On the properties of Hermite series based distribution function estimators (Q2036312)

Hermite series based distribution function estimators have recently been applied in the context of sequential quantile estimation. In this article, the authors generalize their previous theoretical results pertaining to the Hermite series based distribution function estimator, and further explore the CDF estimator through real data and simulation studies. The particular contributions of this article are as follows: \begin{itemize} \item Previous theoretical analysis of the first author, applicable only to the positive half real line estimator, is generalized and new asymptotic convergence results in the mean squared error, mean integrated squared error and almost sure sense, along with rates, for the full real line Hermite series based distribution function estimator are derived. \item Examples of distributions for which the rate of MSE convergence approaches the optimal rate, \(O\big(n^{-1}\big)\), i.e. the normal distribution and finite normal mixture distribution, are provided. \item Bias-robustness of the full real line Hermite series based distribution function estimator is proven, and contrasted with a closely related estimator based on the Gram-Charlier series. \item Finite sample properties of the Hermite series based estimators are investigated through a real data example and simulation study. \item Properties of the Hermite series based distribution function estimator are compared to the smooth kernel distribution function estimator. It is established that in the general (non-sequential) setting the Hermite series based estimator is inferior to the smooth kernel distribution function estimator. \end{itemize}