3 Is the VAT a Particularly Effective and Efficient Tax?
- Liam Ebrill, Michael Keen, and Victoria Perry
- Published Date:
- November 2001
Despite the evident importance of the spread of the VAT, there have been almost no systematic attempts to progress beyond anecdote in evaluating its performance. Is there any evidence that adoption of the VAT has actually produced the benefits claimed by its advocates? These are difficult questions, which the following attempts to begin answering.
Has the VAT Raised More Revenue than the Sales Taxes It Replaced?
The performance of a tax must be gauged by more than the revenue it raises. It must also be assessed in terms of the efficiency and fairness with which it raises that revenue, and the costs incurred by government and taxpayer in doing so. Nonetheless, revenue needs are often a key concern in the introduction of a VAT. Before turning to efficiency and fairness issues, this section therefore focuses on the narrow question of the revenue effects of introducing a VAT (equity issues being considered in Chapter 10 and collection costs in Chapter 5). Specifically, at the time of its introduction, the VAT is generally perceived as an explicit replacement for some preexisting sales tax. Has it in practice tended to raise more revenue than the predecessor tax?
Table 3.1 reports, for eight African countries that adopted the VAT during the 1990s (and whose particular experience is discussed further in Chapter 5), the difference (in percent of GDP) between the average revenue raised by the VAT in its first two or three years of operation (depending on data availability) and the average revenue raised by the predecessor sales tax in the matching period prior to the VAT.
Comparisons of this sort need to be interpreted with care. The introduction of the VAT is often intended to be revenue neutral, in the short term at least (though it would be naive not to acknowledge that commitments to revenue-neutrality often cloak the pursuit of future revenue gains). It may also be that VATs have been adopted at times when fiscal difficulties are so pressing that a significant increase in tax levels becomes politically feasible, and could have been achieved even under the predecessor tax system. Some upward trend in tax revenues can, moreover, be expected as development proceeds. And, not least, in those cases in which a VAT is introduced both to replace a sales tax and tariff revenues lost as a consequence of trade liberalization, any increase in revenues might in part reflect the fact that the VAT is substituting for a both a sales tax and tariffs.33
Despite these caveats, the extent of the revenue gains shown in Table 3.1 is impressive. The average revenue increase in the West African countries, in particular, is more than 1 percent of GDP.
Table 3.2 repeats this exercise—again comparing the two or three years either side of introduction, and subject, of course, to the same caveats—for a wider range of countries: all those that have adopted VATs since 1985 and for which sufficient data are available to compare revenue over the two or three years after the introduction of the VAT with that from the predecessor indirect tax.
|Sub-Saharan Africa||Asia and Pacific||Americas||Central Europe and BRO||EU (Plus Norway and Switzerland)||North Africa and Middle East||Small Islands|
This confirms the impression of a gain, on average, in sub-Saharan Africa. The apparent gain is even larger in the Americas and, especially, the Small Islands. The reduced tax share in CBRO is striking; however, in these cases the VAT was introduced against the background of a trend decline in the ratio of tax revenue to GDP associated with the economic transition itself. Excluding the CBRO countries, the share of VAT revenues in GDP exceeds that of the predecessor tax, on average, by about 1.1 percent of GDP.
Identifying Efficiency Gains from the VAT
As noted, the performance of the VAT is more than just a matter of the revenue it yields. In particular, the case for the value-added tax rests upon the view that a well designed and implemented VAT is a particularly efficient tax. It is this claim that is of central interest in evaluating whether the spread of the VAT has indeed been a desirable development.
It is difficult to quantify the impact on economic efficiency of adopting a VAT. The positive impact on production efficiency arising from the exclusion of intermediate activities from the tax base should in principle be reflected in a higher level of GDP, in so far as the VAT replaces a cascading tax. (Of course, the same benefit attaches to any tax, such as a RST, that excludes intermediate goods and services). The magnitude of this effect is hard to gauge, however, as it depends on the degree of substitutability in input choice. That said, results reported by Bovenberg (1987) from a computable general equilibrium model calibrated to the economy of Thailand of the 1970s suggest that raising an additional 1 percent of GDP by means of a uniform rate consumption-type VAT rather than an intensification of preexisting cascade taxes would raise real incomes by about 0.004 percent of GDP. If a VAT entails the elimination of unrelieved taxes on investment goods, positive growth effects might also be expected as a consequence of a reduced cost of capital. No estimates appear to have been made of the possible extent of this gain, which could exceed the static efficiency gains.34
An alternative to the direct empirical identification of efficiency gains is to look for testable implications of the supposed relatively low efficiency cost of raising revenue by a VAT. One possibility is that a government with a VAT raises more total revenue than would otherwise be the case. This could be investigated by looking at the variation in the ratio of tax revenues—for all taxes, not just VAT—to GDP between those countries with and without a VAT. This approach enables the effects of the VAT to be investigated empirically without specifying any particular alternative: the question asked is not whether VAT is preferable to alternative kinds of sales taxes, or trade taxes, or an expanded income tax, but rather whether the addition of a VAT to the armory of instruments at the disposal of the authorities’ appears to improve the overall efficiency of the tax system.
