Tax Reforms in Singapore H. M. Leung, L. Low, and M. H. Toh, National University of Singapore This paper takes the view that tax reforms in the form of a reshuffling of marginal tax rates are more appropriate policy recommendations than a whole-sale restructuring of a country’s tax system. We adopted a simple model that yields simple and intuitive recommendations on the directions of such tax reforms. This model is estimated using the 1987/ 88 Singapore Household Expenditure Survey. The result also suggests that the recently introduced Goods and Services Tax may be an efficient reform, especially if the government puts limited value on inequality.  1999 Society for Policy Modeling. Published by Elsevier Science Inc. 1. INTRODUCTION Responding to its changing industrial structure and overall economic development, Singapore has gone through a major tax reform. Similar to the experience of many developing countries, the long-run trend has been a movement away from import duties and direct taxes, and towards domestic indirect taxes (see Burgess and Stern, 1993, for a survey). As part of a more general appraisal of such a restructuring of the tax system, we ask in this paper what constitutes an efficient indirect tax system, and suggest directions for improvement to the tax system in Singapore. The intuitive definition of an efficient indirect tax system is, in fact, quite simple. If a tax system is efficient, there cannot exist any reshuffling of its rate structure that improves social welfare. The idea of reshuffling of tax rates comes from the tradition of tax reforms (Feldstein, 1976; Atkinson and Stiglitz, 1980; Newbery and Stern, 1987). The literature on an overall-optimal tax system has been criticized as too unrealistic and impractical at least for policy purposes. The advantage of our definition of efficiency Address correspondence to M.H. Toh, National University of Singapore, Faculty of Business Administration Department of Business Policy, Singapore 0511, Singapore. Received March 1997; final draft accepted December 1997. Journal of Policy Modeling 21(5):607–617 (1999)  1999 Society for Policy Modeling Published by Elsevier Science Inc. 0161-8938/99/–see front matter PII S0161-8938(98)00019-2 608 H. M. Leung, L. Low, and M. H. Toh given above is that when a welfare-improving reshuffling of rates is found, the policy implication is immediate. In general, there are two alternative ways of measuring social welfare improvements. The first one seeks to measure the aggregate welfare change of different households affected by the tax reform, holding the total tax revenue unchanged. The other works in the opposite way, and looks for the change in tax revenue as a result of tax reforms, holding households’ utility, or welfare, unchanged. The two approaches are equivalent, and we somewhat arbitrarily adopt the first in what follows. The intuition of reshuffling tax rates is also quite simple. For instance, we may contemplate raising the tax rate for transportation and lowering the tax rate for food, in such a way as to keep total government tax revenue unchanged. Some households gain and the other lose. Welfare improves if the net result is positive. One problem encountered here is the comparison of different households’ welfare. This is made more precise in the model we seek to develop in the next section. In Section 3 we introduce the data set from Singapore and the procedure to estimate the set of demand elasticities necessary for evaluating the tax system. Section 4 presents and discusses the results. Section 5 concludes the essay. 2. A MODEL OF TAX REFORMS1 To begin with, we assume that producer prices are fixed as tax rates change. The easiest way to justify this is that for small open economy such as Singapore prices are fixed internationally. This simplifies the picture substantially as any (marginal) tax change can be seen as an add-on to the price. In other words, the specific tax (in dollar terms) can be treated as equivalent to a change in price. We think of households as the unit of consumption. Central to the model is the way different households respond to changes in tax rates. Such changes in household consumption directly affect the revenue collected by the government. Revenue from indirect taxes is (Eq. 1) n R5 o tjxj, j (1) 51 1 The model adopted in this section follows that developed in Ahmad, Leung, and Stern (1984). TAX REFORMS IN SINGAPORE 609 where n is the number of commodities consumed, tj is per unit specific tax on good j, and xj is aggregate consumption of good j over all households h 5 1, ... H. Let the indirect utility function of the hth household be uh(xh), and the indirect utility function be vh(p, Mh), where Mh is the lump-sum income of the household. Social welfare is specified as a function of individual