I. A Modified VAT System for Bangladesh

The Government of Bangladesh introduced a VAT at the manufacturing/import stage on July 1, 1991.⁴ Under this new system, the excise duty on domestic production at the production stage and sales tax on imports at the import stage were replaced by a uniform VAT. The change was intended to be revenue neutral, with some exemptions of agricultural and service sectors and zero rating for exports. On domestic production, the new system of indirect taxation is characterized by a basic VAT and supplementary excise taxes on some luxury goods and energy products. On imports, in addition to a basic VAT, there are also customs duties to protect certain economic activities. The basic rate of the VAT was set at 15 percent.

II. Evaluating the Equity Impact of Tax Reform: A Macro Framework

One approach to assessing the impact of a tax reform (e.g., introduction of a VAT) on a developing country is to use the computable general equilibrium (CGE) modeling framework (Shoven and Whalley (1984)). There have been several attempts to model the Bangladesh economy and assess the impact of tax reform using the CGE framework. The most recent study, by Mansur and Khondker (1991), looks at the distributional impact of various tax expenditure policy options, including the VAT, in Bangladesh.⁵ However, CGE models are usually very restrictive in their representation of a developing country and are inappropriate for assessing the impact of tax policy reforms when the analysis of policy impact is more limited, for example, to income distribution.⁶

An alternative approach is suggested by Ahmad and Stern (1987). It uses detailed household expenditure survey (HES) data to identify winners and losers under a policy reform in a population characterized by income inequality. In this approach, the welfare impact of a tax reform is evaluated by specifying an estimated demand/supply response as well as an associated indirect utility function. It uses an exact money measure of welfare change, namely, “equivalent” or “compensating” variation. It also enables one to look at the revenue impact either by identifying the policy package that is revenue neutral or by looking at the trade–off between revenue gain/loss and welfare loss/gain. Policymakers may find the information provided by this approach very useful and easy to follow; however, in comparison with the CGE modeling approach. it uses more detailed data on the household and welfare side, while the production side of the economy is relegated to the background (Stern (1987b)). The basic elements of the Ahmad and Stern (1987) method are discussed below.

The Ahmad and Stern method specifies that the tax element in the final consumer prices is defined by “effective” taxes that capture the cascading effect of all input taxes in the economy so that $t^e=t^{\prime }{(I-A)}^{-1},$ where t^eand t are the effective and nominal tax vector, respectively; A is the input– output flow matrix; and I is the identity matrix. With the fixed producer price vector, p, the final consumer price vector is given by q = p + t^e.⁷

Now the pre‐ post‐reform tax rates are denoted by t^e0 and t^e1, with (q⁰ and q¹ representing the two associated purchaser prices; the pre‐ and post‐reform social welfare and revenue can similarly be represented by W⁰, W¹and R⁰, R¹. One way to approach the problem is to identify changes from t^e0 to t^e1 that yield W¹ > W⁰ and R¹ ≥R⁰(i.e., post‐tax revenue should be at least as high as the pre‐tax revenue).

To avoid any reference to the controversial social welfare function, we calculate the indirect utility levels using the expenditure function approach corresponding to pre‐ and post‐reform situations, i.e., V^h0 and V^h1 for each household (or household group), and thus assess the positive or negative impact on each household. One can express the utility change for household h or (V^h1 — V^h0) by the “equivalent variation” measure, which is defined by the following implicit equation:

Here, M^h0 is the pre‐reform income of the household. Thus, $E_{01}^h$is simply the amount of money that would need to be given to household h, if the pre‐reform prices were ruling, to allow it to reach the post‐reform utility level.$E_{01}^h$can also be defined making use of the explicit expenditure function e^h (q, U^h):

Thus, $E_{01}^h$is a money measure of the benefit (loss) of the reform to household h.⁸

This measure has been used in the Bangladesh context to assess the impact of nonmarginal tax reforms like introducing a specific VAT scheme. One could conceive of many different VAT schemes, but, for the purpose of this paper, they can be broadly classified into two basic schemes. The first is a completely uniform VAT applied to all goods (defined to include services); the second scheme consists of a VAT at a uniform rate but applying only to a subset of goods with a zero rating (and/or exemption) for other goods (and possibly some supplementary excise taxes that could go along with the VAT). The latter form closely resembles the Bangladesh case. Here, for the present purpose, a uniform VAT is simply defined as a tax system that makes the proportion of effective tax in the price of final goods the same for all goods. However, as a variation, it is also possible to consider the case of differential VAT rates applied to different goods. If the VAT rate is uniform at the level “r,” this implies that

for all i. The rate r will be determined by the Government’s revenue requirement. Thus, if $\overline{R}$ is the amount of revenue that needs to be raised from total consumer expenditure, then

