Practical small-sample asymptotics for regression problems

A general method for estimating the distribution of a solution to an estimating equation defined through as sum of independent random variables is proposed. Suggested expansions are valid for the distribution of approximated parameters from general regression models and are not different from saddlepoint approximations in the case of iid random variables. A main advantage of the proposed methodology is that in requires on the conditional cumulant-generating function to be specified.
1. INTRODUCTION Daniels's (1954) seminal paper on saddlepoint approximations in statistics spawned a great deal of research in the general area of small-sample asymptotic approximations. His original paper concentrated primarily [...]