RiskMinds – Model Risk and Variable Dependency Assumptions
December 11, 2008
Professor Paul Embrechts gave the opening talk this morning on quantitative risk management. Main points:
- Paul thinks the largest recent failing is to take VaR for granted – he says that it is only a statistical estimate that has a range of values based upon what assumptions are made.
- Related to the above, whilst improving quantitative risk management is vital he states that risk management is about human judgement.
- He sees "pricing model uncertainty" – the risk that a model for pricing an instrument is badly formed or based on poor assumptions – as a key risk to be addressed going forward (echoing VaR post yesterday)
- On pricing model uncertainty, he pointed out what the regulators are proposing with the recent consultative document from the Basel Committee on fair value calculation practices (click here). This covers things like assessing model sensitivity under periods of market stress, challenging of models, understanding of suitability and appropriateness etc.
- On the incremental risk capital charge guidelines from the Basel Commitee (click here) he raised concerns that a calculation for market risk with a VaR of 99.9% over one year was both very difficult to model and very difficult to backtest (without lots of good data being available, which is not usually the case).
- He ended by spending some time on the re-occuring theme of variable dependency (leading to a lack of expected diversification) and illustrating it with a simple example concerning summations of VaR levels from different business lines at a bank.
Good speaker – main points/concerns were the validity of pricing models and dependency assumptions between variables.