Challenging the popular wisdom. New estimates of unobserved economy. Luisanna Onnis University of Milan Bicocca-DEFAP ([email protected]) Patrizio Tirelli University of Milan Bicocca ([email protected]) December 2008 Abstract We adopt the Modi…ed Total Electricity approach in order to estimate the dynamics and size of unrecorded economy of a large panel of 49 economies for the period 1981-2005. The …rst conclusion from our estimates is that for most countries the relative size of the unobserved economy has decreased during the last decades. This result is broadly consistent with the country-speci…c economic performances and with key episodes of economic reform and liberalization but in contrast with the recent and highly criticized values obtained by Schneider (2004, 2005) with the MIMIC method. We also …nd that the relative size of the unobserved economy falls with economic growth and exhibits a cyclical pattern which is inversely related to that of the o¢ cial economy.

1

Introduction

The phenomenon of the unobserved economy 1 has attracted considerable concern from policymakers and economists, but to obtain uncontroversial estimates of its size has proven a di¢ cult and challenging task. Di¤erent methods have been used for this purpose, and the consensus holds that the unrecorded economy has been on the rise over the last decades (Hill and Kabir, 1996; Schneider and Enste, 2000; Schneider, 2004, 2005). The purpose of this paper is to measure the dimension of the unobserved sector in a large panel of countries and analyze its trend over the period 1981-2005. Challenging a popular wisdom, we show that the unrecorded economy has in fact fallen. Di¤erent techniques may be used to estimate the size of the unrecorded economy. The OECD (2002, p.188) identi…es three main macro-model methods: i) 1 As de…ned by Feige (1990), “the unrecorded economy consists of those economic activities that circumvent the institutional rules that de…ne the reporting requirements of government statistical agencies”.

1

the Currency Demand approach (CDA henceforth) 2 assumes that unobserved transactions are undertaken in the form of cash payments, so as to leave no observable traces for the tax authorities. In this framework non-measured production can be modelled in terms of stocks or ‡ows of money; ii) the latent variable methods (MIMIC model) model the unobserved economic activity in terms of two groups of variables, one group that is assumed to determine the size and growth of non-measured production (“causal” variables) and a second group that provides evidence of the missing activities (”likely indicators”); iii) the Electricity Consumption method (Kaufmann and Kaliberda, 1996) is based on the empirical observation that overall economic activity and electricity consumption move in lockstep. Measures of total (observed and unobserved) GDP growth are therefore obtained by imposing a constant electricity consumption to GDP ratio. All these methods have shortcomings. The CDA is based on the estimate of a currency demand equation where, in addition to conventional controls (income growth, payment habits, etc.), some other variables are included on the basis of theoretical priors about their impact on unobserved transactions. These typically include the direct and indirect tax burden, government regulation, and the social security burden. The main criticisms to this approach are based on the well known di¢ culty of identifying stable money demand functions. 3 The theoretical priors behind the MIMIC model are similar to those who are used to identify the unrecorded economy in the CDA. As ”likely causes”the MIMIC procedure generally uses the burden of taxation, burden of regulation and citizens’ attitudes toward the state (tax morale). As ”likely e¤ects” it uses the development of monetary transactions, development of the labor market and development of the production market. Critics of the MIMIC approach mainly point at the arbitrariness of the variables grouping into causes and indicators (Helberger and Knepel, 1988; Smith, 2002; Hill, 2002; Breusch 2005). Moreover, the use of variables like taxes and government regulation as likely determinants of the unrecorded economic activities impedes to analyze the impact of these factors on the CDA and MIMIC estimates of the unobserved economy. Contrary to the Currency Demand and MIMIC methods, the Electricity Consumption approach does not require theoretical priors on the causes of the unobserved economy. Critics see the assumption of a constant electricity consumption to GDP ratio as a major weakness of the approach. To allow for both country heterogeneity and a time-varying electricity consumption to GDP ratio, Kaufmann and Kaliberda (1996), Johnson, Kaufmann and Shleifer (1997) and Chong and 2 See

Cagan (1958), Gutmann (1977) and Tanzi (1980, 1983). most commonly raised objections to this method are several: i) not all the transactions in the unobserved economy are paid in cash (Isachsen and Strom, 1985) ii) increases in currency demand deposits are due largely to a slowdown in demand deposits rather than to an increase in currency caused by activities in the unobserved economy (Garcia, 1978; Park 1979; Feige, 1996) iii)there is uncertainty about the velocity of money in the observed economy, and the velocity of money in the unobserved sector is even more di¢ cult to estimate (Klovland, 1984; Hill and Kabir, 1996) iv) the assumption of no unobserved economy in a base year is open to criticism (Schneider, 2004). 3 The

2

Gradstein (2006) have imposed ad hoc values for this key parameter. A more innovative and direct measurement procedure has been presented by Eilat and Zines (2002). Their Modi…ed Total Electricity (MTE) approach …lters the in‡uence of additional variables that a¤ect changes in electric consumption in addition to changes in overall economic activity. These additional factors may include changes in electricity prices and changes of industry share. To the extent that the …ltered series of electric usage captures the output-induced changes in electricity consumption, a unitary output elasticity of electricity may then be used for all countries. This paper adopts the Modi…ed Total Electricity approach in order to estimate the dynamics and size of unrecorded economy of a large panel of 49 economies for the period 1981-2005. Since the time series dimension of the panel is signi…cantly long, the choice of the econometric methodology is based on a preliminary analysis about the stationarity and cointegration of the variables. The application of panel unit root and cointegration techniques represents the second important innovative aspect of this study. Contrary to the MIMIC estimates of Schneider (2004, 2005), we show that the relative size of the unobserved sector has decreased for most countries during the last decades. Furthermore, we …nd that there is a negative and statistically signi…cant correlation between annual growth rates of o¢ cial GDP and annual changes of the share of unrecorded income. This is in line with previous theoretical and empirical works (Asea, 1996; Loayza, 1996). We also compute cyclical gaps in the o¢ cial and unrecorded GDP …gures. In line with the theoretical model of Busato and Chiarini (2004) we …nd evidence of a double business cycle, where the correlation between the two gaps is negative and statistically signi…cant for a large number of countries. Finally, we check for country-speci…c structural breaks in the time series of the unrecorded economy relative share. We are then able to explain them on the grounds of observed institutional changes and economic refoms. The remainder of the paper is organized as follows. Section II describes the model and de…nes the empirical methodology. Section III presents the results. Section IV concludes.

2

Model identi…cation, data description and econometric methodology

Eilat and Zinnes (2002) estimate the relationship between changes in electric consumption and changes in overall economic activity taking into account additional variables that could explain changes in electricity usage. These additional factors may include changes in electricity prices, changes of industry share, and changes in the share of private sector production. Changes of industry share may capture changes in the structure of output, whereas changes of private sector share may capture changes in the e¢ ciency in electricity usage derived from privatization and modernization in the set of transition economies that

3

are analyzed in their work. After …ltering out the in‡uence of these variables, Eilat and Zinnes use the residual changes of electric consumption as a proxy for the growth of total economic activity, thereby assuming a unit elasticity. Then using the point estimates reported in Johnson et al. (1997) they calculate the size of the unrecorded income.4 Our application of the MTE procedure is based on the following equation : Elect;i =

i

+

1

Epricet;i +

2

IndGdpt;i + "t;i

(1)

where subscripts t; i are time and country indexes, and Elec, Eprice and IndGdp respectively describe annual percentage changes in electricity consumption, in relative electricity prices and in industry share of GDP. 5 Then we have created the following variable: Elecres t;i =

Elect;i

[b1 Epricet;i + b2 IndGdpt;i ]

(2)

where b1 and b2 are the estimated coe¢ cients for 1 and 2 in equation (1). The …ltered series of electric consumption already accounts for countryspeci…c supply-side e¤ects and for relative price e¤ects, respectively proxied by IndGdp and Eprice, a unit output elasticity of electricity consumption may be used for all countries. Therefore, the discrepancy between the residual growth rate of electricity consumption, Elecres t;i , and the growth rates of o¢ cial income, Gdpt;i , provides an estimate of the unrecorded economy dynamics, SHt;i . SHt;i =

Elecres t;i

Gdpt;i

(3)

Finally, by applying SHt;i to pre-existing base-year …gures we have obtained our estimates of the unrecorded economy.

2.1

Panel composition

We have chosen base-year …gures that have been largely used in the academic literature.In particular, we have adopted the macroelectric estimates of Johnson et al. (1997)- for the transition economies- and Lacko (1996, 1998)- for the OECD and Developing countries. 6 Then, we have made a further selection considering the availability of data about electricity consumption, electricity price and share of industry. Data on electricity consumption, real price of 4 Actually, in Johnson et al. (1997) only the estimates of unrecorded economy for 17 transitional countries are reported. Eilat and Zinnes (2002) do not specify the source of the remaining countries. 5 Unlike Eilat and Zinnes (2002), the change of share of private sector is not taken into consideration. This variable may capture changes in e¢ ciency in electricity use from privatization particularly for countries that shift from a planned economy to a free market economy. Therefore, since the panel in exam only consists of six transition economies, this additional factor is considered not relevant for the entire group of countries. 6 The base-year …gures of Lacko (1996, 1998) have been calculated with the Household Electricity consumption method. The base-year estimate for Tanzania is from Bagachwa and Nasho (1995).

4

electricity, share of industrial income and o¢ cial GDP have been obtained from Energy Information Administration, International Energy Agency, World Bank and United Nations, respectively. 7 For the lack of complete information about electricity consumption and share of industry, we have ruled out some of the countries for which base year macroelectric …gures were available.8 We have obtained a panel of 49 economies for the period 1981-2005.9 Since the time series dimension of the panel is signi…cantly long, the choice of the econometric methodology is based on a preliminary stationarity and cointegration analysis of the variables Elec, Eprice and IndGdp. In fact, the direct use of OLS or GLS estimators to non-stationary time series may produce regressions that are misspeci…ed or spurious in nature (Granger and Newbold, 1974; Engle and Granger, 1987).

