Data

The ITC Bangladesh Project was created in 2008 to evaluate the impact of tobacco control legislation in Bangladesh. The ITC Bangladesh Survey is a face-to-face survey conducted by trained interviewers from the Bureau of Economic Research at the University of Dhaka, Bangladesh, in collaboration with the ITC Project team at the University of Waterloo in Canada.

The analysis in this paper is based on data collected in Wave 1 and Wave 2 of the survey, conducted in 2009 and 2010, respectively. The Wave 1 Survey consisted of a nationally representative probability sample of 2510 adult cigarette and bidi smokers and 2116 adult non-smokers aged 15 years and older selected through a multistage cluster sampling design (sampling with probability proportional to population size at the levels of administrative units such as district, upazila/thana and village/ward). These respondents form a cohort. They were contacted again to answer follow-up surveys in 2010 with an attrition rate of 8.3%.

The smokers were oversampled for the purpose of generating a sufficiently large sample size of smokers. For the present analysis, we have limited the sample to cigarette smokers and non-smokers and excluded the bidi smokers. Thus, the final full sample size of pooled observations of cigarette smokers and non-smokers was 8507 and of cigarette smokers was 3652.

Bangladesh has ample room to raise cigarette tax

The current cigarette tax in Bangladesh is composed of 2 components collected at the producer level: a value added tax (VAT) of 15% of retail price and an excise tax, which is a supplementary duty (SD) imposed as a percentage of the retail price of cigarettes, that varies at different price ranges of cigarette packs of 10 sticks. Between 2009 and 2010, the price bands for the four tiers of cigarette prices were increased and the SD for each tier was raised by 1 percentage point (see table 1). The ranges of price bands are, however, not continuous. The gaps between successive tiers are shown in the row under each tier with corresponding percentage of smokers who reported prices in that range. In order to calculate the average SD, we imputed the tax rate for each price tier to the price gap above that tier up to the lower limit of the next higher tier, in view of the fact that larger percentage of the price reported by smokers falls in the gap above the designated price tier (see last column of table 1). The existence of gaps between price tiers is a definite advantage for the producers because they pay the tax rate corresponding to the lower tier until the retail price reaches the upper tier, while enjoying higher prices and thus greater profit.

After weighting by the number of cigarettes smoked per day as reported by individual smokers, we find that the average SD increased from 38% of retail price in 2009 to 45% of retail price in 2010 and average real price of a pack of 10 cigarettes increased from 17.4 to 19.3 Taka in 2009 prices (see table 2). The average SD and cigarette price are driven down by the concentration of smokers in the lowest two price tiers: 79.8% in 2009 and 76.6% in 2010.

Although the real price of cigarettes increased between 2009 and 2010, this was also a time of significant growth of 5.2% in the gross domestic product (GDP) per capita.12 During this time, the number of cigarettes smoked per day remained almost the same: 10.2 sticks per day in 2009 and 10.5 sticks per day in 2010 (table 2). It is likely that the negative effect of a modest price increase on inelastic cigarette demand was more than offset by a strong positive effect of income growth in Bangladesh.

The minor increase in the share of SD in cigarette price, with almost unaltered average consumption, indicates that the Bangladesh government has yet to gain control over cigarette prices and consumption. At the current rates of SD on cigarettes, the average share of SD in the purchase price of cigarettes has remained far below the WHO recommended level of 70%.13 Thus we find that there is ample room for increasing excise tax on cigarettes.

Econometric model of cigarette demand

In order to estimate the overall impact of a price change on cigarette consumption resulting from lower smoking prevalence as well as lower smoking intensity of existing smokers, we constructed a two-part model following the method of Cragg.14 The total price elasticity of cigarettes is estimated as the sum of the elasticities of smoking prevalence and smoking intensity. Despite having cohort data, we did not use panel data techniques to estimate the effect of price on cigarette demand that would hold unobserved individual level heterogeneity constant. When we ran a sensitivity analysis using a fixed effects model, the effect of price became statistically insignificant, which is attributable to lack of sufficient variation in the price of cigarettes within observations for the same smoker over the 2 years. However, we expected significant cross-sectional variation in price by geographic area (village), source of purchase and mode/volume of purchase (eg, pack or loose). Therefore, we undertook a pooled cross-sectional analysis of the Wave 1 and Wave 2 survey data. We took into account the within-person correlation of observations by correcting the standard errors (SEs) for repeated observations on the same individual using the cluster correction technique.

