I gave a seminar this week at ICRIER (Indian Council for Research on International Economic Relations) where I argued that it is today computationally feasible to implement a risk management system for derivative exchanges that is (a) based on Expected Shortfall, (b) incorporates fat tailed distributions and (c) computes portfolio risk across multiple underlyings (securities or commodities) using non linear dependence models (copulas).
Risk measures like Value at Risk, SPAN and Risk Metrics have their intellectual roots in the early 1990s or earlier when the notion of coherent risk measures had not been developed and risk modelling had not yet embraced fat tailed distributions with non linear dependence structures. For example, current global best practice in handling exposure to multiple underlyings (“inter commodity spreads”) in exchange risk management can only be characterized as crude and ad hoc. Their continued popularity owes much to the inadequacies of correlation based dependency modelling. Similarly, the SPAN framework uses too few scenarios to meet the highly desirable “relevance axiom” for risk measures though computational advances allow us to come very close to fulfilling this axiom.
In India, the regulatory framework for risk management at Indian exchanges is still supposed to be based on the 99% value at risk mandated by the L C Gupta Committee a decade ago. In practice, however, Indian exchanges and their regulators have adopted several features of a fat tailed expected shortfall approach. Risk management practice has thus outgrown the regulatorily mandated value at risk to which it still pays lip service. The time has come to formally discard value at risk from the regulatory lexicon and adopt a more modern vocabulary. This would provide an opportunity to spur new research on improving exchange risk management systems.
My presentation made specific proposals for a modern risk management system and indicated directions for further research. The slides of this presentation are available here.