For fixed revenue, $R^1=R^0=\overline{R}\mathrm{.}$If one assumes total consumer expenditure is unchanged by the reform, then revenue‐neutral r may be simply calculated by dividing $\overline{R}$by the total pre‐reform consumer expenditure.⁹

III. Distributional Impact of a VAT: An Empirical Exercise

Before the introduction of the VAT, there were three basic categories of indirect taxes in Bangladesh: excise taxes, import duties, and the sales tax. The excise taxes were imposed entirely on domestic production, import duties, on imports, and the sales tax, on both domestic sales and imports. In practice, though, almost the entire sales tax revenue came from the taxation of imports.

As the first step of the empirical exercise, the “nominal” and “effective” tax rates for 47 commodity groups were computed, using the 47– sector input–output and tax data for fiscal year 1984/85. To use the resulting estimates for the incidence analysis, a commodity classification, consisting of 15 commodity groups (9 food items and 6 nonfood items), was used. The choice of the commodity classification was governed by the convenience of estimating a complete demand system—the modified LES—for different household groups, making use of household expenditure survey data for Bangladesh. In the second step, the weighted average nominal, effective, and shadow consumption tax rates as well as revenue shares and budget (expenditure) shares for these 15 commodity groups were computed. The third step was to estimate the modified LES for 15 commodity groups (for all the population as well as each of 6 urban and 6 rural household groups) and derive expenditure elasticities and own price elasticities for the commodity groups from the LES parameters. The results are presented in the Appendix in Tables Al and A2.¹⁰

The effective taxes capture the cascading effects of taxes on inputs, of taxes on inputs into those inputs, and so on, and are higher than the nominal taxes. The divergence between nominal and effective tax rates sometimes measures the unintended consequences of government policy.Table Al shows that there are four commodity groups for which t_diff measures (differences between effective and nominal tax rates) are greater than 10 percent: other food (8), clothing (10), manufactured goods (11), and commercial energy (13).

Another critical fiscal indicator in a distorted economy is shadow consumption tax. With shadow prices expressed as accounting ratios (v/q or the ratio of border prices to domestic purchase prices), the shadow consumption tax is simply 1 — vlq. Derived from the shadow prices, these shadow consumption taxes (when normalized) are particularly helpful in assessing how much a good is taxed or subsidized when measured at its shadow price in a distorted economy. For example, as shown in Table A1, commercial energy (13) faces a high effective tax (42.5 percent) when measured at domestic market prices, but it turns out to be highly subsidized (at 36 percent) when the shadow consumption tax measure is used.

As the final step of the empirical exercise, the net indirect tax element in the total consumer expenditure was estimated. For Bangladesh, it was found that the net indirect tax revenue averaged approximately 5.6 percent of total consumer expenditure for the fiscal year 1984/85, for which detailed tax, input‐output, and other data could be gathered.¹¹

The first tax reform option examined here is to replace all taxes with a uniform (proportional) VAT of approximately 5.6 percent of the tax inclusive prices of all goods.¹² Since this is a sizable reform package, one would expect substantial changes in consumption demand. However, since the impact of a uniform (proportional) tax on revenue from all expenditures is being examined here, it is reasonable to assume that if total expenditure is unchanged, then the given VAT will raise the required revenue.

To estimate the impact of the introduction of the VAT on household income, it is assumed that consumer preferences could be represented by a modified LES. Given the target revenue neutral VAT rate, the task is to estimate the equivalent variation $E_{01,}^h$, defined for each per capita expenditure group (with expenditure pattern generated by LES),

where M^h is the per capita expenditure level of household group h; q⁰ and q^l are the pre‐ and post‐tax prices; b, is the marginal budget share of commodity group i and, along with a (which is a constant), is determined by the modified LES estimated for each household group. The parameters b, and a have been derived from the econometric estimation of the LES demand system for 12 urban and rural household groups, using the detailed household expenditure survey data. What follows is a brief discussion of the development of the database and an estimation of the LES.

