Published by Oxford University Press in association with The London School of Hygiene and Tropical Medicine The Author

r 2008; all rights reserved.

Health Policy and Planning 2008;23:200209 doi:10.1093/heapol/czn007

Price Elasticity Estimates for Tobacco Products in India

Rijo M John
Accepted 4 March 2008 The tax base of tobacco in India is heavily dependent on about 14% of tobacco users, who smoke cigarettes. Non-cigarette tobacco products accounting for 85% of the tobacco consumption contributes only 15% of the total tobacco taxes. Though taxation is an important tool to regulate consumption of tobacco, there have been no estimates of price elasticities for different tobacco products in India to date, which can guide tax policy on tobacco. This paper, for the first time in India, examines the price elasticity of demand for bidis, cigarettes and leaf tobacco at the national level using a representative cross-section of households. This study found that own-price elasticity estimates of different tobacco products in India ranged between 0.4 to 0.9, with bidis (an indigenous hand-rolled smoked tobacco preparation in India) and leaf tobacco having elasticities close to unity. Cigarettes were the least price elastic of all. With some assumptions, it is shown that the tax on bidis can be increased to Rs. 100 per 1000 sticks compared with the current Rs. 14 and the tax on an average cigarette can be increased to Rs. 3.5 per stick without any fear of losing revenue. The paper argues that the current system of taxing cigarettes in India based on the presence of filters and the length of cigarettes has no justification on health grounds, and should be abolished, if reducing tobacco consumption and the consequent disease burden is one of the objectives of tobacco taxation policy. It also argues that attempts to regulate tobacco use without effecting significant tax increases on bidis may not produce desired results. Keywords Tobacco, bidis, cigarettes, consumption, elasticity, India

Downloaded from heapol.oxfordjournals.org by guest on August 28, 2011

KEY MESSAGES   Price elasticities for different tobacco products at a national level have never been estimated in India Different tobacco products in India are price responsive and hence taxation could be used as a viable tool of regulating tobacco use in India Effective regulation of tobacco use in India requires substantial increases in tax of bidis and other non-cigarette tobacco products, apart from tax increases on cigarettes The existing system of taxing cigarettes in India based on the presence of filters and the length of cigarettes has no justification on health grounds, and should be abolished

Introduction
Post Doctoral Fellow, Center for Tobacco Control Research and Education, University of California, San Francisco, CA 94143-1390, USA. Corresponding author. 530 Parnassus Ave, Ste 366, San Francisco, CA 941431390. Tel: (Office) 415-476-3139. Fax: 415-514-9345. E-mail: rmjohn@gmail.com

Ever since the huge morbidity and mortality associated with tobacco became evident, some nations have tried to regulate its use by various price and non-price instruments. The process of tobacco regulation has gained momentum in many developing

200

PRICE ELASTICITY ESTIMATES FOR TOBACCO PRODUCTS IN INDIA

Table 1 Excise duty rates for cigarettes and bidis in India, year 200708 Duty as percentage of retail price of the most popular brand

201

Tobacco products Cigarettes Non-filter cigarettes 60 mm >6070 mm Filter cigarettes 70 mm >7075 mm >7585 mm >85 mm* Bidis Handmade Machine-made

Duty per 1000 sticks in Rs.

Proportion of consumption for the year 200506 100%

168 546

34% 44%

7% 24%

819 1323 1759 2163

34% 38% 59% NA

54% 9% 7% NA 100%

14 26

8.8%

$98% $2%

*Filter cigarettes >85 mm constitute only a negligible share of the market and their tax is calculated as Rs. 600 or 300% Rs. 20 whichever is higher. Source: Sunley (2008).

countries, especially after the implementation of the global Framework Convention for Tobacco Control in 2003 by the World Health Organization. Taxation is an important price instrument for regulating tobacco use. As in other countries, cigarettes are taxed in India. However, unlike in other countries, tobacco consumption in India is characterized by extensive use of noncigarette tobacco products, especially in the form of bidis1 and chewing tobacco. In India, 57% of men and 11% of women in the 1549 years age group consume tobacco in some form (IIPS and Macro International 2007). As of year 200405, it is estimated that per capita consumption of bidis in rural and in urban areas is 31.5 and 18.8 sticks per month, respectively, while cigarettes consumption is 0.79 and 2.52 sticks, respectively (Government of India 2006). It is also estimated that there are 119.25 million adult (15) smokers (110.46 million male and 8.79 million female) in India as of the year 2004 (ERC 2005). Cigarettes account for only 14% of the tobacco consumption in India. However, they contribute as high as 85% of the excise revenue collected from tobacco products (ERC 2005). Thus taxation of tobacco products in India is highly skewed in favour of bidis and chewing tobacco products even though they contribute 48% and 38%, respectively, to the total tobacco consumption in India (Sunley 2008). A strong case for increasing taxes on bidis on health grounds has been made by Sunley (2008). He argues that since bidis are at least as harmful as cigarettes, the taxation on bidis should be increased to match at least that of the lower taxed cigarettes. Taxes on cigarettes in India are levied on the basis of cigarette length; the tax ranged from Rs. 168 to Rs. 543 per 1000 sticks for non-filtered cigarettes and Rs. 819 to Rs. 2163 for filtered cigarettes in financial year 200708 (Table 1).2 The tax on 1000 sticks of handmade bidis, on the other hand, is a meager Rs. 14. Machine-made bidis, which account for only 2% of total bidis produced in India, are taxed at Rs. 26 per 1000 sticks. There are no stated reasons why bidis are

