Behavioral Responses and the Distributional Effects of The Russian ‘Flat’ Tax

January 2013

REVISED

Abstract:

This paper simulates the distributional impact of the Russian personal income tax (PIT) following the flat tax reform of 2001 using data from the Russian Longitudinal Monitoring Survey. I decompose the change in the distribution of net income into a direct (tax) effect and an indirect effect. The indirect effect is further decomposed into evasion and productivity effects using existing estimates of these respective elasticities. As expected, the direct tax effect increased net income inequality. Changes in the pre-tax distribution (indirect effect), on the other hand, had a large negative impact on inequality thus leading to an overall decline in net income inequality. I also find that the tax-induced evasion response increased reported net income inequality while reducing consumption inequality. To the extent that consumption approximates actual income, these results demonstrate that the flat tax reform had a much smaller effect on actual income inequality than on reported income inequality. More importantly, relative to non-tax factor, the reform had little overall effect on income inequality. This suggests that objection to flatter tax schedules on the grounds of income inequality is mostly misguided, especially in transitional countries with high levels of evasion.

Keywords: Income distribution, evasion, simulation, taxes, consumption.

JEL Classification: D3, D63, H24, H26, H31

* School of Public and Environmental Affairs, Indiana University - Bloomington, U.S.A. and Institute for the Study of Labor, Bonn, Germany. Email: duncande@indiana.edu,

1

Denvil Duncan*

1. Introduction

An increasing number of countries have adopted or are considering the adoption of a flat

rate PIT schedule. The most popular among these is the Russian flat tax reform of 2001, which

is believed to have acted as a catalyst for other countries in recent years.1 This trend toward

flatter PIT schedules has generated significant debate in tax policy circles.2 The conventional

argument is simple; a flatter PIT schedule reduces the tax burden on the rich relative to the poor,

thus increasing the inequality in net income. Simultaneously, the lower tax burden induces

behavioral responses that not only improve efficiency, but also increases pre-tax income of the

rich relative to that of the poor, which further increases net income inequality. That is, flattening

PIT schedules increases income inequality due to changes in the direct tax burden as well as

through tax-induced changes in behavior (indirect effects). As a result, policymakers with strong

preferences for equality tend to oppose efforts to flatten PIT schedules.

However, it is not clear that the Russian flat tax reform, for example, had the effects

described above. This is because the analysis above ignores the fact that tax-induced behavioral

responses include evasion and avoidance, both of which represent income shifting rather than

real changes in income.3 Given the prevalence of evasion in Russia and other Eastern European

countries, policy conclusions regarding the redistributive impact of flat tax reforms ought to

consider the impact of a flattened PIT schedule on both Reported Net Income and Actual Net

1 Current estimates put the number of independent countries with a flat rate PIT at 31 as at December 31, 2011. This number is up from 14 in 2005. The majority of countries using the flat rate PIT are the former communist countries of Eastern Europe.2 For example, Fuest, Peichl, and Schaefer (2008) and Paulus and Peichl (2009) are among a long list of papers that evaluate the distributional impact of flat taxes.3 Avoidance occurs when taxpayers make strategic decisions to legally reduce their total tax liability. I assume that avoidance is accounted for in our measures of Actual Gross Income. Evasion occurs when taxpayers simply fail to report a portion of their actual gross income to tax authorities. Throughout the paper, I refer to this unreported portion of actual gross income as hidden income. Reported Gross Income represents actual gross income less hidden income.

2

Income; the latter comprises both reported net income and hidden income. While the conclusions

of the preceding analysis still hold for reported net income, the distributional impact of PIT rates

on actual net income inequality is likely to be ambiguous and possibly counterintuitive under

certain conditions. For example, if the rich are induced to report a greater share of their hidden

income, both reported gross and net income inequality will rise while actual net income

inequality will most likely fall. This example demonstrates the need to identify not only the

effects of a tax policy change on reported and actual income inequality, but also the various

channels through which this change affects the distribution of income, as these channels need not

work in the same direction.

I use data from the Russian Longitudinal Monitoring Survey (RLMS) and elasticities of

evasion and productivity in a micro-simulation counterfactual analysis to determine the effect of

Russia’s flat tax reform on income inequality in the years immediately after the reform. This

methodology allows me to distinguish between direct tax effect and indirect behavioral effects,

and identify whether the evasion or the productivity response is the major driving force behind

the indirect behavioral effects.4 Additionally, I am able to answer an important policy relevant

question; do these tax-induced behavioral responses affect reported net income inequality

differently than actual net income inequality? Following the literature, I use consumption as a

proxy for actual net income.

The results show that indirect behavioral responses had a larger effect on the distribution

of income than the direct tax effect. I identify the tax-induced components of the indirect effect

4Following Gorodnichenko, Martinez-Vazquez, and Sabirianova Peter (2009), I classify the behavioral responses into two broad categories: evasion and real productivity effects. The productivity effect is broadly defined to include all the possible behavioral changes that can affect the total income earned except compliance, which is identified separately. The indirect effect also includes non-tax induced changes in behavior. However, the primary focus of this paper is on the distributional impact of tax-induced behavioral responses.

3

and show that the evasion response had a larger impact on inequality than the productivity

response. While the qualitative effect of the productivity responses is the same for both reported

net income and actual net income, I find that the sign of the evasion effect depends on the

income measure; the evasion response lowered actual net income inequality while increasing

reported net income inequality. More importantly, the downward trend in income inequality that

began before the tax reform was unaffected by the flat tax reform.

This analysis makes several important contributions that are of interest to policymakers

and academics. First, it identifies the relative size and sign of the direct and indirect effects

through which the Russian flat tax reform affected the distribution of income. The existing

literature either focuses on the U.S. PIT system (e.g., Alm, Lee, and Wallace 2005; Poterba

2007) or uses hypothetical flat tax reforms in Western Europe (e.g., Fuest, Peichl, and Schaefer

2008). It is also the first, to decompose the tax-induced behavioral effects into evasion and

productivity responses. Existing work in this area has identified parts of the productivity

response (e.g., Altig and Carlstrom 1999), but to my knowledge, no one has so far identified the

evasion effect.

Second, the paper contributes to current tax policy debates regarding the effect of tax

progressivity on the distribution of net income. Recent debates in the U.S. and other countries

seem to suggest that redistribution through the tax code via increased progressivity is an effective

policy for reducing income inequality. My results suggests otherwise. In fact, redistribution via

increased progressivity can lead to results contrary to that intended by the policymaker,

especially for countries with high levels of evasion. More importantly, my results indicate that

that changes in gross income that are unrelated to changes in progressivity tend to have a larger

effect on net income inequality than tax induced changes in gross income. This suggests that

4

investing in programs that improve employability and earning power are likely to be more

effective policies in reducing inequality than simply increasing tax progressivity.

Finally, the results imply that the actual costs of the efficiency gained from switching to a

flatter tax schedule are likely to be much lower than expected. This conclusion is based on the

argument that changes in inequality that arise from income shifting via evasion partly reflect pre-

existing inequality. In other words, observed inequality can increase when a lower tax rate

causes individuals in the right tail of the income distribution to report a relatively greater share of

their income. This increase in inequality represents a shift toward the actual inequality that

mostly existed prior to the tax change. The policy implication of this result is that it may be

optimal to adopt a flatter tax schedule not only because it is more efficient but also because the

true impact on income inequality is smaller than observed. This particular conclusion is

important because flatter tax schedules, though more efficient, are generally resisted because of

their perceived effect on income inequality.

