The effect of low corporate income tax on social

security evasion

Boryana Madzharova ∗

Abstract

We develop a model in which employers and employees cooperate in

declaring lower wages to the tax authorities in order to evade social se-

curity contributions. A lower wage on paper translates into higher cor-

porate profits. Therefore, a low corporate income tax relative to the so-

cial security burden can further enhance contribution evasion, especially

in economies with weak tax administration. We argue that while the

widespread policy of cutting profit tax can have positive effect on taxable

income through decreasing profit evasion, it can trigger more evasion in

another tax base–social security. Using firm-level panel data for Bulgaria

where the problem of contribution evasion is especially severe, we find that

firms facing greater disparity between the contribution rate and the cor-

porate tax rate are more responsive in adjusting their labour expenditure

compared to firms for which these rates are similar.

∗Center for Economic Research and Graduate Education - Economic Institute(CERGE-EI), Politickych veznu 7, 111 21 Praha 1, Czech Republic, E-mail:Boryana.Madzharova@cerge-ei.cz.

1

1 Introduction

The study of people’s responsiveness to changes in the parameters

of a tax system is vital to understanding and predicting revenue

collections. Inevitably, the tax structure itself induces behavioral

responses such as avoidance and evasion. The determinants of tax

evasion and tax compliance have been of particular theoretical and

econometric interest to economists over the years. In 1972, in a sem-

inal contribution, Allingham and Sandmo derived the theoretical

framework, which most tax evasion studies build upon and extend.

Naturally, the fields of major interest have been the personal income

tax (PIT) and the corporate income tax (CIT) due to their impor-

tance to many countries (e.g. Feinstein (1991); Feldstein (1999);

Joulfaian (2000); Chen and Chu (2005)). Surprisingly, social secu-

rity evasion has received little theoretical or empirical analysis. The

issue, nevertheless, is widely discussed in spheres other than eco-

nomics (McGillivray (2002); S.Mitchell et al. (1999)). This lack of

interest may be due to the fact that employers in the United States

are not required to sponsor pension plans and that healthcare is

paid. Employees, on the other hand, at least in a defined contri-

bution system, have a vested interest in contributing. However,

many countries have PAYGO social security systems or some blend

of defined benefit and defined contributions, obliging the employer

to contribute towards the pension of his workers. In this context,

evasion of social security contributions by firms can have serious

implications for current pensions and all public services funded by

these contributions (healthcare, unemployment benefits, etc.). In

addition, social security evasion can spillover to other tax bases if

achieved through the underreporting of workers’ wages. A few pa-

pers investigate taxpayers’ behavior that affects revenue from several

taxes with the main focus being on income shifting between the cor-

porate and personal tax bases (Gordon and Slemrod (1998); Fuest

2

and Weichenrieder (2001)).

Yaniv (1988) considers the underreporting of wages by employers

under a withholding tax system, in which a wage-earner tax liabil-

ity is deducted at source. As far as we are aware, this is the only

theoretical analysis on the subject. Yaniv (1988) derives necessary

conditions for evasion and analyzes employers’ incentives to engage

in fraud under alternative penalty schemes. A recent empirical con-

tribution by Nyland et al. (2006) uses Chinese firm-level data to

study firms’ characteristics that can influence the choice to evade.

Their major finding is that big Chinese firms tend not to pay full

contributions.

Our paper proposes a model of cooperative employer-employee

social security evasion through the understatement of wages. We

explain why cooperation between a firm and its workers is likely

in the case of Bulgaria where hidden labor expenditures amount to

more than 2 billion Euro per year as estimated by the Ministry of

Labor and Social Policy. Although we explain social security evasion

in the context of Bulgaria and the specificities of its economy, such

analysis is applicable to all countries with a weak tax administra-

tion, significant social security rates relative to the corporate burden

and strong public mistrust in the government. We show that low-

ering the CIT rate, while keeping the social security burden high

can worsen social security evasion. The reason is that by declar-

ing lower wages on paper, firms necessarily declare higher profits as

well. Therefore, a substantial reduction of the CIT makes contri-

butions evasion cheaper. We use firm level data to empirically test

the hypothesis that firms facing substantial difference between so-

cial security rate and CIT are more flexible in adjusting their labour

expenditures compared to companies for which the rates are similar.

The paper is structured as follows: section 2 provides details on

the institutional environment, in which Bulgarian taxpayers make

3

decisions to evade or not. Section 3 presents the model. In Section

4 we use panel data to test the main hypothesis generated by our

model. Section 5 concludes.

2 Institutional context

Currently, there is no theoretical framework comparing the costs and

benefits of the welfare state to firms. In particular, what benefits, if

any, do firms derive from social policies? The only answer offered in

the literature thus far is by Isabela Mares who argues that a policy of

social protection induces workers to acquire specific skills and thus,

protects employers’ investment in their workers’ skills (Mares, 2003).

While this argument may hold for high-skill dependent firms, for the

majority of companies the burden of social insurance combined with

the perception of massive evasion by other employers in the economy

is likely to outweigh any other benefits. In that sense, contributions

underpayment, although illegal, is not irrational. The motivation

of employees for participating in evasion schemes, however, requires

more detailed discussion.