2.2

Panel stationarity and cointegration analysis

The stationarity of the variables Elec, Eprice, IndGdp, Elec, Eprice and IndGdp has been tested using the panel unit root procedures developed by Im, Pesaran and Shin (2003), Pesaran (2003, 2007) and Hadri (2000) (see Appendix 1). Panel unit-root tests are more powerful than unit-root tests applied to individual series because the information in the time series is enhanced by that contained in the cross-section data. In addition, in contrast to individual unit-root tests which have complicated limiting distributions, panel unit-root tests lead to statistics with a normal distribution in the limit. The presence of stationarity has been initially tested using the Im, Pesaran and Shin (IPS) test for the null of unit root in heterogeneous panels. Table 1 (Appendix 2) reports the results. The null of unit root for all variables is rejected against the alternative hypothesis that at least one series is stationary. The IPS test is based on the hypothesis that the error terms are independent across crosssections. As noted by Pesaran (2003, 2007), the cross-sectional dependence may lead to over-rejection of IPS test statistics. Therefore, to support the result of the IPS test, we performed the Pesaran test for unit roots with cross-sectional dependence.10 As Table 1 shows, also this test rejects the null of non-stationarity 7 The rates of growth in electricity price for 26 countries have been obtained from a real index of electricity end-use prices for industry and households created by International Energy Agency . For the remaining 23 countries- for which data on electricity prices are not availablethe growth rates of electricity prices are proxied by a world index of total energy published by Unctad. We obtain similar results if the series of world energy price is used for the entire panel of countries. 8 We have ruled out Azerbaijan, Belarus, Croatia, Estonia, Georgia, Kazakhstan, Lithuania, Latvia, Moldova, Russia, Ukraine, Uzbekistan, Cyprus, Mauritius, and Nigeria. 9 Countries in the sample are Australia, Austria, Belgium, Botswana, Bulgaria, Brazil, Canada, Chile, Colombia, Co sta Rica, Czech R., Denmark, Egypt, Finland, France, Germany, Greece, Guatemala, Hong Kong, Hungary, Ireland, Israel, Italy, Japan, Korea, Malaysia, Morocco, Mexico, Netherlands, Norway, Panama, Paraguay, Peru, Philippines, Poland, Portugal, Romania, Singapore, Slovak R., Spain, Sri Lanka, Sweden, Switzerland, Tanzania, Tunisia, Thailand, United Kingdom, United States, and Venezuela. 1 0 We have perfomed a truncated version of the CADF statistics which has …nite …rst and

5

for all variables against the heterogeneous alternative hypothesis. Thus, the IPS and Pesaran tests suggest that not all the series of each relevant variable have a unit root. However, we cannot state that all series are stationary. In fact, the null of unit root is rejected even if only one series is stationary. The results of the Hadry test for the null of stationarity in heterogeneous panels are reported in Table 2 (Appendix 2). The statistics indicate that there is evidence of non stationarity for all variables Elec, Eprice and IndGdp. For the di¤erenced series Elec, Eprice and IndGdp there is evidence of non-stationarity only if the errors are assumed to be serially correlated. Also the Hadri test is based on the assumption of cross-sectional independence of the error terms. As noted by Giulietti, Otero and Smith (2006), the test may su¤er from size distorsions in the presence of cross-sectional dependence.11 To support the results of the Hadri test, this study has also performed separate Kwiatkowski, Phillips, Schmidt and Shin (1992) (KPSS), ADF and PhillipsPerron (PP) unit root tests. According to these tests, a signi…cant portion of series of each relevant variable have a unit root. With non-stationary pooled time series, the application of the OLS estimator may result in biased and inconsistent estimates (Granger and Newbold, 1974; Engle and Granger, 1987). To de…ne an appropriate estimator for equation (3), it has been therefore necessary to turn to panel cointegration techniques in order to determine wether a long-run equilibrium relationship exists among the non-stationary variables. The presence of cointegrating relationships between Elec, Eprice and IndGdp has been tested using the residual-based procedure developed by Pedroni (1999, 2004) (see Appendix 1).12 The Pedroni group tests have a null of no cointegration for all countries of the panel against the alternative hypothesis of cointegration for at least one country. Table 3 (Appendix 2) reports the results. All Pedroni group statistics reject the null of no cointegration. These tests are based on the assumption of cross-sectional independence of the errors. As noted by Pedroni (2004), in order to eliminate some forms of cross-sectional dependence common time dummies can be included in the regression equation. As Table 3 shows, including time dummies we obtain the same results. The null of no cointegration is rejected by all group statistics. The Pedroni cointegration test statistics may su¤er from size distorsions when the time dimension of the panel is not signi…cantly large with respect to the cross sectional dimension second order moments. Pesaran (2003) suggests replacing extreme values of the test statistics by K1 or K2 such that Pr [-K1 < ti (N,T) < K2] is su¢ ciently large, namely in excess of 0.9999. As noted by Pesaran, this truncated test statistc allows to avoid size distorsions, especially in the case of models with residual serial correlations and linear trends. 1 1 Giulietti, Otero and Smith (2006) demonstrate that the Hadri test su¤er from size distorsions in the presence of cross-sectional dependence when N=50 and T=25. However, also their alternative Bootstrap Hadri Test su¤ers from size distorsions in the presence of cross-sectional dependence when N=50 and T=25. 1 2 Pedroni (2004) uses these cointegration tests for testing the weak form of purchasing power parity for the post-Bretton Woods period. In particular, he uses a panel of 25 countries for the period June 1973-December 1994 and reports the results for both annual, T=20, and monthly, T=246, data.

6

(Pedroni, 2004). Therefore, the same cointegration analysis has bee applied to seven subgroups of the panel with T>N. 13 These additional tests con…rm the initial results. The null of no cointegration is always rejected. However, the test of Pedroni rejects the null of no cointegration even if the residuals of a pooled OLS estimation of equation (3) are stationary only for one country. Therefore, to determine whether the residuals of each of the 49 cross-sections of equation (1) are stationary we have performed separate ADF, Phillips-Perron and KPSS unit root tests. These values demonstrate that the OLS residuals are stationary for a signi…cant portion of countries. In particular, there is evidence of non-stationarity in the residuals only for two countries, Canada and Hungary. In the presence of cointegrated time series, the OLS estimator converges to the true value at very fast rate T. However, in the situation where there is serial correlation or the regressors are not strictly exogeneous the estimate may be biased. In this situation the t-statistic is no longer asymptotically normal. Therefore, to apply the OLS methodology for the estimation of equation (1) could not be appropriated. As alternative procedure, we have used the groupmean panel Fully Modi…ed Ordinary Least Squares (FMOLS) method proposed by Pedroni (2000, 2001). 14 Formally, the FMOLS estimator for the i-th member is given by, F M OLS;i

= (Xi0 Xi )

1

(Xi0 yi

T )

(4)

where yi is the modi…ed dependent variable (adjusted for the part of the error that is correlated with the regressor) and is a parameter for autocorrelation adjustment. The group-mean FMOLS estimator (based on the between dimension of the panel) is the simple average of the individual FMOLS estimators:

GF M OLS

1

=N

N X

F M OLS;i

(5)

i=1

The group-mean FMOLS estimator allows for the heterogeneity of the cointegrating vectors. Therefore, the value of can be considered di¤erent for each individual of the panel. In order to eliminate some forms of cross-sectional dependence, we have also included in the regression common time dummies (Pedroni, 2000, 2001). Table 4 (Appendix 2) reports the estimation results. The group-FMOLS estimates suggest that - considering the entire panel of 49 countries- a positive and statistically signi…cant relationship exists between the changes in electric consumption and those in the share of industry. On the 1 3 We

have tested the presence of cointegration in 20 OECD countries, 16 European countries, 6 Transition countries, 23 Developing countries, 10 Latin American countries, 5 African countries, and 8 Asian countries. 1 4 This group-mean panel FMOLS estimator is based on the FMOLS approach developed by Phillips and Hansen (1990) for individual time series. The FMOLS estimator adjusts for the e¤ects of potential long-run endogeneity of the regressors and short-run dynamics of the errors.

7

contrary, a negative and statistically signi…cant relationship exists between the changes in electric consumption and those in electricity price. As noted by Pedroni (2000), the group-mean FMOLS estimator may su¤er from size distorsions in terms of over-rejection of the null (H0 = i = 0 for all i) when N is large relative to T.15 Therefore, we have estimated the same regression equation considering four subgroups of countries with T large relative to N. 16 As Table 4 shows, including or not common time dummies, these results are close to those obtained examining the entire panel. Only for the Developing countries, the relationship between the changes in electric consumption and those in electricity price becomes positive and non-statistically signi…cant in the presence of time dummies.

3

Results

In this section we present results based on country speci…c FMOLS estimators for equation (1).17 The descriptive statistics of our estimates are reported in Table 5. Our estimates show a generalized downward trend in share of the unrecorded economy over the period 1981-2005.

1 5 Using Monte Carlo simulations, Pedroni (2000) demonstrates that, in the bivariate case, the small sample distorsion of the group-mean fully modi…ed OLS estimator tends to be high when N > T, and decreases as T > N. This is a practical consequence of any …xed e¤ects model. 1 6 I have employed the FMOLS estimation for 20 OECD countries, 16 European countries, 22 European countries (including 6 Transition countries), and 23 Developing countries. 1 7 We have also computed electricity consumption series using both the full panel FMOLS coe¢ cients and separate subpanel coe¢ cients (OECD non European, European, and Developing countries) reported in Table 4. The three series computed for electricity usage are quite similar (results available upon request).

8

Table 5 Descriptive statistics of the MTE estimates 1981-2005 OECD Af rican Asian LatinAmerican T ransition 1981-1990 OECD Af rican Asian LatinAmerican T ransition 1990-2000 OECD Af rican Asian LatinAmerica T ransition 2000-2005 OECD Af rican Asian LatinAmerican T ransition

Mean 14.5 43.2 34.2 33.3 20.9 Mean 17 55.7 44.5 37.3 19.7 Mean 13.9 37.2 29.2 31.9 23.9 Mean 11 31.8 24.9 29.3 17.1

Std. 5.2 19,8 22.5 10.9 14 Std. 4.9 21.7 25.2 9.9 12.7 Std. 4.4 13.3 17.7 10.7 15.6 Std. 4.3 11.8 17.6 11.1 11.4

Dev.

Dev.

Dev.

Dev.

Min. 4.1 11 5.2 16 4.4 Min. 8.9 26.6 12.7 16.4 5.1 Min. 4.7 18.6 5.8 18.8 6.7 Min. 4.1 11.1 5.2 16 4.4

Max. 31.8 127.9 105.5 67.5 56.6 Max. 31.8 127.9 105.5 64.8 42.9 Max. 22.9 69.9 75.6 67.6 56.6 Max. 21.7 50.5 62.2 62.4 40.5

Source: own calculations

Several contributions (Johnson, Kaufmann and Zoido-Lobaton, 1998; Asea, 1996; Loayza, 1996) point out that the size of the unobserved sector is negatively related to the economic development level. We have found that- except for Austria and Tanzania- there is a negative and statistically signi…cant correlation between annual growth rates of o¢ cial GDP and annual changes of the share of unrecorded income (Table 6).

