Outliers in functional autoregressive time series

A method for identifying and estimating outliers in a time series is proposed, based on fitting functional autoregressive models. Both additive and innovation outliers may be defined. A simulation experiment and the analysis of some real data sets suggest that the proposed method is effective both for series following some nonlinear models, such as self-exciting threshold autoregressive or exponential autoregressive, and for linear series generated by autoregressive moving average processes.