taxed at such a low rate compared with cigarettes in India, though it is presumed that it is mostly economically weaker people who smoke bidis and thousands of poor households are involved in the production of bidis. The price advantage that bidis enjoy over cigarettes is largely due to the huge discrepancies in taxes on these products and this could be the single greatest reason why bidis account for a major proportion77% by the weight of tobacco (Sunley 2008)of smoked tobacco in India. Apart from bidis and cigarettes, a wide range of alternative tobacco products are consumed in India such as leaf tobacco, hookah, zarda, cheroot and snuff. Some of these are chewed and some smoked. However, bidis, cigarettes and leaf tobacco together account for almost 95% of the tobacco that is consumed in India. Most non-cigarette tobacco products in India are much cheaper than cigarettes. The fact that the relatively expensive cigarettes account for only a small fraction of tobacco use in India is an indication that price hugely affects consumption. If the tobacco products are sufficiently price responsive, an increase in the tax rate of these goods will have the effect of reducing the consumption substantially, while raising the tax revenue. There is overwhelming evidence from other countries (Chaloupka and Warner 2000) that increased taxes on tobacco leads to a reduction in consumption. Estimates of price-elasticity of demand for cigarettes from developed countries range from 0.25 to 0.50. The estimates from low-income and middleincome countries suggest that price elasticity of demand varies between 0.50 to 1.00 (Chaloupka et al. 2000). Analyses from various South East Asian countries have found that short-run price elasticity estimates for tobacco products range from 0.17 to 0.78, while long-run estimates ranging from 0.4 to 1.21 (Guindon et al. 2003). Estimation of price elasticities for different tobacco products has received little attention in India. The only known estimate is for cigarettes for the sample period 198182 to 199293, and is estimated to be 0.67 (Sarma 2000). Comparable and more recent estimates for other tobacco products as well as cigarettes are essential to formulate any comprehensive price control measures. To the best of the authors knowledge there is no study that provides information on the price elasticity of different tobacco products in India. Hence this paper, for the first time in India, examines the price elasticity of demand for bidis, cigarettes and leaf tobacco at the national level. In the next section, a brief discussion of the data on consumption of tobacco in India is made followed by a discussion of the methodology of elasticity estimation used in this paper. The spatial variation in prices of tobacco products is then used to estimate own- and cross-price elasticities of bidis, cigarettes and leaf tobacco.

Downloaded from heapol.oxfordjournals.org by guest on August 28, 2011

Unit values and prices

The National Sample Survey Organization (NSSO), under the Ministry of Statistics and Program Implementation, Government of India, conducts nationwide sample surveys every 5 years on the consumption habits of households in India. The survey conducted from July 1999 to June 2000 collected information on consumption and various household characteristics of 120,309 households spread over 10,140 villages. The consumption goods included a variety of tobacco products along with more than 500

202

HEALTH POLICY AND PLANNING

Table 2 Unit values and budget shares of different tobacco products Rural Commodity Bidis Cigarettes Leaf tobacco Tobacco (Total) % consuming 36.5 3.69 19.42 62.63 Unit values 0.17 1.40 69.86 NA Share 1.72 0.22 0.39 2.66 Urban % consuming 19.84 9.61 7.25 39.69 Unit values 0.21 1.32 74.73 NA Share 1.34 0.97 0.21 2.93

Note: All values are weighted using sampling weights. Total includes all tobacco products including the ones that are not listed here but are in the NSS data. Unit of measurement for bidis and cigarettes is number of sticks and that for leaf tobacco is Kg. Unit values are all in Rupees. Budget shares (averaged over households who bought at least one of the tobacco products) are in percentages. Source: Authors calculation from NSSO (2000) data.

food and non-food items. The survey asked for the quantity purchased as well as the expenditure incurred for consumption of various products over the last 30 days prior to the date of interview. According to this survey, at least one member in 62% of the rural households and 40% of the urban households reported consuming tobacco in some form. One reason for the dearth of estimates of price elasticities for tobacco in India could be the lack of sufficient data on prices and quantity consumed for various tobacco products. While NSSO surveys provide cross-sectional information on households expenditure and quantity consumed of various tobacco products, they do not provide direct information on prices. Various attempts have been made in the literature to estimate price responses in such cases. One approach is to use available regional price data used normally for constructing consumer price indices. These data can be merged with the NSSO data by associating with each household the prices available from the nearest collection centre at the time closest to the reporting period for that household. The combined data can then be used to estimate a demand system with individual households as the unit of analysis.3 However, such a procedure may give inaccurate estimates, as prices are not collected from each and every desirable location (Deaton 1997: 283). Nor are they available at the desired level of disaggregation. Since prices of tobacco products are not available from such collection centres for the level of disaggregation used in this analysis, this approach could not be used in this paper. Another approach would be to treat unit values (expenditure divided by quantity) collected in the surveys themselves as prices. Table 2 gives average unit values, budget shares (both of them averaged only over households who purchased at least one of the tobacco products) and percentage of households consuming each tobacco product. The table shows the weighted averages using inverse sampling probabilities as weights so that the estimates should be representative of the corresponding rural and urban households in India. If we were to consider the unit values as prices it means that the price of bidis, cigarettes and leaf tobacco in rural India is 17 paisa/stick, Rs. 1.40/stick and Rs. 70/kg. respectively. In urban India it is 21 paisa/stick, Rs. 1.32/ stick and Rs. 75/kg. respectively. (One hundred paisa is equivalent to one Indian Rupee.) While the unit values in rural and urban India for both bidis and leaf tobacco are as expected, those for cigarettes are counter to expectation. One would expect a higher average price for cigarettes in urban than in rural India since much of the high priced cigarettes are sold in urban India. However, since more cigarettes smokers are in urban India and if a large number of cigarettes smokers use low priced cigarettes, it

is possible to have a lower average price for cigarettes in urban India relative to that in rural India. As we see later, the variability is very high for cigarettes prices in rural India versus urban India, meaning that some of the cigarettes smokers in rural India are using substantially higher priced cigarettes than others. Having said that, it should be also noted that unit values are not the same as prices for the reasons explained below. Because of this, unit values cannot be used as they are to estimate price elasticities. Unit values are not the same as prices due to two reasons, as Deaton (1997) points out. First, unit values also reflect quality. Since unit values are computed by dividing expenditures by physical quantities, they do not take into account the nature of heterogeneity of the commodity. For example, when households report quantity and value of cigarettes, the interviewer has to overlook the fact that cigarettes are a heterogenous commodity, and that price varies according to the quality chosen. Consumers, in general, adapt to changes in price not only by changing the quantity purchased but also by changing the quality. Hence, an estimate of price elasticity using unit values will lead to overestimation. Secondly, unit values are subject to measurement errors since the measurement error in quantity is transmitted to unit values. Thus, unit values are contaminated by both measurement errors and quality variations. However, unit values also contain price information which can be captured and made use of in estimating price elasticities. A theoretical model was developed by Deaton (1988, 1997), wherein consistent estimation of price elasticities was made possible with the spatial variation in unit values from survey data. This model is used to derive own- and cross-price elasticities for tobacco products in this paper. According to this model, by studying the spatial variation in unit values, one should be able to assess the extent of spatial variation in prices. A simple way of examining the spatial variation in prices, which makes the temporary supposition that quality variations are negligible, is to regress log of unit values on dummy variables representing village/urban units.4 This regression is nothing but a decomposition of variance of unit values over villages, similar to an ANOVA regression. Table 3 provides the extent and significance of the spatial variation in prices (unit values) along with the source of variability separately for rural and urban India. The first column in each panel shows the standard deviation of the logarithms of unit values multiplied by 100 so that the figures can be interpreted as percentage variability. Leaf tobacco seems to have maximum variability in both rural and urban areas. F and R2 are the F-statistics and R2-statistics from this regression. What we

