INCOME INEQUALITY, FISCAL POLICY AND INSTITUTIONAL ...



Income Inequality, Fiscal policy and Institutional Development: On The redistributive effects of IncomeTax Progressivity and Corruption.

ELISA BARONI (UK DEPT. OF WORK AND PENSIONS)

Cathal O’Donoghue (Dept. Economics, National University of Ireland)

Abstract

This paper draws on two existing strands in the public finance literature: one tries to measure and compare the efficiency costs of redistribution through progressive taxation versus public expenditures. The other aims to quantify the effects of institutional failures, e.g. corruption, on inequality.

On the one hand, the paper tests and finds econometric evidence against the main policy tenets stemming from the traditional normative approach (Optimal Income Taxation theories) - which advocates the use of public expenditures rather than tax to deal with equity concerns - supporting instead the idea that both income tax and expenditure progressivity are necessary to reduce income inequality. Furthermore, the relationship between income tax progressivity and inequality is shown to be non-linear. On the other hand, the paper strongly supports theories arguing that countries with higher levels of corruption tend to be more unequal. In particular, it is able to show that the channel is indeed fiscal: corruption reduces the effective degree of progressivity of a given tax system.

Overall, the paper's main technical contribution to the debate lies in the design of an income tax progressivity index. The calculation of such index over a cross-country sample allows us to model explicitely the effects of fiscal policy and corruption on income inequality.

* The first draft of this paper was written as part of the MSc Public Financial Policy degree requirements of Elisa Baroni (2003). She would like to thank her supervisor Dr. J. Leape. We would like to thank W. Asegedesh, H. Davoodi for providing useful data; G. Palomba and R. Puglisi for providing useful comments. This paper was partly written while Cathal O’Donoghue was an ICER fellow at the University of Turin. He gratefully acknowledges the hospitality provided.

** For further queries please contact: elisa.baroni@dwp..uk

Introduction

After much neglect, recent changes in income and wage distribution worldwide have brought inequality back to the fore of academic debate, particularly since they display complex patterns, and an elusive relationship with growth performance.

Overall, evidence since the 1990s both across countries and across time shows a diversity of experiences, characterised by sharp increases in disposable income and/or wage inequality, observable both in some industrialised (OECD, 1996) and in some developing countries (e.g. UK, US, Canada, Mexico).

Although the role of fiscal policy in explaining changes in disposable incomes is unquestionable, the existing literature, both empirical and theoretical, disagrees on the actual effects of the main redistributive instruments – namely taxation and public expenditures - and on how progressive they should be. Furthermore, their effects also depend on the institutional context. Widespread corruption in particular, by affecting the state’s ability to tax and spend efficiently, should have an impact on redistribution and, consequently, on income inequality.

In a nutshell, the aim of this paper is to quantify the effects of different redistributive policies on income inequality, while also testing whether and how institutional development could interfere with such effects.

In Section 1.1 – 1.3, we will provide a review of the literature explaining current trends in inequality, and offer some theoretical motivations behind the resurgence of interest in this topic; we will briefly argue the general rationale for redistribution, and why corruption is bad for redistribution. In Section 2, we will provide an analysis of the traditional policy view, broadly associated with optimal taxation theory, prescribing modest tax progressivity, and encouraging the pursuit of redistributive objectives mostly through the design of well-targeted public expenditures. In Section 3, we will present a theoretical model making the case for the complementarity of progressive taxation and progressive public expenditures, implying that neither alone is sufficient to achieve a given redistributive objective. In Section 4, we will argue why good institutions, in particular, the absence of corruption, are also necessary to make a nominally progressive fiscal policy system effective in redressing inequality. Finally, in Section 5 we will present an econometric model testing the validity of these ideas over a new cross-country data set.

Why Inequality Matters

Inequality should matter to economists, if not per se, because it can have efficiency implications both at the macro level (e.g. more unequal countries grow less), and at the micro level (e.g. inequality can lead to poverty traps in presence of credit constraints).

Issues of inequality and income distribution have not always featured prominently on the economist’s agenda. Classical economists have traditionally been interested in the distribution of factor payments more than personal incomes. Moreover, factor shares have traditionally remained fairly constant. As for income distribution, time-series variations in Gini coefficients of most countries have been usually modest, and in the eyes of economists such as H. Aaron, “following these data was like watching the grass grow”.

Nevertheless, there is overall consensus that, in the 1990s, income distribution “has been brought back in from the cold” . There are two chief reasons for this resurgence in interest shown by the economic profession, one empirical and one theoretical.

(a) Empirical Motivations:

Recent income data for the past two decades, have revealed that:

(i) cross-country inequality (as measured by average GDP per capita) between poor and rich countries is widening (although the claim is subject to methodological debates on what data should be used e.g. consumption- or income-based inequality measures).

(ii) within some countries, particularly industrialised ones, disposable income and earnings distributions are also becoming more unequal, yet without a clear systematic trend. For instance, over 1975-1995, wage inequality rose most dramatically in the U.K., and somewhat in the U.S., Japan, Sweden, but remained stable in Canada, France or Italy. Over the same period however, Canada and the U.S. witnessed a large increase in disposable income inequality; while the UK saw its disposable income inequality measures increase but only after 1984. Many transition economies, e.g. Hungary, also experienced a systematic worsening of their inequality indexes.

Overall, the lack of a clear systematic trend even among countries with similar economic performance begs for some explanation.

By far the most popular explanation for both (i) and (ii) has to do with trade and financial liberalisation, on the one hand, and rapid rates of technological innovation in the developed world on the other, rising the marginal return of skilled human capital (Tanzi and Chu, 1998; Atkinson 1997). While developing countries' adoption of open trade policies is often used to explain the growing income gap worldwide (e.g. between the so-called globalisers and the rest), the concentration of technology-intensive industries predominantly in the North, and the consequent fall in relative wages of unskilled workers, could explain the latest increase in inequality in many developed countries (Wood, 2000). By the same token however, economic liberalisation should lower income and wage inequality in developing countries, due to higher demand for unskilled workers increases there, and higher productivity from transfer of technological know-how.

Clearly, the data do not fully validate this line of argument. On the one hand, there are examples of “open” developed countries that have seen unskilled workers’ real wages actually increase (e.g. Sweden) or not change (e.g. Germany) (Blundell and MaCurdy, 1999). On the other hand, many “open” developing countries, e.g. Mexico, have also witnessed large deterioration in their Gini coefficients over the past two decades (Wood, 1997). Furthermore, the positive effects of liberalisation on growth have been observed mostly in those countries which had previously tackled inequality by substantial redistribution, e.g. by expanding access to education (e.g. East Asia versus Latin America), hinting to a less straightforward relationship between liberalisation, inequality and growth than commonly thought.

Furthermore, international trade theories explain inequality dynamics by looking at changes in relative factor prices, which however are not directly translatable into distributional changes. These rather depend on (i) the net changes affecting all sources of personal income beside earnings, including transfers (ii) the institutional context and the ability to implement fiscal policy in general.

(b) Theoretical Motivations

The second motive for a resurgence of interest in inequality stems from the triumph over the last two decades of the imperfect information and imperfect markets perspective in mainstream economic theory. In contrast with the classical tenets of welfare economics and Pareto efficiency, a key analytical contribution of this perspective has been to treat equity and efficiency as complementary rather than substitute economic goals (Kanbur and Lustig, 1999).

The theoretical argument for the non-separability of equity and efficiency is first due to Mirrlees (1971). More recently, the idea of non-separability has been pushed further to claims that an “equitable” redistribution of income is actually necessary to achieve efficient market outcomes, e.g. given imperfect information in the credit markets and the related problem of credit rationing. Efficiency in fact requires that resources are allocated according to talent and not to an individual’s wealth. However the impossibility for talented people without enough collateral - a screening instrument against moral hazard - to borrow can lead to substantial efficiency loss and poverty traps unless some wealth redistribution takes place. Redistribution should not only reduce credit rationing, but also occupational choice and educational constraints, hence raise total output, and lower the likelihood of intergenerational persistence of poverty and inequality (Benabou 1996a, Banerjee and Newman 1993).

