giovanni calice. spillovers in sovereign bond and cds markets: an analysis of the eurozone sovereign...

IntroductionMethodology

Results

Liquidity Spillovers in Sovereign Bond and CDSMarkets: An Analysis of The Eurozone Sovereign

Debt Crisis

Giovanni CaliceSchool of Management, University of Southampton, England, U.K.

Jing ChenSchool Business and Economics, Swansea University, Wales, U.K.

Julian WilliamsBusiness School, University of Aberdeen, Scotland, U.K.

February 2012

GC,JC & JW Liquidity Spillovers in Sovereign Bond and CDS Markets


Results

Eurozone Sovereign Crisis

Ongoing issue of liquidity and solvency of various EU governments.

Causes are diverse (poor fiscal planning in Portugal, expensive bankguarantees in Ireland, falsified national accounts in Greece)

At present Greece, Portugal and Ireland are in receipt of financialguarantees and liquid capital injections via the IMF, EFSF and assetpurchases by the ECB.

Iceland has also received a substantial ‘bail-out’ after the collapse ofits banking system, earlier on in the crisis.

Causes are well known and are for other discussions.

This paper looks at the mechanism of transmission of liquidity andinformation in the price formation mechanism of Eurozone sovereigndebt during the 2007-2011 period.

The paper provides a table of various macroeconomic indicators for2007, 2008, 2009 and 2010 versus the 2001-2006 average.



Results

Outline of Talk

Brief overview of our research questions and methodology.

Our data and the uniqueness of the data set.

A short tour of some of the main results.

Brief concluding remarks.



Results

Liquidity and Price Formation in Crises

Variables

Let BONDt be the yield (or discount premia) for each countriessovereign debt (for either 5 or 10 year maturities) measured in basispoints.

CDSt is the credit default swap rate, in basis points for eachcountry. BONDDE,t and CDSDE,t are respectively the yield andCDS spread on German sovereign debt of 5 and 10 year maturity.

BONDBIDt is the bid yield in basis points for sovereign bonds andBONDASKt is the ask yield for sovereign bonds, again convertedto basis points.

CDSBIDt and CDSASKt are, respectively, the bid and askspreads for 5 and 10 year sovereign CDS in basis points.



Results

Liquidity and Price Formation in Crises

Our key research question is to establish the dynamics of interactionbetween the credit spread on traded Eurozone sovereign debt with thecredit spread on equivalent maturity sovereign CDS and the liquidityspreads on traded sovereign debt and CDSs.

For each country we compute the BONDCSt, the sovereign bondcredit spread, the CDSCSt, the CDS credit spread, theBONDLSt, the sovereign bond liquidity spread and finally theCDSLSt, the CDS liquidity spread. These are computed as follows:

BONDCSt = BONDt −BONDDE,t (1)

CDSCSt = CDSt − CDSDE,t (2)

BONDLSt = BONDBIDt −BONDASKt (3)

CDSLSt = CDSBIDt − CDSASKt (4)



Results

A Time Varying Vector Autoregression

In this paper we provide results for an endogenous time varying VARmodel of price and liquidity formation for sovereign bond and CDSmarkets during the crisis.

BONDCSt = β1,1,tBONDCSt−1 + β1,2,tCDSCSt−1

+β1,3,tBONDLSt−1 + β1,4,tCDSLSt−1 + µ1,t + u1,t

CDSCSt = β2,1,tBONDCSt−1 + β2,2,tCDSCSt−1


BONDLSt = β3,1,tBONDCSt−1 + β3,2,tCDSCSt−1


CDSLSt = β4,1,tBONDCSt−1 + β4,2,tCDSCSt−1


(5)

the coefficients [βi,j ] are collected into the time varying matrix Bt.



Results

We have developed a least squares based alternative to the Kalmanfilter that is robust to structural change, whilst being able to capturelocal stability in the coefficients.

We call this approach recursive and iteratively re-weighted leastsquares (IRLS), which might be thought of as a specific class of theextended least squares approach.

More specifically, the model is a multivariate extension of the singleequation autoregressive model of Arvastson et al. 2000 which is astandard autoregressive model with time varying coefficientsestimated with exponential forgetting.



Results

The eigenvalues of the time varying matrix B̃t offer valuableinformation on the instantaneous stability of the autoregressivemodel.

Consider the time varying eigenvalues of the 4× 4 slope matrix B̃t,ordered from largest to smallest as {λmax,t, λ2,t, λ3,t, λmin,t}.We have imposed a first order VAR on the time varying coefficients,therefore the eigenvalues of this matrix correspond directly topolynomial roots of the VAR process.

If the range of λmax,t to λmin,t is within the unit circle then theinstantaneous static VAR at time t is stationary. A root equal to oneindicates the presence of at least one random walk in the vectorsystem.

Roots greater than unity indicate an explosive stochastic trend.



