« The Business Damage caused by Risk Management |
Main
| Was I wrong about Gaussian distributions? »
December 06, 2007
A better way to prepare for catastrophic events?
Further to the dialog on modeling risks, it has become increasingly clear that so called “abnormal” or “extreme” events have a tendency to occur far more frequently than many of the “normal” models predict. And at the end of the day, things like bursting bubbles and subprime meltdowns do not care whether they are labeled “abnormal”. They simply happen, and with an alarming frequency. That this frequency is discounted does not offer any protection against the sequelae of such events.
What is also evident from innumerable studies is that the behavior of mos complex systems that we try to model is also chaotic; that is to say that it does not follow a pattern and is completely unpredictable. So what is a risk manager to do?
It occurs to me that the predictions of future events might be somewhat misguided. For discussion, I propose a new way of addressing risk. Instead of applying a stochastic model based on past events and assuming a certain shape to the distribution, risk manager should first engage multiple experts from different areas to outline the possible risks. Second, risk managers would do scenario analysis. Finally, the quant types would be called in, not to model probabilities, but instead to find mathematical methods for measuring and monitoring early warning signs.
This seems to me a more sane approach to dealing with a chaotic system. Let's say you are living in a hut in a part of rural India infested with man eating tigers. Instead of modeling the probability distribution of a possible tiger attack figuring out the likelihood of a day vs. night attack and so on, you would set up a warning system around your house. You would also keep a loaded gun next to the bed. You might not know when the tiger will attack (for this is unknowable) but if one does, no harm will come to you because you will have the time and means to avoid a loss.
This avoids the trap of discounting “extreme” or “abnormal” events. What do you think?
Posted by beaumontv at December 6, 2007 11:49 AM
I couldn't agree with you more. An expected value is of little consolation to an investor who has who has lost his entire investment. What is important is the potential downside risk under any circumstance. Besides, when so called extreme events become the norm rather than the exception, risk mitigation based on historical probability distributions are meaningless.
Posted by: Kalyan Sunderam at December 7, 2007 03:53 AM
I think the key difference between the two approaches you described is a fundamental one. If we see the role of a Risk Manager is the prevention and elimination of risk, then the implementation of cotrols and limits, as described in the second case, is the most relevant approach, especially in the area of Operational Risk Management/Catastrophic Risk. However, when Financial Risk Managers employ, say, VaR or Extreme Value, along with all the probabilistic modeling of risk, including variance/covariance estimations, they are thinking about the quantification of loss as a result of a move in the underlying factor, so that sufficient risk capital can be set aside to keep the firm above a certain risk/loss threshold. This approach typically is difficult to make sense and implement in the Operational Risk domain. (Going back to your example, it's like determining if one will lose an arm or a leg, or one's life due to the tiger attack) In my opinion, due to this fundamental difference between financial risk and operational risk, the strategies adopted to deal with the risks must be different too (hedging vs mitigation, risk optimization vs risk prevention).
Posted by: Eric B. Chang at December 11, 2007 11:03 AM
Whilst I agree that current metrics only give a partial picture they are not truly redundant. What is missing is another piece of the puzzle that should always be reviewed constantly by risk manager/underwriters: the context in which the risk lies.
Seldom, if ever, do underwriters or 'quants' who price the deal review the context and the change their deal would have on it. If they had they would have spotted aggregation, assimilation and cross fertilisation of exposures that trigger probabilities for new loss frequencies.
No amount of advisors need be added if the underwriter balances risk and context change. If you're in a quake zone and there's a lot more building or whether you're creating a new credit product, the attraction to the new area post decision shifts the context in which the original decision was made with consequential shifts in loss probabilities.
You don't drive a car by looking ahead once and hoping you get there unscathed, it's an iterative process until conclusion.
Posted by: Stefan Wasilewski at December 14, 2007 06:04 AM
What I like about Tigers is that once I'v learned their behavior, it always will be the same and thus predictable.
When you suggest "to find mathematical methods for measuring and monitoring early warning signs" as far as I can think of, you are talking about statistical methods to measure early warnings based on information from the past of such extreme events. Mathematical methods are only a tool to implement those statistical results, and so we ended up with the Type II error, which will forecast disasters so often to make investors see Tigers everywhere and avert investments completely.
Posted by: Michal Huller at December 14, 2007 06:36 AM
Post a comment
