Flooding and risk assessment

This story about the Association of British Insurers complaining about inadequate spending on flood defences raises three points showing just how difficult good risk management is:
1. What exactly do we mean by "probability"? The BBC says the summer floods were a 1 in 150 year to 1 in 200 year event.  But how is this estimated? One way is just to look at history: how often such floods have happened in the past. This is the frequentist view - just looking at how frequent such events have been.
But this approach can be misleading if the process generating events changes over time - say, because of climate change. Then, we need a subjectivist view, whereby we  use our knowledge of how the event-generating process changes to generate non-frequentist probability statements, of the sort "I believe there's a one in x chance of a flood."
However, such statements themselves are subject to uncertainty. The government's Future Flooding report estimated that the additional annual losses caused by extra flooding arising from climate change could be between £1bn and £27bn by the 2080s. Risk measurement itself is subject to risk.
Everyday use of the word "probability" glosses over the distinction between frequentist and subjectivist probabilities. We say "there's a 50% probability of this coin flipping heads" and "there's a 50% probability of Hillary Clinton being next US president", unaware that the word "probability" is used very differently in the two sentences.
2. Risk assessments can be biased by incentives. Obviously, the ABI has an incentive to overstate the risk of flooding - because it wants to justify high premiums and push the government into doing more to stop it having to pay out: you don't think insurance premiums would fall proportionately to the lower risk of flooding, do you? But the government too has an incentive to exaggerate the risk, at least in its spending plans.
If it spends to little on flood defences and a flood occurs, it will be fiercely criticized for doing too little. But if it spends too much and a flood doesn't occur, criticism will be more muted - fewer will complain about the hospitals that weren't built or the excessive taxes they paid.
The upshot is a bias to overspending, and to exaggerate risk.
3. Our assessment of risk is coloured by recent events. What are the chances of serious flooding? What are the chances of a major bank collapsing? Most people, I guess, give higher answers to these questions today than they would have a few weeks ago, because the dangers became more obvious over the summer.
It's easy to say this is irrational - just an example of the availability heuristic.
Not necessarily. It might be that the danger of floods or banking collapses really are greater now, and we were wrongly under-rating them in the spring, because the underlying event-generating process has changed. It can sometimes be hard to distinguish between irrational over-reaction and sensible Bayesian updating.
The only message I have here is that risk management and probability assessment is tricker than supposed. As someone once said, everybody thinks they understand probabilities, except those who have spent a lfetime studying them.

Comments

[Bruce]

But would policy decisions about flood defences be made on the back of probability assessments, even if they were possible? I suspect that electoral calculations would still count more.

[Mark Wadsworth]

1. Nobody knows how soon or how bad next floods will be.

2. ABI have vested interest in better flood defences.

3. Homeowners have vested interest in better flood defences and also in lower insurance premiums.

4. Dam builders and the like have vested interest in building dams.

Solution - provided it's homeowners paying for flood defences via land value tax, and their additional land value tax is less than the corresponding reduction in insurance premiums, everybody wins.

Even if it never floods again, it's nice to be able to sleep at night. That must be worth something, even if, with the benefits of 20/20 hindsight in thousands of years time, it turns out it was a waste of money.

[mat]

Here's a solution: don't buy a house on a floodplain, or if you do, recognise you're taking a risk, and get private insurance.

[chrisP]

Surely the pricing mechanism will take care of the problem. After all comparable houses in or out of flood plain areas ought to show the challenge that potential floods bring. The pricing mechanism will then, as it were, provide an inbuilt insurance premium. I.e. the price reduction for a house in a flood plain will be X,000 less & this will 'pay' for the damage caused. This assumes (wild optimism) that the price differential is then invested as a sort of eschew fund for potential future losses.

[dreamingspire]

"3. Our assessment of risk is coloured by recent events."
The "our" here must mean people who are not expert in the area where the risk is to be assessed. Evidence, however, appears to suggest that there is a lag between expert assessment of increased risk in a situation where "the process generating events changes over time" and the application of new or enhanced defence measures. While that may give the appearance of incompetence (and sometimes that is proved later, maybe is on the way now with Lady Whatsit's claim yesterday that this year we had the wrong kind of floods), it is the nature of governments that decisions to mitigate risk often take a long time. The moral is in mat's comment.

[dsquared]

Dreamingspire makes a very good point there Chris - the flooding experts had been warning about these risks for quite a while, and the decentralised wisdom of crowds ignored them. To me, this at least on the face of it looks like an own goal for markets vs. central planners.