Iranian Economic Review, Vol.13, No.

22, Fall 2008

Underground Economy and Tax Gap

∗
Ali Taiebnia
∗∗
Shapour Mohammadi

Abstract
In this paper the relation of underground economy and tax gap in
Iran has been examined. The results show that the underground
economy size is almost 20 percent of GDP in Iran. In this research
three methods of estimation of underground economy size has been
used and combined to achieve a more precise estimation. The
methods that have been used are: cash, fuzzy and latent variable
methods. The size of tax gap in Iran has been over 30 billion dollars
in recent years. Some new payment systems as ATM cause change in
money velocity which makes it necessary that new factors to be taken
into account

Keywords: Underground Economy Size, Fuzzy Sets, Forecast

Combination Informal Economy Size, Latent Variable method.

∗Assistant professor at University of Tehran, Faculty Economic.

∗∗ Assistant professor at University of Tehran, Faculty of Management.
2/ Underground Economy and Tax Gap

1- Introduction
Underground economy is one of the unwanted facts in many countries.
This sector can be defined as legal and illegal sector of economies.
Households which produce goods and services for their own uses are agents
of informal economy but not illegal agents. Black market economy, shadow
economy, and underground economy are almost equal in terms of literature.
However, they are different from informal economy and illegal ones. The
first term has a broader definition than the second one. Economic literature is
familiar with the story of underground or shadow economy. Tanzi (1980) is
the earliest work on the estimation underground economy size. His work is
based on liquidity demand. another study in this area is done by Frey and
Hanlor(1984) that uses unobservable variables for estimation of underground
economy (UE hereafter) size. They find variables such as tax burden,
regulation and tax morality as statistically significant variables. This study
applies Multiple Indicators and Multiple Causes (MIMIC) which is a factor
analysis method for estimating UE as an unobservable variable. From an
econometrics point of view linear structural equations (LISREL) is a
generalization of MIMIC model. LISREL has two parts: measurement
models and structural equations. Measurement model relates unobserved
variables to observable indices and structural equations focus on the relation
between unobservable variables. In the Frey and Hanlor, the unobserved
variable is UE which its size is determined by exogenous variables which are
measured without any error. Another set of variables are indicators that
affect UE size. Other studies in this area, Giles and Caragata (1994), and
Caragata and Giles (1994) examine the response of underground economy to
tax burden and tax mix. In addition, Schneider and Enst (2000) estimated UE
size in developed and developing countries by various methods.
The rest of the paper is organized as follows. The second section briefly
explains methods of estimation and provides our main method in the
estimation of UE size. Third section presents estimation and interpretation of
UE size as well as tax gap in Iran. The last section is devoted to concluding
remarks and policy prescriptions.
 Taiebnia, Ali & Shapour Mohammadi. /3

2- Methods of Estimation for Underground Economy Size

Various methods can be used for the estimation of UE size. The
methods are mainly based on liquidity demand methods mainly and in some
cases energy demand and national accounts imbalances. For example,
demand for money is a method, which supposes that transactions in
underground economy take place with money. According to this method
other medium of exchange can not be used in illegal or informal economy.
Other methods, such as MIMIC model and fuzzy approach also need a
quantitative estimate at first or at least in one of years for UE size. In this
section, we explain these methods in detail.
Methods of estimation can be divided to two categories: direct and
indirect methods. Direct methods are based on survey, tax auditing and
voluntary questioner approach. These methods are micro approach and are
implemented in various countries. The main advantage of these methods is
ability of them in providing broad range of information about structure of
underground economy. However, the results depend on formulation of
questioner and willingness of people to give exact and correct answers. One
of helpful methods in estimating UE size is difference between actual and
declared (stated) tax. The main weakness of this method is its sample
selection bias. When tax auditing's focus is on tax payers, tax avoiders are
ignored and therefore the estimation of underground economy is only a part
of real underground economy and can be used as minimum size of UE in
practice.
Indirect approaches are macro approaches mainly and usually are called
index approaches. These methods use macroeconomic and other factors for
estimation of underground economy size. One of the methods is defined as
difference between national income and national expenditure. These
methods can be reliable if these two methods are used independently and
measurement error is not high. However, in practice both problems are
present. In most of countries difference between income and expenditure is
used for adjustments of national accounts which makes less difference
between these figures in practice and therefore are not reliable information
for UE size measurement. Also some sectors are omitted from national
income accounting and other sectors are measured with error, therefore
statistics of national accounts can nit be used in estimation of UE size
4/ Underground Economy and Tax Gap

(Schneider and Enste, 2000, p.93). For example, UE size of Russia was
reported 74.9 percent for 1995.
The second method is difference between formal and informal
labor force. Change of labor in the formal sector with given rate of
participation can be attributed to change of UE size. This method has at least
two drawbacks. The first one is other factor than change in UE size which
may change labor of formal sector. The second one is possibility of working
in both formal and informal sector simultaneously.
The Third method is transaction method which uses Fisher
equation MV = PT , where M is money, V is velocity of money circulation,
P stands for prices and T is total transaction. The method is proposed by
Feige (1979). In this method supposed nominal GNP is composed from
formal and informal GNP therefore subtracting formal GNP form the GNP
that is calculated by Fisher equation one can obtain informal GNP. In this
method argued that although the informal activities are not included in
formal GNP, transactions show both formal and informal activities. In other
words if formal sector's size is constant, changes in transactions are result of
changes in informal activities. This method supposes that the size of
underground economy is zero in the base year and the ratio of UE to formal
economy is constant. These two assumptions and the necessity of having a
lot of information cause restrictions, which make this method inapplicable.
The Fourth method is cash demand method. For the first time this
method is used by Tanzi. The main reason behind this method is that
informal activities mainly are done by cash, not by banking instruments such
as check and drafts, etc. To extracting the effect of other variables such as
interest rate, payment habits, income and so on, an econometric model
should be estimated. Also some factors such as tax burden and regulations
complexity which encourage economic agents to do their transaction in
underground economy should be entered. The regression was used by Tanzi
(Tanzi 1983) is as follows:

