Enthusiastic about our new release and the progress of the All-New UnRisk project, I have been a bit quiet about reads I find insightful and exciting.
Like another article of Aaron Brown. Risk Intelligentce released in the Wilmott Magazine, March-12.
It is about ignoring real risks and worrying about imaginary ones ...
First, it refers to a 20-year study of the Psychologist Philip Tetlock in which 284 experts in various fields were asked to make 28,000 predictions. The results, in short, is that the experts' answers were near to chance, and worse than basic computer algorithms. If in their expert field or another, although, not surprisingly, their confidence was much higher in their expertise.
Second, Aaron refers to the book Risk Intelligence, by the Philosopher Dylan Evans. It is about How to Live with Uncertainty. To make better decisions, you have to know how uncertain you are. And it makes a difference, if this can be measured. And it can in much more cases than we think.
Provocatively speaking, if you are highly confident, you usually don't. This is the link to Tetlock's results. And people who exhibit high risk intelligence in one field, like professional bridge players, are not better than average in other domains.
How uncertain are we? Can we overcome the biases? Are we all-or-nothing-thinkers? ...
If I am able to make quantitative decisions, with probability estimates, I can test them and consequently benefit from an explorative learning process. Feedback is indispensable. This is one obvious conclusion of the article/the book, ..
If a portfolio has a one-day 5% VaR of EUR 1 Mio, there is a 0.05 probability that it will fall in value by more than EUR 1 Mio over one day if there is no trading, or informally, a loss of EUR 1 Mio or more is expected in 1 day out of 20. The quality of this "expression" can be backtested, more then that we can test the distribution of the break days and what have you.
But UnRisk doses calculate many values that enable the creation of such predictions with probability estimates that can be tested ... instrument and risk factor contributions, VaR deltas, ...
In this respect it is not only a data and analytics management system for risk managers, it is also an explorative, constructive learning platform for them.
