Banks Are Getting Better at Taking Stress Tests
How many banks should fail the Fed's annual stress tests? I guess there is some argument that you need to fail a few banks here and there, to keep them on their toes, or to prove to the public that the tests are super rigorous. But the straightforward answer is "zero." Zero banks should fail the stress tests, just like zero banks should do mortgage fraud or manipulate markets or launder money for terrorists. All of those goals are a bit aspirational, sure, but they are not phrased as goals; they are phrased as rules. So all the banks are supposed to pass the stress tests.
This year they finally did, yaaay:
The Federal Reserve said all 31 big banks subjected to a stress test have sufficient capital to absorb losses during a sharp and prolonged economic downturn.
It’s the first time since the central bank started stress tests in 2009 that no firm fell below any of the main capital thresholds.
The stress tests now, five years in, are just a part of U.S. bank capital regulation: Capital regulation now means that you're supposed to have at least X amount of capital now (on various measures), and at least Y amount in the Fed's annual severe stress scenario. It took a while for all the banks to figure out how (the Fed thinks) the stress scenario would affect their capital, but now they have, and they make sure to have enough capital to meet the minimum. This is true even though this year's test seems to have been harder than last year's.
On the other hand, people often note that this week's stress test (sometimes horribly called the DFAST) is in some sense less important, or at least less challenging, than next week's test (sometimes horribly called CCAR). This week's test is just pass/fail, measuring how banks would withstand a downturn based on their current capitalizations. But next week's test measures the banks' requested capital-return plans against the stress scenario. This week's test measures: Are you a minimally competent bank? Next week's test measures two tougher things:
- How much money can you pay out to shareholders?
- How good are you at reading the Fed's mind?
The first thing is important because it is kind of how all this stress testing is scored. The point of the stress test is to prove that you are so well-capitalized -- so able to withstand even "a sharp and prolonged economic downturn" with aplomb -- that you have money to spare and can give it back to shareholders. If you end up not giving the money back to shareholders, then in some important sense you've lost.
The second thing is a good measure of, are you a clever bank? It is in some sense easy to pass this week's stress test: just have a ton of capital. That way, no matter what the Fed thinks will happen during that "sharp and prolonged" downturn, you'll have enough stressed capital to pass the test.
But for the big universal banks, next week's game is more challenging: You don't just want to be above the minimum stressed capital, you want to be as close as possible to it without going under. You're supposed to operate efficiently, and operating efficiently means, loosely speaking, having as little equity capital as you need to comply with regulations. And since the regulations here are not "you need at least X amount of capital now," but rather "you need at least Y amount of capital after accounting for a severe economic downturn," your ability to ask for the right amount of capital return is a good test of your ability to guess what (the Fed thinks) will happen to you in a crisis.
Also there is a more direct test of competence insofar as banks can also flunk next week's tests on purely subjective grounds, if the Fed thinks they're qualitatively inept or haven't taken the stress tests seriously enough.
Next week we'll find out how the banks did, though they themselves apparently found out yesterday and were given a chance to revise their capital-return ask, if their first attempt was excessive. People seem to think that Goldman's probably was:
Goldman Sachs Group Inc.’s cushion of extra capital in the latest Federal Reserve stress test isn’t enough for the firm to match last year’s payout of dividends and share repurchases.
The Fed projected Thursday that Goldman Sachs’s total risk-based capital ratio would fall to 8.1 percent in a severe economic downturn, weaker than what the central bank estimated in last year’s test and barely above the 8 percent minimum. The figure was below Goldman Sachs’s own calculation of 13 percent, as well as those of analysts including Credit Suisse Group AG’s Susan Roth Katzke, who estimated 11.3 percent.
My rough estimate on that is that Goldman would have about $53.9 billion of total risk-based capital in the stressed scenario, versus a minimum requirement of about $53.3 billion, leaving it free to return $600 or $700 million of cash to shareholders (beyond existing dividends). Last year Goldman spent $5.5 billion buying back stock. So this year seems like it will be rather less.
