WOFRKIN(G P'AF'FRS Country Opwhtoe Europe and Central Asia Country Department I The World Bank October 1993 WPS 1203 Determinants of Value-Added Tax Revenue A Cross-Section Analysis Zeljko Bogetic and Fareed Hassan Empirical analysis of value-added tax revenues on a sample of 34 countries conforms with conventional wisdom from theoreti- cal and case studies. The key implication is that for value-added tax to provide superior revenues, it should be levied in a single rate on as broad a base as possible. And tax administration and enforcement must be tough to ensure compliance. Policy ResearchWotaingPpersdiusaem.ac the findings of work in po and nacowuge the exchange of idea asong Bank staff and alBother interested in devlopntinue These papn dist ibuted bytheRes hAdviKry Staff, cany thenames ofthe authors, reflect only theirviews, and should beused and cited accordingly. Thefindings, interpretations,andconclusions atheauthors'wn. They ghould not be attributed to the Word Bank, its Board of Directors, its managemnat, or any of its manber cosutrics. WPS 1203 This paper-a product of the Country Operations Division, Europe and Central Asia, Country Department I- is part of a largereffort in the department to emphasize public finance reform issues inpolicy dialogue and economic and sector work. Copies of the pcper are available free from the World Bank, 1818 H Street NW, Washington, DC 20433. Please contact Faith Smith, room H5-245, extension 36072 (October 1993, 14 pages). Value-added tax (VAT) has become a major tax the rate, the base, and rate dispersion. The rate instmment in over SO countries and an important and the base coefficients are signiflcant and with element in tax policy advice to developing the expected positive sign in all of the estimated countries. But few studies have empirically versions of the model. An esdmated model is tested some basic hypotheses about the perfor- used with appropriate ctveats to predict VAT mance and key feature of VAT as a revenue- revenue potential in countries (such as Bulgaria) raisiuig instrument, that are thinking of introducing a single rate VAT. Bogeti6 and Hassan examine the main determinants of VAT revenue in a simple cross- They also find that - other things being country framework using data from 34 countries constant - VAT generates higher revenue in to answer certain key questions: What empirical countries with a single VAT rate than in coun- rlationship emerges from existing data on VAT tries with multiple VAT rates. The difference in revenue and VAT rates for countries with a the esdmated models for the two country groups single VAT rate? How much, on averdge, can a is statistically significant, indicating a structural 1 percent increase in the VAT rate be expected change. However this change in the pattem of to raise VAT revenue as measured by VAT-to- VAT revenues cannot be explained exclusively GDP ratio? What key determinants of VAT in terms of differences in rate structure. A revenue emerge from a cross-country analysis of satisfactory explanation must include other the full sample of countries? Is there a statisti- factors, such as the base and tax administration cally significant difference in VAT revenue capacity. performance between countries with a single VAT rate and countries with multiple VAT The key policy implications are simple: to rates? provide superior revenues, VAT should be levied in a single rate on as broad a base as possible. The results of their regressions generally And tax administration and enforcement must be confirn the conventional views on the key tough to ensure compliance. variables influencing VAT revenue performance: The Policy Research Working Paper Series disseminates the frndings of work under way in the Bank. Anobjective of the series is to get these findings out quickly, even if presentations are less than fuDy polished. The findings, interpretations, and conclusions in these papers do not necessarily represent official Bank policy. Produced by the Policy Reseach Dssermiation Center DETERMINANTS OF VALUE ADDED TAX REVENUE: A CROSS-SECTION ANALYSIS 2eUko Bogetid and Fareed Hassan TABLE OF CONTENTS: I. Introduction ............ ............ l II. VAT Rates, Base and Revenue Performance: Some Stylized Facts ...... ........... 2 III. Hypotheses, Models and Results ....... ... . 3 A. The Sample of Single VAT Rate Countries 5 B. Full Sample of 34 Countries ...... . 7 C. Do Countries With Single VAT Rates Mobilize More Revenues? .... .... 9 IV. Conclusion and Policy Implications .... ...... 9 Appendix ............ ............ 11-12 References ......................... 