XVIII. The Distributional Impact of Public Expenditure
- Ke-young Chu, and Richard Hemming
- Published Date:
- September 1991
What data are needed to determine the incidence of fiscal policy? Are such data likely to be available in developing countries?
Who benefits most from the provision of cash benefits and other public expenditure programs? Who bears the burden of taxation?
How should income inequality be measured? How redistributive is fiscal policy? Which programs are most beneficial to the poor?
How efficient is redistribution? Is it possible to achieve the same amount of redistribution with lower taxes?
The government has at its disposal a wide range of instruments with which it can influence both the functional and size distribution of income. Nonfiscal policy instruments include employment, wage, and price controls; the main fiscal policy instruments are taxation and cash transfers, although in-kind transfers and benefits from public goods have an effect on individual welfare that has an income equivalent. The focus of attention in this note is the impact of public expenditure on the size distribution of income. However, the redistribution that results from public expenditure should not be analyzed independently of the redistribution associated with the taxes that pay for it. Because a wide range of efficiency and equity objectives are served by both taxation and expenditure, in attempting to gauge the success with which the government meets these objectives it is necessary to examine the tax and expenditure system as a whole. Moreover, a case can also be made for looking at both nonfiscal and fiscal policies together, and reaching a judgment about the overall redistributive impact of government. This, however, is generally beyond the capacity of available methodology and data. Indeed, assessing the redistributive impact of fiscal policy alone presents as yet unresolved methodological difficulties and is hampered by inadequate data.
The standard approach to measuring the redistributive impact of fiscal policy is based upon fiscal incidence analysis, which involves a comparison of the distribution of income after taking into account taxes, transfers, and benefits with the distribution assumed to hold in their absence. In an ideal world, the methodology of a fiscal incidence study might proceed as follows:
- (i) Beginning with the distribution of original income (i.e., private income from all sources);
- (ii) Allocate taxable cash transfers by income to yield the distribution of total income;
- (iii) Allocate direct taxes by income to yield the distribution of post-tax income;
- (iv) Allocate indirect taxes, nontaxable cash transfers and in-kind transfers by income to yield the distribution of net income; and
- (v) Allocate benefits from public goods by income to yield the distribution of final income.
The terms used here to describe different income distributions are not applied universally, and should be regarded as no more than labels to distinguish between the various income concepts employed.
While this methodology might seem straightforward, the only distributions that are generally available are those described in (ii) and (iii) from tax records; however, in some cases information is available from expenditure surveys and other sources to construct the distribution described in (iv). Major problems arise in constructing the distributions described in (i) and (v). Much of the concern about the efficiency consequences of government activity (see below) derives from the belief that taxes and transfers affect economic behavior; to derive the distribution of original income these behavioral responses should in principle be taken into account. In most practical applications, however, the distribution of original income differs from the distribution of post-tax income only by the exclusion of direct taxes and cash transfers. The distribution of benefits from public goods, and therefore the distribution of final income, cannot be observed and is either inferred or ignored.
Fiscal incidence studies are generally available only on industrial countries, and even then data limitations often mean that their coverage is incomplete or strong assumptions are required to arrive at meaningful results. Rarely in developing countries are even rudimentary analyses feasible. The focus of attention in this note is mainly fiscal incidence and redistribution in industrial countries. However, if suitable data are available in developing countries, the industrial country studies can be replicated. Moreover, a number of the recurring and especially unintended redistributive implications of policies in industrial countries probably carry over to developing countries, and should influence policy design even in the absence of supporting data.
The study of tax incidence is one of the most highly developed subfields of public finance, and there is a large measure of agreement as to the broad incidence of most major taxes. Assessment of tax incidence is carried out on a tax-by-tax basis, and sophisticated economic models are sometimes used to determine who actually ends up paying each tax. The final economic incidence often differs quite appreciably from statutory (or legal) incidence. The main conclusion from tax incidence studies undertaken in industrial countries is that income taxes are usually progressive—the proportion of total income paid in tax increases with total income—while consumption taxes are generally regressive—the share of taxation declines with total income. Most countries have a mix of taxes on income and consumption, and overall tax incidence turns out to be roughly proportional—the share of taxation in total income does not change over most of the income range. However, there is some regressivity at lower incomes where consumption taxes dominate, and some progressivity at higher incomes where income tax rates often climb quite steeply.
