Banks Aren't Too Big to Fail Unless They Fail
Here's the Government Accountability Office announcement and study on whether there's a too-big-to-fail subsidy for big banks. The answer is no-ish but it's complicated. It's complicated in part by the fact that the GAO ran 42 different regression models, and they all got different answers:
All 42 models found that larger bank holding companies had lower bond funding costs than smaller ones in 2008 and 2009, while more than half of the models found that larger bank holding companies had higher bond funding costs than smaller ones in 2011 through 2013, given the average level of credit risk each year (see figure). However, the models' comparisons of bond funding costs for bank holding companies of different sizes varied depending on the level of credit risk. For example, in hypothetical scenarios where levels of credit risk in every year from 2010 to 2013 are assumed to be as high as they were during the financial crisis, GAO's analysis suggests that large bank holding companies might have had lower funding costs than smaller ones in recent years. However, reforms in the Dodd-Frank Wall Street Reform and Consumer Protection Act, such as enhanced standards for capital and liquidity, could enhance the stability of the financial system and make such a credit risk scenario less likely.
I think you need a model for these models. 1 One simple model is that the too-big-to-fail subsidy -- meaning the amount of money that big banks save on funding costs because their creditors assume that if anything goes wrong the government will bail them out -- is like a put option, putting some floor on the value of a bank's assets. 2 When the put option is at-the-money -- when things are bad and the bank looks like it might default without government support -- then it's worth a lot. When the put option is out of the money -- when things are good and the bank is fine on its own -- then it's worth less.
So now it's worth less than it was worth in 2008, or in a hypothetical 2013 with 2008's credit conditions (what?). In fact, it has negative value, which is weird in a put option, but not that weird in life. It's like insurance: You pay a premium every month, and it pays off when things go bad. Some months the premium is worth more than the insurance, in expectation, and some months it's worth less. That's how insurance works: It looks like a bad deal in good times, but a good deal in bad times.
Now you might say: But banks don't pay for the too-big-to-fail subsidy! We need to charge them for it! But I'm not quite sure that's true. Here is a study arguing that post-crisis banking reforms cost the six biggest U.S. banks about $35 billion a year, most of that from increased capital requirements. 3 Some of those requirements apply mostly to the biggest banks, and there are other bigness-focused requirements too -- stress tests, living wills, liquidity buffers -- that further push up the cost of being big. These costs aren't explicitly paid to the government as insurance premium, but that's okay: The government imposes those costs in order to reduce the likelihood that the insurance will have to pay.
A simpler way of putting this is: Big banks actually have higher funding costs than little banks. If you try to isolate their bond funding costs, and then regress against various measures of credit quality, you'll get whatever you get depending on what you regress. But if you just divide funding costs (interest expense and maybe cost of equity) by assets, the big banks have higher costs, because they rely more on expensive long-term funding than on the cheap financing (mainly deposits) that little banks use. 4 That more expensive funding is the premium they pay for their improved odds of getting a bailout.
The Bloomberg View editors remain troubled by too-big-to-fail, and fair enough. But it's worth spending a few minutes pretending to take this GAO report literally, because if you do, its message is something like: Bigness is not the problem. In 2014, in good credit conditions, big banks can ... fail? Probably? Maybe? That's what Dodd-Frank's "living wills" and "orderly liquidation authority" mean: There are mechanisms for regulators to come in, grab a bank, zero its equity, write down its holding-company creditors, and keep its systemic businesses intact and functioning. 5
And this study says: Those mechanisms should, in expectation, work. So if, for instance, Goldman Sachs announces tomorrow that Lloyd Blankfein has stolen $100 billion and lit it on fire, leaving Goldman with a negative capital position and a diminished reputation, then regulators could step in, zero Goldman's shareholders, ding its bondholders, but prevent contagion to the broader financial system. Derivatives contracts would still be honored, repo creditors wouldn't flee, bank depositors would be fine, etc. You may or may not believe that, 6 but the point of the GAO study is that the market does. 7
And every so often huge banks fail idiosyncratically, so I guess that's nice. But the bigger worry, poorly captured by the name "too big to fail," is that a big bank will be in trouble at the same time, and for the same reasons, as every other big bank. The other day I mentioned this Minneapolis Fed economic policy paper called "Too Correlated to Fail," which argues that "the anticipation of bailouts creates incentives for banks to herd in the sense of making similar investments":
Policymakers do not intervene when big banks are threatened simply because those banks are too big. Rather, they intervene because the potential systemic costs resulting from bank failure are considered too big.