Ln(C / M2 )t = β0 + β1Ln(1+TWt ) + β2Ln(WS / Y)t + β3Ln(Rt ) + β4Ln(Y / N) + εt (1)

With expected signs β1 > 0, β 2 > 0, β 3 < 0, β 4 > 0

 Taiebnia, Ali & Shapour Mohammadi. /5

Where C/M2 is cash ratio to demand deposits and time deposits, TW

weighted average of tax rate (a proxy for change in volume of underground
economy), WS/Y share of salary and wage in national income (for taking to
account various patterns of payment and cash holding), R is interest rate for
cost of holding money and Y/N is per capita income. Excess demand for
cash can be result of tax burden and other factors which make people to do
transactions in underground economy. In this method the unrealistic
assumption of zero value for underground economy is not necessary,
however to use this method money velocity should be estimated for formal
sector and be used for informal sector. This assumption that velocity of
money in formal and informal sectors are the same is not so realistic.
However, we cannot estimate velocity of money in the informal sector and
there is no option other than accepting the same velocity for money in formal
and informal sectors:

Ln(C /̂ M2 )t = βˆ0 + βˆ1Ln(1+TWt ) + βˆ2 Ln(WS/ Y )t + βˆ3 Ln(Rt ) + βˆ4 Ln(Y / N) + ε t

With assumption of zero rate tax:

Ln(Cˆ/ M2 )*t = βˆ0 + βˆ2Ln(WS / Y)t + βˆ3Ln(Rt ) + βˆ4Ln(Y / N) +εt ( 2)

Let α be the ratio of cash to deposits in the formal sector, then
Ct* = M 2t eα and with subtracting C∗t from C t , one can get the size of
underground economy CIt = Ct − C*t . In this stage we should estimate
velocity of money (M2 here) and by the formula PT=MV size of
underground economy can be estimated. This method is one of the common
methods in the estimation of underground economy size, however some
critiques of it are as follows:
1- Most of underground economy activities are done by cash, but not
all of them. For example questioner researches show that only 80 percent of
underground economy activities are done by cash in Norway. (Isachsen &
Strom, 1980, 1985)
2- In this method only one of the factors that may lead to growth of
shadow economy, tax burden, is considered, whereas there are some other
6/ Underground Economy and Tax Gap

important factors such as tax morality, and belief of tax payers about tax and
government. Therefore, size of economy in this method is underestimated.
3- Increase in demand for cash is partially a result of decrease in
demand deposits in some countries and cannot entirely be related to the
underground economy.
4- The last critique is related to the assumption of constancy of money
velocity.
The Fifth method is electricity consumption method which was
proposed by Kaufmane and Kaliberda (1996), and Lacko (1998). In the
method of Kaufmane and Kaliberda difference of electricity growth and
GDP growth is used for extension of underground economy. Lacko method
is based on household electricity consumption is used in shadow economy,
therefore household consumption of electricity should be considered. There
are many drawbacks in the use of electricity method, which make it useless
for our study (Schneider, 2000), therefore avoid of further explanations.
Sixth method is related to underground variables, so called model
approach. This method is also called Multiple Indicators and Multiple
Causes (MIMC) approach. In this method unobserved variables such as UE
size is related to observable variables such tax burden, regulation intensity
and unemployment in formal sector. These variables are called causes. On
the other side indicators are effects of unobservable variables. Effects can be
GDP growth of formal sector, participation rate of formal sector and money
variables. The demand for M1 is increases as UE size increases. This method
is used by Frey and Hanneman (in 1984). (Further details of this methods is
presented in Appendix A5)
We will use this method for Iran data in LISREL8 for calculation of
UE size. Initial level of these estimates will be determined by liquidity
approach.
Another method is called econometric method in the context of
underground economy studies. This method is based on Baumol and Tobin
"Optimal Cash Balance (1989)". The model is as follows:

Mt = β0′YRtβ1 Rtβ2 Pt β3 exp(εt ) ( 3)

 Taiebnia, Ali & Shapour Mohammadi. /7

Where M t is money volume, YRt is recorded real income or

production, R t is short term interest rate and Pt is general level of prices.

