Black Swan Theory in Insurance

Black swan theory was bound to show up in the insurance industry, but so far there has been limited discussion.  Do you know what a black swan is?  Black swans are important in assessing future insurance losses, but are they adequately represented in the models? 

The term originates from a book called The Black Swan: The Impact of the Highly Improbable by Nassim Nicholas Taleb (also see here, or see Recent Books on the left of this blog).  Wikipedia provides a good definition (see here):

a large-impact, hard-to-predict, and rare event beyond the realm of normal expectations

The book was written from the perspective of investing (Taleb was a hedge fund manager), but the concepts are particularly important for the insurance industry.  A central thesis of his book is that models are defective and lead to incorrect conclusions because black swan events are not contained in the data and are therefore assigned a probability of zero outcome – obviously incorrect. 

While there has been quite a bit of discussion of insurance modeling, there has been little discussion of black swan events and the ability of predictive models to adequately assess severe catastrophes for insurance purposes (see here and here for two such articles). 

However, two articles came to our attention recently.  The first (see here) is a response to some comments made by Taleb in an interview (see here).  Taleb notes:

My idea in The Black Swan is to make people think of the unknown and of the potency of the unknown, particularly a certain class of events that you can’t imagine but can cost you a lot: rare but high-impact events.

The author makes an excellent pitch for combining predictive modeling techniques with outcomes not envisioned in historical data, and makes 4 suggestions for planners (see here).

In the second (Risk & Insurance commentary, see here) the author is concerned that the black swan theory may result in less risk management and assumes that because black swan events occur we should or will not consider past history. 

I fear they may increase paralysis in risk management and further promote inaction.

But black swan theory is telling us that modeling, which relies on historical data, is only part of the story.  In fact, the future will be a combination of prior history, reflected in historical data, and new events, which can and at some point will lead to more volatility – sometimes better, sometimes worse.

The important point is that black swan events are not contained in historical data, and therefore traditional predictive models cannot provide complete picture of the future.  The second piece is equally important:  these events can be so large that they will dwarf other outcomes if they occur, although the probability of occurrence is extremely small.

Unfortunately, we have seen a number of these black swan events in the insurance business in the last few years, 9/11 and Katrina being two such examples.  Maybe a more flexible approach to modeling and a focus on risk management could be beneficial to the insurance industry.

Brought to you by Tennant Risk Services.

 

Comments

[mickey bruton]

It's obviously a bit late to pick up on Mr Jaggs' hilarious comment. I did chance upon it nearer the time (I think I was trying to google a rather nice yacht called Black Swan I had encountered on the racing circuit.)However,at the time I found myself too weak with laughter to respond. For what it's worth, having happened on this blog again, feel obliged to say (in case anyone could possibly have any doubt)yes, he is a complete c**t(clot, obviously), and in case he is still following the trail, Tim hope your support of the kiddy-fiddlers/Nazis is bringing you the satisfaction you have been unable to find/give elsewhere. Lol- Mickey.

[RamonGustav]

Congratulations! You have so much useful information, write more.

[Texas Adjuster License]

Trust insurance agencies to believe in that! :-)

[Steve]

Follow you on RSS.
Nice Article

[Tim Jaggs]

Isn't the primary difficulty with Black Swan events their fractal nature and that modelers can't therefore include them in a gaussian model? This means we end up being surprised by both the cause and effect of a Black Swan. If insurance modeling included more scenario analysis to accompany the traditional gaussian approach, we might still be surprised by the effect but some of the causes could surprise us a bit less?