Results of the Exercise: Distributional Impact of the VAT

The equivalent variations as proportions of the per capita household group expenditure, i.e., $(E_{01}^h/M^h)$estimates in percentage terms for rural and urban areas (associated with a uniform VAT), are presented in Table 1. The impact of introducing a VAT on household income examined here is measured relative to the pre‐reform tax regime.¹⁶ The results indicate that a uniform (proportional) VAT would harm four urban and four rural groups that are relatively poor and form a large majority of the population. On the other hand, it would benefit two urban and two rural groups that are relatively rich and form a small minority of population. The poorest urban households incur a loss equivalent to a 3.5 percent decrease in their total expenditure while the richest urban households benefit significantly—equivalent to an 8.1 percent increase in their total expenditure. In rural areas, the magnitudes of gain or loss are smaller but the pattern is similar: the poorest households lose the most—about 3.2 percent of their total expenditure, while the richest rural households benefit the most‐about 6.1 percent.

Table 1.

Equivalent Variation $\left(E_{01}^h\right)$ for Uniform VAT

Sources: Bangladesh household expenditure survey data, 1985/86; and author’s estimates.

Note:$E_{01}^h$ and M^hare defined in equation (5).

Table 1.

Equivalent Variation $\left(E_{01}^h\right)$ for Uniform VAT

Group	Per capita expenditure (taka/month)	Percent of household	Sample size	Mean household size	Mean per capita expenditure (Mh)	Equivalent variation (percent) $(E_{01}^h/M^h)$
Urban
1	<249	1.1	42	7.2	210.9	–3.5
2	250–349	2.4	82	7.1	304.8	–3.3
3	350–449	2.4	70	6.5	401.4	–2.2
4	450–599	2.9	108	5.9	523.6	–2.8
5	600–699	1.2	52	6.2	641.8	+4,9
6	700+	4.0	146	5.4	1092.4	+8.1
Rural
1	<249	18.2	219	6.2	201.7	–3.2
2	250–349	27.9	316	6.3	298.3	–2.0
3	350–449	19.4	223	5.6	399.3	–1.9
4	450–599	12.4	149	5.6	504.9	–3.1
5	600–699	3.5	42	4.9	644.4	+4.5
6	700+	4.7	51	5.3	944.7	+6.1
All		100.0	1500	5.9	427.6

Sources: Bangladesh household expenditure survey data, 1985/86; and author’s estimates.

Note:$E_{01}^h$ and M^hare defined in equation (5).

View Table

Table 1.

Equivalent Variation $\left(E_{01}^h\right)$ for Uniform VAT

Group	Per capita expenditure (taka/month)	Percent of household	Sample size	Mean household size	Mean per capita expenditure (Mh)	Equivalent variation (percent) $(E_{01}^h/M^h)$
Urban
1	<249	1.1	42	7.2	210.9	–3.5
2	250–349	2.4	82	7.1	304.8	–3.3
3	350–449	2.4	70	6.5	401.4	–2.2
4	450–599	2.9	108	5.9	523.6	–2.8
5	600–699	1.2	52	6.2	641.8	+4,9
6	700+	4.0	146	5.4	1092.4	+8.1
Rural
1	<249	18.2	219	6.2	201.7	–3.2
2	250–349	27.9	316	6.3	298.3	–2.0
3	350–449	19.4	223	5.6	399.3	–1.9
4	450–599	12.4	149	5.6	504.9	–3.1
5	600–699	3.5	42	4.9	644.4	+4.5
6	700+	4.7	51	5.3	944.7	+6.1
All		100.0	1500	5.9	427.6

Sources: Bangladesh household expenditure survey data, 1985/86; and author’s estimates.

Note:$E_{01}^h$ and M^hare defined in equation (5).

The results of the exercise must, however, be treated with caution, as many details are omitted. For example, if the VAT were to replace import tariffs, benefits to richer urban households, which consume a relatively higher proportion of imported goods, would likely to be even higher. On the other hand, incorporating cash and in kind consumption (which is relevant in rural Bangladesh) in the analysis is likely to indicate lower level of losses to poorer households, even though they would still lose. The basic result appears to be very robust: relative to the pre‐reform tax regime, a uniform VAT would always benefit richer groups and adversely affect the poorer households.