The remainder of the paper is structured as follows. Section two discusses the theoretical

framework. A brief summary of the Russian tax reform is discussed in section three and section

four describes the data. The empirical strategy and results are discussed in section five, and

section six concludes with policy implications.

2. Theoretical Discussion

The objective this section is to describe how the Russian flat tax reform affected the

distribution of reported and actual net income as well as the channels through which inequality

was affected. Duncan and Sabirianova Peter (2012) address these issues in a decomposition

5

framework and identify two clear predictions about the relationship between tax progressivity

and income inequality.

Reported Income Inequality The first prediction is that an increase in taxes on the rich

reduces inequality in reported net income via a direct mechanical channel and two indirect

behavioral channels. The direct channel is straightforward; a higher tax burden on the rich lowers

their net income, holding behavioral responses constant, and thus reduces reported income

inequality. This effect is reinforced by two behavioral responses: higher taxes reduce earned

income via a productivity effect and increase hidden income via an evasion effect. Since the rich

hide a relatively greater share of their income (Johns and Slemrod 2010), and are relatively more

responsive to taxes (Heim 2009; Burns and Zilliak 2012), their reported income falls relative to

that of the poor, which implies that reported net income inequality decreases. Therefore, the

three channels work in the same direction to lead to an unambiguous decline in reported net

income inequality. This result is straightforward, intuitive, and supports advocates of

redistribution via tax progressivity.

Actual Income Inequality However, ignoring the distinction between reported income

and actual income may lead to misguided policies. In particular, Duncan and Sabirianova Peter

(2012) show that the effect of tax rates on actual net income inequality is not only ambiguous,

but also likely to be smaller than the effect on reported net income inequality. This ambiguity is

explained by the evasion response. While the evasion effect reduces reported income inequality,

it increases actual net income inequality because the share of untaxed income among the rich

increases.5 They also show that the net effect on actual net income inequality may be positive if

the evasion effect is very large relative to the productivity effect and the level of evasion is very

5 The direct effect and the productivity response affect reported net income the same way they affect actual net income inequality.

6

high. In other words, it is possible for an increase in tax progressivity to lead to an increase in

actual income inequality due to opposing behavioral responses.

While Duncan and Sabirianova Peter (2012) provide empirical evidence that is consistent

with the discussion above, they are unable to identify the precise contribution of each channel

because they rely on macro data for their analysis. Poterba (2007) and Alm, Lee, and Wallace

(2005) use Micro data to identify the effect of tax rate on income inequality and find that the

indirect effect is larger than the direct mechanical effect. However, they do not identify the

components of the indirect effect. Altig and Carlstrom (1999) and Gramlich, Kasten and

Sammartino (1993) also explore similar questions, but only identify part of the behavioral

response; the former looks at labor supply and capital gains while the latter focuses on labor

supply and savings. Palme (1996) use a micro-simulation method to identify the effect of the

1991 Swedish tax reform on income distribution, but does not account for indirect economic

effects.

These findings highlight the need to distinguish between actual and reported net income

inequality when analyzing the distributional impact of the Russian flat tax reform. In particular,

distributional analyses should separate the direct effect from the indirect effect and decompose

the indirect effect into its evasion and productivity components. This is especially important if

correct policy conclusions are to be drawn from case studies of Russia and other Eastern

European countries that have flattened their PIT schedules, and where tax evasion is prevalent.

Ignoring these distinctions can lead to incorrect results and seriously misguided policy

prescriptions.

7

3. Russia and the Flat Tax

In 2001, Russia replaced its graduated PIT schedule with a linear PIT, representing one of

the most significant PIT reforms of the 21st century. The top two rates of 30 and 20 percent were

eliminated and the threshold was increased from 3,168 rubles to 4,800 rubles. Those earning an

income above 4,800 rubles paid the same flat rate of 13 percent after the reform. From Table 1,

we observe that individuals making over 50,000 rubles were the primary beneficiaries of the

reform. Given that high-income taxpayers benefited the most, this case allows me to answer

specific questions about the impact of the tax reform on those taxpayers that were most likely to

modify their behavior. Focusing on Russia also allows me to identify the distributional impact of

an actual flat PIT reform. Furthermore, the distributional impact of Russia’s PIT reform is of

great practical relevance to other Eastern European countries that have, or plan to, follow

Russia’s lead and adopt similar policies.

Table 1: The PIT Rate Structure Before and After ReformBefore Reform (2000) After Reform (2001-2004)

Taxable Income1 Marginal Rate Taxable Income1 Marginal Rate

Below 3,168 0 Below 4,800 0

3,168 to 50,000 13 Above 4,800 13

50,000 to 150,000 21

Above 150,000 31

Note: (source: Ivanova, Keen, and Klemm 2005). Pre-reform marginal rates include employees’ share of the payroll tax, which was 1 percent. This tax was eliminated as part of the reform.

With respect to data availability, Russia has some limitations that must be addressed for

the study to be feasible. The ideal data set would have comprehensive, longitudinal data on

actual and reported gross income before and after a major tax reform. The data would also

include information on deductions, credits and other allowances, tax liabilities, and hence

8

measures of net income. Unfortunately, these data are not publicly available for any country in

the world.6 However, data collected from the Russian Longitudinal Monitoring Survey (RMLS)

allow for an adequate estimation of the most critical parameters needed for the analysis: the

evasion and productivity elasticities (Gorodnichenko, Martinez-Vazquez, and Sabirianova Peter

2009). Additionally, Gorodnichenko, Martinez-Vazquez, and Sabirianova Peter (2009) and

Ivanova, Keen, and Klemm (2005) argue that consumption is a suitable proxy for actual net

income in Russia and I do the same. The corresponding gross income measures are obtained by

inverting the tax function in each period taking into account the basic deductions that are

available to everyone. More on this below.

4. Data

4.1 Description

The data are taken from the Russian Longitudinal Monitoring Survey (RLMS). It is

widely representative of the Russian population, covering approximately 32 regions, 38

randomly selected primary sampling units, 7 Russian federal districts, and more than 4,000

households and 10,000 adults per year. According to the host website of the RLMS, the

response rate exceeds 80 percent for households and 95 percent for individuals within each

household.7 Besides the relatively large sample size, the data set has a panel feature with two

years before and 4 years after the Russian tax reform, which makes it suitable for the current

analysis.

6 Of course, data sets with the individual variables do exist in most countries. However, public access is usually limited to restricted versions of these data. Furthermore, there is very limited access to datasets with longitudinal information on income, consumption, and taxes. 7 The survey is a joint project between the Population Center at the University of North Carolina and the Russian Academy of Sociology. See http://www.cpc.unc.edu/projects/rlms-hse for a description of the Russian Longitudinal Monitoring Survey. The RLMS samples are replenished over time to deal with attrition (Gorodnichenko, Sabirianova Peter, and Stolyarov 2010). Additional information on sample selection, attrition and the like can be obtained from the host site; http://www.cpc.unc.edu/projects/rlms.

9

The sample used in the empirical analysis is restricted to households in which at least one

individual is between the ages of 25 and 60 years old. Additionally, I focus on the years 2000

(pre-reform base year) and 2002 for my base results. Although the reform became effective on

January 1st 2001 and data are collected in the last quarter of the year, I exclude the year 2001

from the analysis to allow individuals more time to respond to the new tax schedule. Income

distributions are summarized by the Gini coefficient that are calculated using only the non-zero

values of each income measure. All income/consumption measures are converted to December

2002 prices, and household measures are adjusted using the OECD equivalence scale.