13 DETERMINANTS OF VALUE ADDED TAX REVENUE: A CROSS-SECTION ANALYSIS 2eUko Bogetid and Fareed Hassan' I. Introduction Value-Added Tax (VAT) has become a major tax instrument in about 50 countries worldwide contributing to their budgets, on average, 5.1 percent of Gross Domestic Product (GDP) in 1988 (see Table 1 in the Appendix). The global trend to introduce VAT in more countries is continuing. Since 1988, another 8 countries, including Canada and Japan, have introduced VAT, and an additional 15 countries, including the U.S., are seriously considering its adoption (Tait, 1991). VAT has also become an indispensable component of the tax advice and tax reforms in developing countries2. The growing practice of VAT is reflected in the extensive literature on technical, economic, and distributional dimensions of VAT, and there is a growing consensus on the "best practice" and desirable features required of a good VAT system3. However, very few studies have empirically tested some basic hypotheses about key features and the performance of VAT as a revenue-raising instrument. This is surprising since there now exist relatively rich data on the VAT revenue performance (e.g., Government Finance Statistics, IMF), rates and other important characteristics of this tax. In particular, revenue performance and its I/ The authors are a Country Economist and a Consultant at the World Bank, Washington D.C., Department for South-East Europe. Helpful and thought-provoking comments and criticisms from Alan A. Tait, Carlos A. Silvani, Milka Casanegra de J.intscher, Dan Hewitt and R. Kyle Peters are gratefully acknowledged. The authors are solely responsitle for any remaining errors. 2/ See Goode (1993). 3/ See, for example, the collection of napers in a World Bank book edited by Gillis, Shoup and Sicat (1990). A comprehensive treatment of practical problems and options in implementing a VAT is given in Tait (1988). Also, see chapter one by Bird and Casanegra de Jantscher (on tax administration), chapter five by Due and Greany (on the introduction of VAT in Trinidad and Tobago), and chapter eight by Silvani (on compliance), in the recent IMF volume oii tax administration edited by Bird and Casanegra de Jantscher (1992). 2 determinants' has been one issue of special relevance for fiscal authorities. The emerging conventional wisdom, based largely on practice and numerous country case studies, suggests that a single rate VAT (with the rate between 10 and 20 percent), with very few exemptions and, therefore, a broad base is superior to a VAT with multiple rates and many exemptions which reduce its base and complicate administration'. With more quantitative data on VAT now available, there is consiWlerable scrpe for empirical research on VAT, both in comparat,ve and cross-sectional studies. The objective of this paper is to examine the main determinants of the VAT revenue in a simple cross-country framework. Specifically, we seek to answer the following three questions. First, what empirical relationship is emerging from the existing data on VAT revenue and VAT rates for single VAT rate countries, and how much, on average, a one percent VAT increase in the rate can be expected to raise VAT revenue, as measured by a VAT-to-GDP ratio? Second, what are the key determinants of VAT revenue emerging from a cross-country analysis of the full sample of countries (those using single rate and those that practice multiple VAT rates)? And third, is there a statistically significant difference regarding the VAT revenue performance between single VAT rate coun'ries and countries that adopted multiple VAT rates? It is hoped that answers to these questions can provide firmer quantitative background for policy prescriptions regarding VAT. Also, robust estimates of the VAT revenue may provide guidance on revenue potential to policymakers in countries which are currently planning to introduce a VAT. The structure of the paper is as follows. Section II provides a brief, stylized overview of existing VAT rate structures, revenue performances and tax bases. The empirical methodology and the estimated models are presented and discussed in section III. Section IV contains concluding remarks and policy implications. II. VAT Rates, Base and Revenue Performance: Some Stylized Facts An inspection of a sample of 49 countries on which we were able to collect various data on VAT revenues, rates and bases reveals at least five interesting stylized facts about the VAT6. First, a worldwide average VAT rate of 14.4 percent generates 5.1 percent of GDP implying the average revenue productivity ratio7 of 0.35 percent of GDP: each one percentage point of VAT 4/ Apart from the good revenue potential, neutrality and efficiency are also the reasons for superiority of this tax in contrast to other common tax instruments such as a turnover tax. I/ See World Bank (1991). 6/ See table I in the appendix. 7/ See Tanzi (1993) p.18, table 1, which defines revenue productivity as the ratio between the VAT revenue and VAT rate. In that paper, Tanzi quotes Silvani's data on 22 single VAT rate countries with average revenue-to-GDP ratio of 4.3 percent, average rate of 11.6 percent, and the implied revenue productivity ratio of 0.37. Our sample extends Silvani's data to include countries with multiple VAT rates, resulting in somewhat lower overall revenue productivity ratio. 