It is not inevitable that access to a more efficient tax instrument will lead to an increase in the tax-GDP ratio. While a lower marginal efficiency cost of raising revenue can be expected to lead to more revenue being collected, there are at least two additional effects associated with the higher real income presumably due to using a more efficient tax:
If public expenditure is a normal good, the desired level of its provision will rise, pointing to a higher tax ratio.
With an expanded tax base, any given level of expenditure can be financed with a lower tax rate, pointing to a lower ratio of tax revenues to GDP.
However, for the latter effect to dominate, implying that access to a more efficient tax instrument would lead to a lower tax ratio, the income elasticity of demand for public expenditure would have to be sufficiently low. The conventional wisdom, on the contrary, is that, at least over a wide range of incomes, the income elasticity of demand for public expenditure is greater than one. This suggests that a weak income effect is unlikely to more than offset the implications of a reduction in marginal efficiency cost. There are thus grounds for supposing that any efficiency gains associated with the adoption of a VAT may be reflected in a higher ratio of tax to GDP.
Before considering whether countries with a VAT in fact raise more total revenues than those without, note that a positive answer to this question does not necessarily make the VAT a desirable tax. Those who distrust government might regard the ability to raise large amounts of revenue at low marginal cost as an argument against the VAT, not for it: in this view, efficient taxes simply fuel inappropriate spending. Irrespective of the view one takes on this wider issue, however, determining the revenue/efficiency implications of the tax is clearly of considerable importance.
Finally, it would be misleading to test for an efficiency gain ascribable to the VAT by asking whether the ratio of VAT revenues to GDP in countries with a VAT exceeds the ratio of sales tax to GDP in those without. It might be, for instance, that the greater VAT revenues largely offset lower revenue from trade taxes (that are nondistorting in the case of goods that could not be produced domestically), with no necessary implications for efficiency. Nor would considering the sum of VAT/sales taxes plus trade taxes would not resolve the issue: higher ratios in countries with a VAT might simply arise because the VAT is used in part as a substitute for wage taxes, again with no necessary implications for efficiency. Accordingly, the empirical analysis considers a comparison across countries with and without a VAT focusing on total revenue.
Do Countries with a VAT Raise More Revenue?
The question has been little studied. Among the few existing studies, Nellor (1987) compares experiences across 11 European countries between 1955 and 1984, and finds some evidence that the adoption of a VAT did significantly increase tax-GDP ratios.35 However, although suggestive, that study is based on a relatively small sample comprising only developed economies. The rest of this subsection develops new evidence for a wider sample of countries.
The starting point of this exercise was the extensive literature that models international differences in the tax ratio in terms of such variables as per capita GDP (to gauge the possibility that wealthier countries spend a greater proportion of income on publicly provided items) and the share of agriculture in GDP (picking up the degree of development and, perhaps, the breadth of the tax base). The basic strategy in what follows was to modify estimating equations of this kind to allow effects from the presence of a VAT.
There is evidently an issue of causation. While the greater efficiency of the VAT is presumed to underlie any increase in the tax ratio, it could be that implementation of a VAT involves fixed costs that are warranted only if the revenue to be raised is sufficiently high. More generally, of course, the level of taxation and the instruments deployed to collect it are likely to be determined simultaneously. Accordingly, it is only the statistical association that is addressed here.
Moreover, the VAT can take many different forms: a zero-one dichotomy will exaggerate the difference between a crude VAT and a turnover tax that embodies a significant degree of crediting. Nevertheless, given the sample size, the broad distinction makes it worthwhile to ask whether the VAT leaves a trace in the revenue data.