9

Table 6 Correlation coe¢ cients between the growth rates of o¢ cial GDP and changes of unrecorded income

Countries Australia Austria Belgium Botswana Brazil Bulgaria Canada Chile Colombia CostaRica CzechR: Denmark Egypt F inland F rance Germany Greece Guatemala HongKong Hungary Ireland Israel Italy Japan Korea

Corr -0,7* -0,02 -0,8* -0,8* -0,86* -0,84* -0,94* -0,84* -0,41* -0,77* -0,96* -0,59* -0,77* -0,76* -0,78* -0,86* -0,95* -0,45* -0,69* -0,94* -0,72* -0,67* -0,6* -0,8* -0,7*

Countries M alaysia M exico M orocco N etherlands N orway P anama P araguay P eru P hilippines P oland P ortugal Romania Singapore SlovakR: Spain SriLanka Sweden Switzerland T anzania T hailand T unisia UK U SA V enezuela

Corr -0,81* -0,9* -0,95* -0,83* -0,73* -0,83* -0,38* -0,81* -0,69* -0,97* -0,87* -0,87* -0,84* -0,83* -0,75* -0,5* -0,64* -0,87* -0,33 -0,85* -0,6* -0,75* -0,91* -0,96*

Note: A * indicates statistically signi…cant correlation.

We also computed cyclical gaps in the o¢ cial and unrecorded GDP …gures (Table 7). 18 In line with the theoretical model of Busato and Chiarini (2004) we …nd evidence of a double business cycle, where the correlation between the two gaps is negative and statistically signi…cant for a large number of countries.

1 8 The two gaps are obtained detrending the series of unobserved economy and o¢ cial GDP by using the Hodrick-Prescott …lter.

10

Table 7 Correlation coe¢ cients between the detrended series of o¢ cial GDP and unrecorded GDP Countries Australia Austria Belgium Botswana Brazil Bulgaria Canada Chile Colombia CostaRica CzechR: Denmark Egypt F inland F rance Germany Greece Guatemala HongKong Hungary Ireland Israel Italy Japan Korea

Corr -0,71* 0,1 -0,6* -0,73* -0,73* -0,36* -0,8* -0,34* -0,2 -0,3* -0,91* -0,04 -0,59* -0,53* -0,68* -0,69* -0,6* -0,24 -0,2 -0,74* -0,7* -0,45* -0,27 -0,65* -0,01

Countries M alaysia M exico M orocco N etherlands N orway P anama P araguay P eru P hilippines P oland P ortugal Romania Singapore SlovakR: Spain SriLanka Sweden Switzerland T anzania T hailand T unisia UK U SA V enezuela

Corr -0,36* -0,7* -0,83* -0,66* -0,25 -0,58* -0,20 -0,62* -0,33 -0,8* -0,62* -0,52* -0,81* -0,92* -0,61* -0,64* -0,50* -0,66* -0,2 -0,63* -0,19 -0,52* -0,61* -0,87*

Note: A * indicates statistically signi…cant correlation.

Finally, we have calculated the volatility of o¢ cial and total (observed and unobserved) output growth (Table 8). For a large number of countries we have found signi…cant di¤erences between the standard deviation of o¢ cial GDP growth rates and total output growth volatility.

11

Table 8 Output Growth Volatility- O¢ cial and Total GDP Countries Australia Austria Belgium Botswana Brazil Bulgaria Canada Chile Colombia CostaRica CzechR: Denmark Egypt F inland F rance Germany Greece Guatemala HongKong Hungary Ireland Israel Italy Japan Korea

3.1

O¢ cial GDP 1,89 1,15 1,27 4,58 2,99 5,51 2,2 5,15 2,21 3,47 7,56 1,60 2,70 2,84 1,20 1,56 2,17 2,27 3,93 3,78 3,22 2,72 1,31 1,98 3,75

Total GDP 1,37 1,54 0,94 2,5 1,93 3,4 1,83 3,39 2,37 2,41 7,01 1,35 2,96 2,24 0,95 1,12 1,49 2,85 3,39 2,03 2,60 2,11 1,08 1,57 2,67

Countries M alaysia M exico M orocco N etherlands N orway P anama P araguay P eru P hilippines P oland P ortugal Romania Singapore SlovakR: Spain SriLanka Sweden Switzerland T anzania T hailand T unisia UK U SA V enezuela

O¢ cial GDP 4,21 3,32 4,97 1,53 1,65 4 2,62 6,33 3,69 4,49 2,59 5,55 4,09 15,27 1,53 1,85 1,79 1,59 2,42 4,75 2,37 1,4 1,81 6,50

Comparison with ECM and MIMIC estimates

In the following we compare our results with those obtained using the standard Electricity Consumption Method (ECM) and the MIMIC approach. In Figure 4 (Appendix 3) we provide a comparison between the ECM and MTE estimates.The MTE estimates obtained by …ltering out separately the changes in electricity prices- MTE (P)- and changes in output compositionMTE (I)- are also reported. In some countries important di¤erences between the two methods arise as a consequence of the changing weight of the industry share, which is signi…cantly related to changes in electricity consumption. In fact we observe that in transition countries the standard Electricity Consumption Method underestimates the relative size of unobserved sector after the end of communism, when the industry share of GDP decreased. A similar di¤erence is detected in countries like Hong Kong, Italy and Japan, where the service sector as a percentage of GDP has signi…cantly increased during the last decades. By contrast, the development process in countries like Thailand corresponds to an

12

Total GDP 2,94 2,29 1,8 1,29 1,47 2,4 5,4 4,24 2,74 3,18 1,96 4,41 3,43 14,69 1,08 1,53 1,55 1,33 3,04 2,5 1,9 1,12 1,51 4

increase in the industry share of GDP. In this case the ECM overestimates the relative size of the unobserved sector. The relative price e¤ect in energy consumption seems to play a lesser role. We could …nd important di¤erences only for South Korea, where the ECM procedure underestimates the unrecorded income during the 1980s and overestimates it during the 1990s and the early 2000s. Our results are in sharp contrast with those obtained by Schneider (2004, 2005) adopting the MIMIC procedure. As reported in Figure 1 below, the MIMIC estimates show that for all subgroups of countries the relative size of unobserved sector has increased from 1990 to 2003. The discrepancy between our MTE and MIMIC results is particularly apparent if economies are examined individually (Figure 2, Appendix 3). In particular, the Schneider’s MIMIC estimates indicate that for each of the 49 countries the share of unrecorded income has continuously increased during the 1990s, whereas we obtain this result only for Guatemala and Paraguay.

13

Figure 1- MTE and MIMIC estimates of unrecorded economysubgroups of countries for the period 1981-2005 Anglosaxon countries

European countries

20

20

18

18

16

16

14

14

12

12

MTE 10 MIMIC

MTE 10 MIMIC

8

8

6

6

4

4

2

2

0

0

1981

1986

1991

1996

2001

1981

Countries of Latin Europe

1986

1991

1996

2001

Scandinavian countries

25

25

20

20

15

15

MTE

MTE

MIMIC

MIMIC 10

10

5

5

0

0 1981

1986

1991

1996

1981

2001

African countries

1986

1991

1996

2001

Asian countries 60

80 70

50

60

40 50 MTE 40

MTE 30 MIMIC

MIMIC 30

20

20

10 10

0

0 1981

1986

1991

1996

1981

2001

14

1986

1991

1996

2001

Latin American countries

Transition countries

45

30

40 25

35 30

20

25

MTE

20

MIMIC

15

MTE 15 MIMIC 10

10 5

5 0

0

1981

1986

1991

1996

2001

1981

1986

1991

1996

2001

Source: MTE estimates = own calculations; MIMIC estimates = Schneider (2004, 2005).

3.2

Interpreting the dynamics of the shadow economy: preliminary evidence

In several studies it has been found that many factors can a¤ect the unobserved economic activity over time [Johnson, Kaufmann, Zoido-Lobaton (1998), Friedman, Johnson, Kaufmann, Zoido-Lobaton (2000). In particular, the size of the unobserved sector is negatively related to the economic development level. A higher level of development goes together with a greater capacity to pay and collect taxes, as well as a higher relative demand for income elastic public goods and services. On the contrary, the …scal burden is expected to in‡uence the unobserved economy positively. In fact, a higher burden increases the costs of operating in the observed sector. Similarly, strong regulations- especially labour regulations- are an incentive to operate in the unobserved sector. The tax compliance literature has shown the tendency that self-employed workers, such as farmers, are more inclined to evade taxes. Finally, international trade is transparent and more di¢ cult to hide in the unobserved economy. Thus, greater trade openness may lead to a lower unobserved income share. Therefore, according to this literature, structural changes in the unobserved economic activity may be related to changes in the observed economic performance. In the following, we exploit the time-series dimension of our panel to identify structural breaks in the country-speci…c pattern in relative size of the unrecorded economy. Such breaks are then related to speci…c episodes of institutional change, economic reform and economic crisis. To detect structural breaks in the MTE series of unobserved income, we have calculated the Chow test for every year within the period 1984-2003. The plotted Chow F-statistics of each country are reported in Figure 3 (Appendix 3). The null hypotheses of no structural break is rejected for values of F 2; 6. As can be readily seen, there is no evidence of structural changes in the series of unrecorded economy only for Belgium, Czech Republic, Italy and Netherlands. Consider the transition economies where the relative size of unobserved sector has increased immediately after the end of communism (1989). This in15

crease may be related to the economic and institutional disarray during the …rst years after the collapse of communism. The unrecorded income has then begun to decrease since the 1990s, possibly following the consolidation of both the market economy and of the state. For instance, in Poland, the unrecorded income started to decrease in the early 1990s. There has been a structural break in the series of unobserved sector within the period 1991-1997, with the best structural break point in 1994. This structural change may be related to the economic reforms introduced by the Polish government in 1990 directed to remove price controls, to eliminate most subsidies to industry, to open markets to international competition. Similarly, in Hungary, the unobserved economy has increased from 1989 to 1993 and then decreased. The Chow test rejects the null of no structural breaks for the period 1992-1998, with a peak in 1992. Also for Hungary the reduction in the relative size of unobserved sector may be associated to the positive e¤ects of market reforms- based on price and trade liberation measures, tax system and market-based banking system reforms- begun by the government in 1990. In Romania, after an initial increase during the years of recession, the unobserved economy started to decrease in 1993. According to the Chow test, 1993 corresponds with the best structural beak point in the series of unrecorded income. In the same year the Romanian economy started to grow. This recovery was stimulated by government policies based on privatization and trade liberalization. Moreover, Romania signed an association agreement with the EU in 1992 and a free trade agreement with the European Free Trade Association (EFTA) in 1993, codifying its access to European markets and creating the basic framework for further economic integration. In the Slovak Republic there has been a structural break in the series of unrecorded economy within the period 1992-1994, with a peak in 1992. This structural change corresponds to a rapid decrease in the relative size of Slovak unobserved sector, which may be explained by the economic reform- based on privatization, price liberalization and trade openness- begun by Czechoslovakia immediately after the Velvet Revolution in 1989. In Bulgaria, the share of unobserved economy has started to decrease in the late 1990s. The null hypothesis of no structural break is rejected for the period 1998-1999, with a peak in 1998. Even in this case, the structural change may be attributed to the market-oriented economic reforms- begun by the Bulgarian government in 1997- which have signi…cantly contributed to the overall economic recovery in this country. In particular, these policies were characterized by privatization, price liberalization, reform of the country’s social insurance programs and reforms to strengthen contract enforcement. On the contrary, in Czech Republic, the relative size of unobserved sector has ceased to increase since 1992 but without any signi…cant reduction. In fact, there is no evidence of structural breaks in the series of unrecorded income for this country. The absence of a signi…cant decrease in the Czech unobserved income may be therefore explained by the di¢ culties of this country to complete the economic transformation. Turning to the African countries, we begin our analysis with Egypt. After a dramatic drop in the years 1981-1982, the share of unrecorded income has remained stationary until the mid 1980s and decreased since 1987. According 16