This method is alternative form of money demand. Taking log and denoting
logarithm of variables by lowercase letters:
mt = β0 + β1 yRt + β2rt + β3 pt +εt ( 4)

′
Where β 0 = Lnβ 0 . If the underground sector is included in the model

YRH t = (YRt + YHt ) the model can be stated as:

mt = β0 + β1 yRHt + β2rt + β3 pt +εt ( 5)

In this method, using YHt data which is calculated from other

methods model is estimated and velocity of money is calculated for formal

and underground sectors. This model is appropriate for the cases that one
tends to represent the relation between macroeconomic variables when
underground sector is considered as an economy sector. In other words, this
method takes to account effects of putting aside underground sector in policy
analysis.
Giles and Johnson (1999) extend Trandel-Snow model and examine
relation between effective tax rate and the ratio of underground sector to
GDP. They show that in nonconstant range of income and decreasing
absolute risk aversion and nondecreasing relative risk aversion, an increases
in average rate of can increase or decrease both of formal and informal
sectors' size. Giles and Tedds (2000) study indicates that there is a positive
S shape relation between effective tax rate and size of underground
economy.
Seventh method which is used for modeling UE size is fuzzy logic
approach. This method is introduced by professor Lotfi Zade. Fuzzy logic
8/ Underground Economy and Tax Gap

makes set theory more flexible and effective in real world applications. In
classic set theory each element such as, a , may or may not be an element of
set A . While in fuzzy sets each element of a set belongs to the set with a
specific degree of membership. For example degree of membership to A for
element , a , is 0.1. In other words in fuzzy sets each element is related to a
degree of membership.
X = {a , b , c , d }
)
X = {( a , 1) , (b , 0.5 ) , (c , 0.3) , (d , 0.7 )}
)
Where X is a classic set and X is a fuzzy set. In the fuzzy set each
element such as a, b, c, d have a membership degree that states the degree
)
of membership of that element in the set X . This approach in sets and logic
makes approximate reasoning possible for effective modeling real world
phenomena. Although, interval mathematics has most of strengths of fuzzy
logic, fuzzy logic is very simple in practical applications. As father fuzzy
sayes there are other methods that can be used instead of fuzzy logic,
however it is one of simplest methods for similar problems.
The elements of fuzzy sets are similar to a classic set. For the degree
of membership one can use common membership function in fuzzy logic
literature. Appropriate membership function differs for different
applications. In spite of probability, sum of membership degrees is not
necessarily one and can be defined in an interval other than [ 0,1] .

The concepts such as less, medium, more, and also linguistic

variables can be handled with fuzzy logic. Most of operators of set theory
are applicable in fuzzy sets with a little adjustment. For example, union of
sets can be stated with max of membership degrees and intersection of sets
with min of membership degrees and complement with subtracting
 Taiebnia, Ali & Shapour Mohammadi. /9

membership degree from 1 . Therefore, Identity property, Commutative,

Distribution, and DeMorgan rules are satisfied in fuzzy sets.
The fuzzy sets are applied to various problems which are not limited
to specific area of science. One of the applications of fuzzy logic is
measurement of unobservable variables. Giles is from among the first of
researchers that applied fuzzy logic in study of size of underground
economy. Further details are presented in Gills (1999).
When one forecasts economic variables with different methods, he
can combine them to make a more accurate forecast. Methods like
regression and variance-covariance of forecasts are the most common ones.
In the first method, actual value of forecasted variable is given for some
periods and one should determine which weights of various forecasting
methods can lead to best results.
AT + j = β P1,T + j + (1− β )P2,T + j + εT + j ( 6)

Where AT + j stands for actual value in T + j and P1,T + j is forecasted

value of first method. Also P2,T + j is forecasted value of second method. For

estimating β we use the following relation:

AT + j − p 2,T + j = β (P1,T + j − p 2,T + j ) + ε c ,T + j

Estimation of β will give optimal combination of forecasts. The reason

is sum up to 1 condition in forecast combination: α = 1 − β , α + β = 1 . For

generalization to multimodel or multimethod forecasts, it can be written:
AT + j = β1 P1, T + j + β 2 P2 , T + j +.... + β k Pk ,T + j + ε C ,T + j

As sum of weights for various forecasts should be one:

β1 + β 2 + L + β k = 1
10/ Underground Economy and Tax Gap

β k = 1 − β1 − β 2 + L

AT + j − pk ,T + j = β1(P1,T + j −Pk ,T + j ) + β2(P2,T + j −Pk ,T + j ) +L+ βk −1(Pk −1,T + j −Pk ,T + j ) +εc,T + j

or in more compact form:

∗ ∗
A∗ T + j = β1 P1 T +J + β 2 P2 T+ j + L + β k −1 P ∗ K −1,T + j + ε c ,T + j

Where all of the forecasts and actual values are stated as deviation from
kth forecast value. Although this method is very useful it can not be
applicable in estimating the size of underground economy. Because value of
underground economy which is used as actual value is not measured
accurately. As the main concern of this paper is a precise measurement and
not a forecast of UE size, we use the second method i.e. variance-covariance
method.
Variance- covariance tries to combine different forecasts or
estimations to minimize variance of forecasts. This method relies on
combination of forecast error of various methods. (Clements and Hendry
1998, p. 229)

ec , +T = αe1,t + (1 − α )e2,t

Where e1,T + j is forecast error of first method or model in time T + j

and e2,T + j ,T + j is forecast error of the second method in the same time which

combination of them is equivalent to combined error ec ,T + j . Minimizing

variance equation leads to:

V ( ec,t ) = α 2 V ( e1,t ) + (1 − α ) V ( e 2,t ) + 2α (1 − α ) cov ( e1,t , e 2,t )

2
 Taiebnia, Ali & Shapour Mohammadi. /11

Taking derivative respect to α

V( e2,t ) −cov( e1,t ,e2,t )

α∗ = ( 7)
V( e1,t ) + V( e2,t ) − 2cov( e1,t ,e2t )

By putting α∗ in prior equation, one can show that:

MSFE ( Pc ,t ) ≤ min {MSFE (P1,t ), MSFE (P2,t )}

Therefore with combination of forecasts in any way forecast error can

be reduced. This method can be generalized and is applicable for various
types:
ec ,t = β1e1,t + β 2 e2,t + L + β k ek ,t

V ( ec,t ) = ∑∑βiβ j cov ( ei,t , e j,t )

k k

j=1 i =1

As sum of weights ( β1 + β2 + L + βk ) should be 1, optimal values of

βi is given by solving following restricted optimization problem:

( i,t j,t ) ⎢∑βi −1⎤⎥

⎡k
k k
L = ∑∑ββ i jCov e e +λ (8)
j=1 i=1 ⎣ i=1 ⎦

A MATLAB code1 is written by authors for solving the problem.

