GLASS - Global Augmented State Space Error Correction Model

Many economic or financial data, like gross domestic product, inflation or exchange rates, are naturally available as time series, which means that their values are ordered in time. The data frequently contain information on temporal dependencies as well as on how different time series are interrelated. This information is identified and described using econometric time series models. The results are important amongst others for policy advice and economic forecasting. European integration and globalization have strongly increased the interdependencies between countries and time series models have to take into account more complex relationships between countries. Increased computing power and better availability of data allow by now to develop and use such models. Time series models depend on unknown parameters that have to be estimated from data. The usage of more complex models increases the number of parameters drastically, which often implies that the data available may no longer be sufficient to estimate all of these parameters. This is known as curse of dimensionality, a phenomenon appearing in many areas of science and technology. Consequently, simplifying assumptions which lead to parameter reduction have to be made and tested. Economics uses various model classes based on different strategies of parameter reduction (all with strengths and weaknesses): Dynamic Stochastic General Equilibrium (DSGE) models include a small number of parameters by being heavily based on the integration of economic theory while models like Generalized Dynamic Factor Models (GDFMs) achieve parameter reduction based on purely statistical methods. A third class of models, Global Vector Autoregressive (GVAR) models, falls in between these extremes. Here, complexity is reduced in part through structural assumptions based on economic theory, and in part through statistical methods. GVAR models are easy to handle, but very restrictive. Consequently, the GLASS project proceeds by embedding GVAR models in the class of Global Augmented State Space models. This is a comprehensive model class with the additional advantage that the representation as a state space model can by itself reduce the number of free parameters as well as facilitate the use of restrictions based on economic theory. In the project, a comprehensive methodology, ranging from representation to estimation, will be developed. This will allow to apply GLASS models to a number of economic problems. Tested software will be made available to support the dissemination of the methodology developed.