2.1 Why employees cooperate

By agreeing to work for a lower gross wage on paper, a worker

puts more money in his pocket today, but forgoes higher pension

in the future. If put in context, this otherwise ”myopic” behavior

has its logic. First and foremost, the complexity of the Bulgarian

pension system makes it almost impossible to obtain an estimate

of the expected future pension. Worse, frequent changes in rules

and the pension formula create a sense of uncertainty even in the

most diligent taxpayer. Pension uncertainty aside, Table 1 shows

that the average lifespan of the population is very close to the full

retirement age, especially when it comes to men. For example, a

4

Bulgarian man retiring in 2005 at the age of 63 is going to live six

more years on average, in which he is going to receive a pension.

This fact alone creates strong incentives for intertemporal shifting

of income from the future to the present. The general low level of

wages further enhances preferences for present consumption.

Regarding healthcare, lower contributions do not mean that a

person will not be treated. In a comprehensive survey conducted in

1997, Balabanova and McKee (2002) find that ”under-the-counter”

payments and various gifts before or after treatment have become

entrenched into the health sector. People pay to ensure a better

service and access to specialized facilities. The authors note that

informal payments are ”universal” for surgery, childbirth and any

complicated medial procedures. In other words, the healthcare sys-

tem has long ceased to be ”free” with people overcoming its deficien-

cies by out-of-pocket payments. It is not surprising, therefore that

workers would prefer money at hand to contributions given their

expectations to pay for quality treatment. Since free medical care

is a public good, an increased uncertainty about the contributions

of others to the public good is likely to lower an individual’s own

contributions as concluded by Sandler et al. (1987) and Alm et al.

(1992). And when free-riders are not easily punished, people re-

sort to evasion (in our model through their consent to jointly evade

contributions with the employer).

2.2 Overall attitude towards government officials: Are tax

rates irrelevant?

Undoubtedly tax system parameters like tax rates, penalties and

government’s enforcement capacity influence taxpayers’ behavior.

There is, however, one factor, which is difficult to measure but in-

variably affects tax compliance–tax morale. The research on this

topic has surged in recent years with particular emphasis on how dif-

5

ferent social norms, values and attitudes lead to various willingness

to pay taxes across countries (e.g. Torgler (2002); Alm and Torgler

(2006); Cummings et al. (2005)). We argue that the tax morale in

Bulgaria is particularly low because of perceptions of widespread

corruption in the government and the tax administration. A 2004

survey by Coalition 2000 showed that 64.3% of people believed that

almost all or most politicians and political party leaders are involved

in corruption. These perceptions are not unfounded. Only in 2008

and the beginning of 2009 the media has covered at least three ma-

jor scandals that, if anything, entrenched perceptions of ”universal”

corruption.1 Ignoring people’s impression of how their taxes are used

and the morale of the officials who make use of it, would mean ne-

glecting a major driver of tax compliance. Given the economic and

political environment in which taxpayers make decisions to evade or

not, it is not unreasonable to ask if there is a point beyond which

the tax structure does not influence choice.

3 Model

3.1 Perfect enforcement

Below we present a model in which the firm underreports its wage

bill having the consent of its employees and shares the benefits of

evasion with them. The firm simultaneously decides whether to

overpay corporate profits, only part of the increase in profit, or none

of it. For now we assume that the government has the capacity to

detect and punish contribution evasion. Later on, we relax this

assumption.

1Some of these scandals are the European Union’s temporary suspension of millions ofEuros under the pre-accession programs PHARE and SAPARD because of ”mismanagement”and ”inadequate control” of funds. In July a report by the European Commission’s anti-fraudoffice implicating major government officials in having ties to organized crime, leaked to themedia causing anger and frustration among the public. And in the beginning of 2009, thegeneral director of the National Revenue Agency responsible for the collection of VAT, socialsecurity and CIT among others, quit after allegations of syphoning millions of VAT.

6

Let the true profit of the firm be πreal = Y (L) − wRL − wRLts,where wR is the gross wage paid to the employee in the absence