to the Chow test results, 1987 is the best structural break point in the series. This reduction may be associated to the improvements in Egypt’s economic performance boosted by the macro-economic reform program prepared by the government in 1986. In particular, this reform relaxed many price controls, reduced subsidies, cut taxes, and partially liberalized trade and investment. In Tunisia, there is evidence of two structural changes in the series of unobserved economy. The best structural break points are 1987 and 1997. These two structural changes signal the beginning of two decreasing phases of the Tunisian unrecorded income. In 1986 the Tunisian government launched a structural adjustment program directed to liberalize prices, reduce tari¤s, and reorient the country towards a market economy. Moreover, in 1996 Tunisia entered an association agreement with the European Union which removed tari¤s and other trade barriers on most goods. Finally, the Chow test statistics show that also in Tanzania there has been a structual break in the series of unobserved income in the years 1984-1985. In 1984 the relative size of Tanzania’s unobserved economy begun to decrease. This reduction may be associated to the positive e¤ects of an adjustment program undertook by the government in the mid 1980s to dismantle socialist economic controls and encourage more active participation of the private sector in the economy. The program included a package of policies which liberalized the trade regime, removed most price controls, eased restrictions on the marketing of food crops, freed interest rates, and initiated a restructuring of the …nancial sector. We consider now the Asian countries. In Israel there is evidence of two structural breaks in the series of unobserved economy in the years 1985 and 1990. The …rst structural change corresponds to a rapid decrease in the relative size of the unobserved sector during the period 1985-1987 whereas the year 1990 indicates the beginning of a longer decreasing phase of the Israeli unrecorded income. The initial decrease in the unobserved income may be related to a successful economic stabilization plan implemented by the government in 1985 and the subsequent introduction of market-oriented structural reforms. In fact, after a period characterized by a catastrophic economic situation- worsened by the 1983 bank stock crisis- this program has reinvigorated the Israeli economy. In the Philippines, there has been a structural break in 1986. This structural change corresponds with the beginning of a decreasing phase of the unrecorded income after a rapid increse during the …rst half of the 1980s. In 1986 the government decided to restructure the tax system. The 1986 tax reform program was designed to simplify the tax structure, make it more neutral, broaden its base, and raise additional revenue. The MTE estimates indicate di¤erent trends in Thai unrecorded economic activity. The unobserved income has signi…cantly decreased from 1985 to 1995, increased during the years 1996-1998 and restarted to decrease at the end of the 1990s. The Chow tests indicate the presence of two structural changes in the MTE series of unobserved economy within the periods 1986-1991 (with a peak in 1988) and 1996-1997 (with a peak in 1996). The …rst decrease in the relative size of Thai unobserved sector may thus be related to the mid 1980s reform- aimed at opening up the economy to international trade- implemented by the government. This reform is considered 17

one of the main causes of the Thai economic boom that started in the second half of the 1980s. The second break may be explained with the reforms that the government undertook to stimulate the recovery after the 1997 …nancial crisis. These included privatization plans and acts directed to promote free trade. Examining Latin American countries, we …nd that in Chile there has been a structural change during the period 1984-1991, with a peak in 1987. This reduction may be associated to the free-market reforms- aimed at privatization, elimination of protectionist trade barriers and support of foreign investmentsundertaken by the military government after the 1982 Latin American debt crisis. In Mexico, the unrecorded economic activity has followed di¤erent trends. The relative size of unobserved economy has signi…cantly increased during the 1980s, starting to decrease in 1989. Then, the unrecorded income increased in the years 1994-1995 and then decreased again during the second half of 1990s. According to the Chow tests, there have been two structural breaks with a peak in 1990 and 1996, respectively. The reduction in the relative size of Mexican unobserved sector at the end of 1990s may be related to the adoption of a national development plan for 1989-94 which seeked to encourage private investment through denationalization of state enterprises and deregulation of the economy. The economy began to grow in the early 1990s and foreign investment was boosted by United States rati…cation of the North American Free Trade Agreement (NAFTA) in 1994. These improvements in Mexican economic performance may thus explain the reduction in the unobserved economy during the years 1989-1994. The Chow test statistics indicate that, in Peru, there has been a structural break in the years 1993-1994. This structural change corresponds to a signi…cant reduction in the relative size of Peruvian unobserved sector. In the 1990s, the Peruvian government undertook a process of liberalization which put an end to price controls, discarded protectionism, eliminated restrictions on foreign direct investment and privatized most state companies. Since 1990, the Peruvian economy has undergone considerable free market reforms, from legalizing parts of the informal sector to signi…cant privatization in the mining, electricity and telecommunications industries. Thanks to strong foreign investment and the cooperation between the government and the IMF and World Bank, growth was strong in 1994-97. Finally, we consider the OECD countries. In Australia, the relative size of unobserved sector started to decrease in the early 1980s with the best structural break point in 1984. This reduction may be associated to the Australia’s economic improvements due to deregulation of the Australian economy started in 1983. Tari¤s were progressively cut, the Australian dollar was ‡oated and government-run enterprises were privatised. According to the Chow test results, the null hypotheses of absence of structural breaks in the series of the Irish unrecorded income is rejected for the years 1987 and 1988, with a peak in 1988. This structural change may be related to the …rst of a series of national economic programmes- started by the government in 1987- designed ease tax burdens, reduce government spending and reward foreign investment. In Portugal there has been a structural break in the series of unrecorded income within the period 1985-1988, with a peak in 1987. Finally, in Spain, the unobserved 18

income started to decrease in the second half of 1980s with the best break point in 1985. This reduction may be associated to a social and economic program adopted by Spain in 1985 directed to reduce the labor market rigidities and increase the employment level. This policy has been followed by other two labor market reforms in 1994 and 1997, respectively. These reforms are considered the main causes of the signi…cant increase in the Spanish employment level during the last two decades.

4

Conclusions

The …rst conclusion from our estimates is that for most countries the relative size of the unobserved economy has decreased during the last decades. This result is broadly consistent with the country-speci…c economic performances and with key episodes of economic reform and liberalization but in constrast with the recent and highly criticized values obtained by Schneider (2004, 2005) with the MIMIC method. We also …nd that the relative size of the unobserved economy falls with economic growth and exhibits a cyclical pattern which is inversely related with that of the o¢ cial economy. In contrast with popular methods such as CDA and MIMIC, our estimates of the unrecorded economy are not based on theoretical priors concerning the role of taxes and market regulation. As such, they seem better suited for an analysis of the causes of this phenomenon. Consider for instance the set of variables that typically identify a country’s institutional quality. These are also typically related to the size of the public sector and to market regulation. Thus measures of the unrecorded economy based on these two latter variables are by de…nition bound to exhibit a certain relation with measures of institutional quality. Further research should reconsider the issue.

19

References [1] Bagachwa, M.S.D. and A. Nasho (1995), Estimating the Second Economy in Tanzania, World Development, 23 (8), pp. 1387-1399. [2] Bai, J. and S. Ng (2000), Determining the number of factors in approximate factor models, Econometrica, 70, pp. 191-221. [3] Bajada, C. and F. Schneider (2005), The Shadow Economies of the Asia-Paci…c, Paci…c Economic Review, 10 (3), pp. 379-401. [4] Bartlett, B. (1990), The Underground Economy: Achilles Heel of the State?, Economic A¤ airs, 10, pp. 24-27 [5] Breitung, J. (2000), The Local Power of Some Unit Root Tests for Panel Data, in B. Baltagi (ed.), Advances in Econometrics, Vol. 15: Nonstationarity Panels, Panel Cointegration, and Dynamic Panels. Amsterdam: JAI Press, pp. 161-78. [6] Breitung, J. and H. Pesaran (2005), Unit Root and Cointegration in Panels, CESIFO Working Paper, No. 1565. [7] Breusch, T. (2005), Estimating the Underground Economy using MIMIC Models, The Australian National University, Canberra, ACT0 200. [8] Busato, F. and B. Chiarini (2004), Market and Underground Activities in a Two-Sector Dynamic Equilibrium Model, Economic Theory, Springer, 23 (4), pp. 831-861. [9] Cagan, P. (1958), The Demand for Currency Relative to the Total Money Supply, Journal of Political Economy, 66 (3), pp. 302-328. [10] Cebula, R.J. (1997), “An Empirical Analysis of the Impact of Government Tax and Auditing Policies on the Size of the Underground Economy: The Case of the United States, 1973-1994”, The American Journal of Economics and Sociology, 56 (2), pp. 73-186. [11] Chang, Y. (2000), Bootstrap Unit Root Tests in Panels with CrossSectional Dependency, Working Paper, Rice University. [12] Chong, A. and M. Gradstein (2006), Inequality and Informality, Journal of Public Economics, 91, pp. 159-179. [13] Gengenbach, C., Palm, F.C. and J. Urbain, (2006), Cointegration Testing Panels with Common Factors, Oxford Bulletin of Economics and Statistics, Department of Economics, University of Oxford, 68, pp. 683-719. [14] Dell’Anno, R. and F. Schneider (2003), The Shadow Economy of Italy and Other OECD Countries: What Do We Know?, Journal of Public Finance and Public Choice, 21 (2-3), pp. 97-120. 20

[15] Eilat, Y. and C. Zinnes (2002), The Shadow Economy in Transition Countries: Friend or Foe? A Policy Perspective, World Development, 30 (7), pp. 1233-1254. [16] Engle, R. F. and C.W.J. Granger (1987), Cointegration and Error Correction: Representation, Estimation and Testing, Econometrica, 35, pp. 143-159. [17] Feige, E. (1990), De…ning and Estimating Underground and Informal Economies: The New Institutional Approach, World Development, 18 (7). [18] Feige, E. (1996), Overseas Holdings of U.S. Currency and the Underground Economy, in Pozo, S. (ed.), Exploring the Underground Economy. Kalamazoo, Michigan, pp. 5-62. [19] Feige, E. and I. Urban (2003), Estimating the Size and Growth of Unrecorded Economic Actity in Transition Countries: A Reevaluation of Electric Consumption Method Estimates and their Implication, Working Paper, William Davidson Institute, No. 636. [20] Friedman, E., Johnson, S., Kaufmann, D. and P. Zoido-Lobaton (2000), “Dodging the Grabbing Hand: the Determinants of Uno¢ cial Activity in 69 Countries”, Journal of Public Economics, 76, pp. 459493. [21] Garcia, G. (1978), The Currency Ratio and the Subterranean Economy, Financial Analysts Journal, 69 (1), pp. 64-66. [22] Giles, D.E.A. (1999), Modeling the Hidden Economy in the Tax-gap in New Zealand, Working Paper, Department of Economics, University of Victoria. [23] Giles, D.E.A. and L.M. Tedds (2002), Taxes and the Canadian Underground Economy, Canadian Tax Foundation: Toronto. [24] Giulietti, M., Otero, J. and J. Smith (2006), Testing for stationarity in heterogeneous panel data in the presence of cross section dependence, Warwick Economics Research Papers Series 758, Warwick University, United Kingdom. [25] Granger, C.W.J., and P. Newbold (1974), Spurious Regression in Econometrics, Journal of Econometrics, 2, pp. 111-120. [26] Gutmann, P. M. (1977), The Subterranean Economy, Financial Analysts Journal, 34 (1), pp. 24-27. [27] Hadri, K. (2000), Testing for stationarity in heterogeneous panel data, The Econometric Journal, 3, pp. 148-161.

