Örebro University
Swedish Business School At Örebro University
Master’s Degree Project VT 2011
Economics and Econometrics
Supervisor: Jörgen Levin
Examiner: Patrik Karpaty
Session: 2009-2011
A Cross Country Analysis of Tax
Performance with Special Focus on
Pakistan’s Tax Effort
Abstract
This study analyzes tax performance across countries with special focus on Pakistan, utilizing different econometric applications on a panel dataset that covers 104 countries over the period 1996-2005. Generally, openness, per capita GDP, urban population, rule of law and control of corruption are identified to be significant determinants of tax-to-GDP ratio across countries. Also PCSE (Panel Corrected Standard Error) estimates provide statistically better regression results compared to other specifications used in this study. The tax effort indices obtained for Pakistan show an overall decline that suggest the need of good policy measures such as implementation of modified VAT, broaden the tax base and improvement of institutional quality should be taken by government to address this issue.
1. Introduction
The primary aim of this study is to find out conventional and non-conventional determinants of tax-to-GDP ratio across countries and to construct tax effort indices for Pakistan. The conventional determinants are those based on traditional theories such as level of development, foreign aid and structure of the economy (Chelliah, 1971; Lotz & Morss, 1967). Whereas, non-conventional determinants such as governance or institutional variables, shadow economy, tax morale are also considered important in recent literature (Bird, Martinez-Vazquez and Torgler, 2004; Davoodi and Grigorian, 2007). The motivation behind this study is, therefore, to reassess the role of conventional variables in addition to new variables in determination of tax revenue performance.
The variation in tax revenue collection level between developed and developing countries attributed to various reasons in economic literature. These include difference in economic structure, nature of taxes, transparency and effectiveness of tax administration system and taxation policies. Therefore, we would also analyze the factors responsible for the difference in tax revenue performance among countries by dividing the dataset into different income groups (Auriol and Warlters, 2005; Tanzi and Zee, 2000).
Another objective of this study is to construct tax effort indices for Pakistan based on regression results of the total tax revenue. During 2002-10, Pakistan‟s tax-to-GDP ratio remained 10.9% on average which poses an interesting question of whether Pakistan is capable to further exploit its domestic resources or is it already at the verge of full exploitation and no further revenue could be generated from domestic tax collection. It would also help us to critically evaluate the promises of Pakistan with IMF to raise the tax-to-GDP ratio around 14% by 2013-14.
This paper contributes to the literature by examining the effect of governance index or non conventional variables on tax performance in particular and effect of other conventional variables on tax performance in general. In other words, we would analyze the extent to which conventional indicators such as per capita GDP, urbanization, openness, foreign aid (% of GNI) as well as non conventional indicators such as “voice and accountability”, government effectiveness, regulatory quality, rule of law and control of corruption explains variation in tax-to-GDP ratio. This paper also supplements the work of Bird et al. (2004) on the role of societal
institutions in tax revenue performance by using a broader dataset. Therefore, we would examine the impact of governance index on tax revenue performance with additional years added to the dataset by using advance econometric applications.
The outline of the study is as follows: section 2 gives overview of existing literature related to this study. Section 3 sheds light on development of tax structure of Pakistan over time. Section 4 describes research methodology and data used in this study. Section 5 present regression results. Section 6 concludes.
2. Literature Review
In this section, we review the literature on tax revenue performance.1 Stotsky and WoldeMariam
(1997), with panel data from 30 countries over the period 1990-95, find that share of agriculture and mining in GDP have a negative impact on tax revenue. However, export share in GDP and per capita GDP are positively significantly associated with tax revenue performance. They also calculated tax effort index which shows that countries having high tax-to-GDP ratios mostly exhibit high ranks in tax effort index. Another study on developing countries by Bahl (1971) also reported agriculture, export ratio and mining share in GDP as important variables of tax revenue. Piancastelli (2001) found that per capita income and the ratio of trade to GDP are positive strong determinants of tax revenue, whereas, share of agriculture in GDP is negatively associated with tax revenue. He further expands the scope of research by including comprehensive analysis of countries‟ tax performance by dividing them according to their income groups which shows that middle income countries perform worse as compared to low income countries.
Despite acknowledging the role of institutional variables or demand side factors in tax revenue collection, Leuthold (1991) construct a simple model to determine the indicators of tax revenue for eight African countries. He used share of agriculture in GDP, export and import ratios to GDP, share of foreign grants in income, share of mining in income as explanatory variables, out of which, share of foreign grants in income (positively associated) and share of agriculture in GDP (negatively associated) are reported as important variables in determination of tax-to-GDP ratio.
Davoodi and Grigorian (2007) found that the tax collection rate (especially direct taxes) in Armenia did not increase with the same pace as GDP. They also found that institutional quality, urbanization and shadow economic activity are the main factors behind low tax-to-GDP ratio in Armenia. Panel data technique is applied on 141 countries for the time period of 1990-2004. They also construct a tax effort index for Armenia which showed a downward trend since 2000. They conclude that restructuring of fiscal institutions, implementation of transparency mechanism by government officials and improving VAT tax base could have positive effect on tax revenue.
Tanzi and Davoodi (2000) examined the hypothesis that composition of taxes and tax collection mechanism matters to explain the impact of corruption on tax revenue. Due to the fact that different types of taxes are collected by different means in different countries, thereby the level of corruption in tax collection seems to be different across countries. Their preliminary conclusion indicated that those countries that introduced VAT (Value Added Tax) into their tax system experienced much lower level of corruption in subsequent years compared to countries that did not implement VAT.
Auriol and Wartlers (2005) test the hypothesis that fixed market entry fee is positively related with high tax revenue in developing countries, assuming other things being constant. The result of the study confirms the hypothesis and further concludes that GNP, land and population density are also important determinants of tax revenue.
Bird et al. (2004) finds strong impact of non-agriculture share of GDP, governance and ICRG (International Country Risk Guide) indices on tax effort of 110 underdeveloped and transitional economies by using cross sectional data over the period from 1990 to 1999.2 Gupta (2007) applied different econometric techniques on a panel of 105 developing countries over 25 years and concludes that per capita GDP, share of agriculture in GDP, openness, foreign aid, corruption, political stability and specific sources of tax revenue are major indicators of tax revenue.
2 Quality of governance is a six variable index measured by World Bank and ICRG also offers data on quality of
Teera and Hudson (2004) examined the impact of economic variables along with shadow economy on tax performance by dividing countries into different income groups. Put it differently, they captured the effect of these variables on tax collection in different phases of economic development. Shadow economy, openness and time trend used as proxy to capture global trend in taxation turns out to be important indicators of tax revenue. They also analyze the fact that countries in different income groups showed different outcomes about effect of export and manufacturing sectors on tax revenue. Tax effort indices calculated showed that most countries in high income groups are able to collect taxes equivalent to their potential capacity. A number of research papers analyze the impact of a specific variable on tax revenue performance. For example, Baunsgaard and Keen (2009) discussed the impact of trade liberalization on tax-to-GDP ratio. Trade tax revenue being a large portion of total tax revenue in developing countries lowers the overall tax-to-GDP ratio in post trade liberalization era. They analyze whether developing countries would be able to recover the loss of trade revenue incurred after liberalization. They used (unbalanced) panel data technique on 117 countries for the period of 1975-2006. They found that high income countries are able to manage their trade taxes loss through domestic taxes after liberalization. However, middle income countries showed partial recovery of the loss. Finally, low income countries exhibit mixed trend, that is, some countries in the dataset showed substantial recovery others not.