of fraud, L is the number of workers and ts is the social security

rate faced by the employer. The value of output Y (L) is produced

by labour input only, where Y ′(L) > 0 and Y ′′(L) < 0. Denote

te to be the social security rate for the worker and tp the personal

income tax rate. A non-evading firm pays wR(1 + ts) per employee,

while a worker’s after-tax earnings are (1− tp)(1− te)wR. If the firm

underreports wR by an amount u, the gross wage for tax purposes

becomes wR− u. This type of evasion generates benefits from three

different channels: contributions payable by the firm decrease by

utsL, employees’ contributions fall by uteL and finally PIT revenue

goes down by tp(1− te)uL. According to the law, if a worker derives

his income solely from employment, he is not obliged to file a tax

return. It is the employer’s responsibility to withhold the personal

income tax and remit it to the tax collecting agency. At its minimum

no PIT is remitted. This happens when wR− u = wmin. Until 2007

when Bulgaria introduced a flat tax PIT, wmin was not taxed.2

An important question is how these gains are redistributed be-

tween the employer and the employee. Certainly contributions eva-

sion is not possible without the cooperation of the worker. Before a

person is hired, she needs to sign a labor contract stipulating her ex-

act gross remuneration to be wR−u, with the mutual understanding

that she will actually receive wR. The worker’s net wage increases

by tpu + teuL(1 − tp), i.e. we assume that the savings realized due

to evaded employee’s contributions go back to the employee in the

form of higher compensation. Since what the worker gets is not en-

forced by contract, he takes a risk (trusting the employer to deliver

on their non-official agreement). We assume that the employee is

2An alternative mechanism of evasion is to hire workers without a labor contract, i.e.understate the true employment level. In this case, the employer may choose to report thetrue wage of the people he employs legally, while evading all contributions for the illegal ones.Yaniv (1993) develops such model.

7

neither compensated for this risk, nor for expected lower pension

in the future. The diagram below summarizes the savings for both

parties:

No evasion : wR(ts + 1)︸ ︷︷ ︸cost to employer

←− wR −→ (1− tp)(1− te)wR︸ ︷︷ ︸no-evasion net wage

Evasion: (wR − u)(1 + ts)︸ ︷︷ ︸cost to employer

←− wR − u︸ ︷︷ ︸authorities

−→ (1− tp)(1− te)(wR − u)︸ ︷︷ ︸official net wage

(wR − u)(ts + 1)︸ ︷︷ ︸cost to employer

←− wR︸︷︷︸real

−→ wR − (wR − u)(te − tp(1− te))︸ ︷︷ ︸true net wage

The new profit of the firm declared to the authorities becomes

πreported = Y (L)− (wR−u)L− (wR−u)Lts because the firm cannot

expense its true wage bill. The real profit as a result of evasion

increases to

πreal,evasion = Y (L)− wRL− (wR − u)Lts > πreal (1)

The difference πreported − πreal = (1 + ts)uL constitutes an evasion-

driven increase in profit that is taxed at the corporate tax rate tc.

It is at this point that social security evasion creates incentives for

further evasion in the corporate income tax base. Knowing by how

much his profits increase on paper as a result of wage underreporting,

a firm can choose to hide some of its sales so that it does not pay

corporate income tax in excess of its true profits.3 Suppose the firm

chooses to eliminate some of the increase in its profits by subtracting

a fraction φuL(1+ ts) from its taxable income. Thus, if φ = 0, there

is no attempt to reduce profit to its true level and part of the losses

in social security are mitigated by more revenue in the corporate

tax base. Easily, one can conclude that higher corporate revenues3An accountant we consulted extensively for this paper noted that some firms may de-

liberately choose to overpay CIT in a certain year, especially if they have been understatingtaxable income when facing higher rates in previous years. Regarding strategies for loweringtheir sales, firms usually resort to not issuing an invoice so that a cash transaction cannot betraceable.

8

result from lower CIT rate and less evasion, whereas in reality this

rise will be at least partially driven by more evasion in another tax

base. Second, if φ = 1, the cost of social security evasion is totally

eliminated. We do not consider the case when φ > 1. In other

words, we restrict the firm to not evade corporate tax for its own

sake but only in conjunction with social security.4 If the firm engages

in evasion in both bases, the total tax evaded is:

E = (ts + te + tp(1− te))uL︸ ︷︷ ︸PIT+ total social security

+ tcφ(uL+ tsuL)︸ ︷︷ ︸corporate tax

. (2)

The firm’s behavior is constrained by probability of audit and

severe penalties that make evasion costly. Let the probability of

audit be p̄ = p1(φ(uL + tsuL)) + p2( uwR ). p1 is the probability of

being caught for cheating at the corporate tax base. We follow

Slemrod and Yitzhaki (2000) and assume an endogenous probability

of detection that is an increasing function of evaded income in the

corporate tax base. p2, on the other hand, is a function of the ratio of

amount of wage underreported to total wage. A firm that is paying

contributions on minimum wages is more likely to attract attention

than a firm paying the average wage for the economy. If u = 0,

p1 = p2 = 0, which captures our previous assumption that there

will be no corporate tax evasion without social security evasion.

The firm is risk-neutral. Its after-tax profit if not detected is:

πnd = Y (L)− wRL− (wR − u)tsL

− tc[Y (L)− (wR − u)L(1 + ts)− φuL(1 + ts)] (3)

The penalty scheme in this model environment is extremely com-

plicated. First of all, since we have allowed for a mechanism of de-

tection for social security evasion, we further assume that an audit

4If a firm reports zero taxable income, then it eliminates the cost of social security evasionand the corporate income tax, although it faces higher probability of audit.

