ROBUST AUTOREGRESSIVE ESTIMATES USING QUADRATIC-PROGRAMMING

The robust estimation of the autoregressive parameters is formulated i n terms of the quadratic programming problem. This article's main cont ribution is to present an estimator that down weights both types of ou tliers in time series and improves the forecasting results. New robust estimates are yielded, by combining optimally two weight functions su itable for Innovation and Additive outliers in time series. The techni que which is developed here is based on an approach of mathematical pr ogramming applications to 1(p)-approximation. The behavior of the esti mators are illustrated numerically, under the additive outlier generat ing model. Monte Carlo results show that the proposed estimators compa red favorably with respect to M-estimators and bounded influence estim ators. Based on these results we conclude that one can improve the rob ust properties of AR(p) estimators using quadratic programming. (C) 19 97 Elsevier Science B.V.