Developing applications of Global Vector Autoregressive models in financial institutions

Research questions
- What are the possible applications of the GVAR models and how it can be used in financial institutions?
- In particular, how do we specify the model for our purpose? Also how to handle post crisis data as they are more turbulent and may have structural breaks.
- How does the GVAR compare to other global macroeconometric models such as Simultaneous equation models (SEMs) and Dynamic stochastic general equilibrium models (DSGEs) etc.
Research Objectives
- To examine the existing and potential applications of the GVAR model.
- To build upon existing applications of the GVAR model in literature and also from the author's previous dissertation
- To evaluate such applications with comparisons to other global models such as SEM; DSGE.
Prior to this research, most applications are devoted to policy and modelling functions in central banks. Interviews with industry practitioners had confirmed that the GVAR model would be valuable in additional to their existing models.
Upon literature reviews on standard practice in risk management such as the Handbook by Professional Risk Managers' International Association; Moody's KMV Expected default frequency (EDF), CreditMetrics value-at-risk (VaR) by J.P Morgan, we can see that the approach to risk is mostly probabilistic with emphasis to the events' history. Often such approach emphases on correlation between assets and do not link the spill over effects from foreign countries. In fact this was the main motivation when the GVAR was proposed in (Pesaran, Shuermann and Weiner, 2004) with the intention of assessing the impact of credit risk globally.
Following the example in (Dee et el, 2005), we can compute a general model that is adaptable for the purpose of international shock transmission analysis. This is particularly useful for those who work in risk management in central banks and also commercial banks. Portfolio managers would like to analyse their exposures if such economic shocks happen.
A generic model of the VARX* (2,2) i.e. two countries; two lags, for each country in the GVAR is represented by:

Where X is the vector for the domestic economic variables; t being a time variable; and X* for foreign variables. One notable observation on the literature; little attention is paid to the selection of variables in the model. Understandingly, this is the main feature of the GVAR, which allow the users to dump a large amount of variables into the model without worrying too much about the correlation between the residuals and contemporaneous effects.
For our purpose, potential economic variables in the model include: real output (GDP); inflation; real exchange rate; real equity price; short-term interest rate; long-term interest rate and common global factor such as oil price. Countries would include all major economies such as US; China; Japan; UK; Germany; and other countries. Literature review had suggested that potentially any countries that have credible data sources could be employed in the GVAR as the model is no longer restricted by size. The main challenge is dealing with the data structure itself.
As the model in Dee et el was in 2005, prior to the great recession in 2008. This would give rise to challenges in incorporating such turbulent data into the model with structural breaks. Recent euro crisis in 2012 and Brexit in 2016 would also increase the complexity to the model. This will be dealt will comprehensive testing of structural breaks; trade weights determination across countries; lag selection of the model. Outcomes are given in Generalized Impulse Response Analysis (GIRF) and Generalized Forecast Error Variance Decomposition (GFFVD). In the end we could answer questions related to international shock transmission.