SciFinance®

Good Risk Management Techniques Do Already Exist

In the popular press I keep reading about how quant models need to be improved. It's not as simple as that, but nor is it so subtle for the journalists to have to dumb down.

There was an article in the Financial Times last week that said we need more maths, yes, more, lots of it, until we are stuffed to the very brim. No mention of what sort of maths, no mention of the relevance of the maths or the robustness of the resulting models. Just "we need more maths." I have to disagree that it is just a case of more is better. The wrong models are partly responsible for the current mess, and many of them are too mathematical. Well, I've said that many times over the years.

There's also the perception that there are no good models out there. Again, I have to disagree strongly. There are plenty of models that are better than the established models. Some of them are better by virtue of being more accurate in a scientific sense, some are better by virtue of their simplicity. Given that financial models are never going to be perfect it surely makes sense to have models that are transparent and easy to understand (and mend when they go wrong) rather than complex and impenetrable.

One error that keeps being repeated is that there are no models for crashes, when all instruments move together in sync. Well, the much-maligned-and-rightly-so copulas are meant to do that. Clearly they didn't do a great job. There is also the model that Philip Hua and I designed in the mid to late '90s, CrashMetrics.

CrashMetrics ticks all the important boxes: a) accuracy, b) correct level of mathematics, c) understandability, d) robustness, e) flexibility, f) ease of use. We came up with the idea because it was clear that Value at Risk would only work during rather dull markets, that is, perversely at times when it almost wasn't important to measure risk! So something had to be done about the bank-destroying events, the crashes for example. Extreme Value Theory was supposed to do that, but that's just "rearranging the deck chairs on the Titanic." CrashMetrics is a worst-case scenario model, designed to tell you how bad the crash would be for your portfolio, with no reliance on probabilities. Once you know what the worst could be you then decide whether you want to protect yourself against it, and then, thanks to "Platinum Hedging," the methodology also tells you how to do this in the optimal, most cost-effective manner.

Sadly CrashMetrics fails to tick the most important box of all, the so-complicated-that-only-a-handful-of-people-with-17-PhDs-each-can-understand-it box. Silly us!

P