3 rate generates, on average, 0.35 percent of GDP revenue. Average revenue productivity ratio for multiple VAT rate countries (0.35 percent) is slightly lower thar the ratio for the single VAT rates countries (0.37). Therefore, the use of more than one rate does not help raise more revenu^- and, indeed, it seems to reduce the revenue performance of a VAT vis-a-vis countries that use single rates. Second, the most conmmonly used single VAT rate is 10 percent, u-ed in 9 countries representing about 20 percent of the total, followed by the rate of 15 percent, which is used in 3 countries. This is consistent with the recommendations of most tax advisors to adopt the VAT at a single rate between 10 and 20 percent'. Third, twenty five countries (constituting 50 percent of the total) have a single rate (ignoring the zero rate and/or exemption on some services, exports, etc). Seven of the remaining 25 used two rates; nine used three rates; and the remaining nine use more than three rates. These multiple rates offer a greater opportunity to fit tha VAT to various social and political ends. However, rate differentiation raises administrative and compliance costs which undermines the VAT revenue performance (see Tait, 1988). Fourth, the VAT revenue and revenue productivity varies significantly across countries. The minimum revenue productivity ratio of 0.044 is found in Guinea which uses a single VAT rate of 13.6 percent. On the other end of the spectrum is Israel with the ratio of 0.653 -- fifteen times higher than in Guinea - using a single rate of 15 percent. The quality ".d efiiciency of tax administration and the size of the base, the hall.narks of the Israeli VAT system, are important factors behind this performance. Fifth, following Cnossen (1991), the tax base for a VAT can be broadly c!assified into four categories: (i) all goods and services (G+S); (ii) goods and selected services (G+ST); (iii) goods only (G); (iv) consumer goods and capital goods (G+CG); or (v) consumer goods, selected services, and capital goods (G+ST+CG). Two thirds of the countries apply the broad base (G+S), while nine countries use goods and select services (G+ST) as the base. Evidently, there is a wide diversity in the size of the VAT base across countries, but the - neral preference towards a broad based VAT is clear. The above facts are the averages based on the existing practice of VAT. Deviations fromn these averages and the structures of the VAT in countries that use or are planning to introduce a VAT, is a subject of potentially fruitful research. This is the course we take in this paper, focussing on th. VAT revenue performance. III. Hypotheses Models and Results In this section we present the hypotheses and models used to test - using regression techniques - the main determinants of VAT revenue which are suggested by the studies and practice of VAT, and are consistent with the data. The recent contributions to the VAT literature (see Tait 1988, 1991; B/ World Bank (1991). 4 Gillis et al. 1990; Khalilzadeh-Shirazi et al., 1991) provided the theoretical basis for the following set of empiricaily testable hypotheses. Gillis et al (1990), in their survey of the VAT lessons, noted that VAT has developed a worldwide reputation as a governn,ents' ' money machine", as few other single tax instruments can mobilize as large revenue as a well designed and implemented VAT. The experiences of the 49 countries (see Table 1, appendix) show that, over the range of existing rates, with few exceptions, -le VAT revenue rises with the rate. Therefore, we posit a positive relationship between VAT rate (variable: RATE) and revenue. The 'comprehensive" VAT is typically levied on a broad base which includes all goods and services (G+S). However, countries vary in their coverage of the base, particularly with regard to the treatment of services. The negative impact of extensive exemptions on. the size of the base can be quite dramatic. For instance, Kay and Davis (1990) estimate, on the b4 is of a survey of 32 countries, that the complete exemption of all services excludes from the VAT base between 45 to 78 percent of a country's GDP. This, in turn, increases the pressure on the fiscal authorities to use higher rate to mriobilize a target VAT revenue from a smaller base. Also, it is evident that some countries that use almost identical rates experience very different revenue performance. One source of different performances is often the size or the coverage of the base. We therefore define the BASE variable which captures whether a VAT is levied on all goods and services or on some subset of such comprehensive base. The underlying hypothesis is that the wider the base, the smaller is the number of goods and services exempted, and the larger the VAT revenue-to-GDP ratio. A simple dummy variable is used to measure differences in the size of the base. It assumes the value of one if the base comprises all goods and services (G+S), and zero otherwise9. Tait (1988) pointed out that administrative and compliance costs rise dramatically as the number of VAT rates increase. Thus rate differentiation, through higher costs, may adversely affect revenue. Furthermore, single rate is almost always revenue superior to multiple rates with little rate dispersion. Therefore, when countries use more than one rate, rate dispersion tends to be substartial. Consequently, it may be that it is dispersion of rates, rather than the number of rates per se, that may adversely affect VAT revenue. To test this hypothesis we measure the absolute dispersion among countries' multiple VAT rates (variable RANGE), defined as the absolute difference between the highest and lowest VAT rate'°. 