Vector ARMA Models and Macroeconomic Modelling: Some New Methodology and Algorithms

Most multivariate time series models currently used for policy analysis and forecasting, in particular in the area of macroeconomics, are variants of vector autoregressive (VAR) models. However, theoretical and empirical evidence of the inadequacy of VAR models is mounting. In this project, we advance the theory of multivariate time series and develop an algorithm for building a more general class of models that do not have the limitations of VAR models. We show that our methodology provides a smarter approach to extracting information from observed data, and it will lead to models that are more reliable for policy analysis and forecasting.