Minimum distance estimators for count data based on the probability generating function with applications (Q2412758)

Consider a \(d\)-dimensional count data based on a random sample with associated probability generating function (PGF) \(P(\mathbf{t})\). Define a distance measure \(D^{2}_{\alpha}(P_1, P_2) = \int\{P_1^{\alpha}(\mathbf{t})- P_2^{\alpha}(\mathbf{t})\}^2 w(\mathbf{t}) d\mathbf{t}\) between two PGFs \(P_1\) and \(P_2\) with suitably defined nonnegative weight function \(w(\cdot)\) and non-zero \(\alpha\). This paper establishes strong convergence and asymptotic normality of estimators of parameters of a parametric family that minimize the distance \(D^{2}_{\alpha}\) between the empirical and assumed PGFs. The joint limit distributions of such estimators from two families of distributions are also presented. These results and the minimum \(D^{2}_{\alpha}\) criterion are applied to three classical inference problems -- (a) goodness-of-fit tests, (b) model selection, and (c) testing for separate families of distributions. The model selection results parallel the work of \textit{Q. H. Vuong} [Econometrica 57, No. 2, 307--333 (1989; Zbl 0701.62106)], and the results for (c) follow the work of \textit{D. R. Cox} [J. R. Stat. Soc., Ser. B 24, 406--424 (1962; Zbl 0131.35801)]. The results are applied to commonly used count data models.