The basic sample for this exercise comprised a cross-section36 of 183 countries, 99 with a VAT, and 84 without. Several countries that now have a VAT were treated as non-VAT countries for this exercise, since the revenue data available at the time of the work related to a period prior to the introduction of the VAT.37 In some contexts, data availability limited the sample size still further. The data are described in Appendix I.
As implied by the summary statistics in Table 1.5, the impact of a VAT may depend on which measure of aggregate revenue is considered. For instance, since the VAT is generally presumed to be most appropriately levied as a central tax (see also Chapter 17), it may have different effects on central government revenue as opposed to general government revenues (revenue, that is, consolidated over all levels of government). The analysis therefore considers the relationship between the VAT and three different measures of aggregate revenue.
General Government Revenue and Grants
The first measure of revenue is the broadest, and the one that allows the largest sample sizes: general government revenue and grants. This includes receipts at all levels of government, and not only receipts from taxes but also nontax revenue (including grants from abroad) as well. Results are reported in Table 3.3, the dependent variable in each case being (the logistic transform of)38 the ratio of general government revenue and grants to GDP.
|V x ln(Y)||0.25*||0.25*|
|V x AGR||–0.21*|
|V x ln(OPEN)||–0.44*||–0.64*||–0.42*|
|V x AF||0.69*||0.58*||0.37†|
|V x AP||0.04||–0.06|
|V x AS||0.10||–0.09|
|V x NMED||0.52*||0.42†||0.45*|
|V x SI||–0.28||–0.21|
Column A reports an estimating equation of the kind commonplace in the literature on “tax effort” (exemplified, for instance, by Tanzi (1987)), with no allowance for the possibility of an effect from VAT. The result replicates the key stylized facts from the literature (notably Tanzi, 1992): GDP per capita (Y) and the importance of international trade to the economy (OPEN)39—measured as the average of the ratios of imports and exports to GDP—each have a significant positive relation with the tax share.
The positive impact of income on the GDP share of taxes is well known. More interesting is the positive impact of the importance of international trade. Leuthold (1991) and Tanzi (1992) attribute this effect to the administrative ease of collecting revenues on imports and exports as they cross national borders. More recently, Rodrik (1998) has drawn attention to a robust positive association between the size of government and the importance of trade, arguing that this reflects the role of government in cushioning against external shocks, rather than the ease of taxing imports, because the effect remains after controlling for the extent of trade taxes and VAT. Alesina and Wacziarg (1998) postulate that the effect arises from the positive association between the role of international trade and size, with smaller countries having larger governments because of scale effects in the provision of public goods.
Interest here, however, centers on the role of the VAT variable in the context of aggregate revenue performance. Column B begins to explore this by adding a simple dummy variable V reflecting the presence of a VAT: V takes the value unity if a VAT was present throughout the financial year for which tax data are available, and zero otherwise.40 It proves insignificant.
The effect of a VAT may be more complex, however, than a simple shift of intercept. There may, moreover, be idiosyncratic regional effects on patterns of revenue, and indeed the impact of the VAT on wider revenues may also differ by region. Column C therefore reports a more general specification incorporating regional dummies and interaction terms that allow, in particular, for the impact of a VAT to vary with the level of development (as measured by GDP per capita), with the importance of trade, and by region. The regions here are the seven identified in Chapter 1 with EU+ (the EU plus Norway and Switzerland) taken as the reference.
Several striking features emerge from this specification. While the simple VAT dummy remains insignificant, strong effects are apparent in the interactions with income and trade importance. In particular, there is a revenue gain associated with the presence of a VAT that is larger the higher is the level of per capita income. On this score, while all countries gain revenue from the presence for VAT, it is the more developed countries that gain more. The interaction with the importance of trade, however, enters negatively: that is, there is on this account a revenue loss associated with the presence of a VAT that is greater the more important international trade is for an economy. One interpretation of this latter effect is that VAT in itself is less conducive to revenue raising than other devices that it may replace (most obviously trade taxes), so that, considering this effect in isolation, its adoption may weaken revenue collection. This important issue is elaborated upon in the next chapter.
Turning to regional effects, all areas except the Americas, CBRO, and small islands raise significantly less overall revenue—conditional on their real incomes, the importance of international trade, and the presence of a VAT—than do the EU+. Again, however, interesting interaction effects emerge: in both sub-Saharan Africa, and North Africa and the Middle East, countries that have adopted a VAT raise significantly more revenue. Indeed the apparent region-specific gain from adopting a VAT in sub-Saharan Africa roughly offsets the negative effect found for the region as a whole.
