For the last few weeks, Kamakura’s weekly “implied forecast” for the U.S. Treasury yield curve and the U.S. dollar libor-swap curve has commented on the abrupt and stair-step nature of shifts in short term interest rates on U.S. Eurodollar deposits as reported by the Board of Governors of the Federal Reserve in its H15 statistical release. This blog examines the often large discrepancies between this rate and the London interbank offered rates officially released by the British Bankers Association and compares them to default probabilities of the Libor panel banks.
We’ve observed two trends in recent entries on this blog. The first trend is to note the relatively modest volume in corporate and sovereign credit default swaps that has been reported by www.dtcc.com. The second is to note that the implied forecast for interest rate swap spreads has recently been strongly affected by Eurodollar quotations on the Federal Reserve’s statistical release H15. Movements in 1, 3, and 6 month Eurodollar rates reported by the Fed have moved in stair-step fashion and in diverse ways for the three short term rates. These movements have been so concerning that we asked the same question we posed in the CDS market: Are the rates we’re seeing really real?
In order to answer that question, we examined both the default probabilities and funding costs quoted by the Libor panel banks that have been in place since May, 2009, according to www.bbalibor.com:
- Bank of America
- Bank of Tokyo-Mitsubishi UFJ
- Barclays Bank PLC
- Citibank NA
- Credit Suisse
- Deutsche Bank AG
- HSBC
- JP Morgan Chase
- Lloyds Banking Group
- Norinchukin Bank
- Rabobank
- Royal Bank of Canada
- Royal Bank of Scotland Group
- Societe Generale
- UBS AG
- WestLB AG
We note that three of these banking companies do not have publicly traded parent companies:
- Norinchukin Bank (owned by member organizations)
- Rabobank (owned by member organizations)
- WestLB AG (state owned)
We collected the following information daily since June 1, 2009 to May 27, 2010 to insure that changes in the composition of the Libor panel members would not be relevant to any conclusions drawn:
- Quotations by each bank to Reuters, the calculation agent for the British Bankers Association, as reported on Bloomberg.
- The official Libor figures for 1 month, 3 month and 6 month maturities reported via Bloomberg
- The Eurodollar bid rates for 1 month, 3 months, and 6 months as reported on the same days by the Federal Reserve’s H15 statistical release
- The Kamakura Risk Information Services version 5 annualized 1 month default probabilities for the publicly traded parent companies of the 13 Libor panel banks
Before examining the data, we had the following expectations:
- That the London interbank deposit rates reported by the British Bankers Association and the Federal Reserve would be similar, if not identical
- That the spread between these deposit rates and matched maturity U.S. Treasury yields would be highly correlated with the default probabilities of the Libor panel banks
Both of these expectations proved to be wrong in dramatic fashion, at least as far as BBA Libor is concerned.
Our first graph shows the 1 month Eurodollar interbank deposit rates reported by the Federal Reserve and the British Bankers Association from June 1, 2009 to May 27, 2010.
The graph above shows that the 1 month Eurodollar rate quoted by the Federal Reserve was twice as high as the www.bbalibor.com official quotation for 1 month Libor in June 2009 and that it has been substantially higher than official Libor for almost the entire period since June 1, 2009. How can this be?
A recent paper by JP Morgan explores the potential reasons for the differential. In that paper (“Research Note: The Outlook for Libor,” U.S. Fixed Income Strategy, JP Morgan, May 16, 2008), the authors discuss the “credibility crisis” facing the Libor calculation of the British Bankers Association and the claim that the reported Libor figures are much lower than the actual borrowing costs of the Libor panel banks.
The full text of the paper may be found here:
http://www.cmegroup.com/company/history/magazine/pdf/Libor_Outlook.pdf
The process by which Libor is calculated is discussed in the paper and on www.bbalibor.com. The 16 panel banks are polled by Reuters, calculation agent for the BBA, daily. The top 4 quotations and bottom 4 quotations are omitted, and the remaining 8 quotations are averaged to get the official Libor figures. By definition, this calculation process will generally result in Libor quotes that are “too low” for much of the panel and too high for some of the panel banks as well. During this period, the long held image of the Eurodollar deposit market as a homogeneous market where all but the worst banks paid the same rates was clearly inconsistent with the reality of the market place.
Can this calculation process explain the huge gap versus the Eurodollar deposit costs reported by the Federal Reserve? First, we need to know how the data collection process works at the Federal Reserve. The JP Morgan paper summarizes the process as follows:
On the Federal Reserve H15 statistical release, a mere two footnotes describe the Eurodollar interest rates. They do not name the number of banks polled nor their identity. The author has been unable to find other web-based explanation for the Eurodollar deposit statistics collected by the Fed, and a request for further information via the New York Federal Reserve was not answered. By visual inspection, the graph above shows that the Federal Reserve series is much more likely to remain unchanged from day to day and that the Federal Reserve series is much more likely to move in rounded increments like 5 basis points. Neither of these inspires confidence in the quality of the data, except for the fact that the quotations are much higher than the BBA series.