Downloaded from heapol.oxfordjournals.org by guest on August 28, 2011

PRICE ELASTICITY ESTIMATES FOR TOBACCO PRODUCTS IN INDIA

Table 3 Variability in unit values Rural India Items Bidi Cigarette Leaf tobacco SD 59.8 63.0 206.6 F-stat 12.35 4.08 7.67 F-reg 0.76 0.81 0.68 R-sq 71.63 7.52 39.10 Urban India SD 56.3 55.7 227.5 F-stat 4.58 2.82 5.72 R-sq 0.68 0.68 0.76

203

F-reg 23.73 9.75 11.40

Notes: SD refers to 100 times the standard deviation of log unit values calculated over the households reporting positive consumption. F and R-sq are the F-statistics and R-sq-statistics associated with the presence of dummy variables for each village in the survey. F-reg is the F-statistics of an ANOVA of log unit values on dummies for 78 regions. All statistics are significant at 1% level.

are looking for here is evidence that unit values are informative about prices. Since prices should not vary much within villages over a short period of time, there should be significant F-statistics for the village effects. In other words, village dummies should explain a large share of the total variation in logarithm of unit values. We observe that almost 70% of the variation in price is explained by variability between villages for all three goods. Within village variation is thus relatively small. A test for the broad regional effects was also conducted.5 Column F-reg in each panel shows the F-statistics for the regression of log unit values on region dummies and it shows strong evidence of regional price variation. The F-statistics are significant at the 1% level for all regressions. Information from the broad regional variation and stronger inter-village variation of unit values can thus be used to estimate a demand model proposed by Deaton (1997) for the products under consideration. The following section briefly introduces the model of price elasticity estimation.

Estimating price elasticity from unit values

Due to the large volume of equations involved in the model, only a brief introduction to the model is given here. A detailed exposition of the methodology can be found in Deaton (1997). This is a model of consumer behaviour in which households choose both quantity and quality so that expenditure on a good is the product of quantity, quality and price, thereby implying that unit value is price times quality. Thus, this is a model where utility depends not only on quantity, but also on quality, giving an augmented utility function. Commodities are defined as collections of heterogeneous goods and quality is defined as a property of commodity aggregates. Because unit values are defined to include quality, the analysis must take into account price and income elasticities of quality as well. The model requires that households are geographically clustered within the sample. Village demand patterns, as represented by the budget shares, are regressed on the average village prices, as represented by unit values. The following two equations link the budget shares and unit values to household expenditures, other household characteristics and the underlying prices of commodities.
0 WGic 0 0 ln xic G Zic G G N X H1

taken to be a linear function of the logarithm of total household expenditure, x, a vector of household characteristics, Z, and the logarithm of N prices where N is the number of commodities.6 However, coefficients of these equations are not elasticities, which need to be calculated. The derivation for the same can be found in Deaton (1997). The first element of the residual in equation (1), fGc, is a village-level effect that is the same for all households within a village and it can be thought of either as a random effect or as a fixed effect.7 Since both fGc and price are unobserved, it is necessary to assume that the term fGc is uncorrelated with price in order to estimate the influence of the latter. The term u0 is the standard error term representing, Gic among other things, measurement errors in the budget share and taste (quality) variations. The price in equation (1) is not observed but is related to unit value (UV) as given in equation (2). The logarithm of unit value is a function of ln x, household characteristics represented by the vector Z, and price. Since unit value is price multiplied by quality, 1 is the expenditure G elasticity of quality. Differentiating (1) with respect to ln x and defining 2G to be the elasticity of expenditure with respect to quantity, yields @ ln WG/@ ln x 0 /WG 2G 1 1, since the G G logarithm of share is the sum of logarithms of quantity and quality less the logarithm of expenditure. Rearranging it will yield the expenditure elasticity of quantity so that the total income elasticity of quantity and quality together will be 2G 1 . G 0 2G 1 1 G G WG 3

Downloaded from heapol.oxfordjournals.org by guest on August 28, 2011

The non-price parameters such as ,  and in both the equations can be consistently estimated using standard OLS regression with the assumption that market prices do not vary for a given commodity within each village over the relevant reporting period. The model allows for completing the system of demand equations by adding a composite commodity which will then exhaust the total household budget. Symmetry restrictions that add to the precision of parameter estimates are also imposed.8

Empirical results
Households with zero consumption of tobacco are eliminated from the analysis. This is with the assumption that the preferences of tobacco consumers and non-consumers are fundamentally different in the sense that tobacco is not an argument in the utility function of non-users of tobacco for whom tobacco does not give any utility no matter what the price is. This is not to argue that price has no effect on initiation of tobacco consumption. Even though there may be theoretical reasons to include all the consumers in the analysis, the empirical evidence on the

GH ln pHc fGc u0 1 Gic

GH

ln UVGic

1 G

1 G

ln xic

1 G

Zic

N X H1

ln pHc u1 Gic

WGic is the budget share of good G in the budget of household i living in village (cluster) c. The budget share of the household is

204

HEALTH POLICY AND PLANNING

Table 4 Income and household size coefficients and income elasticities Unit value regression Bidis Coeff Rural India lnexp lhsize Observations F-stat Adjusted-R Clusters ExpElsty Urban India lnexp lhsize Observations F-stat Adjusted-R Clusters ExpElsty 0.044 0.018 9139 2.02 0.533 2900 0.28 0.01 0.01 0.29 0.240 0.165 5460 12.34 0.48 2376 1.59 0.00 0.00 0.00 0.103 0.307 3580 2.73 0.630 1266 0.29 0.00 0.29 0.00 0.841 0.228 19229 125.93 0.384 3894 0.00 0.00 0.00 0.893 1.002 19229 76.16 0.340 3894 0.00 0.00 0.00 0.133 0.035 19229 29.59 0.439 3894
Downloaded from heapol.oxfordjournals.org by guest on August 28, 2011

