The scandal over the London Interbank Offered Rate (LIBOR) —one of the most important benchmarks for interbank lending—has been heating up, as governments investigate apparent collusion to manipulate base lending rates by some of the world’s most prominent banks.The British Bankers’ Association calculates the LIBOR rate by polling 18 global banks on every trading day about the price at which they could borrow from other leading financial institutions. Authorities now allege that bankers deliberately misreported lending capabilities around the time of the 2008-2009 financial crisis.
But a research paper from the Federal Reserve’s Jeremy Berkowitz (now a professor at the University of Houston) indicates that the Fed was worried about LIBOR manipulation as early as February 1998.
In that paper, Berkowitz tests alternative methods of calculating LIBOR rates specifically because the Fed has already seen three examples of misreporting from one bank in early 1996. While he writes those incidents off as being “undoubtedly…unintentionally misreported,” the composition of a paper to prevent against such errors suggests a great deal of interest from the Fed in this subject.
The BBA currently controls for misreporting for LIBOR dollar rates by throwing out the four highest and lowest estimates that banks submit and averaging the remaining rates—a “trimmed-mean” approach. Berkowitz expresses concern about the soundness of this practice, and another alternative he suggests (“Winsorized-mean”):
Unfortunately, both the trimmed-mean and Winsorized-mean tend to be quite sensitive to faulty data in small samples. In fact, it is possible to show that a 10% trimmed-mean based on fewer than 20 observations can break down in the presence of even 2 outliers. This means there is no bound on the bias or variance of the estimate, given sufficiently bad outliers.
He calculates that as few as four inaccurate reports (unintentional or manipulative) could bias the BBA LIBOR rate in that system (which at the time saw contributions from 16 banks). This led him to suggest the use of other methods to determine the rate:
The status-quo procedure is expected to perform well as long as there are no more than 4 false reports. The BBA procedure is to always drop the four highest and four lowest prices, thereby, in all likelihood, eliminating the four false reports. This approach should not be accurate if the breakdown point [number of inaccurate reports] of 4 is exceeded. On the other hand, the bootstrapped truncated-mean, the median and the bootstrapped Huber estimator sacrifice some efficiency to guard against larger numbers of false reports…
With a sample of 16 (as with Libor), the Huber-estimate can handle as many as 7 false reports. In fact, Jureckova and Sen (1996) prove that robust M-estimators have the largest breakdown point possible. This suggests a strong a priori justification for considering Huber-estimates of settlement prices where multiple false reports are plausible.
While Rabinowitz’s paper does not imply that banks had been colluding to manipulate rates at the time it was written, it suggests that the Fed was already be concerned about the effects of inaccurate reporting by banks about their lending practices 10 years before the financial crisis.
Further, acknowledgments that a very small contingent of banks potentially could manipulate rates suggests that the Fed may very well have seen this coming.
Traders speculate that LIBOR manipulation has taken place pretty openly for many years, and this is a sign that the Fed acknowledged this possibility very early on.