The role of the CDS market in pricing Eurozone sovereign risk - - PowerPoint PPT Presentation

The role of the CDS market in pricing Eurozone sovereign risk

Richard Portes

London Business School and CEPR The Debt Crisis in the Eurozone Reykjavik University 7-8 October 2011

Road map

 Background and objectives  Procedure and literature  Analytical framework  Data and period  Cointegration tests  VECM (vector error correction model)  Conclusions  Alternative hypothesis

CDS pricing

 CDS contract: insurance against deterioration

• f the credit standing of a bond issuer –

‘credit protection’

 Spread (insurance premium) prices credit risk  So does price of bond (spread between yield

and riskless rate)

 We expect to see close relation between the

two spreads – no-arbitrage condition (Duffie 1999)…

…but maybe not

 Different markets, different market participants  Different maturities: CDS contract on bond typically

won’t have same term as bond itself

 CDS not only a hedge on underlying bond – since

2002-03, CDS market dominated by speculation (‘naked CDS’), in which buyer does not hold underlying bond

 Buying naked CDS similar to short-selling

underlying bond, but easier, and no capital risk

 CDS spread must at least set floor under reference

entity’s borrowing costs

Our aims

 Test long-term accuracy of credit risk pricing

in CDS market

 Investigate price discovery relationship:

which market leads?

 These are limited objectives – there are

bigger issues at stake – but first one wants to see how the data behave

Procedure

 ADF tests on spread series  If spreads cointegrated of order one, then

use Johansen test to determine if bond yield spread and CDS premia move together in long run

 If cointegration, then estimate VECM to

establish price discovery leadership

 Also do Granger causality tests between the

two series

Literature

 Most papers study corporate bonds and CDS on them  Of the few studies on CDS on sovereign bonds, most

focus on emerging markets

 Only two papers on EU sovereign bond CDS  Fontana-Schneider (2010) look at 10 EU countries over

2006-2010, focusing on determinants of difference between the two spreads, split sample between pre- and post-crisis

 Arce et al. (2011) seek evidence of market frictions

that impede arbitrage

 Methodologically closes to us are Zhu (2006), Ammer

and Cai (2007), Blanco et al. (2005)

Pricing framework

 Reduced form approach  With no transaction costs, should have perfect arbitrage

between risky bond, riskless bond, CDS

 Yield of risk-free bond should equal difference between

yield of corresponding risky bond and cost of credit protection as percentage of risky bond nominal value (CDS spread):

 E.g. if risk-free bond yield exceeds difference between

risky bond yield and CDS spread, then buy risk-free bond, short risky bond, sell protection in CDS market

The bond-CDS bas

asis

 Difference between CDS spread and bond yield

spread on same reference entity

 With perfect arbitrage  Note: this assumes no counterparty risk (can

credit protection seller pay off after credit event? recall AIG) – but that seems to have surprisingly little effect on spreads in practice (Arce et al.)

 Note: strong demand from credit protection

buyers would push basis up

Basis trades

 Basis positive if risk premium (spread) on risky bond

is ‘too low’ or CDS spread is ‘too high’

 So short risky bond – but rigidities in cash market

may slow arbitraging

 If basis negative, then short risk-free bond, normally

easier (market more liquid)

 So might expect basis usually positive  Various factors might drive basis up or down

CDS market

 ‘protection buyer’ pays fixed periodic premium to

‘protection seller’

 Premium expressed in bp of reference asset’s nominal

value

 Outstanding gross notional value of CDS contracts at

31 August 2011 was $15 trn, with 2,156,591 trades

 At end-2007, share of sovereign CDS was 5%, of which

90% were in EM

 But by May 2010, sovereign segment was 15%  Main sellers are investment banks (8 dominate market),

main buyers are hedge funds

Data

 Source: CMA  Period: 30 Jan 2004 – 11 Mar 2011 (data for some

countries available earlier, but some appear to be unreliable, and we chose to have identical periods for all countries)

 6 countries: Austria, Belgium, Greece, Ireland, Italy,

Portugal

 Daily data on 5-year maturity (most liquid market

segment), average spreads across dealers

 Bond yield data from Datastream  Risk-free benchmark taken as 5-yr German govt bond

Structural break?

 Stationary model with structural break could be confused

with unit root model (Perron 1989)

 Unit root tests may be unreliable with structural break  Nonstationarity affects results of tests for structural break

(Perron 2005)

 Our sample period longer than other studies – splitting it

would eliminate that advantage and reduce power of tests

 Testing for structural break would require taking account

• f multiple possible break points – e.g. August 2007,

September 2008, spring 2010

 And our results on full sample seem fairly well-determined

Cointegration tests

 ADF does not reject unit root hypothesis for both CDS

and bond spreads for each country

 Equilibrium theory requires cointegration between the

two spreads – otherwise deviations from equilibrium will not be temporary

 We estimate

where

 If credit spreads equal in equilibrium, then cointegrating

vector might be [1,-1] with α = 0

 But if cointegration with α ≠ 0, β ≠ 1, then we

conclude that the two markets may price credit risk differently in the short run but move together in the long run – and this is what we find for all countries

 Standard Johansen test without assuming structural

breaks may cause over-rejection of cointegration – but in our data, no cointegration is rejected for each country, suggesting that structural breaks do not pose a problem

VECM

 Which market leads in price discovery?  VEC representation models changes in each spread as

function of (a) deviation from cointegration relationship (b) lagged values of changes in both spreads

 Coefficient on deviation term is speed of adjustment  If both those coefficients are significant, we infer that

there is significant price interaction, and relative size of coefficients reflects relative importance in price discovery

 If one market always lags the other, then coefficient in

market that leads price discovery should be 0

VECM results



Austria, Belgium, Greece, Italy: price discovery is 2-way, with CDS market more important



Ireland and Portugal: credit risk is priced in CDS market first



But there are short-term pricing discrepancies



They are persistent: estimated adjustment coefficients suggest

• nly 2% of price discrepancy is eliminated within 2 days



Gonzalo-Granger (1995) measure of relative contribution of each market to price discovery suggests that on average, CDS market leads



Granger causality tests not very powerful but generally confirm these results

Conclusions

 CDS and bond spreads appear to be equal in long run, as

theory suggests

 But in short run there is substantial and persistent

divergence

 And even if CDS market prices risk ‘correctly’ in long run,

that does not mean that credit risk as priced by either market reflects ‘fundamentals’

 CDS market usually leads bond market in price discovery,

possibly because it is more liquid

 But there is 2-way causality

An alternative hypothesis

 CDS market may lead in price discovery because

changes in CDS prices affect fundamentals that drive bond spreads

 If CDS spread affects cost of funding – as it must –

then rise in spread will not merely signal but will cause deterioration in credit quality (Portes 2010)

 Need dynamic model with multiple equilibria