A second-order asymptotic distributional representation of M-estimators with discontinuous score functions (Q1097600)

The authors consider a sequence \(\{X_i;i\geq 1\}\) of independent and identically distributed random variables with distribution function \(F(x-\theta)\), where \(\theta\) is an unknown location parameter. An \(M\)-estimator \(\widehat\theta_n\) of \(\theta\) is defined as a solution of the equation \[ M_n(t)=\sum^n_{i=1}\psi(x_i-t)=0,\quad t\in R^1,\;n\geq 1, \] where \(\psi:R^1\to R^1\) is the score function such that \(\psi(x)=\psi_1(x)+\psi_2(x)\), for \(x\in R^1\); \(\psi_1\) is a function absolutely continuous on any bounded interval in \(R^1\), \(\psi_2\) is a step-function and \[ \lambda(0)=0,\text{ where }\lambda(t)=\int_{R^1}\psi(x-t)\, dF(x),\quad t\in R^1. \] For the \(M\)-estimator \(\widehat\theta_n\), it is presented a first-order asymptotic representation of the form \[ n^{\frac12}(\widehat\theta-\theta)=n^{-\frac12}\gamma^{-1}M_n(\theta)+R_n;\;R+n=o_p(1). \] The paper is concentrated on the study of the above asymptotic representation supplemented by the study of the asymptotic distribution of the remainder term and it is shown that, typically, the asymptotic distributioon of \(R_n\) is nonnormal. Asymptotic representations supplemented by asymptotic distribution of the remainder term are referred to as the second-order asymptotic distributional representation. Finally, the results are extended to one-step versions of the \(M\)-estimator \(\widehat\theta_n\). These one-step versions are a convenient approximation to \(\widehat\theta_n\).