Stochastic approximation type estimators in linear models

A new class of stochastic approximation type estimators, different from those obtained via Robbins-Monro procedure, is introduced. Their asymptotic properties are studied. The proposed estimator is on the n-th step defined as the one step version M-estimator, where the estimator from the previous step is used as a preliminary one. This type of estimators is particularly useful in testing of constancy of the regression relationship over time.