A1476
Title: Time-varying multi-seasonal AR models
Authors: Mattias Villani - Stockholm University (Sweden) [presenting]
Abstract: A seasonal AR model with time-varying parameter processes in both the regular and seasonal parameters is presented. The model is parameterized to guarantee non-explosive behavior at every time point and can accommodate multiple seasonal periods. Time evolution is modelled by dynamic shrinkage processes to allow for longer periods of essentially constant parameters and rapid jumps. A robust, fast and accurate approximate sampler based on the extended Kalman filter is proposed and compared to a particle MCMC sampler. The properties of the model are compared to several benchmark models on simulated data. An application to more than a century of monthly US industrial production data shows interesting changes in seasonality over time, particularly during the Great Depression and the recent COVID-19 pandemic.
