More than a decade ago i wrote a paper that compared estimates of VaR from different systems, different methodologies, and data sets. Not surprisingly perhaps there were huge differences in outputs. Why should this be? After these models are trying to estimate the same thing for a given portfolio. The variance in VaR esimates is actually a form of Model Risk, caused from alternative systems, models, data sets, even different users. Unlike most other decision support systems (of which VaR is an example), the variance in VaR estimates can be "added" to the actual VaR estimate to produce a VaR that incorporates not just market volatility but also model risk. Take a look at the paper - can be downloaded from:
http://papers.ssrn.com/sol3/papers.cfm?abstract_id=1212&rec=1&srcabs=148750
The main points still apply today i believe and slowly we are starting to understand the role of people and systems in interpreting risk measures. Measures are rife with uncertainty, ambiguity and even equivocality that needs to be appreciated even if it cannot be completely understood.
Subscribe to:
Post Comments (Atom)
No comments:
Post a Comment