21

[28] Helberg, C. and H. Knepel (1988), How Big is the Shadow Economy? A Re-Analysis of the Unobserved-Variable Approach of B.S. Frey and H. Weck-Hannemann, European Economic Review, 32, pp. 965-976. [29] Hill, R. (2002), The Underground Economy in Canada: Boom or Bust?, Canadian Tax Journal, 50 (5), pp. 1641-1654. [30] Hill, R. and M. Kabir (1996), Tax Rate, the Tax Mix, and the Growth of the Underground Economy in Canada: What Can we Infer?, Canadian Tax Journal/ Revue Fiscale Canadienne, 44 (6), pp. 1552-1583. [31] Im, K.S., Pesaran, M.H., and Y. Shin (2003), Testing for Unit Roots in Heterogeneous Panels, Journal of Econometrics, 115, pp. 53-74. [32] Isachsen, A. J. and S. Strom (1985), The Size and Growth of the Hidden Economy in Norway, Review of Income and Wealth, 31 (1), pp. 21-38. [33] Johnson, S., Kaufmann, D. and A. Schleifer (1997), The Uno¢ cial Economy in Transition, Brookings Papers of Economic Activity, (2 1997), pp. 159-239, Washington, D.C. [34] Johnson, S., Kaufmann, D. and P. Zoido-Lobaton (1998), Regulatory Discretion and the Uno¢ cial Economy, American Economic Review, 88 (2), pp. 387-392. [35] Kao, C. (1999), Spurious Regression and Residual-Based Tests for Cointegration in Panel Data, Journal of Econometrics, 90, pp. 1-44. [36] Kaufmann, D. and A. Kaliberda (1996), Integrating the Uno¢ cial Economy into the Dynamics of Post-Socialist Economies, World Bank Policy Research Working Paper, No. 1691. [37] Klovland, J. (1984), Tax Evasion and the Demand for Currency in Norway and Sweden: Is there a Hidden Relationship?, Scandinavian Journal of Economics, 86 (4), pp. 423-39. [38] Koreshkova, T. A. (2006), “A Quantitative Analysis of In‡ation as a Tax on the Underground Economy”, Journal of Monetary Economics, 53 (4), pp. 773-796. [39] Kwiatkowski, D., Phillips, P.C.B., Schmidt P. and Shin Y. (1992), “Testing the Null Htpothesis of Stationarity Against the Alternative of a Unit Root: How Sure are we that Economic Time Series have a Unit Root?”, Journal of Econometrics, 54, pp. 150-178. [40] Lacko, M. (1996), Hidden Economy in East-European Countries in International Comparison, Laxenburg: International Institute for Applied Systems Analysis (IIASA), Working Paper.

22

[41] Lacko, M. (1998), The Hidden Economies of Visegrad Countries in International Comparison: A Household Electricity Approach, In: Halpern, L. And Wyplosz, Ch (eds.), Hungary: Toward a Market Economy, Cambridge (Mass.): Cambridge University Press, pp. 128-152. [42] Lacko, M. (1999), Hidden Economy- An Unknown Quantity? Comparative Analysis of Hidden Economies in Transition Countries in 1989-1995, Working Paper, vol. 9905, Economics Department, University of Linz, Austria. [43] Lacko, M. (1999), Hidden Economy- An Unknown Quantity? Comparative Analysis of Hidden Economies in Transition Countries in 1989-1995, Economics of Transition, 8 (1), pp. 117-149. [44] Loayza, N. (1996), The Economics of the Informal Sector: A Simple Model and Some Empirical Evidence from Latin America, Carnegie-Rochester Conference Series on Public Policy, 45, pp. 129-162. [45] Levin, A., Lin, C.F. and C. Chu (2002), Unit Root Tests in Panel Data: Asymptotic and Finite-Sample Properties, Journal of Econometrics, 108, 1-24. [46] Maddala, G.S. and S. Wu (1999), A Comparative Study of Unit Root Tests with Panel Data and a New Simple Test, Oxford Bulletin of Economics and Statistics, 61, pp. 631-652. [47] Moon, H.R. and B. Perron (2003), Testing for a Unit Root in Panels with Dynamic Factors. Mimeo, University of California at Santa Barbara. [48] OECD (2002) Measuring the Non-Observed Economy.A Handbook [49] Park, T. (1979), Reconciliation between Personal Income and Taxable Income. Mimeo, Washington D.C., Bureau of Economic Analysis. [50] Pedroni, P. (1999), Critical Values for Cointegration Tests in Heterogeneous Panels with Multiple Regressors, Oxford Bulletin of Economics and Statistics, 61, pp. 653-70. [51] Pedroni, P. (2000), Fully Modi…ed OLS for Heterogeneous Cointegrated Panels, Advances in Econometrics, 15, pp. 93-130. [52] Pedroni, P. (2001), Purchasing Power Parity Tests in Cointegrated Panels, Review of Economics and Statistics, 83, pp. 727-731. [53] Pedroni, P. (2004), Panel Cointegration; Asymptotic and Finite Sample Properties of Pooled Time Series Tests with and Application to the PPP Hypothesis, Econometric Theory, 20, pp. 597-625.

23

[54] Pesaran, H., (2003), A Simple Panel Unit Root Test in the Presence of Cross Section Dependence, Cambridge Working Papers in Economics, No. 0346, Faculty of Economics (DAE), University of Cambridge. [55] Pesaran, H. (2007), A Simple Panel Unit Root Test in the Presence of Cross Section Dependence, Journal of Applied Econometrics, 22 (2), pp. 265-312. [56] Phillips, P. and B. Hansen (1990), Statistical Inference in Instrumental Variables Regression with I(1) Processes, Review of Economic Studies, 57, pp. 99-125. [57] Phillips, P. and D. Sul (2003), Dynamic Panel Estimation and Homogeneity Testing Under Cross Section Dependence, Econometric Journal, 6, pp. 217-259. [58] Schneider, F. (1997), Empirical Results for the Size of the Shadow Economy of the Western European Countries over Time, Working Paper, No. 9710, University of Linz, Austria. [59] Schneider, F. and Enste D. (1998), Increasing Shadow Economies all Over the World- Fiction or Reality: A Survey of the Global Evidence of Its Size and of Its Impact from 1970 to 1995, IMF and University of Linz, August 21. [60] Schneider, F. and Enste D. (2000), Shadow Economies: Size, Causes and Consequences, Journal of Economic Literature, 38, pp. 77-114. [61] Schneider, F. (2004), The Size of the Shadow Economies of 145 Countries all over the World: First Results over the Period 1999 to 2003, IZA DP No. 1431. [62] Schneider, F. (2005), Shadow Economies around the World: What do we Really Know?, European Journal of Political Economy, 21, pp. 598-642. [63] Smith, R. S. (2002), The Underground Economy: Guidance for Policy Makers?, Canadian Tax Journal, 50 (5), pp. 1655-1661. [64] Tanzi, V. (1980), The Underground Economy in the United States: Estimates and Implications, Banca Nazionale del Lavoro, 135 (4), pp. 427-453. [65] Tanzi, V. (1983), The Underground Economy in the United States: Annual Estimates, 1930-1980, IMF- Sta¤ Papers, 30 (2), pp. 283-305. [66] Westerlund, J. and D. Edgerton, (2007), A Panel Bootstrapping Cointegration Test, Economics Letters, Elsevier, 97 (3), pp. 185-190.

24

[67] Williams, C.C. and J. Windebank (1995), “Black Market Work in the European Community: Peripheral Work f Peripheral Localities?”, International Journal of Urban and Regional Research, 19 (1), pp. 23-39.

25

Appendix 1 Panel unit root tests

The Im, Pesaran and Shin test allows for panel heterogeneity and individual unit root processes; i.e., i may vary across cross-sections. As noted in Im, Pesaran and Shin (2003) , the null hypothesis of unit root is tested using the basic Augmented Dickey Fuller (ADF) speci…cation: yit =

i yit 1

+

P

ij

yit

j

+ Xit + "it

(6)

where Xit represents the exogenous variables included in the model, for example the …xed e¤ects or the individual time trends. Given the heterogeneity assumption, the null and the alternative hypotheses for the tests can be written as: H0 :

i

=0

H1 :

i

<0

for all i for i

1

where i = i 1. Under the null hypothesis, each time series has a unit root and it is therefore non stationary. On the contrary, under the alternative at least one series of the panel is stationary1 2 . The Im, Pesaran and Shin test is based on the hypothesis that the error terms are independent across cross-sections (First Generation test). This is a standard assumption in panel-data models. A test for unit roots in heterogeneous panels with cross-section dependence is proposed by Pesaran (2003, 2007)3 4 . This Cross Section Augmented Dickey Fuller (CADF) test is based on the following linear regression: yit =

i yit 1

+

i yt 1

+

i

yt + Xit + "it

1 After

(7)

estimating the separate ADF regressions, the average of the t-statistics for i from the individual ADF regressions is adjusted to arrive at the desired test statistic. Im, Pesaran and Shin (2003) show that a properly standardized statistic has a standard normal distribution. 2 Using Monte Carlo simulations, Im, Pesaran and Shin (2003) demonstrate that- with or without serially correlated errors and individual time trends- the empirical size of their test statistic is very close to the nominal size when T=25 and N=50. They also indicate that their panel test is more powerful than individual ADF unit root tests. 3 Pesaran (2003) indicate that the Im, Pesaran and Shin test is distorted in the presence of cross section dependence. In particular, the extent of over-rejection of the test depends on the degree of cross section dependence. 4 Another approach based on a bootstrap for cases in which the time series dimension is not as long is examined in Chang (2000). Most recently, papers by Bai and Ng (2002), Moon and Perron (2003), and Phillips and Sul (2003) have studied the possibility of using various factor model approaches to modeling the cross-sectional dependence for the purpose of panel unit root tests.

