This is the published version - Deakin Universitydro.deakin.edu.au/eserv/DU:30061932/wang-ratingdowngrade-2013.pdf · JEL Classification: G14, G15. 284 Review of Futures Markets Institute

Thisisthepublishedversion: Chng,MichaelT.andWang,Peipei2014,RatingdowngradeandthepriceimpactofCDSspreadonstockreturn,Reviewoffuturesmarkets,vol.21,no.3,pp.283‐323.

Available from Deakin Research Online: http://hdl.handle.net/10536/DRO/DU:30061932EveryreasonableefforthasbeenmadetoensurethatpermissionhasbeenobtainedforitemsincludedinDeakinResearchOnline.Ifyoubelievethatyourrightshavebeeninfringedbythisrepository,[email protected]:2014,KentStateUniversity

We investigate the time-varying informativeness of credit default swap (CDS)trading on stock returns for 302 US firms from July 2004 to August 2010.Using the Acharya and Johnson (2007) measure, we find that CDS tradingbecomes informative for an increasing number of firms as we approach theglobal financial crisis (GFC). Firm numbers gradually decline post-GFC, butremain high compared to the pre-GFC period. Furthermore, CDS tradingimposes the largest conditional price impact on firms that are recentlydowngraded, regardless of rating levels. Interestingly, this holds during andafter the GFC, but not before. We offer two implications. First, despite post-GFC outcry against the CDS market, our results suggest it exhibits enhancedprice discovery during the GFC. Second, our findings support criticism that,in the lead-up to the GFC, rating agencies are slow in downgrading firms.However, if downgrade decisions made during and after the GFC induceinformed trading in the CDS market, this necessarily implies that during themidst of the GFC, rating agencies have got their act together.

“CDS in theory should mitigate risk in the financial system by allowinginvestors to hedge risks. In practice, they engendered uncertaintyand fear.” — Mark Zandi, 2009 submission to U.S. House Committeeon Financial Services

Credit derivatives, in particular credit default swaps (CDS), are blamed forallowing commercial banks and insurance firms like AIG, Bear Sterns, andLehmann Brothers to take up massive positions that led to massive losses,

which triggered a credit-crunch-induced global financial crisis (GFC). At the 2009

Michael T. Chng and Peipei Wang*

RATING DOWNGRADE ANDTHE PRICE IMPACT OF CDS SPREAD

ON STOCK RETURN

*Michael T. Chng (the corresponding author) is in the School of Accounting, Economics and Finance,Faculty of Business and Law, Deakin University. Email: [email protected] Wang is in the School of Accounting, Economics and Finance, Faculty of Business and Law,Deakin University. Email: [email protected].

Acknowledgements: We thank two anonymous referees and the Editor Mark Holder for commentsthat have vastly improved the paper. A previous draft was presented at the 2012 InternationalConference on Futures and Derivative Markets in Beijing. We acknowledge funding support fromthe Australian Center for Financial Studies.

Keywords: CDS spreads, credit rating, stock returnsJEL Classification: G14, G15

Review of Futures Markets284

Institute of International Finance Meeting, George Soros, chairman of Soros FundManagement, echoed numerous critiques to call for a ban on CDS,1 citing, “they’retruly toxic.” Rating agencies are also blamed for assigning investment-grade ratingsto sub-prime mortgage-backed securities, and for their inertia in downgrading firmsin the lead-up to the GFC.

Despite widespread criticism from regulators and practitioners, we are notaware of any formal study that attempts to evaluate the informativeness of CDStrading and rating changes during the course of the GFC. A fundamental economicrole of the CDS market and rating agencies is to aggregate relevant credit-relatedinformation and provide timely and informed prices and ratings. However, does theprice discovery role cease to function properly under extreme market conditions?What happens to the informativeness of the CDS market as we gradually approachand depart from the height the the GFC? Does rating level or rating change affectthe price impact of CDS spread on its reference entity’s stock return? Is the effectsymmetric for downgrades and upgrades? The motivation of our paper is to shedsome light to these questions.

Studies on information flow between bond-stock, CDS-stock and/or CDS-bond, for example, Hotchkiss and Ronen (2002), Acharya and Johnson (2007) andFung et al. (2008), find that incremental information revealed by the CDS and/orbond market occurs mainly in low-rated firms.2 Das et al. (2012) conclude that theadvent of CDS trading has a detrimental effect on the corporate bond market interms of reduced efficiency and no evident improvements in pricing error or liquidity.Their finding is not surprising if the CDS market is gradually taking over credit riskprice discovery from the corporate bond market. Norden and Weber (2004) findthat CDS spreads anticipate rating downgrades by Standard and Poor’s (S&P) andMoody’s, but reviews for downgrade impose a significant price impact on CDSspreads. Using event-study methodology, Hull et al. (2004) find that the CDS marketanticipates both downgrades as well as reviews for downgrade, but the results arenot significant for upgrade announcements. Flannery et al. (2010) show that CDSspreads respond to new information much quicker than credit ratings. However,their analysis focuses only on 15 investment-grade financial institutions whose creditratings remain quite stable during the 2006–2009 sample period. Chava et al. (2012)test the informativeness of rating downgrades on stock and bond returns. They findthat the price impact of downgrades on stock returns is reduced after CDS tradingon the firm’s debt commences. Furthermore, firms without CDS trading experiencelarger stock and bond market reaction than a comparable sample of firms withCDS trading.

1. A CDS is a derivative contract that is similar in nature to an insurance policy against default. It iswritten on a reference entity, normally a bond that can be issued by a government, state, bank andfirm. The payoff from a CDS contract is triggered by a default event. As such, the CDS marketallows an issuer’s credit risk to be tradable at observable prices.2. The intuition being that informed investors will set CDS or bond positions with huge expectedpayoffs. Firms with weaker credit profile have a larger probability of entering into financial distress.

Price Impact of CDS Spread 285

To gauge the informativeness of CDS trading on stock returns, we utilize theAcharya and Johnson (2007) measure on a sample of 302 US firms from July 2004to August 2010. In the lead-up to and during the GFC, when financial markets weregripped by a systemic credit crunch, we naturally expect to observe higher andmore volatile CDS spreads. A time-series plot of the cross-sectional average CDSspread for our firm sample in Figure 1 (panel A) confirms that this is indeed thecase. Accordingly, one would naturally expect the informativeness of CDS tradingon stock returns to be similarly impaired during the GFC.

Our main results reveal a starkly dissimilar picture on the informativeness ofCDS trading during the GFC. We find that, with the onset of the crisis, CDS tradingsignificantly affects stock returns for an increasing number of firms. As we moveaway from the height of the GFC, the number of firms decline, but remain highcompared to the pre-GFC period. Despite post-GFC outcry from industry groupsand regulatory bodies, we find that CDS trading actually becomes more informativeduring the GFC. Further analysis against an array of credit condition variablesreveals that CDS trading has the largest conditional price impact on firms that havebeen recently downgraded, regardless of their ratings level. Interestingly, this resultholds for estimation periods that correspond to during and after the GFC, but notbefore. While we also document a similar finding for firms that have been recentlyupgraded, the magnitude of the conditional price impact is smaller relative todowngrades.

Our findings show that, regardless of whether CDS trading cause the GFC, itsprice discovery function as a derivative market was not impaired during the creditcrisis. On the contrary, CDS trading becomes more informative for an increasingnumber of firms during the height of the GFC. Without CDS trading to processcredit-related information during the credit-crunch, we postulate that stock returnsmay have been more volatile. Our results support criticisms that, prior to the GFC,rating agencies are indeed slow in downgrading firms. Information flow from theCDS to stock market is unaffected by rating changes. However, if downgradeannouncements made during and after the GFC generate large and statisticallysignificant conditional price impact on stock returns, this necessarily implies that,during the midst of the GFC, rating agencies have gotten their act together.

Our paper proceeds as follows. We discuss institutional details and sampling inSection I. The empirical methodology and results are discussed in Section II. SectionIII concludes.

I. BACKGROUND AND SAMPLING

During the past decade, we have witnessed a remarkable growth in the globalCDS market. According to the International Swaps and Derivatives Association(ISDA), the total outstanding notional amount of CDS contracts has increased 68times from USD 0.92 trillion in 2001 to USD 62.17 trillion in 2007. While the marketsize has reduced to USD 54.6 trillion by June 2008, it remains larger than the USD


Figure 1. Time-Series Plots of the Cross-Sectional Average CDS Trading Variables.

We provide the time-series plots of the cross-sectional average CDS trading variables: CDSspreads (bps) in Panel A; CDS returns in Panel B; absolute bid-ask spread (bps) in Panel C;relative bid-ask spread in Panel D. The cross-sectional values are based on simple averagingof firm level observations within the relevant credit category.


48.36 trillion global bond market during the same period.3 Most of that remarkablegrowth is driven by the US CDS market. Over the years, CDS spreads areincreasingly being regarded by the financial community as a benchmark indicatorof the underlying reference entity's credit risk.4 Hull et al. (2004) discuss theadvantages of CDS spreads over bond yield spreads as a measure of credit risk.

A. Review of Prior Studies

Although it was conceived in the early 1990s, the CDS market did not start togain prominence until 2002. For that reason, academic research on the CDS marketis still in its early stages. Ericsson et al. (2009) and Greatrex (2009) investigate thedeterminants of CDS spreads. Blanco et al. (2005) examine credit risk dynamicsbetween the CDS and bond markets for 16 US investment-grade firms over 18months. They document an evident cross-market credit risk pricing relation, whichconfirms the theoretical relation derived by Duffie (1999). Specifically, Blanco etal. (2005) find that the CDS market is more efficient than the bond market inreflecting credit risk information.

Fewer studies have examined trading interactions between CDS and equitymarkets in the context of credit ratings. An economic link exists between the twomarkets since equity holders own residual claims on a firm upon default. Hence,credit risk matters to equity holders as well, and stock returns should reflect credit-related information relevant to the firm. Theoretically, the seminal work of Merton(1974) establishes a link between equity value and credit risk under structural creditrisk pricing.5 Investment banks and hedge funds already have in place tradingstrategies to take advantage of lead-lag return dynamics between the CDS andequity markets. Yu (2006) finds that at a portfolio-level capital structure arbitrageyields Sharpe ratios that are comparable to other hedge fund strategies.6

Longstaff et al. (2003) examine weekly lead-lag effects amongst the CDS,bond and stock markets. They find evidence of information flow from both CDSand stock markets onto the bond market. However, there is no clear evidence onwhether the CDS market is leading the stock market, or vice versa. Hull et al.

3. Source: BIS Quarterly Review, Dec. 2008.4. See Longstaff et al. (2005), Blanco et al. (2005), Yu (2006), and Acharya and Johnson (2007) fora discussion on CDS spreads as a measure of credit risk.5. Equity-holders are able to retire debt at maturity and claim firm ownership. This is akin to holdinga call option against debt-holders on firm assets, with the face-value of debt as the strike price.Accordingly, the probability of non-exercise is analogous to the probability of default. Any informationthat affects a firm’s credit worthiness will affect the value of equity-holders’ embedded call options,hence its stock price.6. Capital structure arbitrage (CSA) is a convergence strategy that captures mispricing (divergence)between the market-observed CDS spread and model-implied CDS spread extracted from stockprices using, for example, RiskMetrics CreditGrades model. While the CDS position can be quitelarge, CSA does not normally evoke frequent trading. For example, Yu (2006) reports convergencerates of less than 5% for a 180-days maximum holding period. In other words, more than 95% ofCDS positions in the CSA strategies considered in Yu (2006) are held for 180 days. In a current CSAproject that we are working on, we can report that the average CDS holding period across an array ofCSA strategies range between 35 to 43 days.