Franco-Rodriguez, Morrissey and Mcgillivray (1998) reported that aid is negatively associated with tax effort in Pakistan using time series data from 1956-95. This is the first study that has used aid as an endogenous variable. They argue that, in developing countries, governments have the liberty to use foreign aid according to their own choice; hence most funds are allocated to current expenditures instead of fruitful investment. This also reduces their effort to increase revenue collection from domestic sources. Clist and Morrissey (2011), with data from 82 developing countries over the period 1970-2005, further deepen this analysis by examining the effect of loans and grants individually on tax effort. They found that tax revenue is not negatively associated with foreign aid.
On the whole the discussion above summarizes the determinants of tax revenue performance to be share of agriculture in GDP, openness, share of exports and imports in GDP, per capita GDP,
foreign aid, shadow economy, market entry fee, institutional quality, corruption and political stability.
3. Pakistan’s Fiscal Policy Stance
According to Economic survey of Pakistan (2009-10), Government‟s important objective is to increase tax-to-GDP ratio to around 14 percent by 2013-14 but this target is mainly attributed to
increase the efficiency of VAT by implementing its modified versionin 2011. It is interesting to
mention that this targets is also considered as one of important conditions of IMF Stand-By agreement. 3. The achievement of this goal is also subjected to various modifications in tax system which includes imposition of moderate taxes to bring non-tax payers into the tax net, taxpayers‟ training, electronic filing for income tax returns and supporting regional tax offices with advance technology and infrastructure etc.
In general, it is assumed that the tax revenue should increase with the growth of economy; however, this seems not to be true in case of Pakistan. For instance, real growth in tax revenue remained on average around 1.4 percent in contrasts with average real GDP growth of 5.5 percent during 2004-2010. Table 3.1 shows that direct and indirect tax revenue contributed to only 4.5 percent and 6.1 percent to total revenue (% of GDP) in 2009-10. It also showed that sales taxes are major earning source compared to federal excise and custom duties in terms of indirect taxes. Tax-to-GDP ratio remained around 10.9 percent on average during 2002-10 exhibiting the need of pragmatic approach of policy makers to raise the tax revenue level.
Table 3.1: Tax Portfolio of Pakistan: 2002-10 (As a percentage of GDP)
2002-03 2003-04 2004-05 2005-06 2006-07* 2007-08 2008-09** 2009-10** Tax Revenue 11.4 10.8 10.1 10.6 11.0 10.4 11.3 11.9
Direct Taxes 3.1 2.9 2.8 3.0 3.8 3.7 4.0 4.5
Federal Excise Duty 0.9 0.8 0.9 0.7 0.8 0.8 0.9 1.0
Sales Tax 4.0 3.9 3.6 3.9 3.6 3.7 3.8 4.0
Custom Duties 1.4 1.6 1.8 1.8 1.5 1.4 1.2 1.1
Other 2.0 1.6 1.0 1.2 1.3 0.8 1.4 1.3
*Estimated Figures Source: IMF Consultation Report, Various Issues
**Projected Figures
In 2005, Pakistan recorded very low tax-to-GDP ratio of 10.1 percent compared to other middle income countries such as Maldives (18.0 percent) and Sri Lanka (13.7 percent) having much lower GDP level than Pakistan. The most probable reason for low growth rate of tax-to-GDP ratio is that the factors responsible for high GDP growth did not fall into tax net. One such example is agriculture sector which grew at the rate of 10 percent but contributed only 1 percent to tax revenue in 2009-10.
Due to liberalization in 1990‟s countries aimed to reduce their trade taxes to lowest possible level which adversely affects the tax-to-GDP ratio of developing countries due to their greater reliance on trade taxes. Likewise, the average amount of Pakistan‟s trade taxes before liberalization stood at 5.9 percent of GDP (1975-1990) compared to post liberalization amount of 1.7 percent of GDP during 2000-05. Unfortunately, Pakistan is still failing to fill this gap through revenue mobilization. This can be another reason for low tax-to-GDP ratio of Pakistan in recent years. Government of Pakistan documented several other reasons contributing in Pakistan‟s low tax-to-GDP ratio like shadow economy, low tax compliance, inflation, low tax morale, narrow tax base, corruption, weak legal system and inefficient administration.
In the light of discussion above, we are interested to find out the tax effort indices for Pakistan in subsequent sections to analyze its potential to meet the tax revenue target of around 14 percent of GDP by fiscal year 2013/14.
4. Data And Methodology
4.1. Data Description
The longitudinal dataset used in this study cover 104 countries from 1996 to 2005. It is taken from IMF International Financial Statistics (IFS) and World Development Indicators (WDI), World Bank database. However, data on six governance indicators is taken from Worldwide Governance Indicators (WGI), which is available for 1996-2005 out of which observations for 1997, 1999 and 2001 are missing.
Table 4.1 summarizes important statistics about variables used in this study. It shows that average contribution of agriculture in GDP is 16.6 percent whereas tax revenue contributes in GDP by 21.7 percent. The greater values of standard errors for openness, urban population,
agriculture and tax revenue indicate the presence of high divergence from mean values for these variables. This is explained by differences in economic development of different countries in the dataset. For instance, the average value for openness is 82.9%, with minimum value of 12.8% which belongs to Zimbabwe and maximum value of 339.6% which belongs to Singapore exhibits the existence of large variation in the dataset.
Table 4.1: Summary Statistics Dataset (1996-2005)
Variables Source Unit Obs Mean Std. Dev. Min Max
Tax Revenue IFS % of GDP 1015 21.7 11.4 3.8 53.4
Log Per Capita GDP WDI % 1027 7.7 1.7 4.6 10.9
Agriculture Value Added WDI % of GDP 987 16.6 14.5 .1 62.0
Openness WDI % 987 82.9 48.0 12.8 339.6
Foreign Aid WDI % of GNI 1024 5.2 7.8 -3.0 66.0
Urban Population WDI % of total 1040 51.0 24.5 7.4 100
Log Inflation WDI % 898 1.4 1.2 -4.1 6.1
Voice and Accountability WGI Units 728 .1 1.0 -1.8 1.8
Political Stability, No Violence WGI Units 718 .03 1.0 -2.7 1.6
Government Effectiveness WGI Units 703 .2 1.1 -2.4 2.2
Regulatory Quality WGI Units 714 .1 1.0 -2.7 3.4
Rule of Law WGI Units 716 .1 1.0 -1.9 2.0
Control of Corruption WGI Units 703 .2 1.1 -1.8 2.4
Dependent variable in this study is defined as total tax revenue (% of GDP). Based on literature review, we would now describe the explanatory variables used in this study.
GDP per capita attempts to capture the impact of a country‟s development on ability to pay and collect taxes. It is expected that GDP per capita would have positive effect on Tax-to-GDP ratio (Baunsgaard and Keen, 2009; Davoodi and Grigorian, 2007; Gupta, 2007).
Several studies like Chelliah (1971) and Gupta (2007) considered agriculture value added (% of GDP) as one of important conventional explanatory variables. Empirical evidences indicated this sector as hard to tax because of large share of subsistence farming. This could also be considered as one of main reasons behind low tax-to-GDP ratio in developing countries. We expect a negative relationship between tax-to-GDP ratio and agriculture value added (% of GDP).