9

performed for corporate tax evasion will expose wage understate-

ment and vice versa. We explore the consequences of authorities’

inability to expose security evasion later on. An important question

in this context is whether the firm will be reimbursed for its overpay-

ment of profit, not reimbursed, or penalized for engaging in evasion

in general. In the first two cases–full or partial reimbursement and

no reimbursement–the firm faces no penalty for manipulating profits

provided that it is triggered by social security evasion. Therefore, it

pays off to set φ = 1. When not detected, the firm will have elim-

inated the complete cost of its social security fraud. If detected, it

will be punished for social security and personal tax evasion, but not

corporate profit understatement. Thus, even though the profit tax

is overpaid, it makes sense not to reimburse the firm, but to pun-

ish it with a fraction of the tax evaded. The penalty under these

conditions becomes:

P = λ1(ts + te + tp(1− te))uL+ λ2tcφ(uL+ tsuL) (4)

where λ1 > 1 and λ2 < 1.5 The profit of a firm caught cheating is:

πd = Y (L)− wRL− (wR − u)tsL

− tc[Y (L)− (wR − u)L(1 + ts)]− λ1(ts + te + tp(1− te))uL

− λ2tcφ(uL+ tsuL)

= πnd − P (5)

The firm’s problem is to choose L∗, φ∗, and u∗ to maximize expected

profit:

E(π) = (1− p̄)πnd + p̄πd, (6)

5The assumption behind this penalty structure is that the firm bears full responsibility forevaded employees’ contributions. This possibility is not accounted for by the law, so for nowwe will ignore the fact that workers cooperate willingly.

10

The first order conditions for this problem are:

∂E(π)

∂φ= tc − p′1P − (p1 + p2)λ2tc = 0 (7)

∂E(π)

∂u= ts − tc(1 + ts)− (p1 + p2)λ1(ts + te + tp(1− te))

− p′21

wRP

L= 0 (8)

∂E(π)

∂L= (1− tc)(Y ′(L)− wR(1 + ts)) + p′2

u

wRP

L= 0 (9)

From (8) we see that the marginal benefit of underreporting wage

by one Euro, ts, equals the marginal cost of overreporting profit at

the corporate tax base plus the expected penalty. We thus obtain

the necessary condition for social security evasion to be

ts > tc(1 + ts). (10)

(p1 +p2)λ1(ts+te+tp(1− te) is the expected penalty per Euro of un-

derstated wage and depends on the probability of detection for both

social security and corporate tax evasion because the higher the u,

the more sales need to be manipulated. The same logic applies to

expected penalty in eq. (7). The additional term, p′21wR

PL

captures

the fact that an audit for social security fraud can uncover corpo-

rate profit interference. On the other hand, the marginal benefit of

sheltering some of the increased profit, tcφL(1+ ts) cancels out after

we plug in (7) into the second FOC to obtain (8). Concerning the

optimal number of workers, wage underreporting does distort the

choice upwards since the firm is able to decrease the marginal cost

per worker by p′2uwR

PL

.6

Of particular interest is the relationship between wage under-

6We have used the fact that ts − tc(1 + ts)− (p1 + p2)λ1(ts + te + tp(1− te) = p′21

wRPL.

11

statement u and the CIT rate tc. Using (8) we obtain:

du

dtc= −

(1 + ts)(1 + p′21wRλ2φu)

(p′1φL(1 + ts) + p′21wR )λ1(ts + te + tp(1− te))− p′′2 1

(wR)2PuL

< 0

(11)

Eq.(10) shows that a decrease in tc stimulates contributions eva-

sion by decreasing the marginal cost of reporting more profit and

the expected penalty. dφdtc

< 0 as well since more social security

evasion necessitates hiding higher portion of increased book prof-

its. Thus, lower CIT rate can increase taxable income in the econ-

omy through two separate channels: First, lower rate means less

corporate tax evasion; second, lower rate triggers more social secu-

rity evasion with at least some firms overreporting corporate profits.

Therefore, unless the government finds a way to decrease the con-

tribution rates, the policy of lowering the corporate tax burden can

backfire through more evasion in another base. An interesting and

empirically testable implication of this result is that firms facing

lower CIT rate have a bigger incentive to engage in social security

fraud than firms for which the CIT rate is close to their contribu-

tion rate provided that the necessary condition for evasion (10) is

satisfied.

As expected7:

du

dts=

1− tc − (p1 + p2)λ1 − p′1φuLλ1ts − p′2 1wRλ1u

(p′1φL(1 + ts) + p′21wR )λ1(ts + te + tp(1− te))− p′′2 1

(wR)2PuL

> 0

(12)

3.2 Imperfect enforcement

The model we presented in the previous section relied on the as-

sumption that social security evasion can in fact be detected and

7We have plugged in (7) and (8) into the numerator to show that it is positive.

12

punished. In reality, tax authorities have limited capacity for the

conduct of audits, especially if evasion is widespread. Furthermore,

when the incentives of the employer and his workers are aligned so

that they cooperate, it is extremely difficult to prove underpayment

of contributions. In what follows, we consider the case when the

government is not able to enforce social security compliance.