The political economy literature associated with “median voter” models also recognises the non-separability of equity and efficiency. Again, according to this class of models (e.g. Persson and Tabellini, 1994), the greater the degree of ex-ante inequality, the lower is equilibrium growth, because of the inefficiencies generated by the redistributive schedule chosen by the median voter with below-average income (resulting in higher taxation hence lower savings and investment by the richest). The main problem with these models is however their empirical failure with regard to the actual effects of redistribution (e.g. transfers) on growth, which many found to be actually positive (Benabou 2000).

Why Redistribution matters

From the normative viewpoint, redistribution is justified only by non-minimalist views of the state (Atkinson and Stiglitz, 1980). Boadway and Keen (1999) identify three main motives for redistribution: (i) the pursuit of social justice (ii) the achievement of mutually advantageous efficiency gains (against the traditional equity efficiency trade-off) (iii) the exercise of self-interest through the coercive power of the state.

We want to limit our focus to motivation (ii), i.e. on the efficiency gains of redistribution. These gains can be interpreted both as a Pareto improvement, or as an increase in total output. Keeping this in mind, redistribution can lead to both types of efficiency gains whenever market failures are present. For instance, if the utility of the rich depends on the welfare of the poor (i.e. considering the latter as a kind of public good), voluntary redistribution through charity can lead to Pareto improvements. Redistribution can also reduce inefficiency caused by imperfect information, such as undersupply of private insurance or credit rationing. In the latter case, a redistribution of wealth towards borrowers is likely to move investment levels closer to first best. Another example where redistribution can lead to increase in total output is land reform in the presence of informational constraints, tenants’ shirking and rising efficiency wages (Dasgupta and Ray, 1986) . Finally, as previously noticed, contrasting evidence from East Asia and Latin America since the 1970s points to the role played by redistribution in fostering better growth rates (the so-called Lucas Puzzle, 1963).

Why Corruption matters

Corruption broadly refers to “the abuse of public office for private gain” (Kauffman, 1997, Tanzi 1997) and may result in officials and private individuals obtaining a larger share of public benefits associated with certain policies, or bearing a lower share of the costs (Gupta, Davoodi and Alonso Terme, 1998). Essentially, corruption distorts the government’s ability to allocate resources both efficiently and equitably (since higher-income groups benefit mostly from it).

To date, most studies have focused on investigating the efficiency implications of corruption, notably its (now widely acknowledged) negative impact on investments (including foreign direct investments), growth and composition of public expenditure away from productive projects (Mauro, 1995 and 1997; Knack and Keefer, 1996). The distributional consequences have been rather overlooked (Gupta, Davoodi and Alonso Terme, 1998) although there are various channels through which corruption can affect income distribution, e.g. by (i) reducing overall growth rates, which in turn affects the poor disproportionately (Ravallion and Chen, 1997). (ii) increasing tax evasion and reducing the tax base (iii) badly targeting those social programs that benefit the most needy. (iv) inducing high concentration of asset ownership, which in turn may exacerbate the problem of access to capital markets by the poor and create poverty traps (v) impacting negatively on the distribution of human capital, e.g. as the share to fund education may be reduced or biased towards tertiary education.

Beside harming growth and fostering inequality, corruption also threatens fiscal and macroeconomic stability, reduces the impact of development assistance, and provides an incentive to exploit natural resources. It reduces the effectiveness of public administration and distorts public expenditure decisions, channelling urgently needed resources away from sectors such as health and education. It erodes the rule of law and harms the reputation of and trust in the state.[1]

We discussed above the possible links between globalisation, growth, and recent increases in income inequality; indeed, one channel through which market reforms (e.g. privatisation and trade liberalisation) foster growth might be institutional. More specifically, by reducing the prevalence of corruption in the post-reform era (provided that reforms are not just on paper). Indeed, well-designed and implemented reform programs, that truly deregulate the economy, remove trade barriers, foster genuine competition, and create the institutional, administrative and legal framework necessary to control bureaucratic and political discretion (e.g. through civil service reforms), have been shown to reduce corruption (Kaufmann, 1997). Examples include the abolishment of price or foreign exchange controls, subsidies or tax exemptions, secret budgetary accounts or industry-specific barriers to entry; the privatisation of state-owned enterprises; the simplification of the tax system. Beside having a direct impact on growth, lower corruption levels might have an indirect positive effect on growth precisely by generating positive distributional changes.

Why Income Tax Progressivity should not matter: The Traditional View

Taxation is the archetypal policy instrument for financing the costs of redistribution. In so far as it directly affects individuals’ incomes, or relative prices, it affects both individuals’ behaviour, and their welfare. Hence, traditional theories of taxation have analysed tax policy within the classic framework of an efficiency-equity trade-off.

The optimal taxation literature evaluates the effects of taxation through an application of the two basic theorems of welfare economics, yet with some key differences. Indeed, optimal income taxation models acknowledge the limitations of the second welfare theorem (i.e. the ineffectiveness of lump-sum taxes in raising some revenue and the informational constraint faced by the policy maker in observing people’s earning abilities or utility functions). Therefore the optimal income tax schedule derived by the hypothetical planner is a usually non-linear function of income (which is observable). However, by excluding lump-sum taxation, income redistribution becomes distortionary, i.e. it induces behavioural responses such as a lower labour supply, hence lower tax revenues and a greater deadweight loss. Such efficiency costs increase with the progressivity of the income tax system, since the latter reduces the wealthier individual’s marginal return from working proportionally more than the poor’s.

The standard way to illustrate the equity-efficiency trade-off uses the optimal income tax model developed by Mirrlees (1971). Based on his cornerstone results on the optimal marginal tax rate and a number of numerical calculations based on data for the UK, he concluded that: (i) the optimal tax structure should be approximately linear (i.e. a constant marginal tax rate) (ii) the few marginal tax rates should be rather low, always below 40 percent (iii) the income tax is a much less effective tool for reducing inequalities than it has often been thought. Others (see Newbery and Stern, 1987) have modified Mirrlees’ model and corroborated his conclusions: optimal tax rates increase with the revenue to be raised and with inequality aversion, yet only mildly. They decrease with the elasticity of substitution in the individual’s utility function.

The general results of the optimal tax literature have generated important fiscal policy prescriptions: since a highly progressive tax system is inefficient, tax systems should be as neutral and as non-distortionary as possible, with few and low marginal rates. This should imply higher output growth and economic development.

Indeed, many tax reforms in developing countries across the 1980s, e.g. Bolivia, have often reflected the spirit of optimal taxation theory. These reforms have tended to lower tax progressivity and shift away from direct to indirect taxation (Thirsk, 1997). The rationale given is that efficiency considerations are particularly important in developing countries. On the one hand, a reduction in corporate tax rates may provide a needed incentive for foreign direct investments. On the other hand, a highly complex income tax system with many marginal rates might be unsuitable in countries where the initial tax base is small (e.g. due to a large informal economy), the administrative capacity weak, and the incentives for tax evasion high (Tanzi and Zee 2000; Stern and Burgess, 2000). Wherever such institutional or administrative constraints exist, a popular policy recommendation has therefore been to redistribute through a well-targeted public expenditure system (e.g. financing health and education) rather than through the ample use of complex systems of progressive taxation (Harberger, 1998).

Why Public Expenditures Matter Instead

Government expenditures are the other main redistributive tool available to the government. Typically redistributive expenditures comprise items such as education, medical care, housing, as well as social insurance transfers such as unemployment subsidies. The distributive outcome of such benefits can be thought of as the degree of expenditure “progressivity”.

Yet, measuring the progressive incidence of public expenditures is not straightforward. Usually this is done by looking at the income effects of such benefits for different individuals in the economy; or by examining whether poorer income groups benefit from various subsidies or public services in a higher percentage of their income. Usually public spending is considered to be well-targeted if the poorest quintile benefits proportionately more than the richest. Sometime regional patterns of expenditures are also considered.

In theory, the optimal allocation of public funds is also subject to the familiar equity-efficiency trade off. Some argue (Birdsall and James, 1990) that public expenditures’ are actually more efficient when they are more equitable, i.e. more progressive, especially in the developing world, e.g. if they can contribute to reduce high mortality rates among the very poor.

