Bloomberg has an editorial today about how the government is subsidizing the top ten U.S. banks by $83 billion a year and maybe it should stop doing that. Because the editorial is getting a lot of attention, and because it is wrong, let’s discuss it.
Here is Bloomberg:
Lately, economists have tried to pin down exactly how much the subsidy lowers big banks’ borrowing costs. In one relatively thorough effort, two researchers — Kenichi Ueda of the International Monetary Fund and Beatrice Weder di Mauro of the University of Mainz — put the number at about 0.8 percentage point. The discount applies to all their liabilities, including bonds and customer deposits.
Here are Ueda and di Mauro:
[W]hen issuing a five-year bond, a three-notch rating increase translates into a funding advantage of 5 bp to 128 bp, depending on the riskiness of the institution. At the mid-point, it is 66.5 bp for a three-notch improvement, or 22bp for one-notch improvement. Using this and the overall rating bonuses described in the previous paragraph, we can evaluate the overall funding cost advantage of SIFIs as around 60bp in 2007 and 80bp in 2009.
Let’s break that down. Their paper:
- uses Fitch ratings of (1) bank credit quality and (2) likelihood of bailouts, and assumes that the market agrees with Fitch ratings;
- finds that systemically important financial institutions tended to get about three ratings notches of uplift from expected government support in 2007 (pre-crisis), and about four notches or so in 2009 (post-crisis);
- finds that three notches of ratings translates into 5 to 128 basis points of cheaper funding, on average, on a five-year bond; and
- takes the midpoint of that range to assume that banks get about 66.5bps of cheaper funding due to that ratings uplift (~60bps in 2007, ~80bps in 2009).
Fair enough. You might choke a little on the first point, though. Remember how ratings agencies suck? It seems a little problematic to assume that ratings agencies – actually one ratings agency – actually Fitch – know both (1) how likely banks are to default and (2) how likely governments are to step in and save them. As, to be fair, Ueda and di Mauro know:
The use of ratings might be considered problematic because rating agencies have been known to make mistakes in their judgments. For instance, they have been under heavy criticism for overrating structured products in the wake of the financial crisis. However, whether rating agencies assess default risks correctly is not important for the question at hand. All that matters is that markets use ratings in pricing debt instruments and those ratings influence funding costs. This has been the case.
You can go ahead and evaluate that claim if you want; they cite 2005 and 2006 papers on how ratings determine bank spreads. There is probably some truth to the fact that, within the banking sector, ratings correlate with spreads, though which way the causation flows is less clear. I am unaware of any evidence that markets trust Fitch’s “Support Rating” as an accurate indicator of likely external support. This is irrelevant if you believe that the mechanism for bank credit is (1) Fitch determines a bank’s individual credit strength and likelihood of external support, (2) Fitch uses those determinations to make a rating, and (3) the market then reacts to that rating by pricing higher-rated banks tighter than lower-rated banks. That mechanism seems … I’m still going with problematic? But, okay, fine.
A perhaps bigger problem is Bloomberg’s blind use of the midpoint of the effect on funding costs. Here is how Ueda and di Mauro describe the funding advantage of a three-notch ratings uplift: “5-8bp for an A rated bank, 23 bp for a BBB rated bank, 61 bp for a BB rated bank, and 128 bp for a B rated bank.” Then it uses the midpoint and gets 66.5bps, which becomes 80bps for the 4-ish notches of uplift that banks get in 2009. Fine whatever.
Bloomberg applies this 80bps funding uplift to JPMorgan (Fitch holdco rating of A+), BofA (A), Citi (A), Wells Fargo (AA-), and Goldman Sachs (A). See the problem? As Ueda and di Mauro note, “government support is more ‘valuable’ at lower rating levels.” My best guess is that the paper supports about a 31bps funding uplift for those banks (equivalent to going from BBB- to A), not 80bps.1
The biggest problem, though, is Bloomberg’s claim that “the discount applies to all their liabilities, including bonds and customer deposits.” I can’t find any support for that in Ueda and di Mauro’s paper, which in fact says the exact opposite: that the 80bps uplift applies in the case of “issuing a five-year bond.”
