4 Understanding the Revenue Performance of VATs
- Liam Ebrill, Michael Keen, and Victoria Perry
- Published Date:
- November 2001
The focus in the previous chapter was on the contribution that the VAT has made within the wider tax system, not the nature or determinants of the performance of the VAT itself. It is to those issues that the analysis now turns.
Summary Measures of VAT Performance
As can be seen from Table 1.3, countries with similar VATs as measured by the standard rate can have significantly different revenue performance as measured by the ratio of VAT revenue to GDP. In both the Philippines and Fiji, for instance, the standard rate of VAT is 10 percent: yet in Fiji the VAT collects over 6 percent of GDP in revenue whereas in the Philippines the tax yields less than 3 percent of GDP. Clearly there must be significant differences of design, behavior, and/or enforcement between the two VATs. More generally, the “efficiency ratio,”43 defined as the ratio of VAT revenues to GDP divided by the standard rate (expressed as a percentage), has been widely used as a summary indicator of the performance of the VAT and as a useful gauge of the extent to which the VAT bears uniformly upon a broad base. A low ratio, in particular, is typically taken as prima facie evidence of erosion either by exemption and reduced rates within the tax law or by imperfect enforcement.
The first row of Table 4.1 reports the pattern of efficiency ratios across all countries with a VAT of sufficiently long existence for its yield to have been observed.44 The average ratio is 34 percent, implying that a 1 point increase in the standard rate is associated with an increase in the share of VAT revenues in GDP of about 0.34 percentage point. However, the variation around this average is substantial. The highest efficiency ratio, 77 percent, is found in Brazil; next is New Zealand at 65 percent. There is also a marked regional variation in efficiency ratios. At one extreme are the small islands, where the efficiency ratio is on average distinctly higher than elsewhere; at the other is sub-Saharan Africa, where it is distinctly lower.
|Sub-Saharan Africa||Asia and Pacific||Americas||EU (Plus Norway and Switzerland)||Central Europe and BRO||North Africa and Middle East||Small Islands|
Although the efficiency ratio is widely used as a diagnostic tool in evaluating VATs, its limitations are significant. Errors in the measurement of GDP will clearly contaminate the measure, and—in so far as the extent of the error (from, say, the omission of informal activities) varies across regions—may bias comparisons of the kind invited by Table 4.1. More fundamental, however, is a conceptual weakness of the measure: a “perfect” efficiency ratio of 100 percent could be achieved by a product-type VAT levied at a uniform rate. This is misleading, however, since the appropriate benchmark against which to gauge a given VAT would typically be a consumption-type VAT levied at a uniform rate: so, for instance, starting with such a uniform consumption-type VAT, a change that brought investment into the tax base would increase the efficiency ratio—and so appear to be desirable—even though it would be a retrograde step from a tax policy perspective. This difficulty is in principle addressed by working instead in terms of the “C-efficiency” ratio, defined as the ratio of the share of VAT revenues in consumption (rather than GDP) to the standard rate. The appeal of this is that it is normalized by the benchmark of a uniform tax on all consumption: such a tax has a C-efficiency ratio of 100 percent.45 Any other value—higher or lower—indicates deviation from a single rate tax on all private consumption. Zero-rating of some item of consumption, for example, would tend to a C-efficiency ratio of below 100 percent; on the other hand, the inclusion of investment in the VAT base, or a break in the VAT chain resulting in the taxation both of final consumption and some of the constituent intermediate goods, will tend to result in C-efficiency of over 100 percent.
Of course a highly “imperfect” VAT could happen to score 100 percent: extensive zero-rating of final consumption might be offset in revenue terms by a failure to pay refunds promptly. Thus a C-efficiency ratio of 100 percent is a necessary but not sufficient condition for a VAT levied at a single rate on all consumption. The key difference between the efficiency and C-efficiency ratios is that the former is normalized by reference to an income-type VAT and the latter to a consumption type.
The distribution of the C-efficiency ratio is shown in the final row of Table 4.1. As one would expect, since consumption is less than GDP, C-efficiency is on average greater than the regular efficiency ratio: 58 percent rather than 35 percent. Indeed, in several cases, C-efficiency exceeds 100 percent. The C-efficiency ratio also varies more widely and in that sense may prove more informative.