What can be ascertained by an audit of a firm suspected in social

security evasion? As mentioned above, the labour contract signed

by the worker states wR − u as the true gross wage. Besides, once

employed, a worker has to sign monthly payslips, which again list

the amount of the salary, wR − u. As a consequence, a person

who is not willing to participate in an evasion scheme, will simply

not find employment with a company that underpays its workers’

social security. Provided that an employee decided to become a

whistle-blower after being hired, her signature on the contract and

the payslips will officially invalidate her claim. In other words, the

only thing an auditor can show in the presence of social security

evasion is that the employees of a given company work for lower

wages (usually the minimum wage) than the market wages for their

respective professions. This fraudulent behavior is so endemic that

in 2004 the Bulgarian government was forced to introduce so-called

”minimum social security thresholds” for each industry and type of

occupation, on which employers are obliged to pay social security.

Since their implementation, thresholds have been raised annually

with the increase reflecting the government’s perception of what the

true salaries in the economy are (Slavova et al., 2007). This policy is

a clear proof that the tax administration cannot combat or prevent

social security evasion and has attempted to mechanically control

contributions rather than rely on agents’ discretion. Thus, the gov-

ernment compensates for the virtually zero probability of detecting

social security evasion by controlling wR−u instead. A survey of the

13

structure of wages performed by the Bulgarian National Statistical

Institute in 2006 showed that out of 2.2 million labor force, 20%

earned the minimum monthly untaxable income, while 50% of all

employed worked for twice the minimum wage or less. The situation

is worse for the years before the implementation of the thresholds.

The implications of social security evasion are staggering. The

underreporting of true wages by a single economic agent, the firm,

spills over to three bases: the social security, the personal income

tax, and possibly the corporate tax base. Besides significant tax

revenue losses, additional serious distortion is the disproportionate

tax burden born by honest businesses who become less competitive

vis-a-vis their evading competitors generating pressure to engage in

fraud. Similarly, a person who pays taxes on his real wage con-

tributes more to current pensions relative to someone with the same

real wage, but minimum wage on paper. The consequences are low

pensions, barely sustainable PAYGO and corruption in healthcare

among others.

If the government cannot enforce social security compliance, the

probability of detection reduces to p(φuL(1 + ts)). Thus, if a firm

is caught cheating, ironically it will be for the overpayment of cor-

porate tax, but from an auditor’s perspective, there is profit tax

evasion. The penalty decreases too:

tcλφuL(1 + ts), (13)

where this time λ > 1 (λ2 < 1 under perfect enforcement).

The penalty structure assumed here, while corresponding to the

penalties specified by law, may not reflect the practice.8 In partic-

ular, in the event that a firm is caught cheating, the auditor who

prepares the penalty report may be induced to accept a bribe, usu-

8In the case that a firm is caught evading taxes intentionally, it owes the evaded tax,interest on it as well as a fixed fine. This type of evasion is also considered a criminal offense.If evasion is unintentional, however, only the tax and interest are paid.

14

ally less than the amount of the punishment itself. If the auditor

is not susceptible to bribery, the firm can appeal his decision in a

court of law and succeed in paying off the judge, although this sce-

nario is much more complicated and risky. An evading company

certainly takes these possibilities into consideration when evaluat-

ing the risks it is exposed to. According to a survey conducted by

VitoshaResearch (2004), 43% of the general public and 51% of busi-

nesses believed that tax officers are involved in corrupt activities.

In a different survey, the public ranked judges higher than tax offi-

cers in terms of corruptibility Coalition2000 (2005). Even if public

perceptions are incorrect, businesses act upon them. While we will

not account for these prospects in our model, it is important to

bear in mind that the penalty structure may not provide the right

incentives to prevent tax fraud and that the problem of evasion is

manyfold. Chander (1992), for instance, develops a game theoretic

model with corrupt tax administration showing that an increase in

the tax rate can lead to more bribing of auditors and hence, less

revenue collected.

Given that the firm takes no risks when understating wages, its

decision will be entirely based on whether ts > tc(1 + ts) holds

or not. If the condition is satisfied, it pays off to set wR − u =

wmin.9 An auditor can only verify πreported but not πreal,evasion. As a

consequence, πd changes to:

πd1 = Y (L)− wRL− wmintsL

− tc[Y (L)− wminL(1 + ts)]− (λ− φ)tcuL(1 + ts) (14)

The firm is then choosing L and φ to maximize expected profit:

E(π)1 = p(V )πd1 + (1− p(V ))πnd, (15)

9To explain why there are enterprises that pay contributions on true wages, we will haveto include tax morale in the form of cost (disutility) of evading taxes that will serve as adeterrent to full evasion.

15

where V = φuL(1 + ts). The first order conditions for this problem

are:

∂E(π)1

∂φ= 1− λ[p′(V )V + p(V )] = 0 (16)

∂E(π)1

∂L= (1− tc)(Y ′(L)− wR − wmin(ts − tc(1 + ts)) = 0(17)

Because (1− tc)(1 + ts)wR > wR +wmin(ts− tc(1 + ts)), again, more

labor is hired compared to the no-evasion optimum. A decrease of

tc in this case will not have any effect on u as the firm has already

set it at its maximum possible value but it will lead to a higher φ

since the expected penalty for corporate evasion goes down.