Openness is calculated as sum of imports and exports, divided by GDP and tries to capture the effects of international trade on overall government revenue. This is considered as one of high revenue generating source especially in developing countries. The reason behind is exports and imports are quantified (registered) on specified entry and exit points of a country which
diminishes the possibility of escape without paying taxes (Baunsgaard and Keen, 2009; Davoodi and Grigorian, 2007; Gupta, 2007). A positive association between openness and tax revenue is expected.
Due to non stringent repayment conditions, foreign aid in the form of grants usually replaces tax revenue in developing countries, hence negatively related with tax revenue (Gupta, 2007).
However, Clist and Morrissey (2011) argue that foreign aid in the form of grants is infact helpful
to improve the tax-to-GDP ratio of developing countries. Therefore, we are not certain about the expected relationship between aid (% of GNI) and tax revenue performance.
Urbanization could be used as proxy to capture the demand for public services because most government sector activities are concentrated in cities which also results in high tax revenue from urban areas. Hence we expect positive relationship between urbanization and tax-to-GDP ratio (Davoodi and Grigorian, 2007).
Inflation is taken as a regressor to highlight the quality of fiscal and monetary policies in terms of tax revenue. Higher inflation would reduce the tax-to-GDP ratio due to decrease in purchasing power of consumer and investing capacity of investors, thus negative sign is expected between these two variables (Baunsgaard and Keen, 2009; Davoodi and Grigorian, 2007).
Non-conventional studies on tax performance highlight the importance of governance indicators in tax collection. If taxpayers feel that their interests are well represented in assembly and are satisfied with quality and quantity of public goods, their willingness to pay taxes increases. For instance, weak law and order conditions in the economy induce people to avoid the tax payments. Likewise, if corruption is common phenomenon in an economy, large share of business community would like to work underground by paying bribes to avoid high tax payments. Therefore, in order to evaluate the impact of these non conventional variables on tax
performance we focused on six governance indicators computed by World Bank.4 These
indicators are described as: (i) Voice and Accountability, (ii) Political Stability and Absence of Violence, (iii) Government Effectiveness, (iv) Regulatory Quality, (v) Rule of Law and (vi) Control of Corruption.
World Bank assigned them ranked values ranging between -2.5 to 2.5. The basic intuition behind construction of governance index is to collect and express the views of general public and experts to formulate one dataset for comparative analysis of countries (Kaufmann, Kraay and Mastruzzi, 2010). Higher values of governance variables indicates better environment for tax collection. The choice of institutional or demand side variables is fairly based on data availability.
We graphically analyzed the relationship between some of the explanatory variables and tax revenue in Figures 4.1 and Figure 4.2, in which scattered points represent different countries in the dataset. Figure 4.1 depicts that all variables have positive relationship with tax revenue except agriculture value added (% of GDP). Figure 4.2 shows that improvement in governance indicators has positive effects on tax-to-GDP ratio. This preliminary assessment is in line with Gupta (2007).
Source: IFS, World Bank, 2010
Source: IFS, World Bank, 2010
0 10 20 30 40 50 60 0 10 20 30 40 50 60
Agriculture Value Added (% of GDP)
Fitted values Tax Revenue (% of GDP)
0 10 20 30 40 50 60 5 6 7 8 9 10 11
Log Per Capita GDP
Fitted values Tax Revenue (% of GDP)
0 10 20 30 40 50 60 0 50 100 150 200 250 300 350 Openness
Fitted values Tax Revenue (% of GDP)
0 10 20 30 40 50 60 0 20 40 60 80 100
Urban Population (% of Total)
Fitted values Tax Revenue (% of GDP)
Figure 4.1: Tax Revenue (% of GDP) and Conventional Determinants (1996-2005)
0 10 20 30 40 50 60 -2 -1 0 1 2
Voice and Accountability
Fitted Values Tax Revenue (% of GDP)
0 10 20 30 40 50 60 -2 -1 0 1 2 3 Control of Corruption
Fitted Values Tax Revenue (% of GDP)
0 10 20 30 40 50 60 -3 -2 -1 0 1 2 Government Effectiveness
Fitted Values Tax Revenue (% of GDP)
0 10 20 30 40 50 60 -3 -2 -1 0 1 2 3 4 Regulatory Quality
Fitted Values Tax Revenue (% of GDP)
4.2. Methodology
In this section, first we will focus on some potential biases which could affects the significance of regression results and also review the possible measures which could help to avoid them. In subsequent section, we will elaborate the methodology used in this study.
4.2.1. Potential Biases
Generally, previous literature used both fixed and random effect specifications to avoid the most common problem of unobserved heterogeneity in cross country analysis. There are some studies which applied OLS technique in the absence of endogeneity or also include it just for comparison purposes accompanied with other techniques (Davoodi and Grigorian, 2007; Teera and Hudson, 2004). The problem of endogeneity is more sophisticatedly handled in recent literature by employing dynamic panel specifications with the help of difference and system GMM (Baunsgaard and Keen, 2009; Gupta, 2007). This problem usually occurs when an explanatory variable correlates with the error term.
Another important econometric issue is related with the presence of heteroscedasticity in variances of coefficients of estimates which could bias the regression results. However, this problem is somehow solved with the use of robust standard errors for coefficients of estimates. Likewise, problem of serial correlation arises in econometric analysis when disturbances or error term fail to meet the properties of independence and identical distribution. Gupta (2007) deals with both these problems by introducing the Panel Corrected Standard Error Estimates (PCSE). Many studies also face the issue of multicollinearity which arises when two or more explanatory variables appeared to be highly correlated with each other. According to Bird et al. (2004) variables having correlation above 0.80 could definitely be problematic and sometimes even lower values of correlation could cause problem. Hence, Bird et al. (2004) and also Gupta (2007) suggests to use the highly correlated explanatory variables in separate specifications to avoid the possibility of multicollinearity.
After taking brief overview of potential biases and possible measures to handle these econometric issues, we now briefly summarize the economic specifications used in this study.
4.2.2. Fixed Effect
In cross country analysis, there are many unobserved factors (such as language, ethnicity, land locked country) which if do not takes into account properly could lead to biased estimation of the model. Due to rejection of random effect model in favour of fixed effect model by Hausman test, we start with the fixed effect specification as a baseline regression. The equation for fixed effect model is presented as:
it it it it i it
X
INS
Policy
y
(1)Where is tax-to-GDP ratio in a country i during time period t. is the country‟s fixed effect.
The error term it represents unobserved part of regression with respect to tax-to-GDP ratio. ,
and represents structural, institutional and policy variables respectively. The
structural variables include log of per capita GDP, agriculture value added (% of GDP), openness, foreign aid (% of GNI), urban population (% of total). The institutional variables are “voice and accountability”, “political stability and absence of violence”, government effectiveness, regulatory quality, rule of law and control of corruption. Finally, the policy variables include inflation only.
4.2.3. Panel Corrected Standard Error Estimates (PCSE)5
We applied Wooldridge test to check first-order serial correlation in the residuals which,
however, confirm its presence. 6 This issue arises when variance of disturbance of each country is
unique and also each country has its own covariance. To correct these issues, Panel-Corrected Standard Error Estimates (PCSE) are applied to estimate the model. PCSE uses OLS or Prais-Winsten regression, allows for heteroscedasticity and contemporaneously correlated error terms across panels (Davoodi and Grigorian, 2007; Gupta, 2007). In the presence of AR (1), both common coefficient across all panels (i ,i) and specific coefficient for each panel
) ,
(i i j can be derived with this method (Gupta, 2007). Therefore, the empirical model
has been modified as under:
5 See for example, Frain (2006) for further details.
6 xtserial command is used in Stata 11.2 to test for serial correlation. The null hypothesis of “no first order
it it it it X INS y Whereas, it i.it1it
Where is tax-to-GDP ratio in a country i at time period t. is constant term. it is error term consisting of two parts; i.it1 represents non constant part of error term and it shows constant part of error term. is a vector of structural variables which include log of per capita GDP, agriculture value added (% of GDP), openness, foreign aid (% of GNI), urban population (% of
total). Whereas, is vector of institutional variables which include “voice and
accountability”, government effectiveness, regulatory quality, rule of law and control of corruption.