Transfer incidence reflects the distribution of cash and in-kind transfers to individuals. Cash transfers are simply negative taxes, which tend to be highly progressive—implying that they are a decreasing proportion of income-since they are either a flat amount or are targeted toward the poor. When data on transfer receipts are not available these can be assigned on the basis of entitlements. However, in many countries there are problems with low take-up of benefits—reflecting inadequate information, poor administration, and stigma problems—which often undermine their progressivity. An in-kind transfer refers to the receipt of a good that is available in the private market or could be supplied privately, such as food, housing, health care, and education. The usual procedure is to assign program costs equally to identifiable recipients. As a result, the redistributional impact is reflected in the relationship between income and the use of those benefits. The tendency is for food and housing benefits, together with basic health care and primary education, to be progressive, and in the case of food often highly progressive. However, expenditure on more lavish health care and higher levels of education are biased toward the better-off and are therefore regressive (see the notes on Price Subsidies, Health Care, and Education for further discussion).
The equivalent treatment of cash and in-kind transfers, while convenient, can be misleading. Because income is fungible, the receipt of a $1 cash transfer can reasonably be valued at the direct cost of providing it, namely $1. This procedure is less valid in the case of in-kind transfers. Recipients may value the receipt of $1 in food at less than $1; the appropriate valuation should reflect willingness to pay. The procedure of allocating costs is a response to the fact that willingness to pay is unobservable. However, in adopting this approach, the possibility of a significant divergence between the measured distribution of income and the associated distribution of welfare or utility is being introduced.
Public goods have the characteristic of being nonrival in consumption. It is therefore impossible to identify one individual’s consumption of the services associated with a public good. It is, however, clear that individual valuations of the benefits derived from public goods will vary; for example, pacifists presumably place little value in the provision of defense services. Allocating these benefits across individuals is therefore difficult, and attempts to do so are very approximate. While some fairly sophisticated procedures have been developed to place individual valuations on public good provision, the most common approach is to assume that incidence is proportional to income and that total benefit equals total cost. Such an assumption is unlikely to be valid. Indirect evidence from voting behavior, survey results, and information on the demand for close substitutes and complements to public goods, suggests that public good provision is most highly valued by the rich. For example, defense services are worth more to property owners, and environmental protection is important to those who engage in outdoor leisure activities. Furthermore, as in the case of in-kind transfers, there is no direct link between total benefits and total cost; available evidence suggests that the former exceeds the latter across a wide range of programs. While these are serious problems, the proportionality assumption is in effect equivalent to treating benefit incidence as unmeasurable and therefore ignoring the impact of public good provision on income distribution.
Net fiscal incidence
Net fiscal incidence refers to the net outcome of tax incidence, transfer incidence, and benefit incidence. Numerous empirical studies suggest that the pattern of incidence is broadly as shown in Chart 1. Taxes are regressive at the lower end of the income scale, broadly proportional over a wide range of middle incomes, and progressive at higher incomes. Cash transfers are very progressive. In-kind transfers are probably fairly progressive at low income levels and possibly regressive at high income levels, but such a conclusion is sensitive to how benefits from such transfers are calculated. Few attempts are made to characterize the incidence of benefits derived from public goods. The combination of taxes and cash transfers is fairly progressive at low income levels, reflecting largely the impact of transfers, and is then more or less proportional—or mildly progressive—over the range of the income distribution where most individuals are located. Taking in-kind transfers into account probably increases progressivity at low incomes but could give rise to some regressivity at higher incomes.
Chart 1.Fiscal Incidence
In order to translate the pattern of fiscal incidence described above into a conclusion about the redistributive impact of fiscal policy, it is necessary to make some judgement about changes in distribution. It is common practice for such judgements to be based upon the Lorenz curve of the income distribution. The Lorenz curve maps the cumulative share of income received by the bottom X percent of income recipients against X (with X in the range 0-100 percent). An unequal distribution has the characteristic shape shown in Chart 2. The amount of redistribution associated with fiscal policy can be gauged from an examination of shifts in the Lorenz curve. Where Lorenz curves do not intersect, as is the case with Lorenz curves L(x) and L(y) in Chart 2, judgments about redistribution are straightforward. L(x) is unambiguously more equal than L(y). With a given total income, any equalizing (upward) shift of the Lorenz curve can be achieved by redistribution from the richer to poorer individuals. When some income is lost in the process of redistribution—due to administrative costs for example—an upward shift in the Lorenz curve is still equalizing. How the resulting reduction in average income should be reflected in inequality measures is discussed below.