Is it possible that both data series are correct? Could it be that the panels are so different that the answers are simply not the same? Of course, that’s a possibility. There is another explanation that has been much discussed in the press and on the internet—that Libor panel banks have simply not been truthful in quoting funding costs to the BBA. There are a number of reasons that lying may work as a strategy:
- Third party observers of ABC Bank’s quoted funding costs in the BBA survey may believe they are accurate, so lying can serve to help reduce funding costs by misleading other banks
- ABC Bank may note that every other panel bank is understating its funding costs and to not do so would imply that ABC Bank itself is having funding difficulties
- ABC Bank and every other Libor panel bank that is a participant in the U.S. dollar swap market are in a conflict of interest position. If ABC is a net payer of floating rate libor (and net recipient of fixed rate dollars), understating true funding costs will reduce interest rate swap payments if ABC is one of the 8 panel banks whose quotes go into the final libor figures.
These rationales for falsehoods are easy to understand. Is there any penalty for lying? Sadly, no. Like the market for corporate and sovereign credit default swaps, Libor panel participants are not bound by their quotes and can’t be “picked off” by other market participants when they lie.
Does the comparison with allegedly risk free constant maturity Treasury yields for 1, 3 and 6 months shed any light on which series is most realistic? Looking at the results for 1 month, the answer is not obvious:
The high volatility and frequent rise and fall of the one month constant maturity Treasury yield quoted on the H15 report (investment basis) are dramatically different than the pattern of movement in either of the 1 month Eurodollar deposit rate series.
How about default probabilities of the Libor panel banks? Do they shed any light on the issue of realism in the Euro deposit quotations? Keep in mind that four of the 16 Libor panel banks have received very large government subsidies in the 2007-2010 credit crisis:
- Bank of America
- Citibank NA
- Lloyds Banking Group
- Royal Bank of Scotland Group
A fifth bank, WestLB AG, is state owned. The dramatic moves in the 1 month default probabilities in the 16 Libor panel bank members can be seen in this graph:
For each business day, the graph shows the maximum, minimum, average, median, and “panel average” of the Libor panel banks’ parent firm default probabilities. The “panel average” is calculated as follows for the 13 banks with publicly listed parents: from the 13 observations that day on 1 month annualized default probabilities, the 3 highest and 3 lowest are omitted and the average of the remaining 7 default probabilities is reported. This comes as close as possible to the process by which BBA Libor itself is determined.
When we plot this panel average versus the other three interest rate series, we get the following image:
As is apparent from the graph, there is a significant dis-connect between the default risk of the panel banks and the interest rate series. This is not a new phenomenon. In the first few years after the collapse of Long Term Credit Bank of Japan and Nippon Credit Bank, it was common for banks to leave deposits at the Bank of Japan at zero interest rather than depositing at then-current market rates in the yen interbank market because rates had been pushed well below the levels at which credit risk was correctly reflected.
One can make another interpretation as well. Kamakura’s default probabilities represent the ‘probability of failure’ for all classes of liabilities. For any given specific class of liability, there is also the probability of rescue, conditional on failure having occurred. Interbank rates in the USD market may also be implying a very high probability of rescue, even if the probability of failure is high for that bank in particular.
The two credit spread series have only a 50.9% correlation with each other because the underlying Eurodollar deposit quotes bear so little relationship with each other. How do the two series correlate with the default probabilities calculated as the “panel average”?
BBA Libor series correlation with panel average default rate: 35.5%
Federal Reserve H15 series correlation with panel default rate: 85.0%
The graph below shows that there is actually a surprisingly high degree of consistency between the H15 Eurodollar deposit series and the 1 month “panel average” default probability from KRIS. On average, the default probability is approximately 5 times the deposit rate minus 0.16% with an adjusted r-squared of 72%.
All other things being equal, as an analyst, this would indicate to me that the Federal Reserve series, despite its flaws, more realistically reflects the risk of the Libor panel banks than the BBA series. All other things being equal, I would much prefer to tie “Libor” related contracts to the Federal Reserve series than to the BBA series. Unfortunately, as this standard ISDA document makes clear, the Federal Reserve H15 Eurodollar deposit series is not even listed on this reference chart (“ISDA Definitions-Rate Cross Reference Chart”) for deal documentation, even though many other data elements on the H15 statistical release are listed:
http://www.isda.org/c_and_a/pdf/ISDA-Definitions-Rate-Cross-Reference-Chart-051804.pdf
The British Bankers Association has reacted positively to worries that the BBA Libor series is biased low, and for good reason. It responded with the paper “Understanding the construction and operation of BBA Libor-strengthening for the future.” The complete text is available here:
http://www.bbalibor.com/bba/jsp/polopoly.jsp?d=1623&a=15877
This brief analysis shows the same conclusions that we reached in the CDS market: that there is the potential for manipulation of a key financial statistic, and market participants should beware. When in doubt, pay fixed, not floating, on swaps.
Donald R. van Deventer
Kamakura Corporation
Honolulu, June 7, 2010