More important, the two efficiency ratios can convey quite different impressions. Consider, for instance, the measured efficiency of the VAT in New Zealand, which is often held up as a model VAT, and that in Brazil, where there was incomplete crediting of tax on investment goods, so expanding the base relative to a consumption base. The New Zealand VAT has a regular efficiency ratio of about 65 percent, and so scores well below the 77 of the Brazilian VAT. But New Zealand has a C-efficiency ratio of 103 percent, close to the benchmark of 100 percent. Brazil, on the other hand, has a C-efficiency ratio of around 122 percent. Thus a comparison of C-efficiency ratios signals the relative merits of the two more accurately than does a comparison of the usual efficiency ratio.
In terms of regional comparisons, the picture that emerges from C-efficiency ratios (the final row in Table 4.1) is broadly the same in qualitative terms as that from the usual efficiency ratio (discussed above), though more marked in extent. The contrast between Western Europe and the small islands at one extreme and sub-Saharan Africa at the other is now more striking. A lesser but nevertheless notable difference is that whereas Asia and Pacific has a slightly lower efficiency ratio than the Americas, C-efficiency there is on average higher (reflecting, presumably, the higher savings ratio of the former).
When Is the VAT Most Effective in Raising Revenue?
What explains the variation in efficiency ratios described in the previous subsection? Or, put differently, what are the factors—both in terms of tax design and the wider economic environment—that are conducive to revenue productivity of the VAT?
The revenue produced by a VAT depends on three broad sets of factors: the rules describing rates, bases, threshold, and other structural features of the tax; the scale of taxable activities (the amount of final expenditure, for instance, on items taxable at the standard rate); and the degree to which the rules are complied with. The interactions between these factors are important. Tax rates, for instance, are typically set in light of tax bases and revenue requirements. The ease of enforcement will depend on the formal structure of the tax: multiple rates, for example, may lead to the misclassification of items, and a high standard rate may encourage evasion. To understand fully the revenue yield of a VAT these interactions would need to be explored in detail. Information limitations make this difficult. In particular, sufficient information is rarely available to estimate the VAT revenue that would be raised if the rules were implemented perfectly: the extent of evasion is generally unobserved.46
Given the difficulties of disentangling the interactions between rules, base, and enforcement as determinants of VAT yield, the analysis here focuses on the estimation of reduced forms, relating VAT yields to characteristics of both the VAT itself and the economy at large. This serves to highlight some key determinants of yield.
More precisely, in the most general terms, one can conceive of the revenue raised by a VAT as some function R(τ, α) of variables τ describing the rules of the tax system—rates, base, threshold, presence of a large taxpayer unit, etc.—and variables a that describe the wider economic environment (and so affect both the real decisions that determine the base of the tax and the extent to which the rules are enforced). Assume that policymakers choose the tax rules x to maximize some function Ω[R(τ, α), α]—which might (but need not) be the reduced form of a social welfare function—so that the tax rules are themselves some function τ(α) of the environment. Not all tax rules, however, are available in our data set: very basic structural variables, such as rates and (for many countries) thresholds, are, but such potentially important factors as the extent of exemptions or the resources allocated to auditing are not. Distinguishing between those VAT rules that are observed, τ0, and those that are unobserved, τu, one can thus consider revenue to be a function r (τ0, α)=R(τ0,τu(α), α) of the observed tax rules and wider features of the economy. It is regressions of this form that are considered here. The standard rate is always included among the observed tax rules, τ0. Note also that, with this specification, any role found for wider characteristics in α comprises two effects: the direct revenue effect of the characteristic itself (a better educated civil service and business community may facilitate compliance and so lead to higher revenue at any given configuration of tax rules, for example) and indirect effects through induced changes on the unobserved tax rules (better compliance perhaps enabling the government to set fewer exemptions than it otherwise would)47 The data available do not permit a disentangling of these two effects.
The core sample for the exercise is a cross-section of 99 countries meeting the requirements for VAT revenue and other data.48
Representative results are reported in Table 4.2. The dependent variable is in each case the ratio of VAT revenues49 to private consumption. Since the standard rate is always included as a regressor, the exercise is thus, in effect, that of modeling observed patterns of C-efficiency, though the results are—except in one important respect—essentially the same, taking the denominator instead to be GDP.50
The results in column A highlight some of the key themes to emerge from the exercise.
