Title: Mixture of autoregressive moving average models
Authors: Hien Nguyen - La Trobe University (Australia) [presenting]
Abstract: In real-world modelling, time series are often heterogeneous and may take on the features of different regimes over different points in time. Modelling these possible different regimes can be done using a mixture variant of the traditional method of autoregressive moving average modelling. In recent times, autoregressive variants have been proposed due to the ease of estimation of such models using maximum composite likelihood estimation EM procedure. However, the moving average elements cannot be so easily estimated in in this way. We demonstrate how new advances in automatic differentiation can be used to conduct such estimation.