However, data show a clear trend of departure from optimal prescriptions: in highly unequal society, e.g. in Latin America, there is often an inequitable bias of public funding towards the well-off, e.g. university education is highly subsidised while its beneficiaries are disproportionately from the richer income groups. In Brazil, the bottom income quintile receive 7% of total social benefits, while the top quintile receives 41% (World Bank 1989). De Mello and Tiongson (2003) run cross country regressions to check the effects of ex-ante inequality on both government transfers and social security/welfare spending. They confirm that countries with higher ex-ante inequality do redistribute significantly less income through their public policies, and find this relationship to be non-linear.

Incidentally, such finding could be further explained again by the role of corruption, which is likely to be more pervasive in more unequal societies. Indeed, it has been argued that, when corruption is present, merely allocating public resources for “right” goods and services – from the equity point of view – may actually not lead to the desired social or economic outcomes e.g. lower child mortality (Rajkumar and Swaroop, 2002).

Supporting Evidence

Fiscal reforms, particularly in developing countries, have also provided the opportunity to test the results proposed by optimal taxation theory, and more generally, to measure both the individual and joint redistributive effect of progressive taxation and expenditure.

A word of caution with many empirical studies of this sort is however due. Studies measuring the distributional implications of taxes and government spending suffer from several difficulties. For example, very different results can be reached depending on whether inequality measures used are based on consumption, net or gross income, or on whether the unit of measurement is the houselhold, or the individual. Some other limitations of these studies is the likely lack of data on both before and after-tax income for the same country, making it difficult to separate the effects of redistribution on inequality changes from other non-redistributive causes of variation.

The most commonly used source of inequality data is the Deininger and Squire (1996) cross-country dataset for 1970-1990. From this dataset it appears that, on average, developing (and transition) countries tend to have less before-tax income inequality, e.g. for the 1990s, their average before-tax Gini coefficient is 38 percent, against 44 percent in developed countries. Yet, the difference between average before-tax and post-tax Gini seems to be larger in industrial countries, suggesting that in developing countries redistribution might be relatively less effective (Chu Davoodi Gupta, 2000). Although no rigorous conclusion should be drawn from these statistics, they suggest that the effectiveness of redistribution may depend not only on tax structure and progressivity, but also on institutional factors, e.g. the ability to collect taxes or punish evasion, the quality of which varies substantially between developed and developing countries.

A number of tax incidence and expenditure benefits studies for individual developing countries (Bird and De Wolf, 1973; Shah and Whalley, 1991) have endorsed the rationale for tax reforms aiming at reducing tax progressivity on equity as well as efficiency grounds. They often show that a highly progressive tax system is not always associated to a more equal post-tax income distribution (i.e a lower Gini), while expenditure “progressivity” normally is.

Chu Davoodi and Gupta (2000) estimate Gini equations for around 35 countries over the 1970-80s and 1980-1990s decades. They assume various determinants of income distribution such as the ratio of direct to indirect taxes, the ratio of direct taxes to GDP, secondary school enrollment rate (percent) and a number of country characteristics (e.g. dummies for whether the data are for a transition country, or an urbanisation measure to capture migration effects). Their econometric specification is aimed at testing the distributional implications of the nature of tax regime chosen (direct vs. indirect) and of investment in human capital. The null hypothesis is that an increase in both the direct/indirect tax ratio, and in the enrollment rate, lower inequality (i.e. the estimated coefficients should be negative). Thus, one important underlying assumption they make is that direct taxation is inherently more progressive.

Their observations include indistinctively Gini coefficients based on post-tax income, before-tax income or consumption (which proxies for disposable income). They run OLS regressions on a sample of both developing and transition countries, and on one of developing countries only. They conclude that: (i) although always negative and significant, the coefficient of the tax composition ratio is very small (around -0.02) (ii) the ratio of direct taxes to GDP has a very small yet positive effect on Gini (0.0008), i.e. more direct taxation appears to actually have a regressive effect, although insignificant when transition countries are included (iii) the effect of increasing secondary school enrollment rate has a large negative effect, as expected, yet significant only when transition countries are excluded. The authors do claim that these results do broadly suggest a progressive effect of direct taxation on inequality, yet the active use of tax policy for redistribution should be discouraged, given the small magnitude of such effects.

Arguably, the above study cannot really capture the effect of tax “progressivity” on inequality. In fact, as result (ii) above suggests, a tax system which collects more revenues from direct taxes is not necessarily more "progressive".

These results notwithstanding, one reason why progressive income taxation might show no or little effect on some developing countries’ income distribution might be that usually either foreign investors or state-owned enterprises bear the biggest burden of capital taxes. If so, capital tax progressivity would have not much incidence on domestic income distribution. Finally, a progressive tax system may be effectively regressive if the amount of tax collected from the top income brackets is very low, as a World Bank study in the Philippines shows (Devarajanj and Hossain, 1995). Conversely, an apparently more regressive income tax system may end up having a positive effect on income distribution as it might lower the incentives for tax evasion, thus generating higher revenues, potentially available for transfers by the government (Shah and Whalley, 1991).

To conclude, what some of these results would seem to suggest is that (i) tax progressivity is not a sufficient condition for reducing inequality (ii) well-targeted public expenditures are a necessary condition (iii) net incidence is important when examining the progressive or regressive bias of a given system. We now want to test the theoretical and empirical robustness of this relationship.

Why Tax Progressivity should matter: a New View

A starting point for reconsidering the evidence discussed is provided by the so called combined fiscal incidence studies, mostly done for developing countries (Devarajan and Hossain, 1995). They recognise that the central issue in determining the impact of fiscal policy on inequality is its net incidence: even if a tax is nominally regressive, its overall income effect may not be, if the revenue raised is spent in a progressive manner (e.g. Jamaica in 1980-1990s). Conversely, a well-targeted public spending program may not have pro-poor net effects if it is financed by a highly regressive tax. Furthermore, a nominally progressive expenditure system may not have an effective impact if it is not well-targeted. Indeed, Chu Davoodi and Gupta (2000) survey 55 expenditure benefits analysis and find significant regional differences in the quality of targeting.

What emerges from these combined fiscal incidence approach is that taxation and public spending might actually be complementary instruments for redistribution. We review a model which uses the optimal taxation theory framework to argue this case.

Implicit in the argument that equity objectives should be addressed by the expenditure side is the notion that the required costs should be financed by broadly neutral taxes. Zee (1999) shows that actually progressive rather than neutral taxation is optimal when revenues are used to finance perfectly targeted transfers to the poor, and the optimal degree of progressivity increases with the degree of initial inequality. This result holds under different assumptions on the efficiency costs of taxation and society's aversion to inequality. The model supports the notion of complementarity between the two fiscal policy instruments. In fact, the size of transfers, determining the revenue constraint, and the average income tax are simultaneous solutions to the government's optimization problem.

The model is defined over a two-class economy, with an individual A whose pre-tax income yA is greater than individual B's pre-tax income yB. Aggregate pre-tax income is 2z, distributed according to a relative inequality parameter (, with ( = 0.5 when pre-tax incomes of A and B are shared perfectly equally. Hence pre-tax incomes are:

yA= (. 2z and yB = (1-() ·2z

The model assumes a non-linear average tax function:

(1) a = ( - y-( with ( ( (, ln y ( 1, ( ( ( ( (

and a tax revenue function:

(2) R = a·y = (·y - y1-(

From which a marginal tax function is derived:

(3) m = (R/(y = ( - (1- ((·y-( with m > a for ( > 0,

and the marginal rate of progressivity:

(4) MRP = (m/(y = β(1- β)·y-(1+β) > 0 for ( < ( < (

The marginal rate of progressivity however is not monotonic in β. By differentiating MRP with respect to β it can in fact be shown that actually (MRP/( β < 0 for β > 0.5, thus the relationship is non-linear.