In the first step, we estimated the probability of smoking cigarettes as a function of price, demographic characteristics, indicators of socioeconomic status of individuals and rural/urban area of residence. The regression was weighted to adjust for the over-representation of smokers in the survey; 18.9% of the population of adults are cigarette smokers, while this share is 41.8% in the sample. We estimated the smoking probability using the probit model: 1where Φ(.) is cumulative normal distribution and u is random disturbance term.

The price elasticity of smoking participation (b_P) is obtained using the following formula: where φ(.) is the normal density valued at the average levels of the explanatory variables and the estimated parameters of equation (1) and β₁ is estimated from equation (1). Similarly, the income elasticity of smoking participation (b_I) is given by:

In the second step, we estimated the cigarette consumption equation conditional on smoking participation from the following weighted ordinary least squares (OLS) regression: 2where ln is natural logarithm and e is a random disturbance term. This log-linear specification of the conditional demand function is determined by using the Ramsey Regression Equation Specification Error Test (RESET).15 The coefficients of price and household income, α₁ and α₂, estimated from equation (2) need to be multiplied by the average price and income levels, respectively, to calculate the price and income elasticities of the conditional demand for cigarette consumption. The total price elasticity is given by b_P+α₁* Price and the total income elasticity is given by b_I+α₂* Income.

One criticism that is often raised in the context of the estimation of demand equation is that self-reported price is an endogenous variable due to the simultaneity of consumption decision and the self-reported price of consumers.16 The endogeneity of self-reported price can potentially create bias in the estimated effect of price on smoking decision of individuals and their daily cigarette consumption. In order to address this concern, the price variable was constructed by averaging the prices reported by smokers in a specific geographic area of residence (village) for each wave. This price was then assigned to smokers and non-smokers in that area.

In order to address the endogeneity problem, we also estimated a second set of equations using instrumental variable probit model for smoking participation and two-stage least squares (2SLS) model for daily consumption. In the first stage, the price was regressed on the tax variable given by the sum of the SD and the VAT rates along with other socioeconomic and demographic characteristics of individual respondents present in equations (1) and (2). The price variable predicted from the first stage regression was then used in the second stage estimation of smoking participation and daily consumption.

As the SD is ad valorem with four tiers corresponding to four price bands and the rates differ between the 2 years of the survey (table 1), we can identify the price variable in two waves with eight different SD rates. For example, if the price facing an individual was 8 Taka per pack in 2009, the value of the tax variable for that individual is the sum of the corresponding SD rate of 32% and the value added tax of 15%, that is 47%.

The coefficient of the tax variable in the reduced form regression for price is 75.18, which implies that if the tax rate increases by 1 percentage point, the average price per pack of 10 cigarettes increases by 0.75 Taka (see online supplementary table A3). The statistically significant coefficient of the tax variable indicates that it is highly correlated with price. This tax variable was also tested as a valid instrumental for price, as we found a very large F statistic (greater than 10) in the reduced form regression for price. According to Stock and Watson, if there is one right-hand-side endogenous variable, one can use the F statistic from the first-stage regression to test for the significance of the instrument if the F statistic should be greater than 10.17

All the equations were estimated for the tertiles of individuals stratified on the basis of the housing index representing their socioeconomic status (low, medium and high). Thus, we obtained price and income elasticity estimates for the overall population and for population subgroups by socioeconomic status.

Results of estimation

The participation equations for cigarette smoking obtained from probit and instrumental probit estimation are reported in tables 3 and 4, respectively. The conditional demand equation for daily cigarette consumption of smokers obtained from OLS and 2SLS estimation are reported in tables 5 and 6, respectively. The tables in the Appendix present the means of the variables used in the estimation (see online supplementary table A1), the reduced form equation for price used in instrumental variable probit model (see online supplementary table A2) and the reduced form equation for price used in the 2SLS model (see online supplementary table A3).

In the instrumental probit regression, the Wald test of exogeneity of regressors was used to test for the orthogonality of the unobserved disturbances in the decision to smoke and the price equation.18 The Wald χ² statistic reported in table 4 rejects the hypothesis of the exogeneity of regressors in the smoking prevalence equation, indicating that self-reported price is endogenous. Similarly, the orthogonality of the unobserved disturbances in the daily consumption of cigarettes and the price equation was tested in the 2SLS regression for the conditional demand function for all smokers. In this regression, the exogeneity of regressors was also rejected as indicated by the statistically significant robust regression F statistics in table 6.

