The problem is that "too big to fail" isn't about the size of a bank's balance sheet; it's about how tightly coupled that balance sheet is with other institutions. The FDIC can resolve even a huge conventional commercial bank, because as long as the loans are sold and the depositors paid off, that failure doesn't suddenly and massively impair other peoples' balance sheets.
(It may, down the road, if for example a huge portfolio of real estate loans is written down, which casts doubt on the value of the collateral securing the loan books of other banks. But that's different from triggering a bank run.)
It is pretty clear to me that the Fed and Treasury decided to let Lehman (or whatever bank tottered next, rather) fail, pour encourager les autres--and that despite months of preparation, they didn't foresee the meltdown in the money markets that this failure touched off. Hence all the subsequent bailouts: no one could be quite sure what the fallout from further failures might be.
But the degree to which a financial institution is tightly coupled with other parts of the financial markets is a lot harder to measure than its leverage ratio, its balance sheet, or any of the other metrics that we'd like to use to wrap these institutions up into a nice, tidy, too-small-to-fail package. As Economics of Contempt points out, the weakest part of the administration's plan is its reliance on crude metrics like capital levels:
I think the administration focuses too much on capital levels as the relevant measure of a Tier 1 FHC's health. The biggest problem with the PCA regime applicable to commercial banks is that too often commercial banks can go from "well capitalized" to insolvent without ever triggering the PCA requirements. This problem is even worse for Tier 1 FHCs. Lehman had a Tier 1 capital ratio of 11% as of August 31, 2008 -- just two weeks before it filed for bankruptcy. Had Lehman been a commercial bank, it wouldn't have triggered the PCA requirements until it was far too late. The administration's proposal requires that the PCA triggers (which it calls "capital standards") include a risk-based capital requirement and a leverage ratio.
I would make the PCA triggers less focused on capital levels, and more focused on the conditions that make Tier 1 FHCs susceptible to modern-day bank runs. For example, I would make one of the PCA triggers contingent on the tenor of the Tier 1 FHC's overall liabilities. As of August 31, 2008, over half of Lehman's $211 billion tri-party repo book had a tenor of less than one week, which made it remarkably susceptible to a run in the repo markets -- which, of course, is exactly what happened. Lehman was also relying on roughly $12 billion (at least) of collateral from its prime brokerage clients to fund its day-to-day operating business. These conditions had persisted for several quarters before Lehman's bankruptcy.
The Fed should be required to take prompt corrective action once a Tier 1 FHC allows the tenor of, say, 20% of its overall liabilities, or 50% of its daily funding requirements, to drop below one week. (I just pulled those numbers out of the air; the Fed is in a much better position than I am to set then appropriate tenors and percentage of liabilities.) These are the kinds of PCA triggers that would be the most effective. A PCA regime focused on capital levels is unlikely to make much of an impact.
I'd add that overreliance on any metric is likely to cause problems, as well as solve them--I've heard fairly convincing arguments that the Value at Risk regulatory monoculture helped set the system up for catastrophic collapse. One is wary of giving regulators too much discretion, of course, but at some level, at least at the margins, making good decisions about which institutions are in trouble is always going to require a degree of art as well as science.