Estimating the underground economy and tax evasion: cointegration and causality evidence in the case of Cyprus, 1960-2003.

• Author: Fethi, Meryem Duygun
• Position: Report

Abstract

We empirically investigate the size of the underground economy and the amount of tax evasion in the light of Tanzi's currency demand framework for the Cyprus economy. The study covers the period 1960-2003. The paper has two purposes: (1) to find out whether any long run relationship exists for the pairs of measured GDP-underground economy, tax rates, tax evasion and underground economy-tax rates, and (2) m investigate the direction of causality between the pairs Our findings suggest that (i) one co-integrating vector exists between the variables by using the Johansen co-integration approach; (ii) the measured GDP and tax rates are the causes of underground economy and tax evasion respectively whereas bidirectional causality is found between tax rates and underground economy when the FPE Wald, Sim's LR and Holmes-Hutton causality techniques are applied; (iii) significant underground economic activity and changes in tax rates might stimulate greater loss of tax revenue or re tax evasion with a larger budget deficit and slower economic growth.

Introduction

The size of the underground economy and tax evasion have long been of interest and empirically investigated in a number of recent studies. In the relevant literature, a common question is: 'Is there is any relationship between measured GDP and the underground economy?', in other words, 'Are measured GDP and changes in tax rates causes of both underground economy and tax evasion?' It is really hard to provide precise answers for these questions in the first place. However, it is widely agreed that there exists a relationship between measured GDP and the underground economy (see Giles, 1997 and Giles et al., 2002) or between tax rates and the amount of tax evasion (see Kesselman, 1989 and Trandel and Snow, 1999). Policy makers are also aware that significant underground economic activities and changes in the tax rates are associated with slower economic growth, more tax evasion, greater loss of tax revenue and higher budget deficits.

It is important to note that the causality issue between these variables is still controversial and unresolved due to ambiguous evidence in the literature. Nevertheless, some studies on the underground economy provide strong evidence that the direction of causality runs from measured GDP to the underground economy (Giles, 1997) and from tax rates to measured GDP (Scully, 1996). Some evidence also supports the presence of causality between tax rates and the size of underground economy (Giles and Caragata, 2001).

According to the European Union (EU) Report on Undeclared Work in an Enlarged Union (2004), it is highlighted that very little is known about the Cypriot situation regarding the underground economy. There is only one study by Georgoiu and Syrichas (1994) where the size of underground economy in Cyprus is measured for the period 1960-1990. Cyprus has been a new Member of the EU since May 2004 having an estimated amount of $20300 GDP per capita and 3,2 per cent unemployment rate in 2004 (The World Fact Book, 2004). Despite the low unemployment figure, the existence of relatively small, family size enterprises and the rapid increase of illegal immigrants may create favorable conditions for a thriving underground economy (Cyprus National Action Plan for Employment 2004-2006, p.37).

In this paper, we aim to estimate the size of underground economy and the amount of tax evasion in Cyprus by conducting Tanzi's (1980, 1983) currency demand approach over the period 1960-2003. Then, we estimate time series data to examine the relationship between measured GDP and underground economy and between tax rates and tax evasion by employing cointegration and causality techniques. Applying these econometric techniques to the Cypriot case to determine both long- and short-run causal relationship between the variables is the first of its kind to the best of our knowledge.

The rest of this paper is organized as follows: Section II presents the theoretical modeling in estimating the size of underground economy. Section III describes the data and the methodology respectively. Section IV discusses the findings. Finally, Section V provides some concluding remarks.

Theoretical modeling in estimating the size of underground economy

The currency demand approach has been the most influential and widely used or cited method. Cagan (1958) used this method to calculate the correlation between currency demand and tax pressure for the U.S. economy. Gutmann (1977) also utilized a similar approach in finding the ratio between currency and demand deposits in a simpler framework. Cagan's idea was later developed by Tanzi (1980; 1983) to empirically investigate the size of the US underground economy.