Distributional Characteristics of Commodities

The distributional characteristics (D_i) of commodity i are defined as

$\begin{array}{cc}{D}_i={\mathrm{\Sigma }}_h{\beta }^h\mathrm{.}{q}_i{x}_i^h/q_i{X}_i,& \left(6\right)\end{array}$

where β^h is the measure of distributional weights for household h: q_i is the consumer price of commodity i; $x_i^h$is the quantity of ith commodity consumed by household h; and X_i is the quantity of aggregate consumption of commodity i.

Thus, the key data requirement for the calculation of the distributional characteristics of households is the measure of social welfare weights for different households (β^h). This measure indicates the change in social welfare owing to a unit change in household h’s income. The weight ph is defined as

$\begin{array}{cc}{\beta }^h=(\partial W/{\partial U}^h)/({\partial U}^h/{\partial y}^h),& \left(7\right)\end{array}$

where W is the level of social welfare; U^h is the utility of household h; and y^h is the income (or total expenditure) of household h.¹⁷

The welfare weights β^h could he specified in a number of ways. As an example guiding practical application, Ahmad and Stern (1984) suggest the following method:

$\begin{array}{cc}{U}^h\left(I^h\right)=k[I^{1-e}/(1-e)]e\ne 1,e\ge 0=k\log \left(I^h\right)e=1,& \left(8\right)\end{array}$

where I^h is the total expenditure per capita of the hth household; e is a parameter; and e ≥ 0 ensures concavity. Now one can express’ ^h as equal to U^’(I^h), and choose a normalization for ^h (by the choice of k), so that the welfare weight for the poorest household is unity. Under these assumptions one can express

$\begin{array}{cc}{\beta }^h={(I^0/I^h)}^e,& \left(9\right)\end{array}$

where I⁰ is the total expenditure per capita of the reference poorest household.¹⁸ Given this viewpoint, ^{Ahmad and Stern (1984)} suggest that β^hrepresents the marginal social value of a unit of expenditure to individual h relative to a unit that accrues to poorest households. One could argue that the judgment as to the value of e is that of the researcher or the government. When e> 0, β^h < 1 for all households except the poorest, so that the increment to expenditure by the poor is socially more valuable than that by the rich. The ratio β^h/β^h’ increases with e for I^h < I^h’ and thus e may be thought of as an inequality aversion parameter. This method has a particular advantage in that it requires only minimal data to compute β^h across households.¹⁹

The present exercise used the household expenditure survey data for Bangladesh to generate distributional weights for rural and urban households. Apart from the base case (e = 0), in the exercise, four levels of e, e: 0.1, 1, 2, and 5, were chosen to assess the sensitivity of the estimates to changes in the degree of inequality aversion. A value of e = 0 implies that the policymakers value 1 taka of expenditure for the poorest individual as equivalent to 1 taka for the richest. A value of e = 1 implies that a marginal unit to h is worth half as much as a marginal unit to the reference household 1 if the expenditure of h is twice that of household 1. Similarly, values of e in excess of 2 give much greater weight to the poorest, and values of 5 and above begin to approach the “maxi‐min” or Rawisian utility function, by considering the welfare only of the poorest individuals (a marginal unit to the poorest is worth 32 times the value of a unit to someone with twice the expenditure). With no inequality aversion (e = 0), all commodities have distributional characteristics equal to unity. With the introduction of moderate inequality aversion (e = 0.1), the situation changes significantly. In general, the commodities that have high budget shares in the consumption basket of poorer households and/or have low expenditure elasticity acquire relatively higher value than the commodities with low budget share or high expenditure elasticity. This is what is being observed.

The results of the calculation of the distributional characteristics of 15 goods presented in Table 2 show a definite pattern. The value of D, is shown in panel A and its ranking is shown in panel B. These can be interpreted as follows: for example, ranking no. 1 (highest rank) represents the lowest priority and ranking no. 15 (lowest rank) represents the highest priority as a source of extra taxation. With changes in e, the relative rankings of commodities measured by the value of D, change, but the commodities that consistently rank high in the measure of distributional characteristics are rice (rank 2), wheat and grain (rank 3), vegetables and fruits (rank 5), edible oil (rank 4), and clothing (rank 6). This implies that applying a VAT on these commodities will be very regressive. By the same token, the VAT will be less regressive if it is exempted or zero rated for these goods. All of these commodities except clothing are agricultural goods, and, from an administrative point of view, VAT exemption would be more feasible than assigning zero rating. However, clothing, which includes textiles, is an important source of tax revenue under the pre‐reform system; therefore, relinquishing such a large tax base from the VAT does not seem reasonable, given the current narrow revenue base.