Chart 2.The Lorenz Curve
When Lorenz curves intersect—implying that one distribution cannot be arrived at from another purely by transfers from richer to poorer individuals (or vice versa)—it is necessary to attach a welfare weight to different incomes in order to yield a complete ranking of distributions. The problem with many measures of inequality is that they embody implicit weights that may not accord with widely held social values. For example, the popular Gini coefficient, which is closely related to the Lorenz curve, applies rank order weights to successively smaller incomes, independently of the size of these incomes. Rather than accepting such a weighting system, or any other implicit set of weights, it has been suggested that a measure based upon an explicit set of welfare weights be used. One such measure—the Atkinson inequality index—utilizes an inequality aversion parameter to reflect society’s dislike of inequality. The Atkinson index measures the proportionate reduction in total income which, if the remaining income were equally distributed, would yield the same level of aggregate welfare as the existing unequal distribution—in a sense, it measures the welfare cost of inequality. While virtually all of the most widely used inequality indices would record an upward shift in the Lorenz curve as a reduction in inequality, different measures can yield conflicting results when Lorenz curves intersect. Moreover, the Atkinson index, and similar indices, need not yield the same ranking of distributions over different values of the inequality aversion parameter. Unfortunately, there is no fully scientific basis for placing values on this parameter, and the usual approach is to compute the index over a fairly wide range of values to test its robustness.
Poverty alleviation is an important aspect of redistribution policy, especially in developing countries. While there is some precedent for regarding concern about poverty as a limiting case of concern about inequality—for example, an extremely large value of the inequality aversion parameter of the Atkinson index implies a concern with the welfare of only the worst-off members of society (the Rawlsian approach to social justice)—it is more common to treat poverty as a quite separate concept from inequality. To integrate the two robs both concepts of part of their meaning, and to describe the bottom Y percent of the income distribution as the poor is generally inappropriate. That is not to say that the distribution of income is not relevant to defining poverty. But there has to be a possibility of eliminating poverty despite the persistence of inequality. Poverty measurement is discussed in the note on Poverty and Social Security.
A tax and benefit system that is progressive over the entire income range is unambiguously equalizing. It implies that all redistribution is systematically from richer to poorer individuals. As a result, the Lorenz curve shifts upward, and most summary measures will reveal less inequality. When the tax and benefit system has a regressive range Lorenz curves need not intersect, in which case unambiguous conclusions about redistribution are still possible. But if Lorenz curves do intersect, unambiguous conclusions are not possible. Different people will usually be able to appeal to an inequality measure that supports their preferred position about redistributive impact. Clearly, references to the progressivity and regressivity of various taxes and benefits at different income levels are generally not a sufficient basis for distributional judgments, since there are both progressive and regressive influences. Rather, it is necessary to construct Lorenz curves or, where Lorenz curves intersect, to compute the preferred inequality index.
Both fiscal incidence and especially redistribution analysis are complex and highly subjective exercises. Indeed, even the most ambitious studies rarely explore the issues that arise out of the recent discussions of the appropriate way of measuring inequality. Lorenz curves and Gini coefficients remain the basis of most studies. Moreover, the preceding discussion has focused on the extent to which fiscal policy affects income distribution only in the very narrowest sense. To appreciate this, some of the limitations of the standard approach to redistribution analysis should be spelled out.
Limitations of Redistribution Analysis
Variations in average income
While it is clear that the more equal distribution with Lorenz curve L(x) in Chart 2 is preferred on distributive grounds when average income is the same or when associated with a higher average income, it is less obvious that for a given reduction in inequality there is no limit to the reduction in average income that is acceptable to achieve it. One solution to the trade-off is implied by the generalized Lorenz curve, which adjusts the standard Lorenz curve for changes to mean income. This is illustrated in Chart 3, where nonintersecting standard Lorenz curves yield ambiguous welfare comparisons when a more equal distribution is associated with a reduction in average income by a factor α, and generalized Lorenz curves GL(x) and GL(y) cross. It is easy to construct the case where standard Lorenz curves cross but generalized Lorenz curves do not. To generate complete rankings, it is necessary to specify a social welfare function that embodies an inequality-efficiency trade-off, inequality being reflected in an inequality index and efficiency being reflected in average income.
Chart 3.The Generalized Lorenz Curve
The unit of analysis
Reference has been made above only to the distribution of income between individuals—this is simply for convenience of exposition. The unit of analysis, whether it is the individual, the family, or the household, is important. Each alternative has disadvantages which require conclusions based upon them to be qualified. In particular, it is always difficult to take properly into account the extent of income transfers between family and household members and their impact on inequality. The choice of unit is also important when attempting to compare income with needs—as in poverty studies—given that needs vary with family and household size. A common practice is to focus on income per adult equivalent.
Concept of equity
It is usual to distinguish between vertical equity and horizontal equity when discussing income distribution. Vertical equity relates to the relative treatment of individuals (families or households) at different income levels. Horizontal equity relates to the way in which individuals at the same income level are treated. Vertical equity tends to be the focus of income distribution studies. It turns out, however, that often the combined impact of taxes, transfers, and benefits results in individuals at different income levels being treated very unequally, and it is not uncommon for the ranking of individuals in similar circumstances in the income distribution to be markedly affected by fiscal policy. This often reflects well-defined objectives of fiscal policy, such as a desire to encourage particular patterns of consumption, to promote home ownership or saving for retirement, or to foster regional development. It may, however, reflect inadequate administration or capricious discrimination. Most analyses of income redistribution take no account of rerankings, be they intentional or unintentional. However, studies that have attempted to reflect reranking have revealed a tendency for redistribution, especially through taxation, to be overestimated by the standard procedure.