4 Empirical investigation

The model presented in the previous section generates a testable

hypothesis: do firms that face a corporate tax rate close to the

social security rate evade less social security as compared to firms

for which the difference between the two rates is significant. For-

tunately, the Bulgarian tax code until 2002 differentiates between

firms based on their taxable income. Firms with taxable income

above 25,000 Euro face a higher rate than those below this thresh-

old. Table 2 summarizes the statutory CIT rates for companies for

the period 1997-2007. Starting in 2003, a single rate was introduced

for all firms irrespective of profitability. As it is clear from the ta-

ble, the government has embarked upon significant corporate tax

cuts since the beginning of 1998, with the difference between the

top statutory rate in 1997 and the rate in 2007 being 30.2 percent-

age points. Column 3 shows the CIT revenue from non-financial

enterprises as a percent of GDP. While there are minor fluctuations,

revenue collections have remained remarkably stable, a fact, which

the government attributed to less evasion and avoidance of the profit

16

tax.

The situation is rather different when it comes to the social se-

curity burden. The contribution rate faced by employees has been

consistently high up until 2006, although the share ratio between

the firm and the worker has been changing in favor of employers

and is going to be 50:50 from the beginning of 2010. In 2004 the

government introduced the minimum social security thresholds as

a means of controlling contributions since it was unable to com-

bat the endemic evasion of social insurance. As mentioned above,

these thresholds as well as the minimum wage have been increas-

ing steadily over the years so that the ostensibly lower burden on

employers (contribution rate of 24%) from 2006 onwards is ques-

tionable. Table 3 shows the rates faced by employers and employees

from 1997 to 2007 and well as the share ratio. For all years the so-

cial security rate is above the CIT rate for firms with profits above

25,000 Euro, and especially so for the firms below the taxable in-

come threshold. This difference has become particularly pronounced

since 2003.

4.1 Data description

We use firm-level data for Bulgaria from the AMADEUS dataset

provided by Bureau Van Dijk, a European electronic publishing firm.

The data is a panel of firms’ main financial statements’ variables and

contains more than 800,000 firm year observations for our period of

interest 1997-2006. Unfortunately, two thirds of these firms have

missing values for almost all entries in their income statement, so

we end up with 258,000 firm year observations. Firm coverage is es-

pecially good from 1999 to 2004, with fewer firms observed for 1997-

1998 and particularly 2006. Our main variables of interest are the

firm’s total cost of employment, which includes salaries and social

security expenses, taxable income (TI) according to which we will

17

assign each firm into its respective tax bracket, and the number of

employees. Other characteristics of interest are firm size, measured

by number of employees and total assets, degree of indebtedness,

which we capture by the amount of current liabilities, the number

of internal auditors and earnings before interest and tax.

Since we want to differentiate between firms based on their TI,

we first need to recover TI from the data. Once companies calculate

their book profit, they have to add and deduct all items specified by

law in order to obtain their taxable income. As a consequence, our

data contains many firms who have zero or negative book profit but

have paid positive tax. The converse is also true– some firms with

positive profits for a given year, pay no tax. Therefore, book profit

can neither be used in the place of TI, nor is it a good indicator of

it. One peculiarity and major advantage of the Bulgarian tax code

is that the tax liability stated on the firm’s financial statement is

the same as in the firm’s tax return, which is not publicly available.

We are, therefore, able to recover TI by dividing the tax liability by

the respective tax rate.10

One major difficulty with this exercise is that before 2002 we do

not actually know the rate faced by the firm in order to obtain its TI.

To cope with this obstacle, for each year we divide the tax liability

by the lower tax rate and obtain TI1. If TI1 is greater than the

threshold value for taxable income, which we denote by V, then we

assign the firm into the high tax bracket. Similarly, the tax liability

is divided by the high tax rate to obtain TI2. If TI2 < V , then the

firm belongs to the low tax bracket. Inevitably there are firms, for

which TI1 < V and TI2 > V and we are not able to assign them into

a given bracket, so we drop them from the analysis at the expense

of losing 1180 unique firms for 1997-2002. Although we manage to

allocate firms into their tax brackets within a given year, we do not

10Hanlon (2003) provides an excellent analysis of the caveats in inferring taxable incomefrom firms’ financial statements.

18

know to what extent firms will self-select into a particular bracket

across years. We discuss how to deal with this problem later.

4.2 Empirical strategy

To differentiate between the responses of the two categories of firms

that are of interest, we employ the following specification:

ln(costemployeesit) = αi + βln(tsttcit

) + β1ln(tsttcit

)H

+ X ′itΓ + εit, (18)

where tst is the social security tax rate, which is not firm specific,

tcit is the corporate tax rate faced by firm i in year t and the X’s

are other controls that can vary by firm-year (such as total assets,

current liability) or just by year (such as year dummies). H is a

dummy, which equals one if the firm is in the high tax bracket. αi

is a firm-specific fixed effect. We control for changes in the number

of employees by adding interaction terms of number of employees

and the year dummies. We expect that firms who are in the low

tax bracket and for whom the ratio tst

tcitis significantly greater than

one to be more responsive to changes in tst

tcitcompared to firms that

are in the high tax bracket. If tst

tcitincreases, due to an increase in

tst or a decrease in tcit, we expect firms to adjust their reported

labor expenditure downwards. In other words, β < 0 and β1 < 0

or β1 close to zero but β > β1 in absolute value. To see if our

results are not driven by changes in the minimum wage, we calculate

the wage bill for each firm as if it insured all its employees on the

minimum wage for a given year and subtract it from the reported

labor expenditures. For some firms this difference is negative, but

this may be due to employment of part-time workers.In that case, we

simply use the reported wage bill. By focusing on labor expenditure

above the mandatory minimum, we estimate the effect of tst

tcitonly

19

on that part of labor cost, which the firm is able to manipulate by

underreporting the wage.

