LIBOR's poker

Abstract

The recent LIBOR and ISDAfix manipulation scandals have inspired discussions about survey-based financial benchmarks. I investigate surveys as the statistical inference problem plagued by the principal-agent frictions between benchmark administrators and banks. Without knowing the distribution of private signals, an administrator can implement a sufficient expected quadratic penalty through random audits and post-audit fines to minimize reporting errors and induce a distribution-free expected benchmark bias. Sufficiently frequent random audits that discipline the bank with maximum borrowing cost can mitigate the impact of constraints on post-audit fines. Finally, delaying the release of bank reports may not be effective in fighting collusion.