Table 2.

Distributional Characteristics for Commodities, 1984185

A: Distributional Characteristics (D_i)

Source: Author:s estimates, derived from household expenditure survey data.

Note: N^ie refers to expenditure elasticity of commodity group i.

Table 2.

Distributional Characteristics for Commodities, 1984185

A: Distributional Characteristics (D_i)

			Inequality aversion parameter: e
No.	Commodity group	Nie	0.1	1.0	2.0	5.0
1	Rice	0.46	0.8067	0.1277	0.01953	0.00023
2	Wheat and grain	0.65	0.8022	0.1237	0.01910	0.00028
3	Vegetables and fruits	1.01	0.7927	0.1103	0.01533	0.00016
4	Livestock (meat)	1.42	0.7788	0.0930	0.01095	0.00015
5	Fisheries	1.01	0.7910	0.1070	0.01415	0.00011
6	Edible oil	0.99	0.7932	0.1108	0.01545	0.00019
7	Sugar and gur	1.17	0.7837	0.0978	0.01194	0.00006
8	Other food	1.18	0.7882	0.1046	0.01383	0.00013
9	Tobacco products	1.36	0.7856	0.1025	0.01343	0.00009
10	Clothing	0.94	0.7926	0.1113	0.01613	0.00058
11	Manufactured goods	1.42	0.7757	0.0910	0.01113	0.00026
12	Traditional energy	0.27	0.8076	0.1290	0.02014	0.00041
13	Commercial energy	1.19	0.7882	0.1062	0.01455	0.00011
14	Housing	1.80	0.7760	0.0931	0.01168	0.00008
15	Services	1.62	0.7724	0.0874	0.01024	0.00014
B: Ranking of Distributional Characteristics
			Inequality aversion parameter: e
No.	Commodity group	Nie	0.1	1.0	2.0	5.0
1	Rice	0.46	2	2	2	5
2	Wheat and grain	0.65	3	3	3	3
3	Vegetables and fruits	1.01	5	6	6	7
4	Livestock (meat)	1.42	12	13	14	15
5	Fisheries	1.01	7	7	8	11
6	Edible oil	0.99	4	5	5	6
7	Sugar and gur	1.17	11	11	11	14
8	Other food	1.18	8	9	9	9
9	Tobacco products	1.36	10	10	10	12
10	Clothing	0.94	6	4	4	1
11	Manufactured goods	1.42	14	14	13	4
12	Traditional energy	0.27	1	1	1	2
13	Commercial energy	1.19	9	8	7	10
14	Housing	1.80	13	12	12	13
15	Services	1.62	15	15	15	8

Source: Author:s estimates, derived from household expenditure survey data.

Note: N^ie refers to expenditure elasticity of commodity group i.

View Table

Table 2.

Distributional Characteristics for Commodities, 1984185

A: Distributional Characteristics (D_i)

