It's money. On a table. But it's also cookies. Photographer: Andrew Harrer/Bloomberg
It's money. On a table. But it's also cookies. Photographer: Andrew Harrer/Bloomberg

I wrote earlier today about the Fed's stress tests, and how they're a weird exercise in doing arithmetic in a fog: Your job, as a bank, is to:

  1. guess how much capital the Fed thinks you'll have in a crisis,
  2. subtract out how much capital the Fed thinks you'll need to have in a crisis,
  3. and ask to return that much money to shareholders.

Item 1 is the fog: The banks and the Fed calculate their stressed capital differently, and they get very different numbers. Item 2 is straightforward: It's hard to calculate capital, but it's easy to say how much you need; it's nice simple round numbers like 4 or 5 or 8 percent. Item 3 is debatable: You might imagine a bank saying, you know what, we want to have more than the minimum amount of capital in a severely stressed scenario. I don't spend a lot of time imagining that, but you and I probably have very different inner lives and far be it from me to interfere with yours.

Anyway how foggy is the arithmetic? I drew you a graph:


This graph -- which is itself a bit foggy, so read the disclaimers -- shows how much the five most notable non-Citi banks (Bank of America, Goldman Sachs, J.P. Morgan, Morgan Stanley and Wells Fargo) could have paid out, versus how much they did pay out. Citi is omitted because, you know, oops, it's not allowed to pay out anything. Since Bank of America and Goldman Sachs originally asked to pay out more than they were allowed, I threw their initial asks into the chart too.

The gist is that these five banks' capital plans call for about $37 billion of increased payout to shareholders, but that they could have returned $60 billion. So they only got 62 percent of what they could have, and left $23 billion on the table. If the banks are gaming the stress tests, 62 percent is not a great score.

What does this tell you? One potential lesson is that Bank of America is, relatively speaking, a model of efficient capital gaming (this year!): It initially asked to give back shareholders about 124 percent of the "right" number, got rebuffed, and settled on 81 percent of the right number. Goldman, who also initially overshot, ended up paying out only a middling 62 percent of its maximum. Morgan Stanley, which has generally made a fetish of behaving better than is strictly necessary, seems to have been the most conservative of the big banks about its capital return, though that could just mean that my math is wrong.

More generally, though: Nobody here is really a model of efficient capital gaming. Everyone will be keeping a capital cushion of at least a billion or two dollars more than the Fed requires. Perhaps because the banks are optimizing for something other than maximum capital return -- good relations with the Fed, for instance, or extra safety in a crisis (who knows, anything's possible). Or perhaps it's because the Fed's foggy stress tests are working as intended, forcing banks to be more conservative in their capital planning than they'd otherwise be -- and giving them less of a chance to game the tests than they'd prefer.

(Matt Levine writes about Wall Street and the financial world for Bloomberg View.)

  1. Methodology:

    • I look at the Fed's DFAST results, which provide among other things each bank's stressed Tier 1 common equity ratio (Tier 1 common divided by risk-weighted assets), Tier 1 capital ratio (Tier 1 capital divided by risk-weighted assets), Tier 1 leverage ratio (Tier 1 capital divided by total assets, loosely), and risk-weighted assets.
    • From this, you can rough out each bank's stressed Tier 1 capital and Tier 1 common dollar amounts, though this is approximate because the DFAST only gives you risk-weighted assets as of Q4 2015, but the minimum capital may come at an earlier time.
    • Then I look at the CCAR results, which provide the same capital ratios pro forma for each bank's proposed capital plan. From the difference between the DFAST and CCAR results, you can compute how much lower each bank's Tier 1 capital and Tier 1 common amounts are after the bank's capital action. Assuming the bank's capital action is roughly "give back capital," and not thinking too hard about timing or anything, that difference should be the amount of capital that each bank gives back.
    • There are differences in how much each bank appears to be giving back, depending on whether you count Tier 1 common or total Tier 1 capital; for instance, on my math, Wells Fargo seems to be returning $18.6 billion using the Tier 1 capital ratio but $24.4 billion using the Tier 1 common ratio. There are various possible explanations for this but they boil down to, I don't know enough about the methodology or the banks' plans to thoroughly reverse-engineer the stress tests. For the chart, I just took the minimum of those two numbers.
    • Similarly, I looked at the minimum required ratios -- 5 percent Tier 1 common, 6 percent Tier 1 capital, 4 percent Tier 1 leverage -- and turned those into dollar numbers. (Again based on Q4 2015 RWAs, making them inaccurate in some cases.) Subtract these numbers from the DFAST results and you get how much each bank is allowed to give back; again I took the minimum.
    • (Except for Morgan Stanley -- weirdly, Morgan Stanley shows a projected minimum Tier 1 risk-based capital ratio of 7.2 percent under CCAR and 7.1 percent under DFAST. So its capital return plan seems to increase its stressed Tier 1 capital. I don't know what to tell you. I used the Tier 1 common ratio for Morgan Stanley.)
    • You can check these numbers against the banks' actual announced capital plans and they're, you know, mostly close enough. So Bank of America is increasing its dividend to $0.05 (adding 4 cents a share times 10.5 billion shares for let's say six quarters through the end of 2015) and buying back $4 billion of stock, for a total of $6.5 billion, versus my rough math of $6.6 billion. J.P. Morgan's dividend increase and $6.5 billion buyback get you around $7.0 billion of capital return, versus my $7.3 billion. Wells Fargo's numbers come to $18.6 billion (using a $48.50 stock price for its 350-million-share buyback), right on top of mine. Goldman did not disclose its actual plan. Morgan Stanley's numbers come to around $1.6 billion, versus my $0.8 billion; like I said, I don't really get what's going on with Morgan Stanley. (Otherwise, the discrepancies are small and should be about timing, compounding, etc.)

    To be clear, all of these payout numbers are expressed as additions to the pre-stress-test capital plans, i.e. share repurchases plus dividend increases, so they don't count existing dividends.

  2. Again, beyond its current penny-a-share dividend. Citi is also omitted because I don't entirely understand its stress test results either -- see footnote 5 of my post this morning, the Tier 1 capital and Tier 1 common numbers are weirdly different.

  3. Though even if you swap in $1.6 billion for my $0.8 billion of capital return (see the end of footnote 1), that's still under just 50 percent of the $3.2 billion that I calculate Morgan Stanley is allowed to pay out. Higher than the 25 percent that I get using my baseline math, but still at the bottom of this range. J.P. Morgan is the next-lowest percentage, at 54 percent.

To contact the author on this story:
Matthew S Levine at

To contact the editor on this story:
Toby Harshaw at