26

where yt =

(

P

yit ) N

and (

yt =

P

yit ) N

= yt

yt

1

To eliminate the cross-section dependence, the standard ADF regressions are augmented with the cross section averages of lagged levels and …rst di¤erences of the individual series. The null and the alternative hypotheses can be written as follows:

H0 :

i

=0

H1 :

i

<0

for all i

for i

1

The null hypothesis assumes that all series are non-stationary. Analogous to Im, Pesaran and Shin test, Pesaran test is consistent under the alternative that only a fraction of the series is stationary5 6 . So far, we have analyzed panel unit root tests that are based on the null hypothesis of non stationarity. An alternative test for stationarity in heterogeneous panel data is presented by Hadri (2000). This Residual Based Lagrange Multiplier test is based on the following speci…cations: yit = fit + "it

(8)

and yit = fit +

it

+ "it

(9)

where fit is a random walk: fit = fit

1

+ uit

5 Parallel to Im, Pesaran and Shin test, the Pesaran test is based on the mean of individual CADF t-statistics of each unit in the panel. Pesaran (2003) shows that a properly standardized statistic is distributed standard normal under the null hypothesis of non stationarity. 6 Using Monte Carlo simulations, Pesaran (2003) demonstrate that- in the presence of high cross section dependence- the nominal distribution provides a good approximation to the empirical distribution of his test statistic when 20 T 30 and N=50. Only with serially correlated errors in the absence of individual time trends the test tends to slightly under-reject the null of unit root.

27

and "it and uit are mutually independent disturbances. The test statistics are obtained using the residuals from individual regressions of yit with respect to an intercept or a linear trend7 8 . The Hadri test has a null of stationarity. The series may be therefore stationary around a deterministic level, speci…c to the unit (i.e. …xed e¤ect), or around a unit-speci…c trend. On the opposite- parallel to Im, Pesaran and Shin and Pesaran tests- under the alternative at least one series is not stationary9 10 .

Pedroni cointegration tests

The Pedroni test is based on Engle and Granger (1987) residual-based cointegration test. The Engle and Granger test is based on an examination of the residuals of a spurious regression performed using non-stationary variables. If the variables are cointegrated then the residuals should be stationary. On the other hand if the variables are not cointegrated then the residuals will be nonstationary. Pedroni (1999, 2004) extends the Engle-Granger framework to tests involving panel data. The Pedroni test is based on the following linear model: yit =

i

+

i

+

1i x1i;t

+

2i x2i;t

+ ::: +

M i xM i;t

+ "it

(10)

The variables y and x (the vector of regressors) are assumed to be integrated of order one. The parameter i is heterogeneous and indicate cross-section speci…c intercepts. Heterogeneous coe¢ cients and linear time trends are also speci…ed. Finally, the error terms are mutually independent disturbances. Under the null hypotheses of no cointegration, the residuals "it will be nonstationary. The general approach is to obtain the residuals of a pooled OLS estimation of equation (10) and to test their non-stationarity by running either the auxiliary regression: it

=

i it 1

+

it

(11)

or the augmented version of the pooled speci…cation: 7 The

Hadri residual-based test is based on the squared partial sum process of residuals from a demeaning (detrending) model of level (trend) stationary (Hadri, 2000). 8 Hadri (2000) de…nes three possible test statistics. The error process may be assumed to be homoskedastic across the panel, or heteroskedastic across units. Moreover, serial dependence in the disturbances can also be taken into account. 9 The lagrange multiplier test statistic is distributed as standard normal under the null (Hadri, 2000). 1 0 As noted by Hadri (2000), the Monte Carlo simulations indicate that- without time trendsthe nominal distribution provides a good approximation to the empirical distribution of his test statistic when T=25 and N=50. Including linear time trends, the test tends to slightly under-reject the null of stationarity.

28

it

=

i it 1

+

X

ij

it j

+

(12)

it

for each cross-section. Pedroni describes various methods of constructing statistics for testing the following null hypothesis of no cointegration for all the individuals of the panel:

H0 :

i

=1

for all i

The seven test statistics developed by Pedroni are divided in two groups: the Panel Statistics (or within-dimension test statistics) and the Group Statistics (or between-dimension test statistics)11 12 . For the panel statistics, the alternative hypothesis is homogeneous and can be written as follows:

H0 :

i

=

<1

for all i

On the contrary, for the group statistics the alternative hypothesis is heterogeneous and can be de…ned as follows:

H0 :

i

<1

for all i

1

If the null is rejected in the panel case, then the variables are considered cointegrated for all individuals. On the other hand, if the null is rejected in the group case, then cointegration among the relevant variables exists for at least one country13 . 1 1 Pedroni (1999, 2004) shows that the standardized statistics are asymptotically normally distributed. 1 2 Using Monte Carlo simulations, Pedroni (2004) indicates that- in the bivariate case- when N=20 and T<50, the group t-statistic and panel t-statistic tend to be oversided, whereas the panel v-statistic, panel rho-statistic and group rho-statistic tend to be undersized. With N …xed at N=20 and T=150, all these statistics converge to the nominal size. Moreover, when T=250 and N=10, the panel v-statistic and panel rho-statistic converge from above, whereas the other statistics are already close to the nominal size. Speci…cally, the nominal sizes range from around 4.5% to 8.5%. With T …xed at T=250 and N=20 they range from around 5% to 7% (the size distorsion is minimal) and when N=50 they range from around 3% to 6%. Finally, Pedroni (2004) notes that, with N …xed at N=20, the group rho-test reaches 100% power at T=70, and all the other statistics achieves 100% power at T=50. Assuming an alternative close to the null (the autoregressive coe¢ cients of the error terms close to 1), larger values for T are required. 1 3 As noted by Pedroni (1999, 2004), if the data generating process is assumed to require that all individuals of the panel be either uniformly cointegrated or uniformly not cointegrated, then the natural interpretation for the alternative hypothesis is H1 : “all the individuals are cointegrated”. On the other hand, if the data generating process is assumed to permit individual members of the panel to di¤er in whether or not they are cointegrated, then the natural interpretation for the alternative hypothesis should be H1: “a signi…cant portion of the individuals are cointegrated”.

29

The Pedroni test is based on the assumption of cross-sectional independence. The individual processes are assumed to be independent and identically distributed cross-sectionally. In order to accomodate some forms of cross-sectional dependency, Pedroni (2004) proposes to demean data with respect to common time e¤ects. This approach assumes that the disturbances for each member of the panel can be decomposed into common disturbances that are shared among all members of the panel and independent idyosincratic disturbances that are speci…c to each member14 .

1 4 As noted by Pedroni (2004), for many cases this approach may be appropriate, as, for example, when common business cycle shocks impact the data for all individuals of the panel together. In other cases, additional cross-sectional dependencies may exist in the form of relatively persistent dynamic feedback e¤ects that run from one country to another and that are not common across countries, in which case common time e¤ects will not account for all the dependency. If the time series dimension is long enough relative to the cross-sectional dimension, the one practical solution in such cases may be to employ a GLS approach based on the estimation of the panel-wide asymptotic covariance for the weighting matrix. Most recently, Gengenbach, Palm and Urbain (2006) propose a common factor structure to model the cross-sectional dependence for panel no-cointegration tests. Moreover, a bootstrap test for the null hypothesis of cointegration in panel data is presented by Westerlund and Edgerton (2007).

30

Appendix 2 Table 1 Panel unit root test H0 : all 49 timeseries in the panel are non-stationary processes Lag selection: …xed at 1 =individual linear trends EC PE I EC PE I

IPS -14.7* -13.8* -15.7* -29.6* -32.7* -30.8*

PES -11.8* -9.4* -12.5* -24.2* -14.3* -23.8*

IPS( ) -13.1* -16.3* -11.9* -25.5* -29.5* -26.7*

PES( ) -10.3* -5.2* -9.4* -20.9* -9.6* -20.4*

Note: IPS = Im, Pesaran, Shin (2003), PES= Pesaran (2003, 2007). The statistics are asymptotically distributed as a standard normal with a left hand side rejection area. A * indicates the rejection of the null hypothesis of nonstationarity at least at the 5 percent level of signi…cance. Estimations undertaken with EViews6 and Stata10 (only for the Pesaran test).

31

Table 2 Hadri panel stationary test H0 : all 49 timeseries in the panel are stationary processes Homo: homoskedastic disturbances across units Hetero: heteroskedastic disturbances across units SerDep: controlling for serial dependence in errors =individual linear trends EC EC EC PE PE PE I I I EC EC EC PE PE PE I I I

Homo Hetero SerDep Homo Hetero SerDep Homo Hetero SerDep Homo Hetero SerDep Homo Hetero SerDep Homo Hetero SerDep

Z( ) 1.9* 4.8* 3.4* 15.5* 10.9* 7.4* -1.3 0.9 -0.3 -6.3 -5.8 -1.9 -5.5 -4.3 7.9* -6.6 -5.9 -3.1

Z( ) 0.4 4.5* 5.7* 4.6* 7.1* 17.4* 2.8* 3* 6.6* -7.7 -6.9 3.4* -7.8 -6.5 28* -7.6 -6.9 3.6*

Note: The statistics are asymptotically distributed as a standard normal with a right hand side rejection area. A * indicates the rejection of the null hypothesis of stationarity at least at the 5 percent level of signi…cance. Estimations undertaken with Stata10.

32

Table 3 Pedroni residual-based cointegration test H0 : no cointegration Trend assumption: heterogeneous intercepts Lag selection: …xed at 1 Group statistics Rho statistic P P statistic ADF statistic

No time dummies -10.9* -24.2* -22.4*

Time dummies -10.3* -25.1* -22.9*

Note: All reported values are asymptotically distributed as a standard normal. Panel statistics are weighted by long variances. The Pedroni tests are left-sided. A * indicates the rejection of the null of unit root or no cointegration at the 0.05 level of signi…cance. Estimations undertaken with Rats7.

Table 4 FMOLS estimation

T D = Time dummies Dep Var EC Entire panel OECD European European* Developing

I 0.89* 0.72* 0.81* 0.84* 1.04*

(7.1) (3.7) (4) (6.7) (3.7)

PE -0.09* (-5.9) -0.1* (-4.7) -0.1* (-4.4) -0.1* (-4.9) -0.09* (-3.1)

I (T D) 0.84* (6.7) 0.78* (4.7) 0.82* (4.2) 0.78* (6.8) 0.81* (3.6)

P E (T D) -0.02* (-2.9) -0.05* (-2.8) -0.06* (-2.4) -0.07* (-4.2) 0.06 (1.5)

Note: t-stats (in parenthesis) are for H0 : i = 0 for all i vs H1 : i 6= 0. Estimations undertaken with Rats7. European* indicate European countries including Transition economies.

