By James Kwak
There is a common phenomenon in legal disputes over the value of something, be it a company, a piece of land, or a person’s expected lifetime earnings. Each side hires an “expert” who produces an estimate based on some kind of model. And miraculously, every single time, the expert for the party that wants a higher number comes up with a high number, while the expert for the party that wants a lower number comes up with a low number. No one is surprised by this.
Yesterday, the Federal Reserve posted the results of the latest periodic bank stress tests mandated by the Dodd-Frank Act. For these tests, the Fed comes up with various scenarios of how things could go badly in the economy, and the goal is to see how banks’ income statements and balance sheets would respond. The key metrics are the banks’ capital ratios; the goal is to identify if, in bad states of the world, the banks would still remain solvent. If not, the banks won’t be allowed to do things that reduce their capital ratios today, like paying dividends or buying back stock.
For the most part, the results look pretty good: capital levels even under the severely adverse scenario should remain above the levels reached during the 2008–2009 crisis. (Of course, there are several huge caveats here. You have to believe: first, that the scenarios are sufficiently pessimistic; second, that the banks’ current financials are accurately represented; third, that the model is sensible; and fourth, that the capital levels set by current law are high enough.)
But there’s something else going on here. As part of the stress testing routine, each bank is supposed to do its own simulation of how it would respond to the scenarios specified by the Fed, using its own internal model. And—surprise, surprise!—the banks virtually uniformly predict that they will do better than the Fed.
At the high end, Goldman Sachs and Citigroup predict that they would have capital levels 3.9 and 3.0 percentage points, respectively, higher than expected by the Federal Reserve. Since the Fed predicted minimum Tier 1 capital levels for these two banks of 6.8 and 7.0 percent, those are huge differences: 57 percent higher for Goldman and 41% higher for Citi. The other four big banks also claimed their capital levels would be 0.2 to 2.6 percentage points higher than in the Fed’s model.
Now if everyone were being above board here, the expected difference between the banks’ estimates and the Fed’s estimates should be zero. But of course that’s not the case. Banks do things to make money, and in this exercise their goal is to make the case that they can get by with less rather than more capital. There are honest differences of opinion on how to model things, but you can systematically make plausible choices that produce higher rather than lower numbers. And that’s almost certainly what’s going on. Which means that the banks’ estimates aren’t worth the electrons who died (OK, not literally) sending them to you across the Internet.
But, you may be thinking, isn’t the Federal Reserve systematically trying to produce low numbers? Not necessarily. The Fed’s incentive in the first instance isn’t to force banks to maintain more capital; it’s to make sure that banks are holding the right amount of capital. The Fed is supposed to protect the financial system and ensure economic growth, so if you believe in the capital-growth tradeoff, the Fed doesn’t have an incentive to force banks to hold lots of capital. (And if you don’t believe in it, then you should already agree with Admati and Hellwig that every bank should have lots more capital.) In other words, Fed economists don’t make any more money by arguing that banks need more capital.
Although there’s no a priori reason why the Fed as an institution would want banks to hold more capital, it’s also possible that the specific people at the Fed do think that banks should be holding more capital, and they are using the stress tests as a backdoor way to push that agenda. But if that’s the case, there’s another, more powerful tool they should be using: they should be boosting the leverage ratio (which, counterintuitively, is measured as equity over assets, not debt over equity).
Finally, this whole thing proves a point that I argued a long time ago: capital doesn’t exist as an object in the world. It’s inherently probabilistic, since it is based off of the values of things whose value depends on unknown probability distributions. That’s why it’s possible for Goldman to argue with a straight face that its capital will be 57 percent higher in some state of the world than the Federal Reserve thinks it will be. And that’s why, if you’re counting on capital requirements to protect the financial system from disaster, you had better err far on the side of safety.