(2004) find that the CDS market anticipates rating downgrade, and to a certainextent, review for downgrade as well. However, they did not examine the equitymarket. Norden and Weber (2004) find that CDS and equity markets both anticipaterating downgrades, but they did not formally analyze information flow between thetwo markets in the context of rating changes.

Fung et al. (2008) examine information transmission between stock marketand CDS market indices between 2001 and 2007. For the speculative-grade CDSindex, they find two-way lead-lag relations. But for the investment-grade CDSindex, they find that the stock index leads the CDS index. This suggests that theinformation transmission mechanism between the two markets is conditional on thecredit quality of the firm. Using implied CDS extracted from structured credit riskmodel, Forte and Pena (2009) also find that the stock market leads the CDS marketfor the majority of their sample of 17 non-financial firms. In stark contrast, Acharyaand Johnson (2007) find that negative private information is first revealed in theCDS market through insider trading before it is transmitted over to the stock market.Their study is based on 79 U.S. firms between January 2001 and October 2004.However, Acharya and Johnson (2007) analyze rating level rather than rating changes.

Norden and Weber (2009) find that stock return rit exerts a heavier influenceon changes in CDS spread ∆CDSit for firms with lower credit ratings. Hilscher etal. (2012) also document limited information content in CDS trading to affectsubsequent stock return. However, neither study necessarily contradicts our mainfinding. First, the tri-variate VAR estimation in Norden and Weber (2009), and theregression in Hilscher et al. (2012) both impose a linear specification between ritand ∆CDSit. Acharya and Johnson (2007) and Chng et al. (2013) both document anevident nonlinear relation between rit and ∆CDSit, in accordance with the structuralcredit risk pricing framework of Merton (1974). Second, Norden and Weber (2009)examine mainly European firms, but our paper is on US firms. Institutional differencesbetween the US and European CDS and stock markets7 are likely to drive thedissimilar results. Moreover, the sample period is from 2000 to 2002. It is not surprisingthat CDS trading was uninformative in its earlier years. Third, although the Hilscheret al. (2012) firm sample is more comparable, the sample period is from 2001 to2007. Hence our finding of increasing informativeness in CDS trading as theestimation window approaches the GFC in September 2008 does not necessarilycontradict the Hilscher et al. (2012) finding.

Our paper contributes to the existing literature with a comprehensive analysisthat covers about 300 US firms over an extended sample period that covers theGFC. We apply the Acharya and Johnson (2007) methodology to examine howCDS return innovations affect stock returns conditional on rating changes. Lastly,our moving-window estimation allows us to track the time-varying information flowfrom the CDS to stock market for a large firm sample as we progress toward andaway from the height of the GFC.

7. For example, liquidity, contract clauses, trading regulations, and others.


B. Testable Propositions

We discuss the development of two testable propositions to address the researchquestions that we pose in this paper.

Proposition 1A: The informativeness of CDS trading on stock returnsis time-varying. The credit-crunch induced GFC is likely to impose asevere structural break on the data. Hence the price impact of CDStrading on stock returns is likely to be time-varying, especially when theestimation window moves across the GFC.

Proposition 1B: The informativeness of CDS trading on stock returnsis stronger (weaker) as we move toward (away from) the height ofthe GFC. CDS spreads are higher and more volatile surrounding theheight of the GFC. This has a two-fold effect in terms of enhancing thedegree of informed trading in the CDS market. First, uninformed investorsare likely to shun the market, especially given the bad publicity that theCDS market has received during the crisis. Second, the substantially higherand more volatile CDS spreads from mid-2008 onwards present informedinvestors with trading opportunities that did not exist in the pre-GFC period.

Proposition 2A: The price impact of CDS trading on stock returnsconditional on rating changes is insignificant before the GFC, butbecomes significant during and after the GFC. We expect the priceimpact of CDS trading on stock returns to be unaffected by rating changeannouncements in the lead-up to the GFC, consistent with prior studiesand criticisms made of rating agencies. But given the post-GFC publicoutcry against rating agencies for not doing their job, it is possible thatrating agencies will get their act together and provide more timely andinformed ratings.

Proposition 2B: The conditional price impact of CDS trading on stockreturns is likely to be stronger for downgrades compared to upgrades.Prior studies have found that the CDS market is efficient at processingnegative information. Accordingly, the magnitude of the conditional priceimpact is likely to be stronger for downgrades compared to upgrades.

C. Samples and Variables

We define the CDS closing price CDSit as the daily mid-quotes of CDScontracts from July 1, 2004, to August 31, 2010. Following prior studies, we focuson five-year CDS contracts with USD 10 million notional amount written on seniorunsecured debt issued by US firms. Compared to other economies, the US CDSmarket is clearly a larger, mature and more liquid market. CDS bid and ask quotes


are obtained from DataStream. The final sample includes 302 U.S. firms8 for whichdata is available for both CDS and equity markets over the entire sample period.Firm i daily stock returns rit for at time t and matching balance sheet data aretaken from CRSP and Compustat. Monthly S&P ratings on firms’ long-term debtare also downloaded from Compustat.

We download Compustat monthly updated S&P credit ratings for each firm’slong-term debt from June 2004 to August 2010. Table 1 shows the distribution offirms across credit ratings, which we report in six-monthly intervals over the sampleperiod. Table 1 presents the average number of firms in each credit category. Themonthly number of firms in a given rating is averaged over the six-months horizon.9Due to space constraint we report firm numbers for ratings that are bundled intodifferent credit categories.10

The majority of firms fall within AA+ and BB-, with a median rating of BBB.The relevant columns also show that most of the shifts in ratings over time occuramong the three credit categories covering A+ to BB-. In the first sub-sample fromJune 30, 2004, to December 31, 2004, there are 6.7 firms on average with a AAArating. Not surprisingly, that number dropped to 2.13 firms in the sub-sample endingAugust 31, 2010. The next two credit categories also suffered a sharp decline. Theaverage number of AA+ to AA- (A+ to A-) firms fell from 16.67 to 12.33 (83 to76.5) going from the second half of 2008 to the first half of 2009. In the BBB+ toBBB- category, the average number of firms increased from 135.5 to 139.67. Thisis not surprising given a substantial number of firms have been downgraded froman A to a B rating. From BB+ downwards, there does not appear to be a substantialshift by firms across rating categories over time.

In Figure 2, we plot the cross-sectional average of CDSit, ∆CDSit, ASit, RSitfor each credit category over the sample period. Table 1 shows that there are notmany firms rated as CCC+ and below. We observe, as expected, that the lower the

8. The overall firm sample covers 7 industry sectors. Consumer Non-cyclical, Consumer Cyclicaland Industrial contains 44, 35 and 33 firms, respectively. There are 27 Energy firms and 22 firms inBasic Materials. Lastly, there are 7 firms in Technology and 6 firms in Communication.9. For example, consider the 5 AAA-rated firms for the interval starting January 31, 2006. This canbe associated with having 4, 5 and 6 AAA-firms for 2 months each during that particular interval.10. Prime is AAA; High Grade covers AA+, AA and AA-; Upper Medium Grade includes A+, A andA-; BBB+, BBB and BBB- are in Lower Medium Grade; Non-Investment Grade Speculative includesBB+, BB and BB-; Highly Speculative covers B+, B, and B-; Substantial Risk is CCC+; ExtremelySpeculative is CCC. Lastly, the Junk category includes CCC- and below.

We analyze four CDS trading variables: CDS spread (CDSit ) which is the mid-point between the closing bid and ask quotes in basis points (bps), CDS

return ΔCDS it Ln CDSit

CDSit−1

, absolute bid-ask spread ASit , and relative bid-ask

spread RS it

ASit

CDSit . In Figure 2, we plot the full-sample distribution for each of the four CDS variables. The distribution for CDSit is positively skewed with 80% of observations falling between 14.5bp and 200bp. About 90% of ∆CDSit falls between -6.78% to 7.19%. Both ASit and RSit are also positively skewed.


Figure 2. Full-Sample Distributions of CDS Spread, CDS Return, Absolute Bid-AskSpread and Relative Bid-Ask Spread.


Tab

le 1

. Fi

rm D

istri

butio

ns a

cros

s S&

P C

redi

t Cat

egor

ies

over

Tim

e.

Prim

e H

igh

Gra

de

Upp

er

Med

ium

G

rade

Low

er

Med

ium

G

rade

Sp

ecul

ativ

e G

rade

H

ighl

y Sp

ecul

ativ

e Su

bsta

ntia

l ris

k Ex

trem

ely

Spec

ulat

ive

Junk

C

redi

t ca

tego

ry:

AA

A

AA

+ ~A

A-

A+~

A-

BBB

+~BB

B-

BB+

~BB

- B

+~B-

C

CC+

CC

C C

C-~

C

1

2~4

5~7

8~10

11

~13

14~1

6 17

18

19

ab

ove

Jul-D

ec 2

004

6.71

17

.00

95.7

1 12

8.29

37

.29

17

0 0

0

Jan-

Jun

2005

5.

33

18.5

0 92

.33

133.

50

34.6

7 17

.50

0.17

0

0

Jul-D

ec 2

005

5 18

.50

91.1

7 13

7.00

32

.50

17.8

3 0

0 0

Jan-

Jun

2006

5

18.1

7 89

.33

140.

83

31.5

0 17

.17

0 0

0

Jul-D

ec 2

006

5 19

.33

86

139.

17

33.5

0 18

.67

0 0.

33

0

Jan-

Jun

2007

5

19.8

3 87

13

6.67

34

.50

19

0 0

0

Jul-D

ec 2

007

5 20

86

13

6.33

35

.67

19

0 0

0

Jan-

Jun

2008

4

20.5

0 82

.83

137.

17

38.8

3 18

.67

0 0

0

Jul-D

ec 2

008

4 16

.67

83

135.

50

40.8

3 21

.50

0.5

0 0

Jan-

Jun

2009

4

12.3

3 76

.50

139.

67

40.8

3 20

.83

5.17

1.

67

1

Jul-D

ec 2

009

3.5

12.5

0 75

13

8.17

40

23

.67

4 3

2.17

Jan-

Aug

201

0 2.

13

13.8

8 75

.63

138

39.8

8 27

.13

3 1.

25

1.13

The

tabl

e re

ports

the

aver

age

num

ber o

f fir

ms

in e

ach

cred

it ca

tego

ry fo

r a g

iven

six

-mon

th in

terv

al. E

ach

cred

it-ra

ting

is a

ssig

ned

a ra

ting

scor

e (A

AA

=1, A

A+=

2,…

,C=2

0). B

ased

on

mon

thly

S&

P ra

tings

, we

can

wor

k ou

t the

tota

l num

ber o

f fir

ms

in a

giv

en c

redi

t cat

egor

y fo

r ea

ch m

onth

. The

n w

e co

mpu

te th

e m

onth

ly a

vera

ge n

umbe

r of f

irm

s in

a g

iven

cre

dit c

ateg

ory

for e

ach

six-m

onth

inte

rval

.


credit category, the higher the level of CDSit and ASit . Across all credit categories,the level of CDSit and ASit both increase as we approach the GFC. Although thereis an overall decline post-GFC, the levels of CDSit and ASit for all credit categoriesremain high compared to the pre-GFC period.