2/ The shortcoming is that the dummy variable does not capture the variance of less comprehensive bases, which can be quite wide. The ideal variable, for the construction of which our data are not adequate, would be the size of the base, as measure by a percent of GDP covered by the VAT base. A variable defined in this way would make it possible to estimate the relative revenue contributions of increases in VAT rates versus increases in the base through the expansion of base (i.e., elimination of existing exemptions). 1[/ We have also tried olher measures l^ dispersion including standard deviation. However, results were not satisfactory. Ozie reason c'J. ' that since standard deviation measures variation from a mean, in a small sample it may not display sufficient variation. However, when we simply used the number of VAT rates, instead of our RANGE variable, the results were very similar. Therefore, we postulate the following b-cneral model: increases in VAT revenue are due to increases in VAT rates and the coverage and size of the tax base, while rate differentiation z.lises costs and thus negatively affects revenue. That is: REV = F (RA7E', BASE', R4NGE) (1) where: REV variable is defined as VAT revenue as percentage of a country's GDP; RATE Is defined as VAT rate (in percent); BASE is defined as a VAT base variable taking the value of 1 if the base comprises -lI consumer goods and services (G + S), and 0 otherwise; and RANGE is a measure of rate dispersion, defined as the difference between highest and lowest VAT rates. The signs above the dependent variables denote the expected direction of influence. Due to missing observations, mostly on revenue, the original samnple of 49 countries had to be reduced to 34 countries for full sample models, 20 countries for single-late country models, and 14 countries, for multiple rate models. The choice of the year was 1988, lirgely based on the Silvani data as quoted by Tanzi (1993, p.18). We comnpiled additional data on revenue, rates and base from various World Bank and IMF sources (see Table I in the Appendix). We estimated linear versions of the general model (1) using the Ordinary-Least Squares (OLS) technique". The estimated models are given in Table 1 below. The mos. important determinant of VAT revenue is the VAT rate, which is significant at well below one percent, and with the expected sign in all estimated versions of the model (see Table 1) indicating the robustness of the estimates. The RATE variable dominates the results, with a coefficient of determination of 0.75 for the single rate VAT model and 0.57 for all countries model (see Table 1). A. The Sample of Single VAT Rate Countries A simplest estimated relationship between revenue and rate from the sample of 20 single VAT rate countries is depicted in Figure 1. For this group of countries, VAT generates, on average, 4.7 percent of GDP with an average rate of 11.3 percent: each I percent of VAT rate generates 0.50 percent of VAT revenue. The estimated VAT rate that will generate positive revenue, which can be interpreted as the minimum rate from the revenue perspective, is close to 2 percent (see Figure 1)". The RATE coefficient is strongly significant and, in the linear model, explains 71 percent of the variation in VAT revenues". The estimated relationship can be used to predict the potential VAT revenue in a country contemplating the introduction of a VAT, given a proposed single VAT rate. For example, following some other Eastern European countries (i.e., Hungary and former Czechoslovakia), Bulgaria is planning to introduce a VAT rate at the single rate of 18 percent. Given this rate, our model suggests 11/ The exponential form, also tried, does not seem to fit the data. 1/ It should be noted, however, that this is not to be interpreted as the rate at which VAT, as a tax instruments, breaks even, because its costs are not taken into account. 12/ Note that this a univariate model (not reported in Table 1) which excludes the BASE variable. 6 Value-Added Tax Re'venue, 1 988 Countries with Single VAT Rate ,0 - .