Effect of the compensation in abrupt model change detection problem

Abstract

After having detected changes in an ARMA model, one of the problems consists in determining how to compensate for these changes in order to carry on tracking. Different detection techniques are compared and two common compensation of abrupt changes methods in an ARMA model are presented: the first one uses the Fisher information matrix and the second one the parametric error covariance. These compensation methods have a drawback: they can give an important weight to a direction without any failure. A new compensation technique is thus proposed. The compensated detection techniques are compared with each other and discussed.