Our preliminary analysis reveals that the levels of ASit and RSit across creditcategories are the opposite of each other. The B+ to B- categories possess thehighest ASit, but the AAA category has the highest RSit. The RSit measurestandardizes fluctuations in bid-ask spreads using prices, which facilitatescomparisons across assets. However, it is not a suitable measure if fluctuation inRSit is dominated by price volatility in certain assets, but by ASit in other assets.Since the CDSit of lower rated firms are substantially more volatile than those withhigher ratings, we focus on ASit in our analysis. Interestingly, ∆CDSit is more volatilefor AAA firms compared to B+, B and B- firms in the lead-up to the GFC, but theybecome comparable post-GFC. This is a preliminary indication that fluctuations inCDSit over time are not necessarily associated with the level of credit-rating.

Table 2 contains the descriptive statistics of key trading variables. It is worthnoting that CDS trading variables are generally larger compared to stocks. This isbecause CDS prices (mid-quotes) CDSit are in basis points(bps).11 Nonetheless,the high and low values do appear to be quite extreme. For example, the medianCDSit is 62.61 bps, but it has a high of 29,299.21 bps for AMBAC Financial Groupon May 20, 2010. The median ASit and RSit are 12.77 bps and 12.11%, respectively,but they have maximum values of 4,217bps and 199.57%, respectively. In general,a large gap between the median and its minimum and/or maximum values normallyindicates the presence of outliers. Further examination of the sample reveals thatthese extreme values come from firms that are rated CCC+ and below, such asAMBAC Financial Group. Furthermore, Table 1 also shows that these firmsconstitute 5% or less of the firm sample. Accordingly, we drop these outlier firmsfrom our main analysis.

II. METHODOLOGY AND RESULTS

A. Diagnostic Check and Preliminary Analysis

Augmented Dickey-Fuller unit root tests confirm that pit and CDSit are I(1)processes for every firm, such that rit and ∆CDSit are stationary over time. Thisholds in both full- and sub-samples. In Figure 3a, we plot the stock return distributionfor the pooled firm sample. The graph suggests a leptokurtic distribution. Next, weplot the time-series of the cross-sectional average return between AAA firms andfirms in the B+ to B- category in Figure 3b. The two plots show that B-rated firmsdisplay more volatility clustering over time. This suggests that different estimationprocedures may be required when we analyze stock returns across different credit

11. For example, an increase in CDSit from 80bps to 90bps. The increase in only 10bps, but ∆CDSitis 12.5%. For another example, consider a CDS contract that has a best bid and ask quote of 95bpsand 105bps, respectively. This translates to a relative bid-ask spread of 10/100, or 10%.


Tabl

e 2.

Bas

ic F

eatu

res

of K

ey V

aria

bles

bas

ed o

n th

e F

ull S

ampl

e.

r it

CDS i

t ∆C

DS i

t AS

it

RSit

Pan

el A

: Des

crip

tive

stat

istic

s of

key

var

iabl

es

Mea

n 0.

0002

62

.61

0 5.

1 0.

0971

Std

Dev

0.

0298

41

9.89

32

0.06

9 36

.105

6 0.

0957

Med

ian

0.00

04

178.

9295

0.

0007

12

.766

8 0.

1211

Skew

ness

1.

458

11.4

287

1.71

5 32

.426

4.

0743

Kur

tosis

47

.058

7 26

1.08

39

823.

6974

20

04.7

63

39.3

219

Min

-0

.607

9 1

-4.9

454

0 0

Max

0.

9818

29

299.

21

4.96

28

4217

1.

9957

P

anel

B: C

orre

latio

n m

atri

x of

key

var

iabl

es

r it

1

CDS i

t 0.

0055

**

1

∆CD

S it

-0.1

091**

0.

0304

**

1

ASit

0.01

77*

* 0.

4250

**

0.01

28**

1

RSit

0.00

30*

* -0

.623

9**

-0.0

530**

-0

.042

2**

1


categories. For example, we may need to specify a GARCH process when testingthe impact of CDS return on stock returns for firms with lower credit ratings.

Table 1 shows that more than 70% of our firm sample are rated as investment-grade firms (BBB- and above). In general, these are good quality firms withsubstantial analyst coverage. Following Acharya and Johnson (2007) and Fung etal. (2008), we assume that rit incorporates publicly available information. Henceany informed trading activity in the CDS market will deliver a significant priceimpact on the rit of the underlying reference entity.

To gauge the inherent cross-market interactions between the stock and CDSmarkets, we examine the cross-serial correlation structure between rit and ∆CDSitin Figure 4. We plot the correlations between ∆CDSit and rit–k for k = –5, –4, ..., 0,..., +5 for all firms and firms grouped by credit categories. We also plot correlationstructures for the entire sample period as well as sub-samples. The variouscorrelation structures we plot are consistent with those reported in Acharya andJohnson (2007). There is a strong negative contemporaneous correlation between∆CDSit and rit. This is expected since a firm that experiences credit deteriorationwill exhibit both widening CDS spreads and negative stock returns.

Figures 4B and 4C show that correlation between the CDS and stock markets

Note: The left panel shows the stock return distribution for the pooled sample, while theright panel shows the time-series of cross-sectional average stock return for creditcategory AAA and B+~B-.

Figure 3.

2 3 - F e b - 2 0 0 4 0 5 - J u n - 2 0 0 6 1 5 - S e p - 2 0 0 8 2 8 - D e c - 2 0 1 0- 0 . 2

- 0 . 1 5

- 0 . 1

- 0 . 0 5

0

0 . 0 5

0 . 1

0 . 1 5S t o c k R e t u r n B y C r e d i t

A A AB + ~ B -


is evidently stronger in the second half sample period. Fung et al. (2008) alsodocument substantially stronger correlation from the later half of 2007 onwards.For our overall sample, the maximum contemporaneous correlation is –0.452. Themean and median correlation coefficients are –0.153 and –0.150, respectively. Butfor the period July 1, 2007, to August 31, 2010, the mean and median correlationcoefficients are –0.208 and –0.210, respectively.

More importantly, cross-serial correlations between ∆CDSit and future stockreturns has increased when we move from the pre-GFC to GFC sub-sample, andthis is observed across all rating categories. The preliminary evidence indicate thatfluctuations in CDS spreads have become informative during the course of theGFC, regardless of the rating level.

B. Analyzing Credit Risk Information Flow

1. Measuring the Conditional Price Impact of CDS Trading

Acharya and Johnson (2007) utilize a two-stage least square approach toanalyze the informativeness of CDS trading on stock returns. From the first-stageestimation in equation (1), we extract a time-series of CDS return innovation εit.By regressing ∆CDSit against lagged CDS and stock returns, as well as the currentstock return rit that reflects all publicly available information, we can regard εit asa proxy for private information that is traded through the CDS market. Equation (1)also includes a series of interacting variables between lagged stock returns andcurrent CDS spread level CDSit , which allows for the nonlinear impact of CDSfluctuations on stock returns. We use the Schwarz Information Criterion (SIC) todetermine the optimal lag specification (S) on a firm-by-firm basis.

0 10

0 1 2 1 21

[ ( )]

[ ]

Sit s

it i is it s is it s is itits

S

it i is it s is jt it s is it s is jt it s its

rCDS CDS rCDS

r a a r a D r b b D

α α β γ ε

ε ε ω

−− − −

=

− − − −=

∆ = + ∆ + + +

= + + + + +

∑

∑

(1)

(2)

For the second-stage regression in equation (2), we regress rit against the sum of lagged CDS return innovations 1

Ss it-sε=∑ to ascertain if informed CDS

trading has any influence on stock return. If ∑s1S b1is is jointly significant, this

implies that ∑s1S it−s imposes a significant unconditional permanent price

impact on rit. Conversely, if ∑s1S b1is is jointly insignificant, this implies that

CDS trading innovations do not exert a persistent impact on stock returns.

Equation (2) also includes the sum of lagged interacting variables ∑s1S Djtit−s

to further test if the informativeness of CDS trading is affected by a firm’s credit

condition, as indicated by the credit condition dummy Djt. If ∑s1S b2is is jointly


-5 -4 -3 -2 -1 0 1 2 3 4 5-0.2

-0.15

-0.1

-0.05

0

0.05

Lead/Lag

Cross Correlation Between CDS & Stock (All)

-5 -4 -3 -2 -1 0 1 2 3 4 5-0.2

-0.15

-0.1

-0.05

0

0.05

Le ad/Lag

Cross Corre lati on B etw een CDS & S tock (A A & Upper)

-5 -4 -3 -2 -1 0 1 2 3 4 5-0.2

-0.15

-0.1

-0.05

0

0.05

Lea d/La g

Cross Correlat ion Betw een CDS & S tock (A+~ A-)

-5 -4 -3 -2 -1 0 1 2 3 4 5-0.2

-0.15

-0.1

-0.05

0

0.05C ross Corre la tion Be twe en CD S & Sto ck ( BBB+ ~ BBB-)

L ea d /L ag

-5 - 4 -3 -2 - 1 0 1 2 3 4 5-0 .2

-0 .15

-0 .1

-0 .05

0

0 .05

Lea d/ Lag

C ross C orre lat ion Between CD S & S tock (BB+ ~BB-)

-5 -4 -3 -2 -1 0 1 2 3 4 5-0.2

-0.15

-0.1

-0.05

0

0.05

Lead/Lag

Cross Correlation Between CDS & Stock (B & Under)

Figure 4A.

Note: We plot the average correlations between CDS return and 5 lead/lag stock returnsbased on the full sample period. We present correlations based on the entire firm sampleas well as for individual credit categories.


-5 -4 -3 -2 -1 0 1 2 3 4 5-0 . 2

- 0. 1 5

-0 . 1

- 0. 0 5

0

0. 0 5C r os s C o rr er lato n B etw e en C D S & Stock ( All)

Lea d /Lag

-5 -4 -3 -2 -1 0 1 2 3 4 5-0. 2

-0. 15

-0. 1

-0. 05

0

0. 05C ross C orrelat ion Bet wee n C DS & Sto ck ( AA- & u p p er)

Le ag/ La g

-5 -4 -3 -2 -1 0 1 2 3 4 5- 0.2

-0.15

- 0.1

-0.05

0

0.05Cro ss Corre la tion Be tween CDS & S tock ( BBB+ ~ BBB-)

Lea d/ Lag -5 -4 -3 -2 - 1 0 1 2 3 4 5-0. 2

-0.1 5

-0. 1

-0.0 5

0

0.0 5Cross Correlat ion Betw een CD S & S tock (BB+~ BB-)

Lea d/La g

-5 -4 -3 -2 -1 0 1 2 3 4 5-0.2

-0.15

-0.1

-0.05

0

0.05Cross Corre la tion Be twee n CD S & Stock (B+ & unde r)

Lead/Lag

-5 -4 -3 - 2 -1 0 1 2 3 4 5-0 .2

-0 .15

-0 .1

-0 .05

0

0 .05C ro ss Cor r e l a ti on B e tw e e n C D S & S toc k (A + ~ A -)

Le a d / La g

Figure 4B.

