Title: Clustering locally stationary time series using quantile autocorrelations Authors:  Angel Lopez Oriona - King Abdullah University of Science and Technology (KAUST) (Saudi Arabia) [presenting]
Ying Sun - KAUST (Saudi Arabia)
Jose Vilar - Universidade da Coruna (Spain)
Abstract: Locally stationary time series frequently arise in various fields, such as environmental sciences, economics, and seismology. However, statistical methods for analyzing locally stationary time series remain underdeveloped. A clustering approach is presented for local stationary time series that uses a dissimilarity measure based on local estimates of the quantile autocorrelation function at each time point. This distance is then combined with a $K$-medoids-type minimization problem, which incorporates a penalty term driven by the neighborhood size considered for the local estimation. To solve this problem, a three-step iterative procedure that guarantees a decrease in the objective function at each iteration is proposed. Several simulations show that the method generally outperforms some natural benchmarks in terms of clustering accuracy. The potential of the approach is demonstrated through an interesting application involving real-time series.