A0554
Title: Robust estimation of the autocorrelation function in the presence of outliers.
Authors: Antonio Montanes - University of Zaragoza (Spain) [presenting]
Abstract: A robust and computationally efficient method is proposed for estimating the autocorrelation function (ACF) in time series data contaminated by outliers. While traditional estimators of the ACF perform well under normality, they are known to be highly sensitive to non-normality and outliers, leading to biased results. The Hurwicz estimator is revisited, which has been shown to be median unbiased under normal distributions and generalized error distributions. By interpreting the first-order autocorrelation as a ratio of centered normal variables, a robust estimator is derived based on the truncated Cauchy distribution, mitigating the issue of undefined moments inherent to the standard Cauchy. The methodology is extended to higher-order autocorrelations using pairwise transformations and robust statistics, maintaining computational simplicity. Simulation studies confirm the proposed estimators' strong performance in finite samples, even under severe contamination. An empirical example is also provided to demonstrate practical relevance. This approach helps to bridge the gap between robustness and computational efficiency, offering a viable alternative to more complex robust methods in the literature. It is particularly useful for large datasets where computational cost is a key concern.