The revenue function is progressive for ( > 0, proportional for ( = 0 (i.e. as ( →0, R→(y). Similarly the average and marginal tax functions become increasingly proportional as ( →0. The model wants to derive the optimal values (* and (* (with ( interpreted as the proportionality parameter and ( as the progressivity parameter), from maximising an Atkinson social welfare function, depending on post-tax incomes:

(5) Max W = - exp(-(·qA)/ ( - exp((-(·qB)/ ( for ( > 0 (with ( = inequality aversion)

{α,β}

or

(6) Max W = qA + qB for ( = 0 (i.e. no inequality aversion)

{α,β}

where post-tax incomes are:

qA = yA - rA with rA = α·yA - yA1-( (revenues paid by A)

qB = yB - rB + g - c with rB = α·yB - yB1-( (revenues paid by B)

g = rA + rB total revenues

qB = yB + rA – c

The assumption of a perfectly targeted transfer system is captured by the idea that the poorer individual B receives the entire revenues g net of some efficiency costs c(g) of taxation. We can see that A's post-tax income decreases with an increase in either ( or (, while B's post-tax income increases. Also, in the model efficiency costs of taxation c((,() increase with ( and (.

Substituting (. 2z and (1-()2z for yA and yB in equations qA and qB, and in rA, and rB , respectively, two FOC are then derived from (5). Intuitively, the FOC state that (i) the ratio of the effect of a marginal change in ( to that of ( on A's post-tax income (i.e. ((qA/(()/((qA/(()) must be equalised to that some ratio for B’s post-tax income. (ii) the ratio of the effects of a marginal change in ( on both individuals' post-tax incomes (i.e. ((qA/(()/((qB/(()) must be set equal to A's excess income (qA - qB) weighted by (.

The optimal solution for (* turns out to be independent of (, or (, and only determined by the ratio of pre-tax incomes, both in absolute and in relative terms (i.e. ( / (1-()). This suggests that the optimal value of progressivity (* is not governed by equity considerations (i.e. () but only by considerations about how to raise a given amount g most effectively. It is actually the proportionality parameter (* to determine the optimal transfer g, as it depends on (. Furthermore, (* is not well defined for ( = 0.5 (i.e. when pre-tax incomes are equal, (* ( 0), while (* ( 1 as ex-ante inequality ( tends towards its maximum value. At the same time, for a given inequality level (, (* ( 0 as aggregate pre-tax income z ( (. Numerical calculations done by the authors confirm these results. In other words, at an optimum the tax system tends towards more proportionality as the aggregate pre-tax income increases (for a given degree of inequality); it tends towards more progressivity as the degree of relative inequality increases (for a given pre-tax income level).

Based on these optimal values, the calculation of the optimal average and marginal tax rates for A and B is straightforward, for different combinations of ( and z. From these calculations it appears that, unless φ = 0.5, A's average and marginal tax rates at an optimum are always greater than B's, i.e. the optimal tax system is progressive. This difference rises with higher ( and falls with higher z.

We can draw the following interesting policy recommendations from this model: (i) inequality considerations should indeed be reflected in the choice of an optimal transfer level, i.e. addressed by the public expenditures side. (ii) however, tax progressivity becomes more desirable as inequality and inequality aversion rise, since more revenues will be needed to finance the desired transfers. Under a progressive tax system therefore, imposing a higher tax burden on the rich, such financing requirements can be reached with less efficiency cost to society. Yet, as the income of both poor and rich increase, the optimal tax system becomes more proportional as it is able to raise the same amount of revenues for any given inequality level.

In conclusion, in designing equity-driven expenditure programs, the potential useful role of progressive taxation for efficiently raising revenue should not be ignored, particularly by poorer and highly unequal countries.

Why Corruption Matter for Inequality

Many tax reforms in developing countries have been done to account for some of these countries' institutional limitations, including the inability to rise direct taxes (e.g. where the informal economy is large) or to bear the costs of enforcing a tax regime in presence of informational constraints. These costs might include, auditing, or paying commissions to tax inspectors to induce the truthful reporting of taxable incomes, hence discourage them from accepting bribes.

Much political economy literature argues that the pursuit of social welfare and optimal taxation may be constrained by the bureaucracy’s own goals (Atkinson and Stiglitz, 1980). Indeed, corruption can affect the impact of a given fiscal policy regime, as it distorts the government's role in resource allocation in favour of higher income groups, mainly by fostering tax evasion, illicit exemptions, or siphoning of public funds. For instance, a survey in Uganda showed that due to public sector corruption only 13% of resources destined to funding primary school education would actually go to the intended cause; and however it is amply demonstrated that funding higher education has a strong positive effect on reducing the income gap.

Traditional models of tax evasion tend to agree that as income rises, the incentives to evade taxes increase. Hence tax evasion is usually seen to have a regressive bias. The effects of income tax progressivity on the likelihood of evading (hence on the extent of corruption) are however more ambiguous, both in theory and in the data (Andreoni Erard and Feinstein, 1998).

Hindriks, Keen and Muthoo (1998) take the view that a more progressive tax system does lead to more evasion by the wealthier, thus becoming effectively regressive. They model a three stage game of tax extortion (by the tax inspector) and bribery (by the tax payer). Regardless of the penalty to be paid if audited, the rich always gains from evading, hence will collude with the tax authority; while the poor, who faces a lower marginal tax liability, makes no gains from evading. Consequently, the latter will not bribe, but then will be more likely subject to extortion by the authorities. Additionally, as tax evasion reduces government revenues, the poor bear disproportionately more the costs of corruption, as lower revenues entail lower transfers. Therefore, a more progressive tax system is more costly to administer, since, to be effective, it must reduce evasion, i.e. tax inspectors will need to be paid a higher commission in order to offset their incentive to underreport taxable incomes.

The main implication of this model is that a more progressive tax system should be accompanied by higher levels of corruption, leading to higher tax evasion hence to a negative redistributional outcome. This suggests that the interplay of corruption and tax policy might generate some interesting net effects on inequality which are not obvious when the two phenomena are analysed separately. Later we will explore how this idea holds in practice.

In a number of existing studies, corruption is shown to be positively related to higher income inequality. Corruption is found to have a larger impact on direct taxes, hence it is said to reduce the effective progressivity of a given tax system (Tanzi and Davoodi, 2000). Corruption also distorts the composition of government expenditures away from education (Mauro, 1998).

Gupta, Davoodi and Alonso-Terme (1998) attempt to verify empirically the hypothesis that corruption increases income inequality by reducing the progressivity of the tax system, the effectiveness of social spending and of human capital investment. They introduce dummies to account for differences in Gini coefficients based on either before-tax or after-tax income. They run OLS cross-country regressions for 1980-97, and find that a worsening by one standard deviation in their corruption index (2.52 in a scale from 0 to 10) increases the Gini coefficient by 5.4 points, and reduces average secondary schooling by 2.3 years. Furthermore, they show that a one standard deviation in the growth rate of corruption, reduces income growth of the poor by 7.8 percent. These results are robust when controlling for other determinants of inequality such as land distribution, abundance of natural resources, capital productivity, or education inequality. Their hypothesis is again that corruption affects inequality by reducing tax progressivity. They test it by interacting the net income dummy (equal to 1 if Gini is post-tax) with the corruption index. Since this intereaction variable is positive and significant (0.66), i.e. the impact of corruption on inequality is higher when using the post-tax Gini measure, they conclude that corruption increases inequality by making taxes less progressive. Finally, they find that corruption lowers social spending, but that, surprisingly, social spending has no significant effect on income inequality

Hypotheses

We now present a new empirical model to investigate the joint effects of progressive taxation, progressive public expenditure, tax composition, and corruption, on income inequality. The main novelties of this model, compared to the literature reviewed, are that (i) it attempts to capture the effects of tax progressivity on inequality directly, rather than, e.g. from variations in tax composition (see Chu Davoodi and Gupta). To this end, the model addresses some difficulties in measuring tax progressivity which, to the best of our knowledge, have so far never been tackled. (ii) It proposes an alternative way to verify why corruption would worsen income inequality (see Gupta, Davoodi and Alonso-Terme), by allowing to identify more explicitely the effects of corruption on income tax progressivity.

Generally, the following hypothesis will be tested:

1. Both income tax and expenditure progressivity are necessary conditions for lowering income inequality.

2. The distributional effectiveness of public expenditure depends positively on the degree of income tax progressivity, as suggested by Zee (1999).

3. Tax composition (direct versus indirect taxation) matters for inequality, and direct taxation is more effective than indirect taxation to reduce it, as concluded by Chu, Davoodi and Gupta (2000).

