The probability of non-purchasing tobacco of a smoker

This paper addresses the problem of zeroes in tobacco expenditure. Generally, tobacco demand is estimated using limited dependent variable models, i.e. Tobit or Double Hurdle Models, which take into account the zero expenditure problem under the assumption that a relatively important number of smokers declared a zero in tobacco expenditure. Clearly, if all zeroes where from non-smokers then demand estimation could be done using traditional methods over the positive expenditure observations. Based on the Spanish Expenditure Survey we estimate the conditional probability of non-expenditure by a smoker, finding that such probability is extremely small. This suggests that smokers buy quite regularly and hence it is possible to estimate the tobacco demand using only the positive observations.