~ is .. - o., asU 0 Q 7- >2 3 -- >4- 2 (1 _~ 0 2 £ 6 8 10 12 14 16 1i 20 22 Value-Added Tax Rate U Actual VAT Revenue - Predicted VAT Rev. Figure 1 that Bulgaria's VAT has the potential to mobilize approximately 8.1 percent of GDP in VAT revenues once the VAT system is in fully operational. Whiether this revenue potential will turn into actual revenue, however, will depend on other VAT characteristics. In particular, two caveats should be kept in mind when making similar predictions. First, although the estimated relationship is fairly robust, prediction assumes that other important characteristics of Bulgaria's VAT (i.e., size of the base or the number of exemptions) are equal or very close to the average characteristics of the countries in the sample. To the extent a country's proposed VAT features, for instance, larger number of exemptions (and, hence, smaller base), the predicted relationship on the basis of the above model will overestimate the VAT revenue potential; or, more precisely, the actual or expected revenues will be lower than the estimated potential. The second caveat is analogous but has to do with tax administration capacity: the lower this capacity than for an average country in the samrrple, the lower will be the potential revenues than predicted by the model. Nevertheless, the model provides a very rough picture of the order of magnitude when anticipating revenue potential of a new VAT. In the particular case of Bulgaria, it is probably safe to assume that the above potential will not be reached in the near future, since it takes time to make the VAT fully operational. Furthermore, it takes time to make Bulgaria's tax base and administrative capacity more in line with the average country in the sample. Assuming a realistic tax "gap" (ratio oI potential and actual 7 revenue) of around 65-75 percent, that country could probably be expected to raise between 5 and 6 percent of GDP in actual VAT revenues, when the VAT becomes fully functiotial". Extending the linear univariate model by including the base variable improves slightly the goodness of fit (by 4 percentage points), the RATE variable is still strongly significant, and the BASE coefficient has the anticipated positive sign, although it is only weakly significant at 10 percent level (see table 1). This is probably the result of the inability of the dummy variable to captule fully the diversity of tax bases in the sample. B. Full Sample of 34 Countries The regression results from the full sample of 34 countries (see first column in Table 1) give i VAT rate coetficient of 0.389, indicating that the VAT generates revenues of some 0.4 percent of GDP for every percentage point of the ratet4. As expected, VAT revenue as a percent of GDP rises with increases in coverage and size of the tax base: the estimated equation shows a positive and significant oase coefficient at the 6 percent level. The low significance can be attributed to the inadequacy of the available data as a measure of the true base. The base dummy used here has the right sign but does not fully capture the diversity of VAT tax, hence its low significance. For instance, the BASE data in the appendix fail to distinguish between the Israel and UK VAT regimes. Both countries have a single rate of 15 percent and both have a (G+S) VAT tax base. However, Israel enjoys a higher revenue of 9.8 percent of GDP compared to 6 percent in UK (also see Figure 1). The difference, however, does not reflect lower collection efficiency in the U.K vis-a-vis Israel. The difference in revenue is explained by the difference in the size of the bases: Israel is an exception in that its VAT base includes the financial 13/ In fact, there are several subcategories of tax "gaps" including: the gap between potential and declared taxes, between declared taxes and paid taxes, and between taxes paid and those that are received by the treasury. Examination of the nature and sources of these tax gaps is identified as a fruitful area of further research (see: Bird and Casanegra, 1992, p. 1I). 14/ The RATE variable for multmle rate countries uses the standard rate in all regressions. The weighted average rate for these countries would be clearly preferable, but was unavailable at the time of writing. A subsequent comparison of (unpublished) weighted average rates collected by Silvani and the standard rates for ten multiple rates countries reveals that weighted rates are, mainly, lower than the standard rates. This means that the results of oi.r regressions may underestimate somewhat the rate coefficient in the full sample model, and more so in the only multiple rate models, since lower weighted rates generating sarme revenues imply higher revenue productivity. However, it is not clear whether this bias affects the results significantly in the full sarnple model, which is dominated by the larger number of single rate countries (20 as opposed to 14 multiple rate countries). We are grateful to Carlos A. Silvani (International Monetary Fund) for these observations and for sharing his data, which suggested an obvious line of further research. Specifically, we plan to reestimate the above models as more weighted rates data become available and to explore, with Silvani, the -nultiple rate models and tax gap models in greater detail. 8 sector" (Gillis 1990, p. 84), while the U.K. VAT uses extfnsive zero-rating for many services (e.g., food, all housing and some clothing and services), which co'nld comprise 40 percent of the potentially broad VAT base. This means that whan making country comparisons of obFeved differences in revenue productivity, when rates are identical across countries, it is not possible to infer that one VAT is more "efficient" in terms of collection. Observed differences probably reflect largely, if not entirely, the differences in the size of the base which, in turn, ref.ects different distr.butional considerations built into the VAT systems. Table 1: Determinants of VAT RIevenues: Regression Results DEPENDENT (1) (2) (3) (4) VARIABLE: All countries Single rate Multiple Differences VAT revenue to GDP Countries Rates (2 - 3) (%) Countries Censtant -0.880 -1.746 -4.805 3.059 (-0.953) _ (-1. 