4. The relationship between income tax progressivity and income inequality is non-linear, hinting to the possibility that there is a critical level above which progressivity increases income inequality, as suggested by Zee (1999). This non-linearity could be explained by an increase in tax evasion when income tax progressivity raises above the critical level. This implies that, broadly, corruption has a distributional impact.

5. Corruption affects income inequality by reducing the distributional effectiveness of a given degree of tax progressivity (i.e. the revenue it raises for a given level of income), as concluded by Gupta, Davoodi and Alonso-Terme (1998).

The Data

Our dataset comprises information for 47 countries (excluding transition countries) over the period 1980 – 1997. It includes 63 variables in total, of which 9 have been imported from existing datasets. The remaining variables have been calculated mainly using IMF Government Finance Statistics, and a dataset available from the IMF Fiscal Affairs Department . Data collection and preparation have occurred over a three months period prior to the writing of this paper.

The following statistics for each country were selected:

Average income distribution for each country, over the period 1980-1997, measured by the average Gini coefficient, based on either before or after tax per capita income. The source of data is Deininger and Squire (1996). This data-base identifies "high-quality" income distribution data, i.e. drawn from national surveys, and based on various measurements of income and consumption. Yet these high quality data are not unproblematic, e.g. Gini coefficients based on income measurements tend to systematically overstate inequality compared to those based on consumption. Also, comparability of cross-country inequality data is undermined by the use of differing methodologies. Furthermore, average Gini measures for each country are unweighted, although the number of data points available for each country over the relevant period (1980-1997) varies considerably (e.g. some countries have only one data point available).

In the regressions, we experiment using both (i) a sample restricted to "high quality" income-based Gini coefficients, using a net income dummy equal to 1 for those data based on post-tax incomes (ii) for the same countries, a sample of mixed-quality data, which however includes only post-tax income based Gini coefficients.

A measure of income tax progressivity for each country, evaluated at the year for which the first Gini observation (preferably post-tax) is available (within the 1980-1997 range). This is done to avoid biased estimates due to tax endogeneity: before-tax inequality for a given year could in fact determine that same year's degree of tax progressivity. However, the "cost" of our strategy is that our regressions will be able to pick up only an indirect effect of the income tax system on future average gross income distribution, rather than its direct redistributive effect (particularly when using the “high quality” Gini measures). Also, the effects of subsequent income tax reforms on the average Gini measure will not be detected.

No available cross-country progressivity measure was found in the existing literature. We therefore proceeded to build a number of different indexes, which roughly tried to capture the degree of progression, by comparing the average tax rate for two income groups: those with a pre-tax income three times above the average per capita income; and those with a pre-tax income equal to half the average per capita income. This selection of incomes was random.

First, we collected data on average GDP per capita in local currency, relative to the year for which the first Gini observation was available (UN Statistical Yearbook 1997). Then, we calculated the local currency value corresponding to three times and half times the average GDP per capita, respectively. The average tax rate corresponding to these two incomes was calculated by applying that same year's marginal tax rates to each relevant sub-bracket. Marginal rates and corresponding income brackets (in local currency) were obtained from a IMF cross-country data-set for 1980-97.

We use the following notation:

x= GDP p.c.

t1= marginal tax rate applicable to half the average GDP p.c.

t2 = marginal tax rate applicable to the average GDP p.c.

t3 = marginal tax rate applicable to three times the average GDP p.c.

z = tax exempted income (zero bracket)

We build the following indexes to measure tax progressivity:

TAXPROG = ([t3(3x- x)+ t2(x – 0.5x) + t1(0.5x - z)] / 3x( / ([t1(0.5x - z)] / 0.5x( (1)

TAXPROG1 = ([t3(3x- x)+ t2(x – 0.5x) + t1(0.5x)] / 3x( / ([t1(0.5x)] / 0.5x( (2)

The indexes thus calculated should capture the relative average tax rate for the person earning three times above average against the person earning half the average income, in each country, in the relevant year. The greater the index value, the more progressive the income tax system is considered to be (see Annex I for ranges).

Tax system composition, measured by the ratio of direct to indirect taxation, for the year in which the tax progressivity index is calculated. This ratio captures how the tax system raises the majority of its revenues, i.e. a bigger ratio shows an inclination to use proportionately more income taxes to raise a given amount of revenues.

The degree of expenditure progressivity, proxied by two distinct variables: (i) recurrent expenditure on primary education per pupil, as a percentage of per capita income, evaluated in 1985. The source for these data is Barro and Lee (1996). Across the literature, primary education is in fact considered to have a very progressive incidence. (ii) the share of social expenditures, defined as the sum of total expenditures on health, education and welfare, as a percentage of total GDP, evaluated in the year for which the first Gini observation is available (from IMF Government Finance Statistics).

Among the non fiscal determinants of inequality, we include:

Human capital endowment, proxied by the average percentage of population above 15 with completed secondary education (Barro and Lee, 1996)

Inequality in human capital, proxied by the 1980-95 average ratio of the percent of population, aged 15 and over, with no schooling expressed as a fraction of percent of the population, aged 15 and over, with completed secondary and higher education (also from Barro and Lee, (1996)). Lack of access to education is in fact acknowledged as a major determinant of persisting inequality through time.

Inequality in land distribution, proxied by a coefficient measuring the percentage of arable land available per person in 1980 (from FAO Agricultural Production Yearbook 1985).

A measure of corruption, proxied by a Corruption Index used by Gupta, Davoodi and Alonso-Terme (1998). This Index combines two sources frequently used in the literature: the International Country Risk Guide and Business International. These sources reflect foreign investors’ subjective assessment of the degree of corruption in a country, e.g. whether bribing is necessary to obtain a business licence. The combined index ranges from 0 (most corrupt) to 10 (least corrupt).

Summary Statistics. The average Gini coefficient measure for our “high quality” sample is 40.3. This includes both before-tax and post-tax Ginis. The average degree of tax progressivity measures 2.9, both using TAXPROG and TAXPROG1, i.e. the person with three times average per capita income faces an average tax rate 2.9 times bigger than the person with half the average income. The average corruption measure is 5.4.

If we divide the sample between countries whose average per capita income is below $5,000 a year (27 observations) and above $5,000 , we find that average Gini coefficient in the poorer sample is higher (45.47), pointing to higher average inequality, while the mean value for both tax progressivity indexes is lower (e.g. for TAXPROG1, the mean is 1.64 against 2.74 for the richer group). Also, we find that corruption is on average much worse in the former group (3.86 against 7.33). Finally, poorer countries also score worse on our primary education expenditure proxy, with 8.83% of per capita income going to finance this expenditure against a much higher average (19.73) in the richer income group.

Correlation coefficients between our regression variables are generally low. In particular, the correlation coefficient for tax progressivity (TAXPROG) and expenditure progressivity (PREDU) is only 0.10. Indeed there is some positive correlation between tax progressivity and corruption, possibly supporting the theory of Mutoo, Hindricks and Keen, but still it is rather low (0.36).

The Empirical Model

The basic model specification, estimated by OLS, is:

(1) Yi = ( + (1Ti + (2Ei + (3Ci + (4Zi + ui

Where Y stands for either gross or net income average Gini (over 1980-1997), T is the measure of tax progressivity, E is the measure of expenditure progressivity, C measures tax composition, Z is a vector of other control variables, and the subscript i indicates the country. This specification is only able to capture some indirect effects of the progressivity of fiscal policy when the dependent variable Y encompasses Gini indexes on both gross and net income (high quality data). Hypothesis (1) and (3) above will not be rejected if β1- β3 are all negative and significant (a lower Gini implies a more equal distribution).

We then try to test a non-linear specification of tax progressivity and inequality which, to our knowledge, has not yet been tested in the literature. In the presence of non-linearities, equation (1) becomes:

(2) Yi = ( + (1Ti + (2Ti2 + (3Ei + (4Ci + (5Zi + ui

Hypothesis (4) will not be rejected if β1< 0 and β2 > 0.

We also split our sample and run regression (1) and (2) separately for countries with GDP per capita above and below $5,000 a year, to capture whether the effects of a given fiscal policy measure might be different in poorer countries, e.g. because of institutional constraints.

To test hypothesis (2), we also interact tax progressivity with expenditure progressivity so as to capture their joint effect on income inequality:

(3) Yi = ( + (1Ti + (2Ti2 + (5 (E*T)i + ui

Hypothesis (2) will not be rejected if β5 < 0.