utility levels: w(u1, u2, ..., uH), or it can be written as a function of income and prices (Eq. 2) V(p, M1, ..., MH) 5 w[v1(p, M1), ..., vH(p, MH)]. (2) For a welfare-improving tax reform, it must either increase R while keeping w unchanged, or in the case to be modeled here, it must increase w while keeping R unchanged. For tax reforms, it is useful to think of the welfare change arising from an extra dollar of tax revenue. Call this li when the revenue is collected from good i. Equation 3: li 5 ]V ]R / . ]ti ]ti (3) If li is greater than lj, we can raise an extra dollar from good j and reduce a dollar collected from good i to improve social welfare, If ti is raised, li would be negative. From Equation 2 we have (Eq. 4) ]v 5 ]ti H o h 51 ]W h h a xi 5 ]vh H o h bhxih, (4) 51 where bh is the social marginal utility of income to the hth household. From Equation 1, (Eq. 5, 6) ]R 5 xi 1 ]ti n ]x o tj ]pj. j 51 (5) i H li 5 o bh xjh h 1 5 n xi 1 o j51 . ]xj tj ]pi (6) This can easily be put into elasticity terms (Eq. 7) oh bh pi xih li 5 , pi xi 1 o (tj/pj) pj xj jji j (7) where jij is the uncompensated price elasticity of good j with respect to good i. This formulation is also expressed in expenditure 610 H. M. Leung, L. Low, and M. H. Toh terms, which is easy for estimation purposes. The vector tj/pj are existing tax rates. To operationalize the welfare weights we choose an exponential function of relative household incomes (Eq. 8) 1 e bh 5 1MM 2 , h (8) where M1 and Mh are the per capita income of the poorest household and the hth household, respectively. The power e may be interpreted as the degree of inequality aversion. A government indifferent to income inequality will have e 5 0, and a higher e represents greater inequality aversion. We will experiment with a range of different values of e to encapsulate different welfare weights. Given any e, li of Equation 7 can be ranked. A commodity with the lowest l is the least costly to tax, and should be given the priority as far as additional tax revenue is concerned. Apart from the rankings, it is interesting to study the degree of variation between the ls. By definition, a higher degree of variation implies greater gain from taxing certain commodities more strongly than others, thus favoring policy recommendations for differential commodity taxes. A low degree of variation in the ls would, on the other hand, imply that the scope for reshuffling individual tax rates is limited. The policy recommendation in that case would be to change all commodity tax rates uniformly in the manner of a Value-Added Tax, or of a Goods-And-Services Tax, as it was adopted in Singapore in 1994. 3. THE DATA AND ELASTICITY ESTIMATIONS In this section, following the approach of Lluch (1973) and Powell (1973), parameters of the linear expenditure system (LES) are estimated from cross-sectional information contained in the household expenditure survey. They are then used to obtain the price elasticities required in the computation of the li in Equation 7. The demand equation for the ith commodity in the LES is written as (Eq. 9) vi 5 pigi 1 bi(z 2 S), (9) where vi 5 pixi is the household’s expenditure on good i, S 5 Spjgj is the subsistence expenditure, and Z is the total budget. The parameters in Equation 9 satisfy 0 , bi , 1, and Sbi 5 1. TAX REFORMS IN SINGAPORE 611 From the system of equations in Equation 9, the income, ownprice, and cross-price elasticities are given by wi, jii, and jij (i ≠ j) below (Eq. 10): wi 5 bi(z/vi) jii 5 (1 2 bi) (pigi/vi) 2 1 jij 5 2bipjgj/vi for i ? j. (10) Because bi is non-negative, income elasticities are non-negative, implying that all goods are normal. Furthermore, as 0 , bi , 1 and vi . pigi, all own price elasticities are negative. Similarly, all uncompensated cross price elasticities are negative. The LES excludes inferior goods and goods that are gross substitutes. This limitation can be overcome in empirical application by choosing broad enough groups of commodities. In the Singapore context, the LES is estimated using the information available from the Household Expenditure Survey conducted in 1987/88. In the survey, initially there were 7,700 private households, stratified according to the ethnic composition in Singapore. The households covered were divided into approximately 12 equal groups and each group was surveyed for 1 month. The spread of the survey over 12 months has the advantage of lightening the burden of respondents and interviewing staff. It also ensured that seasonal variations in expenditure due to festivities during the year were taken into consideration. To minimize the theoretical limitation of the LES mentioned above, the expenditure on commodities by each household is sufficiently broadly divided to include only ten major items: food, clothing, housing, furniture, medical services, transport and communications, recreation, education, overseas vacation travel, and others. The estimated parameters of the LES function for households in Singapore based on the 1987/88 Household Expenditure Survey (Department of Statistics, 1990) are shown in Table 1. The income and price elasticities derived using the information in Table 1 are presented in Tables 2 and 3. 