907)0 (-1.909) (1.168) RATE 0.389 0.503 0.339 0.164 (6.293)a (7.519)- (3.025) (1.276) BASE 1.426 1.313 4.490 -3.177 (1.930)- (1.937)0 (2.641)b (-1.774)' RANGE -0.071 --- 0.021 -0.021 (-1.863)_ (0.383) (-0.394) R2 0.61 0.77 0.66 R2(adjusted) 0.57 0.75 0.55 F 15.85a 29.83a 6.35a n 34 20 14 Notes: rhe number in parentheses are t-statistics where stands tor signiticance at the Y% level, denotes significance at the 5 % level, and 0 stands for significance at 10 % level. The RANGE variable, as postulated, is negatively associated with the revenue, albeit weakly statistically significant, indicating that rate differentiation - 'es the tax costs and hence may generate losses in terms of revenue. The result seems to provide ea.ipirical support to Tait's argument that both official administrative costs and traders' compliance costs rise sharply as the number of rates multiply, leading to potential revenue losses. Interestingly, in terms of welfare, Ballard, Shoven and Whalley (1982) reached the same conclusion. 15/ Practical problems in including banking services in the base of the VAT have been quoted as the main reason that led EC countr;es to exempt this sector from the VAT base. 9 C. Do Countries With Sing;: VAT Rates MobUlize More Revenues? The countries included in the study represent a diverse group on a wide spectrum of VAT structures, bases and revenues. Ir tern's of VAT structure, two groups of countries are identified: single-rate and multiple-rate countries. Given the revenue performance data and the consensus preference of tax experts for si:'gle rates, we car ask a basic question: do exist:ng data on VAT support the contention that countries with single rate mobilize more revenue than those that i e multiple rates? Separate estimates for single and multipie rates country subgroups are given in Table 1. With the exception of the RANGE variable (which has to be dropped in the single VAT rate model), the results of both country groups support the hypotheses confirmed by the estimates for the full sample of countries. However, the magnitude of the estimated coefficients in the two sub-groups vary. We carried out a structural change test (Chow 1960) to test tie hypothesis that the regression coefficients, taken jointly, are equJ. This hypothesis is rejected at the 5 percent level of significance. Thus, the patern of VAT revenue statistically differs between the two sub-groups of countries. Of particular interest is the difference in' the revenue generation coefficient. For the single rate group, the VAT generat2s 0.503 of GDP for every percentage point of the rate, which is nearly 50 percent higher than the multiple rates group coefficient of 0.339. However, this difference is not statistically significant. The difference can not be exclusively explained in terms of VAT rates. This implies that different revenue performance in the two groups of countries are due not only to the use of single or mu!tiple VAT rates, but also to other factors, which probably include the VAT base and administrative capacity. Our empirical findings indicates that the difference in the base coefficient is weakly statistically significant (see column 4 in Table 1). Our relying on the a- ilable standard rates, rather than weighted rates da!a for multiple rate countries, makes the results of multiple rates models tentative. However, despite needed improvements, our analysis provides preliminary estimates which are consistent with the theoretical and practical wisdom on VAT, with many country case studies, and with the available data. As such, it is hoped to provide a stimulus and a benchmark for further empirical research on VAT. IV. Conclusion and Policy Implications We empirically analyzed determinants of VAT revenue on a sample of 20 single rate countries, a full sample of 34 countries and a small sample of multiple rate countries. The results of the regressions confoi.n to the conventional views on the key variables influencing the VAT revenue performance: the rate, the base, and the rate dispersion. The rate and the base coefficients are significant and with the expected positive sign in all of the estimated versions of the model, indicating their robustness. This robustness, particularly in the single rate models, makes it possible to use an estimated model, with appropriate caveats, to predict potential (and expected) VAT revenues in countries that are contemplating the introduction of a single rate VAT (e.g., Bulgaria). Also, the dispersion of rates is found to negatively affect VAT revenues. Empirical results confirm another conventional view that VAT generates, other things constant, higher revenue in single VAT rate countries than in multiple rates countries. The difference in the estimated models for the two country groups is statistically significant indicating a structural change. However, this change in the pattern of VAT revenues cannot be exclusively explained in 10 terms of different in rate structures. Satisfactory explanation, therefore, must include other factors such as the base and tax administration capacity. The impiications for policymakers are clear: to generate siuperior revenues, a VAT should be levied in a single rate on as broad base as possible; it also must be accompanied by a strong tax administration to ensure enforcement and compliance. 