Budget share regression Leaf tobacco P-val 0.00 0.00 Coeff 0.005 0.005 13169 0.00 0.76 0.589 2843 0.37 0.77 P-val 0.92 0.91 Bidis Coeff 0.719 0.116 43923 197.70 0.441 5865 0.00 P-val 0.00 0.00 Cigarettes Coeff 0.435 0.456 43923 99.78 0.294 5865 0.00 P-val 0.00 0.00 Leaf tobacco Coeff 0.226 0.001 43923 91.80 0.500 5865 0.00 P-val 0.00 0.93 Coeff 0.108

Cigarettes P-val 0.00 0.00

0.048 0.034 24946 3.38 0.695 5029 0.49

0.111 3853

0.00

2.34 0.618 1994 2.37

0.00 0.01

0.00

Notes: Coefficients in the budget share regression are multiplied by 100 for the convenience in reporting. Total expenditure elasticity is given by 2G 1.

effect of prices on the initiation of smoking is mixed (Cawley et al. 2004). Hence, the inclusion or exclusion of consumers reporting zero consumption is largely a matter of empirical significance. A previous study9 using the same data has shown that the large majority of consumers who report zero consumption of tobacco are doing so not because they are unable to afford it but because tobacco does not figure in their utility function no matter what its price is. Hence, based on the existing evidence using this data, this paper excludes all the non-consumers from the analysis. As the non-consumers are excluded, the analysis is conditional in some sense.10 The set of household socio-economic characteristics that are considered for the regression in equations 1 and 2 includes: log of household expenditure, log of household size, ratio of number of adults (14 years of age or more) to household size, ratio of total adult males to household size, average education of the household (total education, in years, received by all the members divided by household size), years of education received by the most educated member in a household, and dummies for religion, social groups and occupational groups. Region dummies were also introduced to eliminate broad regional taste differences, if any, that may be not due to regional price differences. These variables are introduced only with the intention of purging the budget shares and unit values of the household-specific effects so as to allow for the quality effects and enable consistent estimation of own- and cross-price elasticities.

Estimates from unit value and budget share regression

Table 4 shows the estimated coefficients from both unit value and budget share equations along with income (expenditure) elasticities. The coefficient of ln x in the unit value equation gives the expenditure elasticity of quality. As we can observe, in all

cases it is positive and in most cases significant at the 1% level for both rural and urban India. Among the different tobacco products considered, cigarettes have the highest expenditure elasticity of quality, at 0.11 in rural and 0.24 in urban India. This implies that a doubling of the household total expenditure would raise the average price paid for cigarettes by 11% and 24% in rural and urban India, respectively. This is also evidence that lower income households spent more on lower quality cigarettes, possibly ones without a filter.11 Quality elasticity of bidis is around 5% and that of leaf tobacco is insignificant. Coefficients on the logarithm of household size are roughly similar in size and opposite in sign to the coefficients on the logarithm of total expenditure in the unit value regression. This seems to suggest that increases in household size act like reductions in income. The estimated coefficients on household size are smaller in absolute size than the coefficients of total expenditure. With total household expenditure and other household characteristics remaining the same, an increase in household size has a significant effect of decreasing the average price paid by the household. It may mean that given total expenditure, as household size increases, households may increase consumption, therefore resorting to consuming lower quality products which are cheaper. It may be interesting to note here that the average household size is higher among tobacco consuming households (5.4 and 4.92 in rural and urban India, respectively) than other households (4.4 and 4.3 in rural and urban India, respectively) in our data. Budget share equations also show opposite signs in the case of coefficients of log total expenditure and log household size. Keeping expenditures and other variables constant, an increase in household size increases budget shares of bidis and leaf tobacco and it decreases budget shares of cigarettes in rural India. In urban India, on the other hand, household size has an increasing effect on budget share allocated to bidis and

PRICE ELASTICITY ESTIMATES FOR TOBACCO PRODUCTS IN INDIA

Notes: The elasticity in row i, column j estimates the effect of a change in the price of good j on the quantity demanded of good i. Values in parentheses are the bootstrapped standard errors calculated by making 1000 draws from the second stage data, and are defined as half the length of the interval around the bootstrapped mean, and contain 68.3% of the bootstrapped estimates. Assuming the estimates follow a normal distribution, the coefficients with ***, ** and * imply levels of significance at 1%, 5% and 10%, respectively. Residual refers to a composite commodity that exhausts the budget.

205

1.525*** (0.384)

0.262*** (0.001)

0.456 (0.362)

1.54*** (0.386)

0.262*** (0.001)

0.52*** (0.140)

0.581*** (0.095)

0.413*** (0.137)

Composite

0.87*** (0.094)

0.122 (0.113)

0.855*** (0.084)

0.252** (0.126)

0.091 (0.108)

0.0002 (0.000)

0.071 (0.068)

0.000 (0.000)

Urban

a decreasing effect on leaf tobacco and cigarettes. Total expenditure elasticity (sum of the expenditure elasticity of quantity and quality) is less than one for both bidis and leaf tobacco in both rural and urban India and is more than unity for cigarettes. High expenditure elasticities of cigarettes imply that cigarettes are luxury goods both in rural and in urban India. A similar result was obtained by Suryanarayana (2002) though the author had done the estimation using aggregate data unlike the household level data that is used in our exercise. An increase in total household expenditure more than doubles the consumption of cigarettes among rural households. Other socio-demographic variables in the regression exert only occasional and modest effects on unit values and budget shares. The purpose of introducing them in the first stage regression was only to purge their effects on prices so that second stage estimation of price elasticities is not affected. Therefore, the results are not discussed here. However, a few results are worth mentioning. An increase in the male ratio would lead to an increase in budget spent on bidis in rural and urban India. Education has mild but significant decreasing effects on the budget share devoted to tobacco consumption, especially bidis. Similarly, households belonging to the Sikh religious group also exhibited a mild but significant negative effect on budget share of various tobacco products.