			Inequality aversion parameter: e
No.	Commodity group	Nie	0.1	1.0	2.0	5.0
1	Rice	0.46	0.8067	0.1277	0.01953	0.00023
2	Wheat and grain	0.65	0.8022	0.1237	0.01910	0.00028
3	Vegetables and fruits	1.01	0.7927	0.1103	0.01533	0.00016
4	Livestock (meat)	1.42	0.7788	0.0930	0.01095	0.00015
5	Fisheries	1.01	0.7910	0.1070	0.01415	0.00011
6	Edible oil	0.99	0.7932	0.1108	0.01545	0.00019
7	Sugar and gur	1.17	0.7837	0.0978	0.01194	0.00006
8	Other food	1.18	0.7882	0.1046	0.01383	0.00013
9	Tobacco products	1.36	0.7856	0.1025	0.01343	0.00009
10	Clothing	0.94	0.7926	0.1113	0.01613	0.00058
11	Manufactured goods	1.42	0.7757	0.0910	0.01113	0.00026
12	Traditional energy	0.27	0.8076	0.1290	0.02014	0.00041
13	Commercial energy	1.19	0.7882	0.1062	0.01455	0.00011
14	Housing	1.80	0.7760	0.0931	0.01168	0.00008
15	Services	1.62	0.7724	0.0874	0.01024	0.00014
B: Ranking of Distributional Characteristics
			Inequality aversion parameter: e
No.	Commodity group	Nie	0.1	1.0	2.0	5.0
1	Rice	0.46	2	2	2	5
2	Wheat and grain	0.65	3	3	3	3
3	Vegetables and fruits	1.01	5	6	6	7
4	Livestock (meat)	1.42	12	13	14	15
5	Fisheries	1.01	7	7	8	11
6	Edible oil	0.99	4	5	5	6
7	Sugar and gur	1.17	11	11	11	14
8	Other food	1.18	8	9	9	9
9	Tobacco products	1.36	10	10	10	12
10	Clothing	0.94	6	4	4	1
11	Manufactured goods	1.42	14	14	13	4
12	Traditional energy	0.27	1	1	1	2
13	Commercial energy	1.19	9	8	7	10
14	Housing	1.80	13	12	12	13
15	Services	1.62	15	15	15	8

Source: Author:s estimates, derived from household expenditure survey data.

Note: N^ie refers to expenditure elasticity of commodity group i.

It is also found that the following commodities feature low in the distributional characteristics, and thus are particularly attractive candidates for additional taxation with different values of e: tobacco products (rank 10); commercial energy (including electricity, petroleum products, and gas) (rank 9); sugar and gur (rank 11); housing (rank 13); livestock (rank 12); and, to a lesser extent, fisheries (rank 7). Of these, taxing housing, livestock, and fisheries is either politically unacceptable or administratively difficult. Therefore, we experimented with a package of tax reform whereby, on the one hand, there would be zero rating of the VAT for the agricultural products and, on the other hand, tobacco products, commercial energy, and sugar would face high excise taxes at the existing rates (except commercial energy, which faces a much higher rate to eliminate the subsidy element measured at shadow prices), in addition to a proportional VAT. The excises on some of these products are justified, not only for distributional considerations but also because under the present system they raise substantial revenue.

This reform package maintains the previous revenue level, sets a zero rate on foodgrains and vegetables, and applies a proportional VAT at an enhanced rate on the remaining commodities, including tobacco, energy goods, and sugar, which also face high excise taxes.²⁰ The consumption of foodgrains and vegetables was roughly 23 percent of the total private consumption expenditure for 1984/85. If the tax on the foodgrains group is set at zero, and additional excises are applied, a VAT of 9.9 percent of expenditure on all other commodities would be required to keep revenue constant.²¹

Table 3.

Equivalent Variation $\left(E_{01}^h\right)$for Uniform VAT with Zero‐Rated Commodities and Excise Taxes

Sources: Bangladesh household expenditure survey data, 198⅝6; and author’s estimates.Note:$E_{01}^h$and M^h’ are as defined in equation (5).

Table 3.

Equivalent Variation $\left(E_{01}^h\right)$for Uniform VAT with Zero‐Rated Commodities and Excise Taxes

Group	Per capita expenditure (taka/month)	Percent of household	Sample size	Mean household size	Mean per capita expenditure(Mh)	Equivalent variation (percent) $(E_{01}^h/M^h)$
Urban
I	< 249	1.1	42	7.2	210.9	–2.4
2	250–349	2.4	82	7.1	3114.8	–2.7
3	350–449	2.4	70	6.5	401.4	–L2
4	450–599	2.9	I08	5.9	523.6	–2.0
5	600–699	1.2	52	6.2	641.8	–0.3
6	700+	4.0	146	5.3	1092.4	+6.6
Rural
1	< 249	18.2	219	6.2	201.7	–1.2
2	250–349	27.9	316	6.3	298.3	–1.9
3	350–449	19.4	223	5.6	399.3	–1.l
4	450–599	12.4	149	5.6	504.9	+0.1
5	600–699	3.5	42	4.9	644.4	+0.8
6	700+	4.7	5I	5.3	944.7	+4,6
All		100.0	1500	5.9	427.6

Sources: Bangladesh household expenditure survey data, 198⅝6; and author’s estimates.Note:$E_{01}^h$and M^h’ are as defined in equation (5).