As would be expected, revenue increases with the standard rate. If the VAT took the form of a uniform tax on all private consumption, increasing the standard rate would lead to an equiproportionate increase in the ratio of VAT revenues relative to consumption. The coefficient below unity indicates, however, that revenues increase less than proportionately with the standard rate: a 10 percent increase in the standard rate increases revenues, on average, by only 7 percent. (If VAT revenues are initially 10 percent of consumption and the tax rate 15 percent, for instance, then increasing the standard rate to 16.5 percent will increase revenues to 10.7 percent). This cannot be interpreted as reflecting reduced consumption as a behavioral response to the tax, since the dependent variable relates to revenues relative to private consumption. Instead the implication is that higher standard rates tend to be associated with narrower VAT bases.
C-efficiency is somewhat higher, all else equal, the greater the importance of trade in the economy (the variable OPEN being the average of the import and export shares in GDP, as in the preceding chapter). The natural interpretation is that this facilitates the collection of VAT revenues through the collection of taxes on imports. Even though VAT collected on imports is credited further down the production chain, the border is a convenient place to begin the withholding mechanism. The scope for fraud in relation to the zero-rating of exports (see Chapter 15 below) could make trade disadvantageous in collecting revenues: empirically, however, any such effect seems to be dominated by the gain in relation to imports.51
There is no inconsistency between this finding that VAT revenues are higher, all else equal, in economies with higher trade and the conclusion in Chapter 3 that the increase in total revenue associated with the presence of a VAT is lower the greater the share of trade in the economy. The two can be reconciled by supposing that there is also some other tax instrument—trade taxes, for example—whose effectiveness in raising revenue is also greater in economies with more trade. Trade is conducive to good absolute performance of the VAT in such a country but the VAT is not necessary to exploit it.52
Though barely significant, there are indications of a role for the age of the VAT, measured in the number of years since its introduction (AGE). Old VATs raise significantly more, all else being equal, than do new.53 One interpretation is that administration of the VAT, and compliance with it, improves with experience.54 (For a sample of OECD countries, Agha and Haughton (1996) find that the ratio of actual VAT revenues to the yield they estimate would be obtained under perfect enforcement increases with the age of the VAT, suggestive of an experience effect in administration.) Another possible explanation is that unobserved features of the tax move in the direction of increased revenue after introduction of the VAT. It might be, for example, that exemptions are removed subsequent to introduction.
A powerful effect emerges for literacy, included in the exercise as a crude proxy for record-keeping and other administrative abilities of both taxpayers and tax authorities: all else equal, the VAT yields less revenue in less literate economies. This effect emerges even when conditioned on the age of the VAT. That is, the benefit of experience gained from operating a VAT and the benefit from preexisting administrative sophistication represent distinct effects.
The only regional dummy to emerge with significance is that for the small islands group, pointing to especially high VAT revenues there: a favorable effect over and above, it should be noted, that which these economies tend to enjoy as a consequence of generally having a high ratio of trade to GDP (as discussed in Chapter 3).
A disadvantage of the specification in Column A is that use of the literacy variable significantly reduces the sample size. Column B therefore reports the results of deleting literacy (and such regional dummies that are then insignificant). This increases the sample size by 20. The effects of openness, age, and the small island effect emerge even more strongly. The dummy for sub-Saharan Africa becomes significant: the apparent under-performance of the region being attributed, therefore, given the results in Column A, to relatively high levels of illiteracy.
The last two columns of Table 4.2 introduce additional variables relating to the structure of the VAT. The empirical possibilities are somewhat limited, given that many key features of the VAT and its administration—such as the scope of exemptions, the coverage of reduced rates, and/or the presence/absence of a large taxpayer unit—are not available for a sufficiently large sample of countries.55 Having experimented with a variety of measures, results are reported for four: the threshold (for those selling goods, and measured relative to per capita GDP), (THRESH); the range between the highest and lowest (nonzero) VAT rate, (RANGE); a dummy taking the value unity if the VAT extends to the retail stage and zero otherwise (RETAIL); and a dummy taking the value unity if VAT is levied on a broad base of goods and services and zero otherwise (GS).56
The sample size is significantly reduced by the limited availability of information on the threshold, and the results in relation to these structural variables are also not robust. A sense of the results is provided by the last two columns of Table 4.2: Column C adds these variables to the specification of column A, column D adds them to that of Column B.