And Bloomberg’s unsupported interpretation is counterintuitive. Bank deposits in the U.S. are (largely) explicitly insured by the federal government; it makes no sense that an implicit TBTF premium on top of the explicit FDIC insurance would make your deposit any safer, or make you charge a lower rate for that deposit. Also, the notion that TBTF banks are paying eighty basis points less for FDIC insured demand deposits than other banks is sort of facially absurd: if you’re getting 0.8% interest on your checking account at a little bank, let me know and I’ll move my money there.2
Similarly other sources of short-term funding tend to have guarantees that are not TBTF-related. Repo borrowing is overcollateralized by high-quality securities; there is a repo rate determined by the quality of the collateral, not the quality of the borrower. Derivatives liabilities are also often collateralized, though to be fair ratings help determine whether they are or not.
The more reasonable interpretation of Ueda and di Mauro’s paper is that it means what it says: that the ~80bps (really ~31bps!) uplift applies to “issuing a five-year bond.” And since banks fund only a small minority of their assets with long-term unsecured debt, that makes the too big to fail subsidy worth considerably less than Bloomberg thinks. They compute a $17.24bn value of the subsidy for JPMorgan, for instance, by multiplying 0.8% by JPMorgan’s total liabilities of $2,155bn. Here’s how those liabilities break down:
I submit to you that Ueda and di Mauro’s numbers apply to “long-term debt” of $249bn (and assume that that debt has a five-year average maturity), only, and if you use the 31bps uplift actually implied by their paper, then the subsidy is worth about $770 million a year, not $17 billion.3 Lazily extrapolating, that suggests that Bloomberg’s $83 billion subsidy number for the ten biggest banks should really be closer to $3.7 billion, or a little under one-twentieth of the number Bloomberg came up with. Or maybe as much as twice that, depending how you count, but in any case not within an order of magnitude of “an amount roughly equal to their typical annual profits.” It’s pleasingly elegant to think that the too-big-to-fail subsidy provides ~100% of the biggest banks’ annual profits, but it’s also totally wrong. Or at least, there is no evidence for it.
I mean! There is a too big to fail subsidy! Lots of people try to quantify it! Some of those efforts get close to Bloomberg’s $83 billion order of magnitude, though for all banks worldwide, not for just the top 10 U.S. banks. Ueda and di Mauro’s attempt – which never mentions Bloomberg’s $83 billion number – is not a bad one; I don’t love the reliance on Fitch’s judgment of what banks are TBTF and by how much, but I understand where they’re coming from. But Bloomberg’s adaptation of it is probably intended more for entertainment value than as an actual effort to quantify how much the too big to fail subsidy is worth. The $83 billion number is shocking, so it’s got that going for it, but not much else.
1. TThe paper is pretty unclear about whether it means “moving three notches up from A is worth 5-8bps” or “moving three notches up to A is worth 5-8bps.” Also about what a “notch” is, though context clues suggest that A to AA is one notch, not three (viz. not A -> A+ -> AA- -> AA). [Update: I’m becoming convinced that Ueda and di Mauro are just confused and the correct answer is little notches (+/flat/-), meaning that a three/four notch move up to A is from BBB/BBB-, and a three/four notch move up from A is to AA/AA+. I don’t know.]
Intuitively moving from single-B up three notches to A (B->BB->BBB->A) has to be worth more than 5-8bps, so that suggests that the right interpretation is that they mean that moving up three notches from A is worth 5-8bps. Except three notches above A is (1) AA, (2) AAA, (3) AAAA, which is not a thing. So, I have no idea. Probably my eyebrow-raise in the text is wrong and they’re talking about a three-notch move from B to A for these banks being worth 128bps, and the four-notch post crisis move (from CCC to A) being worth even more. Which makes 80bps still wrong, but too low. That said though you should violently distrust a claim that without government support JPMorgan would be rated CCC. I mean, you can do whatever you want, but I will violently distrust that claim. The averages in the Ueda and di Mauro paper don’t necessarily translate directly into implied ratings for the top end of banks.
Best guess is now that the right category to use is the BBB one (i.e. up 3/4 notches up from BBB/BBB- to A), or 23bps for three notches, or I guess ~31bps for four. Not 80bps.
2. Oh let’s be a little more scientific about it. In 3Q2012, JPMorgan paid an average of 0.34% on its deposits (page 6 here). Sterling Bancorp, a NY-based bank with roughly 1/1,000 of JPMorgan’s assets, paid 0.50% (page 56). That is not an 80 basis point difference! It’s more than I’d expected, honestly, but still one-fifth of Bloomberg’s claim.
3. There’s an argument that the subsidy also subsidizes things like CP, “other borrowed funds,” and maybe accounts payable, though there is not, I think, an argument that it’s worth 80bps a year for those short-term liabilities.