We then try to isolate the direct effects of redistribution on post-tax income Gini coefficients by first introducing a net income dummy:

(4) Yi = ( + (GINIPOST + (1Ti + (2Ti2 + (3Ei + (4Ci + (5Zi + ui

where GINIPOST is a dummy equal to 1 for net income Ginis. We also modify (4) by adding slope dummies for every independent variable in the equation. We subsequently try to split the sample to include only observations for GINIPOST = 1, but this seriously limits the number of observations. Therefore we repeat the experiment by using a sample of "lower quality" Gini coefficients which however includes only post-tax income measures. Finally, we run (4) after splitting again the sample among those countries with average per capita income below and above $5,000 a year.

To test our hypothesis (5), we then introduce the corruption index (CORR) into our base-line regression:

(5) Yi = ( + (1Ti + (2Ei + (3Ci + (4Zi + (5CORR + ui

We test the hypothesis that β5 < 0. This tells us whether more corruption (i.e CORR → 0) has a direct negative effect on income inequality (i.e. it rises Gini), which is suggested by the high negative correlation between the two (-0.54). We also regress specifications (2), (3) and (4) adding the corruption index, in the case of (4) to check whether the effect of corruption on average net income Gini is significantly different. If so, this might suggest that corruption affects income distribution through its influence on the implementation of tax policy. To test this hypothesis more precisely, we interact corruption with our measure(s) of tax and expenditure progressivity, Ti and Ei:

6) Yi = ( + (GINIPOSTi + (1Ti + (2Ti2 + (3Ei + (4Ci + (5Zi + ( CORR*Ti + ui

7) Yi = ( + (GINIPOSTi + (1Ti + (2Ti2 + (3Ei + (4Ci + (5Zi + ( CORR*Ei + ui

If corruption worsens income inequality by making a given tax or expenditure system effectively less progressive, we should see γ > 0 and μ > 0. However the interpretation of our result is more meaningful if our Yi only included data for net income Ginis.

Results

Using OLS on our baseline specification (1), but without any control Zi, only the coefficient of the proxy used to measure expenditure progressivity Ei (PREDU) is negative and significant: increasing primary education expenditure by one standard deviation would reduce the value of the Gini index by -10.4 points. This result is confirmed when we control for education inequality or land distribution, although less significantly. When we replace PREDU with the social expenditure proxy however, results are not significant. The effects of either tax progressivity index or tax composition are generally of the right signs but insignificant.

Generally, specification (1) yields low R-squares, and its results may be overestimated, due to omitted variables. Yet they do broadly confirm the “traditional” view of redistribution offered by the Optimal Taxation literature.

Things start to change when we regress non-linear equation (2). We find that all our fiscal policy coefficients beside tax composition are significant and of the expected sign, thus confirming hypothesis (1), which expresses the views expressed in Section 5. Furthermore, the coefficient of the quadratic term is positive, thus confirming also hypothesis (3), and suggesting, on the lines of Zee (1999), that there is a bound above which tax progressivity is actually bad for income equality:

δYi/δTi = β1 + 2 β2Ti > 0 for β1 > - 2β2Ti (i.e. condition for tax progressivity in country i to worsen inequality)

Our index TAXPROG1 yields a much larger negative effect than TAXPROG (although both are significant): a one percent increase in TAXPROG1, would reduce inequality by around 9 points, against 1 point estimated by using TAXPROG. The coefficient for PREDU has a smaller effect, as the same one percent increase in this variable would produce a Gini reduction of only 0.5 points.

When controlling for various other potential determinants of inequality, the coefficient of PREDU looses significance while that of tax composition (DIRTAXINDIRTAX) gains it, although it has an unexpected positive sign, suggesting that more direct taxation might lead to a worse distribution. This might be plausible if the income tax collection system is weak and the tax base small. Therefore, hypothesis (4) should be rejected. Among the other control variables that appear to have significant negative coefficients there is education inequality (EDUINEQUAL), although the latter's significance seems to rely on the presence of including GDP per capita in our regression, which is problematic due to its high correlation with most of the independent variables (e.g. 0.78 with PREDU).

Interestingly, once we split the sample and regress (2) separately by income groups (below and above per capita income of $ 5,000 a year), all coefficients, beside the quadratic term, remain highly significant only for the poorer sub-sample (although its size is small). Therefore, the country's level of economic development seem to influence the effects of fiscal policy. This is relevant to our model in so far as economic and institutional development are highly correlated.

When we regress equation (3), we see that the interaction term (E*T) is negative, as expected, and largely significant. This should support our hypothesis (2), as the effect of tax progressivity on lowering Gini is greater (by a factor of 0.2) whenever it is considered jointly with the effects of expenditure progressivity. This indirectly confirms those studies arguing for the importance of net fiscal incidence.

Regressing eq. (4) to check for direct effects on post-tax income inequality, we find our dummy variable to yield a significant negative coefficient, as expected, indicating that average post-tax gini tends to be lower by approximately 8 points, all other things being equal. This is slightly more than what Chu Davoodi and Alonso-Terme (1998) find (-6.91), yet confirms their overall result. When we build slope dummies for each one of our fiscal policy variables, both tax progressivity and expenditures on primary education have an even greater significant effect on post-tax inequality; however, the coefficient of the intercept GINIPOST looses significance. Using a sample of only "lower quality" net income Gini data does not substantially change this result.

We finally introduce corruption into our model, first to see whether it has a direct effect on income inequality. Estimation of eq. (5), with different controls, results in only corruption and tax progressivity having a significant negative effect on income inequality, with a one standard deviation in the corruption index resulting in a reduction of Gini of 9.8 points. The sign and significance of these results are confirmed when: (i) we add the square of the tax progressivity index into (5). In this case an increase in corruption results in a reduction of Gini of 12.6 points /*to Do: check; this is high compared to Gupta Davoodi paper (ii) we interact the dummy GINIPOST with the corruption index, showing that the significant negative effect of corruption on inequality is even greater when the latter is measured on post-tax income. Furthermore, also the effect of expenditure progressivity becomes significant. These results are confirmed when we use the "lower quality" post-tax Gini sample. Also the introduction of corruption raises our R-square and adjusted R-square considerably (see Annex II).

This provides some support for hypothesis (5), i.e. that the effect of corruption on income inequality occurs through the fiscal policy channel. We test this by regressing (6) and (7), which present interaction terms between corruption and tax and expenditure progressivity, respectively. Indeed, our hypothesis is validated: the interaction term is positive and highly significant, showing that in presence of corruption the coefficient of tax progressivity is reduced by a factor of 1.59. The result is confirmed when we regress the interaction terms on the lower quality data. The R-square statistic for these regressions is very high (e.g. 0.87 for (6)).

Summary of Results: Hypothesis 1, 2, 3, and 5 (see page 22) cannot be rejected. Our results strongly show that, subject to a non-linear relationship between tax progressivity and income inequality, both tax and expenditure systems progressivity (as measured by our tax indexes and public spending on primary education per capita) are necessary to reduce income inequality (measured by the Gini coefficient). In particular, the effectiveness of public spending as a redistributive instrument is positively correlated with adopting progressive taxation. However, there is no evidence that direct taxation be more adequate than indirect taxation from the point of view of redistribution. Quite to the contrary, the data suggest that direct taxation is bad for inequality, possibly since it may lead to more tax evasion if other institutional characteristics allows it. Tax evasion is a likely symptom of corruption, an institutional weakness which our results clearly suggest may be driving, alone, cross-country differences in inequality. Furthermore, our regressions allow us to strongly conclude that the likely mechanism through which corruption affects inequality in a given country is fiscal: a given degree of tax progressivity looses its redistributive effectiveness considerably once corruption is included, i.e. the revenues raised from taxing a given income will be lower.

All our regressions have been double-checked using the Huber-White sandwich estimator for standard errors, mainly to deal with potential problems of heteroscedasticity.

Limitations

Measurement Error: We should stress that this problem affects similarly our results as much as those of previous studies reviewed. It involves:

(i) the dependent variable Yi. As noted by Deininger and Squire, the heterogeneity of Gini data and their collection methodologies hides serious measurement shortcomings. Ideally, we would want to have sufficiently large panel data for each country, both for gross and for net income Ginis, so as to avoid using average measures which disregard differences in pre- and post-tax income. Nevertheless, these measurements error should not bias our estimates.

