![Sample our Mathematics & Statistics journals, sign in here to start your FREE access for 14 days]({"type":"image" , "src" : "/sda/5943/TOC-Subject-Sample-2021-Mathematics-Statistics.jpg"})
Abstract
Using a time-varying cointegration framework, this paper examines the alleged manipulation of the London interbank offered rate (Libor) during the 2007–2009 financial crisis. Bank quotes are found to be poor indicators of their financing costs in the crisis period. The aberration in the estimated values of the cointegrating and error correction parameters governing the long-run equilibrium relationship between bank quotes and the final Libor suggests banks were submitting lower quotes. Further analysis which controls for an individual bank’s credit risk, market wide credit and liquidity risks, and a common market factor, demonstrate possible evidence of Libor rigging during the crisis period.
Acknowledgements
The authors gratefully acknowledge the referee’s useful comments which improve the quality of the paper. Any errors are the responsibility of the authors. The authors also thank Rosa Abrantes-Metz for providing part of the data used in this research.
Notes
No potential conflict of interest was reported by the authors.
1 According to Wrightson, the research arm of interbank brokerage ICAP, there were allegations made by market participants that the one-month Libor was lower than the actual borrowing rates (see Money Market Observer, September 3, 2007).
2 The study that first reported the allegation of Libor manipulation was published on April 16, 2008 in the Wall Street Journal entitled ‘Libor fog: Bankers cast doubt on key rate amid crisis’. The follow-up study entitled ‘Study casts doubt on key rate’ was published in the same newspaper on May 29 that same year.
3 Media reports indicate that Barclay, Rabobank, Royal Bank of Scotland, and the Union Bank of Switzerland have been fined for manipulating Libor quotes.
4 While there are several non-linear ARDL studies established based on the works of Pesaran and Shin (Citation1999) and Pesaran et al. (Citation2001), they focus on a cointegrating non-linear ARDL model in which short- and long-run non-linearities are introduced via positive and negative partial sum decompositions of the explanatory variables. Park and Hahn (Citation1999) propose a cointegrating regression in the spirit of Engle and Granger (Citation1987) with parameters that vary with time. They model the elements of a (single) cointegrating vector as smooth functions of time via Fourier series expansions. They derive the asymptotic properties of the semi-nonparametric sieve estimators involved and propose several residual-based specification tests. Our approach is different from the methods that have been developed so far.
5 The Libor quotes have a much shorter maturity, only one month, compared to the 5-year CDS yields. It would be ideal to obtain a measure of bank’s credit worthiness of the same maturity as the Libor quotes, but due to data paucity we have had to resort to the use of a proxy with a different term structure.
6 The former is based on a comparison between Libor quotes and matched CDS spreads, and the latter focus on the absence of bias in the Euro Libor at the start of the crisis period.
7 Kuo et al. (Citation2012), show that Libor survey responses diverge from two other measures of banks’ borrowing rates, particularly during periods of financial stress. The two borrowing rates are the bank bids at the Federal Reserve Term Auction Facility (TAF) and the Fedwire payments data.
8 The OIS is subtracted from the Libor in order to take out market expectations of future central bank policy rates. The resulting Libor–OIS spread in effect measures the rate required to compensate banks for higher default risk on loans and the rate that reflects the liquidity in the inter-bank loan market, and hence the market-wide credit and liquidity risks.
9 Normal value is defined by the typical value of and
in the period prior to the start of the subprime crisis when little evidence of Libor manipulation has been reported.
10 There is an ongoing debate over whether the Libor–OIS spread should be regarded as a measure of credit risk or liquidity stress. Brunnemeier (Citation2009) reasons that the Libor–OIS measures credit risk because a relatively secured lending cost (OIS) is subtracted from an unsecured cost (Libor) by construction. However, Imakubo et al. (Citation2008) shows that the widening of the Libor–OIS spread during the GFC is mainly driven by liquidity stress based on decomposition estimates from the CDS data. Schwarz (Citation2016), using microstructure data to construct alternative credit and liquidity measures, finds that liquidity explains a large part of the soaring of the spread. Ji and In (Citation2010), and Ji (Citation2012) espouse the view of Imakubo et al. (Citation2008) arguing that in the cross-currency context the spread better represents liquidity stress than credit risk.
11 Both the CDS and the Libor–OIS spreads are obtained from Bloomberg. Unit root tests on both datasets indicate that CDS spreads and the Libor–OIS spreads are I(1) process in levels. Results are not reported here to conserve space, but they are available from the authors upon request.
12 Although it is possible to model changes in the parameters values using a non-Gaussian state-space approach such as the one developed by Kitagawa (Citation1987), it remains controversial whether the adoption of this approach that accommodates abrupt changes in the parameter values is consistent with intuition that banks conceal attempts in manipulating their quote submissions. If we believe that banks manipulated rates in a ‘subtle’ manner to avoid detection, then it is unlikely that the coefficient would change abruptly. Furthermore, the use of a Student-t density would account for aberration in the parameter values and its distribution through its fat tails.
13 The one-month USD Libor is the interquartile trimmed mean of the reported interbank offer rates provided each day by a panel of banks during the London trading session (11am London time). The 15 banks are Bank of Tokyo-Mitsubishi UFJ (BTMU), Bank of America (BOA), Barclays, JP Morgan Chase (JPMChase), Citigroup (Citi), Credit Suisse, Deutsche Bank, HSBC, Lloyds, Norinchuckin, Rabobank, Royal Bank of Canada (RBC), Royal Bank of Scotland (RBS), UBS, and West LB.
14 The rate submissions made by the pre-defined set of banks are the response to the Libor question: ‘At what rate could you borrow funds, were you to do so by asking for and then accepting inter-bank offers in a reasonable market size just prior to 11am?’ (British Bankers’ Association, Citation2012).
15 To compute the t-test for dependent means in a given sample period, we first compute the grand mean, which is calculated from the median of all banks at each period t for a given sample period. The standard error of the t-test is constructed as
where
is the correlation of the median
for bank i and the grand mean. The same method is applied for the error correction parameter
.
16 Various news sources which show the timeline associated with the Libor rigging scandal have indicated that 10 banks were involved with the Libor manipulation. These banks are Barclays Citigroup, JPMorgan Chase, Deutsche Bank, Royal Bank of Scotland, Société Générale, Rabobank, Lloyds, Credit Suisse and the Union Bank of Switzerland. (See http://www.telegraph.co.uk/finance/financial-crime/11767437/Libor-trial-Tom-Hayes-found-guilty-of-rigging-rates.html; http://dealbook.nytimes.com/2013/12/04/e-u-imposes-1-7-billion-euros-in-fines-over-rate-rigging-scandal/; http://dealbook.nytimes.com/2013/10/29/rabobank-to-pay-more-than-1-billion-in-libor-settlement-chairman-resigns/).