3- Data, Estimation, and Empirical Results

In the previous section literature of underground economy and common
methods in estimation of UE size was presented. In this section, based on
previous studies and making some adjustments because of data limitations in
Iran we try to estimate the underground economy size.

1- The code is available through a formal request.

12/ Underground Economy and Tax Gap

The data for estimation of models is the national accounts of 1959-

2003(1338-1382 Iranian calendars) that are collected from various statistical
1
sources such as Central Bank of Iran, web site of Iran economy
2
information and Iran Statistics Center. First of all, a regression model based
on liquidity demand is fitted and tax gap is calculated. Variables of the
model are: C A S H as dependent variable and T A X / G D P , ratio of tax to
gross domestic product, and D IR E C T A X / G D P , ratio of direct tax to GDP,
for modeling the effect of tax burden on liquidity demand, R A T E 3 , one
year bank deposits rate, for taking in to account the opportunity cost of cash
demand and P E R I N C C U R , per capita income in current currency, for
showing the effect of precautionary and transaction demand of money. In
some models the ratio of C A S H / M 2 , ratio of cash to M 2 definition of
money, is used as a dependent variable. Also nominal balance of money is
divided to C P I for converting to real balance. In this case, dependent
variable of regression is denoted by R EALCASH .
Based on previous studies various models are fitted and some results
are presented in following sections.

LO G (C ASH /M 2) = -0.112+1.093LO G (C ASH (-1)/M 2(-1)) + 13.744LO G

(1+D IR EC T AX /G D P)- 0.086LO G (R AT E3) + [M A(1)= -0.997,BAC K C AST =1343]
F= 114.385 R 2 =0.93
Pval.F=0.000000 D W =1.71

This model is one of cash demand models which was presented in

detail in the second section. The dependent variable is the ratio of cash to
M 2 and independent variables are first lag of dependent variable, ratio of
direct tax income to G D P and interest rate. As a matter of fact, various other
variables are entered in model and by backward approach are deleted from
model when they found to be insignificant. In this "general to specific"
specification first type error level is considered to be 5 percent in all of
cases. Complete results are presented in the appendix which show that the
variables in model after deletion of insignificant variable are significant in 1

1- www.cbi.ir
2- www.ieicenter.com
 Taiebnia, Ali & Shapour Mohammadi. /13

percent level or less. Also model is overally significant that is shown by very
low probability of F statistic. In addition diagnostic tests on residual of
regression confirm no serial correlations for error term assumption and
nonspurious regression. In order to achieve a proper model we used three
different specifications for the liquidity model. In the first model (Appendix
1) both of tax variables (ratio of direct tax to GDP and ratio of indirect tax
to GDP ) were entered in the model. After the deletion of the indirect tax
(this variable is not significant at 5 percent level) the direct tax ratio to GDP
became significant at one percent level. This result is in line with theoretical
considerations and empirical results, and confirm that direct tax have
positive effect on liquidity demand and therefore on UE size. This result is
interesting when it reveals that structure of tax (tax mix) can be important in
changes of UE size.
In another model indirect tax ratio to GDP is replaced by ratio of
total tax to GDP (Appendix: model 2). In the model there is multicolinearity
between total tax ratio and direct tax ratio, when the indirect tax is a
component of total tax. In the model, also the direct tax variable ratio is
significant and total tax ratio variable is not significant.
Final model (Appendix model 3) is obtained by omitting the
insignificant variable (total tax ration to GDP) from the model. Summary of
the results of this model which is presented in above indicates that the model
is significant in less than one percent level and signs of the variables are
consistent with theory. Increase in tax causes increase in money demand and
increase in bank deposits rate, reduces liquidity demand. Income variable
has positive sign, but it is not significant in five percent level. Therefore, it is
deleted from the model.
To estimate UE size from the regression mentioned above, tax burden
term is deleted and the real balance of money in the absence of shadow
economy (Theoretical reasons were presented before) is calculated:

L O G ( C A S H / M 2 ) = -0 .1 1 2 + 1 .0 9 3 L O G ( C A S H ( - 1 ) / M 2 ( -1 ) ) - 0 .0 8 6 L O G ( R A T E 3 )
+ [ M A ( 1 ) = -0 .9 9 7 ,B A C K C A S T = 1 3 4 3 ]
14/ Underground Economy and Tax Gap

Then using the following equation size of underground economy is

calculated:

U E S IZ E = (C A S H -e x p ( C A S H f) M 2 ) V 2 / G D P (9)

Where V 2 is velocity of M2 , C A S H is liquidity and C A S H f is

predicted liquidity from final regression model and UESIZE is the size of
shadow economy that is calculated from the equation (9).

