Specific risk guidance is troublesome

I’ve been (re)reading a bit about “risk appetite” again recently. Some of the reading includes messages saying it’s a confusing term, it’s a bad idea, or that it’s a good idea that’s really important to manage properly [pdf].

One of the things that occurred to me during this is that it’s actually very difficult to control risk-taking through an instructional policy.

For example, suppose we had a policy on appropriate investments we might make in start-up companies. One startup might require an investment of £1m, with an 80% chance of losing it all, but a 5% chance of making back fifty times our money. That would be outside the limits of an investment policy which says we’re unwilling to lose all of a single investment with 80% likelihood. However, if we could afford to make 25 such investments (for a total of £25m) the odds of an aggregate success are likely to be very good. So a policy that prevents a bad outcome on an individual basis fails to net a desirable result when we scale it up.

On the other hand a single bet of £25m with same probability profile as the original once again looks poor, if not worse.

We see that the investment policy works in different ways depending on which level it’s applied at, and how widely it’s applied or scaled.

In general probabilities combine in ways that are usually counter-intuitive. It becomes even more difficult when we realise that our startup investments are likely not to be independent. Perhaps the way we choose them has a common theme (maybe a common selection criteria, with subtle biases) and that means our 25-strong portfolio will not perform in the way outlined above.

And this example with a financial investment is simple compared to areas like health and safety, business deals, and projects, where the numbers are not so obvious.

What all this means is that if we’re to tell people across a large company what “acceptable risk taking” is then it’s almost impossible—at least on this superficial reading—to do that in a way that is both specific and achieves the result we intend.

Perhaps a more useful approach would be something that is not very prescriptive, and is combined with a feedback and reaction mechanism that allows people to assess and steer with sufficient speed.

Photo by pattyloof