Additionally, the individual (household) level inequality indices are calculated using the RLMS

individual (household) sample weights to address sample attrition and other sampling errors.8

4.2 Reported and Actual Net Income

The RLMS includes reported net income data at both the individual and household levels.

Individual measures include actual monetary labor income earnings received last month and

contractual monetary labor earnings (received on average over the last 12 months). Actual

income is prone to monthly income shocks, which may result from wage arrears, forced leave,

and sickness, among others. Contractual earnings on the other hand, are more stable as they

reflect the usual income received per month over a one-year period. Contractual monetary labor

earnings are used to create a third income measure: imputed contractual monetary labor earnings.

Using the imputed contractual earnings is advantageous because it provides a more accurate

description of income within households, which is the primary unit of analysis.9 The baseline

results at the individual level use imputed contractual labor earnings at the primary and

8 The RLMS sample weights adjust for sample design factors and deviations from the census characteristics, which implicitly address sample attrition.9 The imputation is for working non-respondents. Because the PIT is assessed on the individual, the imputation is done in an effort to obtain an accurate measure of household net income, which involves summing tax liability across individuals within households.

10

secondary job. Although labor earnings is the only component of income available at the

individual level, it represents over 80% of income and should therefore provide an adequate

estimation of the distributional impact of the tax reform at the individual level. Imputed

contractual labor earnings are summed across individuals within households to obtain a base

measure of household reported net income.

Given the unobservable nature of actual net income, I use consumption as a proxy. I

specifically use non-durable consumption, which includes expenditure data on more than 55 food

items both within and outside the home plus durable expenditure, as my baseline measure of

actual income. While income measures are available at both individual and household level,

consumption is only available at the household level.

4.3 Gross Income Measures

The RLMS does not include information on gross income. Since the analysis requires

these data, I impute them by inverting the tax function for each period. The implicit assumptions

underlying the inversion are that monthly income is received uniformly throughout the year and

that reported net income reflects tax liability actually paid. Starting with net income, I recover

the gross income measures using an iterative process which simultaneously imputes gross

income and the implied tax liability for each individual. Next, I calculate gross income at the

household level by adding household level tax liability to the respective measures of household

net income, where household tax liability is the within household summation of the individual

level tax liability based on imputed contractual earnings.10

10 This procedure is described in more detail in the methodological appendix, which is available upon request.

11

5. Empirical Strategy and Results

I use estimates of the elasticity of actual gross income with respect to taxes and the

elasticity of evasion with respect to taxes to simulate counterfactual net income distributions,

which are then used to decompose the change in the distribution of net income into direct,

evasion, and productivity effects.

5.1 Identification of the Distributional Effect

The distributional impact of the tax reform is obtained using a counterfactual analysis as

in Alm, Lee, and Wallace (2005) and Poterba (2007). Implementation is via micro-simulation

exercises that allow me to examine the effect of taxes on the distribution of income with and

without behavioral responses. The methodology is implemented in several steps. First, I

calculate the Gini coefficient for the pre-reform period (year 2000) and the post-reform period

(year 2002). These two measures are used to calculate the total change in the distribution of net

income between the two periods. I then calculate two counterfactual net income distributions;

net income when the pre-reform tax schedule is applied to the post-reform income and net

income when post-reform tax schedule is applied to pre-reform income.11 The indirect effect is

obtained by comparing the Gini coefficient of the former counterfactual distribution with the

Gini coefficient of the pre-reform net income distribution. Similarly, I obtain the direct effect by

comparing the latter counterfactual distribution with the pre-reform net income distribution.

The tax-induced behavioral effects, which are part of the indirect effect, are identified

using two distinct approaches. The first approach does not distinguish between evasion and real

productivity effects. In other words, I use the elasticity of reported gross income to estimate a

11 Estimating the counterfactuals require several steps of which the most important is the imputation of gross income. The steps are outlined in the methodological appendix, which is available upon request.

12

counterfactual level of income that reflects all tax induced behavioral responses. This

counterfactual income distribution is compared to the pre-reform income distribution to

determine the change in inequality that can be explained by tax induced behavioral responses. By

ignoring the fact that the evasion response may affect the distribution of actual income

differently than it does reported income inequality, this approach is likely to overestimate the

distributional impact of the tax changes.

I correct for this in the second approach, which distinguishes between evasion and real

productivity responses. The evasion effect is estimated by comparing the pre-reform (year 2000)

distribution of net income with the distribution of net income that would exist if the only tax-

induced behavioral response to the tax reform was evasion. The productivity effect is obtained

similarly, except that I assume the only response is through productivity changes. I also estimate

the total behavioral effect by allowing both evasion and productivity to change simultaneously.

This approach allows each behavioral response to affect the distribution of each measure

of income differently, and should therefore improve the accuracy of the estimates and any policy

conclusions that can be drawn from the results. Still, it is worth noting that the framework

described here does not take general equilibrium effects into account. Instead, the analysis

focuses on the immediate effects of the tax reform and is therefore similar to Alm, lee, and

Wallace (2005) and Poterba (2007) in this respect.

5.2 Results

The results presented in Table 2 show that inequality declined between 2000 and 2002,

and that the indirect effect are the primary reason for the decline. Although the direct effect led

to an increase in income inequality, there was a much larger negative indirect effect. This is true

13

for measures of imputed contractual earnings at the individual level as well as household level

reported net income before public transfers, and actual net income. The fact that the indirect

effect is larger than the direct effect is consistent with previous work by Alm, Lee, and Wallace

(2005) and Poterba (2007), and is our first piece of evidence that the oft cited objection to flatter

income tax schedules does not appear to be relevant in the case of Russia.

Table 2: Distributional Impact of the Flat Tax Reform: Direct Vs. Indirect Effect Levels Decomposition

Tax Schedule 2000 2002 2000 2002 Total effect

Indirect effect

Direct effectIncome year 2000 2002 2002 2000

1 2 3 4 5 6 7

Panel A: Individual

Contractual Earnings 0.4812 0.4402 0.4230 0.489 -8.515 -12.091 1.623

Panel B: Household

Consumption 0.495 0.449 0.447 0.497 -9.395 -9.857 0.350Income 0.479 0.445 0.433 0.486 -7.089 -9.616 1.408

Notes: Reported are the Gini coefficients in levels (columns 1 - 4) and percent changes (columns 5 - 6). The sample is restricted to non-zero values for each variable; imputed contractual labor earnings at the individual level and durable plus non-durable consumption and reported income before public transfers at the household level. Decompositions are calculated as follows: the total effect is the change between the first two columns, the indirect effect is the change between columns one and three, and the direct effect is the change between columns one and four.

However, it is important to note that the indirect effect identified in Table 2 includes tax-

induced and nontax-induced responses.12 As a matter of policy evaluation, it is important that we

identify the indirect effects induced by the tax reform. I isolate these tax-induced indirect effects

and present the results in Table 3.

As indicated earlier, correct policy conclusions can only be drawn from these results if

the simulation treats evasion and real productivity responses separately. The importance of this

12 Gorodnichenko, Sabirianova Peter and Stolyarov (2010) provide a detailed discussion of the trends in inequality in Russia between 1994 and 2005 including possible nontax-induced factors that may have contributed to the decline. For example, Russia was in the middle of an economic recovery at the time of the tax reform, and this nontax-induced effect may have affected income inequality. More generally, there is some indication that rapid changes in globalization, which characterizes the sample period, can affect income inequality (Salvatore 2007).