Note: We plot the average correlations between CDS return and 5 lead/lag stock returnsbased on the first half sub-sample Jun 2004 to Jun 2007. We present correlations based onthe entire firm sample as well as for individual credit categories.


Figure 4C.

Note: We plot the average correlations between CDS return and 5 lead/lag stock returnsbased on the second half sub-sample Jul 2007 to Aug 2010. We present correlations basedon the entire firm sample as well as for individual credit categories.

-5 -4 -3 -2 -1 0 1 2 3 4 5-0.25

-0.2

-0.15

-0.1

-0.05

0

0.05Cross Correlation Between CDS & Stock (All)

Lead/Lag -5 -4 -3 -2 -1 0 1 2 3 4 5-0.25

- 0.2

-0.15

- 0.1

-0.05

0

0.05

0.1Cross Correlation Betwe en S tock & CDS (AA- & uppe r)

Lea d/ Lag

-5 -4 -3 -2 -1 0 1 2 3 4 5-0.25

-0.2

-0.15

-0.1

-0.05

0

0.05Cross Correlation Between Stock & CDS (A+~A-)

Lea d/Lag -5 -4 -3 -2 -1 0 1 2 3 4 5-0 .25

-0.2

-0 .15

-0.1

-0 .05

0

0.05C ros s Correlation Between Stock & CDS (BBB+~BBB-)

Lead /L ag

-5 -4 -3 -2 -1 0 1 2 3 4 5-0.25

-0.2

-0.15

-0.1

-0.05

0

0.05Cross Correlation Bet we en S tock & CDS (BB+~BB-)

Lead/Lag -5 -4 -3 -2 -1 0 1 2 3 4 5-0.25

-0.2

-0.15

-0.1

-0.05

0

0.05Cross Corre la tion Betwe en S tock & CD S (B+- & unde r)

Lea d/ Lag


2. Construction of Credit Condition Dummy Variables

significant, this implies that ∑s1S Djtit−s imposes a significant and permanent

conditional price impact on rit. To ensure that any significant finding is not the result of unexplained stock return lag dynamics, equation (2) allows lagged stock

returns to also impose both unconditional (∑s1S a1 is ) and conditional

(∑ s1S a2is ) price impact on rit.

We consider five credit condition dummy variables that are designed to indicate the deteriorating in a firm’s credit quality. Denote Djt = { CDSit

D ,

CDSitD∆ , BBBit

D , downitD , which we sequentially substitute to estimate equation

(2). CDSitD , CDSitD∆ and BBBitD are similar to the credit condition dummies

considered in Acharya and Johnson (2007). In equation (3), CDSitD and CDSitD∆ detect credit deterioration base on a firm’s CDS spread. CDSitD indicates whether

the price of default protection against Firm i exceeds 100bps of the insured debt. CDSitD∆ captures a one-day rise in CDSit of more than 50bps. These are the

same thresholds used by Acharya and Johnson (2007).

The three credit-rating dummy variables are outlined in equation (4). DBBB it is set with BBB rating as the threshold because market participants generally consider a BBB S&P rating as the borderline between investment-grade and speculative-grade status. Ddown it and Dup it correspondingly indicate if a firm has recently experienced a downgrade and upgrade. The end-of-month credit rating of each firm allows us to infer if the firm has a rating change during the month. We set Ddown it 1 ( Dup it 1 ) every day for the coming month for a firm that receives a lower (higher) rating compared to the previous month. For example, consider Firm i that is rated A at the end of Month k – 1. Assume it has been downgraded to B at the end of Month k. For this firm, we set Dd own it 1 for every day during Month k + 1. The construction of Dup it is similarly described.

(4)

1, ;DBBBit 0,

if BBB or lowerotherwise.

=⎧⎪⎪⎨⎪⎪⎩

1 Month k 1, if downgraded; 1 Month k , if upgraded;

0, otherwise 0, otherwisedownit

t + t +1 Dupit. .

D ∀ ∈ ∀ ∈=

⎧ ⎧⎪ ⎪⎪ ⎪=⎨ ⎨⎪ ⎪⎪ ⎪⎩ ⎩

CDS CDS

1, if CDS bp; 1, if CDS bp;0, otherwise 0, otherwise

it itit it

100 50D D

. .∆

⎧ ⎧> ∆ >⎪ ⎪⎪ ⎪= =⎨ ⎨⎪ ⎪⎪ ⎪⎩ ⎩(3)


In Table 3, we report the number of firms that have been detected by each ofthe five dummy variables at least once over the sample period.12 We allocate firmsinto various credit categories based on their time-series average credit-rating. Forthat purpose, we assign numeric values to each of the 21 S&P ratings. AAA isassigned 1, AA+ is assigned 2 and so forth. C, which is the lowest rating, is assigned21.

All five dummy variables exhibit similar firm distributions across creditcategories. 288 out of 302 firms have CDSit exceeding 100bp at least once duringthe sample period. This is expected since Figure 1 shows that CDS spreads soaredduring the GFC. One hundred thirty four of these firms are in the BBB+ to BBB-category. This is followed by the A+ to A- category with 76 firms and BB+ to BB-with 49 firms. 190 out of 302 firms experience a single-day rise in CDS spread ofmore than 50bp at least once during the sample period. The largest category isBBB+ to BBB- with 89 firms, followed by the BB+ to BB- category with 47 firms,and A+ to A- category with 34 firms. For the credit-rating dummies, interestingly,the firm distributions for and are identical. Upon further examinationit turns out that each of the 244 firms has experienced at least a downgrade and anupgrade during the sample period.

3. Full Sample Period Panel Estimation

We estimate equation (1) for each firm over the entire sample period to extractεit. The cross-sectional mean and median R2 are 10.63% and 9.19%, respectively.These values are higher than those reported by Acharya and Johnson (2007), whichis not surprising given we expect a higher degree of co-movement between ∆CDSitand rit for a sample period that contains the GFC. The mean and volatility of εit arealso larger than those in prior studies. Again, this is expected, given that many firmsexperience substantial credit deterioration, especially in the second half of the sampleperiod. The estimation of equation (2) allows us to formally test if the larger and morevolatile CDS return innovations translates into greater price impact on equity returns.

itdownDitupD

12. To note, the dummies are non-mutually exclusive; for example it is possible for a firm to beassigned 1 for , , and .CDSD

it CDSDit∆ DBBBit

Ideally, we would set downitD and upitD corresponding to actual rating

change announcement dates, if we have the data. However, note that Ddo wn it and Dup it are only picking up rating changes with a delay during Month k + 1, rather than Month k. This would actually make any significant findings of conditional price impact associated with rating changes even stronger. As a robustness check, we also consider rating change dummy variables to indicate firms that have been downgraded (upgraded) in the past month, and has not been upgraded (downgraded) in the last three months. Since the main results are consistent with those using downitD and upitD , we focus on reporting results based on downitD

and Dup it .


Tab

le 3

. Dum

my

Var

iabl

e F

irm

Dist

ribu

tions

acr

oss C

redi

t Cat

egor

ies.

bp

Dit

CDS

100

>

bpD

itCD

S50

∆>

it

BBB

D

itdo

wnD

it

upD

Tota

l 28

8 19

0 19

6 24

4 24

4 A

A+~

AA

- 12

4

0 9

9 A

+~A

- 76

34

10

63

63

B

BB+

~BB

B-

134

89

121

110

110

BB

+~BB

- 49

47

49

47

47

B

+~B

- 16

16

16

15

15

Th

e ta

ble

prov

ides

the n

umbe

r of f

irms,

with

in e

ach

cred

it ca

tego

ries

,tha

t hav

ebe

en in

dica

ted

by a

giv

en d

umm

y va

riab

le a

t lea

st o

nce.


We conduct a series of panel estimation of equation (2) by alternating amongeach of the five credit condition variables Djt . Full sample period results are reportedin Table 4 (panel B), while rolling-window estimation results are presented in thenext section. For panel estimation, we need to be mindful of potential biases in thestandard errors that could lead to invalid test statistics. In general, biases in standarderrors are caused by either serial dependence in the residuals of a given firm, orcross-sectional dependence in firms’ residuals at a given time. Pooled panel OLSestimation is potentially exposed to both. For our panel estimation we address thepreceding concern by computing test statistics using the Petersen (2009, RFS Section2.1) adjustment for cross-sectional dependence.14

There are two reasons why we focus on cross-sectional dependence. First,our main analysis on conditional price impact is based on various indicative measuresof each firm’s credit condition. To the extent that change in credit condition issystemic, especially during the GFC, this is likely to cause cross-sectional dependencein the residuals across firms. As one of our main objectives is to investigate time-varying patterns in the informativeness of CDS trading on stock returns as weapproach and depart from the height of the GFC, we need to ensure that anysignificant findings are robust to adjustments for cross-serial dependence. Second,we are less concerned with serial dependence. The panel regression applies toequation (2). Each firm’s εit is extracted from time-series regression of (1), whichcontains lags of both stock and CDS returns. In addition, lags of stock returns andCDS return innovations are also included in the estimation of equation (2).

13. For example, if the firm proportions for the three sets of coefficients are all close to zero, thisraises serious questions about the validity of equation (1), and of the CDS trading innovations thatwe extract.14. We thank the anonymous referee for reminding us about the issue of invalid test statistics and forbringing to our attention the Petersen (2009) paper, which is a very good reference for papers thatutilize panel regressions.

In Panel B, the results show that ∑s1S b1 s is jointly insignificant across all

Djt. However, the conditional price impact ∑ s1S b2 s is jointly significant for all

Djt. This confirms existing studies that informed trading in the CDS market

In addition to the above summary statistics, we report in Table 4 (panel A) the proportion of firms for which each of the three sets of estimated coefficients ∑ s0

S is , ∑s0S is and ∑ s0

S is are jointly significant in equation (1). Our main analysis is based on the estimation of equation (2), which utilizes the residuals extracted from equation (1). Hence it is important to gauge whether equation (1) provides a reasonable specification for our sample.13 Table 4 (panel A) confirms that equation (1) fits our sample reasonably well. Own-lag CDS

returns (∑s0S is ) are jointly significant in more than 50% of firms throughout

the sample period. Current and lagged stock returns (∑s0S is ) are jointly

significant for between 30% and 45% of firms. Lastly, non-linear effects between

CDSit and lagged stock returns (∑s0S is ) are also jointly significant for

between 30% and 47% of firms.


Tab

le 4

. Inf

orm

atio

n Fl

ow fr

om C

DS

Mar

ket t

o St

ock

Mar

ket.