And those systemic costs are much higher in bad times. If you start Lehman Nephews tomorrow, grow it to $600 billion of assets next week by buying Argentinean bonds and knockoff messaging-app stocks, and then blow it up and file for bankruptcy, the financial system will probably be fine. The Lehman Brothers bankruptcy was a problem not just as a cause but as a symptom: Everyone had worrying exposure to mortgages, everyone knew that everyone had worrying exposures to mortgages, then everyone saw that those mortgage exposures blew up Lehman, everyone realized that everyone had exposures to Lehman in addition to their already worrying exposures to mortgages, and everyone panicked. In circumstances like those, pulling out a "living will" and haircutting Lehman's bonds would not have done much to stem the panic.
And the GAO study tells us that too: If credit conditions were now what they were in 2008, the government probably wouldn't let big banks fail.
I guess you can regret that, but ultimately all it means is that the financial system is financially systemic. Panics are bad, and there's a role for central banks to stop them, and that role tends to be profitable in the long term, but look like a bailout in the short term. As Tim Geithner says, "the existence of firehouses doesn't cause fires": You can't make financial crises go away by just promising really hard never to bail anyone out.
But that doesn't mean that nothing has been accomplished. As the GAO says, the new boring well-capitalized banking regime "could enhance the stability of the financial system and make such a credit risk scenario less likely." That would be nice. And it's a more important, and more achievable, goal than convincing the market that no one will ever be bailed out again.
This column does not necessarily reflect the opinion of Bloomberg View's editorial board or Bloomberg LP, its owners and investors.
By the way, the details of the regressions in the report are a little interesting. Obviously you could choose different models, or weight them differently, or whatever. I want to focus on one thing, or one cluster of things. Here are some variables that go on the right-hand side of some of the regression equations:
Volatility: Standard deviation of equity prices over the quarter, option-implied volatility for the quarter, the standard deviation of equity returns over the quarter, the standard deviation of excess equity returns over the quarter, and the standard deviation of earnings.
Some of those things make perfect sense: More volatile earnings, for instance, clearly make you a riskier credit, and so should be part of an explanation of your funding costs. But some, particularly option-implied equity volatility, bug me a bit. I think a plausible model of the too-big-to-fail premium has to be: If things are going badly for you, the government will step in to keep you around as a going concern (or at least sell you to JPMorgan at an equity price above zero). in other words, a plausible TBTF model involves at least some chance of a (partial) rescue for the equity. Which should dampen implied volatility, since your equity probably can't go to zero. Option-implied volatility should almost be a dependent variable. (Actually, the GAO mentions in its literature review (see page 44) that several previous studies have used equity-implied probabilities of default as pretty much the sole independent variable, measuring the difference between CDS-implied probability of default and equity-implied probability of default, and that seems even weirder. I don't think there's a lot of reason to be confident that equity gets zeroed in any bailout.)
Or its debt? How you imagine this option gets a little complicated but doesn't matter much.
It assumes costs of capital for the big banks of 10.5 (Wells Fargo) to 14.1 (Morgan Stanley) percent, and measures increased capital requirements since 2007.
Anyway you don't have to believe that, but if you do, you can sort of roughly convert it to GAO's scale, which measures the too-big-to-fail premium in terms of holding company bond costs. I get around 315 basis points (using Bloomberg's "long-term debt" as a rough proxy for the GAO's "holding company bonds"):
[imgviz image_id:iGfi.5QQ7ph8 type:image]
That overstates the case a bit, perhaps, as some of it just comes from increased capital across the system, not limited to the biggest banks. (On the other hand, it ignores other costs such as liquidity buffers and more long-term debt.) The GAO's numbers are all over the place; some models show subsidies of greater than 600 basis points in 2008, while the models currently show subsidies ranging from over 100 basis points to negative 100-ish basis points.
The Treasury Department frequently makes this argument, e.g. in this speech, or in this presentation noting that large banks have higher borrowing costs than regional banks. I once made a similar argument.
This is explained in various places but one convenient place is the summary on pages 17-18 of the GAO report.
For myself, I guess I believe it for some of the big banks, though not JPMorgan. If you're huge and you run the triparty repo system, you're immortal.
More strictly, the point of the GAO study is that a majority of its regression models believe that the market believes that, but whatever.
To contact the author on this story:
Matthew S Levine at email@example.com
To contact the editor on this story:
Toby Harshaw at firstname.lastname@example.org