So everyone seems to be in reasonable shape except Goldman, though it depends on how much it asked for. In fact, some of these numbers feel downright roomy: Some banks have quite a lot of leeway to return money to shareholders.
Should they use it? I said above that part of the point of the stress tests is to show off each bank's skill and derring-do at guessing how much capital the Fed will let it return, and its commitment to shareholders in returning that capital to them. On that measure, the closer you get without going over, the better you've done. But other approaches are possible. You might think that the Fed's stress scenario is not the only thing that banks should worry about, and that banks should be cautious about capital on their own behalf as well as because the Fed demands it. You might even think that the stress tests would be a pretty empty exercise if they were just about reading the Fed's mind and not about getting the banks themselves to prepare for future risks.
Dodd-Frank Act Stress Test, Comprehensive Capital Analysis and Review, respectively, but it doesn't matter.
In addition to being an easy approach, this has the benefit of being a robust one:
Some banking experts are worried that the emphasis on the stress tests might give rise to a false sense of security. In particular, they contend, it is impossible to know what an actual crisis will look like ahead of time. The Fed assumes that large Wall Street firms have to deal with the default of a large trading partner.
“What are they assuming happens in that default? What are they assuming about contagion?” said Anat R. Admati, a professor of finance and economics at Stanford. “It takes a leap of faith to feel safe because of these tests.”
Disclosure! I used to work at Goldman. I also had some restricted stock until very recently, but then I realized it became unrestricted and I sold it. So these disclaimers will be shorter from here on out. Also, here's a story about Goldman tweeting emojis.
The math here goes like this:
- I took the "minimum" columns for each of the capital ratios in the severely adverse scenarios on page 21 of the stress test results.
- I estimated actual capital amounts by multiplying those ratios by the relevant end-of-period risk-weighted asset numbers in the individual bank results. So I estimated Tier 1 capital as Tier 1 capital ratio times "general approach" RWAs, common equity Tier 1 as CET1 ratio times "standardized approach" RWAs, Tier 1 capital as T1 ratio times "standardized approach" RWAs, total capital as total capital ratio times "standardized approach" RWAs. (See page 16 of the results.) Note that this uses minimum capital ratios and end-of-period asset numbers, so introduces some inaccuracy: A lot of the banks hit their minimums before the end of the period, so I'm comparing somewhat wrong numbers here.
- I also estimated total (non-risk-weighted) stressed assets by assuming that the numerator (capital) for the Tier 1 leverage ratio was the same as the Tier 1 capital amount that I calculated, and then dividing by the stressed Tier 1 leverage ratio to get the stressed total assets.
- I then figured out the minimum stressed capital amounts on each of those measures, by multiplying the minimum ratios (which you can find on e.g. page 13 of last year's CCAR) times the estimated assets.
- Then I subtracted the minimum capital amounts from the estimated capital amounts to get the excess capital on each measure, and took the minimum of those numbers to get the actual excess capital.
- Last year's buybacks are from 10-Ks or Bloomberg.
- I've ignored dividends -- existing dividends are more or less part of the stress test assumptions, see page 14 of the results -- but of course one possibility is a dividend increase instead of a buyback.
This is unscientific in various ways and should not be taken as investment advice! We'll see the actual answers next week! (Here is my version of this math from last year, by the way.)
The Fed test was tougher than many analysts anticipated for firms most reliant on investment banking and trading. Goldman Sachs, Morgan Stanley, and JPMorgan Chase & Co. all fell closer to the minimum than Katzke predicted on at least one measure and appear to lack room to match the payouts she estimated they would request.
This column does not necessarily reflect the opinion of Bloomberg View's editorial board or Bloomberg LP, its owners and investors.
To contact the author on this story:
Matt Levine at email@example.com
To contact the editor on this story:
Zara Kessler at firstname.lastname@example.org