Prediction threshold filtering

This paper is concerned with a modification of the robust filtering algorithm of kleiner, martin, and thomson (1979). this improved version ptf (for prediction threshold filtering) is able to discriminate between outliers of the innovations and of the additive type ( io and ao ). since the effects of the different outliers species on parameter estimation and prediction are different, suggesting data correction for ao and no correction for io, this discrimination could prove important in practice. this paper provides some evidence on the qualities of ptf. first , the power of the filter is investigated using numerical formulae. this approach faces a limit when the assumption of known true parameters is dropped . next, monte carlo simulation is used to reveal patterns based on the temporal dependence of series and decisions which are excluded from the numerical tables. last, some theoretical lemmata explain some of the observed features and extend the results given in the first part of the paper .;