Conditional Scenario Generation with a GVAR Model

Abstract

The stress-testing method formed an integral part of the practice of risk management. However, the underlying models for scenarios generation have not been much studied so far. In past practice, the users typically did not model risk factors for portfolios of moderate size endogenously due to the presence of "curse of dimensionality" problem. Moreover, it is almost impossible to impose the expert views for a future outcome of macroeconomy on the scenario generator without making ad-hoc adjustments.

In this thesis we propose a GVAR-based framework which allows an efficient simulation of risk factors for a complex multi-currency portfolio of various classes of assets conditioning on economic scenarios. Given reasonable sets of economic forecasts, the GVAR model anticipates the trend and codependency of the future path of portfolio risk factors and supports the production of meaningful results from risk analytics.