Thanks to Michael Lewis everyone’s gotten interested in A.K. Barnett-Hart’s honors thesis PDF focused on data analysis of the CDO market. And rightly so! What she did, basically, was run a lot of regressions to see which characteristics of CDOs were associated with defaults. The results that stood out to me: So-called Alt-A mortgages, neither prime nor subprime, were the biggest problem children, especially the Alt-A ARMs. More complicated products did worse than smaller ones. Firms that underwrote tons of CDOs did worse than firms that were more parsimonious. The credit agencies were disastrous.
Another thing the thesis reminds us of is that just because you walk out of the casino with a lot of money doesn’t necessarily mean you had the best strategy. JP Morgan, for example, famously got mostly out of the CDO market before the worst ones were issued in 2006, which has made them a lot of money and made them look smart. But it turns out that when you control for other variables, JP Morgan was actually an unusually bad underwriter. Plausibly, it was precisely the JP Morgan’s inability to do this well that persuaded them to get out of the market. Better underwriters stayed in longer, until a point where the whole product category was lousy.