A comparative study of robust and stable estimates of multivariate location (Q2314463)

Let \(x_1,x_2,\ldots,x_n\) be a random sample of a distribution with density \(f(x-\theta)\), where \(\theta\) is a location parameter. In the case of symmetric density, the center of symmetry \(\theta\) can be estimated by \(M\)-estimate \(T_n\) which is defined as a solution of the following minimization problem \[ T_n=\mathrm{arg}\,\min_\theta\sum_{i=1}^n \varrho(x_i-\theta), \] where \(\varrho\) is a \textit{contrast function}, i.e. an even non-negative function. If the score function is differentiable, then the \(M\)-estimate \(T_n\) is a solution of the equation \[ \sum_{i=1}^n\psi(x_i-T_n)=0, \] where \(\psi=\varrho'\) is the so-called \textit{score function}. The authors of the paper consider properties of general \(M\)-estimates and particular \(M\)-estimates generated by the specially chosen score functions. They present the comparative analysis of a variety \(M\)-estimates by considering random samples of the generalized Gaussian, Student and Tukey gross-error distributions. The multivariate case of these distributions is also studied.