Abstract: A criterion is proposed for selecting the order of the linear interpolator of a stationary time series, which may be useful in the problems of missing values and outlier detection. The criterion is based on a mean-square error similar to that leading to the final prediction error criterion for autoregressive model identification, and is called the final interpolation error criterion. The behavior of the proposed criterion is illustrated by means of a simulation study.
Key words and phrases: Inverse correlations, linear interpolator, missing values, order selection criteria, outliers, time series.