0.002 (0.021)

0.877*** (0.032)

0.003 (0.020)

0.011 (0.011)

0.874*** (0.029) 0.000 (0.119)

0.0001*** (0.000)

0.011 (0.010)

0.179 (0.431)

0.221 (0.414)

0.011 (0.167)

0.063 (0.093)

0.196 (0.428)

0.002* (0.001)

Bidis

0.002** (0.001)

Cigarettes

0.0001*** (0.000)

Leaf tobacco

Downloaded from heapol.oxfordjournals.org by guest on August 28, 2011

Estimates of own- and cross-price elasticities

Table 5 reports own- and cross-price elasticity estimates for bidis, cigarettes and leaf tobacco using only tobacco consuming households. Both symmetry-constrained and -unconstrained estimates are presented here and the results are reported for both rural and urban households. Values given in parentheses below each coefficient are bootstrapped standard errors calculated by making 1000 draws from the second stage data, and are defined as half the length of the interval around the bootstrapped mean, and containing 68.3% of the bootstrapped estimates. Region dummies for the NSSO regions were also used in these regressions to remove any regional taste differences that may affect the estimated coefficients. The elasticity in row i, column j, estimates the effect of a change in the price of good j on the quantity demanded of good i. Elasticity coefficients are relatively well determined and are comparable to other price elasticity estimates available in the literature from other developing countries. As we can observe, all of the own-price elasticities (diagonal elements) are negative and are statistically significant except for cigarettes in urban India. Many of the cross-price elasticities, however, are not statistically significant. The last column and row in each panel shows the estimates for a composite commodity or a residual item for which we have no price information. The elasticities for this were calculated imposing theoretical restrictions. As we can observe, own-price elasticity estimates for rural and urban households are approximately the same, except for cigarettes which are relatively more inelastic in urban India than in rural India. Elasticity for cigarettes is lower than 0.5, which is in consensus with the conservative estimates available for cigarettes, whether from developing or developed countries. On the other hand, there are no comparable studies for bidis and leaf tobacco for which the elasticities were close to unity. This may be reason enough to justify the heavy taxation on cigarettes

Residual

1.845*** (0.196)

0.256*** (0.000)

0.5*** (0.042)

0.411*** (0.067)

1.71*** (0.213)

0.010 (0.009)

0.87*** (0.019)

0.010 (0.032)

0.010 (0.009)

0.00009*** (0.000)

0.871*** (0.018) 0.022 (0.025) 0.036 (0.035) Leaf tobacco

0.021 (0.032)

Table 5 Price elasticity estimates for tobacco products in India

Cigarettes

0.117*** (0.045)

0.41* (0.239)

0.084*** (0.029)

0.338** (0.143)

0.15* (0.078)

0.0001** (0.000)

Bidis

0.91*** (0.046)

0.24* (0.126)

0.067 (0.049)

0.922*** (0.043)

0.0001** (0.000)

0.455* (0.147)

Symmetry-constrained estimates

Unconstrained estimates

Leaf tobacco

Cigarettes

Cigarettes

Residual

Residual

Bidis

Bidis

0.0001*** (0.000)

Rural

0.0002*** (0.000)

0.0001*** (0.000)

Leaf tobacco

0.26*** (0.000)

0.478*** (0.039)

0.509*** (0.039)

206

HEALTH POLICY AND PLANNING

in India, as economic logic supports taxing luxuries which are highly inelastic. The present exercise indicates that consumption of tobacco products in India does respond to changes in prices, though the proportionate increases in price lead to slightly less than proportionate reductions in consumption in the case of bidis and leaf tobacco, while leading to much less proportionate reductions in consumption in the case of cigarettes. Examination of cross-price elasticity from the symmetryconstrained estimates (lower panel in Table 5) indicates that bidis are complementary to cigarettes and leaf tobacco in rural India, and to cigarettes in urban India, though many of the other cross-elasticity coefficients are not statistically significant. An earlier study (John 2006) observed that a large number of households reporting the consumption of bidis were also consuming other tobacco products such as cigarettes and leaf tobacco. It established that increased presence of adults, higher male to female ratio in a household, larger household size and the alcohol consumption habit of households were important predictors for a higher relative probability of consuming a combination of tobacco products. The complementarity between bidis and cigarettes need not be entirely counterintuitive either. Given the huge difference in the price of bidis and cigarettes in India and the fact that both these items cater to totally different socio-economic groups, it is perfectly reasonable to think that consumers who consume either of the products would occasionally try out a few sticks of the other. Moreover, the complementarity here need not necessarily mean that increases in the price of one will lead to a dramatic reduction in consumption of the other. This is primarily because the magnitude of the cross elasticity coefficient is small. If we examine the coefficient in the case of rural India, we see that the magnitude of the effect of an increase in price of cigarettes on bidis, and that of an increase in price of bidis on cigarettes, is different though the directions are the same. The effect of a percentage increase in cigarette price on bidi consumption is very low, whereas that of a bidi price increase on cigarette consumption is higher. This is especially true in rural India, for two reasons: (1) the price of bidis is at a very low base compared with that of cigarettes and (2) a price rise of bidis may have large budgetary implications for rural households, as bidis are the preferred form of smoked tobacco consumption there. Hence any increase in the price of bidis will have greater effects in reducing consumption of cigarettes as well.

merely expository in nature. The following assumptions were made: 1. 2. 3. 4. there are no substitution effects due to price change; change in price is commensurate with change in tax; elasticity is constant across the entire range of prices; there is no tax evasion or smuggling as a result of increased taxes.

Effects of price changes through taxation

Curbing tobacco use is critical for health. However, two counteracting objectives of government with respect to tobacco, namely generating tax revenue and protecting households dependent on tobacco-related employment, become crucial while formulating policies to regulate tobacco use. The public health goal of regulating tobacco use could be yet another objective for policy makers. In this section I address the tax objective of the government while regulating demand for tobacco use. Addressing the employment question is out of the scope of this paper. Using the price elasticity estimates and some key assumptions, it is possible to calculate the movement of tax revenue and consumption of various tobacco products. Calculations here are done only for cigarettes and bidis and are