Results

Variance breakpoint tests

Another helpful by-product of the recursive regression approach isthat a standard matrix equality test can be used to extend thestandard variance break point tests for structural breaks,

By use of a Wishart style covariance equality test, details are in thepaper.

The idea is to identify whether the conditional covariance matrix att is equal to the long run covariance matrix Σ from the modelresiduals.



Results

Data set

Given the controversy surrounding the reporting of various creditspread indices, we have constructed our data set, where possible,from the transaction history.

The data set is sourced from Thomson-Reuters Tick History andDataStream. Sovereign bond data is collected using the ‘SuperRICs’ or Reuters Information Codes.

The super-RICs collect all trades on instruments in the tag range setby the code, i.e. AT5YT=RR literally means pull all yields on tradedbonds with a 5 year maturity from the daily collection date.

We use the same approach for the CDS market, however aggregationis much more complex. Multiple data vendors provide an array ofintra-day and end-of-day information, through Markit and CMA.

The CDS data set is then hand built from these sources andcombined into a daily index.



Results

Countries in sample

We collect all traded sovereign bonds with a maturity of 5 and 10years for the countries selected in the sample.

Originally all Eurozone countries were included in the sample.

However, credit default swaps have only been actively traded on tencountries for a long enough period to permit analysis.

These countries are Austria, Belgium, France, Germany (thebenchmark), Greece, Ireland, Italy, Netherlands, Portugal and Spain.

The next slide lists the various CDS sources that CMA and Markituse when building the index of daily spreads.



Results

ABN AMRO ANZ Investment Bank (Asia)Barclays CDS NYC Barclays TokyoBNP Paribas Citigroup Global MktsDeutsche Bank NY Deutsche Bank SingaporeDZ Bank, Frankfurt GFI Market RecapHandelsbanken HypovereinsbankICAP ING ManilaJ.P.Morgan Mizuho SecuritiesNatexis Nord LB, HannoverRBS Japan SEBStandard Chartered Singapore TIFFETullett Prebon UBS JapanUBS Singapore CMAMarkit



Results

Country Ticks Zero Yields Corrupted Rogue Trading DaysSpain 4,010,003 606 0 24 1,339Austria 5,609,129 348 0 12 1,339Belgium 978,395 55,981 0 0 1,339France 708,122 31,168 0 2 1,339Germany 2,141,828 61 0 2 1,339Greece 2,800,111 18,574 0 4 1,339Ireland 3,151,086 4,982 0 6 1,339Italy 3,800,255 299,131 0 3 1,339Netherlands 4,866,969 593 0 4 1,339Portugal 4,616,628 10,253 0 3 1,339



Results

Results

Large number of results in the paper, appendix and internetappendix.

The results are ordered in the paper as follows:

Breakpoint tests (points at which the market has appeared to changepricing model).Time varying roots (detecting the presence of explosive stochastictrends, helpful for policy makers).Time varying coefficients (direction of price discovery mechanism inthe market).

First: A visual inspection of the data for Greece, Ireland, theNetherlands and France.



Results

Greek credit spreads

Q1−07 Q2−07 Q3−07 Q4−07 Q1−08 Q2−08 Q3−08 Q4−08 Q1−09 Q2−09 Q3−09 Q4−09 Q1−10 Q2−10 Q3−10 Q4−100

200

400

600

800

1000

1200

1400

Credit Spreads 5 Year

BondCDS


200

400

600

800

1000


BondCDS



Results

Greek liquidity spreads

Q1−07 Q2−07 Q3−07 Q4−07 Q1−08 Q2−08 Q3−08 Q4−08 Q1−09 Q2−09 Q3−09 Q4−09 Q1−10 Q2−10 Q3−10 Q4−10−50

0

50

100

150

200

250

Liquidity Spreads 5 Year

BondCDS


20

40

60

80

100

120


BondCDS



Results

Irish credit Spreads


0

100

200

300

400

500


BondCDS


0

100

200

300

400

500


BondCDS



Results

Irish liquidity Spreads


10

20

30

40

50

60

70

80


BondCDS


5

10

15

20

25

30

35

40


BondCDS



Results

Dutch credit spreads


0

20

40

60

80

100


BondCDS


0

20

40

60

80

100

120


BondCDS



Results

Dutch liquidity spreads


5

10

15

20

25

30


BondCDS


5

10

15

20

25

30

35


BondCDS



Results

French credit spreads


0

20

40

60

80

100


BondCDS


0

10

20

30

40

50

60

70


BondCDS



Results

French liquidity spreads


5

10

15

20


BondCDS


5

10

15


BondCDS



Results

Detected First Variance Breakpoints

Austria (AT) Belgium (BE)5 Year 10 Year 5 Year 10 YearFebruary 2008 May 2007 August 2007 May 2008

France (FR) Greece (GR)5 Year 10 Year 5 Year 10 YearMarch 2008 January 2007 January January 2007

Ireland (IE) Italy (IT)5 Year 10 Year 5 Year 10 YearMay 2008 September 2008 March 2008 November 2009

Netherlands (NL) Portugal (PT)5 Year 10 Year 5 Year 10 YearMarch 2008 January 2007 February 2008 March 2010

Spain (ES)5 Year 10 YearMarch 2008 August 2007



Results

Next Few Slides

Document the time varying roots of the first order coefficientsmatrix, for Greece and Portugal.