14

distinction can be seen by comparing the results in column 5, which makes no distinction

between evasion and productivity responses, with the results in columns 6 to 8, which

distinguishes between the two types of responses. First, the results in column 5 indicate that tax-

induced behavioral responses increased reported and actual net income inequality. However, the

magnitude of the effect is overestimated because it treats the evasion and productivity responses

as one and the same.

The results in columns 6 and 7 highlight the magnitude of this bias. The evasion effect

on reported net income is larger than the real productivity effect regardless of unit of analysis.

This suggests that a relatively larger share of the tax-induced increase in reported net income

inequality is being driven by increased reporting. Since the evasion response involves shifting

existing income, it represents an artificial change in the distribution of reported net income,

which leads to an overestimate of the distributional impact of the reform. Because policymakers

can only observe reported income, the results suggest that policy prescriptions should be based

on the contribution of the real productivity effect instead of the combined effect. Therefore, the

relevant comparison for policy purposes is between columns 5 and 7, and here we observe that

the real productivity effect is an order of magnitude smaller than the effect in column 5.

The merits of distinguishing between reported and actual net income inequality are also

evident from the results in panel B of Table 3. I find that the tax-induced indirect effects have a

noticeably smaller effect on actual net income relative to reported net income. Furthermore, this

differential effect is mostly driven by the evasion effect, which is large and positive for reported

net income, but negative and practically zero for actual net income. This is contrary to what is

observed when no distinction is made between evasion and productivity responses (see results in

column 5). These results are line with expectation since evasion only affects actual net income

15

through income shifting while reported net income is directly affected by both evasion and

productivity responses. It follows that ignoring the distinction between evasion and productivity

responses as well as the distinction between actual and reported net income are likely to lead to

incorrect policy prescriptions.

Table 3: Distributional Impact of the Flat Tax Reform: Tax-Induced Behavioral Effects Levels Indirect

effect (A)Indirect effect (B)

Tax Schedule 2000 2000 2000 2000 Evasion effect

Real effect

Combined effectAdjustment None Evasio

n Real Both

1 2 3 4 5 6 7 8

Panel A: Individual

Contractual Earnings

0.481 0.517 0.504 0.536 15.822 7.425 4.739 11.321

Panel B: Household

Consumption 0.495 0.495 0.504 0.503 9.266 -0.100 1.689 1.509Income 0.479 0.505 0.495 0.519 6.563 5.354 3.318 8.471

Notes: Reported are the Gini coefficients in levels (columns 1 - 4) and percent changes (columns 5 - 8). Indirect effect ‘A’ does not distinguish between evasion and productivity responses while indirect effect ‘B’ treats them differently. The sample is restricted to non-zero values for each variable; imputed contractual labor earnings at the individual level and durable plus non-durable consumption and reported income before public transfers at the household level. Decompositions are calculated as follows: the evasion effect is the change between the first two columns (assumes productivity response is zero), the real (productivity) effect is the change between columns one and three (assumes evasion response is zero), and the total effect is the change between columns one and four (assumes both productivity and evasion responds). Column 5 reports the tax-induced indirect effect if no distinction is made between evasion and productivity responses; the elasticity of reported gross income is used to calculate the counterfactuals in this case. Adjustments are made using the following baseline parameters: evasion elasticity 0.26, productivity elasticity -0.04, and reported gross income elasticity -0.21 (Gorodnichenko, Martinez-Vazquez, and Sabirianova Peter 2009); evasion as a share of actual income 0.25 is from Ivanova et al. (2005).

As indicated by the results in Tables A1 and A2, the results discussed here are

qualitatively the same regardless of parameter values or definition of income. I also restrict the

analysis to individuals with non-zero values for imputed contractual earnings and find similar

results. These results are available upon request.

16

6. Conclusion and Policy Implications

Redistribution via the personal income tax code continues to feature prominently in tax

debates in many countries, and is one of the main arguments against the adoption of flatter tax

schedules. However, these debates often ignore the fact the behavioral responses play a role in

the final distributional impact of taxes, and that not all behavioral responses affect income

inequality the same way. The flat tax reform implemented by Russia in 2001 provides an

opportunity to study the effects of flattening PIT schedules on the distribution of income in a

way that addresses these issues, and add much needed content to the current policy debates.

I use data from the Russian Longitudinal Monitoring Survey (RLMS) and elasticities of

evasion and productivity in a micro-simulation counterfactual analysis to determine the effect of

Russia’s flat tax reform on income inequality in the years immediately after the reform. Several

important results emerge from the analysis. First, tax-induced indirect behavioral responses have

a larger effect on the distribution of income than the direct tax effect. Second, tax-induced

indirect behavioral responses have a larger effect on the distribution of reported net income than

on the distribution of actual net income. Most of this difference is driven by the evasion

response. Third, ignoring the difference between evasion and productivity responses lead to

overestimated indirect effects. Finally, nontax-induced indirect effects - such as the general

economic growth that preceded the reform - had a much larger effect on the distribution of

income than the tax reform.

This paper is the first to identify the distributional impact of tax-induced changes in

evasion in Russia. This, along with the distinction between reported income and actual income

inequality, makes the results especially relevant to recent debates concerning redistribution via

increased progressivity of personal income tax codes. There is one general and several narrowly

17

defined policy implications of these results. At a very general level, the results suggest that non

tax-induced changes in the economy tend to be more important for income distribution than tax

progressivity. The empirical evidence is clear that income inequality in Russia, as measured by

the Gini coefficient, began declining in the years leading up to the tax reform, and my results

show that the reform had very little effect on this downward trend. All else equal, these results

suggest that countries considering flatter tax schedules need not be deterred on the grounds of

inequality. This is consistent with the view that governments ought to focus on policies that

improve the equality of opportunity - such as education and training - rather than the equality of

outcomes as measured by income (Fields 2007).

Additionally, it is very important that a distinction be made between evasion and real

productivity effects. Since the evasion response involves shifting existing income, it represents

an artificial change in the distribution of reported net income, which leads to an overestimation

of the distributional impact of the reform. Therefore, failure to distinguish between these two

behavioral responses can result in incorrect policy advice. The results also imply that a

distinction must be made between reported and actual net income when evaluating the

distributional effects of tax reforms. However, because policymakers can only observe reported

income, the results suggest that redistributive policy prescriptions should be based solely on the

real productivity effect. This policy implication is particularly important for policymakers in

countries where tax evasion is prevalent, and possibly explains why the flat rate PIT is popular in

Eastern European countries.

Of course, the tax reform is likely to affect the returns on savings and other types of

investments and thus affect the amount of these activities in the long-run, and these changes are

likely to affect the distribution of income in important ways. For example, increased investment

18

is expected to increase economic output, which in turn will affect incomes. I ignore these second

round general equilibrium effects here and focus exclusively on the short-term partial effects of

the reform. This choice is driven primarily by the lack of information on elasticities measuring

the impact of the tax reform on savings, investments, economic growth etcetera. It would be

constructive to explore the distributional impact of these general equilibrium effects in future

work. Nonetheless, I argue that the insights gained from the partial equilibrium analysis

conducted here remains useful, especially for other countries considering the flat PIT.