Annual consumption of manufactured cigarettes in India was estimated to be 107.5 billion sticks and the tax revenue from cigarettes amounted to Rs. 70.86 billion in the financial year 2006 07 (Sunley 2008). Assuming that the tax is collected from all cigarettes consumed, this yields an average tax per stick of Rs. 659 per 1000. Using the proportion of each type of cigarette out of total cigarette consumption, and the respective ad valorem tax rates given in Table 1, we computed an ad valorem tax rate of 38.3% and a retail price of Rs. 1.72 per stick for an average cigarette in India. Bidis are taxed at a very low rate with a quantity tax of Rs. 14 per 1000 sticks for man-made bidis and Rs. 26 per 1000 for machinemade bidis in the year 200708. Since man-made bidis constitute more than 98% of bidis produced in India (Sunley 2008), Rs. 14 is taken as the initial tax amount. Using the ad valorem tax rate (8.8%) for bidis given in Table 1 and the tax per stick, the price for an average bidi stick was computed as Rs. 0.16 which amounts to Rs. 4 per pack of 25 bidis. Similarly, Sunley (2008) notes that the most common price for a pack of 25 bidis in India is Rs. 4. Annual consumption of manufactured bidis in India was estimated to be between 750 billion to 1.2 trillion sticks in 2007, 5270% of which are not taxed either due to non-compliance or due to small producer exemptions (Sunley 2008). For the purpose of this simulation, the current consumption of bidis was taken to be one trillion.12 Tax revenue from bidis for the year 200607 was Rs. 4.3 billion. If tax was collected from all bidis that were consumed, the tax revenue would have been approximately Rs. 14 billion. The actual tax collection is thus nearly 70% less. In this simulation exercise the initial tax revenue was taken to be Rs. 14 billion. So the reader should keep in mind that for all the levels of projected tax revenues for bidis, actual revenue will be well below the projected revenue unless it is ensured that all bidis consumed in India are taxed. Table 6 shows the changes in consumption and tax revenue as a result of changes in tax amount. Tax shocks (increases in tax as a percentage of the existing tax) are introduced and the changes in consumption, expenditure and tax revenue are calculated using the price elasticity in Table 5. The effects are calculated separately for rural and urban India using the respective elasticities and are later aggregated to obtain the results for allIndia as shown in the table. For this, out of the total cigarettes consumed, 63% were attributed to rural and 37% to urban households as estimated from the NSS data. Similarly for bidis, the proportions were 77% and 23% for rural and urban households respectively. As the table illustrates, revenue from taxation of bidis increases until tax becomes more than 40% of the retail price, which would be equivalent to a multi-fold increase in the current taxes on bidis from the current level of roughly 9% of the retail price. In other words, tax on bidis can be increased to Rs. 100 per 1000 sticks compared with the current Rs. 14 without any fear of losing revenue. At that level, the average price of a pack of 25 bidis would be slightly more than

Downloaded from heapol.oxfordjournals.org by guest on August 28, 2011

PRICE ELASTICITY ESTIMATES FOR TOBACCO PRODUCTS IN INDIA

Table 6 Changes in consumption, expenditures and revenue from changes in taxes Tax Tax per Unit revenue stick Consumption Expenditure Tax price Tax (Rs.) (Billion sticks) (Billion Rs.) (Billion Rs.) shock (Rs.) rate Bidis 0% 20% 40% 60% 80% 0.159 9% 0.162 10% 0.165 12% 0.167 13% 0.170 15% 0.014 0.017 0.020 0.022 0.025 0.028 0.042 0.070 0.098 0.101 1000 984 968 952 936 920 840 681 521 505 159.1 159.3 159.4 159.5 159.4 159.3 157.2 146.5 126.7 124.3 14.0 16.5 19.0 21.3 23.6 25.8 35.3 47.7 51.1 50.9

207

100% 0.173 16% 200% 0.187 22% 400% 0.215 33% 600% 0.243 40% 620% 0.246 41% Cigarettes 0% 20% 40% 60% 80% 1.721 38% 1.853 43% 1.985 46% 2.116 50% 2.248 53%

0.659 0.791 0.923 1.055 1.186 1.318 1.977 3.296 3.493

108 105 103 100 98 96 84 60 57

185.0 194.8 204.0 212.6 220.6 227.9 255.3 263.6 259.5

70.9 83.2 94.9 105.9 116.4 126.2 166.1 199.3 199.0

100% 2.380 55% 200% 3.039 65% 400% 4.358 76% 430% 4.555 77%

Rs. 6 which is tantamount to a 50% increase in the current average retail price. The expenditure spent on bidis starts falling only after the tax increase is more than Rs. 25 per 1000 sticks. Revenue from taxation of cigarettes keeps increasing until tax is almost 76% of the retail price, which effectively means that the tax on an average cigarette can be increased to Rs. 3300 per 1000 sticks compared with the current Rs.659 without fear of losing revenue. The expenditure spent on cigarettes would start falling only after the tax reaches at least Rs.3300 per 1000 sticks. Effectively this would mean a 150% increase on the current average retail price of cigarettes. Even though consumption of both bidis and cigarettes falls for each small increase in taxes, substantial increases in retail price through tax are required for the expenditure to decrease. This is important in the context of India where expenditure on tobacco has crowding-out effects. While these results are subject to the strong assumption of constant price elasticity, they nevertheless point to the potential for increasing taxes in order to curb consumption without losing revenue. This analysis shows the huge potential of taxation as a way of both generating revenue and curbing consumption.

Discussion and conclusion

This paper provides the first ever estimates of own- and crossprice elasticities of bidis, cigarettes and leaf tobacco in India. It was found that a doubling of a households total expenditure