Since current year level of labor expenditure is likely to be af-

fected by last year’s level, we estimate the above specification by

both simple fixed effects and Arellano-Bond linear dynamic panel-

data estimation, in which we include one lag of the dependent vari-

able. A limitation of Arellano-Bond’s GMM estimator is that it

requires that firms’ idiosyncratic errors are not autocorrelated. In

addition, we show that potential bracket creep is not an issue by re-

peating the estimation with firms that have never switched between

brackets from 1997 to 2002 (to be added).

Table 4 lists the results for total labor expenditure as the depen-

dent variable. As expected, declared cost of employment decreases

by 1.87% for each percent increase in the tst

tcitratio for the firms that

are in the low corporate tax bracket in the fixed effects specification.

This elasticity is much smaller and insignificant for the firms facing

higher CIT. This result, however changes once we add one lag of

the dependent variable and estimate the dynamic model. Then the

coefficient for the H group becomes significantly different from zero.

(-.483), while the elasticity of the low-bracket group remains nega-

tive and greater than one. The fact that both types of firms exhibit

negative response to tst

tcitimplies that both engage in understating

their wage bill but the low-bracket group is much more flexible in

doing it. Apart from being suggested by the model, this result is also

intuitively appealing. In case of a suspected fraud, larger firms face

higher probability of audit since more revenue can be recovered. Fur-

ther, the employer-employee relationship is much more impersonal

in big firms than in smaller ones, which may suggest coordination

failure in planning evasion. In Table 5 we have subtracted the mini-

mum wage bill from the reported labor costs in order to concentrate

only on expenditure above the mandatory minimum. The results do

20

not change for the firms in the low-bracket group but the elasticity

of the H firms is almost zero and insignificant for both the fixed

effects and Arellano-Bond specifications.

5 Tax Policy Implications

In the last decade many developing and developed economies have

resorted to cutting their CIT rates in order to attract foreign direct

investment and stimulate businesses at home. While such policy

clearly generates incentives towards the honest disclosure of corpo-

rate profits, it would be hasty to consider its effects in isolation of

other tax bases within the economy. The main goal of our paper is to

point out at the risk that too low CIT can exacerbate social security

evasion if the contribution burden on employers is significant. With

the collaboration of employees, such fraudulent behavior becomes

virtually undetectable and as a consequence it may be more benefi-

cial for a firm to overpay profit tax than to pay its full contributions

expense. Thus, unless the government is able to somehow balance

the corporate and social security burden, such possibility certainly

exists and should not be ignored. Given Europe’s rapidly aging pop-

ulation and significant reliance on defined benefit, it is unlikely that

the social security burden on employers will decrease anytime soon.

Governments should, therefore, carefully consider not only the pros

of low CIT rate, but the cons as well.

21

References

Allingham, M., Sandmo, A., 1972. Income tax evasion: A theoretical

analysis. Journal of Public Economics 1, 323–338.

Alm, J., Jackson, B., McKee, M., 1992. Institutional Uncer-

tainty and Taxpayer Compliance. The American Economic Review

82 (4), 1018–1026.

Alm, J., Torgler, B., 2006. Culture differences and tax morale in the

United States and in Europe. Journal of Economic Psychology

27 (2), 224–246.

Balabanova, D., McKee, M., 2002. Understanding informal pay-

ments for healthcare: the example of Bulgaria. Health Policy (62),

243–273.

Chander, P., 1992. Corruption in tax administration. Journal of

Public Economics 49, 333–349.

Chen, K.-P., Chu, C. C., 2005. Internal control versus external ma-

nipulation: A model of corporate income tax evasion. The RAND

Journal of Economics 36 (1), 151–164.

Coalition2000, 2005. Anticorruption reforms in Bulgaria. Report,

Coalition2000.

Cummings, R. G., Martinez-Vazquez, J., McKee, M., Torgler, B.,

2005. Effects of tax morale on tax compliance: Experimental and

Survey Evidence. CREMA Working Paper 2005-29, Center for

Research in Economics, Management and the Arts.

Feinstein, J. S., 1991. An Econometric Analysis of Income Tax Eva-

sion and its Detection. The RAND Journal of Economics 22 (1),

14–35.

22

Feldstein, M., 1999. Tax Avoidance and the deadweight loss of the

income tax. The Review of Economics and Statistics 81 (41), 674–

680.

