...is this, which Tim Geithner and Larry Summers propose in their Washington Post op-ed today:

The administration's plan will impose robust reporting requirements on the issuers of asset-backed securities; reduce investors' and regulators' reliance on credit-rating agencies; and, perhaps most significant, require the originator, sponsor or broker of a securitization to retain a financial interest in its performance.

The problem is that banks expanded credit to overly-risky borrowers because they knew they could turn around and sell those mortgages to investors via securitization. The thinking, correct in my mind, is that banks will be far less likely to place such bad bets if they have to face the consequences (i.e., defaults), or at least some of the consequences.

Simon Johnson is skeptical, pointing out that:

It was a big unpleasant shock when everyone realized that Lehman, Bear Stearns, and others had retained a large exposure to dubious financial products, some of which they had issued.  We are back to the Greenspan fallacy here – if financial firms have an incentive not to screw up on a massive scale, they won’t.

But, as Felix Salmon notes in response, a lot of what banks kept on their books was the ostensibly risk-less layers of investment vehicles made out of mortgage-backed securities (that is, the portion of the investment vehicle that would supposedly stay afloat even if everything else sunk). Of course, that didn't work out so well--even the risk-less layers turned out to be pretty risky. But there's a big difference between having something you think is risk-less (but isn't) and something you consider so risky you can't wait to pass it off to investors.

The most compelling evidence I've seen that banks would have behaved differently if they'd had to keep some of their riskiest subprime loans comes from this chart in a paper called "Did Securitization Lead to Lax Screening?" (link doesn't seem to be working, but I found it via this post by Matt Yglesias):

That dividing line is a FICO score of 620, which is the minimum score a bank needed on a loan (in most cases) in order to sell it to investment banks and then investors. What you see is that, for essentially identical risks--a FICO score of, say, 619 versus 621--banks seem to have done much less due diligence on the loans they could securitize relative to the loans they had to sit on. That's why you see the former defaulting at a significantly higher rate than the later. Which, in a nutshell, is the problem the Geithner-Summers proposal is designed to fix.

--Noam Scheiber