As PCSE model does not undertake endogeneity problem, therefore, we would deal with this issue by estimating PCSE specifications again by taking lagged values of foreign aid (% of GNI) as an instrument for contemporaneous values of foreign aid. We considered foreign aid as endogenous variable due to its ability to interact with dependent variable specially in case of developing countries which largely depends upon foreign aid to finance their expenditures (Clist & Morrissey, 2011; Gupta, 2007). This creates simultaneous or causal relationship between tax revenue and foreign aid. In other words, if a country could succeed to mobilize domestic resources then it will eventually reduce the demand for foreign aid.
4.2.4.Dynamic Panel Specification
Dynamic panel specification is used to analyze the model in the presence of endogeneity. Arellano and Bond (1991) undertake this issue by introducing Generalized Method of Moments (GMM). It incorporate a lagged dependent variable, a lagged level of endogenous independent variables and the difference of strictly exogenous independent variables in the regression. This method is commonly known as Difference-GMM.
Unfortunately, the major limitation of this method is that highly persistent lagged variables appear to be poor instruments. Arellano and Bover (1995) and Blundell and Bond (1998) overcome this problem with the introduction of another procedure referred as system-GMM. It includes additional moment conditions to enhance efficiency of estimations (Baunsgaard and Keen, 2005; Gupta, 2007).
Furthermore, Baum, Schaffer and Stillman (2003) stated that Hansen J statistic could be used in the presence of conditional heteroscedasticity instead of Sargan statistic for over identification test. The test aims to check the exogeneity of the instruments used in the analysis. The acceptance of null hypothesis of “exogeneous instruments” would verify the correctness of the model. Moreover, the Arellano-Bond test for autocorrelation would also be presented to further strengthen the results. This test has null hypothesis of “no correlation”. According to many research studies first difference process or AR (1) usually fails to accept null hypothesis, however, the second difference process or AR (2) mostly provides desired results due to its orthogonality with current error term.
4.2.5.Tax Effort Indices
The above mentioned regressions would help us to draw conclusions about tax performance of developed and developing countries. Now we turn our attention to analyze the tax effort in Pakistan which is another important objective of this study. Therefore, we need to derive tax effort index by taking ratios of actual and predicted values of tax-to-GDP ratio (Stotsky and WoldeMariam, 1997). If the index is higher than one, it will shows that Pakistan is putting in more effort than its capacity to improve its tax revenue performance and vice versa but if index equals 1 it would suggest that Pakistan is just in line with its capacity to collect taxes. The selection of regression results to estimate predicted values of tax ratio would be made on the basis of their significance as well as economic rationale. This method is generally known as “regression approach” (Teera and Hudson, 2004). It is noteworthy to mention that most studies on tax effort follows this approach, for example, Davoodi and Grigorian, (2007) and Gupta, (2007).
However, this procedure has some limitations. Firstly, this approach is constrained to only those variables included in the regression and ignore other unobserved factors that could have significant effect on tax revenue performance. Secondly, it might be possible that the use of different econometric specification will change values of tax effort index even if applied on the same set of variables. Thirdly, absence of strong theoretical background is another drawback. Despite having limitations, this simple and easy to understand approach to calculate tax effort of countries got attention of many scholars (Davoodi and Grigorian, 2007; Tanzi & Zee, 2000).
According to Stotsky and WoldeMariam (1997), this approach has a power to control tax effort index for variables other than tax bases which should be considered as its merit.
5. Regression Analysis and Tax Effort Indices
5.1: Regression Analysis7
In order to avoid the problem of multicollinearity a series of regressions is done by taking all promising institutional variables separately in all specifications as shown in Table A.4-Table
A.13.8 Further, the influence of agriculture value added (% of GDP) and log of per capita GDP
on tax revenue is also captured separately due to high correlation (0.86) detected between them (Gupta, 2007). Also, robust standard errors have been used to undertake the problem of
heteroscedasticity.9 Regression results are presented in Table A.4-Table A.13, however, the main
results from all econometric specifications are summarized in Table 5.1.
In Table 5.1, fixed effect results are reported for comparison purposes only. Despite having high values of R-squared PCSE model provide statistically significant results as compared to GMM. Therefore, PCSE model is chosen. PCSE results by taking lagged values of foreign aid instead of its contemporaneous values do not show much difference indicating that endogeneity is not a big issue here.
PCSE estimates show that log of per capita GDP has strong impact on tax revenue as one percent change in log of per capita GDP may change tax revenue by 4.27 percent. This finding lends support from the conclusions of the previous studies that ability to collect and pay taxes increases with the level of development (Chelliah, 1971; Gupta 2007). The coefficient on agriculture value added (% of GDP) showed the negative sign and strong significance as expected in both common and panel specific correlation coefficient estimations. Its magnitude implies approximately 0.13 percent drop in tax-to-GDP ratio for one percent increase in agriculture value added (% of GDP).
7 We also run these specifications on structural variables only by using relatively large data set consists of
125 countries over the period of 1975-2005. These regressions produce almost same results as mentioned in this study.
8 Pair wise correlation of explanatory variables is presented in Table A.2. 9 xttest3 command is used in Stata 11.2 to check heteroscedasticity.
The coefficient on openness showed negative sign, when the autocorrelation process is different for each panel. Trade liberalization policies adopted by most countries in post-WTO era could be a possible reason for negative insignificant relationship between openness and tax collection (Bird et al., 2004).