View Table

Table 3.

Equivalent Variation $\left(E_{01}^h\right)$for Uniform VAT with Zero‐Rated Commodities and Excise Taxes

Group	Per capita expenditure (taka/month)	Percent of household	Sample size	Mean household size	Mean per capita expenditure(Mh)	Equivalent variation (percent) $(E_{01}^h/M^h)$
Urban
I	< 249	1.1	42	7.2	210.9	–2.4
2	250–349	2.4	82	7.1	3114.8	–2.7
3	350–449	2.4	70	6.5	401.4	–L2
4	450–599	2.9	I08	5.9	523.6	–2.0
5	600–699	1.2	52	6.2	641.8	–0.3
6	700+	4.0	146	5.3	1092.4	+6.6
Rural
1	< 249	18.2	219	6.2	201.7	–1.2
2	250–349	27.9	316	6.3	298.3	–1.9
3	350–449	19.4	223	5.6	399.3	–1.l
4	450–599	12.4	149	5.6	504.9	+0.1
5	600–699	3.5	42	4.9	644.4	+0.8
6	700+	4.7	5I	5.3	944.7	+4,6
All		100.0	1500	5.9	427.6

Sources: Bangladesh household expenditure survey data, 198⅝6; and author’s estimates.Note:$E_{01}^h$and M^h’ are as defined in equation (5).

The $E_{01}^h/M^h$ measures for this reform package calculated for different urban and rural groups are presented in Table 3. Again, the pattern of results shows that the impact of a selective VAT is beneficial to the rich and harms the poor. But in this case, the magnitudes of loss or gain are less than in the uniform VAT case. The poorest group in urban areas loses the most‐equivalent to about a 2.4 percent decrease in its total expenditure. The richest group, on the other hand, benefits—equivalent to about a 6.6 percent increase in its total expenditure. In rural areas, the loss to the poorest group is 1.2 percent while the benefit to the richest group amounts to 4.6 percent. In this case, the distribution of losses also differs between rural and urban areas. In the uniform VAT case, four of the urban groups and four of the rural groups suffer losses. In the selective VAT case, three of the rural groups suffered lower losses, but now five urban groups lose. This could be attributable to the higher tax on nonfood items, which feature prominently (with a higher budget share) in the budget of urban households.

IV. Conclusion

The findings of our study suggest that among the different possible VAT schemes, a selective VAT scheme with some exemptions (or zero rating) and additional excises is likely to be more acceptable to the general public and policymakers than a uniform VAT.²³ We have studied other interesting cases of selective VATs with a different combination of taxes, but the reform package discussed above seems to be the most feasible one because it is revenue neutral in its impact and its distributional features are relatively attractive to policymakers. However, a nonuniform VAT scheme may create administrative difficulties unless it has only a few categories.²⁴ In particular, even highly advanced European tax authorities find it very difficult to administer the rebate system that accompanies zero rating of agricultural goods. In Bangladesh, if agricultural goods (foodgrains and vegetables) were exempted from the VAT (instead of being zero rated) because of administrative difficulties, this would introduce an additional degree of distributional inequity, adversely affecting the rural households and causing these goods to be consumed proportionately more.

Our analysis is merely suggestive. It is based on detailed household expenditure survey data, and various households were classified only in terms of per capita expenditure. The data in its present form do not distinguish between purchases and expenditure out of production within a household, a phenomenon that is likely to be important in rural Bangladesh, particularly among owner‐occupied farms. Also, many transactions in rural and urban areas do not go through formal markets. Thus, the incidence analysis based on equivalent variation measure is likely to overstate the magnitude of losses or gains to such households. The households with their own farms are likely to be better off, however, than the households that rely primarily on wage income and form part of a growing number of landless laborers.

One crucial formulation underlying the model is that commodity taxes are fully passed through to consumer prices. It is likely that the results are affected by the assumptions of the pricing rule. For some commodity groups, as an alternative to full forward shifting, one can assume full backward shifting, in which domestic prices are determined by world prices so that taxes are passed back, which then compresses factor returns in the taxed sectors. Dahl and Mitra (1991) have shown analytically that the revenue effect and welfare cost also depend on the particular variant (or mix) of the pricing rule assumed. However, an incidence analysis specifying different pricing rules for different sectors requires much more detailed information than that we now have.²⁵