The validity of the use of the tax variable as an instrument for self-reported price is indicated by the estimates of the coefficient of the tax variable that are statistically significantly different from zero in the reduced form price equations corresponding to the instrumental probit and 2SLS regression (reported in online supplementary tables A2 and A3, respectively). The large value of the robust F statistic (greater than 10) in online supplementary table A3 also shows that the tax variable is not a weak instrument for self-reported price. Therefore, we accepted the instrumental variables estimates of the coefficients of the price variable for smoking prevalence and conditional demand functions for the purpose of estimating the price elasticity of demand for cigarettes.

Given the statistical significance and negative sign of the estimates of the price coefficient and the validity of the instrument used in the instrumental variable probit and 2SLS regressions, we used the corresponding estimated coefficients of price and income to calculate the price and income elasticity at the mean price level (table 7). The total price elasticity was −0.49 for the full sample indicating that a 10% increase in the price of cigarettes is expected to lead to 4.9% reduction in cigarette consumption. Most estimates of price elasticity of cigarette demand in low-income and middle-income countries range from −0.5 to −1.0, while those for high-income countries tend to fall in the range of −0.25 to −0.5.19 The present estimate is on the lower side of this range and lies between the short run price elasticity of −0.41 and long run price elasticity of −0.57 obtained in a previous study in Bangladesh.9 Further, we obtained the instrumental variable estimate of the total income elasticity at 0.23 (table 7), which implies that 10% increase in household income is expected to lead to 2.3% growth in cigarette consumption

Impact of tax increases and changes in tax structure on cigarette consumption and revenue

Using the price and income elasticity estimates for cigarette smoking prevalence and conditional cigarette demand, we simulated tax increases and different tax structures to assess the effect of those changes on overall cigarette consumption and tax revenue. The results are reported in table 8. The baseline year was set in the fiscal year 2012–2013 with the existing four-tiered tax structure and the projection was made for 2013–2014. The SD rates are higher in 2012–2013 than those prevailing in the survey years 2009 and 2010 shown in table 1.

Three alternative tax structures were simulated to assess the possible impact of tax policy changes on cigarette consumption and revenue: (1) uniform ad valorem tax at the rate of 61% of retail price of cigarettes; (2) uniform ad valorem tax at the rate of 61% of retail price of cigarettes with a specific minimum of 20 Taka (in 2012–2013 prices) per pack of 10 cigarettes; and (3) uniform specific tax of 22 Taka (in 2012–2013 prices) per pack of 10 cigarettes.

These alternatives were chosen so that the average excise tax per pack of cigarettes is comparable across the three options as shown in the first row of table 8.

The baseline population size is 152 518 015 as projected for 2012 in the Bangladesh Population and Housing Census 2011 and the adult population constitutes 69% of the total population.3 The adult population size was multiplied with the cigarette smoking prevalence rate to estimate the number of cigarettes smokers in Bangladesh in 2012–2013. We projected the negative impact of the tax policy changes onto the number of adult smokers and the annual cigarette consumption using the price elasticities of smoking participation and conditional demand for cigarettes, respectively.

The annual rate of per capita GDP growth is 4.9% in 2013, according to the projection of the IMF.12 The income elasticities of cigarette smoking participation and daily cigarette consumption were used to project the positive impact of income growth on the number of smokers and conditional demand for cigarettes, respectively. In addition, we took into account the growth in the number of adult smokers driven by population growth at the annual rate of 1.24%. The price increases were adjusted for inflation at the annual rate of 8%.

The net changes caused by price and income growth in the number of smokers, annual cigarette consumption and tax revenue are presented in table 8. Overall, the annual projection reveals that cigarette consumption could be reduced and cigarette tax revenue increased significantly by the simulated changes in the tax rates and structure. The highest price increases and decreases in the number of smokers and annual cigarette consumption occur under the uniform specific tax system, while the highest revenue gain and tax share in the retail price occur under the uniform ad valorem tax system. Therefore, it appears that the revenue goal is better served with an ad valorem tax system while the public health outcome is improved under specific tax system. The ad valorem tax with a specific minimum could achieve greater reduction in consumption than the uniform ad valorem excise system and could also narrow the price gap between the lowest and the upper price bands.