Lest squares estimation of a zero-truncated count data regression model

A new approach to limited-dependent variable count data or other model types is considered. Instead of adopting maximum likelihood estimation based on a full distributional assumption or smoothing techniques and semiparametric estimation, the novel idea is to use an approximation to the probability of, say, the zero event. The approximation is based on moments and uses old results for the probability generating function. The approximation is evaluated in a small Monte Carlo experiment. In empirical models of choice set size for Swedish unemployed and of nationalization frequencies for developing countries the results indicate good performance both computationally and resultwise. The results indicate that already quite low order expansions are well-behaved and useful for estimation.