References:

Alm, James; Fitzroy Lee and Sally Wallace. 2005. "How Fair? Changes in Federal Income Taxation and the Distribution of Income, 1978 to 1998." Journal of Policy Analysis and Management, 24(1): 5-22.

Altig, David and Charles T. Carlstrom. 1999. "Marginal Tax Rates and Income Inequality in a Life-Cycle Model." American Economic Review, 89(5): 1197-215.

Burns, Sarah and James Ziliak. 2012. “Identifying the elasticity of taxable income.” Mimeo, University of Kenntucky.

Duncan, Denvil and Klara Sabirianova Peter. 2012. " Unequal Inequalities: Do Progressive Taxes Reduce Income Inequality?," Institute for the Study of Labor Discussion Paper, No. 6910.

Fields, Gary. 2007. “How much should we care about changing income inequality in the course of economic growth?”Journal of Policy Modeling, 29 (4): 577–585.

Fuest, Clemens; Andreas Peichl and Thilo Schaefer. 2008. "Is a Flat Tax Reform Feasible in a Grown-up Democracy of Western Europe? A Simulation Study for Germany." International Tax and Public Finance, 15(5): 620-36.

Gorodnichenko, Yuriy; Jorge Martinez Vazquez and Klara Sabirianova Peter. 2009. "Myth and Reality of Flat Tax Reform: Micro Estimates of Tax Evasion Response and Welfare Effects in Russia." Journal of Political Economy, 117(3): 504-54.

Gorodnichenko, Yuriy; Klara Sabirianova Peter and Dmitriy Stolyarov. 2010. "Inequality and Volatility Moderation in Russia: Evidence from Micro-Level Panel Data on Consumption and Income." Review of Economic Dynamics, 13(1): 209-237.

Gramlich, Edward M.; Richard Kasten and Frank Sammartino. 1993. "Growing Inequality in the 1980s: The Role of Federal Taxes and Cash Transfers," In Uneven Tides: Rising

19

Inequality in America, ed. S. Danziger and P. Gottschalk, 225-49. New York: Russell Sage Foundation.

Heim, B. T. (2009). “The effect of recent tax changes on taxable income: Evidence from a new panel of tax returns.” Journal of Policy Analysis and Management 28 (1): 147-163

Ivanova, Anna; Michael Keen and Alexander Klemm. 2005. "The Russian 'Flat Tax' Reform." Economic Policy, 20(43): 397-435.

Johns, Andrew and Joel Slemrod (2010). “The distribution of income tax noncompliance.” National Tax Journal 63, 397-418.

Palme, Marten. 1996. “Income Distribution Effects of the Swedish 1991 Tax Reform: An Analysis of a Micro-simulation Using Generalized Kakwani Decomposition.” Journal of Policy Modeling, 18(4): 419-443.

Paulus, Alari and Andreas Peichl. 2009. "Effects of Flat Tax Reforms in Western Europe on Income Distribution and Work Incentives." Journal of Policy Modeling, 31(5): 620–636.

Poterba, James M. 2007. “Income Inequality and Income Taxation.” Journal of Policy Modeling, 29(4): 623-33.

Russia Longitudinal Monitoring Survey—Higher School of Economics. 1997. “Sample Attrition, Replenishment, And Weighting in Rounds V-VII.” Retrieved December 1, 2010 from http://www.cpc.unc.edu/projects/rlms-hse/project/samprep.

Salvatore Dominick. 2007. “Growth, international inequalities, and poverty in a globalizing world,” Journal of Policy Modeling 29 (4): 635–641.

20

Appendix

Table A1: Distributional Impact of the Flat Tax Reform: Sensitivity Analysis of Tax-Induced Behavioral Effects Parameters Contractual Earnings Consumption Income

π ε(e) ε(y) Evasion effect

Real effect

Combined effect

Evasion effect

Real effect

Combined effect

Evasion effect

Real effect

Combined effect

0.20 0.26 -0.04 5.7034 4.4585 9.5526 -0.0895 1.6982 1.5429 4.0378 3.1116 7.03240.25 0.26 -0.04 7.4251 4.7386 11.3207 -0.1003 1.6886 1.5085 5.3540 3.3180 8.47100.30 0.26 -0.04 9.2907 5.0562 13.2319 -0.0942 1.6780 1.4899 6.8227 3.5535 10.0651

0.25 0.20 -0.04 5.8393 4.7386 9.9134 -0.0907 1.6886 1.5353 4.1404 3.3180 7.32350.25 0.30 -0.04 8.4427 4.7386 12.2255 -0.1005 1.6886 1.4967 6.1493 3.3180 9.2206

0.25 0.26 0.00 7.4251 0.0000 7.4251 -0.1003 0.0000 -0.1003 5.3540 0.0000 5.35400.25 0.26 -0.10 7.4251 10.8599 16.4149 -0.1003 4.2874 3.9936 5.3540 8.0927 12.8113

Notes: Reported are percent changes in Gini coefficients. The sample is restricted to non-zero values for each variable; imputed contractual labor earnings at the individual level, and durable plus non-durable consumption and income before public transfers at the household level. Decompositions are calculated as described in the notes to Table 3.

21

Table A2: Tax-Induced Behavioral effects: Distributional impact of the flat tax reform

Levels Indirect effect (A)

Indirect effect (B)Tax Schedule 2000 2000 2000 2000 Evasio

n effectRealeffect

Total effectAdjustment None Evasion Real Both

Panel A: Household ConsumptionNon-durable 0.43

7 0.436 0.445 0.443 10.166 -0.318 1.737 1.296

Durable plus non-durable 0.495 0.495 0.50

4 0.503 9.266 -0.100 1.689 1.509

Aggregate plus imputed services from housing

0.429 0.427 0.43

7 0.435 10.869 -0.391 1.876 1.362

Non-durable plus homegrown food

0.414 0.412 0.42

1 0.419 5.254 -0.333 1.850 1.380

Panel B: Household Income

Contractual labor earnings0.47

1 0.500 0.489 0.516 10.582 6.077 3.787 9.544

Income before government transfers

0.479 0.505 0.49

5 0.520 6.563 5.354 3.318 8.471

Disposable Income0.43

5 0.461 0.451 0.477 13.265 6.050 3.745 9.591

Notes: Reported are the Gini coefficients in levels and percent changes. Indirect effect ‘A’ does not distinguish between evasion and productivity responses while indirect effect ‘B’ treats them differently. The sample is restricted to non-zero values for each variable. Aggregate refers to durable plus non-durable expenditure. Imputed services from housing are calculated as 5 percent of the current housing market value. Contractual labor earnings include imputed values for individuals who reported having a job but no income. Income before government transfers includes net private transfers, income from dividends, and interest from bank accounts. Disposable income includes government transfers. Decompositions are calculated as follows: the evasion effect is the change between the first two columns (assumes productivity response is zero), the real (productivity) effect is the change between columns one and three (assumes evasion response is zero), and the total effect is the change between columns one and four (assumes both productivity and evasion responds). Column 5 reports the tax-induced indirect effect if no distinction is made between evasion and productivity responses; the elasticity of reported gross income is used to calculate the counterfactuals in this case. Adjustments are made using the following baseline parameters: evasion elasticity 0.26, productivity elasticity -0.04, and reported gross income elasticity -0.21 (Gorodnichenko, Martinez-Vazquez, and Sabirianova Peter 2009); evasion as a share of actual income 0.25 is from Ivanova et al. (2005). All changes are in percent.

