Robust time series analysis

This paper considers the problem of obtaining estimates of seasonal and trend components of time series. It proposes a simple computational procedure for finding those estimates which minimize the sum of absolute errors for an additive time series model. It also examines a multiplicative model with a compound relative error criterion. The proposed method, which yields estimates which are insensitive to outlying members of the series, is ideally suited to a management science context, especially when large numbers of time series must be analysed on a regular basis.