Table 5.1: Determinants of Tax Revenue Performance: 1996-2005
FE PCSE One-Step GMM1011
Common Correlation
Panel Specific Common Correlation
with Lagged Aid
Panel Specific
with Lagged Aid
Difference System Constant -22.59* (1.69) .39 (0.05) -2.61* (1.73) 20.95*** (24.75) -9.42*** (4.64) 21.01*** (18.88) -2.74** (2.30) 19.89*** (15.48) -5.29*** (3.37) 21.84*** (13.98) 0.55 (0.39) (0.89) 1.08 Lagged Revenue (%of GDP) 0.21** (2.04) (1.64) 0.16 0.90*** (17.54) 0.94*** (15.23)
Log Per Capita GDP 2.94 (1.56) 3.37*** (20.69) 4.27*** (17.77) 3.29*** (18.62) 3.73*** (15.28) -1.89 (0.53) (0.69) 0.18 Agriculture Value Added (% of GDP) -.07 (0.92) -.12*** (10.82) -.13*** (5.83) -.11*** (5.65) -.14*** (4.94) 0.09 (1.07) -0.003 (0.18) Openness .04** (2.48) .04** (2.38) .008 (1.19) .02*** (2.92) -.02* (1.71) -.01 (1.28) .01 (1.37) .03*** (2.84) -.009 (0.82) -.004 (0.35) 0.08*** (4.01) 0.08*** (3.38) 0.005* (1.73) 0.005* (1.65) Aid (% of GNI) .05** (2.00) .02 (0.58) .13** (2.52) .04 (0.96) .17*** (3.47) .06** (2.26) .11** (2.39) .04 (1.20) .10* (1.81) .02 (0.71) 0.002 (0.04) 0.008 (0.19) (1.18) 0.03 (0.59) 0.02 Urban Population (% of total) .36** (2.55) .39** (2.40) -.07*** (6.01) .008 (1.08) -.06*** (8.34) .02 (1.18) -.05*** (8.35) .02* (1.83) -.06*** (7.95) .009 (0.64) 0.86*** (3.23) 0.88*** (3.18) -0.003 (0.33) -0.0002 (0.02) Rule of Law .34 (0.63) .37 (0.60) 4.60*** (9.01) 6.85*** (19.46) 2.60*** (3.92) 5.60*** (10.16) 4.21*** (6.66) 6.57*** (16.27) 2.26*** (2.92) 5.06*** (8.76) -0.72 (0.74) (1.02) -1.07 (1.83) 0.57* (0.86) 0.39 Observations 655 630 655 630 655 630 566 542 566 542 261 246 561 540 Number of Countries 102 100 102 100 102 100 102 100 102 100 96 92 102 100 R-Squared 0.37 0.32 0.69 0.70 0.89 0.87 0.69 0.69 0.87 0.87 AR (1) -0.81 (0.418) (0.477) -0.71 (0.026) -2.23 (0.024) -2.26 AR (2) -0.37 (0.713) (0.920) -0.10 (0.275) -1.09 (0.263) -1.12 J. Hansen Test 53.82 (0.230) (0.219) 53.10 (0.346) 19.76 (0.262) 18.01
*10% significance, **5% significance, ***1% significance level
Robust (absolute) t-values in bracket for FE and Robust (absolute) z-values in bracket for PCSE and GMM. Aid (% of GNI) is used as endogenous variable.
10 Former British Colony, Federal and land are used as instruments for rule of law and control of corruption
(A.9-A.10) when log of per capita GDP is used as a regressor. Federal and former British Colony are both dummy variables having value equal to 1 if a country has federal administration system and is Former British Colony and zero otherwise. Land is measured in square kilometers. The choice of these variables as instrument is based on the fact that countries which are former British colonies still implement some administrative laws inherited from British system. Likewise, federal can also serve as good instrument for rule of law and control of corruption. It is assumed that countries with large land area usually have difficulty to enforce rule of law and also difficult to control corruption, that is why, it is also considered as an instrument for these variables.
In PCSE (common correlation), we find no effect of foreign aid (% of GNI) on tax revenue performance when estimations is done by taking agriculture value added (% of GDP) as a regressor. This may be because in most cases foreign aid received is not disbursed to agriculture sector. However, we find positive effect of foreign aid (% of GNI) on tax-to-GDP ratio in case of panel specific PCSE. We also find that the coefficient on urban population (% of total) is negatively associated with tax performance, when we did regression by taking log of per capita GDP as one of regressors and found insignificant results in case of agriculture value added (% of GDP).
Finally, the effect of rule of law on tax performance appeared to be highly significant which indicates that improvement in governance indicators (rule of law) could be very useful to increase tax revenue specially in developing countries. This result is also in line with Bird et al., (2004).
5.2. Tax Performance Analysis of Countries by Income-Group
In this section, panel specific standard error corrected estimation technique is employed to derive
tax performance of countries according to their income groups.The choice of technique is based
on goodness of fit. The selected results for tax performance analysis of countries according to their income groups are given in Table 5.2. The log of per capita GDP is positively significant in middle and high income groups, whereas, it is negatively significant in low income group. The negative sign in case of low income group might indicate that these countries are exploiting their tax resources to its full potential because after attainment of a certain threshold level of income, potential for further revenue mobilization through taxes became negative (Clist and Morrissey, 2010).
Another interesting finding is related to agriculture sector as it appears to be insignificant in low income group. This conclusion is contrary to findings of Gupta (2007). The results for whole sample computed by PCSE specification are also against this finding showing highly significant negative relationship between agriculture and tax revenue. However, results of middle income group for this variable strongly support our hypothesis.
The findings of high income group regarding openness is different from both low and middle income group giving negative sign for openness. In this case, we lend support from the
conclusion of Baunsgaard and Keen (2009) who suggest that high income countries do not depend on trade taxes for revenue collection. They basically use these taxes as a tool to retain trade at a certain level.
We find no link between aid (% of GNI) and tax-to-GDP ratio in low income countries. However, in middle income countries foreign aid (% of GNI) is significantly positively related to tax revenue. These findings are similar to Clist and Morrissey (2010) but contradictory to Gupta (2007). One possible reason might be that middle income countries have better institutional framework to efficiently use the foreign aid received as compared to low income countries. High income group give mixed results on foreign aid (% of GNI).
The negative sign on coefficient of urban population (% of total) except one specification of middle income countries is contrary to Davoodi & Grigorian (2007) findings and also does not support our hypothesis that urban population require more public sector facilities which motivate them to pay more taxes, thereby, also increase tax revenue. The rule of law appeared to be as important player in determination of tax revenue collection in middle and high income groups. Table 5.2: Income-Wise Determinants of Tax Revenue Performance: 1996-2005
Panel Specific Correlation Coefficient
Low-Income Countries Middle-Income Countries High-Income Countries
Constant 17.43*** (5.98) 9.79*** (10.22) 5.57** (2.46) 14.44*** (14.63) -68.56*** (2.95) 31.97*** (27.96)
Log Per Capita GDP -1.52*** (2.99)
.74*** (3.14)
11.65*** (4.33)
Agriculture Value Added (% of GDP) -.008 (0.59) -.14*** (5.80) .21* (1.95) Openness .09*** (6.46) .10*** (13.65) .07*** (9.77) .07*** (8.88) -.06*** (9.34) -.04*** (6.20) Aid (% of GNI) -.009 (0.34) .006 (0.23) .10*** (7.74) .04* (1.67) 1.67** (2.31) -.37 (0.80) Urban Population (% of total) -.06*** 3.42 -.14*** (10.60) .01 (0.98) -.004 (0.36) -.09** (2.34) -.02 (0.57) Rule of Law .36 (0.78) .001 (0.00) 1.78*** (3.50) 1.63*** (3.01) -.34 (0.22) 4.30*** (3.19) Observations 146 144 318 307 191 179 Number of Countries 22 22 51 50 29 28 R-Squared 0.85 0.88 0.93 0.93 0.96 0.96
*10% significance, **5% significance, ***1% significance level Robust (absolute) t-values in bracket
5.3. Tax Effort Analysis of Pakistan
Now, we focus on construction of tax effort indices for Pakistan to analyze its potential for further tax collection despite having weak economic conditions. We also compare Pakistan‟s tax effort indices with other middle income countries having less income level in terms of GDP and high current tax-to-GDP ratio as compared to Pakistan. We measure two tax effort indices for Pakistan by using regression 6 and regression 12 of Panel Corrected Standard Error (PCSE) estimation with common correlation coefficient given in Table A.5. The difference between two regression results is that regression 6 include log of per capita GDP as one of explanatory variables and regression 12 include agriculture value added (% of GDP) as one of independent variables. The selection of these regressions is based on economic rationale and goodness of fit. Moreover, the 2010 Corruption Perception Index mention Pakistan‟s score at 2.3 on a scale between zero (highly corrupt) to ten (highly clean) which also indicates the importance of choice of this regression to construct tax effort index.
