Gerri Willis and Jeb Hensarling have some trouble with trillions. (Fox Business News) Gerri Willis and Jeb Hensarling have some trouble with trillions. (Fox Business News)

Gerri Willis of Fox Business News just can't get over the amount of support the U.S. government gives big banks. We support them, she said, citing Elizabeth Warren's comments to Ben Bernanke during a Senate hearing yesterday, to the tune of $83 trillion (over what time frame is left unspecified). "Some $83 trillion, $83 trillion," she told House Financial Services Committee Chairman Jeb Hensarling yesterday. "You know, I can't get over that number."

"Absolutely," Hensarling replied.

First, let me reassure Willis and Hensarling: They're off by three orders of magnitude. The estimate they're referring to, calculated by Bloomberg View editorial writers, states that the federal government subsidizes big banks to the tune of $83 billion a year. The U.S. economy produced a little less than $16 trillion last year, so it'd require at least five years of effort by the entire United States to provide $83 trillion in backing.

But the Bloomberg estimate raises a serious question: $83 billion isn't chump change, even in the context of the federal budget. Are we really giving big banks that much?

First off, it's important to be clear on what that number means. The government isn't pulling $83 billion from the budget and handing it out in check form to Goldman Sachs, Morgan Stanley and so forth. What Bloomberg View is arguing is that big banks get an implicit boost from customers, stockholders and other interested parties knowing that if they fail, the federal government will likely bail them out. That's a boost that local savings and loans operations, or even small investment banks, don't get. That boost is worth $83 billion to them — but no one is getting an $83 billion check from taxpayers.

To get this estimate, Bloomberg View turned to two researchers — Kenichi Ueda of the International Monetary Fund and Beatrice Weder di Mauro of the University of Mainz — who released a working paper for the IMF trying to estimate the effect of banks' support from governments on their credit ratings. The idea was that better credit ratings translated into cheaper borrowing in the form of bonds, which reduced the banks' operating costs.

Ueda and Weder di Mauro rely on ratings from Fitch, which issues a "support rating" for banks, indicating the likelihood of a private bank or government bailing them out should things turn south. The rating runs from 1 to 5, with 1 indicating zero chance of support in bad times, and 5 indicating that a bailout is certain. They also put out a "support rating floor" number, which measures the government's ability to provide support; an absence of a ranking here means support is expected to come from another bank instead. The researchers then measured how much changes in the level of government support, as inferred through these measures, affects Fitch's overall long-term rating for the bank.

There's much to be skeptical about here. Rating agencies were notoriously bad at measuring the value of assets before the financial crisis, and Fitch was no exception. In one notorious case, it gave its top rating, AAA, to a collateralized debt obligation (CDO) from Credit Suisse that ended up losing $125 million. Like Moody's and S&P, its business model relies on the companies it rates paying it for the ratings, which provides an incentive to inflate ratings and tends to be less effective than other business models, such as those in which the investors who use the ratings pay for them.

Moreover, how does Fitch actually know what the government will do? If Fitch's government support estimates are wrong, then none of the conclusions follow. Given Fitch's history, assuming they get government support estimates right is quite optimistic.

The next step is even more tenuous. The researchers don't actually compare the ratings Fitch gives banks to the cost of funding for those banks. Instead, they say in a footnote, without any elaboration, "Moody’s and Fitch ratings are comparable," and then proceed to use 2000-vintage Moody's estimates from economist Farouk Soussa (see first paper here).

Based on that method, U.S banks with government support get a 3.6-4.5 notch advantage from Fitch. Since each notch tends to reduce a bank's cost of funding by .22 percentage points, that means the government-backed banks have a 0.8 percentage point advantage. Bloomberg's follow-up method was simple. They multiplied the 0.8 point advantage by the total liabilities of the top ten U.S. banks, and got $83 billion. If the 0.9 point advantage is closer to right, the cost would be closer to $93 billion.

So is the Bloomberg estimate low, or high, or right on? I honestly have no idea. The study only works if the initial government support estimates from Fitch hold. Maybe they do, maybe they don't. Even if they do, the calculations only hold if the relationship between banks and Moody's ratings in 2000 are applicable to banks and Fitch ratings in 2009. Maybe they are, maybe they aren't. There's reason to worry that we've got a garbage-in, garbage-out situation here. The initial numbers — the government support estimates — are only as good as Fitch, and so the $83 billion number is only as accurate as Fitch is in turn.

At any rate, it's not $83 trillion.