Dear Friends,
I am currently attending a joint conference on risk and safety analysis PSAM/ESREL in Helsinki, Finland (my presentation was at an uncertainty session organized by Matthias Troffaes). I have learned a lot about risk analysis, the corresponding problems are very interesting and often misundertood.
Misunderstanding comes from the fact that the word `risk' has two different meanings: in financial applications, any possibility to lose momey -- whether we know its probability or not -- is called risk. In this sense, taking risk into account is a part of normal decision making.
In engineering applications of risk analysis, however, risk usually refers to rare events, events for which normal decision makers are not good in taking them into account. For example, when designing a nuclear power station, an offshore oil platform, or a space station, we need to take into account vey rare events, events that may lead to catastrophic consequences, such as an unusual strong earthquake at the location of the nuclear power station. To properly take such rare events into account, we need to estimate the probability of such an event. For usual events -- e.g., small earthquakes -- we can estimate their probability by looking into the past records and estimating the frequency of similar events in the past. For rare events, we may have one record in the distant past or no records at all, so we cannot estimate their probabilities as simply frequencies. Instead, we can, e.g., estimate the frequencies of earthquakes of different strength s and extrapolate -- e.g., if these probabilities follow the power law A*s^(-a) for some constants A and a -- to the desired strength S. As a result of an extrapolation, we do not get precisely known probabilities, we need imprecise probabilities -- and many imprecise probability techniques such as p-boxes (bounds on cumulative distribution functions) are indeed actively used in risk analysis. In the extrapolation, we can also use expert knowledge -- and several talks described how fuzzy techniques can be used to capture such knowledge.
Risk analysis is an extremely important but still under-developed part of engineering design. I would like to join the conference organizers and participants in encouraging researchers to contribute ideas and results.