Taxes, transfers, and benefits affect economic behavior—the supply of labor, together with consumption, savings and investment are all influenced by fiscal policy variables. A fully articulated redistribution study would compare the distribution of income taking into account fiscal policy with one that would prevail in its absence. At the very least, this would require knowledge of the relevant behavioral coefficients, and the adjustment of certain economic variables to reflect the impact of fiscal policy on them. Ideally, however, account should also be taken of the full general equilibrium implications of fiscal policy on the economy to specify properly the counterfactual that is being used as a baseline for comparison purposes.
Lifetime and intergenerational redistribution
Redistribution as discussed above is a contemporaneous concept, occurring at one particular point in time. However, many public programs—for example, social security—are directed toward redistributing income over individual lifetimes and have implications for redistribution between successive generations. The dynamic issues raised by such programs are difficult ones, but progress is now being made in recasting distributional analyses to reflect these more complex equity objectives.
The Efficiency of Redistribution
An implication of the pattern of redistribution shown in Chart 1 is that there is a large gap between the amount of gross redistribution associated with government programs and the resulting net redistribution. A significant proportion of the benefits resulting from expenditure programs flow back to those who pay the taxes to finance them. It has been argued that the disincentive effects associated with unnecessarily high taxes and the payment of benefits to those who do not need them imply that the efficiency cost of the net redistribution that actually takes place is unjustifiably high. Middle class capture has become one of the most powerful arguments used by those who push for less government intervention. To the extent that the same amount of net redistribution, or more, can be achieved with smaller gross income flows, proponents of this view argue that policies should be put in place to reduce gross income flows, and work effort, saving, and risk-taking will increase as a result. While it is difficult to deny the existence of some inefficiency associated with redistribution, the notion that net and gross income flows can be brought reasonably closely into line may be both inappropriate and unrealistic. As noted above, contemporaneous redistribution is only one aspect of redistribution policy, and to the extent that lifetime and intergenerational redistribution, for example, are important, a large mismatch between gross and net income flows may be justified. Moreover, universality in benefits has advantages where there are difficulties or high costs involved in identifying potential benefit recipients. These and other issues related to reducing gross income flows while fulfilling distributional objectives are discussed in the note on Poverty and Social Security.
Because fiscal incidence studies are more commonplace in industrial countries, the United Kingdom example is included to illustrate the type of analysis that is possible. The Bangladesh example is among the more detailed available for developing countries, but is still fairly rudimentary compared to its United Kingdom counterpart.
Fiscal incidence in the United Kingdom
The results of a fiscal incidence study for the United Kingdom are summarized in Table 1. A number of assumptions and simplifications were necessary to arrive at these results, which together tend to overstate the progressivity or redistributive impact of the tax and benefit system. These include: an assumption that all taxes are borne by those who formally pay them; in-kind benefits are valued on a cost basis, while they tend to be worth more to the rich; and public goods and other general expenditures are not included. Table 1 shows that although taxes rise sharply with income, in relative terms taxes are mildly regressive. Cash transfers, which are targeted toward the poor, are strongly progressive. In-kind transfers are also progressive. The net effect of taxes, cash transfers, and in-kind benefits is that the first few income deciles receive more in benefits than they pay in taxes, and vice versa at higher income levels; as a result the tax and benefit system appears fairly progressive.
Fiscal incidence in Bangladesh
The fiscal incidence analysis for Bangladesh summarized in Table 2 is based upon far less detailed data than those available in the United Kingdom. The tax incidence assumptions are uncontroversial, and the incidence of the tax system reflects the offsetting effect of progressive import duties on luxuries and income tax—from which the poor are exempt—and regressive excise duties. Cash transfers—including food subsidies—are progressive at very low incomes but poor targeting results in some regressivity at higher incomes. In-kind benefits—comprising the imputed value of education, health, agriculture and water development expenditures—tend to be regressive, reflecting mainly the value of state education to the rich. While the ultra poor (with annual incomes below Tk 20,000) receive more in transfers than they pay in taxes, the poor (with annual incomes in the range Tk 20,000-24,000) on average make marginally positive net tax payments.
Atkinson A.B. and J.E. Stiglitz Lectures on Public Economics (New York: McGraw Hill1980).
Saunders P. and F. Klau The Role of the Public Sector OECD Economic Studies No. 4 (Paris: Organization for Economic Cooperation and Development1985).