10:00-11:00 (followed by coffee)
Multivariate time series analysis by mixture modelling
Petros Dellaportas, Athens University of Economics and Business (joint work with Leonardo Bottolo, Imperial College London)
Abstract: We exploit the partial exchangeability structure of the parameters of different time series models to borrow strength for parameter estimation. We adopt a challenging reversible jump MCMC scheme which models the parameters as a finite mixture of normals with unknown number of components. We discuss in detail the careful choice of prior parameters and the construction of the reversible jump algorithm. We illustrate our methodology with stock returns from an S\&P 100 dataset and we find that our forecasts are more robust and offer better out-of-sample forecasting power when compared with those of univariate models.