would have the effect of raising the average price paid for cigarettes by 11% in rural and 24% in urban India. Cigarettes, unlike other tobacco products, were found to be luxury goods in both rural and urban India with income elasticity greater than one. Estimates of own- and cross-price elasticities showed that own-price elasticity estimates of different tobacco products in India ranged between 0.4 and 0.9, with bidis and leaf tobacco having own-price elasticities close to unity. Cigarettes were the least price elastic of all. Analysis of the cross-price elasticities revealed that bidis are a complement to cigarettes. With certain assumptions, it is shown that taxes on cigarettes and bidis can be raised to many times higher than the existing rates without fear of losing tax revenue, which reveals the potential of using taxation as an effective way for both regulating tobacco use and generating tax revenue. The analysis here provides strong support for taxing tobacco products whether it is bidis, cigarettes or leaf tobacco. However, as mentioned in the introduction, taxation of tobacco in India has been predominantly on cigarettes. The tax on bidis of Rs. 14 per 1000 sticks, compared with Rs.819 per 1000 of the least taxed filtered cigarette, is negligible in comparison. Such low taxes on bidis are certainly the most important reason why bidis have such high price advantage over cigarettes, making them one of the cheapest tobacco products in the world. It is then no surprise that tobacco consumption in India is unique with a very high presence of non-cigarette tobacco, especially bidis. It is not clear why cigarettes bear most of the burden of tax on tobacco while bidis and other tobacco products are either very lightly taxed or not taxed at all. There are many arguments for it. One argument is that bidis are largely consumed by economically weaker sections of the community and taxing bidis could be interpreted as being regressive. Attempts to raise tax on bidis are often interpreted as an attack on the poor and have been regarded as politically inexpedient. Secondly, it could also be that taxes on cigarettes are much easier to administer as there are only three to four cigarette manufacturers in India accounting for more than 95% of the production (ERC 2005). Production of bidis on the other hand is highly fragmented and the bidi rolling is done mostly by female home-based workers under contractors, making it extremely difficult to administer excise taxes (Sunley 2008). Concerns about the employment of an estimated 4 million people, 76% of them women and some 8.4% children, who are directly dependent on bidi rolling (Panchamukhi et al. 2007), could be yet another reason why the bidi industry is protected from larger taxes. Whatever the reason, any attempt to regulate tobacco use in India without taking effective measures to reduce consumption of bidis will not produce significant results. Moreover, since bidis are at least as harmful as cigarettes, increasing the tax on bidis to match that of cigarettes is justifiable on health grounds, as noted by Sunley (2008). An increased tax on bidis need not be seen as regressive, as it would have multiple benefits. First, increased taxes would directly translate to reduced consumption which would have direct health benefits. Secondly, as high taxes would lead to a substantial decrease in consumption, it would also have the effect of reducing money spent on consuming tobacco, provided the tax is large enough. Reduced consumption coupled with reduced expenditure would leave the poor with extra disposable

Downloaded from heapol.oxfordjournals.org by guest on August 28, 2011

208

HEALTH POLICY AND PLANNING

income. Finally, tobacco taxes have the potential to increase revenue, a portion of which can be earmarked for rehabilitating poor households who are directly dependent on bidi rolling. There is also no clear reason why there is a huge difference between the rates of tax on cigarettes based on the presence of filters and the length of cigarettes in India. The lowest taxed filtered cigarettes are taxed at Rs.819 per 1000 sticks, while the lowest taxed non-filtered cigarettes are taxed at only Rs.168 per 1000 sticks. According to the National Institute of Cancer, there has been no conclusive scientific study which has shown either that filtered cigarettes are less harmful than non-filtered cigarettes or that any changes in cigarette design resulted in a decrease of disease burden associated with smoking. So cigarettes advertised as light, low-tar, filtered etc. would not actually reduce the harm to a smoker, on the contrary it only gives the smoker an illusion of risk reduction (National Cancer Institute 2001). As it is hard to justify different tax rates based on the presence of filters, it is also hard to justify different tax rates based on the length of cigarettes. The existing duties on filtered cigarettes range from Rs.819 to Rs.2163, as shown in Table 1. One can argue that an addicted smoker would smoke more short cigarettes to take advantage of the lower prices and be able to obtain the same level of tar and nicotine as from a long cigarette. Hence no justification can be provided for having different tax rates either on the basis of the presence of a filter or on the basis of cigarette length. Differentiating cigarettes on such grounds for the purpose of taxation is unwarranted and should be abolished, if regulating tobacco use is one of the objectives of tobacco taxation. This would simplify tax structure for cigarettes, improve the tax administration, and above all would help to communicate the public health message that all cigarettes, regardless of shape and size, are harmful and should be taxed heavily. Past studies (Efroymson et al. 2001) have shown the crowding out potential of tobacco expenditure. Rigorous empirical examination of tobacco expenditures data in India (John 2008) has also shown that increased expenditure on tobacco might result in reduced expenditures on essential commodities such as food and education. Given that price elasticity of tobacco products is less than one, one may argue that an increase in taxes will only result in increased expenditure on tobacco and consequent crowding out. However, as shown in Table 6, a substantial increase in tax has the ability to decrease the expenditure, yielding positive benefits for the tobacco users who reduce consumption. This points to the need for defining the objectives of tobacco taxation itself as a public health policy tool over and above being a tool to generate revenue. It is important to devise a tax policy for all tobacco products, including bidis, cigarettes and chewing tobacco products, that will result in a comprehensive reduction in tobacco use instead of focusing on one particular tobacco product which comprises only a small portion of tobacco used in India.

households as a whole whereas much of the tobacco consumption in India is by adults, especially adult males. Since tobacco is such an important public health problem, collecting data on tobacco consumption habits at the individual level should be a priority in future surveys. Secondly, bidis are smoked by poor smokers and have an estimated price elasticity near unity, whereas cigarettes are smoked by wealthier smokers and, although their price and tax are much higher than for bidis, have much lower price elasticities, which suggests differential price elasticity by income levels. Price elasticity for tobacco is known to be higher among the poor compared with the rich (Farrelly et al. 2001). The exercise in this paper, however, does not distinguish between different income groups. Since price information is not directly available, the model depends on the variability of unit values between clusters. Applying the same model to different income groups separately will not yield sufficient variability in prices. Knowledge of differential price elasticities would be particularly useful in the context of India wherein economic disparities determine the nature of tobacco consumption itself. Collecting large-scale data conducive for detailed economic analysis should be a priority in this respect.
Downloaded from heapol.oxfordjournals.org by guest on August 28, 2011

Acknowledgements
This work was supported in part by a scholarship from Indira Gandhi Institute of Development Research and in part by a postdoctoral fellowship (CA-113710) from University of California San Francisco. I am indebted to Dr. A. GaneshKumar, Prof. Kirit Parikh and Dr. Joy de Beyer for their valuable comments and suggestions. I also thank the participants of 67th Health Economists Study Group Meeting, held at University of Newcastle upon Tyne, UK (29 June 2005) and participants of the 13th World Conference on Tobacco or Health held at Washington DC, USA (12 July 2006) for their useful comments on earlier drafts of this paper. I have also benefited from the discussions with Cecily S. Ray. All errors, however, remain my own.