The threshold never proves significant. GS and RETAIL are respectively each significant in one case but not the other, and in that sense neither case should be given much emphasis.57
Only one structural variable is significant in both cases: the range between highest and lowest rates. The effect is positive: that is, the wider the range the higher is VAT revenue.
This runs counter to the finding of Bogetić and Hassan (1993, 1995)—one of the few empirical attempts to explain VAT yields—that a greater range is associated with a lower efficiency ratio. They rationalize such a finding on grounds of administrative complexity and the scope for tax minimization created by rate differentials, though there are other interpretations. It could simply reflect demand responses. Assume there are two goods, with identical linear demand curves (dependent only upon their own price). Then increasing the tax on one while reducing that on the other by the same amount would have no effect on revenue if demand for each were unchanged; but since demand will shift from the highly taxed good to the more lightly taxed one, revenue will fall.
But the present finding of a positive relation between range and yield—which is based on a larger data set and conditions on important nontax variables—is also easily explained. Suppose, taking an extreme case that suffices to make the point clearly, that the marginal value of public expenditure is very high, so that the proper object of tax policy is essentially to maximize revenue. Suppose too that there are just two goods, differing in their elasticity of demand. Then optimal policy will require setting different tax rates on the two goods, implying that revenue is always lower when the range is zero than under the revenue-maximizing structure, for which the range is positive. Or it may be that the significance of the range variable reflects a positive revenue effect from taxing favored items of final consumption at a positive rate rather than simply exempting them. That said, although striking, little weight should be placed on this result, since it is empirically somewhat fragile.
One other feature of the results is of interest: introducing the structural variables renders the variable AGE insignificant. This suggests that the maturing of a VAT contributes to its yield not through increased experience but through tangible developments in the structure of the tax correlated with those included in the regressions. Indeed the results on the range of VAT rates suggest than even the tendency for the number of rates to increase as the VAT matures, noted in Chapter 1, can be associated with improved revenue performance.
The Importance of Imports in Collecting VAT
The significant role for the degree of trade in an economy in explaining VAT yields is consistent with a key empirical feature of the VAT: revenue collected on imports commonly accounts for a large proportion of total VAT revenues.
Table 4.3 reports the fraction of net VAT revenues accounted for by collections on imports for a sample of 22 developing and transition economies.58 In about two-thirds of them, more than half of all VAT revenue is collected on imports: the average is 55 percent. (The much lower shares in Kyrgyz and Russia reflect their application of the origin principle—under which imports are not taxed—to trade with other members of the Commonwealth of Independent States.)
The implication is not that the economic effects of the VAT are similar to those of a tariff. For insofar as the VAT collected on imports is levied on intermediate goods purchased by registered traders, it will be credited in the usual way against output tax on subsequent sales. And the figures in Table 4.3 still leave much VAT to be collected on domestic production. For simple, small economies, the differences between a VAT and a tariff may, indeed, be relatively slight, a point elaborated upon in Chapter 16. But the primary significance of the stylized fact illustrated in Table 4.3 is for administration: securing VAT collections on imports is generally a crucial part of ensuring effective collection of the tax throughout the chain of production. By the same token, getting collection right at that stage can go a long way to securing the success of the VAT overall.
This chapter has attempted to model the factors that contribute to the performance of the VAT. The “C-efficiency” ratio is to be preferred as a measure of VAT performance because—though by no means perfect—it is better calibrated as a summary measure than is the more traditionally cited efficiency ratio.
There is considerable cross-country variation in the revenue performance of the VAT. Among the factors conducive to strong revenue performance are:
a relatively high ratio of trade to GDP, presumably because of the relative ease of collecting VAT at the point of import;
relatively high literacy, presumably a surrogate for administrative capacity of taxpayers and tax collectors; and
the passage of time, in the sense that the performance of the VAT seems to improve over time.
Firm evidence of effects from specific aspects of VAT structure are hard to find. But there is some tentative suggestion that rate differentiation, as measured by the difference between highest and lowest statutory rates, serves to enhance revenues.