4. ESTIMATION RESULTS Using the estimates just obtained, we now seek to calculate the values of ls given by Equation 7 in Section 2. In the denominator of Equation 7, the terms pixi are sample-average consumptions of good i; similarly, for pjxj. As for the variables indexed by h in the 612 H. M. Leung, L. Low, and M. H. Toh Table 1: LES Parameters Food Cloth Rent Furn Medic TPT Recr Educ Tour Misc bi s.e. (b) t-stat gi se (gi) t-stat 0.1452 0.0501 0.1559 0.0462 0.0445 0.3090 0.0268 0.0448 0.0529 0.1245 0.053 0.026 0.055 0.018 0.009 0.087 0.014 0.031 0.028 0.032 2.764 1.900 2.827 2.624 5.146 3.563 1.875 1.462 1.892 3.901 102.0 9.7 24.8 3.6 1.9 4.3 6.4 8.9 1.6 18.4 12.035 3.527 3.488 1.114 0.547 2.403 1.939 1.940 0.834 2.020 8.472 2.757 7.124 3.228 3.452 1.779 3.279 4.581 1.891 9.108 numerator, we stratified households into 10 income groups. The stratification criteria are shown in the first column of Table 4. Average income, saving, expenditure, as well as estimated subsistence expenditure for these households are given in columns 2 to 5. Correspondingly we have 10 welfare weights to work with. For each (M1/Mh)e we experimented with five es in ascending values of 0, 0.5, 1, 2, and 5. The choice of such values is, of course, somewhat arbitrary, and in principle we can use any other weights as the government sees fit. The value of e 5 0, however, carries the meaning of zero inequality aversion and can be used as a reference point. It reflects the value of a government who does not at all care for inequality, or one who thinks that only economic Table 2: Expenditure Share, Income and Own-Price Elasticities Food Cloth Rent Furn Medic TPT Recr Educ Tour Misc Expend share Expend elasticity Own-price elasticity Compensated own-price elasticity 0.3751 0.0520 0.1454 0.0316 0.0257 0.1515 0.0313 0.0471 0.0285 0.1118 0.3872 0.9636 1.0719 1.4602 1.7324 2.0399 0.8551 0.9517 1.8562 1.1144 20.2995 20.4920 20.5935 20.7200 20.8459 21.0099 20.4284 20.4836 20.9046 20.5965 20.1542 20.4418 20.4376 20.6739 20.8014 20.7009 20.4016 20.4388 20.8517 20.4720 Food Cloth Rent Furn Medic TPT Recr Educ Tour Misc Food Cloth Rent Furn Medic TPT Recr Educ Tour Misc 20.299 20.280 20.311 20.424 20.503 20.593 20.248 20.276 20.539 20.324 20.011 20.492 20.030 20.040 20.048 20.057 20.024 20.026 20.051 20.031 20.027 20.068 20.594 20.103 20.123 20.144 20.061 20.067 20.131 20.079 20.004 20.010 20.011 20.720 20.018 20.021 20.009 20.010 20.019 20.011 20.002 20.005 20.006 20.008 20.846 20.011 20.005 20.005 20.010 20.006 20.005 20.012 20.013 20.018 20.021 21.010 20.010 20.012 20.023 20.014 20.007 20.017 20.019 20.026 20.031 20.037 20.428 20.017 20.034 20.020 20.010 20.024 20.027 20.037 20.044 20.052 20.022 20.484 20.047 20.028 20.002 20.004 20.005 20.007 20.008 20.009 20.004 20.004 20.905 20.005 20.020 20.051 20.056 20.077 20.091 20.107 20.045 20.050 20.097 20.597 TAX REFORMS IN SINGAPORE Table 3: Uncompensated Cross-Price Elasticities 613 614 H. M. Leung, L. Low, and M. H. Toh Table 4: Income, Expenditure, and Saving by Income Class 1 2 3 4 5 6 7 8 9 10 a Income class AVG HH income AVG HH saving AVG HH expend Subsistence expenda L.T. 750 750–1000 1500–2000 2000–3000 3000–4000 4000–5000 5000–6000 6000–7000 7000–9000 G.T. 9000 628.6 944.6 1354.7 1960.3 2715.2 3540.3 4582.0 6165.1 8167.6 13265.4 119.4 193.2 302.2 485.3 742.6 1056.6 1496.1 2235.6 3255.0 6266.7 509.3 751.7 1052.5 1475.0 1972.6 2483.8 3085.9 3929.6 4912.7 6998.7 84.4 162.4 269.7 437.8 656.5 890.8 1167.4 1559.4 2041.2 2922.8 Subsistence expenditure computed from linear expenditure system. efficiency counts. The only remaining variable is (ti/pi), which we take simply as the prevailing tax rates.2 The result is presented in Table 5. In Table 5, a column of rankings accompanies each column of li. The good with the highest ranking, such as food, has the highest l and, thus, the highest social cost per dollar of revenue collected. Note that food has the highest ranking for all of the e values selected. We, therefore, conclude that whenever a tax cut is contemplated, food items should be the prime candidate, irrespective of the government’s view on inequality. It is easy to see why this is so. On the one hand, poorer households spend larger proportions of their income on food, which should, therefore, be subject to less tax if one ascribes to an equalitarian view. On the other hand, food typically has the lowest own price elasticity of