Pane

l A: F

irm-le

vel e

stim

atio

n to

ext

ract

CD

S re

turn

inno

vatio

ns

Estim

ated

co

effic

ient

s RW

1 RW

2 R

W3

RW

4 R

W5

RW

6 RW

7 R

W8

RW9

issα 1

51

∑=

81

.00%

73

.42%

64

.78%

55

.81%

53

.16%

54

.49%

56

.15%

60

.60%

63

.25%

issβ 1

51

∑=

33

.67%

43

.52%

43

.85%

44

.19%

40

.20%

31

.23%

27

.57%

31

.13%

38

.74%

issγ 1

51

∑=

34

.67%

39

.20%

39

.53%

35

.55%

32

.23%

29

.57%

30

.90%

36

.09%

46

.36%


Tab

le 4

, con

tinue

d. In

form

atio

n Fl

ow fr

om C

DS

Mar

ket t

o St

ock

Mar

ket.

Pane

l B: P

anel

regr

essio

n re

sults

on

the

pric

e im

pact

of C

DS

retu

rn in

nova

tions

on

stoc

k re

turn

s

bpD

itC

DS10

0>

bp

Dit

CDS

50∆

>

itBB

BD

it

down

D

*it

down

D

itup

D

0α

0.00

05

0.00

05

0.00

05

0.00

05

0.00

05

0.00

05

(1

.07)

(1

.06)

(1

.05)

(1.0

6)

(1.0

4)(1

.06)

issα 1

51

∑=

-0.0

790

-0.1

122

-0.1

511

-0.1

352

-0.0

899

-0.0

489

(2

.97)

**

(2.4

2)**

(1

.16)

(1.4

2)

(0.3

9)(0

.91)

issα 2

51

∑=

0.00

92

0.05

70

0.11

94

0.10

91

0.09

36

-0.0

925

(0

.59)

(2

.30)

**

(2.8

6)**

(2

.75)

**(3

.18)

**

(2.2

8)**

ss

b 15

1∑

= -0

.001

0 0.

0004

-0

.004

0 -0

.001

9 -0

.004

0 -0

.006

8

(0

.55)

(0

.54)

(1

.15)

(0

.84)

(1

.16)

(0

.92)

ss

b 25

1∑

= -0

.017

3 -0

.015

4 -0

.006

3 -0

.022

4 -0

.062

9 -0

.003

1

(1

.92)

* (1

.88)

* (2

.05)

* (1

.91)

*(1

.34)

(2.6

7)**

Pa

nel A

con

tain

s su

mm

ary

resu

lts fr

om fi

rm-le

vel e

stim

atio

ns o

f equ

atio

n (1

) to

extra

ct C

DS

retu

rn in

nova

tions

. We

repo

rt th

e pr

opor

tion

of fi

rms

for

whi

ch th

e cu

rren

t and

lagg

ed s

tock

re tu

rns

( ∑=

isSα 1

51

), la

gged

CD

S re

turn

s (∑

=is

Sβ 1

51

), an

d la

gged

CD

S sp

read

adj

uste

d

stoc

k re

turn

s ( ∑

=is

Sγ 1

51

) are

join

tly s

igni

fican

t. In

Pan

el B

, we

repo

rt th

e co

effic

ient

s and

test

-sta

tistic

s fro

m fu

ll-sa

mpl

e pa

nel-r

egre

ssio

n of

equ

atio

n (2

) us

ing

alte

rnat

ive

indi

cato

r va

riabl

es to

mea

sure

cre

dit d

eter

iora

tion.

To

valid

ate

that

the

info

rmat

iven

ess

of C

DS

retu

rn

inno

vatio

ns i

s co

nditi

onal

on

dow

ngra

des

only

, w

e ap

ply

varia

nt m

easu

res

of

itdo

wn

D.

*it

down

D e

qual

s 1

if a

firm

exp

erie

nced

a

dow

ngra

de in

the

past

mon

th, a

nd n

o up

grad

es in

the

last

thre

e m

onth

s, a

nd z

ero

othe

rwise

. The

resu

lts a

re ro

bust

to d

iffer

ent i

ndic

ativ

e m

easu

res o

f dow

n gra

de. F

-sta

tistic

s in

pare

nthe

ses;

F-c

ritic

al (1

0%, 5

, >12

0) =

1.8

7); F

-crit

ical

(5%

, 5, >

120)

= 2

.25)

.


Table 4 provides an overview of the conditional price impact of CDS returninnovations on stock returns. In the following two sections, we perform (i) rolling-window and (ii) firm-level estimation to acquire better insights into the nature ofcredit-risk information flow from the CDS market to the stock market during thecourse of the GFC.

C. Rolling-Window Estimation

Acharya and Johnson (2007) find that the CDS market is more efficient thanthe equity market at processing negative private information. It would be interestingto see how the conditional price impact of lagged εit on rit varies as the estimationwindow moves towards and away from the height of the GFC. More importantly,we need to validate if the price impact of εit and ratings downgrades both vary inresponse to the GFC, rather than in association with each other.

We set the estimation window at 700 observations, which we forward shift insteps of 100 observations. This translates into 9 rolling-windows {RW1,...,RW9}

revolves around negative firm information. The magnitude of the conditional

price impacts are also comparatively larger than the corresponding ∑s1S b1s for

all Djt. Take CDSitD∆ , for example. Its Ss 1 2s| b=∑ = 1.54% while S

s 1 1s| b |=∑ =

0.04%; for downitD , the conditional and unconditional price impacts are 2.24%

and 0.19% respectively. The exception is upitD . Intuitively, there exists an

inverse relation between lagged εit and rit. If CDS trading revolves around negative information, then CDS return innovations would have a curtailed impact on stock returns for firms that experience a recent upgrade. More importantly,

although the magnitude of ∑ s1S b2s for Dup it is smaller than ∑s1

S b1s , it is highly significant. On a related note, our sample contains 266 and 170 firm-month downgrades and upgrades respectively. Although there are more downgrades than upgrades, the numbers are comparable.

For variables that are jointly significant, the sum of coefficients are negative, for example, for Ddown itt−s , S

s 1 2sb=∑ = –0.0224, while for

CDS ( )t-sitD ε , Ss 1 2sb=∑ = –0.0173. This is consistent with expectation since a

positive CDS innovation should impose a negative price impact on stocks. The magnitude of the price impact imposed by it ranges from 1.54% for CDSitD to

2.24% for Ddownit . Panel B contains a column of results for an alternative downgrade measure. Denote downit

D∗ = 1 when Firm i is downgraded in the past

month, and it has not been upgraded in the last three months. Table 4 shows that

the magnitude of ∑s1S b2 s increases from 2.24% to 6.29%. However, it is not

statistically significant. We apply a similar measure for upgrades, but there is no substantial difference from the results based on Dup it .


over the entire sample period. RW4 and RW5 correspond to the onset of the GFC.We vary the estimation window (between 500 and 700 observations) as well as therolling-step (between 50 and 150 observations). The main results are robust todifferent configurations.15

15. Due to space constraint, we present results for the 700-window and 100-step configurations,which are representative of results based on other configurations. The latter results are availableupon request.16. The results are readily available upon request.

Table 5 reports the number of firms that has been indicated by a given dummy variable at least once during each RW. For CDSitD , the number of firms increase by 49 going from RW4 to RW5. There is a further increase of 28 firms to 182 from RW5 to RW6. A similar trend is observed for CDSitD∆ , where the

number of firms increase from 105 in RW4 to 154 in RW5, and then to 182 firms in RW6. In contrast, the number of firms indicated by DBBB it is quite stable across RWs. Table 1 suggests there is evident firm movements both in and out of the BBB+ to BBB- credit category over the sample period, which could explain the stability in firm numbers indicated by DBBB it . Dup it and Ddo wn it display contrasting time-trends, with Ddown it gradually increasing until RW5. Then the number of firms jump from 127 to 165 in RW6, and declined slightly to 148 firms in RW9. The number of firms indicated by Dup it between RW1 and RW5 is quite stable at around 90 firms. Dup it then lost 21 firms in RW6, and a further 12 firms in RW8. From November 10, 2008, to April 6, 2009, 38 (21) firms have been downgraded (upgraded).

In Table 6, we present coefficient estimates and test statistics for ∑ s1S b1s

and ∑ s1S b2s across five panels corresponding to Djt. Similar to Table 4, these

results are generated from panel estimation of equation (2) across {RW1,...,RW9}, using the Petersen (2009) adjusted standard errors to compute

test statistics. The results for ∑ s1S a 1s and ∑s1

S a2s are generally consistent across the five panels. Hence we exclude them from Table 6, due to space constraint. Furthermore, they are not the coefficients of interest that relate to CDS trading informativeness.16

For unconditional price impact, the majority of ∑s1S b1s are jointly

insignificant. This is consistent with Acharya and Johnson (2007). For the three

rating dummies, ∑ s1S b1s is jointly insignificant across all rolling windows,

except Panel C RW1. But interestingly, Panels A and B reveal that ∑ s1S b1s

becomes jointly significant correspondingly from RW4 and RW5 onward, which correspond to the onset and height of the GFC. However, the magnitude of unconditional price impacts are small compared to conditional price impacts, and this result is very consistent throughout Table 6.


For conditional price impact, panel A shows a similar time-variation in the

joint significance of ∑s1S b2s over rolling windows as compared to ∑s1

S b1s . It becomes jointly significant from RW5 onwards, corresponding to the height of

the GFC. The magnitude of ∑ s1S b 2s jumps from 0.62% in RW4 to 5.92% in

RW5. It declines to 2.33% in RW6, and stabilizes at slightly above 3% for the rest of the sample period, but these values remain larger compared to the pre-

GFC period. In both panels B and C, ∑s1S b2s is jointly significant for the

majority of rolling windows. However, there is a substantial increase in |∑ s1

S b2s| from RW5 onwards. For example, in panel B, |∑s1S b2s| increases

from 0.06% in RW4 to 0.47% in RW5. In panel C, |∑s1S b2s| jumps from 0.1%

in RW4 to 1.89% in RW5. Similar to panel A, |∑ s1S b2s| for both panels B and

C decline after RW5, only to stabilize for the rest of the sample period, but at a larger magnitude compared to the pre-GFC period.

For downitD in panel D, ∑ s1S b 2s becomes jointly significant from RW5

onwards. Furthermore, instead of stabilizing as in panels A, B and C, |∑ s1S b2s|

continues to increase in subsequent rolling windows. It jumps from 2.94% in RW5 to 7.51% in RW6, and gradually increases towards 12.35% in RW9. Panel D provides strong evidence to suggest that informed CDS trading imposes a significant and increasingly substantial price impact on the stock returns of firms that have been recently downgraded. However, this applies during and after the height of the GFC, but not before. From RW5 onwards, the magnitude of price impact conditional on Ddown it is larger than other credit condition dummies. For

CDSitD , |∑ s1S b2s| range from 0.62% to 5.92%. For CDSitD∆ , |∑ s1

S b2s| range from 3.65% to 5.36%. For firms rated BBB and below, the conditional

price impact range from 1.74% to 3.03%. In contrast, |∑s1S b2s| for Ddown it

range from 2.94% in RW5 to 12.35% in RW9.

For Dupit in panel E, |∑s1S b2s| ranges from 0.25% to 0.48% between

RW6 and RW9, for which it is jointly significant. When discussing Table 4, we explain that if CDS trading revolves mainly around negative information, the magnitude of price impact conditional on a recent upgrade would be smaller. More importantly, consistent with panel D, the conditional price impact of upgrades is also jointly significant after the height of the GFC.