33

Appendix 3 Figure 2- MTE and MIMIC estimates of unrecorded economy- 49 countries for the period 1981-2005 Australia

Austria 18,0

25,0

16,0 20,0

14,0 12,0

15,0 MTE MIMIC

10,0

MTE MIMIC

8,0

10,0

6,0 4,0

5,0

2,0 0,0

0,0 1981

1986

1991

1996

1981

2001

Belgium

1986

1991

1996

2001

Botswana

30,0

120,0

25,0

100,0

80,0

20,0

MTE

MTE

60,0

15,0

MIMIC

MIMIC 10,0

40,0

5,0

20,0

0,0

0,0 1981

1986

1991

1996

1981

2001

34

1986

1991

1996

2001

Bulgaria

Brazil

60,0

45,0 40,0

50,0 35,0

40,0

30,0

MTE

25,0

MTE

MIMIC

20,0

MIMIC

30,0

20,0

15,0 10,0

10,0 5,0

0,0 1981

0,0

1986

1991

1996

2001

1981

Canada

1986

1991

1996

2001

Chile

18,0

60,0

16,0 50,0

14,0 12,0

40,0

10,0

MTE MIMIC

8,0 6,0

MTE 30,0 MIMIC 20,0

4,0 10,0

2,0 0,0 1981

0,0

1986

1991

1996

2001

1981

Colombia

1986

1991

1996

2001

Costa Rica

50,0

45,0

45,0

40,0

40,0

35,0

35,0

30,0

30,0 MTE

25,0

MTE

MIMIC

20,0

MIMIC

25,0 20,0 15,0

15,0 10,0

10,0

5,0

5,0

0,0 1981

0,0

1986

1991

1996

2001

1981

35

1986

1991

1996

2001

Czech Republic

Denmark

25,0

25,0

20,0

20,0

15,0

15,0

MTE

MTE

MIMIC

MIMIC 10,0

10,0

5,0

5,0

0,0

0,0 1981

1986

1991

1996

1981

2001

Egypt

1986

1991

1996

2001

Finland 20,0

140,0

18,0 120,0

16,0 100,0

14,0 12,0

80,0

MTE

MTE

10,0 MIMIC

MIMIC

60,0

8,0 6,0

40,0

4,0 20,0

2,0 0,0

0,0 1981

1986

1991

1996

1981

2001

France

1986

1991

1996

2001

Germany

16,0

25,0

14,0

20,0

12,0 10,0 MTE

15,0 MTE

8,0 MIMIC

MIMIC 10,0

6,0 4,0

5,0 2,0 0,0 1981

0,0 1986

1991

1996

2001

1981

36

1986

1991

1996

2001

Greece

Guatemala 80,0

35,0

70,0

30,0

60,0 25,0

50,0 20,0

MTE

MTE

40,0 MIMIC

MIMIC

15,0

30,0 10,0

20,0

5,0

10,0 0,0

0,0 1981

1986

1991

1996

1981

2001

Hong Kong

1986

1991

1996

2001

Hungary

20,0

60,0

18,0 50,0

16,0 14,0

40,0

12,0 MTE 10,0 MIMIC

MTE 30,0 MIMIC

8,0 20,0

6,0 4,0

10,0

2,0 0,0 1981

0,0

1986

1991

1996

2001

1981

Ireland

1986

1991

1996

2001

Israel

30,0

40,0 35,0

25,0

30,0

20,0 25,0

MTE 15,0 MIMIC

MTE 20,0 MIMIC 15,0

10,0

10,0

5,0 5,0

0,0 1981

0,0

1986

1991

1996

2001

1981

37

1986

1991

1996

2001

Italy

Japan 20,0

30,0

18,0 25,0

16,0 14,0

20,0

12,0 MTE

MTE

10,0

15,0

MIMIC

MIMIC

8,0 10,0

6,0 4,0

5,0

2,0 0,0

0,0 1981

1986

1991

1996

1981

2001

Korea

1986

1991

1996

2001

Malaysia 60,0

100,0 90,0

50,0

80,0 70,0

40,0

60,0

MTE

MTE

30,0

50,0

MIMIC

MIMIC 40,0

20,0

30,0 20,0

10,0

10,0

0,0

0,0 1981

1986

1991

1996

1981

2001

Morocco

1986

1991

1996

2001

Mexico

60,0

60,0

50,0

50,0

40,0

40,0

MTE 30,0 MIMIC

MTE 30,0 MIMIC

20,0

20,0

10,0

10,0

0,0 1981

0,0

1986

1991

1996

2001

1981

38

1986

1991

1996

2001

Netherlands

Norway 25,0

18,0 16,0

20,0

14,0 12,0

15,0 10,0

MTE

MTE

MIMIC

MIMIC

8,0

10,0 6,0 4,0

5,0

2,0

0,0

0,0 1981

1986

1991

1996

1981

2001

Panama

1986

1991

1996

2001

Paraguay

70,0

60,0

60,0

50,0

50,0 40,0

40,0

MTE MIMIC

30,0

MIMIC 20,0

20,0

10,0

10,0 0,0 1981

MTE 30,0

0,0

1986

1991

1996

2001

1981

Peru

1991

1996

2001

Philippines

70,0

70,0

60,0

60,0

50,0

50,0

40,0

MTE MIMIC

30,0

40,0

MTE MIMIC

30,0

20,0

20,0

10,0

10,0 0,0

0,0 1981

1986

1986

1991

1996

1981

2001

39

1986

1991

1996

2001

Poland

Portugal

35,0

25,0

30,0 20,0

25,0 15,0

20,0

MTE

MTE

MIMIC

15,0

MIMIC 10,0

10,0 5,0

5,0 0,0 1981

0,0

1986

1991

1996

2001

1981

1986

Romania

1991

1996

2001

Singapore

40,0

25,0

35,0 20,0

30,0 25,0

15,0

MTE

MTE

20,0 MIMIC

MIMIC 10,0

15,0 10,0

5,0

5,0 0,0 1981

0,0

1986

1991

1996

2001

1981

Slovak Republic

1986

1991

1996

2001

Spain 35,0

25,0

30,0 20,0

25,0 15,0 MTE MIMIC 10,0

20,0

MTE MIMIC

15,0 10,0

5,0

5,0 0,0

0,0 1981

1986

1991

1996

1981

2001

40

1986

1991

1996

2001

Sri Lanka

Sweden 25,0

60,0

50,0

20,0

40,0

15,0 MTE

MTE 30,0

MIMIC

MIMIC

10,0 20,0

5,0

10,0

0,0

0,0 1981

1986

1991

1996

1981

2001

Switzerland

1986

1991

1996

2001

Tanzania

14,0

70,0

12,0

60,0

10,0

50,0

8,0

MTE MIMIC

6,0

40,0

MTE MIMIC

30,0

4,0

20,0

2,0

10,0 0,0

0,0 1981

1986

1991

1996

1981

2001

Thailand

1986

1991

1996

2001

Tunisia

120,0

50,0 45,0

100,0

40,0 35,0

80,0

30,0

MTE 60,0

MTE 25,0

MIMIC

MIMIC 20,0

40,0

15,0 10,0

20,0

5,0

0,0 1981

0,0

1986

1991

1996

2001

1981

41

1986

1991

1996

2001

UK

USA 16,0

25,0

14,0 20,0

12,0 10,0

15,0

MTE

MTE

8,0 MIMIC

MIMIC 10,0

6,0 4,0

5,0

2,0 0,0

0,0 1981

1986

1991

1996

1981

2001

1986

1991

1996

Venezuela 40,0 35,0 30,0 25,0 MTE 20,0 MIMIC 15,0 10,0 5,0 0,0 1981

1986

1991

1996

2001

Source: MTE estimates = own calculations; MIMIC estimates = Schneider (2004, 2005).

42

2001

Figure 3- Chow F-statistics for the presence of structural breaks in the series of unobserved economy (MTE estimates) Australia