Endnotes
1

Limitations
Though the estimates presented here are useful guides to devise suitable tax policy measures on tobacco, it should be noted that the conclusions of the paper are not completely without weaknesses. Much of the weaknesses emerge from the inadequacy of the data itself. First, the estimates are for the

Bidi is an indigenous tobacco preparation in India made by rolling a dried piece of Temburini leaf (Diospyros melanoxylon) with 0.15 to 0.25 g of sun-dried, flaked tobacco into a conical shape and securing the roll with a thread. Bidis contain only a small amount of tobacco compared with cigarettes, yet they deliver as much as 4550 mg of tar and 1.742.05 mg of nicotine compared with 1828 mg and 1.551.92 mg of tar and nicotine, respectively, in Indian cigarettes (Gupta et al. 1992). 2 The exchange rate for Indian Rupees was roughly Rs.40 to one US Dollar during this period. 3 See Alderman (1988) where such an exercise is done for Pakistan. 4 Villages are the First Stage Units (clusters) in the NSSO surveys. For urban areas they are referred to as urban Frame Survey Blocks. There are 6018 villages and 4122 blocks in the data. 5 NSSO divides the entire geographical region of the survey into 78 regions which are called NSSO regions. 6 The budget share equation here closely follows the one suggested by Working (1943) with an extra price term and household demographic terms. This has the theoretical advantage of being consistent with a utility function (Deaton 1997). Though the

PRICE ELASTICITY ESTIMATES FOR TOBACCO PRODUCTS IN INDIA

budget share equation resembles the Almost Ideal Demand System, it is not the same. Note that by distinguishing households according to clusters, the model has a data structure that is similar to a panel data regression, although coming out of a single cross-section. Symmetry-constrained estimates guarantee the unique substitution complementary patterns, ruling out the possibility that good i is a substitute of good j when j is a complement of i. See John (2008) using the same data as in this paper. See also Vermeulen (2003) for a similar exposition using different data. Demand elasticities for tobacco products with all households included are available upon request. Only the results for tobacco consuming households are reported here. There is, however, no conclusive evidence that non-filtered cigarettes are more hazardous than filtered ones. The conclusions from the simulation are not very sensitive to the initial estimate of the consumption.

209

National Sample Survey Organization, Ministry of Statistics and Programme Implementation, Government of India. Guindon GE, Perucic AM, Boisclair D. 2003. Higher tobacco prices and taxes in South-East Asia. HNP Discussion Paper, Economics of Tobacco Control Paper No.11. Washington, DC: World Bank. Gupta PC, Hamner JE, Murti PR (eds). 1992. Control of tobacco-related cancers and other diseases. Bombay: Oxford University Press. International Institute of Population Sciences (IIPS), Macro International. 2007. National Family Health Survey (NFHS-3), 2005-06: India Volumes I & II. Mumbai: IIPS. John RM. 2006. Households tobacco consumption decisions: evidence from India. Journal of South Asian Development 1: 10126. John RM. 2008. Crowding-out effect of tobacco expenditure and its implications on household resource allocation in India. Social Science and Medicine 66: 135667. National Cancer Institute. 2001. Risks associated with smoking cigarettes with low machine-measured yields of tar and nicotine. Smoking and Tobacco Control Monograph No. 13, NIH Publication No.02-5074. Washington, DC: US Department of Health and Human Services, National Institutes of Health, National Cancer Institute. NSSO. 2000. Socio-economic survey, 55th round: July 1999-June 2000, household schedule 1.0: Consumer expenditure. New Delhi: Government of India. Panchamukhi PR, Woollery TA, Nayanatara SN. 2007. Economics of bidis in India. In: Asma S, Gupta PC (eds). Bidi smoking and public health: a monograph. Navi Mumbai: Healis Sekhsaria Institute of Public Health, India; The Centers for Disease Control and Prevention, Office on Smoking and Health, USA; Tobacco Free Initiative, WHO, Switzerland (in press). Sarma A. 2000. Regulating tobacco use: role of taxes. Economic and Political Weekly 35: 461316. Sunley EM. 2008. India: The tax treatment of bidis. Online at: http:// tobaccofreecenter.org/files/pdfs/reports_articles/SunleyBidiTax%20FINAL-5Feb08.pdf, accessed on 20 Feb 2008. Bloomberg Initiative to Reduce Tobacco Use. Suryanarayana MH. 2002. Household consumption of tobacco in India: an economic perspective. Paper presented at the seminar, Tobacco Control Research in India, organized by the World Bank and Tata Institute of Fundamental Research, 10-11 April 2002, New Delhi, India. Vermeulen F. 2003. Do smokers behave differently? A tale of zero expenditures and separability concepts. Economics Bulletin 4: 17. Working H. 1943. Statistical laws of family expenditure. Journal of the American Statistical Association 38: 4356.

10

11

12

References
Alderman H. 1988. Estimates of consumer price response in Pakistan using market prices as data. Pakistan Development Review 27: 89107. Cawley J, Markowitz S, Tauras J. 2004. Lighting up and slimming down: the effects of bodyweight and cigarette prices on adolescent smoking initiation. Journal of Health Economics 23: 293311. Chaloupka FJ, Warner KE. 2000. The economics of smoking. In: Culyer AJ, Newhouse JP (eds). The handbook of health economics, vol. 1B. Amsterdam: Elsevier Science, pp. 1539627. Chaloupka FJ, Hu TW, Warner KE, Jacobs R, Yurekli A. 2000. The taxation of tobacco products. In: Jha P, Chaloupka FJ (eds). Tobacco control in developing countries. Oxford and New York: Oxford University Press, pp. 23772. Deaton AS. 1988. Quantity quality and spatial variation of price. The American Economic Review 78: 41830. Deaton AS. 1997. The Analysis of Household Surveys. Baltimore, MD: Johns Hopkins University Press for the World Bank. Efroymson D, Ahmed S, Townsend J et al. 2001. Hungry for tobacco: An analysis of the economic impact of tobacco consumption on the poor in Bangladesh. Tobacco Control 10: 21217. ERC. 2005. The World Cigarettes I: The 2005 Report. Newmarket, Suffolk, UK: ERC Statistics International plc. Farrelly MC, Bray JW, Pechacek T, Woollery T. 2001. Response by adults to increases in cigarette prices by sociodemographic characteristics. Southern Economic Journal 68: 15665. Government of India. 2006. Level and Pattern of Consumer Expenditure, 2004-05. NSS 61st Round (July 2004-June 2005). New Delhi:

Downloaded from heapol.oxfordjournals.org by guest on August 28, 2011