Testing uncovered interest rate parity using LIBOR

Abstract

We test uncovered interest rate parity (UIP) using London InterBank Offered Rate (LIBOR) interest rates for a wide range of maturities. In contrast to other markets, LIBOR markets have minimal frictions. Whereas most previous studies reject UIP, we find that UIP holds for several short-term LIBOR maturities using block bootstrap panel unit root tests suggested by Palm et al. (2011) and cointegration techniques by Westerlund (2007). Furthermore, the estimation results suggest that the speed of adjustment to the long-run equilibrium marginally differs across the maturity of the underlying instrument, thus supporting the efficient market hypothesis.

Keywords:

• UIP
• LIBOR
• block bootstrap panel unit root test
• panel cointegration

JEL Classification:

• G12
• G15
• F31

Notes

¹ Reviews are provided by Froot and Thaler (1990), MacDonald and Taylor (1992), Flood and Taylor (1996), Isard (1996), Sarno and Taylor (2002), Pasricha (2006) and Alper et al. (2009).

² After we had finished our research, allegations of London InterBank Offered Rate (LIBOR) manipulation by some banks were put forward. However, statistical support for LIBOR manipulation is not very convincing. Although the findings of Snider and Youle (2010) substantiate LIBOR manipulation, Abrantes-Metz et al. (2012) report only some evidence of anticompetitive market behaviour by the participating banks. Kuo et al. (2012) also find some deviations of LIBOR from other bank borrowing costs without emphasizing that this reflects misreporting of LIBOR. Findings of Monticini and Thornton (2013) suggest that the underreporting of the LIBORs by some banks reduced the reported LIBORs specifically, for 1 and 3 months. Intriguingly, markets may have traded on the basis of biased data. This may have serious economic costs. Our research sheds some light on this issue as well, as our results provide support for uncovered interest rate parity (UIP).

³ For example, Reuter DataStream reports 22 types of interest rates with a 3-month maturity for Canadian debt market.

⁴ Using a stochastic process, LIBOR market model (LMM) attempts to predict the behaviour of the LIBOR interest rates. Initially proposed by Brace et al. (1997), Miltersen et al. (1997) and Jamshidian (1997), LMMs are continuously updated.

⁵ The British Bankers Association (BBA) started reporting LIBOR for Danish Kroner and New Zealand Dollar from 16 June 2003 and for Swedish Krona from 23 January 2006.

⁶ LIBOR is based on aggregation of nonbinding quotes, as opposed to actual transactions, which might explain why most researchers have not used this important information source. However, Michaud and Upper (2008) note that the BBA tries to reduce the possible impact of this by eliminating the highest and lowest quartiles of the distribution and averaging the remaining quotes.

⁷ Taylor (Taylor, 1995, p. 14) defines UIP as, ‘if the risk-neutral efficient markets hypothesis holds, then the expected foreign exchange gain from holding one currency rather than another (the expected exchange rate change) must be just offset by the opportunity cost of holding funds in this currency rather than the other (the interest rate differential)’.

⁸ It should be noted that the perfect foresight assumption, in the absence of frictions, may turn UIP into CIP. We are thankful to a referee for highlighting this point.

⁹ Notable among the first generation of unit root tests are Levin et al. (2002) [commonly known as LLC], Breitung (2000), Im et al. (2003) [commonly known as IPS], Maddala and Wu (1999) [commonly known as Fischer test], Harris and Tzavalis (1999) and Hadri (2000).

¹⁰ Widely used second-generation unit root tests are proposed by e.g., Bai and Ng (2004), Moon and Perron (2004), Pesaran (2007), and Choi (2005).

¹¹ In order to provide support to our argument in Section V, we discuss the estimates of the group mean test statistics also. Therefore, this footnote provides a brief note on the construction of these test statistics. Group mean statistics and capture the individual-specific heterogeneity. These statistics test the alternative hypothesis that at least one member of the panel is cointegrated. Similar to panel statistics, the group mean statistics are constructed using a conventional variance estimator and a Newey–West-type long-run variance estimator , where is the conventional SE and is the Newey–West SE.

¹² We interchangeably use the word ‘members’ for the ‘individual cross-sections’ (or currencies) of the panel.

¹³ We refer to Kao and Chiang (2001) for further details of the FMOLS estimation methodology.

¹⁴ LIBORs are, by definition, offered rates. The BBA defines it as, ‘the rate at which an individual contributor panel bank could borrow funds, were it to do so by asking for and then accepting interbank offers in reasonable market size, just prior to 11:00 AM London time.’ For details see http://www.bbalibor.com/rates/historical.

¹⁵ The exchange rates are International Monetary Fund (IMF)-reported rates for various currencies in US Dollar. The IMF website explains these rates as follows: ‘the rates are reported daily to the Fund by the issuing central bank. Rates are normally reported for members whose currencies are used in Fund financial transactions’. Furthermore, the website indicates that the IMF posts representative and SDR exchange rates every 20 minutes from 11:00 am to 6:00 pm US EST Monday to Friday except for the holidays. For details see http://www.imf.org/external/np/fin/data/param_rms_mth.aspx.

¹⁶ To check that this result is not driven by persistence in high-frequency data, we have applied unit root test to weekly and monthly data, as well. The results are not different from those reported in Table 1. For brevity sake, these results are not reported here, but can be provided on request.

¹⁷ As discussed earlier, the coefficient test statistics and have greater power compared with the Newey–West-based test statistics, and in long panels where T is substantially larger than N. Therefore, we regard the significant coefficients for and at the 5% level as stronger evidence for cointegration between the series than the and statistics.

¹⁸ For the slope coefficient, see Table A4 in the Appendix.

¹⁹ The maturity-specific adjustment period is calculated using the reciprocal of the coefficient. For example, for 8-month maturity (−8.141) helps us to arrive at the adjustment time (1/)= −0.123 days, which on further multiplication (with 24 for hours and 60 for minutes) gives 176.9 minutes or 2 hours 56.9 minutes.

²⁰ The negative adjusted R² is in line with previous studies on UIP which generally report a low or even negative R² (Chinn, 2007).

²¹ For example, see discussion on the exchange rate intervention, http://www.project-syndicate.org/commentary/the-logic-of-today-s-brewing-currency-wars-by-mohamed-a–el-erian