Table 1: Share of shadow economy and tax gap (Structural regression)

Tax Gap Shadow Tax Gap Shadow

Year (Billion Rials) Economy's Year (Billion Rials) Economy's
Share Share
1963 1.51 7.31 1983 173.36 21.76
1964 2.03 8.76 1984 204.84 22.79
1965 2.77 9.7 1985 225.49 21.81
1966 3.59 10.34 1986 231.6 22.6
1967 4.27 10.95 1987 224.08 21.75
1968 5.38 10.82 1988 202.39 20.52
1969 6.49 10.72 1989 217.6 18.32
1970 7.47 10.58 1990 290.82 17.16
1971 8.26 10.05 1991 427.62 15.47
1972 10.55 10.28 1992 550.44 14.58
1973 14.06 10.72 1993 566.53 13.95
1974 17.09 10.83 1994 764.07 13.92
1975 31.38 11.59 1995 910.64 12.45
1976 37.75 11.01 1996 1417.23 11.28
1977 48.05 10.83 1997 1980.5 11.42
1978 122.32 26.26 1998 2183.07 11.68
1979 74.92 20.34 1999 2962.09 11.47
1980 73.5 21.59 2000 3714.39 10.09
1981 124.84 22.53 2001 3857.24 9.09
1982 135.11 22.01 2002 4176.02 8.33
 Taiebnia, Ali & Shapour Mohammadi. /15

Due to use of credit cards and automatic teller machines (ATM), it

seems that velocity of the money have been increased in recent years.
However, the hard currency in demand had little growth. This fact can be
important in the estimation of UE size. Since there is no data about ATMs in
Iran, we cannot reduce the bias of the estimation in our research.
Tax gap shows how much income tax could be gathered if the size of
underground economy was zero. If one can shows that the main motive of
underground activities is tax evasion, the policy recommendation is using
indirect tax (especially excise tax) instead of income tax.
In order to examine validity of the estimated results and also to avoid
the spurious regression we take Johanson cointegration test (Table 4 in
appendices). The results of the test show that there is a long run relation
between the ratio of cash to M2 as dependent variable and deposits rate and
indirect tax to GDP ratio as independent variables. This relation is
significant in less than one percent level of first type error. In other words, in
addition to short run relation, there is a positive long run effect of direct tax
to GDP ratio on size of underground economy.

Table 2: Normality Test for Residuals of the Model

The normality test of residuals shows that normality hypothesis can not
be rejected at 5 percent level. Hence, in 5 percent level distribution of
residuals is normal and test-statistics are valid. Also, based on following
table one can concluded that there is no serial correlation.

Table 3. LM Test for Serial correlation

16/ Underground Economy and Tax Gap

In addition to the above mentioned models, other models are estimated,

but because their results are not consistent with theory we do not report them
in this paper. Some of these models are those used by Spiro (1996).
We tried to use MIMIC model, but because of covariance matrix error
for Iran data, LISREL did not turn out any output for general model.
The Estimation of UE size by fuzzy approach is presented in the
following table (this method is based on Gillis methods that was used for
New Zealand). Although these methods underestimate the size of
underground economy in some years they give average size of 16 percent for
whole period which is acceptable and is consistent with other results such as
regression method.
Table 4: Shadow economy's size and tax gap (Fuzzy logic Approach)

Tax Gap Shadow Tax Gap Shadow

Year (Billion Rials) Economy's Year (Billion Rials) Economy's
Share Share

1963 1983 127.2011 15.97

1964 4.578288 19.73 1984 183.7123 20.442
1965 4.56742 15.97 1985 165.0819 15.97
1966 5.54159 15.97 1986 209.9815 20.494
1967 7.99149 20.49 1987 97.51358 9.4655
1968 9.757601 19.63 1988 85.68048 8.6853
1969 11.33765 18.71 1989 86.62404 7.2922
1970 14.52807 20.58 1990 270.6915 15.97
1971 13.12734 15.97 1991 549.4332 19.871
1972 20.73136 20.21 1992 751.0225 19.892
1973 9.879885 7.53 1993 648.5896 15.97
1974 10.74476 6.81 1994 876.8808 15.97
1975 54.35768 20.07 1995 1167.886 15.97
1976 72.18682 21.06 1996 2005.864 15.97
1977 87.74408 19.78 1997 3306.594 19.064
1978 98.53785 21.15 1998 3669.682 19.638
1979 75.1111 20.39 1999 5077.937 19.658
1980 45.38894 13.33 2000 7564.264 20.553
1981 48.29425 8.72 2001 7982.318 18.813
1982 99.62369 16.23 2002 4742.596 9.4585
 Taiebnia, Ali & Shapour Mohammadi. /17

To reduce forecast error one can use combination of two forecasts of

liquidity demand model and fuzzy logic approach. Solving the following
equations gives the share of each forecasting method (42.64 and 57.36
respectively) in combined forecasts.
β i β j Cov [e i , t e ] + λ ⎡⎢ ∑
k k k
⎤
L = ∑∑ j ,t β i − 1⎥
j =1 i =1 ⎣ i =1 ⎦

Figure 1: Combination of the Estimates of Liquidity Demand and Fuzzy logic

Approach

Table 5: Combination of the Estimates of Liquidity Demand and Fuzzy logic

Approach
Year Regression Fuzzy Combined Year Regression Fuzzy Combined
1964 8.76 19.73 15.05446 1984 22.79 20.442 21.44456
1965 9.7 15.97 13.29485 1985 21.81 15.97 18.46191
1966 10.34 15.97 13.56783 1986 22.6 20.494 21.39353
1967 10.95 20.49 16.42214 1987 21.75 9.4655 14.70414
1968 10.82 19.63 15.8748 1988 20.52 8.6853 13.73002
1969 10.72 18.71 15.30123 1989 18.32 7.2922 11.99377
1970 10.58 20.58 16.31778 1990 17.16 15.97 16.47648
1971 10.05 15.97 13.44559 1991 15.47 19.871 17.99244
1972 10.28 20.21 15.97703 1992 14.58 19.892 17.6266
 18/ Underground Economy and Tax Gap

