Good accounting error face. Photographer: Chris Ratcliffe/Bloomberg
Good accounting error face. Photographer: Chris Ratcliffe/Bloomberg

Last month, Citigroup failed the Fed's stress test because, even though it would meet minimum capital requirements in a severely adverse scenario, the Fed wasn't comfortable with Citi's capital planning process. Meanwhile, Bank of America passed the stress test and was authorized to increase its capital return by about $6.5 billion, because the Fed concluded that (1) Bank of America would meet minimum capital requirements in a severely adverse scenario, even after returning the extra $6.5 billion, and (2) its capital planning process was, you know, adequate.

Wrong and wrong! Bank of America delightfully announced today that it's suspending its capital plan for mathematical ineptitude, and the Fed, um, concurred. BofA can resubmit its plan within 30 days, and will do so; it "expects the requested capital actions to be contained in the revised CCAR submission will be less than the company’s previously announced 2014 capital actions."

The press release is a little unclear so let's just do some arithmetic to spell it out. First up: How much capital went missing? The 8-K gives you a variety of choices:

So that's about $2.7 billion of capital that's missing under current transition regulatory capital rules, or $4 billion under fully phased-in Basel 3. By my math, Bank of America had a little bit of wiggle room under the stress tests, but not that much room. Here's how I see its stressed capital ratios after adjusting for this:1

So under the revised math it would fail the stress tests on quantitative grounds -- though just barely. Unclear how it would do on qualitative grounds: On the one hand, oops! On the other hand, it's not like the Fed caught this either. Poor marks all around.

What happened? Amazingly terrible stuff. Here's the Wall Street Journal's version:

When Bank of America acquired Merrill, it assumed some structured notes issued by that firm. When those matured, the bank sometimes realized a loss. At the same time, the bank would have been showing unrealized gains and losses from these notes in its earnings. Banks are required to back out such unrealized changes in the market value of their own credit when determining their regulatory capital.

The mistake occurred when bank officials inadvertently adjusted their capital by excluding the realized loss as well.

Ha! What happens is that, if you issue structured notes, you have income (loss) when your own credit spread widens (tightens). This is fiercely counterintuitive and messes most people up, but not bank accountants. Bank accountants understand that you have a loss on your issued debt if your own credit gets better, and a gain if it gets worse, even though to regular people those gains and losses look pretty fake.

I should say, they look pretty fake to capital regulators too, so capital regulation requires you to back out those gains and losses. Bank of America had lots of phantom losses on the Merrill Lynch structured notes, because Merrill's credit has improved rather dramatically since 2008. So it dutifully took those losses for income accounting purposes and backed them out for regulatory capital purposes.

Now, part of the reason that regulatory capital accounting ignores these losses is that, for a solvent going-concern bank, they tend over time to be zero. Like:

  1. You issue a bond at 100
  2. Your own credit gets worse, so the bond is worth 95, and you have 5 of income.
  3. Your own credit gets better, so the bond is worth 105, and you have 10 of losses.
  4. Eventually the bond matures, and you pay it off at 100, and you have 5 of income.

So, 100 in, 100 out, and zero net gain or loss.2 That's generically how it happens: You issue a note at par, you pay it off at par, and all the swings in between end up being amusing accounting noise but not realized cash expenses.3

The problem is that here some of the losses were realized. Merrill issued some bonds at 100, its credit got worse, its credit got better, and it ended up paying them off at 100, for no ultimate effect. But Bank of America stepped into Merrill's shoes when those bonds were worth, like, 90.4 So when Merrill paid those bonds off at 100, Bank of America had a loss of 10, and that is a "realized" loss, where I put "realized" in quotes because I don't know what it means.

I mean, it's a crystallized loss -- there's no chance that matured bonds can go back to being worth 90 -- but is it a "real" loss? Like, of money? If you try to track the cash flows here you'll go crazy. Merrill Lynch issued debt at 100, paid it off at 100, so didn't lose any money.5 Bank of America bought Merrill when that debt was worth 90, and then paid it off at 100. Did Bank of America lose 10? I guess the question is effectively whether Bank of America in 2008 valued Merrill Lynch assuming that it owed what it owed on its structured notes (a legal-obligation approach), or whether it valued Merrill Lynch as though it owed 90 percent of what it owed on its structured notes (a generally accepted accounting principles approach). That is, did Bank of America pay an extra $2.7 billion for Merrill because its structured notes were trading at a discount?