22

Table A3: Variable Description and NotesVariable Name Definition Notes

Individual IncomeIMP Imputed contractual labor earnings

per monthLabor earnings of working-age non-respondents are imputed as predicted earnings times the predicted probability of working using the full set of interactions between the four age groups (18-60) and two gender groups and controlling for urban and federal district dummies for each year separately.

Household IncomeyL Contractual labor earnings per

month = sum of IMP within each household.

y Household income before government transfers

= yL + net private transfers + financial income received last month.

“Private transfers received” include received alimonies and 11 subcategories of contributions from persons outside the household unit, including contributions from relatives, friends, charity, international organizations, etc. “Private transfers given” include alimonies paid and various contributions in money and in kind given to individuals outside the household unit (6 categories). Financial income includes dividends on stocks and interest on bank accounts.

hhyD: Disposable householdincome

= y +public transfers Public transfers include government pensions, state child benefits, stipends, unemployment benefits, and government welfare payments

Household Consumptionc Non-durable expenditures Sum of expenditures on non-durables in the last 30 days.

Non-durable items include food, alcohol, tobacco, clothing and footwear, gasoline and other fuel expenses, rents and utilities, and 15-20 subcategories of services (such as transportation, repair, health care services, education, entertainment, recreation, insurance, etc.).

cD Aggregate expenditures = c + expenditures on durables in the last 3 months / 3. Durable items include 10 subcategories such as major appliances, vehicles, furniture, entertainment equipment, etc.

This is compared with purchases of goods and services from NIPA

cD+ Aggregate = cD + imputed services from housing Imputed services from housing are calculated as 5

23

expenditures plus services from housing

percent of the current housing market value divided by 12

cH Non-durableexpenditures plusconsumption of homegrownfood

= c +consumption of home-grown food, where the last termis calculated as average monthly quantities of consumedhome-grown food items multiplied by their median price in agiven region

Median prices are determined in the same way as in yH

Source: Reproduced with permission from Gorodnichenko, Sabirianova-Peter and Stolyarov (2010)

24

Simulation Appendix

Behavioral Responses and the Distributional Effects of The Russian ‘Flat’ Tax

Denvil Duncan13

January 2013

13 School of Public and Environmental Affairs, Indiana University - Bloomington, U.S.A. and Institute for the Study of Labor, Bonn, Germany. Email: duncande@indiana.edu,

25

1. INTRODUCTIONThe purpose of this document is to outline the mechanical procedure used to determine the effect of

personal income taxes (PIT) on the distribution of income. I demonstrate how various counterfactuals of net income are calculated and how gross income is adjusted to reflect changes in evasion and productivity.

There are a number of problems that must be addressed in order to complete the analysis. One of the main ones is the fact that the Russian Longitudinal Monitoring Survey (RLMS) does not have direct measures of taxes paid, deductions, or gross income. The data set collects reported net income and reported consumption. While both are likely to suffer from underreporting, consumption is used as an approximation of true net income (Gorodnichenko, Martinez-Vazquez, and Sabirianova-Peter 2009 (GMP hereafter) and Ivanova, Keen, and Klemm 2005 (IKK hereafter)). This implies that gross income (true and reported) can be obtained by inverting the tax function. I use this inversion technique as a starting point to estimate reported gross income, which I then use to calculate counterfactual measures of net income. The limitations mentioned above also imply that analyses focusing on true income can only be done at the household level since consumption data are only available at the household level.14

I discuss net income counterfactuals that do not distinguish among behavioral effects in section 2 and show how I adjust reported gross income so that the evasion effect can be distinguished from the productivity effect. Section 3 highlights the combination of counterfactuals used to identify the different components of the change in income inequality. Section 4 describes the process of inverting the tax function while the variables used in the analysis are discussed in section 5.

2. COUNTERFACTUAL NET INCOME In this section I outline how each counterfactual net income variable is calculated. I first discuss

counterfactuals that allow me to decompose the change in inequality into its direct and indirect components.15 I then discuss how gross income can be adjusted to allow for the evasion and productivity effects. The analysis refers to years 2000 and 2001 only. However, actual implementation includes other post-reform years. I also make reference to counterfactuals at the individual (reported income) and household level (consumption based measures of true income).16

In the expressions below, Y=income, E is hidden income, T is the tax function, εy is the true income elasticity, εe is the evasion elasticity, τ is the statutory marginal tax rate, and the subscripts i, h, n, g, t, and superscript * indicate individuals, households, net, gross, time, and true, respectively.

2.1 No distinction among behavioral effects17:

a. Net income under pre-reform tax schedule with pre-reform income (A):

i. Individual: Y in 2000=Y ig2000−T 2000(Y ig 2000−Di2000 )

ii. Household: Y

hn 2000

¿ =Yhg 2000

¿ −∑i

T 2000(Y ig 2000−Di 2000 )

14 Other problems and assumptions are discussed throughout the text. For example, taxation is at the individual level, which makes household level analyses problematic. I discuss this problem in more detail later in the text.15 The direct component is due to the change in the tax rate and the indirect component is due to the change in income.16Although I do not present them here, the analysis includes reported income at the household level as well.17 Summation is within a given household.

26

b. Net income under post-reform tax schedule with post-reform income (B):

i. Individual: Y in 2001=Y ig 2001−T2001 (Y ig2001−Di2001 )

ii. Household:

Yhn2001

¿ =Yhg 2001

¿ −∑i

T 2001(Y ig 2001−Di 2001)

c. Net income under pre-reform tax schedule with post-reform income (C):

i. Individual: Y inc=Y ig 2001−T 2000 (Y ig2001−Di2000 )

ii. Household:

Yhnc

¿ =Yhg 2001

¿ −∑i

T 2000(Y ig 2001−Di 2000)

d. Net income under post-reform tax schedule with pre-reform income (D):

i. Individual: Y ind=Y ig2000−T 2001(Y ig 2000−Di2001 )

ii. Household:

Yhnd

¿ =Yhg 2000

¿ −∑i

T 2001(Y ig2000−Di 2001 )

2.2 Distinguishing among behavioral effects:

Decomposition assumes evasion takes place and is observed.18 I first define the tax-induced change in both evasion and true gross income.

a. Tax induced change in evasion is

i.ΔEht=Eht×εe×

Δτ i

τ i ⇒ Eh

' =Eh 2000(1+ε e×Δτ i

τ i)

b. Tax induced change in gross income is

i.ΔY

hgt

¿ =Yhgt

¿ ¿ ε y ¿Δτ i

τ i ⇒ Y hgt

' =Yhg 2000

¿ (1+ε y¿Δτ i

τ i)

I then calculate net income allowing each component of reported gross income to change by the tax induced amount. The calculation is done using both pre-reform and post-reform tax schedules.

There are two problems that must be addressed when conducting this analysis. First, taxes are assessed on individuals, not households. Therefore, adjustments for evasion and productivity are required at the individual level when calculating tax liability even if the analysis is at the household level. This poses a problem because the elasticities needed to make the adjustments are estimated at the household level. Second, tax liability for individual

i can be written as T i=T (Y

ig

¿ −Ei−Di) . The behavioral effects are obtained by adjusting Y

ig

¿

and Eig according

to the estimated elasticities outlined above. However, I observe neither Y

ig

¿

norEig ; I am able to estimate18 It is possible to do these calculations even if evasion is not observed. This is illustrated below.

