Megan McArdle has a thought-provoking post up:
The basic point is that, once we’ve dealt with a given risk, we tend to forget that it ever existed — to our misfortune if and when it comes back. McArdle’s first example, quoting a Glenn Reynolds article, is polio in particular and widespread disease in general: now that we’ve widely distributed polio, MMR, and other similar vaccines, we forget why they were necessary in the first place and a little pseudoscience about thimerosal and autism can convince otherwise perfectly good parents to crank their kids’ health care back about a hundred years*.
As you might expect, her next example comes from the credit crisis:
That argument could be made, and perhaps should be made, just as well about financial and regulatory infrastructure. Though I’m not sure that there is any way to prevent 70-year events like the current mess, there are nonetheless decisions that seem lunatic, in retrospect. Why were Goldman, et al, allowed to lever up 30-to-1? Why, for that matter, did they want to? Well, because if you’ve gone for a long time without any problems, all you can see about the safeguards is that they’re costing you money.
I don’t think for a minute that McArdle’s suggesting this was the only motivating factor behind the sort of leverage games that were being played, but it’s interesting so let’s run with it for a little while.
One thing I’ve read in a number of places** is that the quantitative models that green-lit idiotic decisions like the ones mentioned here were originally based on long-term data and worked rather well. I guess if you have seventy years of data on a fairly well-understood chunk of the market, you’ve seen most of what’s likely to happen and can plan accordingly. The problem, these articles went on to explain, is that people slowly forgot that their quantitative models depended upon having lots of well-understood data, and tried to use the same methods with, say, three years of data on heavily-leveraged subprime mortgages in a housing bubble — markets and investment vehicles that hadn’t existed long enough to plug into a model and expect good heuristics to come out. In this case, it’s the risk that your data might not be representative that’s been forgotten. Oops.
The contrapositive to this problem is hyperfocusing on past-tense risks. For example: airport security. I’m not going to seriously suggest that we take down the metal detectors at airports and start waving people onto the planes off the streets***, but I do find it difficult to believe that the vast amount of time, money, and effort spent on incremental improvements to air-travel security couldn’t have been better spent elsewhere. Other targets have been attacked — trains in Madrid; subways in London; hotels in Bombay — but the September 11th attacks left such a deep impression that we have a hard time generalizing past airplanes getting hijacked.
——
* Anyone with a comment about how the CDC and I are dupes of the eee-ville pharmeceutical industry had better have a double-blind peer-reviewed article or three from reputable journals to cite — whether or not you ran for President last year.
** Which I can’t track down any more, so take this with a pillar of salt. Yeah, I’m aware of the irony of demanding sterling references, then supplying an unreferenced assertion of my own.
*** Though I maintain that, post-Flight 93, the best way to prevent hijackings would be to give every passenger a Fairbairn-Sykes at the gate.
0 Responses to “Forgetting about past-tense risks”