In sum, Table 6 reveals some interesting insight into the time-varying informativeness of CDS trading on stock returns. We find that informed trading from the CDS market does revolve around negative news. In panels C, D and E, ∑ s1

S b1s is jointly insignificant across rolling windows. In panels A and B,

while ∑ s1S b 1s becomes jointly significant from the onset of the GFC, their

magnitudes are small compared to the corresponding ∑ s1S b 2s . In panels A, B


Tab

le 5

. Dum

my

Var

iabl

e D

istri

butio

n ac

ross

Cre

dit C

ateg

orie

s.

Dum

my

RW

1:

7/1/

04–

4/

12/0

7

RW

2:

11/2

2/04

–

9/4/

07

RW

3:

4/18

/05

– 1/

28/0

8

RW

4:

9/8/

05 to

6/

19/0

8

RW

5:

2/1/

06 –

11

/10/

08

RW

6:

6/26

/06

– 4/

6/09

RW

7:

11/1

5/06

–

8/27

/09

RW

8:

4/13

/07

– 1/

21/1

0

RW

9:

9/5/

07 –

8-

31-1

0 bp

Dit

CDS

100

>

115

117

162

219

249

288

288

288

288

bpD

itCD

S50

∆>

58

76

93

10

5 15

4 18

2 18

5 18

6 18

9

itBB

BD

16

3 16

3 16

4 16

3 16

6 18

0 18

2 18

0 18

1

itdo

wnD

87

93

10

6 11

2 12

7 16

5 16

5 16

1 14

8

itup

D

90

95

100

91

88

67

65

53

60

This

tabl

e pr

ovid

es th

e to

tal n

umbe

r of f

irms

whi

ch h

as d

efin

ed c

redi

t con

ditio

n by

the

five

dum

my

varia

bles

for e

ach

rolli

ng w

indo

w.


Tab

le 6

. Inf

orm

atio

n Fl

ow fr

om C

DS

Mar

ket t

o St

ock

Mar

ket:

Full

Firm

Sam

ple

Est

imat

ion.

Pa

nel A

: Est

imat

ion

usin

g du

mm

y va

riab

le

bpD

itCD

S10

0>

RW

1 R

W2

RW

3 R

W4

RW5

RW

6 R

W7

RW

8 RW

9

∑=51

1s

sb

0.

0013

0.

0038

0.

0015

0.

0011

-0.0

025

0.00

48

0.00

05

-0.0

033

-0.0

011

(1

.81)

(1

.56)

(1

.54)

(2

.36)

**

(2.2

8)**

(3

.41)

**

(3.5

7)**

(3

.77)

**

(3.5

0)**

∑=51

2s

sb

-0

.009

3 -0

.012

2 -0

.008

5 -0

.006

2 -0

.059

2 -0

.023

3 -0

.030

5 -0

.033

2 -0

.030

8

(1.2

3)

(1.6

0)

(1.5

4)

(1.4

9)

(2.3

2)**

(2

.12)

**

(1.9

6)*

(1.7

9)

(2.1

3)*

Pan

el B

: Est

imat

ion

usin

g du

mm

y va

riabl

e bp

Dit

CDS

50∆

>

R

W1

RW

2 RW

3 RW

4 R

W5

RW

6 RW

7 RW

8 R

W9

∑=51

1s

sb

0.

0001

0.

0025

0.

0001

0.

0006

-0

.004

7 0.

01

0.00

62

0.00

4 0.

0086

(1

.47)

(1

.21)

(1

.13)

(1

.79)

(2

.32)

**

(2.5

8)**

(2

.54)

**

(2.5

8)**

(2

.47)

**

∑=51

2s

sb

-0

.001

-0

.002

1 0.

0001

-0

.003

4 -0

.043

8 -0

.036

5 -0

.045

8 -0

.053

5 -0

.053

6

(2

.16)

**

(1.9

7)*

(2.9

9)**

(1.6

7)(5

.38)

**

(2.9

3)**

(3.0

9)**

(3

.20)

**

(3.1

7)**

Pane

l C: E

stim

atio

n us

ing

dum

my

varia

ble

itBB

BD

RW

1 RW

2 R

W3

RW

4 RW

5 R

W6

RW

7 RW

8 R

W9

∑=51

1s

sb

-0

.000

8 0.

0022

-0

.000

1 -0

.001

-0

.018

9 0.

0024

-0

.004

3 -0

.006

7 -0

.006

4

(2

.20)

**

(1.5

0)

(1.0

1)

(1.6

5)

(1.3

6)

(1.2

5)

(1.0

3)

(1.0

5)

(1.0

5)

∑=5

12

ss

b

0.00

09

-0.0

009

0.00

06

0.00

15

-0.0

03

-0.0

174

-0.0

215

-0.0

303

-0.0

265

(3

.19)

**

(3.6

2)**

(1

.23)

(1

.92)

* (2

.96)

**

(2.1

6)**

(2

.82)

**

(2.7

3)**

(2

.55)

**


T

able

6, c

ontin

ued.

Info

rmat

ion

Flow

from

CD

S M

arke

t to

Stoc

k M

arke

t: F

ull F

irm

Sam

ple

Est

imat

ion.

Pa

nel D

: Est

imat

ion

usin

g du

mm

y va

riabl

e it

down

D

RW

1 RW

2 R

W3

RW

4 R

W5

RW6

RW7

RW

8 RW

9

∑=51

1s

sb

-0

.000

2 0.

0018

0.

0001

0.

0000

-0

.018

8 -0

.002

4 -0

.011

1 -0

.017

4 -0

.015

7

(1

.20)

(0

.92)

(0

.91)

(1

.18)

(1.3

9)

(1.4

6)

(1.4

0)(1

.39)

(1

.42)

∑=5

12

ss

b

-0.0

152

-0.0

100

-0.0

069

-0.0

267

-0.0

294

-0.0

751

-0.0

851

-0.1

073

-0.1

235

(1

.53)

(1

.04)

(1

.36)

(1

.04)

(2.0

2)*

(2.4

9)**

(2.9

9)**

(2

.77)

**

(2.7

6)**

Pane

l E: E

stim

atio

n us

ing

dum

my

vari

able

it

upD

RW

1 RW

2 R

W3

RW

4 RW

5 R

W6

RW7

RW8

RW

9

∑=51

1s

sb

-0

.000

7 0.

0018

0.

0011

0.

0004

-0

.016

8 -0

.004

2 -0

.013

-0

.019

2 -0

.020

1

(0.7

4)

(1.5

0)

(0.9

5)

(1.6

4)(0

.85)

(0.9

0)(1

.16)

(1

.10)

(0

.89)

∑=5

12

ss

b

0.00

42

-0.0

01

-0.0

063

-0.0

045

-0.0

129

-0.0

025

-0.0

03

-0.0

048

0.00

25

(0

.77)

(0

.92)

(1

.86)

(1

.48)

(1.2

4)(1

.96)

* (1

.94)

* (2

.52)

**

(2.9

8)**

The

tabl

e re

ports

coe

ffic

ient

s an

d te

st-sta

tistic

s fro

m n

ine

rolli

ng-w

indo

w p

anel

-OLS

estim

atio

ns o

f equ

atio

n (2

). Th

ere a

re g

ive

five

pane

ls of

resu

lts c

orre

spon

ding

to e

ach

of th

e fiv

e cr

edit

dete

riora

tion

dum

my

varia

bles

. F-s

tatis

tics

are

in p

aren

thes

es; F

-crit

ical

(10%

, 5,>

120)

=

1.89

); F-

criti

cal (

5%, 5

,>12

0) =

2.2

5).


whether a firm has been downgraded the previous month. Panel D results implythat, during and after the GFC, there is heightened informed CDS trading in therelevant firms that have been recently downgraded by S&P. At the very least, thissuggests that rating downgrades have become more timely in the aftermath of theGFC. To a certain extent, Panel E results also suggest that rating upgrades havebecome more timely during the later half of the sample period, after the GFC.

D. Firm-level Estimation

The preceding results are based on the pooled panel regression of (2). In thissection we perfrom firm-level time series estimation of equation (2) to track thenumber and proportion of firms for which is jointly significant across{RW1,..., RW9}. In addition to serving as a robustness check, the firm-levelregressions are designed to provide complementary insights into the time-varyingnature of CDS trading informativeness as we move towards and away from theheight of the GFC in RW5.

In Table 7 we report the mean, minimum, and maximum values for overfour panels corresponding to Djt = We report boththe number and percentage of firms with a significant . The percentagesare based on the total number of firms indicated by each Djt in a given RW.

1 2Ss sb=∑

1 2Ss sb=∑

1 2Ss sb=∑

CDS CDSi i{ , , , }.BBB downt t it itD D D D∆

and C, ∑s1S b 2s is jointly significant for the majority of rolling windows.

However, the magnitude of price impact substantially increases during the height of the GFC, and remains larger (relative to pre-GFC period) until the end of the sample period.

More importantly, our results on Ddown it suggest that information flow from the CDS market is prevalent in firms that were downgraded in the past month, regardless of their level of rating. Furthermore, the price impact of informed CDS trading conditional on downitD is larger than those exhibited by CDSitD , CDSitD∆

and DBBB it . However, this finding applies during and after the GFC, but not before. To reiterate, downitD is an ex-post measure that is defined based on

In panel A, the number of firms for which ∑s1S b2s is jointly significant

shows an evident reverse V-shape pattern over time. The number of firms gradually increases from 11 in RW1 to 49 in RW4, only to jump to 91 firms in RW5. Then, the number dropped sharply from 79 in RW6 to 52 firms in RW7, and stabilizes at around 48 firms for the rest of the sample period. CDSitD∆ in panel B reveals a similar time trend, with ∑ s1

S b2s jointly significant for around 16 firms between RW1 and RW4. In RW5, there is a sharp increase to 62 firms, after which there is a gradual decline to 53 firms in RW6, 40 firms in RW7, and the number stabilizes at 28 and 27 firms in RW8 and RW9, respectively. Both

CDSitD and CDSitD∆ also show that the proportion of firms for which ∑ s1S b2s

is jointly significant is highest in RW5 and RW6, which correspond to the height of the GFC.


In Table 8, we estimate the unconditional price impact of εit on rit withoutcredit condition dummies. The reason is to check if the key findings prevail at thefirm-level without relying on any Djt. Table 8 shows that the number of significantfirms has a less evident time trend compared to Table 7. The largest price impactoccurs in RW5 and RW6, consistent with Tables 6 and 7. Hence the overall findingsin Table 8 support our discussions based on conditional price impact. However, weachieve clearer variations over time and better insights by examining price impactbased on credit-condition dummy variables.

Our main result lends some support in defense of credit rating agencies. In theaftermath of the GFC, credit rating agencies were heavily criticized for being slow

For DBBBit in panel C, the number of significant firms more than doubled from 9 in RW4 to 22 in RW5 and RW6. This is followed by a gradual decline. In panel D, downitD shows a similar trend in the number of significant firms. There

is a jump from 22 to 39 firms going from RW4 to RW5, but a decline gradually to around 20 firms in RW9. While the time trend in firm numbers is similar

between panels C and D, notice that |∑s1S b2s| conditional on DBBB it is on

average higher than Ddo wn it from RW2 to RW4. But from RW5 onwards, Ddown it imposes a substantially larger conditional price impact than DBBB it . For

example, in RW5, |∑ s1S b 2s| is 2.96% for DBBBit compared to 9.8% for

Ddown it . This is similarly described for RW6 to RW8, although they become comparable in RW9.