Tanzi (1983) proposed a basic regression equation which contains weighted average tax rate, proportion of wages and salaries in national income, interest rate on savings deposits, and per capita income as a function of currency ratio in circulation to broad money. The model assumes that cash or currency is the main factor that determines underground economic activities. The second assumption is that the velocity of money in an official economy equals the velocity of money in an unofficial economy. The third stems from a tax burden or very high tax rate that causes the underground economy in that individuals thus prefer to work in the underground economy to avoid high tax burden. The main idea in the model is that a rise in the underground economy will cause an increase in demand for money. In order to find out the size of the underground economy (i.e. excess money in money demand), the currency demand regression should be run over time (see Thomas, 1999 and Bhattacharyya, 1999 for a detailed criticism of the assumptions of currency demand model). The modeling framework of the Currency demand approach employed in this study is as follows:

Using Tanzi (1980; 1983) the currency demand approach is considered in the following equation:

[MATHEMATICAL EXPRESSION NOT REPRODUCIBLE IN ASCII] (1)

Where RCM2 is real currency in circulation to money supply (M2) ratio, TY is the average tax calculated as direct taxes on income expressed as a percentage of GNP, WSY is the share of wages and salaries in GNP, IR is one-year nominal interest rate on saving deposits, INF is the growth rate of consumer price index and PGNP is the real per capita GNP. Ln and [[epsilon].sub.t] denote natural logarithms and error term respectively.

Data and Methodology

1. Data

   The data (1) we have employed for the Cypriot economy are annual figures covering the period 1960-2003. The first data set is for the exploitation of the currency demand approach: RCM2 is real currency in circulation to money supply (M2) ratio, TY is the average tax calculated as direct taxes on income expressed as a percentage of GNP, WSY is the share of wages and salaries in GNP, IR is one-year nominal interest rate on saving deposits, INF is the growth rate of consumer price index and PGNP is the real per capita GNP. For the use of cointegration and causality analyses, the variables in the second data set, namely underground economy (UGE) and tax evasion (TEVA), are estimated by the authors. GDP and tax rate series are extracted from Cyprus Statistical Abstract. All variables in the second data set are deflated by a GDP deflator.

2. Methodology

   Time series data can be non-stationary (trended) and this kind of data can be regarded as a potentially major problem for applied econometric studies. It is well known that trends may cause spurious regression problems. The presence of a trend can be determined by examining the existence of unit roots in time series data. A number of tests were widely recommended for the existence of unit roots in the time series data in the relevant literature such as the Augmented Dickey-Fuller (ADF) (1981), Phillips and Perron, (PP) (1988) and Kwiatkowski et al. (KPSS) (1992).

   The Augmented Dickey-Fuller (ADF) and unit root tests are usually employed respectively in the empirical literature to test the stationarity of the variables for the sake of confirmation. The most reliable and general model of the ADF test can be reformulated as follows (2):

   [DELTA][y.sub.t] = [a.sub.0] (2) + [gamma][y.sub.t-1] + [a.sub.2]t + [p.summation over (i=2)] [[beta].sub.j][DELTA][y.sub.t-i-1] + [[epsilon].sub.t] (2)

   Where [DELTA] is the first difference operator, y is the series; t = time (trend factor); a = constant term (drift); [[epsilon].sub.t] = Gaussian white noise and p = the lag order. The number of lags "p" in the dependent variable is chosen by the Akaike Information Criteria (AIC) to ensure that the errors are white noise. One problem with the presence of the additional estimated parameters is that it reduces the degrees of freedom and the power of the test.

   To confirm the test results obtained from the ADF test, Kwiatkowski Phillips, Schmidt and Shin's test (1992) (KPSS) is suggested to avoid a possible low power against stationary near unit root processes which occurs in the ADF test. The KPSS test complements the ADF test in which the null hypothesis of KPSS test is that a series is stationary. This means that a stationary series is likely to have insignificant KPSS statistics and significant ADF statistics.

   The KPSS test is based on an assumption that a series can be investigated in three parts: a time trend, a random walk and a stationary error in the following equation:

   [y.sub.t] = [rho]t + r[w.sub.t] + [[epsilon].sub.t] (3)

   Where r[w.sub.t] = r[w.sub.t-1] + [v.sub.t] and [v.sub.t] is i.i.d (0, [[delta].sub.v.sup.2]). Basically, the regression above can be run in two ways: first with a constant in the case of level stationary, second both a constant and a trend in the case of trend stationary. We then use the residuals [[epsilon].sub.t] from the regression and compute the LM statistics in the following equation:

   [MATHEMATICAL EXPRESSION NOT REPRODUCIBLE IN ASCII] (4)

   Where...