Roots above unity indicate the presence of explosive trends.

Roots equal to one indicate that there is at least one random walk inthe vector process.

In our internet appendix, we document the results for every countryand adjust the nuisance parameters in the weighting system toillustrate the robustness of the results.

For the smallest root, if it is very large, then this indicates a jointlyexplosive trend.



Results

5 year Greek model roots.


0.2

0.4

0.6

0.8

1

1.2

1.4

Roots

Largest RootSmallest Root



Results

10 year Greek model roots.

Q1−07 Q2−07 Q3−07 Q4−07 Q1−08 Q2−08 Q3−08 Q4−08 Q1−09 Q2−09 Q3−09 Q4−09 Q1−10 Q2−10 Q3−10 Q4−10−0.2

0

0.2

0.4

0.6

0.8

1

1.2

Roots




Results

5 year Portuguese model roots.

Q1−07 Q2−07 Q3−07 Q4−07 Q1−08 Q2−08 Q3−08 Q4−08 Q1−09 Q2−09 Q3−09 Q4−09 Q1−10 Q2−10 Q3−10 Q4−100.2

0.3

0.4

0.5

0.6

0.7

0.8

0.9

1

1.1

1.2

Roots




Results

10 year Portuguese model roots.


0.2

0.4

0.6

0.8

1

1.2

1.4

Roots




Results

Greek 5 year model first equation

2008 2009 20100.5

0.6

0.7

0.8

0.9

1

1.1

β1,1,t

2008 2009 2010

−0.2

−0.15

−0.1

−0.05

0

0.05

0.1

0.15

β1,2,t

2008 2009 2010

−0.4

−0.2

0

0.2

0.4

0.6

β1,3,t

2008 2009 2010

−0.4

−0.3

−0.2

−0.1

0

0.1

0.2

0.3

0.4

0.5

β1,4,t



Results

Greek 5 year model second equation

2008 2009 2010

−0.05

0

0.05

0.1

0.15

0.2

0.25

0.3

0.35

0.4

β2,1,t

2008 2009 20100.5

0.6

0.7

0.8

0.9

1

1.1

β2,2,t

2008 2009 2010

−0.6

−0.5

−0.4

−0.3

−0.2

−0.1

0

0.1

0.2

0.3

β2,3,t

2008 2009 2010−0.4

−0.2

0

0.2

0.4

0.6

0.8

1

1.2

β2,4,t



Results

Greek 5 year model third equation

2008 2009 2010

−0.4

−0.2

0

0.2

0.4

0.6

β3,1,t

2008 2009 2010−0.5

−0.4

−0.3

−0.2

−0.1

0

0.1

0.2

0.3

0.4

β3,2,t

2008 2009 2010

−0.2

0

0.2

0.4

0.6

0.8

β3,3,t

2008 2009 2010−2.5

−2

−1.5

−1

−0.5

0

0.5

1

1.5

β3,4,t



Results

Greek 5 year model fourth equation

2008 2009 2010

−0.2

0

0.2

0.4

0.6

β4,1,t

2008 2009 2010

−0.5

0

0.5

β4,2,t

2008 2009 2010

0

0.5

1

1.5

2

2.5

β4,3,t

2008 2009 20100

0.2

0.4

0.6

0.8

1

β4,4,t



Results

Observations

Most important take home messages:

Explosive trends present at times in almost all Eurozone countriesand in particular Greece, Ireland and Portugal.At this point the market has ceased to function in the normalmanner.Without intervention the discount rate would have been driven toinfinity.There is a time varying transmission effect from the CDS liquidityspread to the bond market credit spread (violates the nearlycomplete market condition of Jarrow-Protter 2005).



Results

Policy Implication

Setting the effective rate of interest using the market rates, justprior to bailout is inappropriate.

At this point the market has ceased to price new information anddefault is already priced in, before it has happened.

This is most certainly a liquidity effect.

At points this liquidity effect is NOT from the bond market, butfrom the CDS.

Which the authors believe is part of a case for banning what shouldbe a redundant asset.


giovanni calice. spillovers in sovereign bond and cds markets: an analysis of the eurozone sovereign...

Economy & Finance

time t

t cdscst1

t cdslst1

t bon dcst1

t bon dlst1

sovereign bond liquidity

year sovereign cds

cds liquidity