Some Questions about the New York Fed's TBTF Research

I read the paper written by a Vice President of the New York Fed, João Santos, “Evidence from theBond Market on Banks’ ‘Too-Big-to-Fail’ Subsidy” after seeing a few stories about it in more generalist media yesterday. The paper concludes that the top 5 banks (the ones assumed to be "too big to fail") have lower funding costs than smaller banks and that this cannot be explained solely by a size premium, because the author investigates size premia in bond offerings of other types of issuers and concludes that the large banks' funding cost advantage is statistically significantly greater than large non-bank issuers' advantage.  

The paper on first blush is a little better than some other work in this area, because it shows an awareness of the size premium in capital markets.  As I laid out earlier this year in another post, all types of capital markets -- leveraged loans, high yield bonds, and publicly traded equities, to name just three  -- reflect size premia, i.e., the cost of debt or equity, as the case may be, tends to be lower for large companies than small, all else being equal.  Thus, the fact that a large bank has a lower cost of debt, whether bond or deposit, than a small one does not prove that the difference arises because the creditor believes the larger bank has an implicit government guarantee, given that the same difference exists in numerous types of industries where no one has argued there is an implicit government guarantee.  As Santos himself says, his paper's consideration of the size premium is a "novel" analysis and thus improves on the other work in this area.

That said, the paper raised for me a number of questions, which unfortunately cannot be answered without access to his data.    In addition, his methodology struck me as paradoxical, if not self-contradictory,

The paper analyzes the spread paid by the "top 5" issuers (by assets) compared to the other issuers, in three different categories: banks, nonbank financial corporations and nonfinancial corporations, from 1985 through 2009, which is a fairly long timeframe in which a lot of M&A activity occurred in the banking industry.  It does not identify the issuers by name which might be useful but the likely suspects are fairly easy to identify -- Citi, B of A, Chase Manhattan and its successors among the banks;  GE and Philip Morris among the nonfinancials; and one presumes Merrill among the nonbank financials.

The paper confirms that the top 5 issuers in each category, as a group, had lower spreads over Treasuries than the other issuers in their category, confirming the sie premium effect.  For large banks, specifically, their bonds carry average spreads of 44 bps than the other banks.   The author further claims that, after he "controls" for the size, maturity and S&P rating of each issue, the discount drops to 41 bps.  The author does not explain what he means by "controlling" for those variables.  It may seem pedantic to ask what  he means, but when one considers his sample size, it actually becomes a more worrisome question -- his entire sample size of bonds issued by banks consists of 438 bonds over a 25-year period, and he eliminates from his analysis nearly 75 of those, consisting of issues bearing less than an "A" rating from S&P, so really the analyzed sample size is more like 360 issues, or 1.2 per month of the period sampled. When you break that down by rating, as he does, it becomes even smaller: roughly 15% of the sample is "AA" rated, which amounts to only about 2.5 issues a year.  So I question what the "control" and how legitimately meaningful it was.  Put another way, how many small bank "AA" issues were there that had the same size and maturity as, and were issued contemporaneously with, an issue from a large "AA" bank?  I suspect too few to make the claim of comparability meaningful.

Moreover, in the pool of the "top 5" banks, a "15%" share of the pool equates to just 1 bank!   Could it be that, in addition to a size premium, bond buyers are willing to pay a "best in class" premium for the top 1 bank?  I could easily imagine that to be the case, but the author does not examine the possibility.

He next applies the same analysis to the other two categories, non bank financial issuers and non-financial issuers.  He finds that the largest issuers in those two categories had, on average, a 79 bps discount and a 76 bps discount, respectively, in the spread they were charged versus their smaller peers, in each case, before adjusting for controls.

Without more, these calculations would greatly undermine any argument that there exists a "TBTF subsidy" for large banks, since the largest issuers in both nonbank subsets received a larger discount from the credit markets than the 44 bps the largest banks received.   

But the author again "controls" for size, maturity and rating among the nonbank issuers and concludes that the discounts large issuers received were much smaller: 22 and 47.  The size of the reduction in the discount generated by the adjustment points up the importance of understanding what it means for him to "control" for these variables and how legitimate the control is.  But even so, after the adjustments he makes, the 47 bps discount received by the top 5 nonfinancial issuers still exceeds the 41 bps earned by the top 5 banks, again making it difficult to argue the top 5 banks were getting a "TBTF subsidy" as opposed to a widespread size premium. It's an odd premium that shows up in one comparison and not in another.

To reach his conclusion that the top 5 banks are in fact getting a TBTF subsidy, the author breaks the sample down into isolated rating subsets; that is, he compares (1)  the difference in spread on AA-rated large bank issues vs  all other AA-rated bank issues to (2) the difference in spread on AA-rated large nonbank issues vs the difference in spread on all other AA-rated nonbank issues.  And then he repeats this analysis at the "A"-rated level. 