(ii) the independent variables, particularly the tax progressivity indexes Ti. For instance, in our model each relevant average tax rate is calculated on the basis of only three representative brackets (namely (3x-x), x, and 0.5x) while in reality there may be many more brackets above and below these which should be included to achieve a better index calculation. Measurement errors in the independent variables might bias our estimators. Additionally, the greatest shortcoming of our indexes is the lack of information about the tax liability distribution of the population. Our assumption about a uniform distribution is seriously restrictive, while we would need to weight each one of our representative tax rates, reflecting how many people are actually liable to pay them.

Omitted Variables: although generally our regressions yield high R-squares, it is possible that they omit important variables, biasing our coefficients and invalidating the test statistics. An important omitted factor in the model might be the recent increase in globalisation of many developing or transition economies (Slemrod and Bakija, 2000). Beside changing relative wages, increased openness may in fact also increase the “cost” of progressivity: as capital is more mobile, higher tax progressivity could then imply lower revenues through capital flight, hence worse inequality. This could bias our progressivity estimates downwards. Also, the presence of credit constraints might have a negative effect on current and future inequality (i.e. poverty traps). This effect could be picked up by our progressivity index, also resulting in a downward bias.

Overall, this model is only a preliminary exercise, and as such, it suffers from serious limitations, ranging from the small number of observations in our sample, to the need to reduce measurement errors. However, the results obtained seem sufficiently promising to motivate future improvements to the model. For instance, beside including a measure of trade openness, and one of credit constraints, we could control for the size of total government expenditures, relative to the size of the economy, to see whether this has a significantly different effect on inequality outcomes, given a degree of tax progressivity. Finally, we could control for the structure of the economy (e.g. size of the agricultural sector or informal economy against industrial or service sector), to see whether this also influences inequality, given a likely correlation between structure of the economy and size of the tax base.

Conclusion

Against the backdrop of recent trends in income inequality, we hope to have satisfactorily shown that theories pointing to globalisation and increased factor mobility, technological progress and changes in the relative wage of skilled and unskilled workers, are not sufficient to explain such trends. Any assertion about changes in income distribution must necessarily take into account changes to the progressivity of the tax (and expenditure) system, and how these changes may in time affect not only net but also gross income distributions. For instance, recent tax reforms adopted by many countries, reducing tax progressivity for the highest income groups, might have induced behavioural responses, ranging from an increase in the labor supplied by these groups, hence an increase in their pre-tax incomes, to a greater willingness to report higher taxable incomes. More generally, changes to the progressivity structure of a fiscal system could unleash behavioural responses particularly among wealthier individuals, which could eventually result in a more unequal distribution of reported gross incomes. Indeed, many have found such effects sizeable (Slemrod and Bakija, 2000).

Despite some data set limitations, our model brings some new evidence challenging the classic notion of an inescapable trade off between equity and efficiency in taxation. The recurrent conclusion of optimal taxation theories is in fact that changes to tax progressivity have a small benefit in terms of equity and income distribution, with a large efficiency cost.

Instead our model has hopefully suggested that the role of tax progressivity in redressing income inequality cannot be easily dismissed. We want to suggest that a progressive tax system is a necessary condition for lowering income inequality hence for promoting long-term growth. However the quadratic relationship tested in the model should also warn us that this is only so up to a certain critical level.

We also want to suggest that the presence of institutional failures can seriously undermine the revenue-raising capacity of a given redistributive policy, and therefore their effect should be accounted for when modelling an optimal tax schedule. From the data available, our model suggests that indeed corruption reduces the effectiveness of a given degree of tax progressivity, possibly because of a higher likelihood of tax evasion.

The main policy suggestion emerging from our model is that a country which manages to tackle corruption might actually be able to also reduce inequality in its income distribution without changing its tax policy, i.e. without having to make it more progressive. The appeal of this conclusion is that addressing corruption, beside being worthwhile per se, might entail an additional gain: it might substantially reduce the classic equity-efficiency trade-off in taxation.

References

Alesina, A. and Rodrik, D. (1994). Distributive Politics and Economic Growth, Quarterly Journal of Economics, Vol. 109, 465-490.

Andreoni, J., Erard,B. and Feinstein, J. (1998). Tax Compliance. Journal of Economic Literature, Vol. 36, 818-860.

Atkinson, A. (1997). Bringing Income Distribution in from the Cold. Economic Journal, Vol. 107, 297-321.

Atkinson, A. and Stiglitz, J. (1980). Lectures in Public Economics. McGraw-Hill.

Auerbach,A. and Hines,J. (2001). Taxation and Economic Efficiency. NBER Working Paper 8181.

Banerjee, A. and Newman, A. (1993). Occupational Choice and the Process of Development. Journal of Political Economy, Vol. 101, 2, 274-298.

Barr,N. (1992). Economic Theory and the Welfare State. Journal of Economic Literature, Vol. 30.

Barro, R. and Lee, J. (1996). International Measures of Schooling Years and Schooling Quality. American Economic Review, Papers and Proceedings, Vol. 86, 2, 218-223.

Benabou, R. (1996a). Inequality and Growth. NBER Annual, 11-74.

Benabou,R. (1996b). Unequal Societies. NBER Working Paper 5583.

Benabou, R. (2000). Unequal Societies: Income Distribution and the Social Contract. American Economic Review, Vol. 90, 1, 96-129.

Bird, R. and De Wulf, L. (1973). Taxation and Income Distribution in Latin America: a Critical Review of Empirical Studies. IMF Staff Papers, Vol. 20, 3, 639-682.

Birsdall, N. and James, E. (1990). How and Why Governments Misbehave. PRE Working Paper, the World Bank.

Blundell, R. and MaCurdy,T. (1999). Labor Supply: A Review of Alternative Approaches. IFS Working Paper W98/18.

Boadway, R. and Keen, M. (1999). Redistribution. In Handbook of Income Distribution, Atkinson and Bourguignon eds.

Cooter, R. and Helpman, E. (1974). Optimal Income Taxation for Transfer Payments under Different Social Welfare Criteria. Quarterly Journal of Economics, 88, 656-70.

Chu,K.Y., Davoodi, H., Gupta,S. (2000). Income Distribution and Tax and Government Social Spending in Developing Countries. IMF Working Paper WP/00/62

Deininger,K. and Squire, L. (1995). Inequality and Growth: Results from a new Dataset. World Bank Mimeo.

Devarajan, S., Hossain, S. (1998). The Combined Incidence of Taxes and Public Expenditures in the Philippines. World Development, Vol. 26, 963-977.

De Mello, L. And Tiongson, E. (2003). Income Inequality and Redistributive Government Spending. IMF Working Paper, WP/03/14.

Gupta,S., Davoodi,H., Alonso-Terme, R. (1998). Does Corruption Affect Income Inequality and Poverty ? IMF Working Paper WP/98/76

Hindricks, J., Keen, M., Muthoo, A., (1998). Corruption, Extortion and Evasion. Journal of Public Economics 74, 395-430

Hongyi,L., Squire, L., and Zou,H. (1998). Explaining International and Inter-temporal Variations in Income Inequality. The Economic Journal, Vol.108, 26-43.

IMF, 1980-1997, Government Finance Statistics Database (Washington)

Kanbur, R. and Lustig, N. (1999). Why is Inequality back on the Agenda ? (Draft), The World Bank.

Mauro,P. (1995). Corruption and Growth. Quarterly Journal of Economics, Vol. 110, 3, 681-712.

Mirrlees, J. (1971) An Exploration in the Theory of Optimal Income Taxation. Review of Economic Studies, Vol. 38, 175-208.

Myles, CD. (1995). Public Economics. Cambridge University Press.

Newbery, D. and Stern, N. eds. (1987). The Theory of Taxation for Developing Countries. Oxford University Press.

Persson, T. and Tabellini, G. (1991). Is Inequality Harmful for Growth ? Theory and Evidence. American Economic Review, 48, 600-621

Stern,N., Ahmad, E., The Theory of Reform and Indian Indirect Taxes (1984). Journal of Public Economics, 25, 259-298

Sachs, J. and Werner, A. (1997). Foundamental Sources of Long-Run Growth. American Economic Review, Papers and Proceedings, Vol. 87, 2, 184-88.

