Estimating the inverse autocorrelation function from outlier contaminated data (Q1424610)

Let \(X_t\), \(t\in (-\infty,\infty)\), be a stationary process with zero mean and spectral density \(f(l)\). Denote by \(R_i(u)=\int_{-\infty}^{\infty}\exp(ilu)/f(l)\,dl\), \(u=\dots,-1,0,1\dots\), the inverse autocovariance function. Robust estimates of \(R_i(\cdot)\) can be applied to a wide range of problems which include robust model selection for time series. This paper demonstrates how a number of commonly used estimates of the inverse autocovariance function can be modified to deal with outlier contaminated data. The robust versions of the orthogonal and interpolation procedures appear to be new and provide an alternative to the robust autoregressive approach suggested by \textit{R. D. Martin} [Lect. Notes Math. 757, 117--143 (1979; Zbl 0423.62073)]. The author studies the performance of procedures in a large scale of numerical experiments. The results show significant improvement in performance of robust estimates in the case where the data are contaminated by additive outliers, while there was no uniformly best robust technique. The experiment supports the use of the autoregressive approach to avoid catastrophic reduction in performance, and robust interpolation for short series corrupted by few outliers.