1973 10.72 7.53 8.889317 1993 13.95 15.97 15.10844

1974 10.83 6.81 8.52439 1994 13.92 15.97 15.09392
1975 11.59 20.07 16.45347 1995 12.45 15.97 14.47007
1976 11.01 21.06 16.77569 1996 11.28 15.97 13.97167
1977 10.83 19.78 15.96467 1997 11.42 19.064 15.80396
1978 26.26 21.15 23.32698 1998 11.68 19.638 16.24578
1979 20.34 20.39 20.36905 1999 11.47 19.658 16.16536
1980 21.59 13.33 16.85256 2000 10.09 20.553 16.09262
1981 22.53 8.72 14.60839 2001 9.09 18.813 14.66748
1982 22.01 16.23 18.69413 2002 8.33 9.4585 8.97668
1983 21.76 15.97 18.44084

4- The relation between underground economy size and income

taxes
Income taxes in most countries show the resource which is available for
government in provision of utilities and gives a sense about government’s
liberty in governance and developmental policies. When there is a negative
effect of the UE size on the GDP and government's revenue and type of tax
can be important in determination of UE size, we interested to study the
relation between tax mix and UE size. In this paper we examine the relation
of UE size and tax structure in a cross country framework. The question of
this section is “what is the relation between tax mix and UE size?” Answer
to this question gives us a better understanding of the possibility of reduction
in UE size by changing tax mix. It is important to note that importance of the
tax mix is not affected by the presence or absence of the relation between the
underground economy and tax mix. Importance of the indirect tax is the
availability of it even from underground economy activities. In other words,
even when there is no empirical relation between direct tax and underground
economy, because of the possibility of tax gathering indirectly from
underground economy and impossibility of gathering income tax from this
sector changing the tax mix is still recommended. One of the main motives
of underground economy’s activities is income tax evasion. Therefore if
income tax could be applicable (irrespective of possibility of the income tax
 Taiebnia, Ali & Shapour Mohammadi. /19

gathering from this sector) for some of agents there is no motivation to enter
to underground economy. Underground economy such as formal economy
uses goods and services therefore there is no way for evasion from indirect
tax and this not only leads extension of tax base, but it also shrinks UE size
in future. The graphical illustration is as follows:

Economic Agents: Can I

get ride of tax by going
to underground economy

Going to underground What kind of tax is

D
economy imposed?

Indi

size of Growth in the countuniu to

hiddern economy activate in formal
economy

Figure 2: Behavior of Economic Agents in tax regimes

Therefore, changes in the tax mix can lead to change in share of

informal economy in economy. Costs of increase in informal economy size
are not solely economic costs, and social costs of this sector are one of the
concerns in the most of the countries.
In most of the studies, panel data or cross section data is used for
country comparisons. Time series data is not proper for the examining the
relation between tax types and UE size, because of the estimated nature of
UE size and also special conditions of Iran’s economy. It may argue that
estimated nature of the UE size may have bias effects in cross country
 20/ Underground Economy and Tax Gap

studies. Although the argument is not incorrect, as the number of countries

increases biases may cancel out each other and mean of them be less than the
study that relies only on Iran’s economy estimates. Therefore, in this paper
we use cross country data for the examination of the tax mix and UE size.
As we know, no study has been done about relation between tax mix
and UE size in Iran and also in other countries there is no study in panel data
form. We use following form for our panel data approach in the study of the
tax mix.
UE it = α + β1T1,it + β 2T 2,it + K + β k T K ,it + ε it (10)
UEit is the size of the underground economy for country i
Where
T
in time t , 1,it is the tax of the first type (income tax for example) for
T
country i in time t , and k ,it is the share of the tax of the kth type for
country i in time t . Because, only the information of 1996 is available t
will be 1996.
We use following data for running a regression of tax mix and its
relation to UE size. The estimated equation show that value added tax is not
significant at 5 percent level and should be deleted from the regression.

Table 6. Data of informal sector size as percent of GDP and tax and Insurance
burden on employers
Informal Value added Direct Social Secutrities Social Secutrities Social Secutrities Social Social
Sector Size as Tax (percent) Tax(Percewnt) Payments(employee) Payments(employer) Payments(Total) Secutieties+Direct Secutieties+Direct
Country
Percent of Tax Tax+Value Added
GDP Tax
Greec 28.5 18 11 15.8 27.5 43.3 54.3 72.3
Italy 27 19 12 9.9 32 41.9 53.9 72.9
Spain 22.9 16 13 6.6 31.6 38.2 51.2 67.2
Belgium 21.9 21 19 10 26 36 55 76
Sewden 19.2 25 20 4 29.6 33.6 53.6 78.6
Norway 18.9 23 19 7 12.8 19.8 38.8 61.8
Denmark 18.3 25 36 9 0 9 45 70
Ireland 15.9 21 20 7.2 12.3 19.5 39.5 60.5
Canada 14.6 7 21 7 8 15 36 43
Germany 14.5 15 18 16.1 16.1 32.2 50.2 65.2
France 14.3 20.6 6 13 31 44 50 70.6
Hetherlands 14 17.5 10 31 8.8 39.8 49.8 67.3
UK 13.1 17.5 16 10.7 10.2 21.4 37.4 54.9
USA 8.8 3 17 7.6 13.8 21.4 38.4 41.4
Austria 8.3 20 8 18.2 24.2 42.4 50.4 70.4
Switzerland 7.5 6.5 10 11.6 11.6 23.2 33.2 39.7
 Taiebnia, Ali & Shapour Mohammadi. /21