Sadka, E. (1976). On Income Distribution, Incentive Effects and Optimal Income Taxation. The Review of Economic Studies, Vol. 43, 2, 261-267.

Shah, A., Whalley, J. (1990). An Alternative View of Tax Incidence Analysis for Developing Countries. PRE Working Paper N. 462, The World Bank.

Slemrod, J. (1990). Optimal Taxation and Optimal Tax Systems. Journal of Economic Perspectives, 157-178.

Stern, N., Burgess, R., Taxation and Development (1993). Journal of Economic Literature, Vol. 31 -2, 762-830

Tanzi, V., Chu, K.Y. eds. (1998). Incombe Distribution and High-Quality Growth, The MIT Press, Cambridge.

Tanzi, V., Davoodi, H. (2000). Corruption, Growth and Public Finances, IMF Working Paper WP/00/182

Tanzi,V., Zee, H. (2000). Tax Policy for Emerging Markets: Developing Countries. IMF Working Paper, WP/00/35

Thirsk, W. (1997). Tax Reform in Developing Countries. The World Bank Regional and Sectoral Studies.

Tuomala, M. (1990). Optimal Income Tax and Redistribution. Oxford University Press.

United Nations Statistical Yearbook, 1997

Wood, A. (2000). Globalisation and Wage Inequalities: a Synthesis of Three Theories. DFID.

Wood, A. (1997). Openness and Wage Inequality in Developing Countries: the Latin American Challenge to East Asian Conventional Wisdom. The World Bank Economic Review, Vol. 11, 1, 33-57.

Zee, H. (1999). Inequality and Optimal Redistributive Tax and Transfer Policies. IMF Working Paper WP/99/60.

Annex I. Summary Statistics

Variable | Obs Mean Std. Dev. Min Max

-------------+-------------------------------------------------------------------

country 47

logland71 | 45 -.3955556 1.313001 -5.07 3.64

land80 43 4.551159 9.603987 .0238494 53.05667

gini 47 40.34532 9.457731 25.68 62.3

ginipost 47 .4680851 .5043749 0 1

-------------+-------------------------------------------------------------------

gini2 | 28 35.14286 7.969513 22.51 54.39

gini3 | 47 39.66021 10.03921 22.51 62.3

realgdppc | 38 6742.211 4640.492 914 15518

gdppc | 45 5054.489 5333.804 278.9 25377.51

-------------+-------------------------------------------------------------------

t1 | 36 .1447222 .13129 0 .45

t2 | 36 .6475 2.246349 0 13.7

t3 | 36 .0991667 .1003245 0 .35

-------------+-------------------------------------------------------------------

avtaxrate 3gdppc 36 .4896115 1.492916 0 9.133333

avtaxrate 0.5 gdppc 36 .0574884 .1025599 -.2950706 .35

taxprog | 27 2.986005 3.21417 -5.461982 11.92177

taxprog1 | 28 2.163596 1.649129 0 9.233334

taxprogdiff1 | 35 .4021429 1.524122 0 9.133333

taxprog1_2 | 28 7.303644 15.61148 0 85.25445

dirtaxindirtax | 44 2.146818 4.263585 .23 25.63

-------------+-------------------------------------------------------------------

socialexp | 24 .1725 .2269888 .02 1.13

predu | 23 14.99565 9.65738 3.6 43.5

eduinequal | 45 7.687333 14.5962 .01 69.42

sec | 47 12.00851 8.369765 .8 33.4

corrupt | 45 5.408444 2.646792 .38 9.94

-------------+-------------------------------------------------------------------

taxprog1*corr | 26 14.82286 16.83134 0 86.33167

-------------+-------------------------------------------------------------------

ginipost*taxprog1 | 28 .9017826 1.907867 0 9.233334

ginipost*taxprog1_2 | 28 4.32317 16.09114 0 85.25445

ginipost*predu | 23 8.24913 16.99921 0 69.43

ginipost*dirindirtax | 44 .6634091 1.438109 0 9.03

ginipost*corrupt | 45 2.501556 3.259673 0 9.48

taxprog1*predu | 22 46.63008 81.39076 7.457142 401.65

taxprog1*socialexp | 17 .5387484 .7683407 0 2.69944

| |

|Annex II. Dependent Variable: Gini Coefficient (High Quality) , whole sample |

| |

|(I) (II) (III) (IV) (V) (VI) (VII) |

| |

| |

|Ginipost -8.5*** -8.91*** -6.90*** |

|-6.68** |

|(3.09) (3.53) (2.03) (2.65) |

|Taxprog -0.5 |

|(0.49) |

|Taxprog1 -0.05 -9.21** - 8.46** -10.62** -12.7*** -10.97*** -11.34**-9.17** -19.29*** -11.11*** |

|(1.3) (4.56) (3.68) (4.57) (4.08) (4.5) (3.1) (3.69) (3.9) (3.36) |

|(Taxprog1)2 0.91** 0.77** 1.98*** 1.3*** 1.14*** 1.18*** 0.85** 0.51 1.09*** |

|(0.43) (0.34) (0.65) (0.39) (0.43) (0.31) (0.36) (0.34) (0.33) |

|Dir. Tax/ Indir.Tax -0.003 0.005 0.04 0.41* 0.07 -0.09 -0.01 -0.13 -0.04 -0.2 -0.1 |

|(0.29) (0.29) (0.27) (0.20) (0.27) (0.23) (0.26) (0.18) (0.21) (0.15) (0.19) |

|Prim. Educ. Exp. -0.42 ** -0.42** -0.53** 0.32 -0.55*** -0.34 -0.49* -0.21 -0.32* 0.14 |

|(0.17) (0.22) (0.2) (0.25) (0.17) (0.28) (0.24) (0.22) (0.15) (0.68) |

| |

|Taxprog1*Prim. Educ. -0.24*** |

|Exp. (0.09) |

| |

|Sec. Educ. -0.04 -0.23 0.13 |

| |

|(0.24) (0.29) (0.21) |

|Educ. Inequal. -0.17** 0.03 0.15 |

| |

|(0.07) (0.1 ) (0.1) |

|Gdp per capita -0.002*** |

|(0.00) |

|Land80 0.07 |

| |

| |

|Corruption -0.78 -1.71*** -4.87*** |

|-1.14 |

|(0.55) (0.58) (1.24) (0.1) |

|Corruption*Ginipost -1.09*** |

|(0.32) |

|Corruption*Taxprog1 1.59*** |

| |

|(0.57) |

|Corruption*Prim. Educ. |

|-0.04 |

|Exp. |

|(0.91) |

| |

|Cost. 47.61*** 46.23*** 62.36*** 62.22*** 60.03*** 71.1*** 67.99*** 70.52*** 69.83*** 95.1 *** 70.46*** |

|(3.61) (3.55) (8.42) (7.1) (7.96) (7.83) (8.79) (6.31) (7.05) (8.8) ( 9.92) |

| |

|R- Square 0.29 0.25 0.40 0.79 0.42 0. 59 0.63 0.85 0.69 0.87 0.80 |

|Adj.R Square 0.17 0.13 0.26 0.66 0.29 0.47 0.44 0.75 0.58 0.80 0.68 |

| |

| |

|*= significant at 10 percent level |

|**= significant at 5 percent level |

|***= significant at 1 percent level |

| |

| |

| |

| |

Annex III.

Incidence of Post-Transfer Povertya, Mid 1980s

| |Pre-Transfer |Post-Transfer |Percentage Reduction |

Australia 21.3 10.8 49.3

Canada 21 11 47.6

Germany 24.2 5.8 76.0 Netherlands 25.1 7.2 71.3

Sweden 29.7 5.3 82.2

Switzerland 15.6 7.4 52.6

United Kingdom 21.4 7.9 63.1

United States 23.4 18.1 22.6

a: Percentage of individuals in households with net income below 50 percent of adjusted median income.

Source: Barr, N. (1992).

-----------------------

[1] See

................
................

In order to avoid copyright disputes, this page is only a partial summary.

Google Online Preview   Download