Column D reports the results of excluding variables not significant at 10 percent or less from the specification in Column C. The effects just described remain.
The pattern of VAT effects that emerges is thus complex. The revenue gain associated with the presence of a VAT is greater in richer countries, and seems also to be have been greatest in sub-Saharan Africa and the Middle East. Against this, however, is a loss associated with the VAT that increases with the importance of trade to the economy, perhaps reflecting the availability there of other effective revenue-raising devices. With revenue effects pointing in opposite directions, it is not a priori clear whether the presence of a VAT is associated with higher revenue or lower. Using the point estimates in Column D to compare, country by country, predicted revenues with and without a VAT, one finds that predicted revenue is indeed higher with a VAT in 55 of the 99 countries having a VAT, and that for these countries the average revenue gain associated with the presence of a VAT was around 0.4 percent of GDP. In all but two of the countries without a VAT, however, predicted revenue would be reduced by the presence of a VAT.
These calculations themselves thus provide only mixed evidence that the adoption of the VAT has been associated with higher revenues, and hence of an underlying efficiency gain associated from its use. But these results must be interpreted with great care. They assume, unreasonably, that the unexplained difference between predicted and actual revenues would be the same both with and without a VAT. They are of course open to all the wider criticisms of this approach—such as the neglect of self-selection issues—that are touched on later. Moreover, the results are sensitive to plausible differences in the magnitudes of the point estimates used. If, for instance, the point estimates in Column C are used—which involve a smaller negative interaction effect with international trade—one finds that predicted revenue is higher under the VAT in 93 of the 99 countries with a VAT (by an average of 4 percent of GDP), and in 64 of the 84 without (their average gain being over 4 percent). Finally, the results naturally reflect the fact that they are based on VATs that have been implemented in practice, rather than some ideal VAT. Actual VATs incorporate a range of non-recommended exemptions and other factors (see below) that will serve to undermine the efficiency of the tax.
The last column, E, reports a specification of the same general form as in Column C but replacing GDP per capita by the share of agriculture in GDP (AGR). As would be expected, more agricultural countries raise less revenue, all else equal. More interesting for present purposes, the simple dummy V is now significantly positive, pointing to a gain associated with the presence of a VAT. The interaction term with the share of agriculture itself is significantly negative. How to interpret this is not clear: the agricultural share may be a proxy for development and the ability to comply with and administer the tax; or the effect may simply indicate the difficulty of raising significant sums from this sector (see Chapter 9). Other interactions are as before: greater reliance on international trade is associated with a smaller gain from the VAT; there is a distinct revenue gain associated with the VAT in sub-Saharan Africa that again tends to offset—though now only partially—a general tendency for revenue to be lower in the region; and there again appears to be a distinct gain from the VAT in North Africa and the Middle East.
With effects again pointing in opposite directions, the question remains whether, on balance, the VAT is associated with higher or lower revenues. For the specification in Column E, however, the answer is clear-cut: for all countries, both those with and those without a VAT, predicted revenue is higher in the presence of a VAT. The positive coefficient on the VAT dummy is an order of magnitude larger than the negative interaction effects.
General Government Tax Revenue
Table 3.4 repeats the exercise for a narrower measure of receipts (now excluding nontax revenues and grants): general government tax revenue. This measure is only available, it should be noted, for a far smaller number of countries, so that the sample size is now only around 70. The specifications reported are analogous to those in Table 3.3.
|V x ln(Y)||0.53*||0.29*|
|V x ln(AGR)||–0.34*|
|V x ln(OPEN)||–0.06|
|V x AF||0.98|
|V x AP||1.24†||0.62||0.56|
|V x AS||0.39|
|V x NMED||0.78|
|V x SI||0.21|
Columns A and B differ little from those for general government revenue and grants. In particular, the simple presence of a VAT again proves insignificant. The more general specification in Column C again shows a positive interaction between the presence of a VAT and the level of per capita income. The negative interaction with the international trade variable now vanishes. The VAT dummy itself, however, now becomes negative and (almost) significant at 10 percent. The only other feature to emerge from Column C is that the distinct gain from the presence of a VAT for sub-Saharan African countries is no longer present, though there does now appear to be a gain in Asia and the Pacific.