At the "AA" level, he concludes that the large banks obtained a discount that was 91 bps greater than large non-bank financial issuers got vs their smaller peers, while at the "A" level, it was just 16 bps and the latter, he admits, was "not statistically significant".  Although a statistically significant finding of a 91 bps advantage seems dramatic, recall that the size of the "AA" sample amounted to only about one issue per month, or alternatively to just one of the 5 largest banks under study; I question how tightly such a sparse dataset can be compared.  Whereas, by contrast, "A" rated bonds amount to almost 80% of the issues by the top 5 banks which he samples, and he cannot find a statistically significant advantage compared to nonbanks in that much larger pool.  That suggests his conclusion about the "AA" set may be much less meaningful and robust than he believes.

Measured against the size benefit for large nonfinancial issuers, Santos concludes that the largest banks obtain 53 bps more of a discount at the "AA" level and 50 bps more at the "A" level.  However, here, he admits the former conclusion is not statistically significant, only the latter.   But there is an anomaly about the latter that needs exploring. 

In contrast to every other subset he slices and dices the data into, the category of the largest "A" rated nonfinancial issuers, instead of obtaining a discount vs. their peers, pays, according to Santos, a 10 bps premium to borrow money versus their smaller peers (see Table 4, column 4 of his paper).  Meanwhile the issues in the adjacent tiers, AA and BBB, received discounts of 14-17 bps.  

That seems very odd and hard to fathom.  One has to believe that the CFO's of "A" rated nonfinancial corporations were systematically failing over 25 years to negotiate the kind of discount that CFOs at "AA" or "BBB" rated nonfinancial corporations, as well as CFO's at banks and other financial issuers with the same "A" rating, were systematically winning from bond buyers.  It's theoretically possible, but hard to believe, especially when you note that the "A" rated category is far and away the largest category -- almost 41% -- of the bond issues by the 5 largest nonfinancial corporations being sampled, so the failing has to have been really widespread.  I think some investigation of the data is warranted there:  for example, was there one issuer who, while "top 5" and indeed A-rated, had some kind of contingent liability risk that the market persistently priced higher than the rating implied?

So to sum up the paper's analysis:

(1) it finds no evidence of a TBTF subsidy at the "all bonds" level, nor at the "all bonds adjusted for size, maturity and rating" level;

(2) it finds no statistically significant evidence of a TBTF subsidy for AA-rated large banks compared to nonfinancial issuers of similar size and rating; and

(3) it finds no statistically significant evidence of a TBTF subsidy for A-rated large banks compared to nonbank financial issuers of similar size and rating.

But,

(4) it finds statistically significant evidence of a TBTF subsidy for AA-rated large banks compared to nonbank financial issuers of similar size and rating, however,  that pool is a very, very small one - it may be literally just one bank - and one can question how much weight should be given to a conclusion based on such a limited amount of data; and

(5) it finds statistically significant evidence of a TBTF subsidy for A-rated large banks compared to nonfinancial issuers of similar size and rating, but there is a striking anomaly in the data used for comparison that suggests caution be given to the conclusion.  

Those are the questions I had about the report, but I also mentioned that the paper's methodology seemed paradoxical or even contradictory at the big picture level.   By that I mean that I don't see how anyone can claim that the bond market is pricing bonds as TBTF when 80% of the supposedly subsidized bonds are only rated  "A".  TBTF means the bond buyers think the government is going to bail the issuer out.  So if that were the case, buyers would be bidding that debt up to AAA-level prices, Fannie Mae/Freddie Mac type pricing.  Why is the bond just rated A and priced at an A rated price if the market supposedly sees so little risk to repayment?  Remember the "A"" rated category is by far the biggest so this is not a quirk, it's the gist of the question.  If an "A" rated bank is TBTF, what about the smaller banks that had higher ratings throughout this period?  What about the nonbank financial issuers and  nonfinancial issuers that had higher ratings throughout this period?  How does one explain the way their bonds are priced relative to the "A" rated larger banks with no TBTF premium?  It doesn't make any sense.  The paper does not consider this conundrum.   I think this paper may suffer from a forest-for-the-trees problem that research is often prone to.

Another point to make in closing is that the paper estimates the "value" of the subsidy to be in the range of $60 million for the average AA-rated issue and $1.5 million for the average A-rated issue, in each case over the life of the issue.  He does not explain how he calculated those and I could not figure them out from the data he supplied.  But, given that the vast bulk of the large bank offerings he examined were "A"-rated, his analysis, even if correct, indicates a really, really, really small problem, a factor to be kept in mind when listening to those populist politicians who have portrayed this as something of great significance, requiring governmental intervention of some unprecedented nature and scale.