The Term Structure of Interbank Risk
Social Science Research Network
We use the term structure of spreads between rates on interest rate swaps indexed to LIBOR and overnight indexed swaps to infer a term structure of interbank risk. We develop a dynamic term structure model with default risk in the interbank market that, in conjunction with information from the credit default swap market, allows us to decompose the term structure of interbank risk into default and non-default components. On average, from August 2007 to January 2011, the fraction of total
... on of total interbank risk due to default risk increases with maturity. At the short end of the term structure, the non-default component is important in the first half of the sample and is correlated with various measures of market-wide liquidity. Further out the term structure, the default component is the dominant driver of interbank risk throughout the sample period. These results hold true in both the USD and EUR markets and are robust to different model parameterizations and measures of interbank default risk. The analysis has implications for monetary and regulatory policy as well as for pricing, hedging, and risk-management in the interest rate swap market. and firstname.lastname@example.org. Both authors gratefully acknowledge research support from NCCR FINRISK of the Swiss National Science Foundation. "The age of innocence -when banks lent to each other unsecured for three months or longer at only a small premium to expected policy rates -will not quickly, if ever, return". Mervin King, Bank of England Governor, 21 October 2008 2 is not due to default risks. To the extent that liquidity risk is correlated with default risk, for instance if liquidity hoarding is more prevalent when aggregate default risk is high, the residual component captures the component of liquidity risk that is orthogonal to default risk. Since a long-term IRS-OIS spread reflects expectations about future short-term LIBOR-OIS spreads, the term structure of IRS-OIS spreads reflects the term structures of the default and non-default components of the LIBOR-OIS spreads. To identify the default component, we use information from the credit default swap (CDS) market. At each point in time we construct a CDS spread term structure for the representative panel bank as a composite of the CDS spread term structures for the individual panel banks. Assuming that CDS spreads are pure measures of default risk of the underlying entities, the CDS spread term structure for the representative panel bank allows us to identify the process driving the default component of LIBOR-OIS spreads. More specifically, we develop a general affine model. Depending on the specification, two factors drive the OIS term structure, one or two factors drive the default component of LIBOR-OIS spreads (i.e., the risk of credit quality deterioration of the bank that represents the panel at a given point in time relative to the respective banks that represent the future panels), and one or two factors drive the nondefault component of LIBOR-OIS spreads. The model is highly tractable with analytical expressions for LIBOR, OIS, IRS, and CDS. We apply the model to study interbank risk from the onset of the financial crisis in August 2007 until January 2011. We utilize a panel data set consisting of term structures of OIS rates, rates on interest rate swaps indexed to 3M as well as 6M LIBOR, and CDS spreads -all with maturities up to 10Y. The model is estimated by maximum likelihood in conjunction with the Kalman filter. We first conduct a specification analysis, which shows that a specification with two factors driving the OIS term structure, two factors driving the default component of the LIBOR-OIS spread, and one factor driving the non-default component of the LIBOR-OIS spread has a good fit to the data. We then use this specification to decompose the term structure of interbank risk into default and non-default components. We find that, on average, the fraction of total interbank risk due to default risk increases with maturity. At the short end of the term structure, the non-default component is important in the first half of the sample, while further out the term structure, the default component is the dominant driver of interbank risk throughout the sample period. 2 To understand the determinants of the non-default component of interbank risk, we regress the residual factor on a number of illiquidity proxies for the fixed-income market including the spread between the 3M OIS rate and the 3M Treasury bill rate (Krishnamurthy (2010)), the yield spread 2 In principle, our analysis also allows us to determine how much of the variation in the term structure of interbank risk is due to variation in risk premia. In practice, however, risk premia are quite imprecisely estimated given the short sample, so we refrain from making too strong statements in this regard. It does seem, however, that participants in the interbank market require a premium for bearing exposure to both default and non-default (liquidity) risk. (2004) ), the "noise" measure recently introduced by Hu, Pan, and Wang (2010), and the notional amount of Treasury delivery fails reported by primary dealers used by Fleckenstein, Longstaff, and Lustig (2011) and others as a measure of disruptions in fixed income market liquidity. The factor is significantly related to all four illiquidity measures and jointly they explain a large fraction (about 70 percent) of the variation in the factor. This suggests that the non-default component is strongly related to market illiquidity. We conduct a variety of robustness test which show that the results hold true for alternative model parameterizations and measures of interbank default risk. By using CDS spreads to identify the default component of interbank risk, our approach is reminiscent of Longstaff, Mithal, and Neis (2005), Blanco, Brennan, and Marsh (2005), and Ang and Longstaff (2011), among others, who use CDS spreads as pure measures of default risk. However, a number of recent papers, including Buhler and Trapp (2010) and Bongaerts, de Jong, and Driessen (2011), have found that CDS spreads may be affected by liquidity effects. 3 Since we mostly use CDS contracts written on large financial institutions, which are among the most liquid contracts in the CDS market, and since we aggregate individual CDS spreads, which reduces the effect of idiosyncratic noise in the individual CDS spreads, we believe it is reasonable to use the composite CDS spreads to infer the default risks of the representative panel banks. Nevertheless, we also consider two alternative measures of default risks that correct for possible liquidity effects. First, we measure default risk by 90 percent of the composite CDS spreads, which, given the results in Buhler and Trapp (2010), seems to be a reasonable lower bound on the default component of CDS spreads. And, second, we measure default risk by composite CDS spreads constructed solely from the banks with the most liquid CDS contracts. None of these alternative measures substantially change the decomposition of the term structure interbank risk. between 10Y Refcorp bonds and 10Y Treasury notes (Longstaff Throughout, we also report results for the EUR market. Not only does this serve as an additional robustness test, but this market is interesting in its own right. First, by several measures, the market is even larger than the USD market. Second, the main shock to the interbank market in the second half of our sample emanated from the Eurozone with its sovereign debt crisis. And, third, the structure of the EUR market is such that the reference overnight rate in an OIS exactly matches the average cost of unsecured overnight funding of EURIBOR (the EUR equivalent of LIBOR) panel banks providing a robustness check of this assumption. Interestingly, results for the EUR market are very similar to those of the USD market. Our analysis has several practical applications. First, the framework could be a valuable tool for central banks and regulatory authorities, as it provides market expectations about future stress in interbank markets. In addition, the decomposition into default and non-default (liquidity) components can help guide appropriate policy responses (recapitalization of banks, termination/introduction of 3 While it is possible that CDS spreads are also affected by counterparty risk, Arora, Gandhi, and Longstaff (2009) find that this effect is minimal, which is consistent with the widespread use of collateralization and netting agreements.