'A 11 Appendix Table 1. Value Added Tax Rates, Revenue, and Base, Worldwide (1988) Country VAT Revenue Standard VAT Tax Base' Other VAT Rates' as % of GDP' Rate %b __ Argentina 0.8 16.0 G+ST 25 Austria 6.1 20.0 G+S 10,32 Belgium 7.2 19 G+S 1,6,17,25, 33 Bolivia 2.6 10.0 G+S single rato Brazil NA 17.0 G+CG single rate Chile 8.8 16.0 G+S single rate Colombia NA 10.0 G+ST 4,6,15,20,35 Costa Rica 3.8 10.0 G+ST single rate Cotf d'Ivoire NA 25.0 G+SG 11.1,35.1 Denmark 9.5 22.0 G+S single rate Dominican Rep 1.6 6.0 G+ST+CG single rate Ecuador 2.5 10.0 G+ST single rate Finland 8.4 19.1 G+ST+CG single rate France 11.9 18.6 G+S 2.1,4,5.5, 22 Germany 3.8 14.0 G+S 7 Greece 8.9 18.0 G+S 3,8,36 Grenada 4.5 6.0 G+S single rate Guatemala 2.4 7.0 G+S single rate Guinea 0.6 13.6 NA single rate Haiti 1.8 10.0 G+S+CG single rate Honduras NA 7.0 G+ST 6 Hungary 9.1 25.0 G+S 15 Indonesia 4.5 10.0 G+ST single rate Ireland 8.5 21.0 G+S 2.2,3.3,12.5 Israel 9.8 15.0 G+S single rate Italy 5.8 19.0 G+S 38- Korea 3.3 10.0 G+S single rate 12 Luxembourg 6.8 12.0 a+s 3,6 Madagascar 1.5 15.0 G+S single rate Mauritius 2.1 5.0 G single rate Mexico 3.4 15.0 G+S 6,20 Morocco NA 9.0 G+S 7,12,14,19,30 Netherlands 8.0 18.5 G+S 6 New Zealand 6.7 12.5 G+S single rate Nicaragua 2.3 10.0 G+ST 6,25 Niger NA 25.0 G+S 15,35 Norway 9.4 20.0 G+ST single rate Panama 1.1 5.0 G+S single rate Peru 2.0 13.0 G+ST single rate Philippines NA 10.0 G+S single rate Portugal 6.6 17.0 G+S 8,30 Spain 4.6 12.0 G+S 6,33 Sweden 7.5 25.0 G+S single rate Taiwan 2.6 5 G+S single rate Togo NA 14.0 G+S 3- Tunisia 2.8 17.0 G+S 6,29 Turkey 1.7 12.0 G+S 1,6,8,20 UK 6.0 15.0 G+S single rate Uruguay 7.0 22.0 G+S 12 Average 5.1 14.4 ources: a. Government Finance Statistics, Yearbook, International Monetary Fund, 1992. The 1988 revenue values were used;b, c & *. Sijbren Cnossen (1991), 'Design of the Value Added Tax: Lessons from Experience' in J. Khalilzadeh- Shirazi and A. Shah (eds), Tav Policy in Developing Countries, A World Bank Symposium, Washington DC, 1991, pp 74-75. For countries with single VAT rates, revenue and rate data are taken from Silvani as quoted by Tanzi (1993, p.18); d. Alan Tait (1991), 'VAT Policy Issues: Structure, Regressivity, Inflation, and Exports" in Alan Tait (editor), Value-Added Tax: Adninistrative and Policy Issues, IMF Occasional Paper #88, Washington DC, 1991, pp 2-3. NA= not available. It was pointed out to us by Silvani and Casanegra de Jantscher that the revenue data for Argentina from GFS is probably an underestimate. We used published GFS data whenever possible for consistency purposes. 13 References Bird, Richard and Milka Casanegra de Janstcher (1992) (eds). Improving Tax Administration in Developing Countries, International Monetary Fund, Washington D.C. Ballard, Charles L., John B. Shoven, and John Whalley (1982). " The Welfare Cost of Distortions in the United States Tax System: A General Equilibrium rpproach, NBER Working Paper # 1043 (Cambridge, Massachusetts: National Bureau of Economic Research, Decembe.r 1982). Chow, G. (1960). "Tests of Equality Between Sets of Coefficients in Two Linear Regressions", Econometrica, Vol. 28, pp. 591-605. Cnossen, Sijbren (1991). "Design of the Value Added Tax: Lessons from Experience' in Khalilzadeh- Shirazi, Javad and Anwar Shah (1991)(eds). Tax Policy in Developing Countries, A World Bank Symposium, Washington D.C., pp 74-75. Due, John F. and Francis P. Greany (1992). "The Introduction of a Value Added Tax in Trinidad and Tobago", in Bird, Richard and Milka Casanegra de Janstcher (1992) (eds). Improving Tax Administration in Developing Countries, International Monetary Fund, Washington D.C., 168-199. Gillis, Malcolm (1990). "The VAT and Financial Services" in Gillis, Malcolm, Carl S. Shoup, and Gerardo P. Sicat (1990) (eds.). Value Added Taxation in Developing Countries. A World Bank Symposium, The World Bank, Washington D.C. pp. 83-94. Gillis, Malcolm, Carl S. Shoup, and Gerardo P. Sicat (1990) (eds.). Value Added Taxation in Developing Countries: A World Bank Symposium, The World Bank, Washington D.C. Goode, Richard (1993). "Tax Advice to Developing Countries: An Historical Survey", World Development, Vol.21, No.1, pp. 37-53. Khalilzadeh-Shirazi, Javad and Anwar Shah (1991) (eds.). Tax Policy in Developing Countries. A World Bank Symposium, The World Bank, Washington D.C. Kay, John A. and Evan H. Davis (1990). " The VAT and Services" in Gillis, Malcolm, Carl S. Shoup, and Gerardo P. Sicat (1990) (eds.). Value Added Taxation in Developing Countries. A World Bank Symposium, The World Bank, Washington D.C., pp. 72-82. Sandford, Cedric and Michael Godwin (1990). "VAT Administration and Compliance in Britain" in Gillis, Malcolm, Carl S. -'%oup, and Gerardo P. Sicat (1990) (eds.), Value Added Taxation in Developing Countries. A World Bank Symposium, World Bank, Washington D.C., pp. 207-215. Silvani, Carlos A. (1992). "Improving Tax Compliance" in Bird, Richard and Milka Casanegra de Jantscher (1992) (eds). Improving Tax Administration in Developing Countries, International Monetary Fund, Washington D.C., 274-305. 