income Taxes on Household Income. In: The Review of Economics and Statistics 81 (1999), Nr. 4, 681–693 Feldstein, M.: The Effect of Marginal Tax Rates on Taxable Income: A Panel Study of the 1986 Tax Reform Act. In: Journal of Political Economy 103 (1995), Nr. 3, 551–572 Giertz, S. H.: The Elasticity of Taxable Income over the 1980s and 1990s. In: National Tax Journal 60 (2007), Nr. 4, 743–768 Goolsbee, A.: What Happens When You Tax the Rich? Evidence from Executive Compensation. In: Journal of Political Economy 108 (2000), Nr. 2, 352–378 Gottfried, P./Schellhorn, H.: Empirical Evidence on the Effects of Marginal Tax Rates on Income - the German Case. IAW Discussion Paper, Februar 2004 Gruber, J./Saez, E.: The elasticity of taxable income: evidence and implications. In: Journal of Public Economics 84 (2002), 1–32 Hansson, A.: Taxpayers’ responsiveness to tax rate changes and implications for the cost of taxation in Sweden. In: International Tax and Public Finance 14 (2007), Nr. 5, 563–582 Heim, B. T.: The Effect of Recent Tax Changes on Taxable Income: Evidence from a New Panel of Tax Returns. In: Journal of Policy Analysis and Management 28 (2009), Nr. 1, 147–163 Holmlund, B./Söderström, M.: Estimating dynamic income responses to tax reforms: Swedish evidence. Working Paper 28, 2008, Institute for Labour Market Policy Evaluation (IFAU) Kopczuk, W.: Tax bases, tax rates and the elasticity of reported income. In: Journal of Public Economics 89 (2005), 2093–2119 16
Lindsey, L.: Individual taxpayer response to tax cuts: 1982-1984, with implications for the revenue maximizing tax rate. In: Journal of Public Economics 33 (1987), 173–206 Saez, E.: The effect of marginal tax rates on income: a panel study of ’bracket creep’. In: Journal of Public Economics 87 (2003), 1231–1258 Selén, J.: Taxable Income Responses to Tax Changes - Panel Analysis of the 1990/91 Swedish Reform. FIEF Working Paper No. 177, August 2002 Sillamaa, M.-A./Veall, M. R.: The Effect of Marginal Tax Rates on Taxable Income: a Panel Study of the 1988 Tax Flattening in Canada. In: Journal of Public Economics 80 (2001), 341–356 Slemrod, J./Kopczuk, W.: The optimal elasticity of taxable income. In: Journal of Public Economics 84 (2002), 91–112 Statistisches Bundesamt: Finanzen und Steuern - Jährliche Einkommensteuerstatistik 2004. 2008
17
