the average of investor sell prices of bondi on dayt. As inFeldh¨utter, Hotchkiss, and Karakas (2016), we define the RTC of a bond as the median of available daily RTCs over the past two weeks.
We plot average roundtrip costs (RTC) in Figure3.2for the guaranteed bonds along with average RTCs for bonds issued by the same financial institutions and that were not guaranteed. We see that RTCs are much smaller for the guaranteed bonds compared to the unguaranteed bonds; the average RTC for unguaranteed bonds is 0.38% in the period January 2009-December 2012 while the average RTC for guaranteed bonds in the same period is 0.05%. Thus, guaranteed bonds are more liquid compared to unguaranteed bonds when measured as the cost of trading the bonds. Using the same methodology to compute transaction costs, Chakravarty and Sarkar (2003) estimate average RTC for U.S. Treasury bonds in the period 1995-1997 to be 0.08%.5 This suggests that the liquidity of the guaranteed bonds is comparable to the liquidity of U.S. Treasury bonds.
The guaranteed bonds are liquid and free of default risk, but nevertheless are corporate bonds and therefore do not enjoy the convenience yield of U.S. Treasuries.6 These bonds are therefore a near-ideal measure of the appropriate riskfree rate when calculating corporate bond spreads. Figure 3.3 shows a time series of the guaranteed bond yield, swap rate, and Treasury yield. Guaranteed bond yield is calculated as the average of the daily yields on all fixed coupon guaranteed bonds where daily yields are calculated as the median over daily values. Swap rates and Treasury yields are similarly calculated as the average of rates maturity matched to all guaranteed bonds in our sample. We see that the swap rate tracks the yield on guaranteed bonds quite closely while there is a gap between the yields on guaranteed bonds and Treasury bonds. The average yield on the guaranteed bonds during the period covered by the graph (November 2008 - May 2011) is 1.18%, the average swap rate is 1.22%, and the average Treasury yield is 0.85%. This shows that even during the 2008-2009 crisis swap rates are more appropriate than Treasury yields to use as riskfree rates when
5Chakravarty and Sarkar(2003) measure RTC asPitbuy−Pitsellwhile we measure RTC as P
buy it −Pitsell
Pitbuy
but since most bonds likely trade close to par value due to their relatively short maturity, the difference between the measures is small.
6Benefit of Treasury bonds compared to guaranteed corporate bonds are for example lower risk weights and broader collateral eligibility.
calculating corporate bond yield spreads.7
4 Price Discovery in the Corporate Bond and CDS Market
Blanco, Brennan, and Marsh (2005), Zhu(2006), and Norden and Weber (2009) test the lead-lag relationship between CDS and corporate bond spreads using the Granger causality test, Hasbrouck’s measure, and Gonzalo and Granger’s measure and find that CDS leads the corporate bond market. In Section2we show that results of these methods are biased when autocorrelation of one of the tested time series is non-zero.
Figure3.4shows histograms of autocorrelation in changes in CDS spreads and changes in corporate bond quotes of the 1244 firms in our sample with bonds quoted by Merrill Lynch, and a histogram of changes in end-of-day corporate bond transactions for the 763 firms with bonds trading persistent enough for us to utilize TRACE transaction.
The median autocorrelation for CDS spread, Merrill Lynch corporate bond spread, and TRACE corporate bond spreads is 0.01, -0.13, and -0.37 respectively. That is, autocorrelation of the CDS spreads is close to zero, but the autocorrelation of bond spreads is negative. If the autocorrelation pattern is the same in samples used in the above mentioned papers, it is likely that their results are biased towards favoring price discovery in the CDS market.
We test the lead-lag relationship between CDS and corporate bond spreads using a Granger causality test and using the unbiased model presented in Section 2.3 in which results are unaffected by negative autocorrelation in corporate bond spreads.
Our sample covers 11 years from 2002 to 2012. We split the sample into calendar years such that we can capture time variation in the relative price discovery in the corporate bond and CDS markets. Each year we use the yield spreads of the most recently issued corporate bond whichRonen and Zhou (2013) show are the most informative.
The unbiased model builds on estimating Equation (3.11) and (3.12), and the Granger causality test builds on estimating Equation (3.1) and (3.2). To make our analysis simple and transparent we choose to estimate the models with five lags (p= 5) for all firms.8
7
Table3.2 shows, for each calendar year and for both tests, the percentage of firms where CDS price leads, the percentage of firms where bond price leads and the differ-ence between the two. In the first four years of the sample the unbiased model finds that the CDS market price leads for 20% of the firms on average and that the corpo-rate bond market price leads for 24% of the firms on average. Hence the corpocorpo-rate bond market is 4% more often price leading than the CDS market. This result sug-gest a small tendency towards the corporate bonds marked being more price leading than CDS market in the first part of the sample. This contradicts findings of Blanco, Brennan, and Marsh (2005), Zhu (2006), and Norden and Weber (2009) who study a sample covering the same period. Interestingly, their findings are in line with the result of the Granger causality test. The Granger causality test finds that CDS price leads for 27% of the firms on average and that bond price leads for only 20% of firms on average. That is, the Granger causality test both overestimates the contribution to price discovery from the CDS market and underestimates the contribution to price discovery from the bond market. This finding illustrates that it is not a sample selec-tion that give us a different result from those in earlier studies, it is the methodology used for testing price discovery.
After 2005 the relative price discovery between the CDS and the corporate bond market changes such that the CDS market becomes more price leading relative to the corporate bond market. According to the unbiased test, the difference between price discovery in the CDS market and price discovery in the bond market is 8% on average.
The difference is 17% on average in the Granger causality test. Though, we conclude in both analyses that most price discovery arises in the CDS market, we overestimate the contribution from the CDS market more than 100% if we rely on the Granger causality test.
Next, we investigate if price discovery differs between the financial sector and non-financial sectors and if price discovery differs between investment grade and speculative grade rated firms. Panel A of Table3.3show that CDS is relatively more price leading for financials than for non-financials in all sub periods, in line with the fact that the CDS market is far more developed for financials than other corporates. Panel B of Table 3.3 shows that CDS is relative more price leading for speculative grade rated firms than for investment grade rated firms.
4.1 Improving Price Discovery in the Corporate Bond Market with