Wednesday 20 March 2013, 4.15PM to 5.45pm
Speaker(s): Dr Matthew Greenwood-Nimmo, University of Melbourne
Abstract: One of the key benefits of the Global VAR framework is that it largely circumvents the so-called curse of dimensionality, thereby providing a means by which to estimate large high dimensional models using available macroeconomic datasets. However, in turn it may introduce a secondary curse of dimensionality in relation to the high volume of statistical outputs that it may generate. In many cases, the limits of the modeller’s ability to process the output will become a binding constraint. We therefore differentiate between the traditional curse of input dimensionality and our proposed curse of output dimensionality. This paper develops a family of generalized connectedness measures (GCMs) designed to summarise the output of large multi-country multi-variable models at a user-defined level of aggregation and thereby provides a means to alleviate the curse of output dimensionality. Our GCMs are derived from normalized generalized forecast error variance decompositions in a manner that extends the multi-country univariate connectedness measures developed by Diebold and Yilmaz (2011). We demonstrate the usefulness of our approach using the 26 country 172 variable GVAR model developed by Greenwood-Nimmo, Nguyen and Shin (2012, Journal of Applied Econometrics).
Location: Economics Staff Room (A/EC/202)
Admission: All welcome to attend