Probably not,6 but that is not the issue here at all. You can't, like, try to figure out what reflects economic reality and then use that to drive your regulatory capital accounting. Does Bank of America actually have $2.7 billion less equity cushion than it had thought? That is a meaningless question. It has $2.7 billion less of regulatory capital under transitional Basel 3 rules, or $4 billion less under fully phased-in rules, or like $720 million less of common equity tier 1 capital. Even capital accounting can't give you a consistent account of what happened here, and the various accounts that it gives are inconsistent with GAAP. Asking for consistency with economic reality would be madness.

You have to sympathize a little with Bank of America here. They totally messed this up and should be made fun of: The people doing your regulatory capital calculations really ought to know how to do regulatory capital calculations! "The rules are hard" is not a good excuse for a $2 trillion bank. But it's a genuinely weird situation: Every bank has unrealized own-credit gains and losses, but big realized losses are less common. And it seems to me that their (wrong!) approach might better reflect economic reality than the correct regulatory capital (and GAAP) approach.

And, y'know, the Fed missed it too -- which I suspect means less that the Fed also independently got this accounting question wrong and more that the Fed did not review Bank of America's capital calculations in any detail. Which is -- really weird? Like, the banks and the Fed run independent parallel calculations of the stress tests, getting very different numbers for the banks' stressed capital ratios. But they each start with the banks' calculations of their current capital ratios, apparently without too much in the way of checking by the Fed. If the banks get those wrong, the Fed's inputs are wrong too. Don't you feel great about the rigor of the stress tests now?

The lesson here is the usual one. Nobody knows what a bank is, or how big it is, or how much capital it has. Various people will write down various numbers, but those numbers won't tell you what you really want to know: how much the bank is worth, or how safe it will be in a crisis. If you're lucky, they'll answer dry semi-related technical questions like "how much tier 1 common equity does Bank of America have under fully phased-in Basel 3 rules?" If you're unlucky, they won't even do that.

1 Math takes the stressed ratios from the CCAR results, then reduces them by the relevant amounts in this revision. Because the stress test results are rounded to one decimal, and because of timing/etc. differences, I can't vouch for this, but the gist -- that BofA is quantitatively fine under every ratio except leverage ratio, but just fails that test -- seems right.

2 I'm ignoring everything but the simple own-credit math. In particular, structured notes have embedded derivatives linked to the S&P or whatever, which I assume were accounted for correctly (regulatory capital backs out only unrealized gains/losses on own credit, not on other derivatives). And I'm ignoring current-pay interest, who cares.

3 You can realize a gain or loss if you buy back the notes in between: If they're worth 105, and you pay 105, then you have 5 of actual cash realized losses. This happens to regular companies all the time, but it is perhaps less common in the bank structured-note business, which is after all designed to take advantage of credit mispricing.

4 Real round numbers: At the end of 2008, Merrill had about $58.5 billion of structured notes outstanding (see page 113 of its 10-K). At the end of 2013, it had about $30.5 billion outstanding (see page 219 of the Bank of America 10-K; I'm attributing all "Bank of America Corporation" structured notes to Merrill, which may not be right). So about $28 billion of notes have matured or been repurchased, with missing capital of around $2.7 billion, or about 10 cents on the dollar of those notes. (The missing capital is cumulative: It's all maturities since buying Merrill.)

5 I am aggressively ignoring the possibility of above-par repurchases and assuming that it's all maturities. Obviously -- from the press release -- there were some early redemptions, so this is approximate.

6 I would assume that a rational person, in deciding whether to buy a bank, would:

  1. add up that bank's assets, at fair value;
  2. subtract what that bank owes, at the face amount of what it owes; and
  3. pay the difference;

but that seems too simple, doesn't it?

To contact the writer of this article: Matt Levine at mlevine51@bloomberg.net.

To contact the editor responsible for this article: Zara Kessler at zkessler@bloomberg.net.