The overall results in Table 7 support our main findings in Table 6 to suggest that CDS innovations are most informative around the middle of the sample period, which corresponds to the height of the GFC. The proportion of significant firms is the highest in RW5 and RW6 across all four panels. Furthermore, the conditional price impact of εit is most substantive in firms that have been downgraded in the past month rather than current low-rated firms. The price impact conditional on downgrade is surprisingly larger than those based on

CDSitD and CDSitD∆ . For example, in RW5, 1 2Ss sb=∑ = –9.8% in panel D,

compared to –7.85% and –7.08% in panels A and B, respectively. In RW6 (RW7), 1 2

Ss sb=∑ = –6.3%(–5%) in Panel D, compared to –5.3% (–0.65%) and

–3.7% (–2.77%) in panels A and B, respectively. Table 7 reiterates the interesting implication that rating downgrades affect

the informativeness of CDS innovations on stock returns. In terms of the number of significant firms, rating downgrade is not as relevant as CDSitD or CDSitD∆ in

explaining credit risk information flow from the CDs to stock market. However, for those firms that are significant conditional on Ddown it , the magnitude of the conditional price impact is larger that credit-condition dummy variables based on the CDS market. Furthermore, the finding holds for downgrades, but not upgrades. This is despite that our overall sample contains a comparable number of downgrades and upgrades.


Tab

le 7

. Con

ditio

nal I

nfor

mat

ion

Flow

from

CD

S M

arke

t to

Stoc

k M

arke

t: F

irm

-lev

el E

stim

atio

n.Pa

nel A

: Est

imat

ion

of ∑

=51

2s

isb

base

d on

dum

my

varia

ble

bpD

itC

DS10

0>

RW

1 R

W2

RW

3 R

W4

RW

5 RW

6 RW

7 R

W8

RW

9

Mea

n -0

.051

2 -0

.019

2 -0

.054

2 0.

0096

-0

.070

8 0.

0370

0.

0277

0.

0305

-0

.020

7

Min

-0

.502

3 -0

.379

5 -2

.961

1 -0

.402

4 -1

.052

3 -0

.436

7 -0

.628

2 -1

.688

7 -4

.565

5

Max

0.

2612

0.

5771

1.

2189

1.

1537

2.

1941

1.

4074

1.

3278

1.

3812

0.

9578

No.

of f

irm

s 11

23

35

49

91

79

52

49

46

% o

f firm

s 3.

64%

7.

62%

11

.59%

16

.23%

30

.13%

26

.16%

17

.22%

16

.23%

15

.23%

Pan

el B

: Est

imat

ion

of ∑

=51

2s

isb

base

d on

dum

my

varia

ble

bpD

itCD

S50

∆>

R

W1

RW

2 RW

3 R

W4

RW5

RW6

RW

7 R

W8

RW9

Mea

n 0.

0768

0.

0888

0.

0467

0.

0291

-0

.078

5 0.

0526

-0

.006

5 -0

.011

1 -0

.018

0

Min

-1

.056

1 -0

.445

8 -0

.535

9 -0

.525

2 -1

.718

8 -0

.750

8 -0

.571

9 -0

.512

5 -0

.987

3

Max

3.

0680

0.

7304

0.

6188

0.

5680

8.

2019

1.

4744

0.

5561

0.

4907

0.

6141

#. o

f fir

ms

14

17

14

17

62

53

40

28

27

% o

f firm

s 4.

64%

5.

63%

4.

64%

5.

63%

20

.53%

17

.55%

13

.25%

9.

27%

8.

94%


Tab

le 7

, con

tinue

d. C

ondi

tiona

l Inf

orm

atio

n Fl

ow fr

om C

DS

Mar

ket t

o St

ock

Mar

ket:

Fir

m-le

vel E

stim

atio

n.

Pane

l C: E

stim

atio

n of

∑=5

12

sis

b b

ased

on

dum

my

vari

able

it

BBB

D

R

W1

RW

2 RW

3 RW

4 RW

5 RW

6 R

W7

RW

8 R

W9

Mea

n 0.

0246

0.

0042

-0

.013

4 -0

.031

6 -0

.029

6 -0

.026

8 0.

0056

-0

.025

7 -0

.012

2 M

in

-0.0

720

-0.2

972

-0.2

489

-0.9

518

-0.4

145

-0.8

026

-0.3

946

-0.5

011

-0.4

024

Max

0.

5049

0.

1191

0.

2037

0.

7949

0 .

3077

1.

3629

0.

8356

0.

3751

0.

5287

N

o. o

f fir

ms

8 16

8

9 22

22

15

13

15

%

of f

irms

2.65

%

5.29

%

2.65

%

2.98

%

7.29

%

7.28

%

4.67

%

4.31

%

4.97

%

Pane

l D: E

stim

atio

n of

∑

=51

2s

isb

base

d on

dum

my

vari

able

it

down

D

R

W1

RW

2 RW

3 RW

4 RW

5 RW

6 R

W7

RW

8 R

W9

Mea

n 0.

0590

0.

0011

-0

.009

7 -0

.012

8 -0

.098

0 -0

.062

9 -0

.050

0 -0

.072

5 0.

0104

M

in

-0.8

456

-1.4

003

-1.7

040

-2.8

846

-2.1

957

-2.2

410

-2.2

301

-2.3

790

-2.2

070

Max

2.

5846

0.

5048

1.

3648

2.

8657

0 .

7236

1.

3707

0.

8337

1.

8870

2.

7769

N

o. o

f fir

ms

14

30

23

22

39

43

25

18

19

% o

f firm

s 4.

64%

9.

93%

7.

62%

7.

28%

12

.92%

14

.24%

8.

28%

5.

96%

6.

29%

Th

e ta

ble

repo

rts d

escr

iptiv

e st

atist

ics

for

the

distr

ibut

ion

of c

ondi

tiona

l pr

ice

impa

ct m

easu

res

base

d on

fir

m-b

y-fir

m e

stim

atio

n of

Eq

uatio

n (2

) fo

r ea

ch o

f th

e ni

ne r

ollin

g w

indo

ws

(RW

). W

e re

port

the

cros

s-se

ctio

nal

mea

n, a

vera

ge F

-sta

tistic

s, m

in a

nd m

ax f

or

∑=5

12

sis

b,

as w

ell

as t

he n

umbe

r of

firm

s in

eac

h R

W f

or w

hich

∑=5

12

sis

b i

s jo

intly

sig

nific

ant

at t

he 5

% l

evel

. T

he f

our

pane

ls

corre

spon

d to

it

CDS

D,

itCD

SD∆

, it

BBB

D, a

nd

itdo

wnD

.


at downgrading firms. Our analysis shows that, at least during and post-GFC,downgrades by S&P trigger information-based trading in the CDS market, which ismanifested in a significant conditional price impact of on stock returns.

E. Robustness Check for Informed Trading Activity

Microstructure theory predicts that bid-ask spreads will widen to reflect theheightened risk of trading against informed investors. Our main analysis indicatesthe presence of heightened informed trading activity through the CDS marketconditional on a firm’s recent credit deterioration. In this section, we perform arobustness check by examining how absolute (ASit) and relative RSit bid-ask spreadsfluctuate over time. We associate heightened informed CDS trading with widerbid-ask spreads. We find generally similar patterns between ASit and RSit.Furthermore, since ASit fluctuates with the level of CDS spread, they may not beas reflective of informed trading activity as compared to RSit. Hence we focus ourdiscussion on relative bid-ask spreads.17

We use firm-level estimation results in Table 7 to sort firms into two groups. Inthe CDS (Equity) Group are firms for which CDS trading imposes a (in)significantprice impact on stock returns, conditional on a recent credit deterioration. Withineach group, we separately compute the daily cross-sectional average relative bid-ask spread for each credit category in each rolling window. This procedure iscarried out from RW1 to RW9, which allows us to compare the magnitude ofbetween the two firm groups over time. Our main results suggest heightenedinformed trading in the CDS market in the second half of the sample period, inparticular RW5 to RW7. Hence we expect the of the CDS Group to graduallybecome larger than the Equity Group as we move from RW1 to RW7, followed bya gradual decline.

In Table 9 we report the time proportion of each RW for which the of theCDS Group is larger than the Equity Group. The four panels correspond to priceimpact results conditional on each of the four credit deterioration dummies ,

, , and . Our earlier analysis indicates heightened informedCDS trading from RW4 and RW5, with a slight decline to follow from RW7 onwards.Hence we expect the time proportion results to display an N-shape pattern acrossrolling windows. To aid the discussion, we plot the Table 9 results in Figure 5.

Out of the five rating categories, AA+ to AA- is the only category where thereis no evident time-trend in the time proportions across the four panels. This is notsurprising given that CDS informed trading is less likely to occur in firms with highcredit quality. But for the other rating categories, there are some evident resultsthat support out main findings. In panel A, the three middle-band rating categoriesall display an upward time-trend towards RW7, with the A+ to A- and BB+ to BB-categories showing a gradual decline after RW7. In panel B, an upward time-trendis observed for the BBB+ to BBB- and BB+ to BB- categories, while an N-shapepattern is observed for the A+ to A- category. Panels C and D display the strongest

down it-sitD ε

RSt

RSt

17. We thank an anonymous referee for pointing this out to us.

CDSitD

RSt

RSt

itD CDS∆ itBBBD

itdownD


Tab

le 8

. Unc

ondi

tiona

l Inf

orm

atio

n Fl

ow fr

om C

DS

Mar

ket t

o St

ock

Mar

ket.

R

W1

RW

2 RW

3 RW

4 RW

5 RW

6 R

W7

RW

8 R

W9

Mea

n -0

.003

1 0.

0008

-0

.000

7 0.

0010

-0

.021

3 -0

.021

1 -0

.011

0 -0

.019

1 -0

.013

2

Min

-0

.135

3 -0

.128

3 -0

.145

3 -0

.128

6 -0

.399

2 -0

.438

2 -0

.518

2 -0

.547

1 -0

.613

6

Max

0.

1126

0.

0872

0.

1471

0.

1515

0.

1107

0.

1931

0.

2475

0.

2633

0.

2446

No.

of f

irms

24

38

28

24

42

43

25

33

34

The

tabl

e re

ports

des

crip

tive

statis

tics f

or th

e un

cond

ition

al p

rice

impa

ct v

aria

ble ∑

=51

1s

isb

acro

ss fi

rms,

whi

ch a

re b

ased

on

OLS

esti

mat

ions

of

equa

tion

(2) o

n in

divi

dual

firm

s ov

er n

ine

rolli

ng-w

indo

ws.

We

repo

rt th

e cr

oss-

sect

iona

l mea

n, F

-sta

tistic

s, m

in a

nd m

ax fo

r ∑=51

1s

isb

, as

wel

l

as th

e nu

mbe

r of f

irms i

n ea

ch ro

lling

win

dow

for w

hich

for ∑

=51

1s

isb

is

signi

fican

t at t

he 5

% le

vel.