Source: Author’s Calculation Source: Author’s Calculation
Figure 5.1 and Figure 5.2 shows the actual and predicted values of tax-to-GDP ratio of Pakistan over the period of 1996-2005. With log of per capita GDP, Figure 5.1 shows highest gap between actual and predicted tax revenue in 2003 with a value of 2.0 percent of GDP, however, if we takes into account the presence of agriculture value added (% of GDP), the discrepancy between actual and potential tax revenue was highest during 2000 which was equivalent to 2.4
0 2 4 6 8 10 12 14 16 1996 1998 2000 2002 2003 2004 2005
Figure 5.1: Pakistan's Actual and Potential Tax Revenue (% to GDP) Based on Log of Per Capita GDP Estimation: 1996-2005
Actual Tax Revenue (% of GDP) Predicted Tax Revenue (% of GDP)
0 2 4 6 8 10 12 14 16 1996 1998 2000 2002 2003 2004 2005
Figure 5.2: Pakistan's Actual and Potential Tax Revenue (% to GDP) Based on Agriculture
Value Added (% of GDP) Estimation: 1996-2005
Actual Tax Revenue (% of GDP) Predicted Tax Revenue (% of GDP)
percent of GDP as shown in Figure 5.2. In order to derive tax effort indices for Pakistan, we divide actual tax-to-GDP ratios by predicted tax-to-GDP ratios as shown in Table 5.3.
Both tax effort indices presented in Table 5.3 exhibits value above unity during 1996-1998 followed by a declining trend. One explanation is that effective measures such as policy reforms taken by government of Pakistan during early 1990‟s could not be sustained in later periods. The other explanation is that regression results used to estimate potential tax revenue are biased due to greater number of missing observations for institutional variables and relatively short time duration. However, we also computed average and predicted tax-to-GDP ratios for a period of 1975-2006 presented in Figure A.1 and Figure A.2, which shows almost similar declining trend of tax-to-GDP ratios as we obtain in Figure 5.1 and Figure 5.2. Hence, overall decline in tax effort indices is evident. Table A.3 shows tax effort indices of Pakistan computed by different studies which also refers the effectiveness of tax policies during early 1990s because Pakistan shows tax effort ratio of above unity during 1990s.
With log of per capita GDP as one of explanatory variables in regression, the value of average tax effort index stood at 0.98 over the period from 1996-2005. Whereas, average index of tax effort reached at 0.96 when we look at regression results with agriculture value added as one of explanatory variable over the same time period. However, the important point to be noted is that the tax effort indices shows an overall declining trend.
Table 5.3: Tax Effort Indices of Pakistan: 1996-2005
Year Tax Effort with
log of per capita GDP Agriculture Value Added Tax Effort with (% of GDP) 1996 1.23 1.24 1998 1.14 1.14 2000 0.85 0.83 2002 0.91 0.88 2003 0.88 0.83 2004 0.93 0.93 2005 0.90 0.89
Table 5.4:Average Tax Effort Indices for Middle Income Countries: 1996-2005
Country Per Capita GDP Agriculture Share Country Per Capita GDP Agriculture Share Index
No
Rank Index No
Rank Index No Rank Index No Rank Antigua and Barbuda 0.58 26 0.62 26 Lesotho 2.13 1 1.80 1
Belize 0.91 18 0.96 17 Mauritius 0.65 25 0.64 25
Botswana 1.42 6 1.3 7 Morocco 1.11 13 1.02 15
Cameroon 1.55 3 1.46 4 Pakistan 0.98 16 0.96 16
Côte d'Ivoire 1.14 12 1.03 14 Peru 0.76 21 0.69 22
Dominica 1.06 14 1.05 12 Seychelles 1.37 8 1.37 5
Dominican Republic 0.71 24 0.67 24 Sri Lanka 0.73 22 0.74 21
Egypt 0.92 17 0.94 18 St. Lucia 0.85 20 0.88 19 Gabon 1.69 2 1.67 2 St. Vincent and Grenadines 1.06 15 1.04 13 Ghana 1.29 10 1.06 11 Swaziland 1.32 9 1.26 8 Grenada 0.86 19 0.87 20 Syria 1.24 11 1.26 9 Guyana 1.5 5 1.6 3 Uruguay 0.72 23 0.67 23 Kiribati 1.55 4 1.3 6 Zambia 1.38 7 1.25 10
Source: Author’s Calculation
Table 5.4 shows comparison of Pakistan‟s tax effort indices with the tax effort indices of selected middle income countries. Despite having lower income level, Gabon, Cameroon and Guyana are performing better than Pakistan. It seems that these countries are exploiting their domestic resources to generate tax revenue more than their capacity. Both tax effort indices for Pakistan indicate that there still exist potential to further increase tax revenue which obviously require policy reform with a focus on raising the efficiency of existing VAT structure.
6. Conclusion
The main objective of this paper was to investigate the role of governance indicators along with other traditional determinants in tax revenue performance of 104 countries over the period from 1996-2005. The other objective was to construct the tax effort indices of Pakistan for the same time period to analyze existing potential for revenue collection from domestic resources.
Among various techniques PCSE provide significant results. Log of per capita GDP, agriculture value added (% of GDP), foreign aid (% of GNI), rule of law and control of corruption appeared as significant variables of tax-to-GDP ratio. The primary contribution of this study to the literature has been to extend the paradigm of research on non-conventional determinants by employing different econometric specifications and also taking into account the possible biases in the regressions.
The construction of tax effort indices for Pakistan based on global regression results of tax revenue performance are rather abstract and highly model dependent but it gives good insight of tax revenue performance of Pakistan. It also indicates the presence of overall downward trend in Pakistan‟s tax-to-GDP ratio. It clearly recommend that serious efforts are required to raise the tax-to-GDP ratio of Pakistan. The results obtained in this study also exhibits the need of structural reform of taxation system in Pakistan to collect taxes from other domestic sources with the same magnitude as collected from trade taxes. One way to obtain this goal is to enhance the efficiency of VAT. Structural reforms with the aim to broad the tax base could also be another option to increase tax-to-GDP ratio. Last but not the least, there is vital need to improve institutional quality in order to enhance the tax revenue and to build the confidence of general public on government‟s policies which could also motivate them to willingly pay their tax payments.
Acknowledgements: I am extremely grateful to Mr. Jörgen Levin (supervisor) for his support and guidance to complete
this project. I am thankful to Mr. Patrik Karpaty for his valuable suggestions which helped me to improve my work. I also want to thank my father (Abdul Ghani) and sister (Arshia Ghani) whose support and prayers were with me all the time. Last but not the least, thanks to my brother Jaffer Bhai whom I want to dedicate my work because I would not even be able to continue my studies without his moral support and financial help. May Allah bless you all. Ameen.
References
Agbeyegbe, T.D., Stotsky, J. and WoldeMariam, A., 2004. Trade Liberalization, Exchange Rate Changes, and Tax Revenue in Sub-Saharan Africa. International Monetary Fund, Washington DC, IMF Working Paper No. 178.
Arellano, M. and Olympia, B., 1995. Another Look at the Instrumental Variable Estimation of Error-Components Models. Journal of Econometrics, Vol. 68, pp. 29–51.
Arellano, M. and Steve, B., 1991. Some Tests of specification for Panel Data: Monte Carlo Evidence and an Application to Employment Equations. The Review of Economic Studies, Vol. 58, pp. 277–97.
Auriol, E. and Wartlers, M., 2005. Taxation Base in Developing Countries. Journal of Public Economics, April, 89(4), pp. 625-646.
Bahl, R.W., 1971. A regression approach to tax effort and tax ratio analysis. International Monetary Fund, Washington DC, IMF Staff Papers 18: 570–612.