27

Y ig=Yi g¿ −E ig .19 As such, it is not possible to directly adjust the amount of evasion or true gross income as

indicated above.

Both problems are addressed by assuming that the percentage change in the amount of evasion at the household level applies to each working member of a given household. This implies that the new level of reported income - due to the change in evasion - can be written as

Yig

' =Yig¿ −Ei(1+εe¿

Δτ i

τ i)

which then implies that the percentage change in individual reported gross income due to the change in evasion is

%ΔY ig=(Y ig' −Y ig )/Y ig=−ε h×

Δτ i

τ i× π

1−π; π=E/Y

Similarly, the percentage change in individual reported gross income due to the change in productivity is

%ΔY ig=ε y×Δτ i

τ i× 1

1−π .

Finally, the percentage change in individual reported gross income due to the change in both evasion and

productivity is

%ΔY ig=Δτ i

τ i×

ε y−εe π1−π

Although I do not knowπ , it is possible to compute the counterfactuals based on different values of π .20 With this in mind, I am able to write out the following;

1. Reported gross income adjusted for evasion is Y ig

e =Y ig+Y ig×[−εe×Δτ i

τ i× π

1−π]

2. Reported gross income adjusted for productivity is Y ig

y =Y ig+Y ig×[ε y Δτi

τ i(1−π )]

3. Reported gross income adjusted for evasion and productivity is

Y igey=Y ig+Y ig×[

Δτ i

τ i×

ε y−ε e π1−π

]

Using this same procedure, I calculate the change in true gross income (consumption) at the household level as21

Y hgy =Y hg

¿ +Y hg¿ ×ε y×

Δτ i

τ i

19 This is done by inverting the tax function. See variable definitions (Table 1) and section 5.20 GMP consider 1/3 to be a reasonable upper bound forπ . It is possible to allow π to vary between 1/5 and 1/3, for example. 21 Note that true income is only adjusted for changes in productivity. The implicit assumption here is that changes in evasion do not affect the amount of income earned except through its effect on individual tax liability. This is quite reasonable since evasion usually involves a reallocation of what is earned. Evasion may still have an indirect effect on how much is earned. For example, an increase in the ability to evade taxes may act as an incentive to increase earnings (Slemrod 2001). However, I ignore these cross effects since there are no estimates available for them.

28

Reported household gross income is adjusted similarly to individual income. The above procedure allows me to write down counterfactuals that I use to determine the size and sign of the evasion and productivity effects.22 I estimate the evasion effect, by comparing the pre-reform (year 2000) distribution of net income with the distribution of net income that would obtain if the only tax induced behavioral response to the tax reform was evasion. The productivity effect is obtained similarly, except that I assume the only response is through productivity changes. I also estimate the total behavioral effect by allowing both evasion and productivity to change simultaneously (calculations are summarized in Tables 1 and 2).

3. CHANGE IN INEQUALITY The change in inequality is determined by comparing several counterfactual measures of net income

distribution (see Table 2). The primary reason for using these counterfactuals is that they allow me to separate the total change in the distribution of income into the components of interest. This approach is widely used in the literature (eg., Alm, Lee, and Wallace 2005 and Poterba 2007).

3.1 No distinction among behavioral effects:

Panel C of Table 2 illustrates how the various counterfactual net incomes are combined to decompose the change in income inequality into its various components. The direct (tax) effect is defined as the change in income inequality that results from a change in the tax schedule holding the pre-tax distribution constant. Therefore, the direct effect can be estimated by comparing D and A (pre-reform income held constant) or B and C (post-reform income held constant). The indirect (behavioral) effect, on the other hand, is the change in the distribution of income that follows from a change in the distribution of pre-tax income with the tax schedule held constant. Again, this is estimated holding either the pre-reform tax schedule constant (compare C and A) or the post-reform tax schedule constant (compare B and D).

3.2 Decomposing the indirect (behavioral) effects:

The indirect effect obtained using the approach above included behavioral changes along many different dimensions. While some of these responses are most likely induced by the change in the tax schedule, others are totally unrelated and would have taken place even if no reform took place. The central objective of this essay is to determine how much of the indirect effect is tax-induced. For example, I ask the question, what would be the net income distribution if individuals were induced to change only the amount of income they evade? Similarly, what would be the resulting net income distribution if only tax-induced productivity responses took place? These questions are answered by using the counterfactuals in Panel B of Table 2.23

For example, the distribution of net income when evasion is the only response, E1, is compared with the one that would have obtained had there been no change in the pre-reform income, A.24 This is illustrated in panel C of Table 2 where I compare E1 with A (holding pre-reform tax schedule constant) and E2 with D (holding post-reform tax schedule constant). The productivity effect is similarly calculated by allowing productivity to be the only tax induced behavioral response and adjusting pre-reform pre-tax income accordingly. Here I compare F1 with A, which holds the pre-reform tax schedule constant, and F2 with D, which hold the post-reform tax schedule constant.

22 The adjustments all use income in year 2000 as the base. Additionally, I hold the tax schedule constant so that any change must be due to the change in income only; base calculations are done using the pre-reform tax schedule. 23 The counterfactuals are discussed in section 2.2 and summarized in Table 1.24 The net income distributions are calculated using both pre-reform (E1) and post-reform tax schedules (E2). As such, the evasion effect can be obtained by comparing E1 with A, or E2 with D. The same procedure is followed to obtain the productivity effect.

29

4. IMPLEMENTATIONIn this section I discuss the steps used to implement the analysis. The analysis requires six steps:

1. Determine the amount of deduction for each individual

2. Invert the tax function to obtain gross income

3. Adjust gross income for evasion and productivity

4. Calculate the counterfactual net incomes outlined above

5. Calculate the indices of net income distribution

6. Calculate the change in distribution

Each step is discussed in more detail below.

Step 1: Determine the amount of deduction for each individual

Deduction for the year 2000 is summarized as follows:

1. 264 rubles per month every month for which accumulated income (up to that month) is less than or equal to

20,000 rubles

2. 132 rubles per month every month for which accumulated income is less than or equal to 50,000 rubles and

greater than 20,000 rubles.

3. Zero for remaining months

Deduction for the post-reform period is summarized as follows:

1. 400 rubles per month every month for which accumulated income is less than or equal to 20,000 rubles

2. Zero for remaining months

The expressions below are based on the following assumptions:

3. Since information is only available on income earned last month, I assume that income is received evenly

throughout the year when accounting for these complex deduction rules (IKK).

4. The rule for year 2000 also applies to 1998

Let d1 equal deduction while accumulated income is less than or equal to 20,000 rubles and d2 equal deduction while accumulated income is less than or equal to 50,000 rubles but greater than 20,000 rubles. Given the rules above,

d1=¿ {20000∗12Y ig

∗(264 ) if Y ig≥20000 ¿¿¿¿

d2=¿ {(50000∗12Y ig

−20000∗12Y ig )∗(132 ) =30000∗12

Y ig∗(132) if Y ig≥50000 ¿ ¿¿¿

30

Define C=20000∗12

Y ig and simplify deductions to get:

Deductions before reform:

D=d1+d2=¿ {12∗(264 ) and C≥12 if Yig≤20000 ¿ {C∗(132 )+1584 and 5≤C<12 if 20000≤Y ig <50000 ¿ {C∗(462 ) and 2≤C<5 if 50000≤Y ig <150000 ¿¿¿¿Deductions after reform:

D=¿ {12∗400 if Y ig≤20 ,000 ¿ ¿¿¿Step 2: Invert the tax function to obtain gross income

Gross income for each individual is imputed based on the PIT tax schedule summarized in Table 3 and the deductions outlined in step 1.

