• Ingen resultater fundet

7. Empirical Results

7.4. Price Discovery

61

The lowest half-life is estimated for AAA- and AA-rated companies, which show a half-life of only 3.3 days. Surprisingly, A-rated show a relatively high half-life of 21.9 days, while a deviation in the basis spread of BBB-rated seems to halve within 6.0 days. Lastly, financial companies exhibit a longer half-life (16.0 days) than non-financial companies (5.9 days). This conforms to the larger observed average basis spread for financial companied compared to non-financial companies.

These results seem to confirm the expected arbitrage relationship, but they should be interpreted with due care. As explained in the previous section, microstructural noise could cause the results to be biased. Furthermore, as stated, the average explained variance of credit spreads is only 4% while the basis spreads are in fact not able to explain any of the variance in CDS spreads. This gives rise to the existence of other more important factors that influence credit spreads. Finally, the OLS regressions in this case share the disadvantage that they only incorporate one past period in the estimations. The effect of previous periods is omitted, which can severely disturb estimation results.

The results in this section of the paper should therefore be treated rather as supporting evidence further confirming the main evidence presented in the previous and next sections of the paper.

62

section 4.1.3. None of the price measures is universally superior to the other. Thus this paper will present both measures.82

To estimate both measures of price discovery it is necessary to first estimate the following VECM:

( )

t

p

j

j t CS j p

j

j t CDS j t

CS t

CDS t

CDS p p p p

p 1,

1

, 1 1

, 1

1 , 1 0 1 , 1

, =λ −α −α + β ∆ + δ ∆ +ε

∑ ∑

=

=

and

( )

p t

j

j t CS j p

j

j t CDS j t

CS t

CDS t

CS p p p p

p 2,

1

, 2 1

, 2

1 , 1 0 1 , 2

, =λ −α −α + β ∆ + δ ∆ +ε

∑ ∑

=

=

where ε1,t and ε2,t represent i.i.d. shocks. If the cash bond market contributes to the price discovery process, then λ1 is ought to be negative and statistically significant such that the CDS market adjusts to new information from the cash bond market. In contrast, if λ2 turns out to be positive and statistically significant, the CDS market contributes to the price discovery process. It is also possible, that both coefficients are statistically significant, in which case both markets contribute to price discovery. The cointegration relationship between the markets implies that at least one of the markets has to adjust by the Granger representation theorem. That market reacts to publicly available information and is thus inefficient.83

The Hasbrouck and Gonzalo-Granger measures reveal how much the CDS market contributes to price discovery. Table VI presents the results of these estimations of the price discovery measures.

The results confirm previous research, which has found evidence in favour of CDS market leadership. At the 10% level λ2 is significant in 16 out of 20 cases, which indicates that CDS markets are relevant for price discovery. The relevance of cash bond markets for price discovery seems limited when considering the entire sample. Only 11 out of the 20 companies show significant λ1 coefficients. British Telecom, Deutsche Telekom, Enel and Telefonica provide exceptions to this case, where bond markets seem to be the main source of price discovery.

82 see Bai et al. (2012), pp. 1-61; Baillie et al. (2002), pp. 309-321; Blanco et al. (2005), pp. 2255-2281, Booth et al.

(1999), pp. 619-643; Chu et al. (1999), pp. 21-34; Harris et al. (1995), pp. 563-57; Harris et al. (2002), pp. 277-308;

Harris et al. (2002b), pp. 341-348; Hasbrouck (2002), pp. 329-339; Jong (2002), pp. 323-327; Lehmann (2002), pp.

259-276; Ronen, Yaari (2002), pp. 349-390.

83 Engle, Granger, (1987), pp. 251-276.

63

The results present new interesting insights, when considering different groups of companies.

Interestingly, while the Hasbrouck and Gonzalo-Granger measure provide conflicting signals for a few individual cases, they are remarkably consistent when considering group averages. First of all, the importance of CDS markets seems to decrease for lower ratings. While all AA, A and A-rated companies show significant λ2 coefficients, only 50% of BBB-rated companies do provide significant λ2 coefficients. Accordingly, the share of significant λ1 coefficients increases with lower ratings, i.e. 25% of AAA- and AA-rated, 38% of A-rated and 88% of triple-B-rated companies show significant λ1 values. This is also confirmed, when considering the Hasbrouck and Gonzalo-Granger measures of the group. AAA- and AA-rated companies show a Hasbrouck measure of 0.93 while this value falls to 0.74 for A-rated companies and finally to 0.50 for BBB-rated companies. When considering the Gonzalo-Granger measure, the respective values are 0.90, 0.69 and 0.48. This provides evidence for the hypothesis, that investors rely more on bond prices than on CDS with lower ratings.

Furthermore, the CDS market seems to be more relevant for U.S. companies than for EU companies. All λ2 coefficients for U.S. companies are significant, while this is only the case for 64% of the EU companies. Accordingly, the cash bond market’s importance for price discovery is limited in the U.S. case, a fact that is reflected in the significance of only 33% of λ1 coefficients for U.S. companies compared to 73% for EU companies. Both the Hasbrouck as well as the Gonzalo-Granger measure confirm this evidence, with U.S. companies showing values of 0.84 and 0.91 respectively, compared to a value of 0.49 for both measures in the EU case.