Fuest, C., Weichenrieder, A., 2001. Tax competition and profit shift-

ing: On the relationship between the personal and corporate tax

rates. CESifo Working Paper 781, CESifo.

Gordon, R. H., Slemrod, J., 1998. Are ”real” responses to taxes

simply income shifting between corporate and personal tax bases.

NBER Working Paper 6576, National Bureau of Economic Re-

search.

Hanlon, M., 2003. What can we infer about a firm’s taxable income

from its financial statements? National Tax Journal 56 (4), 831–

864.

Joulfaian, D., 2000. Corporate income tax evasion and managerial

preferences. The Review of Economics and Statistics 82 (41), 698–

701.

Mares, I., 2003. The sources of business interest in social insurance:

Sectoral versus national differences. World Politics 55 (2), 229–

258.

McGillivray, W., 2002. Contribution evasion: Implications for so-

cial security pension schemes. International Social Security Review

54 (4), 3–22.

Nyland, C., Smyth, R., Zhu, C. J., 2006. What determines the extent

to which employers will comply with their social security obliga-

tions? Evidence from Chinese firm-level data. Social Policy and

Administration 40 (2), 196–214.

Sandler, T., Sterbenz, F., Posnett, J., 1987. Free Riding and Uncer-

tainty. European Economic Review 31, 1605–1617.

23

Slavova, Z., Shisheva, M., Zlatanov, S., 2007. Social Security in Bul-

garia in 2007. Annual report, National Social Security Institute.

Slemrod, J., Yitzhaki, S., 2000. Tax avoidance, evasion, and ad-

ministration. NBER Working Paper W7473, National Bureau of

Economic Research.

S.Mitchell, O., Myers, R. J., Young, H. (Eds.), 1999. Prospects for

Social Security Reform. University of Pennsylvania Press.

Torgler, B., 2002. Speaking to theorists and searching for facts: Tax

morale and tax compliance in experiments. Journal of Economic

Surveys 16, 657–683.

VitoshaResearch, 2004. Corruption monitoring system (business sec-

tor). Report, Vitosha Research.

Yaniv, G., 1988. Withholding and non-withheld tax evasion. Journal

of Public Economics 35, 183–204.

Yaniv, G., 1993. Payroll tax evasion. Discussion Paper 51, National

Insurance Institute.

24

Year Full retirement age Average life expectancy

2000

2001

2002

2003

2004

2005

2006

2007

Men Women Men Women

60.6 55.6 68.15 75.34

61 56 68.5 75.2

61.6 56.6 68.54 75.37

62 57 68.68 75.59

62.6 57.6 69 76.3

63 58 69 76.3

63 58.6 69.1 76.3

63 59 69.2 76.3

Source: National Statistical Institute

Table 1: Full retirement age and average life expectancy 2000 - 2007

Year Statutory CIT rate % Revenue % GDP**

1997

1998

1999

2000

2001

2002

2003*

2004

2005

2006

2007

Profit <=50,000 BGN Profit >50,000 BGN

28 40.2 3.09

28 37 2.34

28 34.3 2.26

28 32.5 2.26

23.5 28 2.5

23.5 28 2.69

- 23.5 2.75

- 19.5 2.2

- 15 2.11

- 15 2.39

- 10

*Starting from 2003, the CIT rate is the same for all firms irrespective of profit

** Revenue from non-financial companies only

Table 2: Corporate Income Tax Rates and Revenue

Year Social Security % Share ratio

1997

1998

1999

2000

2001

2002

2003

2004

2005

2006

2007

By employer By employee

44 2 -

44 2 -

42.4 4.7 -

36.7 9 80:20

34.3 8.4 75:25

34.16 8.5 75:25

34.16 8.5 75:25

32.2 10.5 75:25

30.65 12.45 70:30

24.075 12.425 65:35

24.075 12.425 65:35

Table 3: Social Security Contributions

Pooled OLS Fixed Effects Arellano-Bond GMM

Lag Cost of empl

log(SS/CIT)

log(SS/CIT)*H

Year Dummies

Number of employees* year dummies

Observations

0.382 0.014

-4.67 0.227 -1.87 0.138 -1.14 0.188

6.11 0.301 -0.197 0.16 -0.483 0.169

Yes Yes Yes

Yes Yes Yes

23, 563 23, 563 10, 503

All standard errors are robust and are reported in the second column for each estimation technique. The sample covers 1997-2002.

Table 4Response of Total labor expenditure

Pooled OLS Fixed Effects Arellano-Bond GMM

Lag Cost of empl-min. wage bill

log(SS/CIT)

log(SS/CIT)*H

Year Dummies

Number of employees* year dummies

Observations

0.349 0.02

-5.05 0.244 -1.78 0.158 -0.786 0.307

7.18 0.323 0.014 0.169 0.0711 0.282

Yes Yes Yes

Yes Yes Yes

23, 563 23, 563 9393

All standard errors are robust and are reported in the second column for each estimation technique. The sample covers 1997-2002.

Table 5Labor Expenditure Above the Mandatory

Minimum