Baum, C.F., Schaffer, M.E., and Stillman, S., 2003. Instrumental Variables and GMM: Estimation and Testing. Department of Economics, Boston College, Boston, February, Working Paper No. 545.
Baunsgaard, T. and Keen, M., 2010. Tax Revenue and (or?) Trade Liberalization. Journal of Public Economics, Vol. 94, pp.563-577.
Bird, R.M., Martines-Vazquez, J. and Torgler, B., 2004. Societal Institutions and Tax Effort in Developing Countries. International Studies Program, Andrew Young School of Policy Studies, Georgia State University, Georgia, Working Paper 04-06.
Blundell, R. and Steve B., 1998. Initial Conditions and Moment Restrictions in Dynamic Panel Data Models. Journal of Econometrics, Vol. 87, pp. 115–43.
Chelliah, R.J., 1971. Trends in Taxation in Developing Countries. International Monetary Fund, Washington DC, IMF Staff Papers 18: 254–325.
Clist, P. and Morrissey, O., 2011. Aid and Tax Revenue: Signs of a Positive Effect Since the 1980s. Journal of International Development, Vol. 23, pp. 165-180.
Davoodi, H. and Grigorian, D., 2007. Tax Potential vs. Tax Effort: A Cross-Country Analysis of Armenia's Stubbornly Low Tax Collection. International Monetary Fund, Washington DC, May, IMF Working Paper No. 106.
Franco-Rodriguez S., Morrissey O., and Mcgillivray M., 1998. Aid and the Public Sector in Pakistan: Evidence with Endogenous Aid. Centre for Research in Economic Development and
Frain, J.C., 2006. STATA Commands for Unobserved Effects Panel Data. Trinity College, Dublin.
Government of Pakistan, Fiscal Policy Statement. Various Issues. Debt Policy Coordination Office (DPCO), Ministry of Finance, Islamabad.
Government of Pakistan, Pakistan Economic Survey. Various Issues. Ministry of Finance, Islamabad.
Gupta, A.S., 2007. Determinants of Tax Revenue Efforts in Developing Countries. International Monetary Fund, Washington DC, July, IMF Working Paper No. 184.
IMF, Pakistan: Article IV Consultation and First Review Under the Stand By Arrangement-Staff Report; Staff Statement; Public Information Notice and Press Release on the Executive Board Discussion; and Statement by the Executive Director for Pakistan. Various Issues. International Monetary Fund, Washington DC.
IMF, 2010. Pakistan: Request for an Extension of the Stand By Arrangement-Staff Report; Press Release on the Executive Board Discussion. International Monetary Fund, Washington DC. IMF Country Report No. 10/384.
IMF, IMF Financial Statistics (IFS), International Monetary Fund Washington DC. CD-ROM database.
Kaufmann D., Kraay A. and Mastruzzi M., 2010. The Worldwide Governance Indicators: Methodology and Analytical Issues. Global Economy and Development, The Brookings Institution, Washington DC.
Khan, A., 1992. Determinants of Income Tax Base in Pakistan: A Policy Review. Pakistan Development Review, Winter Issue, 31(4), Part II, pp.1123-1142.
Leuthold, J.H., 1991. Tax Shares in Developing Countries: A Panel Study. Journal of Development Economics, Vol. 35, pp. 173–185.
Lotz, J.R. and Elliott R.M., 1967. Measuring „Tax Effort‟ in Developing Countries. International Monetary Fund, Washington DC, IMF Staff Papers 14: 478-99.
Piancastelli, M., 2001. Measuring the Tax Effort of Developed and Developing Countries: Cross Country Panel Data Analysis, 1985/95. The Institute for Applied Economics Research (IPEA), Brasilia, Texto Para Discussao No. 818.
Roodman, D., 2006. How to Do xtabond2: An introduction to "Difference" and "System" GMM in Stata. Center for Global Development, Working Paper No. 103.
Stotsky, J.G. and WoldeMariam, A., 1997. Tax Effort in Sub-Saharan Africa. International Monetary Fund, Washington DC, IMF Working Paper No. 107.
Tanzi, V. and Davoodi, H.R., 2000. Corruption, Growth and Public Finances. International Monetary Fund, Washington DC, November, IMF Working Paper No. 182.
Tanzi, V. and Zee, H.H., 2000. Tax Policy in Emerging Markets: Developing Countries. International Monetary Fund, Washington DC, IMF Working Paper No. 35.
Tanzi, V., 1987. Quantitative Characteristics of the Tax Systems of Developing Countries. In: D.Newbery and N.Stern, ed. The Theory of Taxation for Developing Countries. New York: Oxford University Press. pp. 205–241.
Teera, J.M. and Hudson, J., 2004. Tax Performance: A Comparative Study. Journal of International Development, Vol.16, pp. 785-802.
Transparency International: The Global Coalition Against Corruption, 2010. Corruption Perception Index 2010, Berlin.
World Bank, 2010a. World Development Indicators (WDI), The World Bank, Washington DC. World Bank, 2010b. Worldwide Governance Indicators, The World Bank, Washington DC.
Appendix
(%)
Source: Author’s Calculation
(%)
Source: Author’s Calculation
0 2 4 6 8 10 12 14 16 18 20 1975 1976 1977 1978 1979 1980 1981 1982 1983 1984 1985 1987 1988 1989 1990 1991 1992 1993 1994 1995 1996 1997 1998 1999 2000 2001 2002 2003 2004 2005 2006
Figure A.1: Pakistan's Actual and Potential Tax Revenue (% to GDP) Based on Log of Per Capita GDP Estimation: 1975-2006
Actual Tax Revenue (% of GDP) Predicted Tax Revenue (% of GDP)
0 5 10 15 20 25 1975 1976 1977 1978 1979 1980 1981 1982 1983 1984 1985 1987 1988 1989 1990 1991 1992 1993 1994 1995 1996 1997 1998 1999 2000 2001 2002 2003 2004 2005 2006
Figure A.2: Pakistan's Actual and Potential Tax Revenue (% to GDP) Based on Agriculture Value Added (% of GDP) Estimation: 1975-2006
Table A.1: Summary of Non-Conventional Studies of Tax Revenue
Study Dependent Variable
Explanatory Variables Comments
Davoodi and Grigorian (May 2007)
[Armenia, panel data]
Total Tax revenue as a percent of GDP
(1) GDP per capita (2) Agriculture share in GDP (3) Urban population (4) Openness (5) Inflation (6) oil (7) Shadow (proxy to low tax morale and willingness to pay) (8) composite (proxy to institutional quality) (9) VAT
Significant: (3), (7) and (8) Insignificant: (2) and (4) Bird, Martines-Vazquez and Torgler (2004) [Developing countries; cross section data, 1990-99] (i)Tax revenue as a share of GDP (ii) Current revenue/GDP (excluding grants)
(1) GDP per capita (2) Rate of population growth (3) Openness (4) Tax morale and Shadow economy (5) Institution Six governance index by world bank (6) ICRG index (corruption in the government, rule of law, bureaucratic quality, ethnic tension) (7) Regulation to entry (number of procedures and official time to complete the process) (8) Non agriculture share of GDP