Column D reports the consequence of eliminating all variables insignificant at 10 percent (other than the constant) from the regression in Column B. The qualitative results remain the same. Column E then replaces per capita GDP in the specification of Column D41 by the share of agriculture. The significant negative interaction term again suggests that the gain from the presence of a VAT is lower in more agrarian countries.
The point estimates in column D imply that the positive effect from the interaction between the VAT dummy and the level of GDP per capita will offset the negative fixed effect associated with the presence of the VAT at income levels over $2,343 per annum ($1,260 in Asia and Pacific). This is a disconcertingly high cut-off, but for the reasons noted in connection with Table 3.3—now compounded by the small sample size for this set of estimates—should not be interpreted too literally. In addition to the small sample size and sensitivity of results to the specification, such calculations are highly sensitive to plausible variations in parameter values. If the coefficient on the interaction with real income were one standard deviation higher, the critical income level would fall to $620 ($333 in Asia and Pacific).
Central Government Tax Revenue
The final set of regressions relate to a still narrower notion of revenues: tax receipts of the central government. The results, which are set out in Table 3.5, for the same series of specifications as for the other revenue measures, are easily summarized: apart from a couple of the interaction terms with regional dummies (pointing to a loss from the VAT in the Americas and a gain in North Africa and the Middle East, and, perhaps, in the small islands) there is no sign of any very marked effect on central tax revenues. In particular, there is no discernible systematic structural effect associated with the presence of a VAT.
|V x ln(Y)||–0.06|
|V x AGR||0.74|
|V x ln(OPEN)||0.02||–0.25|
|V x AF||0.26||0.43|
|V x AP||0.15||0.26|
|V x AS||–0.65*||–0.33*||–0.64*|
|V x NMED||1.67*||1.51*||2.18*|
|V x SI||0.38||0.71*|
The marked difference of results depending on revenue measure employed is striking. Moreover, the differences run counter to the simple intuition that suggests that since the VAT is best suited to be an instrument of central government, any revenue effect arising from its presence is most likely to be found in central government revenues.42 On the contrary, the results here suggest that the center shares any revenue gain associated with the presence of a VAT with lower levels of government by vacating other tax bases. A resolution of these issues is likely to require, among other things, a greater understanding of the decision to adopt a VAT.
It is hard to gauge directly the extent to which the spread of the VAT has increased the efficiency with which productive resources are allocated. This chapter has focused instead on the search for indirect signs of an efficiency gain, in the form of an increase in the ratio of tax revenues to GDP associated with the presence of a VAT. The empirical results that have been presented here must be interpreted with great caution. They rely only on cross-section variation, so that the presence of a VAT may be proxying for country effects—such as administrative sophistication—that are correlated with revenue performance. Panel data are not available, however, to control for this. There is, moreover, an endogeneity problem: those countries that have adopted a VAT may be marked by characteristics known to them, but unobserved in the data, making them particularly well suited to the VAT. Alternatively, they may be countries that turned to a VAT in the face of a deteriorating revenue situation. Progress will require modeling the decision to adopt a VAT, an important topic for future work.
Caveats aside, however, a number of suggestive results do emerge from the analysis:
There is some evidence that the presence of a VAT is associated with a higher ratio of general government revenue and grants to GDP; weaker evidence that, except perhaps for the poorer countries, it is associated with a higher ratio of general tax revenue; and no evidence at all that (apart from region-specific effects) it is associated with a higher (or lower) ratio of central tax revenues.
There is thus an important, albeit limited, sense in which the supposed ability of the VAT to bolster revenues in an efficient manner is borne out by the data. The extent of the effect, however, cannot be estimated with any precision.
The absence of an impact on central tax revenues is striking, particularly since—for reasons related to the difficulties associated with interjurisdictional trade (see Chapter 11)—it is widely regarded as a tax that is properly levied only by central government. The implication of the lack of significance found for the VAT in relation to central tax revenues is that any direct addition to central tax revenue is offset by a reallocation of other tax instruments to the lower levels of governments.
Two other lessons stand out.
The revenue gain associated with the VAT increases with the level of GDP per capita and decreases with the share of agriculture in GDP.
While these correlations are strong, the results are not totally clear since the effects remain once literacy has been controlled for, implying that the variables are not capturing administrative sophistication as such. In the case of agriculture, the result may simply reflect the typical exemption of agricultural output.
The revenue gain also appears lower, all else equal, the more important international trade is for an economy, though this finding is more tentative.
This may reflect the availability in such economies of other devices—most obviously tariffs—that are no less effective at raising revenue than the VAT.