Tab

le 9

. Tim

e Fr

actio

n fo

r w

hich

the

Rel

ativ

e B

id-A

sk S

prea

d of

the

CD

S G

roup

is L

arge

r th

an E

quity

Gro

up:

Con

ditio

nal G

rou p

ing.

RW1

RW2

RW3

RW4

RW5

RW6

RW7

RW

8 RW

9

Pane

l A: C

redi

t det

erio

ratio

n in

dica

ted

by

bpD

itCD

S50

∆>

A

A+

to A

A-

0.93

29

0.84

00

0.97

57

0.91

43

0.30

86

0.38

00

0.45

57

0.63

71

0.68

86

A+

to A

- 0.

4943

0.

5214

0.

6329

0.

6814

0.

6400

0.

7114

0.

8029

0.

5714

0.

4929

B

BB+

to B

BB

- 0.

7914

0.

7214

0.

6943

0.

6871

0.

7986

0.

8343

0.

9371

0.

9329

0.

9014

B

B+

to B

B-

0.45

57

0.33

00

0.51

43

0.68

71

0.58

43

0.60

43

0.60

86

0.43

00

0.41

43

B+

to B

- 0.

7600

0.

7543

0.

7257

0.

7186

0.

5743

0.

5114

0.

4157

0.

5286

0.

5329

Pane

l B: C

redi

t det

erio

ratio

n in

dica

ted

by

bpD

itC

DS

100

>

AA

+ to

AA

- 0.

9329

0.

6357

0.

6986

0.76

570.

7243

0.

788 6

0.

4557

0.52

710.

5171

A

+ to

A-

0.61

29

0.48

00

0.38

71

0.50

86

0.88

71

0.76

57

0.77

14

0.54

71

0.53

57

BB

B+ to

BB

B-

0.48

00

0.68

71

0.51

430.

5600

0.59

29

0.62

71

0.64

290.

7714

0.85

29

BB

+ to

BB-

0.

4557

0.

7186

0.

6086

0.60

140.

7114

0.

675 7

0.

7214

0.75

710.

7243

B

+ to

B-

0.76

00

0.62

29

0.56

86

0.57

14

0.60

86

0.53

71

0.49

14

0.47

71

0.46

43


Tab

le 9

, con

tinue

d. T

ime

Frac

tion

for

whi

ch th

e R

elat

ive B

id-A

sk S

prea

d of

the

CD

S G

roup

is L

arge

r th

an E

quity

Gro

up:

Con

ditio

nal G

rou p

ing.

Pa

nel C

: Cre

dit d

eter

iora

tion

indi

cate

d by

it

BBB

D

AA

+ to

AA

- 0.

9471

0.

8629

0.

9771

0.

9771

0.

5829

0.

6100

0.

9229

0.

9543

0.

9429

A

+ to

A-

0.72

71

0.52

43

0.40

86

0.46

29

0.52

43

0.60

43

0.87

43

0.78

43

0.78

29

BB

B+ to

BB

B-

0.43

86

0.73

14

0.39

14

0.47

71

0.94

57

0.96

57

0.78

43

0.75

43

0.78

29

BB

+ to

BB-

0.

0500

0.

3400

0.

0800

0.

1843

0.

7457

0.

6414

0.

4643

0.

3614

0.

3043

B

+ to

B-

0.70

86

0.75

43

0.71

43

0.71

86

0.74

57

0.64

14

0.54

29

0.41

43

0.35

00

Pan

el D

: Cre

dit d

eter

iora

tion

indi

cate

d by

it

down

D

AA

+ to

AA

- 0.

9329

0.

6357

0.

6986

0.

6043

0.

6086

0.

6800

0.

6043

0.

8800

0.

6886

A

+ to

A-

0.49

43

0.46

00

0.44

570.

5257

0.

5029

0.74

86

0.66

86

0.66

00

0.49

71

BB

B+ to

BB

B-

0.79

14

0.54

86

0.78

71

0.88

57

0.65

14

0.74

57

0.83

57

0.72

86

0.70

71

BB

+ to

BB-

0.

4557

0.

4371

0.

6700

0.

7129

0.

7443

0.

6971

0.

5100

0.

2171

0.

2129

B

+ to

B-

0.76

00

0.68

00

0.56

000.

7186

0.

6857

0.49

43

0.51

00

0.41

57

0.33

57

We

sort

firm

s int

o tw

o gr

oups

for e

ach

rolli

ng-w

indo

w. T

he C

DS

(Equ

ity) G

roup

con

tain

s firm

s fo

r w

hich

the

cond

ition

al p

rice

impa

ct o

f C

DS

tradi

ng o

n sto

ck r

etur

n is

join

tly s

igni

fican

t (in

sign

ifica

nt),

base

d on

eac

h of

the

four

cre

dit c

ondi

tion

dum

my

varia

bles

to

indi

cate

cr

edit

dete

riora

tion.

The

tabl

e sh

ows

the

time

frac

tion

for w

hich

the

cros

s-se

ctio

nal a

vera

ge re

lativ

e bi

d-as

k sp

read

for

the

CDS

grou

p is

larg

er th

an th

e Eq

uity

gro

up.


Figure 5.

Note: We plot the time proportion of each rolling-window for which the cross-sectional average relative bid-ask spread tRS for the CDS Group is larger than the Equity Group. The firm groupings are based on Table 7 results on whether the CDS innovations provide a jointly significant conditional price impact on stock returns. We plot four separate graphs base on each of the four indicative measures of credit deterioration.


evidence of N-shape patterns in the time proportions across various rating categories.Indeed, Figure 5D, which is based on rating downgrades, shows that the majority ofrating categories display an evident N-shape pattern in the time-proportions overtime. This is consistent with Table 7, which shows that the informativeness of CDStrading is strongest conditional on recent downgrades.

III. CONCLUDING REMARKS

One would instinctively expect the price discovery mechanism of any credit-related market to cease functioning properly during a systemic credit-crunch,including, and especially, the US CDS market. In this paper, we find that credit riskinformation flow from the CDS market to the stock market did not weaken duringthe course of the GFC. In fact, CDS trading is most informative during the height ofthe GFC. Our significant findings are conditional on recent credit deteriorations, aswith Acharya and Johnson (2007). However, we find that CDS innovationsconditional on recent downgrades impose a larger price impact compared to othercredit condition dummy variables, including those used in Acharya and Johnson(2007). We do not find significant price impact conditional on upgrades. This isconsistent with the literature’s finding that the informational efficiency of the CDSmarket is skewed towards negative news.

Our paper offers an interesting finding that informed trading in the CDS marketis associated less with currently low-rated firms, and more with firms that havebeen downgraded in the past month. This necessarily implies that downgradedecisions by S&P, in particular those decisions made during and post-GFC, haveinduced CDS traders to further examine those downgraded companies, trade theirprivate information through the CDS market, which subsequently flow onto thestock market. Put differently, S&P downgrade decisions provide incrementalinformation to the CDS market about shifts in the credit quality of recentlydowngraded firms.

Indeed, criticism of rating agencies displaying strong inertia in downgradingfirms is justified. Our analysis shows that CDS innovations do not display significantprice impact conditional on recent downgrades in the lead-up to the GFC. However,our paper also lays support in defense of rating agencies in the aftermath of theGFC, since their downgrade decisions have provided incremental information forCDS traders to have a closer look at the recently downgraded firms. Indeed, atleast the rating agency S&P seems to have gotten its act together in the aftermathof the GFC.

References

Acharya, V. and Johnson, T., 2007, Insider Trading in Credit Derivatives. Journalof Financial Economics, 84(1), 110-141.

Blanco, R., Brennan, S., and Marsh, I., 2005. An Empirical Analysis of the DynamicRelation between Investment-grade Bonds and Credit Default Swaps. Journalof Finance, 60(5), 2255-2281.


Chava, S., Ganduri, R., and Ornthanalai, C., 2012, Are Credit Ratings Still Relevant?SSRN-FEN working paper series.

Chng, M., Fang, V., and Xiang, V., 2013, Transmigration across Price DiscoveryCategories: Evidence from the U.S. CDS and Equity Market. Journal ofFutures Markets, June, In-press.

Das, S., Kalimipalli, M., and Nayak, S., 2012, Did CDS Trading Improve the Marketfor Corporate Bonds? SSRN-FEN working paper series.

Duffie, D., 1999, Credit Swap Valuation. Financial Analysts Journal, 55(1), 73-87.

Ericsson, J., Jacobs, K., and Oviedo-Helfenberger, R., 2009, The Determinants ofCredit Default Swap Premia. Journal of Financial and QuantitativeAnalysis, 44(1), 109-132.

Flannery, M., Houston, J., and Partnoy, F., 2010, Credit Default Swap Spreads asViable Substitutes for Credit Ratings. University of Pennsylvania Law Review,158, 2085-2123.

Forte, S. and Pena, J., 2009, Credit Spreads: An Empirical Anlaysis on the InformationalContent of Stocks, Bonds, and CDS. Journal of Banking and Finance, 33,2013-2025.

Fung, H., Sierra, G., You, J., and Zhang, G., 2008, Are the U.S. Stock Market andCredit Default Swap Market Related? Evidence from the CDX Indices. Journalof Alternative Investments, 11(1), 43-61.

Greatrex, C., 2009, Credit Default Swap Market Determinants. Journal of FixedIncome, 18(3), 18-32.

Hilscher, J., Pollet, J., and Wilson, M., 2012, Are Credit Default Swaps a Sideshow?Evidence that Information Flows from Equity to CDS Markets. SSRN-FENworking paper series.

Hotchkiss, E. and Ronen, T., 2002, The Informational Efficiency of the CorporateBond Market: An Intraday Analysis. Review of Financial Studies, 15(5), 1325-1354.

Hull, J., Predescu, M., and White, A., 2004, The Relationship between Credit DefaultSwap Spreads, Bond Yields, and Credit Rating Announcements. Journal ofBanking and Finance, 28, 2789--2811.

Longstaff, F., Mithal, S., and Neis, E., 2003, The Credit Default Swap Market: IsCredit Protection Priced Correctly? Working paper, University of California.

Longstaff, F., Mithal, S., and Neis, E., 2005, Corporate Yield Spreads: Default Riskor Liquidity? New Evidence from the Credit Default Swap Market. Journalof Finance, 60(5), 2213-2253.

Merton, R., 1974, On the Pricing of Corporate Debt: The Risk Structure of InterestRates. Journal of Finance, 29(2), 449-470.

Norden, L. and Weber, M., 2004, Informational Efficiency of Credit Default Swapand Stock Markets: The Impact of Credit Rating Announcements. Journal ofBanking and Finance, 28(11), 2813-2843.


Norden, L. and Weber, M., 2009, The Comovement of Credit Default Swap, Bondand Stock Markets: An Empirical Analysis. European Finanical Management,15(4), 529-562.

Petersen, M., 2009, Estimating Standard Errors in Finance Panel Data Sets:Comparing Approaches. Review of Financial Studies, 22, 435-480.

Yu, F., 2006, How Profitable is Capital Structure Arbitrage? Financial AnalystsJournal, 62(5), 47-62.

Documents

This is the published version - Deakin Universitydro.deakin.edu.au/eserv/DU:30061932/wang-ratingdowngrade-2013.pdf · JEL Classification: G14, G15. 284 Review of Futures Markets Institute