A1340
Title: Efficient outlier detection in heterogeneous time series databases
Authors: Pedro Galeano - Universidad Carlos III de Madrid (Spain) [presenting]
Daniel Pena - Universidad Carlos III de Madrid (Spain)
Ruey Tsay - The University of Chicago Booth School of Business (United States)
Abstract: A fast and powerful approach is proposed to identify univariate outliers in a large dataset of time series. The approach is highly flexible, as it can handle databases with different definitions, frequencies, and sample sizes across the marginal series. The proposed method examines the residuals of the observed series after a robust model fitting to detect outliers, uses saturated regression models to consider all observations as potential outliers, and uses the orthogonal greedy algorithm to detect significant outlying effects. The method is automatic and has been implemented to run in parallel in the texttt R package, allowing fast and efficient identification of outliers in large datasets. The performance of the proposed procedure is investigated by several simulations and the analysis of an empirical example.