Finally, the empirical evidence seems to suggest that investors rely slightly more on CDS markets for price discovery of financial companies than of non-financial companies. According to the share of significant λ2 coefficients, the CDS market is important for price discovery for all financial companies while it is only for 43% of non-financial companies. With regards to the bond markets, λ1 is significant for 62% of non-financial companies and for only 43% of financial companies.

Again, this is confirmed by the companies’ Hasbrouck and Gonzalo-Granger measures. Financial companies exhibit a Hasbrouck measure of 0.72 and a Gonzalo-Granger measure of 0.71, while non-financial show values of 0.61 and 0.66 respectively.

12 out of 32 companies reject cointegration when using daily observations. Rejection is possibly due to a CTD option, binding short sale constraints or too much microstructural noise because of

64

daily sampling. In this case the VECM representation is not valid. For these companies, the simpler concept of Granger causality using a vector autoregressive model is tested as explained in section 4.2. Table VII presents the results of these estimations. 5 out of twelve companies support the hypothesis of CDS spreads Granger-causing credit spreads, while 2 of these cases indicate bi-directional causality. Further 2 cases seem to indicate that only credit spreads Granger-cause CDS spreads. Finally, 5 cases indicate no existing Granger-relationship between the two markets.

Table VII: Price Discovery Measures (daily)84

84 The interpretation of GG values greater than unity or smaller than zero is not clear and unambiguous. Thus they were respectively treated as one and zero when computing averages. Significance is measured at the 10% level. λ1

significant at least at the 10% level.

λ1 Z-stat λ2 Z-stat Lower Upper Mid GG

American Express -0.03 -2.2 0.21 3.7 0.62 0.78 0.70 0.87

Bank of America -0.02 -1.0 0.11 7.2 0.75 0.99 0.87 0.84

Comcast -0.02 -2.4 0.14 2.3 0.46 0.51 0.49 0.88

GE -0.02 -2.1 0.05 3.1 0.54 0.74 0.64 0.71

Johnson & Johnson 0.00 0.5 0.39 3.9 0.98 0.99 0.99 1.01

Kraft Foods 0.00 -0.4 0.38 4.3 0.96 0.99 0.98 0.99

Morgan Stanley -0.01 -0.6 0.15 5.8 0.82 0.99 0.90 0.93

Pfizer 0.00 -0.5 0.21 5.5 0.98 0.99 0.99 0.99

Wal-Mart 0.00 -0.6 0.67 5.6 0.97 0.99 0.98 1.00

Atlantic Richfield -0.02 -1.3 0.09 5.3 0.93 0.95 0.94 0.83

Barclays -0.03 -2.5 0.08 4.5 0.55 0.82 0.69 0.69

British Telecom -0.04 -2.5 0.02 1.5 0.14 0.62 0.38 0.35

Credit Agricole -0.01 -1.1 0.02 2.7 0.33 0.95 0.64 0.56

Deutsche Telekom -0.06 -3.8 -0.01 -0.4 0.01 0.05 0.03 -0.22

Enel -0.06 -3.5 0.02 1.8 0.11 0.58 0.35 0.19

GlaxoSmithKline -0.04 -3.8 -0.08 -2.3 0.12 0.33 0.23 2.06

Nokia -0.05 -2.7 0.03 2.2 0.18 0.75 0.46 0.34

Santander -0.07 -3.7 0.02 2.2 0.13 0.61 0.37 0.23

Standard Chartered -0.01 -0.6 0.04 3.9 0.76 0.98 0.87 0.86

Telefonica -0.04 -2.1 0.03 1.9 0.16 0.81 0.48 0.37

Hasbrouck

Means Lower Upper Mid GG λ1 sign. % λ2 sign. %

Cleansed (20 companies) 0.53 0.77 0.65 0.68 11 55% 16 80%

AAA-AA (4 companies) 0.87 0.93 0.90 0.93 1 25% 4 100%

A (8 companies) 0.55 0.83 0.69 0.74 3 38% 8 100%

BBB (8 companies) 0.33 0.64 0.48 0.50 7 88% 4 50%

US (9 companies) 0.79 0.89 0.84 0.91 3 33% 9 100%

EU (11 companies) 0.31 0.68 0.49 0.49 8 73% 7 64%

Financial (7 companies) 0.57 0.87 0.72 0.71 3 43% 7 100%

Non-Financials (13 companies) 0.50 0.72 0.61 0.66 8 62% 9 69%

Hasbrouck

65

To sum it all up, the evidence in this case is slightly less in favour of CDS markets than compared to previous research. Again one likely reason for this is the larger microstructural noise in daily data. Evidence of this is reflected by the fact that all five companies that reject Granger-relationships between the two markets in the case of daily data do confirm cointegration in the case of weekly data.

Table VIII: Granger-Causality (daily)