Just getting around to reading Wired’s recap on the Gaussian Copula, or the formula that destroyed the world banking system. Or, really, the formula that allowed the world banking system to destroy itself.
The GC, as I like to call it, is not news; the Wall Street Journal ran a prescient story with the headline “How a Formula Ignited Market That Burned Some Big Investors” back in 2005. I’d picked that up from perusing the ever-informative blog, Information Processing, in… Still, the Wired story is a good summary. Basic story: David Li, a former Chinese peasant farmer who became a mathematician at JPMorgan Chase, published a paper, “On Default Correlation: A Copula Function Approach,” in The Journal of Fixed Income. Who doesn’t read the JFI these days, right? Li offered a formula that allowed the market to price default risk by correlation rather than historical data. The false sense of security engendered by having a formula — with numbers even! — is what created the $60 trillion fabulism known as credit default swaps. Not to mention the $4 trillion in regular CDOs. What everyone who was getting rich of these things when they going was good was that it would not always be so; they chose to forget as that the key numbers in Li’s formula are not immutable — gammas, not pi; calculations, as opposed to constants. Woops!
By the way, the WSJ pointed out all the problems in 2005, long before the subprime crackup. The original is behind a paywall but the key passages are excerpted on the Information Processing post, with highlights of the original article’s clear articulations of the model’s flaws and the invisible markets it created.
Most interesting point, mostly missed by the Wired story: the whole idea is borrowed from the actuarial science of broken hearts. People die sooner when their spouse dies. They’re seemingly unrelated events that are related, and have a correlation, and statisticians have used copulas to correlate them for some time. This was the idea that Li borrowed that idea for formula. Broken hearts indeed!