UEˆit =−7.54 + 0.09VAT + 0.62DIRETAX + 0.23SOCIAL1+ 0.54SOCIAL2

With strategy of general to specific, the insignificant variable

SOCIAL1 also is deleted from the model. The final model is overally
significant at 5 percent level.

UEˆ it = − 0.47 + 0.5 DIRETAX + 0.49 SOCIAL 2

F = 5.69 Pvalue ( F ) = 0.0168
R 2 = 0.47

Also, Lagrange multiplier test for serial correlation shows that the
model is acceptable in common levels of the significance. The above
regression show that there is no relation between value added tax and UE
size, whereas direct tax has significant positive effect on the UE size.
In the case of Iran, we can use M 2 as a proxy for underground
economy size when it is highly correlated with underground economy size.
Therefore the relation between UE size and tax variables can be studied via
money aggregates such as M 2 .

Expa Increase
Direct nsin of UE s in money
tax size demand

Figure 3: Relation between direct tax and UE size and money demand

As the t-statistic of the tax coefficient in the estimated regression is not

significant can be concluded that indirect tax has no effect on underground
economy size in Iran. But, the direct tax variable has significant positive
relation with liquidity demand and therefore with the underground economy
size.
22/ Underground Economy and Tax Gap

Table 7: Results of the regression of the money demand on the direct tax,
indirect tax, deposits rate and lags of the dependent variable

Results for Iran are consistent with results of other studies. There is a
positive and significant relation between direct tax and UE size. Also this
study is in the line with (Bovi 2003, p. 65) that show a positive relation
between direct tax and UE size by a correlation analysis for all types of tax.

5- Conclusion and policy prescriptions

Studies done in different economies show that almost in all of them part
of economic activities is ignored intentionally or unintentionally from the
economic statistics. Size of the underground economy for Iran is estimated
about 18.5 percent by Schneider. Studies of Iranians also are in line with this
estimate. However, studies in Iran give a range of 7 to 83 percent for UE
size. All in all, the reasonable range for Iran’s underground economy size is
12-25 percent. Previous studies give valuable information about Iran’s
underground economy size that were considered in this study. However, due
to get update estimates and more accurate estimates two methods of the
liquidity demand and fuzzy logic approach were used. Also, we combined
the results of these two methods to reduce the error of the estimates (this
feature of results were not in the previous studies). Also, we used the results
for estimation of the tax gap and in determination of the tax mix and its
relation with underground economy’s size.
The results of our estimation for UE size is about 15 percent for recent
years. Results of this study show high share of underground economy for
special years such as 1957(the revolution of Iran) that confirm the robustness
of the estimations for outlier observations.
The results of the forth section show there is a significant positive
relation between the UE size and direct tax. Therefore changing the tax mix
 Taiebnia, Ali & Shapour Mohammadi. /23

in favor of indirect tax can reduce the size of the underground economy and
increase the attainable tax income for government. The reason behind of the
indirect tax-dominated tax mix is that the indirect tax can be imposed on
both of formal and informal activities and it reduces motivations of the
economic agents to convey their activities in underground economy for tax
evasion.

References
1- Frey, S. B. weck - Hameman, H.(1984). The Underground Economy as
an on observed variable. European Economic Review 26, 33-53. North -
Holland.
2- Frey, S.B.Weck ,H.(1983). Estimating the shadow Economy A"NAÏVE"
Approach, Oxford Economic papers, Vol 35, MARCH 1983, Number 1.
3- Tanzi, v.(1980).Underground Economy Built on Illicit pursuits is
Growing , Concern of Economic policymakers ,IMF Survey (Febr,4),p.34-7.
4- Isachser, Arne and Steiner storm 1980. The Underground Economy, the
Labor market and Tax Evasion, Scandinavians. Econ. 82, pp.304-11.
5- I sachsenr, Arne)and steinar storm 1985 The size and Growth of Hiden
economy in Norway ,Rev. Income weath, 31:1, pp. 21-38
6- Giles, D.A (1999). Measuring the Underground Economy: Implications
For Econmetric Modeling. The Economic Journal, 109(June),F371-F380.
7- Diebold, F.(2001).Elements of Forecating,
8- Clement, Michael, p. and David F. Hnadry (1998). Forecastling
Economic Time Series. Cambridge University Press.
9- Aigner, penis,) and A. S. Goldberger (1977). Latent Variable in Socio -
economic Models. North - Holland.
10- Draeseke, Robert & David E. A. Giles(1999).A Fuzzy Logic Approach
to Modelling the Underground Economy. Derpatment of Economics,
University of Victoria, Canada. Working paper.
11- Zimmerman .H. j.(1996).Fuzzy Set Theory and its Application. Kluwer
Academic publishes.
12- Giles,D.A.E. (1997)The Rise and Fall of New Zealand Underground
economy: Are the Responses Symmetric? (1994)
13- Bajada Christopher(2002), How reliable are the estimates of the
Underground economy? Economic Bulletin, Vol. 3, No.14 PP.1-11
24/ Underground Economy and Tax Gap