14 Tait, Alan A. (1991). 'VAT Policy Issues: Structure, Regressivity, Inflation, and Exports' in Tait, Alan A. (1990) (editor). Value-Added Tax: Administrative and Policy Issues, IMF Occasional Paper #88, International Monetary Fund, Washington D.C., pp. 2-3. Tait, Alan A. (1988). Value Added Tax: International Practice and Problems, Intemational Monetary Fund, IMF, Washington D.C.. Tanzi, Vito (1993). "A Primer on Tax Evasion", IMF Working Paper, Fiscal Affairs Department, International Monetary Fund, Washington D.C., March 1993. Tanzi, Vito (1993). Transition to Market: Studies in Fiscal Reform, International Monetary Fund, Washington D.C. World Bank (1991). Lessons of Tax Reforms, World Bank, Washington D.C. Polly Reaarch Working Paper Serie Contact Title Author Date for paper WPSI 178 Productivity of Public Spending, John Baffes September 1993 C. Jones Seoral Alkoaton Choices, and Anwar Shah 37699 Economic Growth WPSI 179 How the Market Transitn Affected BartlombJ Kaminski September 1993 P. Kokila Export Performance In the Central 33716 Europen Economies WPS1 180 The Financing and Taxation of U.S. Harry Huizinga September 1993 R Vo Direct Investmnnt Abroad 31047 WPS1 181 ReformIng Health Care: A Case for Zeljko Bogetic September 1993 F. Smith Stay-Well Health Insurance Dennis Heffley 36072 WPS1 182 Corporate Governance in Central Cheryl W. Gray September 1993 M. Berg and Eastern Europe: Lessons from Rebecca J. Hanson 31450 Advanced Market Economies WPSI 183 Who Would Vote for Inflation in Cheikh Kane September 1993 T. Holiestelle Brazil? An Integrated Framework Jacques Morisett 30968 Approach to Inflation and Income Distribution WPS1 184 Providing Social Benefits in Russia: Simon Commander September 1993 0. del Cid Redefining the Roles of Firms and Richard Jackman 35195 and Government WPS1 185 Reforming Hungarian Agricultural Morris E. Morkre September 1993 N. Artis Trade Policy: A Quantitative David G. Tarr 38004 Evaluation !PS1186 Recent Estimates of Capital Flight Stijn Claessens September 1993 R. Vo David Naud6 31047 WPSI 187 How Should Sovereign Debtors Andrew Warner September 1993 J. Queen Restructure Their Debts? Fixed 33740 Interest Rates, Flexible Interest Rates, or Inflation-indexed WPS1 188 Developmentalism, Socialism, and Mario Marcel September 1993 S. Florez Free Market Reform: Three Decades Andr6s Solimano 39075 of Income Distribution in Chile WPS1 189 Can Communist Economies Alan Gelb September 1993 PRDTM Transform Incrementally? China's Gary Jefferson 37471 Experience lnderjit Singh WPS1 190 The Govemment's Role in Japanese Yoon Je Cho September 1993 T. Ishibe and Korean Credit Markets: A New Thomas Hellmann 37665 Institutional Economics Perspective Policy Resarch Working Paper Series Contact Title Author Date for paper WPSI9I Rent-Sharing In the Muti-Fibre Geoffrey J. Bannister September 1993 A. Daruwala Arrangement: The Case of Mexico 33713 WPSI 192 Effects of Tax Reform on Argentina's Jacques Morisset September 1993 G. Carter Revenues Alejandro lzqulerdo 30603 WPS1 193 The Armenian Labor Market In Milan Vodoplvec September 1993 S. Florez Transition: Issues and Options Wayne Vroman 39075 WPS1 194 How Fast Has Chines Industry Tom Rawski September 1993 E. Khine Grown? 37471 WPS1 195 The Entvrprise Sector and Mark Schaffer September 1993 E. Khine Emergence of the Polish Fiscal 37471 Crisis, 1990-91 WPS1 196 Corporate Tax Structure and Jeffrey Bernstein September 1993 C. Jones Production Anwar Shah 37699 WPS1 197 Determinants of Inflation among Bruno Boccara September 1993 C. Jones Franc Zone Countries in Africa Shantayanan Devarajan 37699 WPS1 198 Enterprise Reform in China: The Natalie Lichtenstein September 1993 M. Rangarajan Evolving Legal Framework 81710 WPS1 199 Public Pension Governance and Olivia Mitchell October 1993 D. Evans Performance: Lessons for Developing 37496 Countries WPS1200 The Life-Cycle Distributional Jane Falkingham October 1993 D. Evans Consequences of Pay-As-You-Go Paul Johnson 37496 and Funded Pension Systems WPS1 201 Five Criteria for Choosing among Margaret E. Grosh October 1993 M. Quintero Poverty Programs 37792 WPS1202 Pnvatization and Foreign Investment Frank Sader October 1993 Rose Vo in the Developing World, 1988-92 31047 WPS1203 Determinants of Value-Added Tax Zeliko Bogetic Octobtr 1993 F. Smith Revenue: A Cross-Section Analysis Fareed Hassan 36072 WPS1204 Structural Adjustment, Economic Nisha Agrawal October 1993 K. Rivera Performance, and Aid Dependency Zafar Ahmed 34141 in Tanzania Michael Mered Roger Nord