demand (true to our sample, see Table 2 above where it is -0.2995). There are little substitution possibilities if food is taxed, precipitating little dead-weight welfare loss. The high ranking of food means that the second consideration dominates the first. In contrast, a commodity bearing the lowest l, such as furniture in Table 5 when e is zero, has the lowest social welfare cost per dollar of revenue collected. Furniture should be the prime 2 Strictly speaking, a vector of effective tax rates could be calculated by taking account of excise duties, and any subsidies on imported and exported goods. We did not embark on such an exercise because for an open economy like Singapore, the difference is likely to be insignificant. e5 li 0 Rank li 0.5 Rank li 1 Rank li 2 Rank li 5 Rank Food Cloth Rent Furn Medic TPT Recr Educ Tour Misc 1.045 1.034 1.032 1.031 1.032 1.032 1.036 1.035 1.032 1.033 1 4 7 10 9 6 2 3 8 5 0.477 0.401 0.399 0.363 0.349 0.338 0.412 0.403 0.351 0.390 1 4 5 7 9 10 2 3 8 6 0.259 0.178 0.196 0.149 0.141 0.133 0.189 0.187 0.135 0.179 1 6 2 7 8 10 3 4 9 5 0.116 0.052 0.087 0.041 0.040 0.038 0.059 0.064 0.030 0.063 1 6 2 7 8 9 5 3 10 4 0.048 0.011 0.045 0.009 0.010 0.012 0.013 0.020 0.004 0.021 1 2 3 4 5 6 7 8 9 10 TAX REFORMS IN SINGAPORE Table 5: Computed Values of li for Different Values of e 615 616 H. M. Leung, L. Low, and M. H. Toh candidate for tax increases if extra revenue is required. But unlike food, the commodity that has the lowest l changes when different values of e is used. It changes from furniture to transportation, tourism, and the miscellaneous category when e is raised from zero to five. Our recommendation changes according to the government’s value on inequality aversion. The meaning of the other ls is self-explanatory. Now we turn to examine the degree of variation in each set of ls. First, such variation in each column of ls is limited. This is, of course, comforting, for otherwise the existing indirect tax system would have been much less efficient. Second, the variation in ls rises significantly with e. To see this, calculate the difference between the highest and the lowest l as a percentage of the highest l, we get values of 1.3, 25, 43, 69, and 90 percent for e 5 0, 0.5, 1, 2, and 5, respectively. The resulting policy recommendation is that a uniform rate VAT or GST is a more palatable reform when e is low. This is hardly surprising. We know that a VAT or GST is regressive. It will be considered a good policy is a low social value is put on equality. 5. SUMMARY AND CONCLUSIONS This paper takes the view that tax reforms in the form of a reshuffling of marginal tax rates are more appropriate policy recommendations than a wholesale restructuring of a country’s tax system. We adopted a simple model that yields simple and intuitive recommendations on the directions of such tax reforms. We also believe that as a by-product of this model, we could derive some simple measure as to the desirability introducing a uniform indirect tax such as the Goods and Services Tax recently introduced in Singapore. This paper also estimated the model just described using the 1987/88 Singapore Household Expenditure Survey. It is found that food items should be the prime candidate for indirect tax cuts, and this is true even for governments with significantly differing inequality aversion. The commodities that are prime candidates for tax increases vary according to the government’s degree of inequality aversion. The policy recommendations are summarized by a single numerical measure, l. Because the variance of ls are quite small; we think this suggests that a Goods and Services Tax is not grossly inefficient. Furthermore, because such variance decreases with TAX REFORMS IN SINGAPORE 617 government’s inequality aversion, the Goods and Services Tax is more appropriate if the government has low degrees of inequality aversion. REFERENCES Ahmad, A., Leung, H.M., and Stern N.H. (1984) Demand Response and the Reform of Indirect Taxes in Pakistan, DERC working paper #50, Warwick University. Atkinson, A., and Stiglitz J. (1980) Lectures on Public Economics. New York: McGrawHill. Burgess, R., and Stern N.H. (1993) Taxation and Development. Journal of Economic Literature 16:762–830. Department of Statistics. (1990) Report on the Household Expenditure Survey 1987/88, Singapore: Ministry of Trade and Industry. Feldstein, M. (1976) On the Theory of Tax Reform. Journal of Public Economics 6:77–104. Lluch, C (1973) The Extended Linear Expenditure System. European Economic Review 15:21–32. Newbery, D., and Stern N.H. (1987) The Theory of Commodity Taxation for Developing Countries. New York: Oxford U. Press. Powell, A.A. (1973) Estimation of Lluch’s Extended Linear Expenditure System from Cross Section Data. Australian Journal of Statistics, 15.