Title: Dynamic Bayesian estimation of time-varying cointegration parameters
Authors: Basile Marquier - University of Sheffield (United Kingdom) [presenting]
Abstract: Many models involving financial or macroeconomic models are changing over time and cointegration offers the possibility of identifying relationships in a set of economic time series. The aim is to estimate a time-varying cointegrated model. We discuss the possibility of time-varying cointegration by proposing a Forward Filtering Backward Sampling scheme for the parameters of the Vector Error correction model. The time-varying Vector Error Correction Model (VECM) is first derived from the dynamic VAR model. The long-run relationships matrix becomes a dynamic parameter and its rank, the cointegration rank, which is estimated from the cointegration matrix, is then also time-varying. Therefore we have the possibility of seeing the independent cointegration relations evolving over different time periods. This new dynamic approach is shown with several synthetic data sets, in which we change the cointegration relations and the rank over different time periods. Finally an application on different sectors of the Dow Jones is studied.