A Procedure for Obtaining M‐estimates in Regression Models With Serially Dependent Errors

Applications where error terms in a regression model display both non‐normal and serially dependent behavior are considered. For the estimation of the parameters, an iterative Cochrane‐Orcutt type M‐estimator is proposed. A proof of convergence of the iterative procedure is given. In a simulation experiment, where the least absolute error criterion is applied, the performance of the estimator is tested and the theoretical convergence properties illustrated. In particular, the existence of multiple stationary points in an iterative process is discussed.