Pre-reform (1998 - 2000):

Y ig=¿ {Y in−0. 13∗12∗2640 .87 if 3,168<Y ig≤20 ,000 ¿ {Y in−0 .13∗(C∗132+1584 )

0 .87 if 20,000<Y ig≤50 , 000 ¿ {Y in+(50 ,000−C∗462)∗0 . 13−50 , 000∗0 .210 .79 if 50,000<Y ig≤150 ,000 ¿ ¿¿¿

Post reform (2000-2004):

Y ig={Y in−0 .13∗C∗4000 .87

if Y ig>4 , 800

Step 3: Adjust gross income for evasion and productivity

The estimated gross income is adjusted for behavioral responses and used to determine counterfactual net income as described in section 2.2. The baseline adjustments set the evasion parameter (π) at 25 percent, evasion elasticity at 0.26, and the productivity elasticity at -0.04.25 The tax rates and change in tax rates are determined from

25 Each of these parameters is adjusted in robustness checks. Simulations allow π to be equal to 20 percent and 30 percent.

31

Table 3. I place each individual (household) into a tax bracket based on their gross income and then assign the relevant tax rate.

Step 4: Calculate the counterfactual net incomes outlined in section 2

The estimated gross income is used to determine counterfactual net income as described in section 2.1 while the adjusted gross income is used to determine counterfactual net income as described in section 2.2 and Table 1.

Step 5: Calculate inequality indices for net income

With counterfactuals of net income determined, I calculate various measures of income inequality. These include the Gini coefficient, the coefficient of variation (CV), the relative mean deviation (RMD), and variance of log. I also calculate measures of effective progressivity. All income/consumption measures are converted to December 2002 prices, and household measures are adjusted using the OECD equivalence scale. Additionally, the individual (household) level inequality indices are calculated using the RLMS individual (household) sample weights to address sample attrition and other sampling errors.26

Step 6: Calculate the change in inequality

The decomposition involves two steps. First, I identify the direct and indirect effects using the counterfactuals in panel A of Table 2. Second, I use the counterfactuals in panel B of Table 2 to decompose the indirect effect into its evasion and productivity components. This is similar in spirit to the approach taken by Poterba (2007) and Alm, Lee, and Wallace (2005). In effect, I am able to see how the distribution changes when, say, evasion changes, ceteris paribus. The changes are calculated as percentages.

Parameter values and source

Adjustments are made using the following baseline parameters: evasion elasticity 0.26, productivity elasticity -0.04, and reported gross income elasticity -0.21 (Gorodnichenko, Martinez-Vazquez, and Sabirianova Peter 2009); evasion as a share of actual income 0.25 is from Ivanova et al. (2005).

REFERENCESAlm, J., Lee, F., & Wallace, S. (2005). How Fair? Changes in Federal Income Taxation and the Distribution of

Income, 1978 to 1998. Journal of Policy Analysis and Management, 24 1, 5-22.

Gorodnichenko, Y., Martinez-Vazquez, J. and Sabirianova Peter, K. (2009). ‘Myth and Reality of Flat Tax Reform: Micro Estimates of Tax Evasion Response and Welfare Effects in Russia’, Journal of Political Economy, 117(3), pp. 504-554

Ivanova, A., Keen, M. and Klemm, A. (2005). ‘The Russian 'Flat Tax' Reform’, Economic Policy, 20(43), pp. 397-435.

Poterba, J. M. (2007). Income Inequality and Income Taxation. Journal of Policy Modeling, 29 4, 623-633.

26 The RLMS sample weights adjusts for sample design factors and deviations from the census characteristics, which implicitly address sample attrition.

32

TABLE 1: DERIVATION OF COUNTERFACTUAL MEASURES OF NET INCOME DUE TO BEHAVIORAL RESPONSES

Level of analysis Evasion effect Productivity effect Combined effect

E1 F1G1

Individual Yin

e =Yig

e −T2000 (Y ige −D2000 )

Yin

y =Y igy −T 2000(Y ig

y −D2000 ) Yin

ye =Y igye−T 2000(Y ig

ye−D2000 )

Household Y hn¿ e=Y hg 2000

¿ −∑i

T2000 (Y ige −D2000 ) Y hn

¿ y=Y hg¿ y−∑

iT 2000 (Y ig

y −D2000 ) Y hn¿ ye=Y hg

¿ y−∑i

T 2000(Y igye−D2000 )

E2 F2 G2Individual Y

in

e =Yig

e −T2001 (Y ige −D2001 ) Y

in

y =Y igy −T 2001(Y ig

y −D2001 ) Yin

ye =Y igye−T 2001(Y ig

ye−D2001)

Household Y hn¿ e=Y hg 2000

¿ −∑i

T2001 (Y ige −D2001) Y hn

¿ y=Y hg¿ y−∑

iT 2001(Y ig

y −D2001) Y hn¿ ye=Y hg

¿ y−∑i

T 2001(Y igye−D2001)

Notes: The top panel (E1, F1, and G1) uses the pre-reform tax schedule to calculate net income while the bottom panel (E2, F2, and G2) uses the post-reform tax schedule. Superscripts e and y indicate that income has been adjusted for evasion and productivity, respectively. Consumption based measures of household income is adjusted for productivity only. Household tax liability is first calculated at the individual level and then summed over individuals within the household.

33

TABLE 2: SUMMARY OF COUNTERFACTUAL MEASURES OF NET INCOME

Panel ATax schedule Pre-reform Pre-reform Post reform Post reform - -Income year Pre-reform Post reform Pre-reform Post reform - -

(YN)A C D B - -

Panel BTax schedule Pre-reform Pre-reform Pre-reform Post reform Post reform Post reform

Income Adjust E Adjust Y Adjust Y&E Adjust E Adjust Y Adjust Y&E(YN) E1 F1 G1 E2 F2 G2

Panel CTax Behavior Tax and

BehaviorEvasion Productivity Productivity and

evasionD-A C-A B-A E1-A F1-A G1-AB-C B-D E2-D F2-D G2-D

Note: (YN) is a summary measure of net income distribution (GINI, income share of deciles, coefficient of variation etc.). Counterfactuals in Panel A are used to separate the direct (tax) effect from the indirect effect while those in Panel B are use to decompose the indirect (behavioral) effect into two components (evasion and productivity); these are illustrated in Panel C. For example, the direct (tax) effect is calculated by holding the pre-tax distribution of income constant while allowing the tax schedule to change. This can be done by comparing D with A (pre-reform income held constant) or B with C (post-reform income held constant). The counterfactuals E1 to G2 use income in year 2000 as the base; E1 through G1 uses the pre-reform tax schedule to calculate net income while E2 to G2 uses the post-reform tax schedule.

34

TABLE 3: THE PIT RATE STRUCTURE BEFORE AND AFTER REFORMBefore Reform (2000) After Reform (2001-2004)

Taxable Income1 Marginal Rate Taxable Income1 Marginal Rate

Below 3,168 0 Below 4,800 0

3,168 to 50,000 13 Above 4,800 13

50,000 to 150,000 21

Above 150,000 31

Note: (source: IKK). Marginal rates include the 1% payroll tax.

35