Austria

12

12

10

10

8

8

F statistic

6

4

4

2

2

0

0 1984

F statistic

6

1989

1994

1984

1999

Belgium

1989

1994

1999

Botswana

3

10 9

2,5

8 7

2

6

F statistic

1,5

F statistic

5 4

1

3 2

0,5

1

0 1984

0

1989

1994

1999

1984

Brazil

1989

1994

1999

Bulgaria

6

5 4,5

5 4 3,5

4

3

F statistic

3

F statistic

2,5 2

2

1,5 1

1

0,5

0 1984

0

1989

1994

1999

1984

43

1989

1994

1999

Canada

Chile

6

25

5 20

4 15

F statistic

3

F statistic 10

2 5

1

0

0

1984

1989

1994

1999

1984

Colombia

1989

1994

1999

Costa Rica

35

8

30

7 6

25

5

20 F statistic

F statistic

4

15 3

10 2

5

1

0 1984

0

1989

1994

1999

1984

Czech Republic

1989

1994

1999

Denmark 14

2 1,8

12

1,6

10

1,4 1,2

8 F statistic

F statistic

1

6

0,8 0,6

4

0,4

2 0,2

0

0 1984

1989

1994

1984

1999

44

1989

1994

1999

Egypt

Finland

35

4

30

3,5 3

25

2,5

20 F statistic

F statistic

2

15 1,5

10 1

5

0,5

0

0

1984

1989

1994

1999

1984

France

1989

1994

1999

Germany

4

25

3,5 20

3 2,5

15

F statistic

2

F statistic 10

1,5 1

5

0,5 0 1984

0

1989

1994

1999

1984

Greece

1994

1999

Guatemala

16

4

14

3,5

12

3

10

2,5 F statistic

8

F statistic

2 1,5

6 4

1

2

0,5 0

0 1984

1989

1989

1994

1984

1999

45

1989

1994

1999

Hong Kong

Hungary 10

6

9 5

8 7

4

6 F statistic

5

F statistic

3

4 2

3 2

1

1 0

0 1984

1989

1994

1984

1999

Ireland

1989

1994

1999

Israel

8

7

7

6

6 5

5 4

F statistic

4

F statistic 3

3 2

2

1

1 0

0

1984

1989

1994

1999

1984

Italy

1989

1994

1999

Japan 6

2 1,8

5

1,6

4

1,4 1,2

F statistic

3 F statistic

1 0,8

2

0,6

1

0,4 0,2

0

0 1984

1989

1994

1984

1999

46

1989

1994

1999

Korea

Malaysia

4

6

3,5 5

3 4

2,5 F statistic

2

F statistic

3

1,5 2

1 1

0,5 0 1984

0

1989

1994

1999

1984

1989

Mexico

1994

1999

Morocco

8

5

7

4,5 4

6

3,5

5 3

F statistic

4

F statistic

2,5 2

3

1,5

2

1

1

0,5

0

0

1984

1989

1994

1999

1984

Netherlands

1989

1994

1999

Norway

3

4,5 4

2,5 3,5

2

3 2,5

F statistic

1,5

F statistic 2

1

1,5 1

0,5 0,5

0 1984

0

1989

1994

1999

1984

47

1989

1994

1999

Panama

Paraguay 18

9

16

8

14

7

12

6

10

5 F statistics

F statistic

4

8

3

6

2

4

1

2 0

0 1984

1989

1994

1984

1999

Peru

1989

1994

1999

Philippines

8

14

7

12

6

10

5

8 F statistic

F statistic

4

6 3

4 2

2

1

0

0 1984

1989

1994

1984

1999

Poland

1989

1994

1999

Portugal

6

14

5

12 10

4

8

F statistic

3

F statistic 6

2 4

1

2

0 1984

0

1989

1994

1999

1984

48

1989

1994

1999

Romania

Singapore

3,5

14

3

12

2,5

10 8

2

F statistic

F statistic 1,5

6

1

4

0,5

2 0

0 1984

1989

1994

1984

1999

Slovak Republic

1989

1994

1999

Spain

6

12

5

10

4

8

F statistic

3

F statistic

6

2

4

1

2

0

0

1984

1989

1994

1999

1984

Sri Lanka

1989

1994

1999

Sweden

25

8 7

20 6 5

15 F statistic 10

F statistic

4 3 2

5 1

0 1984

0

1989

1994

1999

1984

49

1989

1994

1999

Switzerland

Tanzania 7

5 4,5

6 4

5

3,5 3

4 F statistic

F statistic

2,5

3

2 1,5

2

1

1 0,5

0

0 1984

1989

1994

1984

1999

1989

Thailand

1999

Tunisia 12

6

10

5

8

4

F statistic

3

F statistic

6

2

4

1

2

0

0 1984

1994

1989

1994

1984

1999

UK

1989

1994

1999

USA 10

6

9 5

8 7

4

6 F statistic

3

F statistic

5 4

2

3 2

1

1 0

0 1984

1989

1994

1984

1999

50

1989

1994

1999

Venezuela 3

2,5

2

F statistic

1,5

1

0,5

0 1984

1989

1994

1999

Source: own calculations.

51

Figure 4- A comparison between the standard ECM estimates and the MTE results Australia

Austria

25,0

20,0 18,0

20,0

16,0 14,0

ECM

15,0

ECM

12,0

MTE MTE (I) 10,0

MTE (P)

MTE 10,0 MTE (I) 8,0

MTE (P)

6,0

5,0

4,0 2,0

0,0 1981

0,0

1986

1991

1996

2001

1981

Belgium

1986

1991

1996

2001

Botswana

30,0

120,0

25,0

100,0

20,0

ECM

80,0

ECM

MTE 15,0 MTE (I) MTE (P)

10,0

5,0

MTE 60,0 MTE (I)

20,0

0,0 1981

MTE (P)

40,0

0,0

1986

1991

1996

2001

1981

Bulgaria

1986

1991

1996

2001

Brazil

60,0

35,0

50,0

30,0 25,0

40,0

ECM MTE

ECM 20,0

MTE

30,0 MTE (I)

MTE (I)

15,0

MTE (P)

20,0

MTE (P) 10,0

10,0

5,0

0,0 1981

0,0

1986

1991

1996

2001

1981

52

1986

1991

1996

2001

Canada

Chile 60,0

16,0 14,0

50,0 12,0 ECM

10,0

40,0

ECM

MTE

MTE 30,0

8,0 MTE (I) 6,0

MTE (P)

MTE (I) MTE (P)

20,0

4,0 10,0 2,0 0,0

0,0

1981

1986

1991

1996

2001

1981

Colombia

1986

1991

1996

2001

Costa Rica

45,0

45,0

40,0

40,0

35,0

35,0

30,0

ECM

30,0

25,0

MTE

25,0

MTE

20,0

MTE (I)

20,0

MTE (I)

MTE (P)

15,0

10,0

5,0

5,0

1981

MTE (P)

15,0

10,0

0,0

ECM

0,0

1986

1991

1996

2001

1981

Czech Republic

1986

1991

1996

2001

Denmark 25,0

9,0 8,0

20,0

7,0 6,0

ECM

5,0

MTE

4,0

MTE (I) MTE (P)

3,0 2,0

ECM

15,0

MTE MTE (I) 10,0

MTE (P)

5,0

1,0

0,0

0,0 1981

1986

1991

1996

1981

2001

53

1986

1991

1996

2001

Egypt

Finland

140,0

20,0 18,0

120,0

16,0

100,0

14,0

ECM

ECM

12,0

80,0

MTE MTE (I)

60,0

MTE (P)

MTE 10,0 MTE (I) 8,0

MTE (P)

6,0

40,0

4,0

20,0 2,0

0,0 1981

0,0

1986

1991

1996

2001

1981

France

1986

1991

1996

2001

Germany

16,0

25,0

14,0 20,0

12,0 ECM

10,0

ECM

15,0

MTE

MTE

8,0 MTE (I) 6,0

MTE (P)

MTE (I) 10,0

MTE (P)

4,0 5,0

2,0 0,0 1981

0,0

1986

1991

1996

2001

1981

Greece

1986

1991

1996

2001

Guatemala

25,0

80,0 70,0

20,0 60,0

ECM

15,0

ECM

50,0

MTE MTE (I) 10,0

MTE (P)

MTE 40,0 MTE (I) 30,0

MTE (P)

20,0

5,0

10,0

0,0 1981

0,0

1986

1991

1996

2001

1981

54

1986

1991

1996

2001

Hong Kong

Hungary 60,0

25,0

50,0 20,0

40,0

ECM

ECM

15,0

MTE

MTE 30,0 MTE (I)

MTE (I) 10,0

MTE (P)

5,0

MTE (P)

20,0

10,0

0,0

0,0 1981

1986

1991

1996

1981

2001

Ireland

1986

1991

1996

2001

Israel

30,0

40,0 35,0

25,0

30,0

20,0

ECM

ECM

25,0

MTE 15,0 MTE (I) MTE (P)

10,0

MTE 20,0 MTE (I) 15,0

MTE (P)

10,0

5,0 5,0

0,0

0,0

1981

1986

1991

1996

2001

1981

Italy

1986

1991

1996

2001

Japan

25,0

20,0 18,0

20,0

16,0 14,0

ECM

15,0

ECM

12,0

MTE MTE (I) 10,0

MTE (P)

MTE 10,0 MTE (I) 8,0

MTE (P)

6,0 4,0

5,0

2,0 0,0

0,0 1981

1986

1991

1996

1981

2001

55

1986

1991

1996

2001

Korea

Malaysia 60,0

100,0 90,0

50,0

80,0 70,0 ECM

60,0

40,0

ECM

MTE

MTE 30,0

50,0 MTE (I) 40,0

MTE (P)

30,0 20,0

MTE (I) MTE (P)

20,0

10,0

10,0 0,0 1981

0,0 1986

1991

1996

2001

1981

Morocco

1986

1991

1996

2001

Mexico

60,0

60,0

50,0

50,0

40,0

ECM MTE

30,0

40,0

ECM MTE

30,0 MTE (I)

MTE (I) MTE (P)

20,0

10,0

10,0

0,0

0,0 1981

MTE (P)

20,0

1986

1991

1996

1981

2001

Netherlands

1986

1991

1996

2001

Norway 12,0

20,0 18,0

10,0 16,0 14,0 ECM

12,0

MTE 10,0

8,0

ECM MTE

6,0 MTE (I)

MTE (I) 8,0

MTE (P)

6,0 4,0

MTE (P)

4,0

2,0

2,0

0,0

0,0 1981

1986

1991

1996

1981

2001

56

1986

1991

1996

2001

Panama

Paraguay

50,0

60,0

45,0 50,0

40,0 35,0 ECM

30,0

40,0

ECM

MTE 25,0 MTE (I) 20,0

MTE (P)

15,0

MTE 30,0 MTE (I) MTE (P)

20,0

10,0 10,0

5,0 0,0 1981

0,0

1986

1991

1996

2001

1981

Peru

1986

1991

1996

2001

Philippines 70,0

60,0

60,0

50,0

50,0 40,0

ECM

ECM MTE

40,0

MTE

30,0 MTE (I)

MTE (I)

30,0

MTE (P)

MTE (P)

20,0

20,0 10,0

10,0 0,0

0,0 1981

1986

1991

1996

1981

2001

Poland

1986

1991

1996

2001

Portugal

30,0

25,0

25,0 20,0

20,0

ECM MTE

ECM

15,0

MTE

15,0 MTE (I) MTE (P)

10,0

MTE (P)

5,0

5,0

0,0 1981

MTE (I) 10,0

0,0

1986

1991

1996

2001

1981

57

1986

1991

1996

2001

Romania

Singapore

25,0

25,0

20,0

20,0

ECM

15,0

ECM

15,0

MTE

MTE

MTE (I) 10,0

MTE (P)

5,0

MTE (I) 10,0

MTE (P)

5,0

0,0

0,0

1981

1986

1991

1996

2001

1981

Slovak Republic

1986

1991

1996

2001

Spain

12,0

35,0

10,0

30,0 25,0

8,0

ECM MTE

ECM 20,0

MTE

6,0 MTE (I)

MTE (I)

15,0

MTE (P)

4,0

MTE (P) 10,0

2,0

5,0

0,0

0,0

1981

1986

1991

1996

2001

1981

Sri Lanka

1986

1991

1996

2001

Sweden 14,0

60,0

12,0

50,0

10,0 40,0

ECM

ECM MTE

8,0

MTE

30,0 MTE (I)

MTE (I)

6,0

MTE (P)

MTE (P)

20,0

4,0 10,0

2,0 0,0

0,0 1981

1986

1991

1996

1981

2001

58

1986

1991

1996

2001

Switzerland

Tanzania

14,0

60,0

12,0

50,0

10,0 ECM 8,0

MTE

40,0

ECM MTE

30,0 MTE (I)

MTE (I)

6,0

MTE (P)

MTE (P)

20,0

4,0

10,0

2,0

0,0

0,0 1981

1986

1991

1996

1981

2001

Thailand

1986

1991

1996

2001

Tunisia

120,0

60,0

100,0

50,0

80,0

ECM

40,0

ECM

MTE 60,0 MTE (I) MTE (P)

40,0

20,0

MTE (I) MTE (P)

20,0

10,0

0,0 1981

MTE 30,0

0,0

1986

1991

1996

2001

1981

United Kingdom

1986

1991

1996

2001

USA

25,0

18,0 16,0

20,0

14,0 ECM

15,0

MTE MTE (I) 10,0

MTE (P)

12,0

ECM

10,0

MTE MTE (I)

8,0

MTE (P)

6,0 4,0

5,0

2,0 0,0 1981

0,0 1986

1991

1996

2001

1981

59

1986

1991

1996

2001

Venezuela 40,0 35,0 30,0 ECM

25,0

MTE 20,0 MTE (I) 15,0

MTE (P)

10,0 5,0 0,0 1981

1986

1991

1996

2001

Source: own calculations

60