We investigate the behaviour of S - estimators in the linear regression model, when the error terms are long-memory Gaussian processes. It turns out that under mild regularity conditions S - estimators are still normally distributed with a similar variance - covariance structure as in the i.i.d - case. This assertion holds for the parameter estimates as well as for the scale estimates. Also the rate of convergence is for S - estimators the same as for the least squares estimator and for the BLUE.
Linear regression model long - range dependence robustness