Very Significant: (5), (6) and (8) Insignificant: (3)
Auriol and Wartlers (2005)
[Developing and High Income countries] Central Govt Tax Revenue as a percentage of GDP divided by (100-Tax revenue)
(1) GNP per capita in dollars 1996 (2) Dev: dummy that equals 1 if country has per capita GNP greater than $10,000 (3) Agriculture value added as a percentage of GDP 1996 (4) Population per square kilometer divided by 10,000 (5) Country population in millions (6) Free Trade Openness (7) Fuels, minerals and metals as share of 1993 merchandise exports (8) Time Cost: minimum time in days required to meet government requirements for establishing a new company (9) Africa Dummy equals 1 if country is on African continent (10) OECD: dummy equal 1 if country belongs to OECD (11) Common Law: dummy equals 1 if country has common law system (12) Democratic 46: dummy equals 1 if country has been democratic in 46 years (13) Federal: dummy equals 1 if country has federal structure (14) Former British Colony: equals 1 if the country is a former British colony (15) HIPC: dummy equals 1 if country was classified as heavily indebted poor by IMF and World Bank in May 2001 (16) Land: 1995 land area in millions square kilometers (17) Sunk Cost: Direct cost as a fraction of 1997 GDP per capita (18) Transition: dummy equal 1 if country is in transition (19) TISCORE: 1996 Transparency scores
Significant: (1), (4), (9), (12) and (16-18) Gupta (2007) [Developing Countries] Central Govt Revenue (excluding grants)
(1) GDP per capita (2) Agriculture share in GDP (3) Manufacturing share in GDP (4) Share of imports in GDP (5) Tax revenue from trade (% of total revenue) (6) Tax revenue from exports (% of total revenue) (7) ICRG index (political stability, economic stability, corruption, law and order and government stability) (8) Ratio of debt and Aid to GDP (9) Tax revenue from goods and services (%of total revenue) (10) Tax revenue from income, profits and capital gains (% of total revenue) (11) Highest marginal tax rate, individual rate (%) (12) Highest marginal tax rate, corporate rate (%) (13) Average tariff
Significant: (1), (2), (7) and (8)
Teera and Hudson (2004) [Developed and Developing countries]
Tax-to-GDP ratio
(1) GDP per Capita, in international dollars (2) Ratio of agriculture to GDP (3) Ratio of manufacturing to GDP (4) Population density (people per sq km) (5) Openness (6) Shadow indicating tax evasion (7) Time trend to capture global trend in taxation (8) Ratio of aid to GNP (9) Ratio of total external debt to GDP
Table A.2: Pair-Wise Correlation of Variables
tr lpcgdp agr Open Aid_GNI Urban linf va pv ge rq rl cc
tr 1.00 lpcgdp 0.68 1.00 agr -0.59 -0.86 1.00 Open 0.24 0.27 -0.29 1.00 Aid_GNI -0.30 -0.63 0.64 -0.08 1.00 Urban 0.44 0.76 -0.69 0.09 -0.52 1.00 Iinf -0.34 -0.47 0.35 -0.19 0.26 -0.34 1.00 va 0.68 0.72 -0.58 0.11 -0.33 0.48 -0.39 1.00 pv 0.63 0.71 -0.59 0.34 -0.27 0.46 -0.45 0.74 1.00 ge 0.70 0.87 -0.70 0.17 -0.47 0.66 -0.47 0.81 0.76 1.00 rq 0.60 0.83 -0.67 0.17 -0.48 0.62 -0.48 0.77 0.71 0.91 1.00 rell 0.70 0.84 -0.68 0.19 -0.44 0.60 -0.49 0.83 0.83 0.95 0.89 1.00 cc 0.70 0.84 -0.66 0.16 -0.41 0.64 -0.47 0.81 0.78 0.95 0.89 0.95 1.00 Note: tr, lpcgdp, agr, open, aid_GNI, urban, linf, va, pv, ge, rq, rl, cc stands for Tax Revenue (% of GDP), Log per Capita GDP, Agriculture Value Added (% of GDP), Openness, Foreign Aid (% of GNI), Urban Population (% of Total), Log Inflation, Voice and Accountability, Political Stability_No Violence, Government Effectiveness, Regulatory Quality, Rule of Law and Control of Corruption respectively.
Table A.3: Pakistan’s Inter-Temporal Comparison of Tax Effort Indices Tax Effort
Index 1966/68* Index 1969/71** Tax Effort Index 1972/76*** Tax Effort Index 1985/95+ Tax Effort Average Index Index 1975/06++ Tax Effort
(a) (b) (c) (d) d=(a+b+c+d)/4
0.75 0.73 0.96 1.29 0.93 0.96
*Chelliah Source: Cited in Piancastelli, 2001, p.13 and Author’s calculation **Chelliah
***Tait +Piancastelli ++present study
Table A.4: Determinants of Tax Revenue Performance: 1996-2005 (Fixed Effect) 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 Constant -14.31* (1.95) -23.84* (1.80) -23.98* (1.75) -23.56* (1.89) -21.60 (1.65) -22.92* (1.68) -22.85* (1.73) -22.59* (1.69) -23.99* (1.75) 24.87*** (32.09) 4.58 (0.54) 6.53 (0.77) 1.05 (0.12) .37 (0.04) -.83 (0.09) .35 (0.04) .39 (0.05) -1.22 (0.13)
Log Per Capita GDP 4.68*** (4.92) 3.53* (1.88) 3.81** (2.06) 3.12* (1.75) 2.83 (1.51) 2.89 (1.53) 3.02 (1.61) 2.94 (1.56) 2.96 (1.55) Agriculture Value Added (% of GDP) -.17*** (3.67) -.10 (1.52) -.13 (1.63) -.08 (1.06) -.06 (0.87) -.06 (0.79) -.07 (0.95) -.07 (0.92) -.07 (0.83) Openness .03*** (2.80) .04*** (3.07) .04** (2.43) .04** (2.42) .04** (2.47) .04** (2.50) .04** (2.48) .04** (2.53) .03** (2.33) .04** (2.49) .04** (2.31) .04** (2.32) .04** (2.39) .04** (2.41) .04** (2.38) .04** (2.45) Aid (% of GNI) .06*** (2.73) .06*** (2.71) .05** (2.02) .05* (1.97) .04 (1.33) .05** (2.11) .05** (2.00) .04 (1.28) .04 (1.27) .03 (1.19) .01 (0.31) .01 (0.41) .004 (0.13) .02 (0.61) .02 (0.58) .003 (0.10) Urban Population (% of Total) .31** (2.21) .26* (1.93) .36** (2.52) .36** (2.51) .37** (2.51) .36** (2.46) .36** (2.55) .38** (2.53) .33** (2.10) .29* (1.90) .38** (2.40) .39** (2.43) .41** (2.41) .39** (2.45) .39** (2.40) .42** (2.44) Log Inflation .21 (1.49) .26* (1.75) Voice and Accountability -.10 (0.16) .31 (0.54) Political Stability, No Violence .38 (1.10) .55 (1.47) Government Effectiveness .30 (0.46) .49 (0.63) Regulatory Quality -.03 (0.06) .25 (0.41) Rule of Law .34 (0.63) .37 (0.60) Control of Corruption .60 (1.24) .60 (1.14) Observations 1004 968 847 666 656 642 653 655 642 964 934 822 640 630 617 628 630 617 Number of Countries 103 102 96 102 102 102 102 102 102 101 100 94 100 100 100 100 100 100 R-Squared 0.4616 0.3884 0.4556 0.3610 0.3679 0.3616 0.3598 0.3677 0.3691 0.3511 0.3214 0.3382 0.3199 0.3257 0.3199 0.3100 0.3204 0.3248
*10% significance, **5% significance, ***1% significance level Robust (absolute) t-values in brackets
