members, quantifying their possible impact on the ﬁnal rate. In line with the argument of
Kyle and Viswanathan (2008), we deﬁne manipulation to be any submission that differs
from an honest and truthful answer to the question asked of the panel banks.
Furthermore, we explicitly take into account the possibility of collusion between several
market participants. Our setup allows us to quantify such effects for the actual rate-
setting process in place during our sample period, and compare it to several alternative
rate-ﬁxing procedures. Moreover, we can determine the effect of the panel size on
manipulation outcomes. These results allow us to comment on important details of the
rate-setting process, as well as on broader questions, such as the use of actual transaction
data as an alternative information source.
A dispassionate appraisal of the events of the past few years and the discussion among
market professionals, journalists and regulators suggests that two conceptually distinct
issues became conﬂated in the heat of the discussion. The ﬁrst relates to the potential
for manipulation of Libor and Euribor –which are both determined by similar
methodologies, but subject to the supervision of different bodies –under the current
method of eliciting quotes from a given panel of banks. This issue naturally leads to a
discussion of how the effect of manipulation might be mitigated, if not eliminated, by the
use of an alternative deﬁnition of the rate, without altering the method of collecting the
basic data from the panel of banks. The second and logically separate issue relates to
changing the nature of the data themselves, for example, by only collecting data on
actual transactions rather than using submitted quotes and, thus, introducing greater
transparency and reliability into the process. The latter would be a much more
fundamental change, and raises additional questions about how the liquidity of the rates
for different maturities and currencies would be affected under the restriction of being
based on transactions data.
Within the context of the current rate-setting process, there are three factors that
potentially affect the cross-sectional and time-series variation in the submissions, which,
in turn, inﬂuence the computation of the trimmed mean used to set the rate. The ﬁrst is the
variation in the credit quality of the banks represented by the panel. Depending on the
particular question asked of the panel banks, the rate submitted by a bank reﬂects, to a
certain degree, the credit risk premium built into the borrowing rates.
If the banks have
very different credit qualities in the judgement of the market, the rates submitted could
reﬂect this variation. The second is the variation in the liquidity positions of the banks in
the panel, which reﬂect their need for additional funding. If some banks are ﬂush with
funds of a given maturity in a currency, while others are starved of them, the rates they
submit for this currency/maturity should be very different, even if their credit standings
are similar. The third is due to the potential manipulation of the rates, as has been alleged
and even demonstrated in at least some cases, by regulatory and legal action. Since it is
impossible to disentangle the effect of manipulation from the credit risk and liquidity
effects, without detailed data on the other two effects, we address questions based solely
Kyle and Viswanathan (2008) deﬁne manipulation as any trading strategy that reduces price
efﬁciency or market liquidity.
In 2014, the Market Participants Group on Reforming Interest Rate Benchmarks lead by
Darrell Dufﬁe submitted its ﬁnal report discussing various of these issues, see Market
Participants Group on Reforming Interest Rate Benchmarks (2014).
See section 2 for details of the underlying questions used in the rate-setting process.
© 2017 John Wiley & Sons, Ltd.
606 Alexander Eisl, Rainer Jankowitsch and Marti G. Subrahmanyam