A Mechanism for LIBOR

Brian Coulter is Adjunct Professor at the Western University Richard Ivey School of Business, Joel Shapiro is Associate Professor of Finance and Peter Zimmerman is a DPhil Candidate in Financial Economics at the University of Oxford Saïd Business School. This post is based on a recent article by the authors published in the Review of Finance.

The London Interbank Offered Rate (LIBOR) is intended to reflect the average rate at which banks can borrow in the unsecured market. It is computed by taking a trimmed mean of the daily reported borrowing rates of the banks on a panel. But panel banks may have incentives to manipulate LIBOR to profit off of their exposure to the benchmark. Manipulating one of the rates by even a fraction of a basis point can bring substantial gains: the market for derivative and loan products that use LIBOR rates has been estimated at greater than $300 trillion.

Over recent years, it has come to light that several banks and their employees have been complicit in the manipulation of LIBOR. This ongoing scandal has already resulted in fines over $9 billion. This has pushed regulators to decide that the benchmark must be based on transactions rather than reports. However, the recent drying-up of the interbank unsecured market has convinced regulators that there are not enough transactions to continue with LIBOR. On 27 July Andrew Bailey, Chief Executive of the Financial Conduct Authority (FCA), indicated that the replacement of the sterling LIBOR benchmark with the Sterling Overnight Index Average (SONIA) rate would occur by 2022. The US is likely to replace the US dollar LIBOR benchmark with a new rate based on repo transactions.

But introducing benchmarks based on transactions will not eliminate incentives to manipulate the markets. As we have seen in many other benchmarks, market participants can manipulate transactions. From “banging-the-close” in gold, to timing the 60-second window in FX, market participants have repeatedly gamed transaction-based benchmarks. Tying the benchmark to transactions makes gaming more costly, but recent experience tells us that, as long as the benefits continue to outweigh the costs, incentives to manipulate will remain. Furthermore, tying the benchmark to a particular market—such as overnight lending or repurchase transactions (“repos”)—presupposes that the market will always be liquid. In our article, A Mechanism for LIBOR, we propose a mechanism which addresses manipulation problems and produces a reliable benchmark even when markets are not liquid.

We take a transactions-based benchmark as our starting point, and then apply two refinements. First, we propose a reporting mechanism—which we call the ‘revealed preference algorithm’ (RPA)—which elicits the rates at which the banks on the LIBOR panel would lend to one another at a given point in time. Second, we create a comparison rate using the elicited rates and the set of transactions. The comparison rate is used to assess which banks’ transactions appear to be manipulated. These banks are then fined. This reduces manipulation and produces an unbiased estimate of the true rate. We set the fines and the comparison rate to minimize the variance of this estimate. The optimal choice may involve using only the elicited rates for fines—and it will always involve those rates—so the mechanism can still operate even when markets are illiquid.

In our model, banks make their transactions, taking into account that they can potentially manipulate LIBOR but may be fined for doing so. The administrator sets LIBOR using only the banks’ transactions. Then, the administrator elicits rates at which each bank would lend to one another using the “revealed preference algorithm” (RPA). To ensure truthful reporting, a threshold rate is then chosen randomly, and if the offered rate was below this threshold rate, then the offering bank must lend with positive probability. [1] [2] As the LIBOR calculation does not include RPA rates, banks cannot influence LIBOR by their actions in the algorithm.

The RPA does not prevent banks from deliberately transacting at an uneconomic rate in order to manipulate LIBOR. We show that levying fines on transactions that appear to be manipulated leads to a more accurate benchmark, though we cannot reduce manipulation to zero. The mechanism can be made revenue-neutral, so that all banks pay a mean fine of zero.

The LIBOR manipulation scandal triggered investigations into the manipulation of other benchmarks. We discuss how our model applies to the proposed pure transaction-based benchmarks (SONIA and the US proposal) and to other benchmarks that have been subject to manipulation, such as the WM/Reuters 4pm Foreign Exchange fix and ISDAFIX, a common reference rate for interest rate swaps. In short, our model could also be used quite easily in those fixings with very little adaptation.

The complete article is available here.


1Lending would be to the LIBOR administrator, and the promised repayment would be made only if an equivalent loan to the bank would have been repaid. Thus, a synthetic bank loan may be created, which market participants will view as equivalent to lending directly to the bank in question.(go back)

2Making participants state rates and potentially lend at those rates (i.e. making them “put their money where their mouth is”) is similar to approaches in other jurisdictions, such as Australia and Israel.(go back)

Both comments and trackbacks are currently closed.