14- Modeling the underground Economies in Canada and New Zeeland: A

Comparative Analysis.
15- Schnider, F. (2002). Size and Measurement of the Informal Economy in
110 Countries.Workshop of Australian National Tax Centre, ANU,
Canberra, Australia,
16- Spiro,Peter.(1996). Monetary estimates of the underground economy:
17- a critical evaluation. Canadian Journal of Economics XXIX April .
18- Kaufmann, Daniel and Kaliberda, Aleksander (1996), Integrating the
unofficial economy intothe dynamics of post socialist economies: A
framework of analyses and evidence,Washington, D.C., The Worldbank,
Policy research working paper 1691 .
19- Lackó Mária (1998): The underground economies of Visegrad countries
in international
20- comparison: A household electricity approach , In: Halpern, L. and
Wyplosz, Ch. (eds.),Hungary: Two wards a market economy, Cambridge
(Mass.): Cambridge University Press, p.128-152.
 Taiebnia, Ali & Shapour Mohammadi. /25

Appendices

A1: Regression of the liquidity demand and all types of tax and rate

A2: Regression of the liquidity demand and direct tax and other variables
26/ Underground Economy and Tax Gap

A3: Long Run Relation between Liquidity demand and, tax and interest rate

A4. LM test full results for serial correlation test

 Taiebnia, Ali & Shapour Mohammadi. /27

Table 1: Correlations between the Underground Economy and list

determinants

Source:
Bovi , Maurizio (2003).The Nature of the Underground Economy –
Some Evidence from OECD Countries,JIIDT Vol. 7.
A5. Details of MIMIC Model

The MIMIC model can be presented in econometric specification.

Let η be an unobserved scalar variable (Underground variable which
measures size of under ground economy), Y ′ = ( y1 , y 2 ,..., y P ) vector of
( )
indicators of η and X ′ = x1 , x 2 ,..., x q vector of causes for η . Also
suppose λ ( p×1) and γ ( q×1) as vectors of parameters, and ∈( p×1) and ζ as error
term. All of the Elements of ∈ and ζ are normally distributed and,
Var (ζ ) = ψ Cor (∈) = θ ε .
Then the MIMC model can be stated as
28/ Underground Economy and Tax Gap

y = λη + ε (a1)
η = γ ′α + ζ (a2)
By replacement of ( a2 ) in ( a1) , the model can be written as
y = πx + z , where π = λγ ′ . Also z = λζ + ∈ and Cov(z ) = λλ ′ψ + θ ε ,
since ∈ and ζ are independently distributed. Rank of matrix of regressors is
one and error matrix is constrained; numerical estimation of parameters is
not possible. Estimates are function of parameters and represent relative
importance of parameters. Then, normalization of parameters is required.
Since x and y are given, the equation can be estimated by
maximum likelihood method. By estimated value of π one can get γ = λ .
Putting ζ = E ( ζ ) = 0 we can forecast η and UE size. This estimated value
for UE size is an ordinal value in order to reach the cardinal value at least in
one year UE size should be determined cardinally by other methods. This
fact is the main limitation of this method.
One of conventions in analysis of underground variables is LISREL1
method. In this method two types of observable variables are
X = (x 1 , x 2 ,K , x q )′,Y = ( y 1 , y 2 K , y p )′ which are in deviation from
mean form. It is supposed that variables X, Y satisfy the conditions of
factor analysis models. Common factores can be represented by
η = (η1 ,η 2 , Kη m )′ and ξ = (ξ1 , ξ 2 , K, ξ n )′ , and unique factors
δ = (δ1 , δ 2 ,K δq )′ and ε = (ε1 , ε 2 ,K , ε p )′ . So:
Y = Λ Yη + ε
X = ΛXξ +δ
Where Λ x , Λ y are loading matrixes with q xn , p xm dimensions.
Traditional assumptions of factor analysis are satisfied in our problem
E (η ) = 0, E (ξ ) = 0, E (ε ) = 0, E (δ ) = 0
E (ηε ′) = 0, E (ξδ ′) = 0, E (εε ′) = Θ ε , E (δδ ′) = Θ δ , E (εδ ′) = 0
2 2

Where Θ δ , Θ ε are diagonal matrices. Factors ξ , η are correlated in

whithin and between sets. In order to estimate parameters let Ω be
covariance matrix of (η ′, ξ ′)′ :

1- Linear Interdependent Structural Relationship

 Taiebnia, Ali & Shapour Mohammadi. /29

⎛ Ωηη Ωηξ ⎞
Ω = ⎜⎜ ⎟
⎟
⎝ Ω ξη Ω ξξ ⎠
Ωηη = B ΓΦΓ ′B ′ −1 + B −1 ΨB ′ −1
−1

Ωηξ = Ω′ξη = B −1ΓΦ

Ω=Φ
The model can be estimated by maximum likelihood method.
Variance-covariance matrix of ( y ′, x ′) is as follows:
⎛ ∑ yy ∑ yx ⎞
∑ = ⎜⎜ ⎟
⎝ ∑ xy ∑ xx ⎟⎠
Where
∑ yy = Λ y Ωηη Λ ′y + Θ4
2

∑ xy = ∑ ′xy = Λ y Ωηξ Λ ′x
∑ xx = Λ x Ω ξξ Λ ′x + Θ 2 δ